You are on page 1of 498

This page intentionally left blank

Numerical Notation
This book is a cross-cultural reference volume of all attested numerical notation
systems (graphic, nonphonetic systems for representing numbers), encompassing
more than 100 such systems used over the past 5,500 years. Using a typology that
defies progressive, unilinear evolutionary models of change, Stephen Chrisomalis
identifies five basic types of numerical notation systems, using a cultural phylogenetic framework to show relationships between systems and to create a general
theory of change in numerical systems. Numerical notation systems are primarily representational systems, not computational technologies. Cognitive factors
that help explain how numerical systems change relate to general principles, such
as conciseness and avoidance of ambiguity, which also apply to writing systems.
The transformation and replacement of numerical notation systems relate to specific social, economic, and technological changes, such as the development of the
printing press and the expansion of the global world-system.
Stephen Chrisomalis is an assistant professor of anthropology at Wayne State University in Detroit, Michigan. He completed his Ph.D. at McGill University in
Montreal, Quebec, where he studied under the late Bruce Trigger. Chrisomaliss
work has appeared in journals including Antiquity, Cambridge Archaeological Journal, and Cross-Cultural Research. He is the editor of the Stop: Toutes Directions
project and the author of the academic weblog Glossographia.

Numerical Notation
A Comparative History

Stephen Chrisomalis
Wayne State University

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,


So Paulo, Delhi, Dubai, Tokyo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521878180
Stephen Chrisomalis 2010
This publication is in copyright. Subject to statutory exception and to the
provision of relevant collective licensing agreements, no reproduction of any part
may take place without the written permission of Cambridge University Press.
First published in print format 2010
ISBN-13

978-0-511-67934-6

eBook (EBL)

ISBN-13

978-0-521-87818-0

Hardback

Cambridge University Press has no responsibility for the persistence or accuracy


of urls for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.

Contents

Acknowledgments

page vii

Introduction

Hieroglyphic Systems

34

Levantine Systems

68

Italic Systems

93

Alphabetic Systems

133

6 South Asian Systems

188

Mesopotamian Systems

228

East Asian Systems

259

Mesoamerican Systems

284

10

Miscellaneous Systems

309

11

Cognitive and Structural Analysis

360

Contents

vi
12

Social and Historical Analysis

401

13

Conclusion

430

Glossary

435

Bibliography

439

Index

471

Acknowledgments

Although the history of scholarship on numeration is lengthy and includes such


illustrious figures as Alexander von Humboldt, Alfred Kroeber, and Oswald Spengler,
its temporal and spatial breadth inevitably means that its practitioners frequently
operate in a seeming near-vacuum. For this reason I am doubly grateful for the
assistance I have received over the decade since this works inception.
This book had its genesis during my time at McGill University. The late Bruce
Trigger was the shepherd and guiding hand behind this book, beginning in its
formative stages and continuing almost to the final draft. The central premise of
this book stems from Bruces conviction that comparative research is not only
possible but indeed necessary in order for anthropology to be theoretically meaningful. Without Bruces mentorship and support for me throughout this decidedly
unorthodox anthropological pursuit, this book would not exist. Bruces death in
2006 was a momentous loss for the discipline and for comparativism.
At McGill, in addition to Bruce, Michael Bisson, Andre Costopoulos, Jim Lambek,
and Jerome Rousseau read the manuscript and provided useful suggestions for
improvement at various stages, as well as providing invaluable moral support to
me. Funding at this stage of the research was provided by a Social Sciences and
Humanities Research Council (Canada) doctoral fellowship. I wish to thank particularly the interlibrary loan staff at McGills McLennan Library, who went well
beyond the call of duty in tracking down obscure material.

vii

viii

Acknowledgments

Further refinements and a new draft of the book were produced under a SSHRC
postdoctoral fellowship at the University of Toronto. While I was in Toronto,
Richard Lee, Trueman MacHenry, and David Olson were particularly helpful
to me and provided useful insights on the theories and concepts underlying my
work, forcing me to clarify my own positions in ways that I had not previously
done. Bob Bunker, John Gilks, Heather Hatch, Andy Pope, and Shana Worthen
read portions of the manuscript at this stage and provided very useful editorial
advice.
A work of this scope inevitably relies upon the individual and collective experience of regional specialists in the writing systems and mathematical practices of
various regions and periods, and of theorists working in cognitive and psychologically oriented anthropology and linguistics. I have benefited tremendously from
the specialized expertise of Priskin Gyula, Christopher Hallpike, Jim Hurford,
Joel Kalvesmaki, Eleanor Robson, Nicholas Sims-Williams, Matthew Stolper, and
Konrad Tuchscherer. A School of Advanced Research Advanced Seminar entitled
The Shape of Script was the key to moving my work into its final completed form,
and introduced me to many additional regional specialists whose advice has been
of assistance: John Baines, John Bodel, Beatrice Gruendler, Stephen Houston,
David Lurie, Kyle McCarter, John Monaghan, Richard Salomon, Kyle Steinke,
and Niek Veldhuis.
Scholars of numeration include historians, anthropologists, archaeologists, linguists, mathematicians, and psychologists, to name only a few, and it is all too
easy in such a disparate crowd of research traditions to lack a sense of disciplinary
cohesion and of ones scholarly influences. I therefore acknowledge my intellectual forebears in the comparative study of numerals, most notably Florian Cajori,
Genevieve Guitel, Karl Menninger, Antoine Pihan, and David Eugene Smith.
Although I disagree with his conclusions in many places, I thank Georges Ifrah,
whose gargantuan and important Histoire universelle des chiffres (1998) inspired me
to produce this volume.
Eric Crahan and Frank Smith at Cambridge University Press deserve great
credit for their skillful guidance of my work through the editorial process at all
stages. Russell Hahn guided the complex copyediting masterfully, and Leah Shapardanis prepared the index and read proofs. Many thanks to the fourteen anonymous reviewers who read and commented on one or more chapters on behalf of
the Press, and to the entirety of the production staff for their handling of dozens
of specialized typefaces.
To my family, all my love and thanks. Arthur Chrisomalis provided useful
firsthand insights into the childhood acquisition of lexical and graphic numeration, and rekindled his fathers wonderment at the magic of numbers. Finally, this
work is dedicated with love to my wife, Julia Pope, for her patience with me over

Acknowledgments

ix

the past decade, her keen editorial eye, her endless willingness to reread manuscript chapters, and her ongoing conviction that this work is worthwhile.
Despite the advice and assistance of the abovementioned, and any others I have
forgotten, I have doubtless made many errors of fact and interpretation, and I
eagerly anticipate the opportunity to broaden my knowledge of numerical notation systems in the future.

Notes on Style
Throughout the book I have used the conventions bc and ad to refer to chronological periods. Where no era indicator is associated, ad dates are assumed; I do
so only when the interpretation of a date is obvious.

chapter 1

Introduction

The Western world is a world of written numbers. One can hardly imagine an industrial civilization functioning without the digits 0 through 9 or a similar system.
Yet while these digits have pervasive social and cognitive effects, many unanswered
questions remain concerning how humans use numerals. Why do societies enumerate? How does the representation of numbers today differ from their representation
in the past? Why does the visual representation of number figure so prominently in
complex states? What cognitive and social functions are served by numerical notation systems? How do numeral systems spread from society to society, and how do
they change when they do so? And, despite their present ubiquity, why have the
vast majority of human societies not possessed them at all?
If you look up from this page and examine your surroundings, I am certain that
you will encounter at least one instance of numerical notation, probably more.
Moreover, unless you have a Roman numeral clock nearby, I am nearly certain
that all of the numerals you encounter are those of the Hindu-Arabic or Western1
system. Numerals serve a wide variety of functions: denotation Call George,
1

The conventional term used in popular literature, Arabic numerals, and the term used
in most scholarly literature, Hindu-Arabic numerals, can lead to considerable confusion because the scripts used to write the Hindi and Arabic languages use numerical
notation systems that differ from those of the West in the shape of the signs. I use the
term Western numerals to refer to this system because it developed in Western Europe
in the late Middle Ages, while fully acknowledging its Indian and Arabic ancestry.

Numerical Notation

8765000; computation 21.00 1.15 = 24.15; valuation 25 cents; ordination 1. Wash dishes, 2. Sweep floor, 3. Finish manuscript; and so on. Most of
the thousands of numerals we see each day barely register on our conscious minds;
regardless, we encounter far more written numbers in our lifetime than we do
sunsets, songs, or smiles. Until the past few centuries, the opposite was true for
most people.
These ten digits are so prevalent that it is easy to equate our numeral-signs
with the set of abstract numbers. In this view, 62 does not merely signify the
abstract concept sixty-two it is the raw form of the number itself, the stuff of
pure mathematics (or perhaps pure numerology). That these signs are frequently
encountered and used in mathematical contexts contributes to the prevalence
of such attitudes. According to this view, our numeral-signs constitute abstract
number, and other systems (when recognized as such) are simply archaic deviations from the abstract entity comprised by these signs.
This view is erroneous, and rests on the confusion of a mental concept (signified) with its symbolic representation (signifier). Our numerical notation system has an extensive history, as do the more than one hundred systems that
have existed over the past five thousand years. Still, the worldwide prevalence of
Western numerical notation is undeniable. Most literate individuals worldwide,
as well as a sizable number of illiterates, understand them. Nor does any competing system have any reasonable chance of supplanting our system in the near
future. This has led many scholars to assert its supremacy solely on the evidence
of its near-universality (Zhang and Norman 1995; Dehaene 1997; Ifrah 1998).
Nevertheless, this situation does not imply that our system will dominate the
whole world forever. The study of numerical notation remains mired in a theoretical framework that has much more in common with late nineteenth-century
unilinear evolutionism in anthropology than it does with early twenty-firstcentury critiques of unfettered scientific progress.
Despite this theoretical weakness, numerical notation as a topic of academic
study is a relatively common pursuit, with linguists, epigraphers, archaeologists,
anthropologists, historians, psychologists, and mathematicians all making significant contributions to the literature. These studies are mostly restricted to the
analysis of one or a few numerical notation systems, although a small number
of synthetic and comparative works dealing with numerical notation exist
(Cajori 1928; Menninger 1969; Guitel 1975; Ifrah 1998). However, such works
rarely consider more obscure numerical notation systems, such as those of subSaharan Africa, North America, and Central Asia. Similarly, social scientists such
as the anthropologist Thomas Crump (1990), the psychologist David Lancy (1983),
and the ethnomathematicians Marcia Ascher (1991) and Claudia Zaslavsky (1973)
have undertaken major comparative research on numeracy and mathematics in

Introduction

non-Western societies. Yet numerical notation has not been a primary focus of
this body of research.
This study is a comparative analysis of all numerical notation systems known
to have existed throughout history approximately one hundred distinct systems,
most of which can be grouped into eight distinct subgroups. By presenting a
universal study of such systems and examining the historical connections and
contexts in which they are encountered, I will develop a framework that accounts
for cultural universals, identifies evolutionary regularities, and yet remains cognizant of idiosyncratic features, seeking to determine, rather than to assume, the
amount of intercultural variability among them. I will distinguish several major
types of numerical notation, evaluate their efficiency for performing specific functions, link their features to human cognitive capacities, and relate systems to their
sociopolitical contexts.

Definitions
A numerical notation system is a visual, relatively permanent, and primarily nonphonetic structured system for representing numbers. Signs such as 9 and 68, IX
and LXVIII, are part of numerical notation systems, but numeral words such as
nine and achtundsechzig are not. Though there are ties between numeral words
and numerical notation, a lexical numeral system, or the sequence of numeral
words in a language (whether written or spoken), has a language-specific phonetic
component. Every language has a lexical numeral system of some sort, while numerical notation is an invented technology that may or may not be present in a
society.2 Some numerical notation systems contain a small phonetic component,
as in acrophonic systems whose signs are derived from the first letters of the appropriate number-words in a language. However, since such systems are still comprehensible without having to understand a specific language, they are numerical
notation systems.
Numerical notation systems must be structured. Simple and relatively unstructured techniques, such as marking lines on a jailhouse cell to count ones days or
piling pebbles in a basket, are largely or entirely unstructured. They rely on oneto-one correspondence, in which things are counted by associating them with an
equal number of marks or other identical objects. A numerical notation system,
by contrast, is a system of different discrete numeral-signs: single elementary symbols, or, in the terminology used in writing systems, graphemes, which are then
2

I will leave aside for the moment discussions of counterevidence questioning the assumption of the universality of lexical numeral systems (Hurford 1987: 6878; Gordon
2004; Everett 2005).

Numerical Notation

used in combination to represent numbers.3 A numeral-phrase is a group of one


or more numeral-signs used to express a specific number (e.g., MMDXXV);
numeral-phrases such as 8 or Roman L are nonetheless complete even though they
only use one sign apiece.
All numerical notation systems (and most lexical numeral systems) are structured by means of powers of one or more bases. A power is a number X multiplied
by itself some number of times (its power); 101 = 10, 102 = 100, 103 = 1000, etc.
By mathematical definition, a number raised to the power 0 equals 1. A base is
a natural number B in which powers of B are specially designated. While mathematicians normally require that a base be extendable to an infinite number of
powers of B (e.g., 10, 100, 1000, 10,000, ... ad infinitum), most numerical notation systems are not infinitely extendable. It is sufficient that some powers of B are
specially designated within a numerical notation system. Western numerals and
many other systems use a base of 10, but this is not universal. In addition to its
base, a numerical notation system may have one or more sub-bases that structure
it. The Roman numeral system has a primary base of 10 with a sub-base of 5. Unlike bases, the powers of sub-bases are not specially designated; there are no special
Roman numerals for 25 or 125. It is, rather, the products of a sub-base and the powers of the primary base that are specially designated for the Roman numerals,
50 (5 10) and 500 (5 100).
Two topics that I will present only peripherally are number and mathematics.
Number is an abstract concept used to designate quantity. For the purposes of my
study, a simple (if philosophically nave) definition will suffice. Questions such
as whether numbers are real or Platonic entities, or the connection of the set of
natural numbers to formal logic, are beyond its scope. The distinction between
cardinal numbers denoting quantity but not order and ordinal numbers
designating ordered sequences is extremely important for lexical numerals,
where many languages use different series of words (e.g., two versus second ) for
the two concepts. This distinction also has implications for our understanding
of the origin of numerals and numerical concepts in humans (Crump 1990:
610), but has little influence on numerical notation. In defining mathematics
as the science that deals with the logic of quantity, shape, and arrangement, I
am consciously employing a simple definition for a complex term. In order to
understand numerical notation, one needs no mathematical ability save some
knowledge of basic arithmetic. While some parts of mathematics make frequent
3

A few numeral-signs are more complex in that they graphically combine two or more
signs into one in order to represent multiplication, but they are treated as elementary
numeral-signs because their use is identical to that of all other simple signs in the systems in question.

Introduction

use of numbers (number theory being the most obvious example), large parts
of the discipline have only infrequent or peripheral encounters with numerical
notation. Numerical notation systems are not necessarily designed with mathematical purposes in mind. Even in contemporary industrial societies, where
mathematical ability is more extensive than in any other historical or modern
society, most numerical notation is nonmathematical.

Universal Comparison
The present study is, as far as possible, a universal one. I have not excluded any numerical notation system intentionally save where data are not plentiful enough to
undertake a reasonable analysis. Most comparative research in anthropology aims
to discover generalizations and patterns in human behavior, but using the universe
of cases is neither possible nor desirable in most cross-cultural studies. In order to
use most analytical statistics on cross-cultural data, each case must be independent
of the others, which requires that each case may not be historically derived or diffused from any other case. This issue, known as Galtons problem, is the thorniest
methodological issue in statistical cross-cultural research (Naroll 1968: 258262).
The establishment of correlations between traits among historically independent
societies is enormously useful, and is the basis for most cross-cultural research in
modern anthropology.
Yet to do so in a study such as this one, in which there are perhaps only seven
independently invented numerical notation systems, would be pointless. Firstly,
seven cases would be too small a sample to analyze statistically. Secondly, by
studying all cases, I am able to show that the total observable variability among
numerical notation systems is far greater than has previously been believed. This
variability cannot be understood by studying only a fraction of numerical notation systems. To paraphrase the old fable, if we study only the elephants trunk
or tail, we ignore most of the animal. Thirdly, I wish to explain structural variation among historically related systems, which frequently differ considerably from
their relations. This would be impossible using a sampling technique that omitted
related cases. Finally, were I to omit related cases, I could not analyze how systems
change over time or how new systems develop out of existing ones. By taking
events of change, rather than static systems, as the units of analysis in my comparisons, I am able to elucidate both synchronic and diachronic patterns among
numerical notation systems. It is worth noting that Galtons problem does not
apply to events of change of the sort I am analyzing, since every event is essentially
independent of every other, and can thus be analyzed statistically, where relevant.
I reject as false the dichotomy in anthropology between universalism (Tylor 1958
[1871], White 1949, 1959; Steward 1955; Harris 1968) and relativism (Lowie 1920,

Numerical Notation

Boas 1940, Sahlins 1976, Geertz 1984), both of which presume rather than evaluate
the degree of regularity that social phenomena display. While numerical notation
systems display remarkable regularities and even universals, historical contingencies also played a major role in shaping the cultural history of numerical notation.
Yet the only way to determine which features of numeration are cross-culturally
regular and which are idiosyncratic is to undertake cross-cultural comparison. The
best way to deal with the messiness of the world less universal than universalists
would like, less relative than relativists prefer is through a body of theory that
deals with constraints.
Most anthropological theory is predicated on the existence of very strong constraints on the forms possible within human societies. Some of these constraints
are so strong as to produce cross-cultural universals (Brown 1991). Most cultural
relativists dismiss these universals as minimally true, but facile, irrelevant, and useless for understanding humanity (cf. Geertz 1965, 1984). The denial of comparativism on this basis is an overly negative position, given that those who criticize
comparativism most harshly are very often those who have not undertaken it. One
of the most crucial theoretical contributions of anthropology should be to indicate
the degree to which human societies are alike and the degree to which they differ.
While some aspects of human existence are truly universal, and others are almost
infinitely variable, most of the really intriguing domains of activity fall somewhere
in the middle.
In the early 1900s, Alexander Goldenweiser developed his principle of limited possibilities, which stated that for any social or cultural phenomenon,
there are a limited number of possible forms that can be expressed in human
societies (Goldenweiser 1913). Goldenweiser was particularly interested in the
limitations imposed by human psychology on the expression of cultural traits,
although, given the inchoate nature of psychological theory at the time, he was
unable to describe these mechanisms precisely. Bruce Trigger (1991) has rejuvenated the idea of constraints, proposing that anthropologists should use the
concept of constraint to describe the limitations on human sociocultural variation whether those constraints are biological, ecological, technological, informational, psychological, or historical in order to analyze statistical regularities
among cultures without implying determinism. We must be cautious, with both
the limited possibilities and the constraint approaches, not to restrict our
formulations and assume the restricting influence of various factors to be more
important than positive (enabling) effects. A very strong propensity in favor
of some trait is not the same thing as a very strong constraint against all other
possibilities. Constraints and inclinations can and do coexist, and the negative
limitations of one variable must be weighed against the positive inclinations of
another. Despite this caveat, I find a constraint-based approach to be the most

Introduction

promising theoretical perspective for explaining the regularities found in numerical notation systems, something to which I will return in Chapters 11 and 12.
In much of my analysis, I follow Joseph Greenberg (1978), whose analysis of significant regularities in lexical numeral systems presents a list of fifty-four generalizations. Unlike much of his later work, Greenbergs study of numerals is universal
and cognitive in orientation rather than phylogenetic. It is synthetic, based on the
detailed empirical work of earlier scholars, such as the German linguist Theodor
Kluge, who spent years compiling sets of numeral terms in languages throughout
the world (Kluge 193742). While many of Greenbergs regularities are extremely
complex4 or have some exceptions, others reveal truly universal and nontrivial
features of every natural language; for instance, every numeral system contains
a complete set of integers between one and some upper limit each system is
finite5 and has no gaps (Greenberg 1978: 253255). Similarly, no natural language
expresses two as ten minus eight or twenty as one-fifth of one hundred.
While every language has a set of lexical numerals, most pre-modern societies
functioned quite well without numerical notation. It is possible to conceive of
a world in which there are many regularities in lexical numerals, but in which
numerical notation systems are highly specific and unique responses to local
needs. We do not live in such a world. There is considerable uniformity among
the worlds numerical notation systems, and they display many synchronic and
diachronic regularities.
In fact, the number and variety of conceivable numerical notation systems is
far greater than what is attested historically. To take only a very limited example, a
numerical notation system can very easily be imagined that is just like the Western
system but instead of being a decimal system having a base of any natural6
number of 2 or higher. Yet most numerical notation systems have a base-10 structure (and those that do not use multiples of 10). This does not preclude the existence of binary and hexadecimal numerical notation for specialized computing
purposes. Similarly, while there are only five basic principles of numerical notation
systems found historically (as described earlier), it is easy to imagine other types
that could have existed: a system where the size of a numeral-sign is relevant to its
4

For instance: 37. If a numeral expression contains a complex constituent, then the numerical value of the complex constituent itself in isolation receives either simple lexical
expression or is expressed by the same function and in the same phonological shape,
except for possible automatic phonological alternations, stress shifts, or overt expressions
of coordination (Greenberg 1978: 279280).
This is not true of numerical notation systems, some of which (like our own) are truly
infinite.
Or even, as discussed in some aspects of number theory, having a fractional or negative
base!

Numerical Notation

value, or where all nonprime numbers are expressed multiplicatively using prime
number numeral-signs. Several modern writers, abandoning traditional principles
of numerical notation, have created new systems ex nihilo that rely on rather different principles than do the systems discussed in this study (Harris 1905; Pohl
1966; Dwornik 198081). Explaining regularities from a constraint-based perspective allows us to speculate about why certain numerical notation systems flourish
while others do not. Instead of denying the existence of exceptions, I use general
rules to explain why special cases are special, and why some imaginable systems are
unattested in the ethnographic and historical records.
Yet one might wish to contend that comparison of any sort, much less the
universal type of this study, is misleading because each culture, and hence each
numerical notation system, is a product of unique historical circumstances. If so,
comparing Egyptian hieroglyphic numerals to Shang oracle-bone numerals and
Inka khipus might be misleading. At best, even if there is a core of features common to all numerical notation systems, I would be labeling oranges apples in order
to compare them to other apples. At worst, if these systems are entirely different
phenomena, I am trying to make apples out of abaci. Yet the relative ease of intercultural communication refutes the claim that all cultures are incommensurable.
The intercultural transmission of ideas relating to numerical notation systems is
frequent and poses a serious challenge to this degree of relativism. Prior to comparing phenomena among multiple societies, one cannot assume either that the phenomenon is cross-culturally regular or that it is not. Having compared numerical
notation systems on a worldwide basis, I regard the systems as being sufficiently
similar to warrant their theoretical analysis as variations on a single theme.
I regard numerical notation as translatable cross-culturally without significant
loss of information or change of meaning. The number 1138 is practically identical
in referent to MCXXXVIII or t\s\rrr\qqq\qqq\qq or any other representation. These
systems have very different structures, but, in Saussurean terms, the various signifiers refer to the same signified (Saussure 1959). Although the linguistic and symbolic signifiers for numbers may differ greatly (23, dreiundzwanzig, XXIII, viginti
tres, etc.), the correlation of both numeral-phrases and lexical numerals with natural numbers is not culturally relative. Yet, while seemingly uncontroversial in the
exact sciences, the cross-cultural universality of number concepts has been criticized recently by relativistic anthropologists and sociologists. In his recent work
on Quechua number and arithmetic, Urton (1997) asserts that Western concepts
such as odd/even are not appropriate to the Quechua arithmetical experience,
and that the Quechua use a fundamentally different ontology of numbers than the
Western one. Yet Quechua numbers can be understood in the same way as any others, and the Inka numerical notation used by Quechua speakers (Chapter 10) can be
compared to others without any particular difficulty. Relativist philosophers such

Introduction

as Restivo (1992) claim that 1 + 1 cannot equal 2 in any absolute manner, because
if one were to take a cup of popcorn and add a cup of milk to it, the result would
not be two of anything, but somewhat more than a cup of pulpy mush. Resisting
the temptation to describe such casuistry as pulpy mush, I simply point out that
addition is an arithmetical function that can only represent adding discrete objects
of a like nature. Such evidence does not convince me that the number concepts
of non-Western societies are incommensurable with our own. On the contrary,
my own research suggests that these differences are relatively inconsequential in
comparison to the commonalities observed in all societies.
I acknowledge that, by treating all numerical notation systems purely as systems for representing number, I do not do justice to the complex symbolism that
complements many of them or to the scholarship on numerology (Hopper 1938,
Crump 1990). The arrival of the year 2000 was not simply another cause for celebration (or trepidation); rather, the nature of our numerical notation system and
the rolling over of the calendrical odometer on 2000/01/01 held great symbolic
and even mystical significance for much of the worlds population. My decision
to underemphasize numerology is based partly on space limitations, but also on
my theoretical interest in the comparable core of features underlying all lexical
numeral systems and numerical notation systems. These interesting differences do
not affect the validity of cross-cultural comparisons, but merely highlight the need
to establish, rather than assume, the level of regularity in sociocultural phenomena. It may be true, as Geertz (1984: 276) famously asserted, that [i]f we wanted
home truths, we should have stayed at home, but if we want human truths, we
must compare.

Structural Typology of Numerical Notation


The systematic classification of numerical notation systems helps to identify their
relevant features, distinguish independent inventions from cultural borrowings,
and determine how their features relate to their uses. The goal of typology is not
simply to develop a scheme into which every case fits, but to do so in a way that
allows us to ask and answer questions that could not otherwise be considered.
When poorly done, typology is descriptive but nonanalytical, and thus largely
useless; when well done, it organizes knowledge in a way that answers inquiries.
Any classificatory scheme is inherently theory-laden, and answers only some of the
questions that might be asked of a set of data. The typology presented here represents all the major principles by which numbers are represented and emphasizes
the features of numerical notation that are cognitively most important. It removes
each system from its temporal, geographic, and spatial contexts and examines how
numeral-signs are combined to represent numbers.

10

Numerical Notation

Any natural number can be expressed as the sum of multiples of powers of some
base. In Western numerals, 4637 is 4 1000 + 6 100 + 3 10 + 7 1 or, to use
exponential notation, 4 103 + 6 102 + 3 101 + 7 100. Because the Western
numerals use the principle of place-value, the value of any numeral-sign in the
phrase is determined by its position position dictates the power of the base that
is to be multiplied by the sign in question. If the order changes, the value changes,
so that 6437, 3674, and so on mean different things than 4637. We could also write
the number out lexically as four thousand six hundred and thirty seven. Instead of
using place-value, the powers (except for 1) are expressed explicitly thousand, hundred, -ty. Because each multiplier corresponds to a word for a power, we could in
theory move each power and its multiplier to a different spot without introducing
ambiguity; German lexical numerals, among others, do exactly that viertausend
sechshundert sieben und dreizig four thousand six hundred seven and thirty.
Some numerical systems, however, do not use multiplication at all. To use the
Roman numerals, one simply adds up the values of all the signs: MMMMDCXXXVII 1000 + 1000 + 1000 + 1000 + 500 + 100 + 10 + 10 + 10 + 5 + 1 + 1.
Although there is no logical requirement that systems like the Roman numerals
list their powers in order, they almost universally do so. The Roman numeral
CCLXXVIII could be unambiguously read even if it were written as VIIICCXXL,
or even as XLIVIXCIC, if we omit the slight complexity of the occasional use of
subtraction. The fact that such disordered phrases are not valid tells us something
about systems that lack place-value, however they too are structured as the sum
of multiples of powers. We can thus interpret the Roman numeral MMMMDCXXXVII as (1000 + 1000 + 1000 + 1000) + (500 + 100) + (10 + 10 + 10) + (5 + 1 + 1).
I will return in a moment to the issue of the signs for 500 and 5, and how they
affect our understanding of such systems.
The only major systematic attempt to date to classify numerical notation systems is Genevive Guitels Histoire compare des numrations crites (1975).7 Guitel
classifies approximately twenty-five systems (drawn from about a dozen societies)
according to whether they use addition alone to form numeral-phrases (Type I,
like Roman numerals), addition and explicit multiplication (Type II, like English
lexical numerals), or implicit multiplication with place-value (Type III, like Western numerals) just as I have done here. Each type is further subdivided according
to the systems base(s) and other features. Despite an admirable attempt, Guitels
analysis fails the most basic test of classification, which is that it must classify
similar systems together and separate dissimilar ones. It is problematic because its
primary division is made only on the basis of the degree to which multiplication
7

See also Zhang and Norman (1995). Ifrahs (1985, 1998) popular studies on the subject
follow Guitels typology.

Introduction

11

is used in forming numeral-phrases. Her typology assumes that, since positional


notation is very important to our own numerical notation system, positionality is
the primary criterion by which all numerical notation systems should be judged. It
is not factually incorrect in any significant way; systems that are identically structured do end up in the same category. Yet because it is tied to the misleading question To what extent does system X use multiplication to form numeral-phrases?,
it fails to represent fully the similarities and differences among numerical notation
systems.8
There is another way to approach the subject, however. Consider a different
system, namely, the Babylonian cuneiform positional system, which expresses
numbers as combinations of signs for 1 (1) and 10 (a). Like the Roman numerals, the Babylonian system relies on the repetition of like symbols the number
37 is written as three signs for 10 followed by seven for 1. However, the system
also has a base of 60, and multiples of 60 and powers of 60 are written using
the principle of place-value. It shares the use of implied multiplication through
place-value with the Western numerals but shares the use of repeated added
signs with the Roman system. Yet according to Guitels typology there is nothing in common between the Babylonian and Roman systems. The resolution to
this difficulty is that when one writes a numeral-phrase in any system, one is
actually doing two things: expressing the value associated with each power of
the base, and then organizing all the powers in a numeral-phrase into a single
total value. Thus, I distinguish two separate dimensions of numerical notation
systems, which I call intraexponential and interexponential, in order to analyze
them adequately.
Intraexponential organization determines how numeral-signs are constituted
and combined within each power of the base. The major types of intraexponential
organization are cumulative, ciphered, and multiplicative. Cumulative systems
are those in which the value of any power of the base is represented through the
repetition of numeral-signs, each of which represents one times the power value of
the sign, and which are then added. For instance, XXX = 30 in the Roman system
because the sign for 10 (X) is repeated three times. Ciphered systems, on the other
hand, use at most a single numeral-sign for each power represented, with different signs being used to represent different multiples of the power. The Western
numerals are ciphered: a number that has as 103 (thousands) as its highest power
(e.g., 1984) will require at most four symbols, one for each power of the base. Multiplicative systems have two components for each power represented: a unit-sign
(or sometimes multiple signs), which represents the quantity of that power needed
to represent the number, and a power-sign, which represents a power of the base.
8

See Chrisomalis (2004) for a more detailed critique of Guitels typology.

12

Numerical Notation

The product of the two signs determines the value of that power. The multiplicative principle is often used only to structure higher powers of the base.
Interexponential organization determines how the values of the signs for each
power of the base are combined to symbolize the value of each entire numeralphrase. It is subdivided into additive and positional subtypes. Additive systems
are those in which the sum of the intraexponential values in a numeral-phrase
produces its total value. For instance, the Roman numeral CCLXXVIII consists of
two 100s (102), one 50 (5 101), two 10s (101), one 5 (5 100), and three 1s (100), for
a total of 278. Positional or place-value systems, of which the Western system is the
best known, are those in which the value of a numeral-phrase is determined not
only by its constituent numeral-signs but also by the place of each sign within the
phrase. The intraexponential values within a numeral-phrase must all be multiplied by the appropriate power-values before the sum of the phrase can be taken.
All numerical notation systems are structured both intra- and interexponentially,
creating six theoretically possible pairings of principles. However, it is logically
impossible for a multiplicative-positional system to exist because multiplicative
systems represent the required positional value (10, 100, 1000, etc.) intraexponentially, leaving only five possibilities, as detailed in Table 1.1.
Cumulative-additive systems, such as Roman numerals, have one sign for each
power of the base; the signs within each power are repeated and their values added,
and then the total value of the phrase is the sum of the signs. Cumulative-positional
systems likewise use repeated signs to indicate the value of each power, but this value
is then multiplied by the place-values (in the Babylonian example used earlier, 60
and 1) before summing the phrase. In order to be entirely unambiguous, some sort
of placeholder or zero sign is required. Ciphered-additive systems have a unique sign
for each multiple of each power of the base (19, 1090, 100900, etc., in a base-10
system like the Greek alphabetic system); the values of these signs are added to obtain
the value of the numeral-phrase. Ciphered-positional systems like Western numerals
have unique unit signs from one up to but not including the base (e.g., 1, 2, 3 ... 9)
and a zero sign; the unit-value is multiplied by the power-value indicated by its position, and then the sum of these products gives the total value. Finally, multiplicativeadditive systems (like the traditional Chinese system shown earlier, but also for that
matter spoken English lexical numerals, e.g., three thousand six hundred and twenty
four) juxtapose a unit-sign (or signs) and a power-sign, which are multiplied together,
and then the sum of those products gives the total value of the phrase.
Most numerical notation systems use only one of these five combinations
throughout the entire system. However, some additive systems use one intraexponential principle (either cumulative or ciphered) for lower powers of the base,
and then use the multiplicative principle thereafter. These systems, which I call
hybrids, comprise about 30 percent of those I examine in this study. Systems that

Introduction

13

Table 1.1. Typology of numerical notation systems


Additive
The sum of the values of each
power is taken to obtain the total value of the numeral-phrase.

Positional
The value of each power
must be multiplied by a
value dependent on its
position before taking the
sum of the numeral-phrase.

Cumulative
Many signs per
power of the base,
which are added
to obtain the total
value of that power.

Classical Roman

Babylonian cuneiform

1000 + (100 + 100 + 100 + 100) +


(10 + 10 + 10) + (1 + 1 + 1 + 1)

(10 + 10 + 1 + 1 + 1) 60 +
(10 + 10 + 10 + 10 + 10 + 1 +
1 + 1 + 1) 1

Ciphered
Only one sign per
power of the base,
which alone represents the total value
of that power.

Greek alphabetic

Khmer

1434 = /auld

1434 =

((1 )1000 + 400 + 30 + 4)

(1 1000 + 4 100 + 3
10 + 4 1)

Multiplicative
Two components
per power, unitsign(s) and a powersign, multiplied
together, give that
powers total value.

Chinese (traditional)

LOGICALLY
EXCLUDED

1434 =

MCCCCXXXIIII

1434 =

1434 = b3

e4

(1 1000 + 4 100 + 3 10 + 4)

use two principles are not exceptions to my typology. They simply need to be
analyzed in two parts, with each part of the hybrid being assigned the appropriate principle. For instance, the version of the Greek alphabetic system shown in
Table 1.1 is ciphered-additive for powers below 1000 and multiplicative-additive
for those above 1000.9 No numerical notation system employs more than two of
the five basic types, and no positional system uses more than one type.
Systems that have a sub-base as well as a base require further typological clarification because they may use two intraexponential principles: one for units up
to the sub-base, and another for multiples of the sub-base up to the base. For instance, the cumulative-positional Babylonian system shown in Table 1.1 has a base
9

This feature is the one that leads Guitel to place this version of the Greek alphabetic
numerals in her Type II as opposed to Type I.

14

Numerical Notation

of 60 and a sub-base of 10. In this case, we must know both how units from 1 to 9
are expressed and how tens from 10 to 50 are expressed in order to fully describe its
intraexponential structure. In this case, both the sub-base and the base use the cumulative principle, so we might more properly describe this system as a (cumulativecumulative)-positional system. However, no system uses a different intraexponential
principle for its sub-base than for its base, so this elaboration is mostly unnecessary.
Again, none of this affects the interexponential structure of these systems.
Because it reflects both intra- and interexponential principles, this typology
shifts the focus of analysis from systems to the structural principles that build
systems, and thus allows a more nuanced comparison of systems structures.
Moreover, it allows us to ask fruitful questions regarding the cognitive effects and
historical development of numerical notation.

Cognition and Number


Cognitive psychology examines how the brain processes information, including
the study of sensation and perception, concept formation, attention, learning, and
memory. Its methodologies are primarily experimental: because neuroscience cannot yet fully observe the workings of the brain directly, cognitive psychologists study
the brain by its observable outputs the behavior of humans under controlled conditions. Information processing is crucial for human survival, and the ability to form
concepts is a major part of our species evolutionary adaptation. At the same time,
however, these concepts are not perfect representations of reality, because the act of
conceptualization requires some information to be emphasized, and because errors
in information processing reflect the imperfect conceptual abilities of the brain.
In this regard, cognitive psychologists would agree with the archaeologist Gordon
Childes argument that humans do not adapt to the world as it really is, but rather
to the world that they perceive as mediated through culture (Childe 1956: 6568).
Still, Childe insisted, human perceptions must correspond reasonably well to reality
or else we would not survive. Two cognitive questions inform the study of numerical
notation. Firstly, there is the question of origins how did numerical abilities originate, and how do they relate to the origins of numerical notation? Secondly, there is
the question of what cognitive effects, if any, numerical notation has on its users.
George Millers seminal paper on the magic number 7 2 remains an essential
work for understanding how the brain processes number (Miller 1956). Miller
asserts that in several related aspects of human cognition, our capacity for processing information lies between five and nine units. Two aspects of his research are
particularly relevant to the study of number. Firstly, using research conducted by
Kaufman et al. (1949), Miller discusses subitizing, in which small quantities of
figures or objects are perceived directly, while larger quantities must be encoded

Introduction

15

by counting, a more time-consuming process. This experiment involved showing


groups of dots to subjects for one-fifth of a second, after which they would indicate how many were present; up to five or six dots, few errors were made (subjects
were subitizing), while above that number subjects had to estimate and hence
made more errors. In more recent studies, the limit of subitizing has been found to
be somewhat lower than six, ranging around three or four for most experimental
subjects under typical conditions (Mandler and Shebo 1982).
Closely related to subitizing is chunking, which organizes large quantities
of objects into smaller groups, thereby enabling the brain to process the larger
number as a certain number of the smaller sets rather than requiring each object to
be cognized independently. North American telephone numbers of ten digits are
divided into three chunks such as 212-555-2629 rather than written 2125552629,
in part to distinguish the area code, local exchange, and individual phone line
but also to facilitate memorization and recall. Chunking normally involves the
division of a collection of objects into groups of three or four units each, which
speeds up the process of perception and accurate quantification by the brain. The
perception of larger units as gestalts thus maximizes the brains efficiency within
the limits of its biological evolution.
A third element to be considered, related to the first two, is the principle of
one-to-one correspondence defined earlier. This capacity has been studied primarily through research on infants and children (Piaget 1952, Lancy 1983, Wynn
1992). Adults use one-to-one correspondence when they hold up eight fingers to
represent eight coconuts, put aside twenty-seven pebbles to count their flock of
sheep, or mark twelve lines on a sheet of paper to indicate the number of pints
of beer consumed before staggering out of the local pub. Counting (as opposed
to subitizing) cannot take place without one-to-one correspondence. One-to-one
correspondence can be used in combination with chunking to increase the ease
of representation and cognition. After my fifth pint, I might place a horizontal
stroke through the four existing strokes to indicate a group of five; my twelve pints
would thereby be rendered as two groups of five strokes followed by a group of
two (rather erratic) strokes. By extension, numerical notation systems, particularly
cumulative ones, rely on one-to-one correspondence.
Much of the debate on cognitive domains relating to mathematics and its origins takes place in the realm of comparative ethology, specifically studying number
concepts in animals in order to create meaningful analogies with the abilities of
human infants and adults (Fuson 1988, Gallistel 1990, Dehaene 1997, Butterworth
1999).10 There is much skepticism about the ability of animals to count, after it
10

These authors go into far more detail on the various research programs undertaken to
study animal and human infant perception of numerosity than is warranted here.

16

Numerical Notation

was shown that the mathematical abilities of Clever Hans and other animal calculators were the result of subconscious cues passed from human trainers to these
purported prodigies (Fernald 1984). Yet, following in the footsteps of Koehlers
(1951) work on counting among birds, and the enormous literature studying primate numeracy (Matsuzawa 1985; Boysen and Berntson 1989, 1996), we now know
that many animal species are able to perceive quantity at least accurately enough
to perform tasks involving small quantities, mostly up to three to five units. It is
not yet known whether animal quantification is a homology inherited from an
ancestral species, a specific convergent adaptation in many species to the requirements of similar physical environments, or a general cognitive response of animals
of a certain level of neurological complexity. Many experiments involving many
different species have confirmed that something more than a Clever Hans phenomenon is being observed. The same is true in the case of human infants, who are
able to distinguish small numerical quantities (Gelman and Gallistel 1978, Wynn
1992). While our hominid ancestors did not need numerical notation, the ability
to distinguish between two gazelles and three gazelles would have been cognitively important and evolutionarily adaptive. As the survival of early hominids was
strongly predicated upon the ability to function in groups, the number concept
likely developed relatively early in human prehistory, although direct evidence is
limited. It is highly probable that by the time of the Upper Paleolithic (40,000
to 10,000 years ago), Homo sapiens sapiens possessed languages including two or
more numeral words and the ability to conceptually distinguish cardinal and
ordinal quantities (Marshack 1972; Wiese 2003, 2007).
Any human being (save those suffering from certain types of brain damage or
other serious mental deficiencies) has the capacity to learn how to use numerical notation. As a technology invented in particular historical contexts, however, its use is
limited to those who have encountered it. Anatomically and cognitively modern humans survived for millennia without any need for numerical notation, and the variability among numerical notation systems cannot be explained fully by the universal
human mathematical ability. Even so, this does not prevent us from considering the
possible effects of human cognitive capacities on the types of numerical notation
system that have been developed historically. It is very likely that the evolved capacity of some primates to distinguish five from six bananas is related to the human
visual capacity to distinguish five from six strokes on a tally or knots on a cord. Three
biological characteristics of humans pertain to the development of the concept of
number, which in turn is necessary for the development of numerical notation.
1. Perception of discrete external objects. The ability, common at least to all animals, to
distinguish foreground from background, to perceive the borders of external objects, is
necessary to the creation of the concept of oneness.

Introduction

17

2. Perception and cognition of concrete quantity. The ability to distinguish the quantity
of sets of objects is present in human infants and some animals, but is generally restricted to small quantities.
3. Possession of language. The ability to identify numbers using linguistic symbols, as
opposed to the pre-linguistic quantitative abilities possessed by infants and animals,
permits the conceptualization of number through a series of lexical numerals, each
greater than the previous by one unit.

While numerical notation systems are useful because they enable the human
brain to conceptualize quantities efficiently, we must not assume that their structure and evolution can be derived entirely from the principles of cognitive psychology. Some neuropsychologists examine the development of numerical notation from a cognitive perspective (Dehaene 1997, Butterworth 1999). Dehaene
(1997: 115117) uses a stage-based unilinear scheme to describe the development of
numerical notation from its beginnings in one-to-one correspondence, through
chunked groupings of notches and ciphered numerals, to the ultimate stage of
positional notation with a zero. However, I am very suspicious of such schemes
in the absence of significant historical documentation. The contention that the
history of technology can be understood as a sequence of ever-superior inventions
the better to fit the human mind and improve the usability of numbers is untested at best, ethnocentric at worst (Dehaene 1997: 117).
There are three sociocultural features that are likely prerequisites for the development of numerical notation. These are nonuniversal and derive from contingent historical circumstances, so it is possible to establish whether they are necessary conditions for the development of numerical notation using my universal
cross-cultural methodology.
1. Presence of organizing principles that structure the number line. This refers to the
ability to structure the natural numbers in a manner most convenient to thought,
usually taking the form of a numerical base. No known numerical notation system has
ever been developed by speakers of any of the worlds many languages whose lexical
numerals have no base.
2. Presence of a nonstructured tally-marking system based on one-to-one correspondence. Often claimed as the earliest stage of numerical notation through which
all societies must pass, tallying is a very intuitive way to represent number visually.
There is evidence for this form of representation as early as the Upper Paleolithic
(Marshack 1972).
3. Social need for long-term recording and communication of number. The social need
for a relatively permanent record of numbers is essential to the development of numerical notation. One of its main functions is to assist memory, so the social need to preserve

18

Numerical Notation
numbers beyond the ordinary limits of memory for whatever specific purpose is
probably necessary to its development. Related to this is the need to communicate
number outside a local community. While verbal numbers suffice for local communication, the ability of numerical notation to communicate numbers across barriers
of geography and language is an important feature that would make its development
likely in such circumstances.

Because numerical notation is a human invention, it must be subject to the


constraints imposed by our cognitive abilities. Yet, because it is an invention deriving from specific historical contexts, I study its historical development inductively
before turning to cognitive approaches. A full explanation of the origin of numerical notation must consider both cognitive and sociohistorical factors.
Turning from causes to consequences, I believe that numerical notation has important cognitive effects on its users. These consequences, I suspect, are of a similar
nature to Goodys (1977) suggestions regarding the cognitive consequences of literacy.
Goody himself believes this to be the case, as seen from his observations regarding the
process of counting cowrie shells among the LoDagaa (1977: 1213). The LoDagaa
separate large groups of cowries into smaller groupings of five and twenty cowries to
facilitate the counting of the larger group. While this is not numerical notation, since
it does not represent large numbers using new signs for a base and its powers, it is
an efficient way of counting a large group of objects. Yet, while LoDagaa boys were
expert cowrie counters, they had little ability to multiply, a skill they had only begun
to acquire recently in school. The very existence of multiplication tables, a technique
used by almost all Western children to learn to multiply, links literacy and the use of
numerical notation. While Goody is careful not to overextend this distinction into a
rigid dichotomy, he rightly insists that the formalization of numerical knowledge that
accompanies written numeration is a more abstract way of using numbers.
The comparison of the cognitive abilities of groups who lack numerical notation
and those who possess it would best be done through the ethnographic study of
a group before and after its members learned such a system, or in a group where
some but not all members use numerical notation. To date, no such study exists, although Saxe (1981) has done so for the body counting system used by the Oksapmin
of Papua New Guinea. I discuss only societies that possess numerical notation, and
even then, there is rarely specific contextual information about how the numerals
were used. However, it may be possible to determine whether different types of
numerical notation have different cognitive effects on their users. It is often assumed
that cumulative systems such as the Roman numerals represent concreteness in
numeration because of their iconicity, while positional systems represent abstraction because of their infinite extendability (Hallpike 1986: 121122; Damerow
1996). The existence of cumulative-positional systems is highly problematic for this

Introduction

19

dichotomy. All associations of numerical structure with cognitive ability are untested,
and rely on the equally untested assumption that numerical notation develops from
concreteness to abstraction over time. By examining the diachronic patterns that
actually occurred in the evolution of numerical notation, I will show that these
patterns are multilinear, not unilinear. By comparing the structure of systems to
the functions for which they were used, I will examine the cognitive framework
within which different groups used numerical notation, keeping in mind that it
is only one part of a cluster of techniques that includes mental calculation, lexical
numerals, finger numbering, and computational artifacts.
Rather than assigning labels such as concrete and abstract to numerical notation systems, or identifying any other single factor on which the utility of a system should be judged, I focus on a constellation of features of numerical notation
systems that have cognitive consequences. This approach is similar to that adopted
by Nickerson (1988), who lists the relevant criteria as being ease of interpretation,
ease of writing, ease of learning, extensibility, compactness of notation, and ease
of computation. A set of nonhierarchical criteria for evaluating systems from a
cognitive perspective is a very valuable tool. Nickerson notes usefully:
If one accepts the idea that the Arabic system is in general the best way of representing numbers that has yet been developed, one need not believe that it is clearly
superior with respect to all the design goals that one might establish for an ideal
system. It may be, in fact, that simultaneous realization of all such goals is not possible. (Nickerson 1988: 198)

There is no ideal numerical notation system; rather, each system is shaped by


a set of goals that its users and inventors seek to attain, and that they can achieve
only by compromising on other factors. There may be patterns of change among
systems, but the burden of proof lies with those who wish to maintain that numerical notation evolves in a unilinear sequence.

Numerals and Writing


The scholarly analysis of numerical notation has often been pursued by scholars
interested in writing systems. Therefore, numerical notation systems are usually
regarded as a subcategory of writing systems (Diringer 1949, Harris 1995, Daniels
and Bright 1996, Houston 2004). Most numerical notation systems are associated
with one or more scripts, and conversely, most scripts have some special form
of numerical notation. Numeral-signs are graphemes that undergo paleographic change over time, just as phonographic signs do. The process of recovering
instances of numerical notation archaeologically and interpreting them thus

20

Numerical Notation

inevitably involves epigraphers, paleographers, and other scholars of writing.


However, the uncritical acceptance of a close connection between numerical notation and writing can lead to unfounded assumptions.
There are three basic ways that number is expressed by human beings: a set
of spoken lexical numerals, the written expression of those words in scripts, and
the graphic expression of number through numerical notation systems. We can
divide these three types into auditory systems (verbal lexical numerals) and visual
ones (written lexical numerals and numerical notation). Alternately, we might
distinguish lexical (verbal and written numerals) from nonlexical (numerical notation) means of expressing number. If the similarities between the two visual
representations are more significant than the similarities between the two lexical
representations, then the connection between numerical notation and writing is
strong. However, four differences between lexical and nonlexical representations
of number suggest that this distinction is the more important one.
Firstly, lexical numerals are linguistic, while numerical notation represents number
translinguistically. Numerical notations followed the evolution of language chronologically, and could not have occurred in a nonlinguistic species, but they are not
inherently linked to any language structurally or semantically. The distinction between
writing and not-writing is an issue of great debate among modern scholars, particularly in Mesoamerican (Marcus 1992, Boone 2000) and Andean (Urton 1997, 1998)
studies. The most restrictive approach holds that only phonographic scripts those
whose signs can represent phonemes constitute writing. Accordingly, the Maya
glyph system is a true script, while the Aztec system is a pictographic system that
requires a great deal of context in order to be interpreted, and the Inka khipu notation
is a numerical notation system with some undeciphered non-numerical component.
A broader approach holds that phoneticism is not an essential feature of scripts, and
classifies pictographic representational systems, numerical notation, and even pictorial art under the rubric writing. Gelbs classic definition of writing as a system of
intercommunication by means of conventional visible marks (Gelb 1963: 253) would
suggest that a numerical notation system is a script, although I do not believe that
Gelb meant to imply this. I am sympathetic to the argument that because classifying
societies as illiterate can be used to denigrate them, a broad definition of writing helps
to counteract ethnocentrism, but there is enormous theoretical value in distinguishing phonographic from nonphonographic representation systems.
I do not consider numerical notation to be writing in this narrow sense. Because numerical notation is nonphonetic, it transcends language and can traverse
linguistic boundaries more easily than scripts. It is also learned much more
readily than scripts. To use the terms proposed by Houston (2004b), numerical
notation systems are open and can thus be employed by many groups, as opposed to closed notations that are accessible to one or few linguistic or cultural

Introduction

21

communities. Once an individual learns a numerical notation system, he or she


can communicate numerically with any other individual familiar with the system,
regardless of their linguistic differences. This does not imply that they exist completely outside of culture. As Houston (2004a: 226) and others have noted, quantification systems exist within cultural contexts, which is why, even though we can
identify Inka khipu (Chapter 10) as encoding specific numbers, we know little about
the communicative acts or information systems underlying them. Yet the fact that
we can read Linear A, or Indus Valley, or Inka numerals even though the rest of
those representation systems have eluded decipherment is telling. This suggests that
numbering is separate from writing, more decipherable and less bound to culturally
conventional encoding than other forms of notation.
Secondly, numerical notation systems are not limited to societies possessing
scripts, nor do societies with scripts necessarily possess numerical notation systems. Unfortunately, while scholars such as Ifrah (1998) and Guitel (1975) mention
the existence of tallies, knotted strings, and other such technologies, they are considered solely as peripheral and/or ancestral to numerical notation proper. However, the khipu and several other tallying systems lie within the scope of this study
because they are structured by a numerical base and its powers. One problem with
studying such systems is that they are notched on wood, drawn in sand, or knotted on ropes or strings, all of which are unlikely to survive archaeologically, while
written numerical notation is often found on durable metal, stone, or clay. Far
more numerical notation once existed in nonwritten contexts than has survived.
Moreover, just as numerical notation is not necessarily encountered in conjunction with writing, many scripts have no corresponding numerical notation system.
For instance, the Ogham script of Ireland, the Canaanite script, the early alphabets of Asia Minor such as Carian and Phrygian, and the indigenous scripts of the
Philippines all lack numerical notation and instead express numbers lexically. In
societies that possess both scripts and numerical notation systems, there are often
strong norms prescribing the means of representing number depending on social
context. Throughout the Western world, lexical numerals are preferred in literary
or religious contexts, while numerical notation is preferred in commercial transactions and accounting. In cases where both systems are found in a single text, there
is often a functional division between the two. For instance, the text of the Bible
is written using lexical numerals, but chapters and verses are numbered using numerical notation. In writing checks, numerical notation predominates, but dollar
amounts are written out in full to prevent forgery. Such contrasts suggest that
lexical and nonlexical representations should be treated separately.
Thirdly, numerical notation systems and scripts exhibit very different patterns
of geographical distribution and historical change. In part, this may be because
scripts are largely phonographic, so their diffusion can be constrained by patterns of

22

Numerical Notation

language use. Numerical notation, by contrast, is largely nonphonetic and translinguistic, and may diffuse more readily than scripts. The Western numerals diffused
initially from India and passed through the Arab world before reaching Europe,
while the Roman alphabet is of Greek and Phoenician ancestry. This historical differentiation is not uncommon; the path of diffusion of numerical notation is often
radically different from that of the diffusion of scripts. Yet there may be a connection between the indigenous development of writing and numerical notation. In
several historically unrelated cases (Mesopotamia, Egypt, China, and Mesoamerica),
the independent invention of numerical notation immediately preceded or coincided with the development of a full-fledged script. Perhaps the social need for
numerical notation and a phonetic script tends to arise under similar circumstances
(i.e., during the formative phases of early civilizations). Alternatively, the idea of
numerical notation, once developed, might naturally suggest to its users that other
domains might also be represented visually. I return to this subject in Chapter 12.
Finally, the structures by which written lexical numerals and numerical notation
express number are quite different. The simple fact of being denoted visually is not
as important as the different principles used in the two symbol systems. Lexical
numerals (whether written or verbal) share a common structure that is very different
from that of numerical notation.11 For instance, while the cumulative principle is
commonly employed in numerical notation, it is nearly absent from lexical numeration. No known language expresses thirty as ten ten ten, even though cumulative
numeral-phrases (like the Roman XXX = 30) are quite common. In lexical numeral systems that have a base, multiplicative-additive structuring is overwhelmingly
prevalent, whereas numerical notation systems are only occasionally multiplicativeadditive. To take a familiar example, let us compare Western numerical notation
with English lexical numerals. The English lexical numerals eleven and twelve do not
follow the regular pattern for numbers between thirteen and nineteen, and words
like dozen and score add further complexity. Our numerical notation system is base10 and ciphered-positional, while English lexical numerals use a mixed base of 10
and 1000 (one million = 1000 1000; one billion = 1000 1000 1000), a situation
that becomes even more complex if we include British English, in which one billion
normally means one million millions (1012). Finally, while Western numerals are infinitely extendable one can add zeroes to the right of a number ad infinitum English
lexical numerals are only potentially infinite, since one needs to develop new words
to express higher and higher values. The highest number in many English dictionaries is decillion (1033 in American English, 1060 in British English).
11

The major exception to this disjunction is the classical Chinese numerals, which due to
the somewhat logographic nature of the Chinese script serve both as lexical numerals
and as numerical notation. I will return to this definitional issue in Chapter 8.

Introduction

23

The relationship between the origins of writing systems and numerical notation
systems is similarly complex. Visual number marks clearly precede phonetic writing by many millennia. A wealth of evidence from Upper Paleolithic portable artifacts (e.g., notched bones and stones) suggests that one-to-one marking of numbers for calendrical or other mnemonic purposes has roots extending back at least
30,000 years (Absolon 1957, Marshack 1972, dErrico 1998, dErrico et al. 2003).
This may in turn have been related to the early use of the fingers and hands as a
visual, though nonpermanent, numerical system around the same time (Rouillon
2006).12 Schmandt-Besserat (1992), on the basis of controversial interpretations
of Mesopotamian evidence from the proto-literate period, has been the strongest advocate for an evolutionary sequence from numeration to writing. Houston
(2004: 237) argues that most writing systems emerged as word signs bundled with
systems of numeration that probably had a different and far-more-ancient origin,
and this may be correct. However, numerical notation (as opposed to non-basestructured tallies) does not greatly precede, if at all, the earliest writing. As I shall
show, in all the independent cases of the development of numerical notation, written numerical systems with bases emerge alongside other conventionalized signs,
not as a unilinear predecessor to them.
For these reasons, to analyze numerical notation systems as adjunct components of scripts does not do them justice. Nevertheless, throughout this study I
will sometimes refer to numeral-signs and numeral-phrases as being written.
When I do so, it is mere conventionality, and this usage does not indicate any
specific relationship between numerals and scripts.

Diffusion and Invention


Numerical notation systems develop out of purposeful human efforts to perform
tasks related to the visual representation of number. In a handful of instances,
they developed independently of influence from other numerical notation systems, while in the vast majority of cases systems were borrowed wholesale or with
modification from one society to another. I want to explain the origin, transformation, transmission, and decline of systems, not merely to describe a sequence of
historical events. This does not mean that technical and functional aspects should
be given priority over social factors; rather, social context and historical contingencies must be incorporated into analyses of the histories of systems. It does require,
however, that I distinguish analogies similarities that derive from independent
operation of cause and effect from homologies similarities that derive from the
12

It is absolutely clear that manual counting and numerical notation are connected in later
societies known through written and oral evidence (see Chapter 12).

24

Numerical Notation

descent of cultural features from a common ancestor or the borrowing of features


from one society to another.
In anthropology, analogies and homologies are normally seen as dichotomous,
with materialists (Steward 1955, Harris 1968) preferring analogical explanations
and regarding independent invention as common and idealists (Elliot Smith 1923,
Driver 1966, Rouse 1986) assigning priority to homologies and using borrowing
to explain most cross-cultural similarities. There is no serious scholar who denies
that some features are developed independently multiple times, and similarly no
one doubts that societies borrow extensively from one another. Despite this, the
anthropological effort to distinguish cultural analogies and homologies has not
been especially fruitful (Steward 1955, Kroeber 1948, White 1959, Driver 1966,
Tolstoy 1972, Jorgensen 1979, Maisels 1987, Burton et al. 1996).
Harris (1968) has attempted to circumvent the debate by noting that regardless
of how a trait was exposed to a society, it must still be accepted and integrated into
that society, even if it is borrowed from elsewhere. He thus asserts that diffusion
is a sterile nonprinciple that is not only superfluous, but the very incarnation
of antiscience (1968: 378). Harris is right that simply classifying an innovation
as representing either diffusion or independent invention is insufficient, but he
is wrong in implying that it does not really matter whether a trait was of internal
or external origin. In practice, Harriss rejection of diffusion leads him to assume
that cultural adaptation is a unitary process and that analogical explanations are
the only ones worthy of scientific consideration. Yet if the social consequences of,
and motivations for, adopting diffused numerical notation systems and adopting
independent invention are different as I believe them to be then we must
instead compare the two different circumstances while keeping an open mind as
to potential differences.
Diffusion is often implicitly taken to represent a largely benign transfer of
features from one group to another, followed by a period in which the recipient
society evaluates the innovation, followed by its acceptance or rejection. This extremely nave view of processes of cultural contact denies entirely the role of imperialism, peer-polity networks, and power structures. For instance, many numerical
notation systems developed in societies just as they began to enter into longdistance trading relationships with larger, more politically complex state societies
that already possessed numerical notation. Numerical notation, in this instance,
is not simply something that happens to be transmitted due to cultural contact;
it is a medium through which contact takes place and a feature that becomes important just as societies are becoming integrated into larger intersocietal networks.
Power relations are always involved in such cases, and we need to understand
how the social statuses of individuals and groups within given contexts affect the
transmission process and the eventual outcome. Yet the existence of historical

Introduction

25

contingencies need not be fatal to the development of a cultural-evolutionary


theory of numerical notation. In order to demonstrate empirically the cultural
evolution of numerical notation, we must examine how systems change in a patterned way, comparing analytically the conditions under which numerical notation systems are invented, transmitted, and adopted.
In this study, I address three basic contextual questions regarding each numerical notation system:
1. What antecedent(s) does the system have, if any? I establish whether each system is
descended from antecedent numerical notation systems. Numerical notation was independently invented six or seven times, and these pristine systems stand at the head
of cultural phylogenies, but are certainly not the norm. Independent invention should
not be the null hypothesis for any account of the origins of a system, but neither
should it be restricted only to very ancient systems. Most systems have one antecedent
only, while a few systems blend features of two antecedents.
2. Does the new system supplant one or more older systems? I establish what happens
when a new system is introduced into a society that already uses numerical notation.
Four outcomes are possible: a) the newly introduced system replaces the existing one;
b) the new system is used in conjunction with the older one, normally with some sort
of functional division between the two; c) elements of the original and new systems are
commingled to create a third system; d) the new system is rejected entirely, while the
older system is retained. All these outcomes are attested multiple times.
3. Does the new system use the graphic symbols and/or the structural principles of its
antecedent(s)? I establish how specifically the new system resembles its antecedent(s),
either in the form of its numeral-signs or in its structure (base, interexponential and
intraexponential principle[s], and additional signs). Resemblances among closely related systems, in conjunction with other historical evidence, help to specify the exact
connection between them.

To answer these questions, criteria must be adopted to distinguish endogenously


invented systems from ones introduced from outside a society and to specify connections between ancestral and descendant systems. Discerning historical relations among cultural phenomena can be extremely contentious, particularly when
only archaeological data are available. Rowe (1966) would permit diffusionary
explanations only when abundant evidence of colonies, trading posts, or traded
objects independently confirms contact between two regions, while Tolstoy (1972)
deemed it sufficient to show that a particular combination of features is probabilistically unlikely to have occurred independently. This question is unresolvable in
the abstract, because the ease of demonstrating cultural transmission depends on
the nature of the specific trait or phenomenon being studied.

26

Numerical Notation

Few inventors of numerical notation systems have ever provided detailed information about the contexts of their inventions. Thus, I must build a circumstantial
case for the origins of most systems. In order to demonstrate cultural affiliations
between numerical notation systems, I use both internal (structural and graphic)
resemblances between systems and external (contextual and historical) considerations. The main criteria I use are as follows:
1. Use of the two systems at the same point in time. This criterion is nearly
unavoidable; some chronological overlap in the periods during which two systems
are used is needed to sustain a hypothesis of cultural transmission. An extinct system might conceivably be revived and modified by a later society (for instance, on
the basis of old inscriptions), but this is hardly a sufficient basis for a hypothesis of
cultural transmission. Alternately, a system that is not attested to have survived may
in fact have done so; this is the basis of the controversial theory that the Mycenean
Linear B numerals (Chapter 2) gave rise to the Etruscan numerals (Chapter 4).
Such hypotheses cannot be dismissed immediately, if other factors suggest that
they could be true, but they require much more evidence.
2. Similarity in structural features. Because there are only three intraexponential principles (cumulative, ciphered, multiplicative), two interexponential principles (additive, positional), three common bases (10, 20, 60), and two sub-bases
(5, 10), no one aspect that is similar in two systems is sufficient to prove a connection. However, when two systems are alike in all or most of these respects, cultural
contact becomes a much more likely explanation for the resemblance. Many of
the cultural phylogenies of systems that I discuss share a common structure; for
instance, all the Italic systems (Chapter 4) are cumulative-additive with a base of
10 and a sub-base of 5. This does not mean that all identically structured systems
must be placed in that phylogeny the Ryukyu sho-chu-ma numerals (Chapter
10) and modern Berber numerals (Chapter 10) do not fit because they were used
much later and have different numeral-signs. The use of structural features as evidence of contact suffers from the weakness that, if many systems in a phylogeny
are identical or similar, it is often impossible to choose between several equally
likely candidate ancestors.
3. Similarity of forms and values of numeral-signs. Because many graphic
symbols are very complex, they are unlikely to have developed independently.
If the forms of numeral-signs used in two systems are identical or very similar,
and if those signs represent the same numerical values in the two systems, it is
likely that cultural contact resulted in the invention of the later system based on
the earlier one. The more signs that are shared between two systems, the more
likely it is that there is a historical connection between them. However, when two
systems use similar signs for different numerical values, this is not good evidence
of such a connection. For instance, and J represent 10 and 20 in the Kharoh

Introduction

27

numerals (Chapter 3) but mean 7 and 9 in the Brhm numerals (Chapter 6).
In this instance, even though the two systems were used in the same region at
the same time (the Indian subcontinent in the fourth century bc) and have two
similar numeral-signs, the dissimilarity of their values reduces the likelihood of
a historical connection. Caution must be exercised when invoking this criterion
for very simple symbols vertical and horizontal lines, dots, circles, crosses, and
the like because such designs are cross-culturally common. This is especially
true in the case of the use of lines and dots with the value of one, since these
signs may have been part of tallying systems before being used in numerical
notation systems. Cases where signs are similar but not identical must also be
treated with caution. There is no general paleographic principle for identifying
relations among graphically similar signs; hence, such efforts usually proceed on
an intuitive basis.
4. Known cultural contact between the regions where the two systems are
used. In general, where one cultural trait is transmitted from one region to
another, multiple traits are likely to have been transmitted. Thus, where there
is a known pattern of shared non-numerical features in two societies, or where
there is substantial evidence of interregional trade, migration, or colonization,
such evidence supports a postulated ancestor-descendant relationship between
two numerical notation systems. Determining whether known cultural contact is sufficient to postulate the diffusion of a numerical notation system is
always a tricky matter and involves an evaluation of various lines of evidence.
For instance, one of the difficulties in postulating that the Brhm numerals
(Chapter 6) are descended from the Egyptian demotic ones is that, despite
structural and graphic resemblances between the two systems, Egypt is well
down on the list of areas with which ancient India had contact. In no case do
I postulate a connection between two systems solely on the basis that they
were used at approximately the same time and in a single region. There must
always be some structural or graphic resemblance between postulated ancestor
and descendant systems.
This problem is made more complex by stimulus diffusion, a complex blend of
inventive and diffusionary processes in which awareness of an invention is transmitted, but, because of some obstacle to transmission or acceptance, the actual
invention does not take hold in the adopting society (Kroeber 1948: 368370).
However, because the general principle is seen as useful, some members of the
adopting society, stimulated by the original idea, invent their own version of the
innovation. The most widely cited example of stimulus diffusion is the development of the Cherokee syllabic writing system by Sequoyah in the nineteenth century, based on his rudimentary knowledge of the Western alphabet. While several
numerical notation systems resulted from stimulus diffusion (e.g., the abortive

28

Numerical Notation

Cherokee numerals, never used in the syllabary), no principles exist to help identify stimulus diffusion. It is tempting to postulate stimulus diffusion even when
the basic fact of incomplete transmission cannot be established. However, I use
stimulus diffusion as an explanation only when it can be established that the form
of cultural contact that occurred between two regions fits Kroebers model.
5. Use of ancestor and descendant systems in similar contexts. If two systems
serve similar purposes, on similar media, or among similar social groups in their
respective societies, this can serve as further confirmatory evidence that the two
systems are related historically. This factor, while useful, is never sufficient on its
own to demonstrate such a connection, but it may provide further support. For
instance, the spread of the Greek alphabetic numerals into Armenia and Georgia
(Chapter 5), though poorly documented, is confirmed not only by the striking
similarities in the systems but also by the systems use in Bibles and other liturgical texts. The similarities among some of the cuneiform systems of Mesopotamia
(Chapter 7) rest on their common use of a wedge-shaped stylus on clay media as
much as on specific resemblances in the numeral-signs or their organization.
6. Geographic proximity of the regions where two systems were used. All
other factors being equal, a system is more likely to have been modeled on one
that is used by neighboring groups than on one used more distantly. This is a
particularly dangerous criterion to invoke, especially where there is less cultural
contact between neighboring regions than with more distant regions. Many times,
two very different and unrelated systems are used in proximity to one another, and
other times, closely related systems are used at considerable distances from one
another. Geographical proximity is such a weak measure that I will use it only as a
last resort, and never as the sole factor for hypothesizing transmission.
Establishing links between ancestor and descendant systems, within the limits
of the available data, allows me to describe phylogenies of related systems. These
are, however, analytical descriptions, which allow the explanation of evolutionary
patterns of change in numerical notation systems. These explanations are analogical, because they describe independent recurrences of cause and effect. However,
they are also explaining homological processes resulting from cultural contact and
the transmission of knowledge among many societies. This is a paradox only if we
accept the notion that these two concepts stand in opposition to one another. A
phylogenetic perspective is both homological and analogical, seeking to describe
particular historical contexts, but also to derive general processes by which numerical notation systems are related to one another. Diffusion may be, as Harris
contends, a nonprinciple, but it is not a nonprocess. Comparing ancestor and descendant systems, and understanding the nature of the process of borrowing and
adoption of cultural features, is absolutely essential to an evolutionary perspective
on cultural change.

Introduction

29

Technology, Function, and Efficiency


Modern historians of science such as Thomas Kuhn (1962) have effectively demolished the myth of linear progressivism in science. While in some fields, the accumulation of knowledge leads to a better understanding of reality, and technical
innovations likewise have antecedents, the burden of proof has now rightly shifted
to those who wish to demonstrate that progress occurs. Progressivist schemes that
assume rather than demonstrate the superiority of new technologies, imply that
where such progress exists it implies moral superiority, or argue teleologically that
present achievements can never be exceeded, have no scientific credibility. Yet
these preconceptions abound among scholars of numerical notation; this shift in
our conception of progress has not yet taken hold. Consider the following collection of recent laudatory statements regarding Western numerals:
Sa perfection va bien au-del de la civilisation indienne puisquaucune autre numration de Type III na jamais t en mesure de lgaler. (Guitel 1975: 758)
If the evolution of written numeration converges, it is mainly because place-value
coding is the best available notation. So many of its characteristics can be praised:
its compactness, the few symbols it requires, the ease with which it can be learned,
the speed with which it can be read or written, the simplicity of the calculation algorithms it supports. All justify its universal adoption. Indeed, it is hard to see what
new invention could ever improve on it. (Dehaene 1997: 101)
Our positional number-system is perfect and complete, because it is as economical
in symbols as can be and can represent any number, however large. Also, as we have
seen, it is the most efficacious in that it allows everyone to do arithmetic. . . . In short,
the invention of our current number-system is the final stage in the development
of numerical notation: once it was achieved, no further discoveries remained to be
made in this domain. (Ifrah 1998: 592)

Such perspectives accept without proof that the Western numerals are the most
efficient ever developed, and are not only the best in existence but also perfect the
best that could ever be conceived. Their adoption by the vast majority of human societies today is perceived as a natural and inevitable consequence of this
superiority, only minimally mediated by social factors. Other, more cumbersome
systems are to be evaluated in relation to the Western system, and in particular to
their utility for arithmetical calculations and higher mathematics. Since so many
modern technologies require mathematics, Western numerical notation is a partial cause of these evolutionary developments. The corollary of this proposition,
often left unstated, is that those societies that did not develop or adopt Western
numerals failed to compete politically with the West in part because of this.

30

Numerical Notation

I consider the decimal, ciphered-positional system of numerical notation developed in India in the sixth century ad and transmitted by Arab scholars to Western
Europe to be a very remarkable invention. Its brevity, unambiguity, and ease of
learning make it conducive to the practice of written arithmetic and mathematics. How well numerical notation systems represent number strongly affects the
development of new systems, their acceptance after being transmitted, their modification over time, and their eventual abandonment. This pattern of long-term
sociocultural change can meaningfully be called evolutionary.
The primary difficulty with the assumption of the evolutionary progress of numerical notation is not the notion of evolution itself. The problem is that the efficiency of numerical notation systems cannot be evaluated in the abstract, but only
by considering the purposes for which they were developed and used. It is often
assumed that the function of numerical notation is to perform written computations. For instance, Ifrah, whose work is the most popular and influential study of
the history of numerical notation, writes:
To see why place-value systems are superior to all others, we can begin by considering
the Greek alphabetic numeration. It has very short notations for the commonly used
numbers: no more than four signs are needed for any number below 10,000. But that
is not the main criterion for judging a written numeration. What matters most is the
ease with which it lends itself to arithmetical operations. (Ifrah 1985: 431)

This view is entirely erroneous. Numerical notation was a necessary condition


for the development of modern mathematics, but it is ethnocentric to argue from
this that its purpose was to facilitate the development of mathematics. The efficiency of any technology can be evaluated only in terms of the purposes for
which it was developed and/or used. There is thus no eternal abstract standard of
efficiency for any technology. It smacks of teleology to argue that Western numerical notation is wonderful because it enabled modern mathematics to develop. The
origin of Western numerals had little to do with mathematical computation and
much to do with writing dates on ancient and medieval southern Asian inscriptions. The primary function of numerical notation is always the simple visual
representation of numbers. Most numerical notation systems were never used for
arithmetic or mathematics, but only for representation. Even when they are used
in mathematical contexts, they frequently simply record the results of computations performed in the head, on the fingers, or with an abacus. Even in industrialized societies, computation remains a secondary function of numerical notation. I
am looking at a not-so-crisp Canadian five-dollar bill, on which numerals indicate
a monetary value (5), the date the bill was designed (1986), a serial number with
some letters to render it unique (GPA6537377), and the number 64 penciled in

Introduction

31

one corner (probably to record the number of five-dollar bills received at some
event). None of these numeral-phrases was actually ever used to compute.13 Numbers denote far more often than they reckon, even in our highly numerical society.
This was doubly true in pre-industrial contexts.
In defining a numerical notation system as a system for representing numbers, I
am explicitly making a functional statement. At minimum, whatever else numerical notation may mean in a particular society, it must always express number as
one of its functions. I am not saying that a numerical notation system must be
fully integrated with other sociocultural phenomena, that it must be perfectly
adapted to serve social needs, or that the purpose for which it is used must be that
for which it was developed. The representational function of numerical notation
is general enough that it can be stimulated by a variety of social or political needs.
While trade is the most obvious one making transactions possible over long distances, enabling monetary calculations, or recording results to facilitate accurate
bookkeeping it is not the only one. For instance, the main impetus behind the
origin of the Mesoamerican numerical notation systems was probably astronomical and calendrical, while the Shang numerals were first used in the context of
Chinese divination. We should not expect to find a single specific domain of activity correlated with the origin of numerical notation, and we should be skeptical of
universal or unilinear schemes.
If we wish to compare the efficiency of various numerical notation systems, we
must compare systems that served a common purpose in terms of how well they
served that purpose. Because all systems represent number visually, some general
criteria can be used. A system that represents numbers using few number-signs is
more efficient than one that requires many signs. One could then argue that the
Roman numerals are not as efficient for representing number as Western numerals
are because 1492 is much shorter than MCCCCLXXXXII (or MCDXCII). While
the situation is slightly more complex MMI is shorter than 2001, for instance the
Roman numerals are more concise for only a small fraction of all natural numbers.
Two other criteria that are relatively easily definable are a systems sign-count (how
many signs it uses in total) and extendability (the highest number expressible).14
I return to these criteria in Chapter 11 and show how they can be used to ask

13

14

One might protest that the numeral on the bill is used in doing arithmetical computations such as providing change for purchases. To refute this, one need only go into a bank
and ask for $100 in five-dollar bills, and see whether the teller looks at the number on
each bill, or whether in fact he or she merely counts out twenty bills while doing mental
arithmetic. The numeral on the bill denotes its value, but is not used in calculation.
See Nickerson (1988: 189197) for a different, but related, list of criteria used in comparing numerical notation systems.

32

Numerical Notation

fruitful questions that help explain synchronic and diachronic patterns among
attested systems.
By contrast, efficiency for computation is a Western-centered and historically inaccurate benchmark for comparing numerical notation systems. Zhang
and Normans (1995) paper on the visual representation of numbers through numerical notation is a major step forward in our understanding of how numerical
notation systems work. They analyze how specific systems visually represent (or
fail to represent) numerical information, describe three general means by which
numerical notation systems are structured (shape, quantity, and position), and
then examine how these features are combined in numerical notation systems. Yet
their analysis falls apart because they compare and evaluate different numerical
notation systems based on their ability to aid in multiplication. Even if Western
numerals are the best system for doing arithmetic (which would best be resolved
through the use of the systems rather than abstract theorization), most other systems were never designed or used for such a purpose. The situation is analogous
to denigrating screwdrivers for being inefficient hammers. The fact that one can
use a screwdriver handle to drive in nails does not justify that comparison, just as
the fact that one might use Egyptian hieroglyphic numerals to multiply does not
justify comparing them to systems such as the Western numerals. To add insult
to injury, even though Zhang and Norman recognize that calculation technologies such as the abacus are frequently better than numerical notation for doing
arithmetic, they suggest that part of Western numerals superiority is that they are
used for both calculation and representation, while other societies employed two
separate systems (Zhang and Norman 1995: 293). They thus blame the carpenter
for using both a hammer and a screwdriver where just the screwdriver would do.
Such arguments are little more than elaborate rationalizations for a historical fact
(the near-universality of Western numerals) that eludes simple explanation.
The only way to compare numerical notation systems fairly is to use functional
criteria that apply to all systems (namely, those related to simple representation).
Yet, even where there are definite answers to these efficiency-related questions,
this does not mean that individuals testing out a new system will immediately
perceive its advantages and disadvantages. A familiar but in some respects inefficient system, so long as it is not entirely unworkable, may be retained, despite
its obvious inferiority. There may be a steep learning curve preventing the easy
adoption of the alternative system, or there may be cultural or political reasons
for retaining ones present system. Moreover, numerical notation, as a system for
communicating information to others, requires not only that specific individuals
adopt it, as would be the case with a more efficient plough or a better mousetrap,
but also that an entire social group learn it before its usefulness will be evident.
A system with many users is functional for that reason alone, because it can be

Introduction

33

used to communicate with more people than one with few users. Thus, it is inappropriate to evaluate numerical notation systems only in terms of their structural
features. Rather, these features must be considered in the broader social context in
which systems develop and are used. I examine these social factors and show how,
far from negating structural factors, structural and social explanations of regularities combine to produce a more complete understanding of numerical notation
than has previously been possible.
I turn in the following chapters to the body of data itself. I endeavor to highlight the ways in which the general theoretical principles just discussed relate to
the data. I have organized these data according to cultural phylogenies of related
systems, presenting the earliest systems first, leading forward to systems developed more recently. The first five phylogenies are probably related to one another
historically, so I treat them together, but no other principle has been used in the
ordering of chapters. The eight major phylogenies, each of which merits a full
chapter, are as follows:
Chapter 2: Hieroglyphic systems historically descended from the Egyptian hieroglyphic
numerals;
Chapter 3: Levantine systems used in the Levant, descended from the Aramaic and
Phoenician numerals;
Chapter 4: Italic systems used in the circum-Mediterranean region, descended from
the Etruscan numerals;
Chapter 5: Alphabetic systems whose signs are mainly phonetic script-signs, descended
from the Greek alphabetic numerals;
Chapter 6: South Asian systems originating on the Indian subcontinent and descended
from the Brhm numerals;
Chapter 7: Mesopotamian systems used in Mesopotamia, descended from the protocuneiform numerals;
Chapter 8: East Asian systems descended from the Shang numerals;
Chapter 9: Mesoamerican systems descended from the Mesoamerican bar-and-dot
numerals.

Chapter 10 is devoted to miscellaneous systems and cultural isolates that do not


fit into any of these phylogenies, and also to the numerous systems invented in
colonial contexts over the past two hundred years. Chapter 11 analyzes synchronic
and diachronic regularities among numerical notation systems in a structural and
cognitive framework, while Chapter 12 tempers these findings with considerations
relating to social context.

chapter 2

Hieroglyphic Systems

A recognizable phylogeny of numerical notation systems was used in conjunction


with a group of related scripts and their descendants, beginning with the Egyptian
hieroglyphic numerals as early as 3250 bc, which thus rivals the Mesopotamian
family (Chapter 7) as the oldest attested numerical notation anywhere. Among
these, I include the Egyptian hieroglyphic system, obviously, but also the Hittite hieroglyphic, Cretan hieroglyphic, Minoan Linear A, Mycenean Linear B,
and Cypriote numerals. In addition, I include the Egyptian hieratic and demotic
systems, which are cursive reductions of the Egyptian hieroglyphic numerals, even
though they are structurally closer to the alphabetic systems (Chapter 5), to which
they are ancestral. I use the term hieroglyphic simply because several systems
discussed in this chapter are associated with hieroglyphic scripts, rather than to
imply anything about the systems structure. The hieroglyphic systems are summarized in Table 2.1.1
Of these systems, the Egyptian hieroglyphic has been discussed most extensively, though it has often been misinterpreted, while others, such as the Cypriote
system, are severely understudied and poorly known. Identifying these systems
and distinguishing them from other, superficially similar ones helps explain the
diffusion of numerical notation throughout the ancient Mediterranean region.
1

The hieratic and demotic systems are too complex to be included on this chart; consult
their individual entries for their numeral-signs.

34

Hieroglyphic Systems

35

Table 2.1. Hieroglyphic numerical notation systems


System

10

100

1000

10,000 100,000 1,000,000

Egyptian hieroglyphic

q
\=

s
0

u\

Cretan hieroglyphic
Minoan Linear A
Mycenean Linear B
Cypriote syllabic
Hittite hieroglyphic

The hieroglyphic phylogeny of numerical notation systems is ancestral to the


Levantine (Chapter 3), Italic (Chapter 4), and Alphabetic (Chapter 5) phylogenies,
but its systems differ sharply from those of its descendants.

Egyptian Hieroglyphic
The hieroglyphic script is the best-known ancient Egyptian script. It was used
between about 3250 bc and 400 ad, making it the longest surviving of all scripts
(Loprieno 1995). However, its use was restricted geographically to the Nile Valley
and nearby areas under Egyptian control. While the hieroglyphic script may well
have arisen because of stimulus diffusion and trade with Mesopotamia, the scripts
in these two areas emerged essentially simultaneously and show no substantial
resemblances. Hieroglyphic inscriptions are written from top to bottom, left to
right, or right to left, with the last of these three options being the most common (Ritner 1996: 80). The script is mixed in principle, with both phonograms
(consisting of one, two, or three consonants) and logograms indicating words
nonphonetically (Ritner 1996: 74). The later hieratic and demotic scripts used to
write the ancient Egyptian language, as well as the Meroitic hieroglyphic script,
are directly derived from the hieroglyphic, while the early scripts of the Levant and
the Aegean are probably its less direct descendants.
Numbers other than one are very rarely expressed through lexical numerals in
Egyptian, making it difficult to determine their structure, although evidence from
some Old Kingdom Pyramid Texts and later Coptic writings establishes that they
had a purely decimal structure with words for each power of 10 up to 1,000,000
(Loprieno 1995: 71). Most hieroglyphic inscriptions express numbers using numeralsigns rather than words, however, with separate signs corresponding to each power.
These signs are shown in Table 2.2 (cf. Gardiner 1927: 191; Allen 2000: 97).

Numerical Notation

36

Table 2.2. Egyptian hieroglyphic numerals


1

10

100

1000

10,000

100,000

1,000,000

R-L

q
q

r
r

v\

Lex.

fn

L-R

68,257 =

qqqq rrr \\\\\\\


qqq\ \\rr\\\\\\\ \\\\\\\\\\\

The system is purely decimal and cumulative-additive, with each sign repeated
up to nine times as necessary, and ordered from highest to lowest rank. The direction in which a numeral is read is always the same as the direction of writing, but
varies depending on the inscription in question. The set of signs in the top row of
Table 2.2 are those used when the direction of writing is from left to right; when
right-to-left writing is used, the signs are mirrored (i.e., q\r\\\\). Occasionally,
when days of the month are being expressed, the signs for 1 and 10 were placed on
their side: or ^ instead of r or q (Gardiner 1927: 191). Numeral-signs could
be used either cardinally or ordinally, with ordinals from second through ninth
adding the ending nw (masculine) or nwt (feminine) to the numeral-phrase, and
those from tenth upward adding m (masculine) or m t (feminine). To aid in
reading long numeral-phrases, five or more identical signs were usually grouped in
sets of three or four rather than placed on a single line. Thus, 5 is written as a row
of three signs above a row of two signs, 6 as a row of three above a row of three, 7
as a row of four above a row of three, 8 as a row of four above a row of four, and 9
either as a row of five above a row of four or as three rows of three.2
The sign for 1 is a simple vertical stroke. Gunn (1916: 280) suggests that in early
well-executed inscriptions, the sides of the vertical bar are curved inward slightly,
thus making a biconcave bar, and postulates that it may represent a small object
of bone or wood used in some kind of tally or aid to reckoning, but I tend to
think that it is simply an abstract stroke. The sign for 10 has been described as a
heel bone (Kavett and Kavett 1975: 390), a tie made by bending a leaf (McLeish
1991: 42), or even, anachronistically, as a croquet wicket (Boyer 1959: 127). In fact,
it corresponds to the phonetic value m (masc. m w, fem. m t) hook, handle,
2

However, other groupings were sometimes used when it was more convenient for the
scribe to do so.

Hieroglyphic Systems

37

and is a rebus for the Egyptian lexical numeral for 10 (m ) (Sethe 1916: 2). The
higher power signs also have specific representational qualities and can also represent phonetic values in Egyptian apart from their use as numerical symbols. The
sign for 100 ( t) is probably a coiled length of rope; that for 1000 ( ) is a lotus
plant; the sign for 10,000 ( b ) is an extended finger; and that for 100,000 ( fn)
is a tadpole. These numeral-signs, as well as the overall structure of the system,
remained remarkably stable throughout its history. In some older instances in
which the sign for 1000 occurs, rather than grouping the signs in clusters of three
to five separated signs (as in the numeral-phrase mentioned earlier), multiple lotus plants were depicted as emerging from a single bush (e.g., 3000 = ). The sign
for one million ( ) could also mean multitude or a countless quantity, just as
the Greek word myriad can mean a group of ten thousand or, more generally, a
large quantity (Loprieno 1995). After the Early Dynastic period, this nonspecific
lexical sense predominated over the specific numerical value. In most other respects, Predynastic hieroglyphic numerals would have been completely intelligible
to Late Period scribes.
The earliest known Egyptian hieroglyphic numerals are those from Tomb
U-j at Abydos, which dates to around 3250-3200 bc (late Naqada II or early
Naqada III period), and also contains the earliest examples of Egyptian writing (Dreyer 1998). Numeral-signs occur on many drilled bone and ivory tags
found in this royal tomb, which were probably once attached to containers of
goods. Other tags have other signs that resemble later Egyptian hieroglyphs,
but only a very few contain both numerals and hieroglyphs (Baines 2004:
154157). Some tags have six to twelve vertical or horizontal strokes, others the
sign for 100, and one has both a sign for 100 and a sign for 1 (Dreyer 1998:
113118). This system has three distinctive features as compared to the mature
hieroglyphic system: it uses both horizontal and vertical strokes for units;
there is no attested numeral-sign for 10; and there are tags with more than
nine unit-strokes. Dreyer (1998: 140) explains the first two of these differences
simultaneously by noting that on some Old Kingdom linen-lists, horizontal
strokes stand for 10. The Tomb U-j tags are very similar to others found at
Naqada and Abydos that date from the Naqada III and Early Dynastic periods, which contain the sign for 10 and use only vertical strokes for 1 (Dreyer
1998: 139). The very early date of the tags suggests that the system was developed independently of Mesopotamian influence, although the U-j tags are
essentially contemporaneous with the Uruk IV tablets. The margin of error
and discrepancies in the different radiocarbon dates from Tomb U-j are large
enough that no definite conclusion regarding priority can be reached (Baines
2004: 154). Even though the U-j tags are apparently administrative or commercial, the context of the discovery (a royal tomb) suggests instead that the

38

Numerical Notation

Figure 2.1. A scene from The Narmer mace-head, a late Predynastic ceremonial artifact
bearing early but recognizable Egyptian hieroglyphic numerals. At the far right, a quantity
of 120,000 prisoners is indicated, while the lower register indicates quantities of 1,422,000
goats and 400,000 cattle. Source: Quibell 1900: Plate XXVI B.

signs were part of a nascent visual high culture of interest to elites seeking to
legitimate their authority (Baines 2004: 170171).
While we have no evidence for numeral-signs higher than 100 from the Tomb
U-j tags, by the Early Dynastic period the system was fully developed. Figure 2.1
depicts the Narmer mace-head found at Hierakonpolis, which may describe the
unification of Upper and Lower Egypt by around 3100 bc, and which demonstrates
that even the very highest signs were being used at that time (Arnett 1982: 42).
The mace-head indicates an exaggerated tally of 400,000 oxen, 1,422,000 goats,
and 120,000 humans (Quibell 1900: 89, Pl. XXVI). This is traditionally interpreted as an exaggerated and propagandistic tally of booty and prisoners acquired
through Narmers military victories. Millet (1990), however, provides an alternate
interpretation of the mace-head inscription as a year-identifier and suggests that
the numbers are purely artificial, meant only to signify the taking of a census.
Another early example of hieroglyphic numerals is found on the Second Dynasty
statue of Khasekhem indicating the slaughter of 47,209 of the pharaohs enemies
(Guitel 1958: 692). These large numerical values figure prominently in early hieroglyphic inscriptions, further supporting the idea that early Egyptian monumental
writing was primarily oriented toward display purposes relating to the ideological
justification of the kings authority (cf. Baines 2004). Other numerical inscriptions from the Early Dynastic include tags for commodities from Naqada like the

Hieroglyphic Systems

39

earlier ones found at Abydos, but using ordinarily structured hieroglyphic numerals
(Imhausen 2007: 14).
In the Old Kingdom (25752134 bc), variants on the basic hieroglyphs were not
uncommon. Clagett (1989: 5657) discusses a variant of the hieroglyphic numerals
used on the Palermo Stone (a Fifth Dynasty / 2400 bc pharaonic annal), in which
certain notations of the aroura measure of land are represented with unit-strokes
in a quasi-positional manner. However, this system is not used regularly throughout the Palermo Stone and is not found in any other inscriptions; hence its value
for understanding the hieroglyphic numerals is somewhat limited. Another unusual Old Kingdom system has been proposed by Posener-Kriger (1977) to have
been used in papyrus documents from Gebelein indicating area measures of fabric on so-called linen-lists (mentioned earlier). In this system, a single cubit sign
meant one square cubit, horizontal strokes ten, and long vertical strokes meant
one hundred each of which were followed by short vertical strokes indicating
how many of the requisite units were denoted. Both of these variant systems were
employed solely for enumerating particular types of goods, and were never used
more generally for denoting numbers.
In the Ptolemaic era (33230 bc), the hieroglyphic numerals became more
complex. The sign for 1,000,000 was reintroduced into the numerical sequence,
though it is unclear whether its numerical meaning was truly understood. In a few
inscriptions from this period, a ring sign is found in the sequence between
v and w. While Sethe (1916) believed that the ring sign was a meaningless addition, Gunn (1916: 280) protested that perhaps, in order to lengthen the series of
numerals without assigning the god w a subordinate place, was assigned the
value of 1,000,000, while w either shifted upward in value to ten million or else
retained its lexical meaning of an uncountable number. Curiously, on the stela
of Ptolemy Philadelphos (r. 282246 bc) at Pithom, the sign used for 100,000 is
not v but rather , with the ring sign placed underneath the ordinary tadpole sign
(Sethe 1916: 9).
Another curious change in the late hieroglyphic numerals is the occasional
use of cryptographic ciphered numeral-signs for many numbers, as shown in
Table 2.3. These signs replaced the standard cumulative sets of signs with single
signs whose association with the number was homophonic, pictorial, religious, or
related to the corresponding hieratic numeral-sign. They were used as early as 950 bc
on a wooden votive cubit rod of Sheshonk I, but are found on no artifacts between that point and the Ptolemaic era (Priskin 2003). The most common of these
signs is that for 5, a five-pointed star, which often combines with unit-strokes in
the same way as V = 5 in Roman numerals (Sethe 1916: 25). However, unlike the
Roman numerals and related systems, no signs were developed for 50, 500, or
other half-powers. The origin of this sign is almost certainly pictorial, from the five

Numerical Notation

40

Table 2.3. Ptolemaic-era cryptographic hieroglyphic numerals


5
7
9
60
80

Q
U
Z
A
B

points of the star. Other common signs were a human head for 7, from the Egyptian understanding of the head as having seven orifices, and a scythe for 9, from
the resemblance between that sign and the hieratic numeral-sign for 9 (Sethe 1916: 25).
In addition to the signs for the units 1 through 9, there were cryptographic hieroglyphs for 60 and 80, both of which were derived from resemblances to hieratic
numerals (Fairman 1963). These new signs never led to a fully ciphered-additive
set of hieroglyphic numeral-signs, and were often included in otherwise perfectly
ordinary cumulative numeral-phrases.
Hieroglyphic numerals are largely written on monumental inscriptions, but not
exclusively so. Texts including hieroglyphic numerals include seals, funerary stelae
and tomb inscriptions, annals, lists relating to conquest and plundered goods, and
certain administrative texts. An often-overlooked source of hieroglyphic numerals
is the wide variety of stone balance-weights bearing inscriptions indicative of their
weight (Petrie 1926, Petruso 1981). Numerals indicated dates, weights and measures, and a wide variety of quantities of goods, animals, and people. In all of these
texts, the numerals are formed in the ordinary fashion just described.
The hieroglyphic numerals were rarely if ever used for mathematics and calculation. The vast majority of Egyptian literary texts, and all Egyptian mathematical
texts, are written in the hieratic or later demotic scripts (cf. Gillings 1978: 704705).
Nor are there hieroglyphic numerals marked on potsherds, tallies, or other such
media that would suggest their use as an intermediate step in performing calculations. Some hieroglyphic numerals are used in an inscription from the tomb of
Methen (Fourth Dynasty, twenty-sixth century bc), which indicates the calculation
of the area of a rectangle, but this inscription indicates only that the calculation was
done; the numerals were not actually used in the calculation process (Peet 1923: 9).
Yet, because Egyptologists regularly transliterate documents in the hieratic
script into regularized hieroglyphs, historians of mathematics have sometimes

Hieroglyphic Systems

41

inferred wrongly that the Egyptians calculated using hieroglyphic numerals. The
hieroglyphic system is cumulative-additive, while the hieratic system is cipheredadditive, but since this difference in structure was underemphasized by Egyptologists of earlier generations (e.g., Gardiner 1927: 191; Peet 1931: 411), historians of mathematics frequently presume that the hieroglyphic numerals were the
only ones available to Egyptian scribes (cf. McLeish 1991: 42; Guedj 1996: 3435;
Palter 1996: 228229; Dehaene 1997: 97). It is to be hoped that the presence of
new Egyptological literature may remedy this deficiency (Ritter 2002, Imhausen
2007). It is necessary to treat the hieroglyphic and hieratic systems separately, not
despite their very strong historical connection, but because of that connection,
inasmuch as the two systems were different in structure and used in entirely different functional contexts.
A very few hieroglyphic inscriptions express large numbers (particularly multiples of 100,000) through multiplicative formations instead of purely additive
ones. In one Ptolemaic-era text, the number 27,000,000 is expressed by placing
a single sign for 100,000 above the ordinary additive hieroglyphic phrase for 270
(Brugsch 1968 [1883]: III, 604).3 The only other way to write 27,000,000 would
have been to use 270 signs for 100,000 or 27 signs for 1,000,000, neither of which
is an attractive option. Such phrases would be unappealing from the perspective
of Egyptian aesthetic canons, in addition to the clear economy of symbols enjoyed
through multiplicative phrases. In a second instance (from the time of Amenhotep
III, around 1400 bc), 100,000 is expressed multiplicatively using the tadpole-sign
v placed above a vertical stroke 100,000 1 = 100,000 (Sethe 1916: 9; Loprieno,
personal communication), thus, curiously, requiring more signs than the standard
numeral-phrase. Finally, multiplicative hieroglyphs are found on a number of votive cubits from the New Kingdom and later, cubit-long polygonal stone objects
inscribed with metrological and religious information, with clear multiplicative
phrases for multiples of 100,000 and possibly also for 100 and 1000 (Ritter 2002:
308309). Yet there is no evidence that this multiplicative-additive structure was
widespread. The ciphered-additive hieratic numerals use multiplicative forms far
more frequently, and earlier, than do the cumulative-additive hieroglyphs (Sethe
1916: 810; Mller 1936: I, 59). Yet because Egyptian grammars mention the hieratic examples (e.g. Gardiner 1927: 191; Allen 2000: 97) but transcribe the numerals
as hieroglyphic numerals, it is easy to conclude that multiplicative expressions are
common in the hieroglyphic numerals, when in fact almost all such expressions
come from hieratic texts.
3

While Brugsch interprets this figure as 100,270, the figure being represented is the
amount (in arouras) of land in Egypt, for which 27,000,000 is the only reasonable interpretation (cf. Kraus 2004: 225).

42

Numerical Notation

The question of when this borrowing took place remains open; the first hieratic
documents to use this structure date to the Middle Kingdom (2040 to 1652 bc),
while the first hieroglyphic example (mentioned earlier) dates to about 1400 bc.
Because the hieratic multiplicative numeral-phrases are more common and earlier
than the hieroglyphic, I think it likely that the ancestral system (hieroglyphic)
borrowed the feature from its descendant (hieratic). Because hieroglyphic numerals were used only for monumental purposes at that time, numbers higher than
100,000 would have been expressed only infrequently. It is entirely possible, given
the small number of hieroglyphic inscriptions using multiplication, that it was an
exceptional response to the occasional requirement for expressing high numbers
in hieroglyphic numerals. There is no evidence supporting Guitels assertion that
this occasional use of multiplication, which is paralleled in certain Aztec texts
(Chapter 9), represents even an abortive step toward a fully positional notation
(Guitel 1958: 692695; 1975: 44). Rather, it represents an alternative means of
increasing the conciseness of some (but not all) numeral-phrases and extending a
systems capacity to write numbers while retaining its basic structure.
The Egyptian hieroglyphic script possessed two distinct systems for representing
fractional values, both of which normally expressed only unit-fractions those in
which the numerator is 1. The first such system, the standard system for expressing
fractional quantities, simply required the scribe to place the sign r, which could
also mean part or mouth, above any hieroglyphic numeral-phrase to indicate
the corresponding unit-fraction (Loprieno 1986: 1307). If the mouth sign was too
small to place over the entire phrase, it was simply placed over the signs for the
highest power of 10. This system also used special symbols for some of the most
commonly used fractions: 1/2 (gs), 1/4 (r-4), 2/3 (rwj), and 3/4 (hmt-rw) (Sethe
1916: Table II; Allen 2000: 101). The last two of these are not unit-fractions, and
are thus exceptions to the general rule. This system was not used in the Predynastic
era, but is found in abundance during the Old Kingdom and thereafter. While,
in theory, this system could express any fraction, most have denominators smaller
than 20.
The second system was used primarily for measurements of volume of grain,
fruit, and liquids by indicating fractions of the heqat ( t), a measure probably
equal to 292.24 cubic inches, or roughly 4.8 liters (Chace et al. 1929: 31). This notation is sometimes known as Horus-eye fractions because the six hieroglyphic
symbols for fractional values can be combined to form the glyph of the w t or
eye of Horus (#), a symbol of health, fertility, and abundance. The sum of these
signs is only 63/64; symbolically, the remaining 1/64 would be supplied magically by the god Thoth when he healed the Eye of Horus, thus producing unity
(Gardiner 1927: 197). Ritter (2002) makes a persuasive case, however, that these
signs were originally nonpictographic, hieratic submultiples of metrological units

Hieroglyphic Systems

43

Table 2.4. Egyptian hieroglyphic fractional ideograms


1/2

1/4

2/3

3/4

Horus-eye fractions
1/2

1/4

1/8

1/16

1/32

1/64

& *

rather than pure numerals, and that they were not originally associated with the
Eye of Horus at all, and thus he renames them capacity system submultiples.
Despite the early caution of Peet (1923), the Horus-eye interpretation has misleadingly become standard among historians of mathematics. The hieroglyphic
forms of these signs are shown in Table 2.4, presuming a left-right direction of
writing (Sethe 1916: Table II; Gardiner 1927: 197).
This systems binary structure was probably most useful for dividing and multiplying by two, a standard operation needed when manipulating volumes of goods.
The system probably originated in an earlier hieratic series of fractional signs, of
which the earliest example is from the Abusir Papyri of the Fifth Dynasty, and
only later did the signs assimilate to resemble the parts of the Horus-eye symbol
(Reineke 1992: 204). Other than one possible sign for 1/2 from the Fifth Dynasty,
the earliest hieroglyphic Horus-eye fractions are from the Nineteenth Dynasty or
later (Priskin 2002: 76; Ritter 2002: 304).
The Egyptian hieroglyphic numerical notation system has several direct descendants, the most direct of which is the ciphered-additive Egyptian hieratic
system (to be discussed later), which developed as early as the First Dynasty as
a scribal shorthand for the hieroglyphs (Peet 1923: 11). Egyptian scribes would
have learned both the hieroglyphic and hieratic numerals during their education,
and used both systems in the appropriate contexts the hieroglyphs on stone
monuments, and the hieratic numerals written in ink on papyrus and ostraca
(inscribed potsherds). It is also very likely that the civilizations of the Aegean
used the Egyptian hieroglyphic numerals as the model for their own indigenous
numerals the Cretan hieroglyphic system, the Linear A and B numerals, and the
Hittite-Luwian hieroglyphic numerals. There was considerable commercial and
political interaction between Egypt and the Aegean in the second millennium
bc, when the Aegean numerical notation systems began to emerge (Cline 1994).
Despite the dissimilarity in the numeral-signs of the two systems, they are identically structured, and thus a hypothesis of diffusion is likely correct. A less direct
descendant of the Egyptian hieroglyphs is the Phoenician-Aramaic system, which

44

Numerical Notation

was developed around 750 bc, blending the numeral-signs and script tradition of
the Egyptian hieroglyphs with the structure of the Assyro-Babylonian cuneiform
numerals (Chapter 7). This development marks the formation of the Levantine
systems (Chapter 3), reflecting the intermediary position of the Levantine civilizations between the larger polities of the eastern Mediterranean.
By the Greco-Roman period, the use of the hieroglyphic script and numerals
had declined greatly, and both writing and numerals had increased in the number
of signs used and the complexity thereof, to the point where it was considered to
be a purely symbolic or cryptographic script by outsiders (Ritner 1996: 81). By the
third century ad, Egypt was becoming increasingly Christian in its religion, and
its language was being written in the Greek and Coptic scripts. The latest dated
hieroglyphic inscription is on the temple of Isis at Philae, and dates from August
24, ad 394 (Griffith 1937: I, 126127). By the fifth century, knowledge of how to
read and write hieroglyphs had disappeared. The hieroglyphic numerals, as well
as their immediate descendants, were replaced by the Coptic alphabetic numerals
(Chapter 5).

Egyptian Hieratic
The hieratic script was developed around 2600 bc by Egyptian scribes as a sort
of cursive shorthand for the earlier hieroglyphic script (Loprieno 1995) and, like
its forerunner, used a mixture of logographic and phonographic components.
However, unlike the hieroglyphs, hieratic writing was designed for cursive writing
on papyrus and on ostraca, making it suitable for administrative and literary
purposes. Furthermore, while the hieroglyphs could be written in a variety of
directions, hieratic texts are always linear and written from right to left. While the
form of the hieroglyphs was very regular and formalized, hieratic writing varied
greatly by period, location, and the idiosyncrasies of the scribes handwriting. The
Old Kingdom divergence of Egyptian scripts into monumental (hieroglyphic)
and cursive (hieratic and demotic) variants continued throughout the remainder
of ancient Egyptian history.
A base-10 ciphered-additive numerical notation system accompanied the hieratic script. The hieratic numeral-signs, like the script itself, changed considerably over the systems extensive history. The paleographic development of hieratic
numerals is traced in the charts provided by Mller (1936). In Tables 2.5, 2.6, and
2.7, I present three distinct sets of numerals, the first and earliest from the Kahun
papyrus (Twelfth Dynasty / 20001800 bc), the second from Pap. Louvre 3226
(fifteenth century bc), and the third from the Harris papyri (twelfth century bc)
(Mller 1936, vol 1: 5963, vol. 2: 5559). These three texts contain mostly complete sets of numeral-signs at least as high as 1000, and are thus very useful for

Hieroglyphic Systems

45

Table 2.5. Hieratic numerals (Kahun papyri, Twelfth Dynasty)


1

a
b
10s
j k
100s
s t
1000s
1 2
10,000s
: ::
100,000s >
4367 = gou4
1s

c
l
u
3
:::

d
m
v
4
=

e
n
w
5

f g h
o p q
x y z
6
8

i
r
0
9

Table 2.6. Hieratic numerals (Pap. Louvre 3226, Eighteenth Dynasty)


1

A B C D E F G H I
10s
J K L M N O P Q R
100s
S T U V W X Y Z [
1000s
\ ]
657 = GNX
1s

Table 2.7. Hieratic numerals (Pap. Harris, Twentieth Dynasty)


1

a b
10s
j k
100s
s t
1000s
| }
10,000s
8 9
100,000s ,
56,207 = gt4;
1s

c
l
u
1
0

d
m
v
2
:

e
n
w
3
;

f
o
x
4
<

g
p
y
5
=

h
q
z
6
>

i
r
{
7
?

46

Numerical Notation

comparative purposes. The Harris papyri numerals, from Table 2.7, include all of
the numbers up to 100,000; this is the only text to do so.
Looking only at the signs for 5, 6, 7, and 9, the three series appear remarkably
distinct. At the same time, however, most of the hieratic numeral-signs show remarkable continuity. Many of the hieratic signs used in the Old Kingdom would
have been perfectly comprehensible to a scribe in the Late Period or even the Ptolemaic era. Many of the numeral-signs are very similar to others from the same period; for instance, it is very difficult to distinguish 400 from 600 or 3000 from 5000 in
Table 2.7. When used to express days of the month, hieratic numerals, like hieroglyphic numerals, were often rotated ninety degrees counterclockwise to reflect
their function. Given the nature of the Egyptian calendar, these forms exist only for
numerals less than 30. To write fractional values, a small dot was placed above the
numeral-phrase for an integer to indicate the appropriate unit fraction (1/x).
The hieratic system is primarily ciphered-additive, and its signs each represent a
multiple of a power of 10. Many of the hieratic numeral-signs bear a clear relationship to their cumulative-additive hieroglyphic forerunners, seen particularly in
the signs for 1 through 4, 10, 10,000 through 40,000, and 100,000. Other hieratic
numerals show no clear correspondence with their hieroglyphic ancestors except
in very early periods. The ciphered-additive hieratic system thus shows traces of its
cumulative-additive ancestry. For this reason, I include the hieratic system in this
chapter even though it is structurally different from its hieroglyphic ancestor.
In some hieratic texts, irregular numerical systems were used in conjunction
with grain measures (Allen 2000: 102). One early Middle Kingdom system notated
sacks of grain using regular numerals, and heqats (1/10 sack) using one to nine
dots in a cumulative fashion. Later in the Middle Kingdom, a notation developed
whereby ordinary numerals placed before a heqat unit indicated multiples of 100,
those after the numeral multiples of 10, and then one to nine dots for the units.
These systems were not used outside of this metrological context. As discussed
earlier, hieratic fractions were frequently written in unit-fraction form or through
capacity system submultiples (Ritter 2002).
For writing many numbers above 10,000, multiplicative notation was used in
the hieratic numerals; for instance, the sign for 60,000 is written by placing the
sign for 6 below the sign for 10,000. This principle is not used for 10,000 through
30,000, but was used occasionally for 40,000, and normally for 50,000 through
90,000 and for values above 100,000. While the multiplicative principle is seemingly used for certain values of the hundreds and the thousands, paleographic
analysis of the numeral-signs shows that the sign for 300 represents the abbreviation of the first two of three cumulative 100-signs and the extension of the third
rather than the juxtaposition of 3 and 100. Imhausen (2006: 2526) discusses an

Hieroglyphic Systems

47

Table 2.8. Evolution of cursive from linear Egyptian numerals

300

Hieroglyphic

Old Kingdom

Kahun papyrus
(Dyn. 12)

P. Louvre
3226 (Dyn. 18)

P. Harris
(Dyn. 20)

qqq
qqq
qqq
qqq
qqq

ostracon from the New Kingdom workers village of Deir el Medina, a scribal
exercise in which 600,000, 700,000, 800,000, and five, six, and seven million are
expressed multiplicatively using the hieratic signs for 100,000 and one million,
respectively.
The regular use of multiplicative-additive structuring allowed very high numbers
above 100,000 to be expressed easily in hieratic numerals by placing the appropriate
multiplier below the tadpole sign. The earliest multiplicative hieratic numerals are
from the twentieth century bc Kahun papyrus fragments, which are arithmetical
problems and accounting documents, suggesting that this technique originated in
the context of mathematical or arithmetical practice (Griffith 1898: 16). The development of hieratic numerals was thus a highly creative process involving both the shift
to ciphered notation and the use of multiplication where it was deemed useful.
The strong similarities between the hieratic numerals and the earlier hieroglyphic numerals, coupled with the indisputable historical connections between
the two scripts, prove the historical relationship of the two systems. Whereas the
hieroglyphic numerals are found in Predynastic inscriptions, hieratic numerals
first appear in the First Dynasty (Peet 1923: 11). Their use became widespread from
the Old Kingdom onward, with the two systems (hieroglyphic and hieratic) being used for parallel purposes. The earliest hieratic numerals were little more than
cumulative-additive cursive forms of the appropriate hieroglyphic numerals. Over
time, the numerals became increasingly removed from their hieroglyphic ancestors as multiple strokes were condensed into single strokes, probably for greater
ease of writing. By the Fifth Dynasty, the numerals written in the Abusir papyri
(archives of royal funerary cults) had acquired a strongly cursive character that had
moved away from the original cumulative signs (Goedicke 1988: xvixvii). Table 2.8

48

Numerical Notation

compares how the numbers 6, 9, and 300 were written in Old Kingdom hieratic
to the numeral-signs from the three sets of numerals presented earlier.
While ciphered signs were the ordinary ones, the systems origins were not completely forgotten; cumulative hieratic numeral-signs were occasionally employed
even into the New Kingdom. No single individual invented ciphered notation
in Egypt; rather, its development was a process of abbreviating and combining
cumulative signs by scribes over many centuries until, by the Late period, very few
hieratic signs bore any resemblance to their hieroglyphic counterparts. It is even
possible that the scribes making these changes were not really aware of the importance of the new structural principle they were using. Hence the origin of ciphered
notation may, in some sense, have been accidental. Strikingly, in some hieratic
documents from the Ptolemaic era, there is a reversion in the numeral-signs away
from the ciphered signs used in older hieratic texts and back to the common use of
the cumulative principle. In several texts (Leinwand, P. Bremner, Isis-N., Leiden
J. 32, and P. Rhind),4 hieratic units were expressed with repeated vertical strokes,
tens with horseshoe-shaped curves, and hundreds with coils, in an exact imitation of the hieroglyphic numeral-phrases of the same value (Mller 1936: vol. III,
5960). While some of these documents retained the ciphered signs for some values, there is a trend over time toward the use of cumulative numeral-signs in these
late hieratic documents. Some scribes may have forgotten the ciphered signs; more
likely, however, the reversion to cumulative-additive numerals was a deliberate
archaism, resulting from the desire to emulate hieroglyphs more exactly.
Egyptian scribes would have learned both hieroglyphic and hieratic writing
and numerals during their education, and used whichever was appropriate according to the context. Accordingly, while the functions of the hieratic numerals
are quite distinct from those of the hieroglyphic numerals, the users of the two
systems would have been the same individuals. For the hieratic numerals, two
functions stand out above all others: administration and mathematics. Almost all
extant Egyptian legal, commercial, educational, and literary texts from 2600 to
600 bc are written in hieratic, and numerals abound on such documents. While
hieroglyphic numeral-phrases were very lengthy, requiring an enormous number
of symbols to express many small values, hieratic numerals were highly concise,
facilitating their use in accounting, commerce, and law, as well as for expressing
dates and cardinal quantities. Because they would have been learned and used by
only a small and well-educated segment of the populace (i.e., the scribes), their
main disadvantage the large number of signs one needed to learn in order to use
the system would not have been a serious problem.
4

P. Rhind does not refer here to the famous Rhind Mathematical Papyrus, but to a different text dating to 9 bc and having nothing to do with mathematics.

Hieroglyphic Systems

49

A limited but interesting set of hieratic texts directly concern mathematical subjects. The hieratic numerals were first used in Egypt for arithmetic and
mathematics in the late Middle Kingdom and the early Second Intermediate
Period (Twelfth and Thirteenth Dynasties). The well-known Reisner, Berlin,
Kahun, and Moscow mathematical papyri all date from the nineteenth century bc
(Gillings 1978: 704705). Later, around 1650 bc, during the period of Hyksos domination, the Egyptian Mathematical Leather Roll and the famed Rhind Mathematical Papyrus were written using hieratic numerals, though the latter may be a copy
of an earlier document. Egyptian mathematics was never perceived as a separate
field of activity, but was thoroughly enmeshed and embodied within daily scribal
practice, so to search for pure mathematics divorced from administrative activities is futile and ethnocentric (Imhausen 2003: 386). While the most thoroughly
mathematical texts from ancient Egypt date from roughly 1900 to 1650 bc, these
texts are but a minuscule fraction of the total number of hieratic texts containing numerals, and cannot represent the full scope of mathematical practice over
four millennia of Egyptian history. A full discussion of the mathematics of ancient
Egypt is well beyond the scope of this work (cf. Peet 1923; Neugebauer 1957; van der
Waerden 1963; Gillings 1972, 1978; Rossi 2004; Imhausen 2003, 2006, 2007).
Tracing the diffusion of the hieratic numerals is quite difficult. As the system was
primarily used for administrative purposes, it spread wherever Egyptian domination
extended for instance, into Canaan in the Nineteenth and Twentieth Dynasties
(Millard 1995: 189190). The early Israelites used a minor variant of hieratic numerals
(to be described later) starting in the tenth century bc. In addition, the hieratic numerals gave rise to two distinct descendant systems. By the eighth century bc, the
hieratic of Upper Egypt (abnormal hieratic) was no longer mutually legible with
that of Lower Egypt, which is now known as demotic, and which eventually
replaced its ancestor. In addition, the Meroitic cursive script, found on ostraca in the
Sudan starting in the third century bc, contains numeral-signs to which Griffith
(1916: 23) assigns ancestry from the hieratic numerals. While the hieratic numerals
have relatively few direct descendants, through their demotic descendant they are
ancestral to a great number of systems.
In the Twenty-sixth Dynasty (664 to 525 bc), the demotic script and numerals,
which had only begun to diverge from hieratic a century or so earlier, were accorded royal preference for most purposes. After that point, demotic began to replace
hieratic for more and more functions throughout Egypt. By the early Christian
era, when hieratic was encountered by the Greeks, it was used only in religious
texts by which means it got its name, hieratikos sacred. The name that we
now give to this script and numerical notation system is, ironically, taken from a
purpose for which it was rarely used throughout over two millennia of its history.
By around 200 ad, even these religious functions had ceased.

50

Numerical Notation

Hebrew Hieratic
In the ninth century bc, the Egyptian scribal tradition, including the use of the
hieratic script and numerals, was adopted by the ancient Israelites, who incorporated a great deal of Egyptian learning into their own thought (Millard 1995,
Rollston 2006). Prior to this point, there is no evidence that the Israelites used
any numerical notation whatsoever, although some may have become familiar
with Egyptian notations while in Egypt. From around the ninth5 through the
sixth centuries, Hebrew scribes combined their own (Paleo-Hebrew) script with a
variant form of hieratic numerals, which I will call Hebrew hieratic for simplicity, although they differ only paleographically from ordinary Egyptian hieratic
numerals (Kletter 1998: 142).
The earliest Hebrew inscriptions containing hieratic numerals are the Samaria
ostraca from the late ninth or early eighth century bc. The most notable and
complete example of Hebrew hieratic numerals is a large (30 22 cm) ostracon,
KBar6, excavated from Tell el-Qudeirat (Kadesh-barnea) in 1979, depicted in Figure 2.2, with the numerals transcribed in Table 2.9. The ostracon contains a very
complete series of hieratic numerals; only the signs for the units 1 through 9 and
60 were missing, blurred, or unreadable (Lemaire and Vernus 1980; Cohen 1981:
105107). It was probably a scribal exercise or practice text in writing numerals and
measures (Dobbs-Allsopp et al. 2005: 251260). The number 10,000 is expressed
by combining the Egyptian hieratic sign for 10 with the lexical numeral thousand
in Paleo-Hebrew script, and there are also Hebrew units of measurement (heqat,
shekel, gerah) on the ostracon, so it cannot have been written by an Egyptian
scribe. The numeral-signs are paleographically very similar and structurally identical to the late hieratic ones, indicating that these numerals were directly borrowed
under conditions of economic domination by and cultural contact with Egypt.
Ostraca found at Arad and Lachish and mostly dating to the late seventh and
early sixth centuries record weights and measures in letters and accounts, and
served an administrative function (Levine 2004: 433).6 While Gandz (1933: 61)
argued that the numerals from the Samaria ostraca had an Aramaic origin, the numeral-signs are in fact hieratic in origin (Lemaire 1977: 281). The Samaria ostraca
signs for 5, 10, and 20 are , , and , respectively, which closely resemble the
late hieratic e, j, and k, but not Aramaic aaa\aa, A, and B.

Sheas claim (1978) that this system is attested on a Late Bronze Age jar handle from
the thirteenth century bc has not been addressed by Semitic epigraphers, but both the
paleographic interpretation and the date are questionable.
Some of the Arad ostraca record quantities in a West Semitic (Aramaic-Phoenician)
notation rather than in hieratic numerals (see Chapter 3).

Hieroglyphic Systems

51

Figure 2.2. The KBar6 ostracon from Kadesh-barnea, most likely the work of a student practicing writing numerals in Hebrew hieratic numerals. Source: Cohen 1981: 106.
Reprinted with the permission of the American Schools of Oriental Research.

The Hebrew hieratic numerals were also used on Judaean inscribed limestone
weights, a metrological system that reflected the growing commercial importance
of Judah within the context of Near Eastern commerce between the late eighth
and early sixth centuries bc. Kletters (1998) study of 434 of these Judaean weights
represents the most thorough examination of these artifacts to date. Although it
was once argued that these notations were cumulative-additive and indigenous in
origin (Allrik 1954, Yadin 1961), apart from the paleographic variability one would
expect when transferring a cursive notation onto stone, they are clearly hieratic,
Table 2.9. Hebrew hieratic numerals from KBar6 ostracon from Kadesh-barnea
1
10s
100s
1000s

52

Numerical Notation

ciphered-additive numerals like those on the ostraca (Aharoni 1966). The numerals on these weights range from 1 to 50 to indicate multiples of the shekel (approx.
11.3 grams), but curiously, other than the 1- and 2-shekel weights, the numerals are
in a 5:4 ratio to the expected masses of the weights. This puzzle has been the source
of considerable recent debate among metrologists and remains incompletely resolved, but in any case the numerical interpretations of the signs are paleographically secure (Ronen 1996, Kletter 1998).
The ostraca and shekel weights are of a relatively early date in the history of
Hebrew writing, and must be understood in the context of growing administrative needs in the Iron Age states of the Levant. This period marks the first in which
literacy was relatively widespread in Judah (Kletter 1998: 144). Rollston (2006)
rightly sees the presence of hieratic numerals and artifacts such as KBar6 as positive (if not conclusive) evidence for the introduction of formal schooling in late
Iron Age Israel. After Judah lost its independence in 586 bc, the system appears to
have become obsolescent. The Hebrew hieratic numerals are not directly ancestral
to the cumulative-additive Levantine systems that emerged in the eighth century
bc, among Phoenicians and Aramaeans. These systems were used contemporaneously with the hieratic numerals and, for instance, occur in the same contexts on
ostraca from Arad. There is no evidence of an indigenous Hebrew numerical notation system until about 125 bc, when the use of the familiar alphabetic numerals
(Chapter 5) began.

Meroitic
The kingdom of Mero, which flourished from roughly 300 bc to 350 ad, made
use of two distinct scripts. The first, Meroitic hieroglyphs, were based on Egyptian
hieroglyphs and were used on some stone monuments. The attested Meroitic hieroglyphic inscriptions contain no numerals. The other script, the Meroitic cursive, was
written from right to left on ostraca as well as on stone, and was accompanied by a
set of numerals. Almost all of our information on the Meroitic numerals rests on the
work of F. Ll. Griffith, the original decipherer of the Meroitic scripts. Unfortunately,
because the Meroitic language has no known relatives, we are largely unable to read
Meroitic inscriptions, even though the values for the signs of the cursive script are
more or less fully deciphered. Griffith (1916: 22) first presented the interpretation
of the Meroitic numeral-signs shown in Table 2.10. On structural and paleographic
grounds, the values for the units, 10, and all of the hundreds are unquestionable, and
the remainder of the numeral-signs are fairly certain.
This system is ciphered-additive and decimal, and written from right to left
(like the Meroitic cursive script). The number 2348 as shown in Table 2.10 appears
on the stela of Akinidad, which dates to the late first century bc (Griffith 1916: 22).

Hieroglyphic Systems

53

Table 2.10. Meroitic numerals


1

a
b
10s
j k
100s
s t
1000s A
B
2348 = hmuB

1s

c d e f g
l m n
p
u v w x
C

As with the hieratic numerals, the signs for the units, low hundreds, and possibly
1000 through 3000 are somewhat cumulative. There is only one case (again, from
the Akinidad stela) where a number greater than 10,000 is expressed; interestingly, where hieratic uses a single sign for 10,000 (8), Meroitic appears to use a
multiplicative formation (10 1000). However, this evidence is far too limited to
conclude that the Meroites regularly used multiplicative-additive structuring to
express higher powers. In addition, cumulative sets of one to nine dots apparently
indicated tenths from 1/10 to 9/10, while a dot in a semicircle (0) represented
1/20 (Griffith 1916, 1925). Griffith (1916: 2223) believed this system to be purely
metrological, representing tenths and twentieths of some larger unit of measure
rather than abstract numbers.
By the time of the development of the Meroitic scripts, the hieratic script and
numerical notation system had largely been replaced by demotic throughout
Egypt. Nevertheless, on paleographic grounds (citing especially the signs for 6,
10, and 20, but also the cumulative unit-signs), Griffith (1916: 23) argued that
the Meroitic numerals resemble the late hieratic numerals (eighth to third centuries bc) more closely than the demotic forms, even though the characters of the
Meroitic cursive script are almost certainly derived from a demotic rather than a
hieratic prototype (Millet 1996: 85). More paleographic analysis is desirable to
settle this question.
The Meroitic numerals were used for administrative purposes such as tax records
and mensuration, as well as in funerary and monumental contexts indicating yeardates and quantities of individuals. Griffith suggests that something akin to the
Egyptian heqat or artaba measures, used to indicate volumes of produce such
as corn or dates, was probably indicated on some ostraca (1916: 23). There is no
evidence that the Meroitic numerals were ever used for arithmetic or mathematics. Even on ostraca upon which multiple numerals have been written, Griffith
was unable, except in one instance, to establish any arithmetical correspondence

54

Numerical Notation

between the numerals that would indicate that a tally or sum had been taken
(1916: 24).
The Meroitic numerals were used until the fourth century ad, but did not
outlast the kingdom of Mero. Millet (1996: 84) suggests that the script may have
continued in use until the introduction of Coptic Christianity in the sixth century, but there is no textual evidence to establish whether the Meroitic numerical
notation system existed during this late period. The Coptic and Ethiopic numerals, both of which are derived from the Greek alphabetic numerals, were used
widely in the region from the sixth century onward.

Egyptian Demotic
The demotic script developed in the late eighth century bc (Twenty-fifth Dynasty) and began to replace the hieratic script about a century later. It was a cursive script consisting largely of consonantal characters, derived from the business
hand hieratic used in the Nile Delta (Ritner 1996: 82). During the Late period
and the Ptolemaic era, demotic writing was used very widely for administrative
and literary purposes, and more sporadically throughout the Roman period. A
set of ciphered-additive, base-10 numerals accompanied this script throughout
its history. As with the hieratic numerals, there is a great deal of variation in the
demotic numeral-signs; the ones presented in Table 2.11 (after Sethe 1916: Table I)
are typical of those found in papyri of the Late and Ptolemaic periods. Griffith
(1909: 415417) provides an interesting paleographic comparison of the demotic
numeral-signs found on a selection of papyri dating from the Twenty-sixth
dynasty to the Roman period.
The demotic numerals are a base-10, ciphered-additive system, written from right
to left. They are less reliant on the cumulative principle than their hieratic ancestor
(compare hieratic c and demotic C for 3). Some of the signs for the thousands may
be vaguely multiplicative, as there is a general resemblance between the signs for the
hundreds and the corresponding signs for the thousands, but it is more likely that
they are simply further reductions of the nonmultiplicative hieratic signs. Sethe
(1916: Table I) suggests that additive phrases incorporating two lower signs (3000 +
2000, 4000 + 3000) were used for the missing 5000 and 7000 signs. Above 10,000,
the demotic numerals, like the hieratic ones, are multiplicative (though such large
expressions are fairly rare); for instance, Parker has found multiplicative expressions
for 90,000 (7, = 9 10,000) and 100,000 (8, = 10 10,000) in his study of demotic
mathematical papyri (Parker 1972: 86). As in the hieratic numerals, a small dot
placed above a numeral-sign indicated the corresponding (1/x) unit fraction.
The demotic numerals are directly derived from the hieratic forms used in the
eighth century bc; as the hieratic numerals were used as late as 200 ad, the two

Hieroglyphic Systems

55

Table 2.11. Demotic numerals


1

A B C
10s
J K L
100s
S T U
1000s @
# $
6268 = HOT&

1s

D E
M N
V W
%

F G H I
O p Q R
X Y Z !
&
* (

systems were used side by side in Egypt for nearly a millennium. This long coexistence can be explained in part by regional variations, with Upper Egypt retaining
the abnormal hieratic numerals and Lower Egypt using demotic. Unlike the corresponding writing systems, the hieratic and demotic numerals would have been
largely mutually intelligible until the Ptolemaic period at least, which may have
facilitated communication between different parts of Egypt. The demotic script
and numerals were accorded royal preference in the Twenty-sixth Dynasty, and
thus they were used for most royal functions thereafter, while the hieratic system
was retained primarily for calligraphic religious texts (Ritner 1996: 8182).
Unlike the hieratic script and numerals, which were rarely written on stone
except at the very end of their history, demotic inscriptions are found on stone as
well as ceramics and papyrus. Like their predecessor, demotic numerals were used
for a wide variety of commercial, legal, and other administrative functions, as well
as for indicating dates. A number of demotic mathematical papyri have survived
from the Ptolemaic period, confirming the suitability of the system for arithmetical and mathematical purposes (Parker 1972, Gillings 1978). However, as with the
hieratic numerals, most demotic texts that contain numerals serve no mathematical function. Much of our paleographical knowledge of the demotic numerals
comes from administrative texts, such as dowry records and educational papyri
(Griffith 1909). An extensive set of demotic numerals is found in P. Tsenhor, the
private archive of a sixth-century woman (Pestman 1994).
The importance of the demotic numerical notation system lies not in any structural feature or unusual function, but rather in its historical role as the immediate
ancestor of several other numerical notation systems. The demotic numerals are
almost certainly ancestral to the Greek alphabetic numerals (Chapter 5). These
numerals, which are structurally identical to the demotic numerals, first appear
in the sixth century bc in Ionia and Caria, at which time Greek trade with Egypt
was beginning in earnest, and when the Ionian trading city of Naukratis in the

56

Numerical Notation

Nile Delta was the major center for trade between Egypt and Greece (Chrisomalis
2003). Furthermore, the alphabetic numerals became common in the late fourth
century bc, at which time Egypt came under Ptolemaic control. Remarkably, the
similarities between the demotic and Greek alphabetic numerals have been substantially ignored over the past century, with most scholars inclined to treat the
latter system as a case of independent invention (but cf. Boyer 1944: 159). Secondly, there are strong similarities between the demotic numerals and the Brhm
numerals (Chapter 6), which began to be used in India around 300 bc. In this
case, the historical connection between the two regions is not as clear, but the
structural similarities between the two systems suggest some connection. While
trade between Egypt and India became common only in the Roman period, there
are strong indications of overseas trade dating from the Ptolemaic period and perhaps even somewhat earlier. Again, few historians of mathematics have proposed
this connection, although it has held some popularity among Indologists for over
a century (Bhler 1896, Salomon 1998).
By the Roman period (30 bcad 364), the demotic numerals were used increasingly rarely, as the general decline of Egyptian cultural institutions continued apace.
However, even though Roman imperialism was the immediate circumstance surrounding the decline of the demotic numerals, they were not replaced with Roman
numerals, but rather with the Coptic numerals, which were themselves descended
from the demotic through the Greek alphabetic numerals. As Christianity began to
take hold in Egypt, and the Coptic script and numerals became more widespread,
demotic suffered a fatal decline. The last text with demotic numerals is a graffito
on the temple of Isis at Philae (the same temple that contains the last evidence for
hieroglyphs), which dates to December 2, ad 452 (Griffith 1937: I, 102103). The last
known demotic text of any sort (dated in Greek) was written nine days later.

Linear A (Minoan)
The Linear A script was the standard script used in the Minoan civilization of
Crete between 1800 and 1450 bc (Bennett 1996: 132). It is perhaps the most famous
of all undeciphered scripts, having foiled decades of effort to interpret it. Only the
numerals and a few other ideograms for commodities can be deciphered. Linear A
is written from left to right and is almost certainly a mixture of syllabograms and
logograms. Its well-attested numeral-signs are shown in Table 2.12 (Sarton 1936b:
378; Ventris and Chadwick 1973: 36).
The Linear A numerical notation system is decimal and cumulative-additive,
and is written from left to right with the powers in descending order. Where appropriate, signs are grouped in two rows of up to five signs each rather than placing them in an uninterrupted row. The variant dot symbol for 10 is found only in

Hieroglyphic Systems

57

Table 2.12. Linear A numerals


1

10

100

7659 =

1000

qqqqq
qqqq

early Linear A documents and is probably related to the identical numeral-sign


for 10 in the contemporaneous Cretan hieroglyphs (see the following discussion).
Other than this, however, the system remained unchanged throughout its history. While Evans (1935: 693) suggested that there may have been a sign R or
that stood for zero, this was later shown to be a sort of check-mark or sign for
completion of an item, or perhaps served some other bookkeeping function (Bennett 1950: 205). Using statistical methods, Daniel Was (1971) has postulated the
existence of a complex base-24 system for representing fractions in Linear A. If
correctly deciphered, this system is likely to have been metrological in function.
While Struik (1982: 56) suggests that this system is related to the Egyptian unitfractions, no real resemblances exist between the two fractional systems.
The Linear A script and numerals were probably borrowed in some manner
from the identically structured Egyptian hieroglyphic system (cf. Sarton 1936b:
378). Trade between Egypt and Crete was extensive in the Middle Minoan II period
(ca. 18001700 bc), when Linear A developed (Cline 1994). Admittedly, there is
no real similarity between the numeral-signs of the two scripts, except in the use
of vertical strokes for the units, which is common to almost all systems used in
the Mediterranean region. Whereas Egyptian hieroglyphic numerals are pictorial
representations with phonetic values mostly originating as homonyms of lexical
numerals, Linear A numerals are abstract and simplified. However, we would not
expect the Minoans to adopt the Egyptian signs, because the signs would have
no such phonetic associations for them. I am unconvinced by the isolationist
position with regard to Minoan literacy (e.g., Dow and Chadwick 1971: 35). The
link suggested between Linear A and the Proto-Elamite numerals (Chapter 7) of
fourth millennium bc Iran, however, requires implausibly great chronological and
geographical gaps (Brice 1963). Egypt is the only plausible ancestral region for the
Minoan numerals.
The abstract and geometric character of the numeral-signs makes it impossible, however, to exclude an independent origin for the system. Branigan (1969)
speculates that concentric circles on sealings from Phaistos may have represented
tens, hundreds, and thousands, and may be a geometric precursor to the Linear A

58

Numerical Notation

numerals. A similar system of small circles and large circles inscribed on cylindrical
stone weights from the palace at Knossos may have indicated one and ten units of
some metrological value (Evans 1906). While either of these systems could be related
to Linear A, at present the hypothesis of borrowing from Egypt best explains the
structure of Linear A numerals, with the numeral-signs developed indigenously.
Numeral-signs are the only known means of representing numbers in Linear
A; although it remains possible that lexical numerals were written using syllabic
signs, the closely related (and deciphered) Linear B script does not do so, suggesting that this is unlikely. The vast majority of Linear A documents are clay tablets
having an accounting or bookkeeping function, and thus we have many examples
of the use of numerals. Vertical strokes that probably represented numbers have
been found in other contexts for example, on Minoan balance weights; these
marks, however, do not show any clear relation to the Linear A signs found on the
clay tablets and are probably simply unstructured unit-marks or tallies (Petruso
1978). What are likely Linear A numerals occur on a number of pieces of pottery
from Bronze Age Cyprus (Grace 1940). Stieglitz proposes that a numerical graffito
found at Hagia Triada and containing the sequence of numbers (1, 1 1/2, 2 1/4,
3 3/8), in which each number is 1.5 times the previous one, represents a series
of musical notes or tunings for a stringed instrument (Stieglitz 1978). I think it
equally likely that the series served an economic function such as calculating interest. Since we do not have significant literary or monumental texts in Linear A, we
do not know if the numerals were ever used in other contexts.
While the Cretan hieroglyphic numerals were formerly thought to be ancestral to
Linear A, it now appears that Linear A predates the Cretan hieroglyphs, perhaps by
as much as a century. The exact historical relationship between the two numerical
notation systems is unclear, but I believe it most likely that the Cretan hieroglyphic
numerals were a local variant of the Linear A system. The Linear B Mycenean script
used on Crete and the Greek mainland definitely derived from Linear A. Its numerals
(to be discussed later) are nearly identical to those of Linear A. The precise relation
between the peoples using the Linear A and B scripts is still unclear, as is the question
of the cause of the collapse of the Minoan civilization in the fifteenth century bc.
Presumably, during this period, the Greek-speaking Myceneans adapted Linear A for
their own language, resulting in Linear B. The two scripts coexisted in Crete from
about 1550 to 1450 bc, after which time Linear B replaced Linear A completely.

Cretan Hieroglyphic
The Cretan Hieroglyphic or Pictographic script was first identified by Sir Arthur
Evans (1909) based on his work at Knossos. While it was once considered ancestral to the other Aegean scripts, it probably developed about the same time as, or

Hieroglyphic Systems

59

Table 2.13. Cretan hieroglyphic numerals


1

10

100

1000

\\=

8357 = \000\\\====

===
slightly later than, the Linear A script. Its use is generally thought to have lasted
from 1750 to 1600 bc (Bennett 1996: 132). It is found on around 300 attested
seal-stones and clay documents (Olivier et al. 1996). While the script is still undeciphered, it is probably of a mixed syllabic and logographic structure, like other
Aegean scripts. Among the few Cretan hieroglyphic signs that can be interpreted
securely are the numerals, which are shown in Table 2.13 (Evans 1909: 258; Sarton
1936b: 378; Ventris and Chadwick 1973: 3031).
The system is cumulative-additive and decimal, and most often written from
left to right, although right-to-left numeral-phrases are also attested. Groups of
multiple repeated signs were sometimes organized using two rows, one above the
other, each with no more than five signs, but this rule was not strictly applied, and
in other cases the organization of signs was more haphazard. Figure 2.3 depicts a
Cretan clay rectangular bar on which numerals are written on three of the four
long sides plus the base (Evans 1909: 177). While the number 483 is written at the
bottom of side (b) according to this principle, for instance, many of the other
numerals are oriented irregularly, grouping signs in clusters of six or more. Evans
(1909: 257) assigns the uncommon sign P the value 1/4 because it is repeated
not more than three times at the end of a few numeral-phrases, while Dow and
Chadwick (1971: 12) suggest a quite different fractional system with signs for 1/2,
1/4, and 1/8.
Since the Cretan hieroglyphs are largely undeciphered, it is difficult to speculate on the history of their numerals. As with other Aegean scripts, an Egyptian
origin for the system has been proposed (Sarton 1936: 378), though this cannot be
demonstrated conclusively. There is limited similarity between the numeral-signs
for the Cretan hieroglyphs and any other system, except that the use of the dot for
10 is common to some early Linear A inscriptions. Dow and Chadwick (1971: 14)
suggest that the differences between the Cretan hieroglyphs and Linear A are attributable to the fact that the former were designed for chiseled inscriptions while
the latter were intended to be written with ink. The Cretan hieroglyphic numerals
are probably a local variation of the Linear A numerals or, less plausibly, a direct
borrowing from the Egyptian hieroglyphic numerals. The contexts in which the

60

Figure 2.3. Cretan hieroglyphic inscriptions on a clay bar containing numerals. For instance, at the bottom of face b, the
numeral 483 is represented with four diagonal strokes, eight dots, and three curved strokes. Source: Evans 1909: 177.

Hieroglyphic Systems

61

numerals are found are similar to those for Linear A. The Cretan hieroglyphic
inscriptions include information on commodities such as wheat, oil, and olives
and thus are probably records of transactions, inventories of goods, and similar
administrative documents (Ventris and Chadwick 1973: 31). By around 1600 bc,
Cretan hieroglyphs had been entirely replaced by Linear A.

Linear B (Mycenean)
The Linear B script was used on Crete and the Greek mainland in the middle to
late second millennium bc to write an archaic Greek dialect on clay administrative tablets. It is written from left to right, and consists of a syllabary with a large
repertory of logograms and taxograms (classifiers), including a numerical notation
system. The Linear B numerals are shown in Table 2.14.
The Linear B signs are mostly identical with the Linear A signs, except that the
sign for 10 is always a horizontal stroke (never a dot), and there is a sign for 10,000
that is not found in the earlier system. The 10,000 sign is probably a multiplicative
combination of the signs for 10 and 1000. The structure of the system is cumulative-additive and decimal, with the highest powers on the left, written in descending order and with five or more identical signs divided into two rows.
Unlike the Linear A numerals, Linear B lacks a separate system for expressing fractions; instead, specific logograms express divisions of metrological units
and then combine with numeral-signs as appropriate (just as one might say 10 cm
instead of 0.1 m). Ventris and Chadwick (1973: 5455) note that some of the
Mycenean logograms for metrological units resemble the Minoan signs for fractions, and may have originally indicated specific ratios of two types of units, which
further shows the indebtedness of Linear B to its Minoan forerunner.
The Linear B system definitely originated through direct contact with the
Minoan civilization and the Linear A numerals. The earliest Linear B inscriptions
date from the sixteenth century bc, so the two scripts coexisted on Crete for about
a century. Their numerical notation systems are so similar that some authors do
not distinguish between the two (Ventris and Chadwick 1973: 53; Struik 1982).
The distinction between the two is not nearly as great as between the two scripts,
which record different languages. Throughout the history of the Linear B numerical notation, there is no observable change in the form of the numeral-signs or in
the structure of the system.
Linear B numerals are found almost solely on clay tablets serving accounting and financial purposes (Olivier 1986: 384386). Numerals are used both for
counting discrete objects (men, chariots, etc.) and for measures of dry and liquid
volume and weight. Almost all Linear B documents relate to administrative and
bookkeeping functions, suggesting a very limited level of literacy and numeracy

Numerical Notation

62

Table 2.14. Linear B numerals


1

10

100

1000

10,000

68,357 = \\\
\\\\\\\\\\\\\\\\ \\\\
throughout Mycenean society. Even so, the consistency of the numerals throughout several centuries and across a substantial geographic area suggests that some
sort of scribal education system was in place to transmit knowledge of both the
Linear B script and its numerals. We do not know if Linear B numerals were written on papyrus or other materials, though such uses are certainly possible.
We also do not know whether the Myceneans used their numerals for arithmetical purposes. Andersons (1958) theory on the means by which such calculations
could be undertaken suffers from the defect that it involves aligning and manipulating numbers as one would in Western arithmetic, although there is no evidence
that such a procedure was ever undertaken. Dow (1958: 32) and Anderson (1958:
368) both point to a clay tablet found at Pylos (designated Eq03) in which tallying
in groups of five units is used to reach 137. Other tablets from Pylos discussed by
Ventris and Chadwick (1973: 118119) show that the Myceneans could successfully
compute complex ratios in order to determine the contributions of goods required
from towns of different sizes. Rather than proving that the Myceneans used numerical notation for arithmetic, however, these examples indicate that tallying by
units and in groups of five, rather than the purely decimal-structured numerical
notation, was the method used for computation. None of this denies that clay tablets recorded the results of rather complex computations done mentally, through
tallying, or perhaps by some other method.
There is no relationship between the Mycenean numerals and either of the later
Greek numerical notation systems (the acrophonic and alphabetic systems). It is
conceivable, however, that there is some relationship between the Mycenean and
Etruscan numerals (Chapter 4). Both Haarmann (1996) and Keyser (1988) have
raised this claim, which will be discussed in detail when considering the origins of
the Etruscan system. Mycenean settlements have been found in Sicily and southern Italy, providing one possible locus for cultural contact. However, this theory
is controversial, not least because of the time elapsed between the latest known
Linear B documents (twelfth century bc) and the first Etruscan ones (seventh century bc). A more likely descendant of Linear B numerical notation is the Hittite
hieroglyphic system, which was invented around 1400 bc and used by Hittite and

Hieroglyphic Systems

63

Luwian speakers in Anatolia. The Hittite signs for 1 and 10 are identical to the
Linear B ones, and at the time when the Hittite numerals were developed, there
were Mycenean settlements in western Anatolia (such as at Miletus) and on
Cyprus that were engaged in trade throughout the eastern Mediterranean. The
contemporaneity of the two systems makes this scenario plausible, if not proven.
The perplexing and apparently violent end of the Mycenean civilization in the
twelfth century bc, and the repeated razing of major sites such as Mycenae and Pylos, marks the end of the Linear B inscriptions and the start of the Dark Age of
Greek civilization. No writing or numerical notation of any kind is attested from
the Aegean region between 1100 bc and the introduction of the Greek alphabet a
few centuries later.

Hittite Hieroglyphic
The Hittites lived in central Asia Minor from about the end of the third millennium bc. The Hittite and closely related Luwian languages are the first Indo-European languages for which we have solid textual evidence. By the middle of the second millennium bc, two distinct scripts were in use in the Hittite Empire. Firstly,
a cuneiform script (borrowed from Mesopotamia) was used to write the Hittite
language. Its numerals are closely related to the Assyro-Babylonian cuneiform system, and so will be treated in Chapter 7. Additionally, an indigenous hieroglyphic
script was used to represent the Luwian language on monumental inscriptions,
on a few lead tablets, and probably also on wooden tablets that have not survived
(Melchert 1996: 120). This script was used from about 1500 to 1200 bc, during the
apogee of the classical Hittite Empire, and then is found only sporadically until the
rise of the Neo-Hittite kingdoms between around 1000 and 700 bc, during which
time it was again common (Hawkins 1986: 368). This script is known as Hieroglyphic Hittite or Hieroglyphic Luwian, and has a mixed syllabic and logographic
structure. Among the purely ideographic signs, the Hittites used a set of written
numerals as shown in Table 2.15 (cf. Laroche 1960: 380400).
The system is purely cumulative-additive and uses a base of 10. Numeral-phrases
were written from left to right, right to left, or top to bottom, depending on
the overall direction of the inscription. As in the Egyptian and Aegean systems,
Hittite numeral-signs were sometimes but not always grouped in clusters of three
to five unit-signs. Laroche (1960: 395) indicates that 9 was variously written using
three rows of three strokes, a row of five above a row of four, or simply with nine
strokes in sequence on a single line.
The Hittite hieroglyphic numerals were most likely based on one of the Aegean
numerical notation systems. Both the Linear A and Linear B scripts were in use
around 1500 bc, when the first Hittite hieroglyphic inscriptions are found, but

Numerical Notation

64
Table 2.15. Hittite numerals
1

10

100

1000

^
(
)
3635 = )))((((((*qqqqq
Linear A was almost extinct by that time. Like the hieroglyphs, the three Aegean scripts
use a combination of syllabograms and logograms. The Linear A, Linear B, and Hittite hieroglyphic numerical notation systems are all decimal and cumulative-additive,
and use a horizontal stroke for the units and a vertical stroke for the tens. There
was a significant degree of intercultural contact between the Aegean and Asia
Minor during this period. The Myceneans had settlements in western Anatolia
and traded throughout the eastern Mediterranean, and were possibly the Ahhijawa (Achaeans) mentioned in the Hittite archive from Bogazkoy. Because the
Luwian language was spoken primarily in western Asia Minor and only later was
used in the Hittite Empire, the transmission of the numerals from the Aegean to
western and then central Anatolia is plausible (Hawkins 1986: 374). An alternate
hypothesis is that the Hittite system was based directly on the Egyptian hieroglyphic numerals, since the Hittites were in contact with Egypt at that time.
Due to the paucity of extant examples, little can be said about the function
and use of the system. The numerals are found on a variety of stone inscriptions and lead tablets. Most notable among these are the Kululu lead strips (mid
to late eighth century bc), which record village census data using an abundance
of numerical signs (Hawkins 2000: 503505). The Hittite numerical notation is
used far more frequently than lexical numerals, which is also true of the Egyptian
hieroglyphs and Aegean scripts. There is no discernable change in the structure
or sign-forms of the system throughout its history, even though there is little evidence for its use between 1200 and 1000 bc, following the invasion of Phrygians
and others who ended the classical Hittite kingdom. During these two centuries,
the hieroglyphs were likely used only on perishable materials, such as wooden
tablets (Hawkins 1986: 374).
A few inscriptions on clay jars found at the Urartian site of Altintepe (in eastern Asia Minor) use a syllabary closely related to the Hittite hieroglyphs to write
single words in the Urartian language, starting in the early eighth century bc
(Laroche 1971, Klein 1974). Many of these inscriptions contain numeral-signs
for small numbers using either pitted dots or vertical strokes to represent units
(i.e., 5 = 554 or 11111), but never to express numbers larger than eight, making
this system an unstructured tally system having no base. Klein (1974: 93) accurately states that this usage should thus be viewed as an isolated and short-lived

Hieroglyphic Systems

65

phenomenon, possibly not outlasting the career of a single (foreign?) scribe. The
numerals that accompany the Cypriote syllabary, which was invented around
800 bc, are also potentially derived from the Hittite hieroglyphic numerals. The
proximity of the Neo-Hittite kingdoms to Cyprus, the extensive trade relations
between the regions, and the identical structure of the two systems all suggest
that such a derivation is likely. However, there are too few Cypriote syllabic inscriptions containing numerals to establish an accurate chronology or even to
secure values for certain numeral-signs. Less plausible descendants of the Hittite
hieroglyphic system are the earliest Levantine systems, Phoenician and Aramaic
(Chapter 3). However, these systems developed around 750 bc, at the very end of
the Hittite systems history, and are structurally distinct from it, since they have a
sign for 20 and are multiplicative-additive above 100.
The subjugation of the Neo-Hittite kingdoms under the Assyrian empire ended
the use of Hittite hieroglyphic numerals around 700 bc, and the system was replaced for all functions by the Assyro-Babylonian common numerals. Later numerical notation systems developed for related peoples of Asia Minor, such as
the Lycians, were based on a Greek model and display no obvious relation to the
Hittite hieroglyphs.

Cypriote Syllabary
As its name suggests, the Cypriote syllabary was a syllabic script used only on the
island of Cyprus. It was used between about 800 and 200 bc for writing the Greek
language, and thus coexisted with the much more prominent and long-lasting
Greek alphabetic script (Bennett 1996: 130). Cypriote is always written from right
to left. None of the synthetic works concerning numerical notation have dealt
with the (admittedly small) evidence for a distinct Cypriote numerical notation
system. However, Masson (1983: 80), whose discussion of the Cypriote syllabary is
the most detailed currently available, presents about a dozen inscriptions in which
the system shown in Table 2.16 was used.
This rudimentary system was decimal and cumulative-additive and, like the
syllabary itself, was written from right to left. The numbers expressed using the
system are very small; unless certain undeciphered signs are in fact numeral-signs (as
discussed later), the largest number expressed in any Cypriote inscription is 22. This
system parallels the Aegean Linear systems from which the Cypriote numerals
are probably derived. This is strongly suggested by the use of the Cypro-Minoan
script, which was very probably borrowed from Linear A, on Cyprus as early as
1500 bc. However, eastern Cyprus was under Phoenician domination well into the
period of the use of the syllabary, and the Phoenician numerical notation system
is also written from right to left, and uses vertical strokes for units and horizontal

Numerical Notation

66

Table 2.16. Cypriote numerals


1

10

strokes for tens. Furthermore, Masson (1983: 80) notes the use of two unusual
symbols: , found in but a single inscription but possibly indicating 100 on the
model of the Aegean systems, and , also in only a single document, but possibly
signifying 20. It is notable that the Phoenician system used @\and D at various
times as the sign for 20. Because Cypriote inscriptions do not contain dates, it is
often difficult to place them in chronological context, but it is possible that the
Cypriote system is either ancestral to or descended from the Phoenician system.
A final complexity is that the Hittite hieroglyphic numerals, which were still in
use in the Neo-Hittite kingdoms in 800 bc, also use a vertical stroke for 1 and
a horizontal one for 10. Trade between Cyprus and Asia Minor was substantial,
and it would have been an extremely short sea voyage between the two regions.
None of this material categorically excludes the possibility that the aberrant signs
found by Masson are non-numerical and that the Phoenician, Hittite, and Cypriote numerals are unconnected except by their temporal and geographic proximity
on the island of Cyprus. The corpus of inscriptions containing numerical signs is
simply too limited, and the numbers expressed too small, to resolve the issue of
their origin.

Summary
Despite the enormous amount of work being done in the archaeology of the eastern Mediterranean, the genetic relations among the systems of this phylogeny have
not been analyzed adequately in the past. The connections between the Egyptian
hieroglyphic, hieratic, and demotic systems are well established, but more data
are needed to establish the specific links between the Egyptian and Aegean systems. Nevertheless, on the basis of a shared set of features that distinguish it from
other, superficially similar phylogenies such as the Levantine (Chapter 3) and Italic
(Chapter 4), the inclusion of all the hieroglyphic systems in a single group is warranted. First, all the hieroglyphic systems have a base of 10, but they do not use a
sub-base of 5 or any additional structuring signs. Second, they mostly have a cumulative-additive structure, although the hieratic, demotic, and Meroitic systems
are ciphered-additive reductions of the original structure. Third, large numbers of
cumulative signs in a numeral-phrase are grouped in sets of three to five. Fourth,
their direction of writing can be quite variable (left-right, right-left, top-bottom,

Hieroglyphic Systems

67

or boustrophedon). Finally, hieroglyphic numerical notation systems are used far


more frequently than lexical numerals for expressing numbers.
While no hieroglyphic systems survived past 400 ad, its less direct descendants
include the Roman numerals and probably even Western numerals (though greatly transformed). In the following four chapters, I will discuss a) the Levantine
systems (Chapter 3), the Phoenician-Aramaic numerals and related systems; b)
the Italic systems (Chapter 4), the Etruscan and Roman numerals and their descendants; c) the Alphabetic systems (Chapter 5), the Greek alphabetic numerals and
related systems; and d) the South Asian systems (Chapter 6), the Brhm system
and its descendants. While they are distinct enough to warrant placing them in
separate families, all originate ultimately from hieroglyphic systems.

chapter 3

Levantine Systems

The first millennium bc was an era of considerable interregional commerce,


warfare, and colonization in the Levant. This region was peripheral to both
Egypt and Mesopotamia and thus exposed to multiple cultural influences.
The various Levantine numerical notation systems that developed in the first
millennium bc share several common features that reflect their debt to both
Mesopotamia and Egypt, while demonstrating their indigenous creators considerable inventive energy. While this phylogeny of numerical notation systems was developed and most widely used in the Levant, it would eventually
be adopted in various script traditions in Asia Minor, Arabia, Iran, the Indian
subcontinent, and Central Asia. The Aramaic notation is the most important of the Levantine family of systems, which also includes the Phoenician,
Palmyrene, Nabataean, Kharoh, Hatran, Old Syriac, Middle Persian, Sogdian, Manichaean, and Pahlavi systems. The most commonly used signs of
these systems are shown in Table 3.1.
Unfortunately, despite their widespread use over a large geographical area, these
systems remain poorly analyzed in recent scholarship, so we must turn to the earlier work of epigraphers and paleographers such as Schroder (1869), Duval (1881),
Lidzbarski (1898), Cooke (1903), and Cantineau (1930, 1935) for analyzing Levantine numerical notation. Despite the age of these works, there is no reason to
question the data presented. However, this tradition of scholarship was primarily
oriented toward the study of the texts of specific societies. Issues of diffusion and
68

69

Pahlavi

Manichaean

Sogdian

Middle
Persian

Old Syriac

Hatran

Kharoh

Nabataean

Palmyrene

Phoenician

Aramaic

a
a
a
a
a
a
a

10

a b d

C D

G
A

A
A
H A
N K

> A
> A

R
g

BA

Table 3.1. Levantine numerical notation systems

C
B @ D
J
J
J

J
@

20

F\

A
I
L

100

500

1000

10,000

Numerical Notation

70
Table 3.2. Aramaic numerals
1

10

20

100

1000

A C
F
G
2894 = \0\aaa\A\CCCC\F\aa\aaa\aaa\G\aa
cross-cultural comparison have not previously been addressed, and much work
remains to be done.

Aramaic
The Aramaeans, who originally inhabited a large portion of modern-day Syria,
are first recognizable in the archaeological and written records around the end of
the second millennium bc. During the ninth and eighth centuries bc, Aramaeans
ruled a number of small states in the Levant, until these came under the domination of the Assyrian empire. Around this time, they developed a consonantal script
on the model of the pre-existing Phoenician consonantary. By the eighth century
bc, Aramaic inscriptions began to include numerical signs, shown in Table 3.2.
The system is purely cumulative-additive for numbers up to 99, written (like
the script itself ) from right to left, using signs for 20, 10, and 1. The unit-signs are
grouped in threes, since up to nine such signs could be required. Occasionally,
when an ungrouped unit-stroke was present in a numeral-phrase, it was written
at a slight angle (so that 7 would be 0\aaa\aaa). Because there was a sign for 100, no
more than four 20-signs and one 10-sign would ever be required, obviating the
need for such groupings for higher values. The 10-sign appears to have originally
been a simple horizontal stroke, with a tail added cursively. The 20-sign is almost
certainly a ligatured combination of two 10-signs, as shown by the occasional use
of a variant form B. There is a gradual trend over time toward the use of a special
sign for 5 (H), which Lidzbarski (1898: 199) notes appearing on an Assyrian clay
tablet as early as 680 bc. However, the majority of Aramaic numeral-phrases do
not use a symbol for 5.
Above 100, the Aramaic numerical notation system is multiplicative-additive
rather than cumulative-additive, and it is thus a hybrid system. To form 800, for
instance, eight unit-signs (appropriately grouped) were placed in front of the sign
for 100 in order to indicate that the values should be multiplied. The same principle was followed for the thousands. There were apparently two signs for 1000;
the first, G, is actually no more than the final two letters of the Aramaic lexical

Levantine Systems

71

numeral LP thousand (Gandz 1933: 6970), while the second, , is the same as
the corresponding Phoenician numeral-sign (Lidzbarski 1898: 201202).
While there is no distinct sign for 10,000 in the Aramaic system used in the
Levant (though see the following discussion for Egyptian variants), rarely numbers greater than 9,999 were written using 10- and 20-signs in conjunction with
the sign for 1000. Fractions are apparently found in a handful of inscriptions in
which ungrouped unit-strokes aaaa and aaaaa mean 1/4 and 1/5, and one inscription
contains a special sign for 2/3 () (Lidzbarski 1898: 202), but they were normally
written out lexically.
The first Aramaic inscription with numerical notation is an eighth-century bc
ostracon from Tell Qasile, in which 30 is expressed as three horizontal strokes (*)
rather than the normal form (Lemaire 1977: 280). However, it may be a Hittite
hieroglyphic numeral-phrase (Chapter 2), since that system was still in use in the
eighth century bc in the Neo-Hittite kingdoms to the north. The earliest uncontestable examples are from the Assyrian bronze lion-weights found at Nimrud by
Layard in the nineteenth century. These eighth-century inscribed weights have
texts in Aramaic and Akkadian; on the largest (BM 91220; CIS II/1, 1), dating to
the reign of Shalmaneser V (726722 bc), the number 15 indicates the objects
weight of fifteen minas in three different ways on its three lines of text: in Aramaic
lexical numerals, as fifteen ungrouped single strokes, and according to the structure detailed above (aa\aaa\A) (Fales 1995: 35). This threefold repetition using different
methods of representation suggests that the system was unfamiliar, either because
of its novelty or because it was intended for speakers of several languages.
Structural similarities between the Aramaic system and the Assyro-Babylonian
common system (Chapter 7), with which it shares a decimal base and the use of
multiplicative-additive structuring for the hundreds and thousands, suggest a historical connection (Gandz 1933: 69; Ifrah 1985: 356). The conquest of the Aramaeans in 732 bc by the Assyrian empire establishes a clear historical context in which
this transmission could have taken place. The lion-weights from Nimrud may
well have been taken from the Levant as war booty shortly after this time
(Fales 1995: 54). Yet the Aramaic system is also similar to the Egyptian hieroglyphic
system. Aramaic speakers would certainly have had considerable contact with
Egypt in the eighth century bc, and by the sixth century bc the Aramaic script was
being used by settlers in Egypt at Elephantine and Saqqara. There are a number
of similarities in the forms for signs. Like the Egyptian hieroglyphs but unlike the
Assyro-Babylonian system, Aramaic uses vertical unit-strokes grouped in threes to
express the units. A relationship between Aramaic A and hieroglyphic r (both
signifying 10) has also been postulated (Schroder 1869: 186), although it is more
likely that the hooked Aramaic sign is simply a cursive alteration of a horizontal
stroke. Regardless, both signs are very different from the cuneiform Assyro-Babylonian

Numerical Notation

72

Table 3.3. Aramaic, Egyptian hieroglyphic, and Assyro-Babylonian numerals

Aramaic

424 =

\aaaa\C\F\aaaa
4

Hieroglyphic

424 =

qqqqrr
4

Assyro-Babylonian

424 =

20 100 4

20

400

4i\b4
4 100 20 4

system. The Aramaic use of unit fractions along the Egyptian hieroglyphic model,
including the exception of having a special sign for 2/3, further suggests Egyptian
borrowing. West Semitic accounts, like those in the Egyptian hieroglyphic and
Aegean scripts, are written with the item being enumerated placed before the
numeral, in contrast to Mesopotamian texts, which follow a quantity + item
order (Levine 2004: 435). Finally, Egyptian hieroglyphic numeral-phrases are primarily written from right to left, as in Aramaic, whereas the Assyro-Babylonian
system runs in a left-right direction, although of course the Aramaic script is also
written right to left, so this cannot be taken as positive evidence in its own right.
These differences are compared in Table 3.3.
To muddy the waters even further, two other Hieroglyphic numerical notation
systems were used in the eastern Mediterranean around 750 bc and could potentially have been known to the early users of Aramaic numerals. The Neo-Hittite
kingdoms, although on the wane by that time, were still present in southeastern
Anatolia, immediately abutting the Aramaeans. Moreover, the Cypriote numerals
were invented just before that time, and there was enormous trade between Cyprus
and the Levantine coast. But while the Cypriote and Hittite systems are cumulativeadditive and decimal and use vertical strokes for 1 and horizontal strokes for 10,
they lack the other characteristics that might identify them as potential ancestor
systems.
The Aramaic numerals were likely developed under a dual cultural influence
from Egypt and Mesopotamia. The systems basic structure is very similar to
the Assyro-Babylonian common system, but many paleographic and contextual
similarities are far more similar to the Egyptian hieroglyphs. Geographically and
historically, the Aramaeans and other Levantine peoples were peripheral to both
civilizations in the mid first millennium bc, at the time of the systems invention.

Levantine Systems

73

Although cultural phylogenies for scripts and numerical notation systems are usually arranged in accordance with a biological taxonomic scheme, cultural phenomena may have multiple origins, each making a contribution to the descendant,
much as biological parents contribute to a childs genetic makeup. If this explanation is correct, we must ask why the Egyptian hieroglyphic numerals, rather than
the hieratic, would be chosen as a model for the Aramaic numerals. As I discussed
in Chapter 2, the hieratic numerals were widely used in the Kingdom of Judah
in the first half of the first millennium bc. Ostraca from Arad in the Negev dating to the late seventh and early sixth centuries attest both hieratic and Aramaic
numerals (Levine 2004: 433). Like Millard (1995: 190191), I find the failure of the
Aramaeans to adopt the hieratic numerals to be rather curious.
The existence of a distinct sign for 20 in Aramaic, and the recombination of
features of two quite different systems, demonstrates that the Aramaeans were
numerically inventive. In most of the Semitic languages, the word for twenty
is etymologically the plural of ten for example, Hebrew eser ten versus esrim
twenty (Menninger 1969: 14). This may explain why the graphic etymology of
the Aramaic numeral-sign for 20 is two ligatured 10-signs. This development of
a special sign for 20 outside the regular decimal base of the numerical notation
system is a unique development of the Levantine numerical notation systems; neither the Assyro-Babylonian system nor the systems of the Hieroglyphic phylogeny
have this feature.
Like the script to which it was attached, the Aramaic numerical notation was
used in the Levant, Egypt, Mesopotamia, and farther afield throughout the second
half of the first millennium bc. Segal (1983) gives ample evidence for the use of
the system among Aramaic texts from Saqqara in Lower Egypt throughout the
fifth and fourth centuries bc, and Aramaic papyri found at the fifth-century bc
military colony at Elephantine demonstrate the use of the system in numerous
administrative documents. While the system as used in the Levant had no special
sign for 10,000, the Aramaic papyri found at Saqqara and Elephantine do use such
a sign (), which obeys the multiplicative principle in the same way as detailed
earlier for 100 and 1000 (Segal 1983: 131; Ifrah 1985: 335). An alternate sign for 100
() was also used in Egyptian Aramaic, but it resembles none of the signs used
in the Levant and is not similar to any of the Egyptian demotic or hieratic signs
used at that time.
The Aramaic script was widely used throughout the Achaemenid Empire from
the sixth to fourth centuries bc on clay administrative and legal tablets, stone
monuments, and leather and papyrus. While the scripts used in official royal proclamations and dedications were Old Persian, Babylonian, and Elamite, Aramaic
was the lingua franca of the empire and served most administrative functions. As
such, it was used as widely as Lower Egypt, Asia Minor, and the Transcaucasus and

74

Numerical Notation

even as far east as the Indus River. Throughout its history, the Aramaic numerical
notation was used extensively on monumental inscriptions, ostraca, and administrative papyri. In literary and religious texts, however, numbers were more often
written using lexical numerals only. Aramaic numerals were used to record the
results of calculations used in commerce and administration, but none of the
extant inscriptions demonstrate the use of written arithmetic.
The end of the Achaemenid Empire did not spell the end of Aramaic influence over the Middle East; however, it did result in the fragmentation of what
previously had been a unified script and numerical notation into several regional
variants. During this period, Greek alphabetic numerals were often used administratively, although Aramaic numerals continued to be used in a variety of contexts. By the second century bce, political and ethnic divisions in the Near East
had led to the emergence of variant numerical notation systems. The Hellenized
Palmyrene, Nabataean, Hatran, and Edessan Syrian populations of the Levant
each possessed its own variant numerical notations based on Aramaic. In these
variants, the use of a distinct sign for 5 was far more prominent than in Aramaic
numerals. The Kharoh numerical notation used in parts of modern Afghanistan
and Pakistan, and the Middle Persian and Pahlavi systems used in Iran, are also
variant forms of Aramaic.

Phoenician
The Phoenicians, who inhabited various cities (Tyre and Sidon foremostly) along
the Levantine coast in the first millennium bc, were perhaps the greatest mercantile people of the ancient Mediterranean. The Phoenician consonantal script
was a descendant of the earlier Canaanite consonantary that diverged from its
ancestor late in the second millennium bce. However, none of the earliest Phoenician inscriptions contain numerical notation. While the Aramaic writing system
developed from the earlier Phoenician, the Aramaic numerals appear to be slightly
prior to the Phoenician, and there is no reason to assume that script and numerical notation must be borrowed jointly. The Phoenician numerical notation system
is similar in structure to the Aramaic, with distinct signs for 1, 10, 20, 100, and
1000. These signs (including some paleographic variants) are shown in Table 3.4
(cf. Schroder 1869; Lidzbarski 1898; Gandz 1933; van den Branden 1969: 4243).
Like Aramaic, this system is purely decimal with the exception of the 20-sign,
cumulative-additive below 100 and multiplicative-additive thereafter. Unit-signs are
simple vertical strokes, although a left-slanting stroke is often used for ungrouped
single strokes, and are grouped in threes, as in the Egyptian hieroglyphic and
Aramaic systems. Like the Phoenician script itself, numeral-phrases are nearly
always read from right to left, although van den Branden (1969: 43) notes at least

Levantine Systems

75

Table 3.4. Phoenician numerals


1

10

20

A B @
D
697 = 0\aaa\aaa\ADDDD\\aaa\aaa

100

1000

one exception (CIS.87,ph) in which left-to-right ordering is used, probably in


error. The most notable feature of Phoenician numerals is the wide variety of
forms for number-signs, particularly for 20 and 100. Schroder (1869: 188189 and
Table C) lists over twenty variants each for these two numbers, some of which can
be attributed to differing individual scribal styles, while others may reflect regional
or diachronic variation. I list only the more common forms for the sake of brevity. The 1000-sign is extremely rare; Lidzbarski (1898: 201) reports only a single
instance from Tyre. There is no evidence whatsoever of the use of a distinct sign
for 5, in contrast to later Levantine systems, nor is there any evidence of numeralsigns for fractions.
For numbers greater than 100, a multiplicative-additive structure is employed
as in Aramaic; a group of cumulative unit-signs preceding a single 100-sign indicates multiples from 100 to 900, with any additional signs to the left indicating
the component of the number less than 100. It is likely that the rare sign for 1000
also combined multiplicatively with sets of grouped unit-signs.
It is sometimes claimed that the Phoenicians used an alphabetic (presumably
ciphered-additive) numerical notation system as early as 900 bc (Dantzig 1954:
295; Zabilka 1968: 117119). The myth of Phoenician alphabetic numeration has
been repeated for more than a century, but there is no foundation for this assertion. The first alphabetic numerals were developed by the Greeks in the late sixth
century bc (Chapter 5). Zabilka (1968: 118) claims that the first ten letters of the
Phoenician alphabet were used on coins minted at Sidon to represent the numbers 1 through 10, based on Harris (1936:19), who is, however, referring only to
Alexandrine coins. By this time, the Greek alphabetic numeral system was used
throughout the Levant by speakers of both Indo-European and Semitic languages.
Even this does not prove the existence of a true alphabetic numerical notation
system among the Phoenicians in the fourth century bc; it could, rather, indicate
a system of letter labeling as used by the Greeks (Tod 1979: 98105), which is not
really different from the practice of modern writers who label points of discussion
A, B, C, and so on.
The first example of numerical notation in a Phoenician inscription is the Karatepe inscription of around 750 bc, which contains a single stroke for 1 (Millard
1995: 191). If this is a true example of the system just described, then its appearance

76

Numerical Notation

is virtually simultaneous with that of the Aramaic system. However, one unit-stroke
is scant evidence for this. From the seventh century onward, however, Phoenician
texts containing numerals are relatively common, including inscriptions on stone,
ink writings on clay, administrative documents on papyrus, and, at a somewhat
later period, inscriptions on coins. Phoenician numerals were often used to enumerate regnal years and for record keeping of commodities. However, between
the seventh and first centuries bc, Phoenicia was politically dominated in turn
by the Assyrian, Neo-Babylonian, Achaemenid, and Alexandrine Greek empires.
The Phoenician numerical notation thus predominated in the Levant only during its early history. However, in the Phoenician colonies in North Africa and
Spain (including, most importantly, Carthage), the Phoenicians continued to use
the system detailed here, without significant regional variation, until Roman and
Greek conquests in the second century bc effectively ended its use. Coins from
Akko, Tyre, and Sidon used Greek alphabetic numerals as early as 265 bc, though
at Arvad and Marathus, Phoenician numerals were used on coins until about 110 bc
(Millard 1995: 193).
The fruitful transmission of the Phoenician consonantal script throughout the
Aegean and the Middle East has led some to speculate as to the transmission of
its numerical notation system. Millard argues that Phoenician may have been the
model for the Greek acrophonic (base-10, sub-base 5, cumulative-additive) numerical notation system (Millard 1995: 192). However, the acrophonic systems sub-base
of 5, coupled with the more obvious derivation of acrophonic numerals from the
very similar Etruscan system, makes such an origin unlikely. Schroder (1869: 187f )
suggests that the Lycian numerals (Chapter 4) are a variant of Phoenician, but
that system much more closely resembles the Greek acrophonic numerals than the
Phoenician. It is entirely possible that one or more of the later Levantine systems
have a Phoenician as opposed to Aramaic origin, but there is no good way to
demonstrate this in most circumstances. The lack of a symbol for 5 in Phoenician
numerals suggests that most of these later systems were Aramaic-based. Phoenician numeration, then, unlike the Phoenician writing system, is essentially a side
branch of the broader Levantine family.

Palmyrene
Palmyra was an important mercantile city located in modern Syria around
200 km northeast of Damascus, and whose inhabitants, Aramaic-speaking Semites,
managed to retain considerable control over their own affairs despite Greek and
Roman influence in the area. Palmyrene inscriptions are found dating from the
first century bc to the mid third century ad, continuing the tradition of the earlier
Aramaic script. Palmyrene numerical notation retained much of the structure of

Levantine Systems

77

Table 3.5. Palmyrene numerals


1

a
178 =

10

20

H
A
J
aaa\\H\\\A\\J\J\J\\A\a

the older Aramaic system, while introducing new numeral-signs. Despite their
relative obscurity, the Palmyrene numerals were first analyzed over a quartermillennium ago by Swinton (175354). The Palmyrene system had distinct signs
for 1, 5, 10, and 20, as shown in Table 3.5.
These four symbols express any number less than 100. While in earlier Aramaic
scripts the sign for 5 appeared only sporadically, it was a fundamental part of the
Palmyrene system. Because of this, only four unit-signs were required at most, so
there was no need to group sets of unit-signs into threes. Like its Aramaic ancestor, Palmyrene numerical notation is base-10 and cumulative-additive below 100.
For numbers greater than 100, Palmyrene, like Aramaic, is multiplicative-additive,
with the complexity that the sign for 100 is identical to that for 10. The possibility of confusion is avoided by the requirement of having one or more unit-signs
before the 100-sign, whereas no such signs could precede a 10-sign.
While this feature resembles the use of the positional principle, such phrases are
multiplicative, not positional. To represent 100, the sign A had to be combined
with unit-signs; alone, it always meant 10, not 100. Cantineau (1935: 36) contends
that the original Palmyrene sign for 100 was a horizontal stroke placed above a
dot, but that it was later reduced until it was identical to the 10-sign. If so, the
identity of the two signs may be largely coincidental. In monumental inscriptions
on stone, Palmyrene numerals are among the clearest and most unambiguous of
all the Levantine systems. Figure 3.1 is a memorial inscription dating to the year
492 (ad 181); the phrase is clearly visible on the last three lines of the inscription
(Arnold 1905: Plate IV).
Palmyrene numerical notation was restricted geographically and temporally to
the city of Palmyra during the period from about 100 bc to 275 ad. During that
time, however, it was used widely on inscriptions and records of commercial transactions, though not normally in literary contexts. We do not know the extent to
which it may have been used in a broader range of genres due to the poor survival
of evidence.
The importance of Palmyra as a commercial center rested on its strategic location on the Roman frontier and its trade ties with peoples outside the empire.
Despite considerable Hellenisation and Latinisation, Palmyra retained its script
and numerical notation through the third century ad, though Greek alphabetic

78

Numerical Notation

Figure 3.1. A Palmyrene memorial inscription dating to ad 181; the year-date (492) is
clearly visible on the final three lines of the inscription, including multiplicative use of the
sign for 100. Source: Arnold 1905: Plate IV.

and Roman numerals came to be used more frequently for administrative and
mercantile purposes. In 273 ad, following the short-lived independent rule of
Queen Zenobia over the province (266272 ad), Palmyra was destroyed by the
Roman emperor Aurelian, abruptly ending its importance as a commercial center.
Thus, political factors, rather than criteria of function and efficiency, led to the
complete replacement of the Palmyrene numerical notation system by those of
Greek and Roman colonizers. It has sometimes been argued that Palmyrene is
ancestral to the Syriac numerical notation, though I will show below that this is
only one of many possible scenarios of transmission.

Nabataean
The Nabataeans were a South Semitic people of Arabian ancestry who inhabited
the area between Syria and Arabia in the southeastern Levant in the late first millennium bc and into the Christian era. Though not Aramaeans, they came under
considerable Aramaean influence and adapted the Aramaic script for their South

Levantine Systems

79

Table 3.6. Nabataean numerals


1

R
N
178 = aaa\\NA\\J\J\J\\I\a

10

20

100

Semitic language, including a variant of its numerical notation. This system was
used from approximately 100 bc to 350 ad in inland areas of the Levant (modern southern Syria and Jordan) including the cities of Damascus and Petra, and
even as far south as the port of Aqaba on the Red Sea. Its signs are indicated in
Table 3.6.
As with all the Levantine systems, Nabataean is decimal, cumulative-additive
below 100 and multiplicative-additive above, with additional signs for 4, 5, and 20.
Unit-strokes are grouped in threes where necessary and are sometimes joined together
at the base in cursive writing. The sign for 4 is used only in some inscriptions, and then
only in numeral-phrases for 4 (never 5 through 9); 8 is expressed as 111N (5 + 3) or
11\111\111, but never (to my knowledge) as RR. Lidzbarski (1898: 199) argues that its
shape represents four unit-strokes placed in a cross, strictly on graphic principles,
but this is unproven. Its historical connection with the identical Kharoh sign for
4 is still unclear, but some link seems probable, as the Nabataeans were frequently
engaged in commerce with peoples to the east. However, Gibson (1971: 13) notes
that the eighth-century bc Samaria ostraca, in which the Hebrew variant of the
Egyptian hieratic numerals (Chapter 2) predominates, contain a + or X-shaped
sign for 4, which would antedate either the Nabataean or Kharoh symbol by
several centuries. Finally, Cantineau (1930: 36) and Lidzbarski (1898: 199) believe
the signs for 4 and 5 to be quite late inventions, possibly independent of any other
system.
The Nabataean sign for 10 is a more arched version of the hooked horizontal
stroke used in most Levantine systems, while the 20-sign can easily be shown to
derive from the Aramaic form. In one inscription from Egypt, the year-number 160
(ad 266) is written irregularly as 100 20 10 20 10 instead of the expected 100 20
20 10 10; this seems unlikely to be a scribal error but is otherwise unexplained (Littmann and Meredith 1953: 16). The sign for 100 is not obviously related to that of any
other notation, though Cantineau (1930: 36) argues for its possible derivation from
Phoenician . The sign for 100 combines with signs for 1, 4, and 5 multiplicatively.
Accordingly, the 4-sign is used to express 400, as in an inscription from Dumr (near
Damascus) from 94 ad in which the number 405 is expressed as NIR (4 100 + 5)
(Cooke 1903: 249). Such numeral-phrases make the system look less cumulative
than it actually was. No Nabataean writings contain numbers higher than 1000.

80

Numerical Notation

The Nabataean numerical notation system is found on inscriptions dating


from around 100 bc to the late fourth century ad, primarily in the inland Levant
from Damascus south to Petra. Throughout its history, it was used in inscriptions
on edifices, on ostraca, and on coins. I do not know of any attested Nabataean
numerals in the poorly attested cursive script tradition, which varies from the
monumental script in several paleographic respects. The Nabataean legal papyri
from the Cave of Letters dating to the early to mid second century ad express all
numerals lexically (Yadin et al. 2002). The Greek alphabetic numerals (Chapter 5)
and Roman numerals (Chapter 4) were also well known and frequently used in
otherwise Nabataean inscriptions. Though the Nabataeans were politically subordinate to Rome throughout most of the period under consideration, they held a
monopoly over the caravan trade that passed from inland Arabia to the Levantine
coast. Nabataean numerical notation has been found on economic documents
and inscriptions from Greece, Italy, and Egypt.
In the fourth century ad, the Nabataean numerals began to be replaced by the Greek
alphabetic and, to a far lesser extent, Roman numerals. While the Nabataean script is
ancestral to the earliest Arabic script, there is no connection between the Nabataean
numerals and the systems used by the early Arabs. Millard (1995: 193194) reports the
use of Nabataean numerals on the pre-Islamic Arabic inscriptions from En-Namara
(dated 328 ad), and possibly on the sixth-century ad Zabad and Harran inscriptions.
Such late occurrences become increasingly rare, however, as Greek alphabetic numerals and systems based thereupon predominated throughout the Middle East, and by
the time of the introduction of Islam, no trace of the Nabataean system remained.

Hatran
A variant Aramaic script was used in the region around the city of Hatra (modern
Al-Hadr, in northern Iraq), an outpost of the Parthian Empire and later the capital
of the small autonomous state of Araba. The Hatran script, for which inscriptions
have been found dating from about 50 bc to 275 ad, possessed a distinct numerical
notation system with signs for 1, 5, 20, and 100, as shown in Table 3.7.
As with all Levantine systems, the Hatran numerical notation is decimal, cumulativeadditive for numbers less than 100, multiplicative-additive above 100, and written
from right to left. The precise relation of the Hatran system to the other Levantine
systems is unclear, but it is descended in some way from the Aramaic system used
around Hatra in the centuries prior to the development of the Hatran script, given
the similarity of signs for 1, 10, and 20 to earlier Aramaic forms. The sign for 5 is
identical to that of the Old Syriac script used around Edessa at that time. Finally,
the 100-sign is of entirely mysterious origin, though a case could be made that it
is related to the Phoenician .

Levantine Systems

81

Table 3.7. Hatran numerals


1

10

20

>

100

697 = 11>AJJJJ1>

The Hatran numerals were probably ancestral in some way to the Middle
Persian numerals, at least giving rise to some of the Middle Persian numeralsigns. Hatran numerical notation was used widely on ostraca, on inscriptions on
stone, and in economic documents. Unlike the Palmyrene and Nabataean states,
which were subjected to Roman political and economic domination for most of
their history, Hatra remained independent from both Roman and Parthian control until 272 ad, when the Middle Persian king Shapur I conquered the region.
After this time, Hatran inscriptions are more rarely encountered, and the Middle
Persian script and numerals (and its successors) replaced them.

Old Syriac
A consonantal script was used at the ancient city of Edessa (modern Urfa, in
southeast Turkey) in the early years of the Christian era. Based on an Aramaic
model, this script, which resembles the estrangela Syriac script that emerged in
the later manuscript tradition, was used to write the Old Syriac language, a close
relative of Aramaic. A large number of Old Syriac inscriptions on stone, mosaics,
and parchment have survived, dating from the beginning of the Christian era to
500 ad, and largely found in northern Syria and southern Turkey.1 The Old Syriac
numerical notation system used signs for 1, 5, 10, 20, and 100, and possibly also 2
and 500, as shown in Table 3.8 (cf. Rdiger 1862, Duval 1881, Segal 1954).
Numeral-phrases, like the script itself, are always written from right to left. The
system is decimal (with a special sign for 20), cumulative-additive for numbers
less than 100, and multiplicative-additive for higher values, all of which points to
its membership in this phylogeny and its close relationship to Aramaic. However,
it has some curious features. The sign for 2, as defined by Duval (1881: 1415), is
simply a ligatured form of two unit-strokes, a paleographic convenience that was
never used consistently or regarded as a structural feature of the system. The sign
for 5 is identical to that of the Hatran system, and the sign for 20 to one variant
form used in Phoenician. The sign for 5 was not consistently used in numeral
1

The earliest dated Syriac inscription is from ad 6, although Drijvers and Healey (1999: 17)
argue that it may have originated earlier.

Numerical Notation

82

Table 3.8. Old Syriac numerals


1

10

>
A
697 = >A@@@@

20

100

500

phrases for 5 through 9 (Segal 1954: 35). For instance, in three separate inscriptions
dating to ad 165, the date 476 of the Seleucid era is written in three different ways,
as indicated in Table 3.9 (Drijvers and Healey 1999). The earliest dated Old Syriac
inscription (As55, dating to ad 6) expresses the year-date (317) using seven unitstrokes (Drijvers and Healey 1999: 141). This suggests that the sign for 5 may have
been a later development, and at best an occasional one.
Duval (1881: 14) argues that the 100-sign is a slightly modified form of the 10sign, resembling the Palmyrene numerical notation in this respect. The Old Syriac
symbol for 500 is rare, partly because numbers of this magnitude are infrequent
in Syriac writings. Duval (1881: 14) insists that it ought to be understood in many
numeral-phrases where it is not written, such as in year-dates. However, in the
Old Syriac slave sale contract, P. Dura 28, written at Edessa in ad 243, 700 is written multiplicatively as seven unit-strokes (some ligatured) followed by the powersign for 100, rather than with a sign for 500. The notion of a sign for 500, if not its
form, may have been borrowed from the Roman numerals, given the cultural and
political dominance of Rome in Syria throughout the period.
Old Syriac numerical notation was used on numerous inscriptions on stone
around Edessa, but not in the short-lived tradition of mosaic inscriptions from the
third century ad. It evidently originated as a variant of the Aramaic system, but
its specific relationship to the other Levantine numerical systems remains unclear.
Three Old Syriac legal texts written on parchment survive, dating from ad 240243;
of these, only P. Dura 28 (just discussed) contains Syriac numerals (Drijvers and
Healey 1999: 232235; Goldstein 1966). In many inscriptions and manuscripts,
lexical numerals were used instead of numerical notation. Although the Edessan
Table 3.9. Old Syriac year-dates for 476 / ad 165
Inscription

Date

As29

a>A\\\aaaa

As36

aaaaaaA\\\aaaa

As37

aaA\\\aaaa

Transliteration
(1 + 1 + 1 + 1) 100 + 20 + 20 + 20 +
10 + 5 + 1
(1 + 1 + 1 + 1) 100 + 20 + 20 + 20 +
10 + 1 + 1 + 1 + 1 + 1 + 1
(1 + 1 + 1 + 1) 100 + 20 + 20 + 20 +
10 + 1 + 2 + 1 + 2

Levantine Systems

83

Christians were subject to Roman imperial authority throughout most of the history of their script and numerical notation, and although both Greek alphabetic
and Roman numerals were widely used in Syria, the Old Syriac numerals were not
displaced by either system.
By the fifth and sixth centuries ad, however, the older numerals began to be
replaced by the ciphered-additive Syriac alphabetic system (Chapter 5), which
assigned numerical values to the twenty-two letters of the Syriac consonantary. This
gradual obsolescence corresponds to the development of the indigenous Syriac Christian manuscript tradition, particularly oriented toward liturgical subjects. Many texts
use the two systems side by side. A seventh-century Syriac religious commentary (BM
Add. 14,603) contains several lines of Old Syriac numerals that are incomprehensible
until the numeral values are converted into their corresponding values in the Syriac
alphabetic numerals; the resulting alphabetic signs can then be read as the authors
epigraph (Wright 1870: II, 586587). While this demonstrates that the numerals were
still in use, the cryptographic nature of the note suggests that they may not have been
well known. The very latest evidence of the Old Syriac system is from the eighth century, after which only alphabetic numerals were used (Duval 1881: 15).

Kharos.t.h
The Kharoh script was used in the region of Gandhara in eastern Afghanistan and
northern Pakistan from around 325 bc to 300 ad and, from the second century ad
onward, in parts of Central Asia. Given that this region was under the control of the
Achaemenid Empire (for which Aramaic was a lingua franca) from 559 to 336 bc, the
similarity in form and value of many of the signs in the two scripts, and their common right-to-left directionality, Kharoh is clearly descended from Aramaic. During
the earliest periods of its use (before about 100 bc), Kharoh inscriptions containing
numerals are quite rare, being found in only a few royal inscriptions of the Mauryan
King Aoka, who reigned from about 273 to 232 bc. Only the numbers 1, 2, 4, and 5
are represented, and they are always formed using simple unit-strokes. In the later Saka,
Parthian, and Kusana inscriptions (dating from about 100 bc onward), a more complex
system was used, and much larger numbers were represented. This system possessed
unique signs for the numbers 1, 4, 10, 20, 100, and 1000, as shown in Table 3.10 (cf. Das
Gupta 1958: Table XIV; Salomon 1998: Table 2.6; Glass 2000: 139143).
In common with all the Levantine systems, Kharoh is purely cumulativeadditive up to 100 and multiplicative-additive thereafter. As in the script as a whole
(and in other Aramaic-derived scripts), the direction of writing is always from right
to left. Unlike other Levantine systems used at the time, Kharoh has no special
sign for 5; numbers from 4 through 9 were always expressed through combinations
of units and 4-signs. Unit-signs in the cursively written texts of Central Asia are

Numerical Notation

84

Table 3.10. Kharoh numerals


1

20

100

J
697 = aaa\R\\JJJJ\\aaR

10

1000

usually ligatured together in groups of two and three. The sign for 1000 is found
only in the late (perhaps fifth century ad) texts from Inner Asia (Das Gupta 1958:
259; Mangalam 1990: 48; Glass 2000: 143). It is most likely a variant of the similar Aramaic sign, which was a conventionalized version of the lexical numeral for
1000, LP (Salomon 1998: 64). The signs for 100 and 1000 combine multiplicatively
with signs for units less than 10, with the units to the right (before) the power-sign.
Figure 3.2 is the obverse of a leather text written in cursive script found at Niya by
Sir Aurel Stein; the numerals 3 and 25 (20 + 4 + 1) are readily visible at the bottom
of the text (Boyer, Rapson, and Senart 1920: 120, Plate V).
The Aokan-period system of vertical strokes may or may not be of Aramaic
origin, though the geographical proximity of its users, coupled with the obvious
relation of the Kharoh alphasyllabary to the Aramaic consonantary, suggests
that it was. In its fully developed form, however, it is definitely part of the Levantine phylogeny, and not related to the Brhm numerals (Chapter 6) used in India.
Kharoh shares with the other systems the right-to-left direction of writing, the
use of vertical strokes for units, similar forms for the numeral-signs for 10 and 20,
and the use of the multiplicative principle for 100.
The use of X\ for 4 is common to both Kharoh and Nabataean, and this is
unlikely to be coincidental, since they share a common sign for 20 as well, and
both systems developed around 100 bc. These are the only two cumulative systems worldwide ever to use a special sign for 4. As mentioned earlier, the Hebrew
hieratic ciphered sign for 4 was + or X, suggesting transmission from west to east.
However, Datta and Singh (1962: 23) argue that the sign may have developed by
rotating the Brhm sign for 4 (+) by forty-five degrees, and may then have been
transmitted westward to the Nabataeans. Buhler (1896: 73), in turn, contends that
the Nabataean and Kharoh signs were invented independently of one another.
The Kharoh numerical notation system was used primarily on inscriptions on
stone and on copper, but there are also surviving documents from Inner Asia written on wood, palm leaf, birch bark, and leather (Salomon 1996: 378). Throughout
its history, it was in competition with its rival, Brhm numerals (Chapter 6),
the system used on the Mauryan inscriptions of the Indian heartland. The use
of Kharoh was tied to the political independence of the Greek, Scythian, and
Parthian kingdoms, which looked to Bactrian and Iranian traditions rather than

85

Figure 3.2. A Kharoh leather text found by Sir Aurel Stein at Niya. The numeral 25 is clearly visible at the bottom left of the page (20 + 4 + 1).
Source: Boyer, Rapson, and Senart 1920: Plate V.

86

Numerical Notation

to Indian ones. By the late third century ad, the Bactrian and Indo-Scythian polities
of the Kharoh heartland were seriously weakened, and the advent of the Gupta
Empire in the fourth century ad heralded the predominance of Brhm throughout
the Indian subcontinent. However, Kharoh survived longer in the small states of
Inner Asia. Inscriptions on wooden documents from the city of Niya date to as late as
the seventh century ad, and contain a variant of the Kharoh script and numerals.

Middle Persian
The Persians of the Achaemenid, Seleucid, and Parthian Empires used the Old Persian numerals (Chapter 7), the Greek alphabetic numerals (Chapter 5), or the common Aramaic numerals already described. The Sassanian dynasty began in ad 228
when Ardashir I destroyed the Parthian Empire, which had ruled much of the territory of modern Iraq and Iran for several centuries prior. For several centuries, the
Sassanian empire was a rival of Rome and later Byzantium to the west, and of the
Gupta Empire in India to the east, and was the dominant power in Mesopotamia
and Persia until the Islamic conquest. The Middle Persian language (the ancestor of
modern Farsi), the language of Sassanian administration and commerce, was written
in an Aramaic-derived consonantary, reflecting the legacy of Achaemenid, Seleucid,
and Parthian rule in the region. The numerical notation system associated with the
Middle Persian script is shown in Table 3.11 (Frye 1973).
Like its ancestor, the Aramaic system, the Middle Persian system was cumulativeadditive and decimal for numbers below 100, and written from right to left with
the highest powers at the right. The signs for 1, 10, and 20 resemble closely the
signs used in the other contemporary Levantine systems, while the sign for 100
resembles only that of the Hatran system. Unlike most of the later variants of
Aramaic, however, there was no sign for either 4 or 5, and units from 5 to 9 were
written using grouped sets of three or four unit-strokes. The sign for 1000, as in
Aramaic and Kharoh, is a reduced version of the Aramaic lexical numeral LP
thousand. For numbers above 100, the signs for 100 and 1000 combined multiplicatively with cumulative numeral-phrases, as in the other Levantine systems.
Thus, in the Qaba inscription of Kartir, the number 6798 is written as shown in
Table 3.11 (Frye 1973: 4).
It is impossible to determine the precise historical affiliation of the Middle
Persian system to the other Levantine systems, other than to note that it is most
definitely descended from the Aramaic system in some way. The lack of a sign
either for 4 or 5 is quite unusual for such a late descendant of Aramaic, as all of the
other contemporary Levantine systems have some such sign. The Middle Persian
script is most closely affiliated with the Hatran script, and the two systems share
similar signs for 100, suggesting a historical connection. Middle Persian numerals

Levantine Systems

87

Table 3.11. Middle Persian numerals


1

10

20

100

1000

2
3
4
5
6
6798 = 1111\1111\24444\5\111\1111\7\111\111

were employed on silver bowls and plates to indicate weights, inscribed on stone
texts, and written in ink on ostraca.
As the Middle Persian period progressed, the script and numerals tended to be
written increasingly cursively, with signs ligatured together. The Middle Persian
Empire came to an abrupt end in ad 637 after the child-king Yezdigird III was
overthrown by the Islamic Umayyad caliphs. By that time, the Middle Persian
script had diverged into several variants, one of the more important of which
was Book Pahlavi. The Book Pahlavi numerals are sufficiently different from their
Middle Persian ancestor to warrant separate treatment.

Sogdian
The Sogdian language was an Iranian language closely related to Middle Persian
but spoken further to the north, in modern Uzbekistan and Tajikistan. Sogdian
was written using three separate scripts: the Sogdian script descended from Middle Persian, the Manichaean script used by Sogdian followers of that religion and
descended from the Estrangelo Syriac script, and the Christian Sogdian script
used by Nestorian Christians and descended from Nestorian Syriac (Skjaerv
1996). This religious and scriptal pluralism greatly complicates the history of Central Asian Iranian scripts and numerals. The Sogdian script is first attested from
the Ancient Letters dating to ad 312313 found by Stein in Chinese Turkestan,
but may have originated in the third century. The Sogdian script proper and the
Manichaean script had distinct numerical notation systems of the basic Levantine
structure, which I will treat in turn, while the Christian Sogdian script used lexical numerals or Syriac alphabetic numerals (Sims-Williams, personal communication). There has been no systematic comparative treatment of Sogdian numerals
to date, and minimal paleographic work. The Sogdian numeral-signs are shown in
Table 3.12 (cf. Sundermann and Zieme 1981).
Table 3.12. Sogdian numerals
1

10

20

100

1000

697=

G\GGG\GGGHIIII\JGGG\GGG

88

Numerical Notation

The Sogdian system has numeral-signs for 1, 10, 20, and 100. It is cumulative-additive below 100 and multiplicative for the hundreds and thousands, with
numeral-phrases always written in descending order from right to left. The sign
for 1 is never used to write 1 alone but always phonetically as yw (Sims-Williams,
personal communication). Units from 2 to 9 are usually ligatured, although they
can be arranged in groupings of two to four units, or sometimes ungrouped. There
is no special sign for 5, in contrast to many of the Levantine systems (including Manichaean), but in common with Middle Persian, Kharoh, and Pahlavi.
While the sign for 10 is similar to the Sogdian letter that begins the word s ten,
the similarity is probably the result of later paleographic assimilation rather than
being indicative of an alphabetic origin for the sign. Paleographically, the sign for
20 was originally two superimposed signs for 10, and in the Ancient Letters 30
was occasionally expressed using three such signs (Sims-Williams, personal communication). By the seventh and eighth centuries the Sogdian letters dleth (d)
and ain () had become assimilated to the forms of the numerals 20 and 100,
respectively (Livshitz 1970: 259). The sign for 1000 is not a numerical-sign per
se, but, as in Aramaic and Middle Persian, an abbreviated ideographic form of
the Aramaic word LP thousand; similarly, 10,000 is written using an ideogram
RYPW (Aramaic ribb) (Sims-Williams, personal communication). Fractions are
poorly understood, although Grenet, Sims-Williams, and de la Vaissire (1998:
96) suggest that there is a sign for 1/2 that had previously been interpreted as a
variant for 100.
The majority of the texts in which this system was used are religious in nature
(the so-called Sogdian sutra script), although the Ancient Letters are personal
correspondence, and there are a few inscriptions on stone from Pakistan (Skjaerv
1996: 517). Numerals are used ordinally and cardinally in texts in various ways.
Sundermann and Zieme (1981) discuss some fragmentary lists of sequential numbers in the Sogdian-Turkish word lists used as translation glossaries, one of which
simply lists numerals from 88 to 100, and another of which contains undeciphered
(non-Sogdian) numerical symbols associated with the Turkish numeral words
one through five. Numeral-signs could be combined with lexical numerals, as
in the phrases 100 t 108 and wy 100 20 220 in the Padmacintmai-dhrastra dating to around ad 700 (Mackenzie 1976: 1217). Multiplicative powerideograms could also be combined together, as in the phrase 100 1LPW RYPW
100,000 myriads (= billions) in the Dhyna text (Mackenzie 1976: 7273).
Following the conversion of most of the Central Asian peoples to Islam, the
Sogdian script was used increasingly infrequently. In the eighth and ninth centuries Sogdian writing was adopted by Buddhist Uygurs, around which time the
script was rotated ninety degrees counterclockwise, resulting in vertical columns
(Coulmas 1996: 472473). By this time, however, it appears that no numerical

Levantine Systems

89

symbols were used (numbers were written lexically). It had largely fallen out of
use by the tenth century ad.

Manichaean
The development of a specifically Manichaean writing system is usually attributed
to Mani himself in the third century ad, although this tale is likely mythical, and,
the script may indeed be older than the religion (Skjaerv 1996). Manichaean
writing is derived from the Estrangelo variety of Syriac, and its numerical system
owes much to its ancestor. It was used to write a wide variety of languages, including Middle Persian, Sogdian, and Uygur. The best-known Manichaean texts are
those from the oasis of Turfan along the Silk Road, which date from the eighth
and ninth centuries ad. As with many of the Central Asian scripts, Manichaean
numerals remain understudied paleographically and comparatively; the signs
shown in Table 3.13 derive from the manuscript published by Mller (1912).
Manichaean numerals are always written from right to left with the highest
powers written first. The system is cumulative-additive below 100, with signs for
1, 5, 10, 20, and 100, while the hundreds are expressed multiplicatively. Units are
generally ligatured together, with long flourishes at the end (left) of the phrase.
Figure 3.3 depicts a section of a hymn book (Mahrnmag) in Manichaean script
dating to ad 76162; the year-number 546 (dated from Manis birth) is found
on lines 12 (5 100 20 20 5 1), while the last line contains the year-number 162
(100 20 20 20 1 1), reckoning from the death of Mari Schad Ormizd (Mller 1912:
36, Taf. II). The presence of a distinct sign for 5 very similar to that of Syriac, as
well as similarities in the sign for 20, suggest that Iranian and Central Asian scripts
such as Middle Persian and Sogdian played little role in shaping the Manichaean
numeral-signs. However, all of the Manichaean numeral-signs, with the exception
of the upright stroke for 1, are assimilated to letters of the Manichaean consonantal script (5 = ; 10 = h; 20 = p; 100 = m), so reconstructing the diachronic paleographic history of the system is complex (Sims-Williams, personal communication). It is unclear whether Manichaean had any signs for 1000 or higher powers.
While the Manichaean religion flourished for several centuries after its peak,
texts in the Manichaean script became less numerous after the tenth century, by
which time it seems to have acquired a dignified and prestigious but also arcane
quality (Sundermann 1997). The Manichaean numerals do not appear to have left
any descendant systems.

Pahlavi
Following the Islamic conquest, the Arabic script was normally used for writing
the Middle Persian language. The Zoroastrian Persians, however, continued to use

Numerical Notation

90

Table 3.13. Manichaean numerals


1

10

20

100

BA
C
D
697 = BACDEEEEFAC

their own Aramaic-derived script for their religious texts and for other purposes.
No Persian texts on papyrus survive from the early Middle Persian period, but late
in the Middle Persian period, and following the Islamic conquest, Persian began
to be written using a cursive, highly ligatured version of the earlier Pahlavi script,
known as Book Pahlavi. Alongside this script, a set of numerals was employed
(which I will call simply Pahlavi), shown in Table 3.14 (Abramian 1965: 285;
Mackenzie 1971: 145).
The Pahlavi system is decimal and written from right to left with the highest powers at the right. Frye (1973: 46) established conclusively that the Pahlavi
numeral-signs are cursively derived from those of the earlier Middle Persian system. The signs for 1 through 9 are ligatured and cursive reductions of unit-strokes,
and the Middle Persian practice of grouping unit-signs in groups of three and four

Figure 3.3. A portion of the Manichaean Mahrnmag of ad 76162,


with the numeral 546 spanning the
first two lines and 162 at the beginning (reading from right to left) of
the last line shown. Source: Mller
1912: Taf. II.

Levantine Systems

91

Table 3.14. Pahlavi numerals


1

a b c d bc cc cd dd ccc
10s
f j l m n
o
p
q
r
100s
s
1000s u
4697 = cdrsccud

1s

strokes can also be seen in the phrases for 5 through 9. Like the Middle Persian system, the Pahlavi system is multiplicative-additive for the hundreds and thousands.
Yet the Pahlavi numerals are structurally quite divergent from their ancestor. The
signs for the tens, in particular, show almost no trace of their cumulative ancestry,
and the unit-signs have largely become ligatured into single signs or, in the case of
5 through 9, into combinations of two or three signs. Combinations of tens and
units were usually ligatured together. The Pahlavi system is thus, for all intents,
ciphered-additive rather than cumulative-additive below 100. This transformation
from cumulation to ciphering occurred when the epigraphic Middle Persian script
and numerals, written mostly on stone and metal, were transferred to papyrus,
which is more amenable to cursive and ligatured writing. This transformation
is directly analogous to the derivation of Egyptian hieratic numerals from their
hieroglyphic ancestor (Chapter 2), corresponding to the switch in medium from
stone to papyrus. These two instances, in fact, are the only two known cases where
a cumulative system directly gave rise to a ciphered one.
The Zoroastrian Persians continued to use the Book Pahlavi script for their
religious writings and for new pieces of literature into the tenth century ad. Most
surviving Pahlavi numerals are found in these papyrus texts, although there are
also epigraphic texts on stone and metal. After about ad 1000, the Middle Persian
language underwent a set of changes that led to its transformation into Modern
Persian (Farsi). By this time, the abjad numerals (Chapter 5) and Arabic positional
numerals (Chapter 6) had completely replaced the Pahlavi system.

Summary
The Levantine phylogeny is descended from the Aramaic and Phoenician systems
developed around 750 bc, based on the dual model of the Egyptian hieroglyphic
system and the Assyro-Babylonian common system. Over the second half of the
first millennium bc, the Aramaic system and its descendants spread throughout

92

Numerical Notation

Assyria, Persia, Egypt, Asia Minor, and even into India and Central Asia. While
these systems were used for over a millennium, they ceased to be used once the
polities in which they predominated (most significantly, Achaemenid and Seleucid Persia) declined in importance. Levantine numerical notation systems steeply
declined in frequency of use after the rise of Islam, as the systems were replaced
by those of the alphabetic (Chapter 5) and South Asian (Chapter 6) phylogenies.
They persisted the longest in Central Asia, where Islam was somewhat slower to
take hold.
The central features of Levantine numerical notation systems are as follows:
a) a decimal base; b) a special sign for 20 (sometimes a combination of two 10signs); c) the use of vertical strokes for units and horizontal strokes (usually with
some degree of curvature) for tens; d) a cumulative-additive structure for numbers
smaller than 100; and e) the use of multiplicative-additive notation for expressing multiples of 100 (and also of 1000 and 10,000, where appropriate). Signs for
4 are found in Nabataean and Kharoh. The presence of a sign for 5 used in late
Aramaic, Palmyrene, Nabataean, Hatran, Old Syriac, and Manichaean helps to
clarify some of the relationships among the systems of the family. The late Pahlavi
system, which is heavily ligatured, is essentially ciphered-additive and thus somewhat anomalous, but it shares all the other structural features of this phylogeny,
and is clearly derivative of a cumulative-additive ancestor.
While these systems were used extensively for administrative and mercantile
purposes, as well as on inscriptions, there is no direct evidence that any Levantine
numerical notation system was ever used as a computational aid. We simply do
not know by what means the users of these script traditions performed arithmetic, but there is no reason to assume that it was done with pen and paper. There
are issues relating to the survival of perishable materials such as wooden tablets,
papyrus sheets, and leather scrolls, none of which survive well in the archaeological record. Yet, even in surviving texts, numbers were often written out lexically in
religious and literary contexts and even occasionally in economic documents. As
such, numerical notation occupied a less significant role in the script traditions of
these societies than would otherwise have been the case. Moreover, in comparison
to the incredibly wide diffusion of Aramaic-derived scripts throughout Europe,
the Middle East, and South Asia, Levantine numerical notations spread only sporadically, and their imprint was impermanent. The reason for this deserves careful
attention, and I will return to the question in Chapter 12, after looking at the history of these systems competitors.

chapter 4

Italic Systems

The Roman numerals are undoubtedly one of the better-known numerical notation
systems, and have received a tremendous amount of scholarly attention. Nevertheless, they constitute only a part of a larger phylogeny of numerical notation systems
that originated, not among Romans, but among Etruscans and Greeks on the Italian
peninsula around 600500 bc. The name Italic refers only to this geographical origin, and thus does not reflect any shared linguistic or cultural affiliation. Italic systems
flourished between 500 bc and 500 ad throughout the Mediterranean region, Western
Europe, and North Africa, under conditions of Greek and Roman cultural hegemony and political domination. Ironically enough, however, the collapse of the Roman
Empire brought about the greatest expansion of one particular system the Roman
numerals in medieval Europe, and ultimately throughout the modern Western world.
The most common variants of the Italic numeral-signs are shown in Table 4.1.

Etruscan
The Etruscans were a non-Indo-European people whose civilization had its center
in north central Italy, in the region of modern Tuscany (whose name is taken from
the Latin Tusci, meaning Etruscan). The origins and civilization of the Etruscans
are poorly understood, and large parts of their language remain undeciphered.
Yet Etruscan civilization was the most potent political force on the Italian peninsula between around 800 and 300 bc, and significantly influenced Roman culture
93

Numerical Notation

94

Table 4.1. Italic numerical notation systems


System

10

50

100 500 1000 5000 10K 50K 100K

Etruscan

1 Q R ; \ \

Greek archaic

Greek acrophonic

1 b c d e f g h

Greek Argos/Nemea

d\

Greek Epidaurus

Greek Olynthus

Lycian

1 < 1

Roman

1 P R S\ U W Y\ ,

. ~

Roman multiplicative

1 P R S U W a

d e g

Arabico-Hispanic

Calendar numerals

A E J

South Arabian

1 ! @ # $

i j

throughout the Republic and even later. The Etruscan alphabet, developed in the
early seventh century bc on the model of the archaic Euboean Greek alphabet,
usually runs from right to left (Bonfante 1996).
The Etruscan lexical numerals were probably base-10 with a special term for
20, zathrum (but not for 40, 60, 80 ...), and it appears that subtractive structures
formed the words for 17 through 19 (Lejeune 1981; Bonfante 1990: 22). However,
these irregularities are not reproduced in the Etruscan numerical notation system,
shown in Table 4.2.
Table 4.2. Etruscan numerals
1

10

50

100

500

1000

5000 10,000

1 Q R :\ \ ;\ VaU U /\
1378 = aaaQRR;;;

?\

Italic Systems

95

This system is cumulative-additive with a base of 10 and a sub-base of 5, very


much like the Roman numerals, but is most often written from right to left, with
the highest values at the right side of the numeral-phrase. Each power-sign of the
primary base (10) may be repeated up to four times, but the half-decade values
may occur only once in any numeral-phrase. The numeral-signs for 1, 5, 10, and 50
are quite regular throughout the systems history. The sign ; for 100, though rarer,
is found in many inscriptions from a relatively early date; the use of C is seen by
Keyser (1988: 542) as a development occurring between 250 and 200 bc. It may be
that C = 100 arose first in Latin inscriptions and found its way into Etruscan only
when the Etruscan system was already declining. The signs for 1000 and 10,000
are encountered only rarely (Keyser 1988, Bonfante 1990). The signs for 500 and
5000 are unattested, but because the numeral-signs 5 and 50 are the bottom halves
of the signs for 10 and 100, we would expect this same graphic principle to be
followed (Buonamici 1932: 244; Keyser 1988: 544545). This theory is given some
support in that the earliest Roman sign for 500 is X (later to become W). Furthermore, two Etruscan inscriptions contain an unidentified sign that might have
had such a value (Keyser 1988: 545). A special sign, , is used primarily on coins
to indicate (Bonfante 1990: 48).
While the Etruscan script is attested from around 700 bc, there is no evidence
of Etruscan numerals until the late sixth century bc. Their invention may have
been entirely independent of other base-10, cumulative-additive systems used
around the Mediterranean in the early first millennium bc. No earlier system used
a mixed base of 5 and 10, and most of the numeral-signs can be derived from successive crossings and circlings of tally marks for 1, 5, and 10. The most common
sign for 100 is simply 10 with a vertical line through it, while 50 is made by drawing a straight line from the apex of the upside-down V 5-sign (Keyser 1988: 533).
This theory is closely allied to Zangemeisters (1887) theory regarding the origin of
Roman numerals. Tallying practices in which numbers were marked sequentially,
then crossed off as appropriate, could thus have led to a numerical notation system. However, because tally sticks are normally wooden, no evidence survives that
the Etruscans ever used tallies in such a manner.
Alternatively, the Etruscan numerals may be descended from the Mycenaean
Linear B system (Chapter 2). Peruzzi (1980) argues that Mycenaean Greece
influenced some aspects of Etruscan culture but does not discuss the numerical evidence. Mycenaean settlements have been found in southern Italy and
Sicily, though these are too early to have had much direct cultural contact
with the Etruscans. Keyser (1988: 542543) notes that the Aegean systems, like
Etruscan, use strokes to represent the lower powers of the base and strokes in
conjunction with circles for higher powers. Unfortunately for this theory, 100
is ; in Etruscan but in Linear B. The similarity between Etruscan and

96

Numerical Notation

Linear B for 10,000 is notable, but it is the only numeral-sign to be relatively


close in form and value in both systems, other than their historically meaningless use of a vertical stroke for 1.
I think it likely that the Etruscan system arose relatively independently of other
systems, but with some continuity or influence from Linear B numerals. Base-10,
cumulative-additive systems abounded in the Mediterranean between 1100 and
650 bc the Egyptian hieroglyphic system, the Aramaic and Phoenician systems,
the Hittite hieroglyphic numerals, the Cypriote numerals, and any remnants of
the Linear B numerals. Regardless, the invention of a mixed base of 5 and 10 is
an important development, and the use of halved signs for the sub-base of 5 is an
ingenious means of deriving sign-values, suggesting that whatever system(s) the
Etruscans knew, their numeral-signs are of their own invention.
The early history of the Etruscan numerals is, in essence, a shared history with
that of the Greek acrophonic numerals to be described later. While the numeralsigns of the mature systems are quite different, they are structurally identical. The
ancestral role of the Greek scripts with respect to Etruscan is now very widely
accepted, and many other aspects of Etruscan culture owed much to contact with
Greek traders. Many early Greek numerals are found in the late sixth century bc in
south Italy in the context of contact with the Etruscans (Johnston 1975: 362364;
Johnston 1979: 31). Yet it is impossible to assign chronological priority to one or
the other. Instead, we might recognize a single ancestral system in the late sixth
century bc, which later diverged into Greek and Etruscan variants. The Etruscan
system is, however, the direct ancestor of the Roman numerals (Rix 1969, Keyser
1988). Etruria was politically dominant over Rome throughout its early history
and remained a potent force in Roman culture well into the Republican period.
Similarly, several Indo-European languages of the Italian peninsula, including
Oscan, Umbrian, and Faliscan, adopted scripts and numerical notation systems
based on an Etruscan model; their numerals are essentially identical to the Etruscan numerals.
Etruscan numerals were used in a wide variety of contexts. Out of nearly 10,000
Etruscan inscriptions known to Kharsekin (1967), about 200 contain numerical
notation, while only 40 contain the (as yet poorly understood) Etruscan lexical
numerals. Numeral-signs were often used on funerary inscriptions to indicate the
age of the deceased; other inscriptions on stone make use of numerical notation,
but more rarely. From the fifth century bc onward, Etruscan coins were stamped
with the numeral-signs for , 1, 5, 10, 50, and 100 in various combinations. As
well, many graffiti inscribed on potsherds contain Etruscan numerals. These graffiti often recorded the quantity or value of goods in containers. A lead tablet
whose purpose has not been reliably established contains the numeral-signs for
1000 and 10,000 (Keyser 1988: 544, Fig. 9).

Italic Systems

97

Figure 4.1. The Etruscan abacus-gem (CII 2578 ter) showing a figure seated at a board
working with Etruscan numerals. Source: Fabretti 1867: 224.

Figure 4.1 depicts the so-called Etruscan cameo or abacus-gem (CII 2578 ter),
a small gem (1.5 cm high) dating from the fifth century bc that depicts a seated
individual working at a large board upon which rows of Etruscan numerals have
been inscribed, including the elusive signs for 1000 and 10,000, but not 500 or
5000 (Fabretti 1867: 224; Keyser 1988: 545). This demonstrates the association of
the numerals with pebble-board computation at an early date. Similarly, Etruscan
numerals may have been used on wooden tallies and similar perishable materials,
despite the lack of evidence for such a function. There is no evidence, however,
for the use of the Etruscan system for performing arithmetical calculations as we
would (on papyrus or slate). Computations would have been done in the head,
with the fingers, or on a counting board. Large numbers and long numeral-phrases
are very rarely encountered; even the sign for 100 is relatively uncommon.
The demise of the Etruscan numerical notation system was a direct consequence
of the rising fortunes of the Roman republic. The lack of Etruscan political unity
in the third century bc, coupled with the political advantages of association with
Rome, led to the slow but steady assimilation of the cities of Etruria into the
Roman political and cultural milieu. While the Etruscans remained a culturally

Numerical Notation

98

Table 4.3. Post-Etruscan tally numerals


1 5

10

50

100

500

1000

1 Q P\U R O A ; Z
distinct people at least until the beginning of the Roman Empire, by 100 bc they
were entirely within the Roman political sphere. This inevitable trend was accompanied by the slow replacement of the Etruscan language, script, and numerical
notation with those of the Romans. Given the similarity of the two numerical
notation systems, there would have been little difficulty in making the change to
the new system. The last certainly dated examples of Etruscan numerical notation
are from the second century bc.
However, A. P. Ninni (188889) first presented the theory that the Etruscan
numerals survived into the nineteenth century. While studying the tally marks
used by fishers along the coast of the Adriatic Sea, near the town of Chioggia (near
Venice), Ninni discovered a numerical notation system that he called cifre chioggiotte (numerals of Chioggia). This potential vestige is cumulative-additive, with a
mixed base of 5 and 10, like both the Roman and Etruscan systems. Its numeralsigns are shown in Table 4.3 (Ninni 188889: 680).
Ninni noted that the signs of the cifre chioggiotte more closely resemble the
Etruscan numerals than the Roman numerals, and on this basis proposed that
they were of ancient origin. Furthermore, in both the Etruscan system and the cifre
chioggiotte, several of the sub-base numeral-signs are halved versions of the signs for
powers of ten (Ninni 188889: 680681). Could this system in fact be a survival,
over 2,000 years, of an Etruscan tradition among modern fishers? At present, there
is simply not enough surviving data to speculate on the possibility of such longterm cultural survivals, particularly in a region, such as Italy, that has experienced
immense social change over two millennia. On the one hand, certain signs of the
cifre chioggiotte (e.g., the first signs in Table 4.3 for 50 and 100) are identical to the
Etruscan numeral-signs for those numbers, but are quite dissimilar to the intervening Roman numerals. On the other hand, Chioggia is not in modern Tuscany,
and there is no evidence that the systems users believed the cifre chioggiotte to be
ancient. Since no information has come to light for over a century, perhaps we have
lost our opportunity to learn more about this system.

Greek Acrophonic
Between 750 and 500 bc, what we now call archaic Greece was a conglomeration
of small city-states in mainland Greece, the Aegean islands (including Crete), the
southern half of the Italic peninsula (known as Magna Graecia), and western Asia

Italic Systems

99

Table 4.4. Greek acrophonic numerals


1 5

10

50 100

500 1000

5000 10,000

50,000

1 b

d e

f g

36,849 = iiihgfeeeccccbaaaa

Minor, sharing in common only the use of Greek dialects. A tremendous number
of local scripts, known as epichoric scripts (from Greek epi-, upon, over, and chora,
place, country), were used during this period, all of which were based on the
model of the Phoenician consonantary around 800 bc. In their earliest phases,
some of these alphabets were written from right to left or in alternating directions (boustrophedon), although by around 500 bc all the epichoric scripts were
written from left to right. Adjoining these scripts were two very distinct types
of numerical notation: the acrophonic, to be described here, and the cipheredadditive alphabetic numerals (Chapter 5). For our present state of knowledge of
these two systems, we are greatly indebted to the tireless and unparalleled work of
Marcus Niebuhr Tod.1 The acrophonic system as used in classical Athens is shown
in Table 4.4 (Tod 191112: 100101).
The system is cumulative-additive, uses vertical strokes for units, has a base of
10 with a sub-base of 5, and is always written from left to right, with numeralphrases in descending order of numeral-sign value. The acrophonic system is so
named because the signs for many numbers are taken from the first letter (akros =
highest, outermost; phone = sound) of the corresponding (classical) Greek word.
Other names for this system, now largely rejected, include Herodianic and decimal (Tod 191112: 125127). The signs for 50, 500, 5000, and 50,000 combine
the sign for 5 with the sign for the appropriate power of 10. Whether we choose
to see these sub-base signs as single signs or as two ligatured multiplicative ones
is largely a matter of definition, and does not substantially affect how we classify
the entire system.
Similar acrophonic signs were used in large portions of the Hellenic world, the
only difference being that the appropriate letters from each epichoric script were
used in place of the letters used in the Attic inscriptions. Dow (1952) notes that
the variety of acrophonic Greek numerical notation systems stands in sharp contrast to the Greek alphabetic system (Chapter 5), which is remarkably consistent
1

Tods six papers on Greek numerical notation (Tod 191112, 1913, 192627, 193637,
1950, 1954) have been reprinted in one volume (Tod 1979). My citations are taken from
the original papers.

Numerical Notation

100

Table 4.5. Non-acrophonic archaic Greek numerals


1

10

50

100

throughout its geographic and temporal range. This degree of variation among
local systems is far greater than the variety of lexical numerals used in the Greek
dialects. However, the differences in sign-forms were probably not great enough
to affect their comprehensibility (Tod 193637: 246).
For expressing monetary values, the acrophonic numerals were often modified to
reflect the forms of currency being expressed; for example, in Attica, (talanton = 1
talent = 6,000 drachmas), Z (mna = 1 mina = 100 drachmas), (1 stater), (1 drachma),
I (1 obol), (1/2 obol), or (1/4 obol), and R (1/8 obol) (Threatte 1980: 111). These
could sometimes be ligatured to the sign for 5, just as the ordinary acrophonic powers
of 10, to express multiples of units of currency. While there is some potential for confusion ( can mean 1 talent or obol; Z can mean 1 mina or the numeral 10,000, etc.),
numeral-signs are always listed in descending order, which averts most ambiguities. In
some regions, special signs were used to indicate monetary values that did not fit easily
into the standard system. For instance, a system found in inscriptions from Thespiae
(in Boeotia) uses numeral-signs for 30 and 300, which consist of a sign (for triobole,
or 3 obols) ligatured to the appropriate sign for 10 or 100 (Tod 191112: 109; Feyel 1937).
Other acrophonic subsystems used cumulative signs related to systems of weight or
volume, such as those described by Lawall (2000) on graffiti from the Athenian Agora
from the last quarter of the fifth century bc; for example, EEEE = 4 hemichoes.
Despite the name of the system, not all numeral-signs used in the Greek epichoric scripts are acrophonic, and in fact the earliest ones are nonacrophonic.
Johnston (1975, 1979, 1982) has found several instances of a very early Greek
cumulative-additive but nonacrophonic system with a mixed base of 5 and 10
dating from the sixth and fifth centuries bc. The signs of the system are shown
in Table 4.5 (cf. Johnston 1979: 2930; Johnston 1982: 208). Johnston argues that
this system was built up systematically by cumulatively adding oblique lines to a
vertical stroke to obtain higher numeral-signs. Curiously, he does not note that
the signs for the sub-base (5 and 50) are the right halves of the appropriate primary
bases (10 and 100). This structure parallels the halving of Etruscan numeral-signs,
which is notable because many of the examples of this pre-acrophonic system
are of South Italian provenance. Johnston (2006: 17) notes several sixth-century
Greek inscribed vases where X = 10, paralleling the Etruscan practice but in contrast with later Greek acrophonic practice.
A very unusual numerical notation system used only to express monetary values
is found in five fourth-century bc inscriptions from the Greek colony of Cyrene

Italic Systems

101

Table 4.6. Cyrenaic numerals


20,000 10,000 5000 1000 500 100

20

b 2 R

1/5 1/10 1/50

Z 2 c

(in modern Libya). These numeral-signs are nonacrophonic, and their interpretation is controversial (Tod 192627, Oliverio 1933, Tod 193637, Gasperini 1986).
Our best evidence comes from the temple of Demeter at Cyrene, where inscriptions list the prices of various goods and the temples revenues and expenditures
(Tod 193637: 255). They present a dual series of figures in which each numeralsign has both a higher and lower value; the specific amount must be inferred from
the context within the numeral-phrase. Normally the higher is 5000 times the
value of the lower, but this breaks down for some of the lower signs. The relative
values of different units of currency used in Cyrene during this period (drachmas,
staters, minas, and talents, where 1 talent = 50 minas = 1250 staters = 5000 drachmas) help explain its unusual structure. The interpretation presented by Oliverio,
Tod, and Gasperini is derived from an analysis of the maximum number of times
each sign is repeated (and is thus open to question if more inscriptions are found).
This system is shown in Table 4.6.2
Still another aberrant acrophonic system is found in fourth-century bc inscriptions from Olynthus (in the northern Chalcidice region). There, a system was used
that is nonacrophonic and lacks a sub-base of 5 (Tod 193637: 248249; Graham
1969). The signs for 10, 100, and 1000 (R, , and , respectively) are the last three
letters of the western Greek alphabet used in the region. Of course, R = 10 is common to the Roman and Etruscan systems as well.3 On this basis, Graham (1969:
356) argues that the Roman/Etruscan system was borrowed from the Chalcidian colony at Cumae (in southern Italy). This theory, while attractive, has several
flaws, many of which derive from Mommsens (1965 [1909]) flawed lost-letter
theory of the Roman numerals discussed later. Moreover, the fourth-century bc
numeral-signs of Olynthus cannot have spread to the sixth-century bc Etruscans
by means of a colony at Cumae that never used the numeral-signs in question.
I suspect that the Greek letters were borrowed for the higher powers, just as the
Romans began with nonalphabetic numeral-signs, but later modified their signs
into alphabetic ones for mnemonic purposes.
2

The numbers listed are amounts in drachmas, based on the assumption that the lower
Z sign represents one drachma, without which the absolute value of each sign would be
indeterminate.
No significance should be attributed to the fact that the sign _, a common Roman
numeral-sign for 1000, is rotated ninety degrees from the Olynthian sign for 100.

Numerical Notation

102

Table 4.7. Epichoric Greek numerals


System

10

Standard Acrophonic

1
1

c
d
None
g
2 None
\\ d\\b

Olynthus
Epidaurus
Argos and Nemea

None
None
None

50

100

500

1000

g
g

None
None
None

A similar system was used in Epidaurus, on the southern Greek mainland (Tod
191112: 103104). It is acrophonic for 100 and 1000 but not for the lower powers.
Nearby, in Argos and Nemea, a closely related system was used that apparently
had a sign for 50, but not for 5 (Tod 191112: 102103; Ifrah 1985: 235). The systems
of Epidaurus and Argos, alone among the Italic numerical notation systems, use
a dot rather than a vertical stroke for 1. Table 4.7 compares the numeral-signs of
these irregular systems to the standard acrophonic system.
Most scholars explicitly or implicitly assume that the acrophonic system was
invented independently of the Roman, Phoenician, and other systems used at the
time (e.g., Ste. Croix 1956: 52). Because of the use of the acrophonic principle,
the numeral-signs are often Greek letters, which makes reconstructing the history
of the system rather difficult. It could be argued that the acrophonic nature of
the system suggests that it could only have been invented in Greece. Yet like the
Roman numerals, the earliest Greek acrophonic numerals are not phonetic signs
at all, which provides crucial evidence allowing us to reconstruct their origin.
While the traditional and widely quoted dates given for the use of the acrophonic system in Athens are 454 to 95 bc (Heath 1921: 30), there is indisputable
evidence of an earlier origin for the system. Tod argues, solely on logical grounds,
that a seventh-century bc origin is not unreasonable, as the system was fully developed by the middle of the fifth century bc (Tod 191112: 128). Mabel Lang mentions a seventh-century bc decorated Greek amphora inscribed with three vertical strokes, but this does not prove that the numeral was part of the acrophonic
system; it might have been part of an unstructured tallying system or almost any
numerical notation system in use in the Aegean at the time (Lang 1956: 3). For the
second half of the sixth century bc, however, there is more promising evidence
of the acrophonic system. Johnston (1979: 2729) discusses three different variations of the pre-acrophonic system mentioned earlier, used in the sixth century
bc in southern Italy, Sicily, western Asia Minor, the Aegean islands, and various
parts of mainland Greece almost the entirety of Greek civilization during that
period. Several vases from southern Italy and Sicily, which Johnston dates to the
last quarter of the sixth century bc, bear marks used in commercial transactions

Italic Systems

103

( Johnston 1975, 1979, 1982). It is telling that so many variations of the acrophonic
system were used in the fifth and fourth centuries bc, suggesting an initial period
of experimentation followed by consolidation and agreement on a single form of
the numerals.
The Greek acrophonic numerals likely originated on the Italian peninsula
around 575550 bc, around the same time and in similar contexts as the Etruscan
system. As I accept Keysers (1988) contention that the Etruscan numerals developed relatively independently as an outgrowth of tally marks, the obvious conclusion is that the Greek system developed on the model of the Etruscan numerals
in southern Italy and Sicily, an area of considerable Greco-Etruscan commercial
and cultural contact. It is difficult to believe that two cumulative-additive, quinary/
decimal numerical notation systems developed on the Italian peninsula in the
second half of the sixth century bc independently of one another. Yet because
the Etruscans owe their script to contact with the Greeks, it is counterintuitive to
think of the transmission of numerals moving in the opposite direction. In any
event, separating out questions of chronological priority of the two systems is virtually impossible. There may also have been some influence from the Phoenicians,
who were in contact with both the Greeks and the Etruscans in the sixth century
bc. In at least one document, tablet V from Entella in west central Sicily, the early
acrophonic numerals for 10, 50, and 100 were written with the smallest numbers
on the left and ascending to the right perhaps in emulation of the right-to-left
direction of the Phoenician system (Nenci 1995). The Phoenician system, however,
has a special sign for 20, is a hybrid multiplicative system above 100, and does not
have a sign for 50 at all.
In the early classical period, acrophonic numerals were used in Asia Minor,
the Aegean islands, North Africa, southern Italy, and Sicily, in addition to mainland Greece, but their spread to the non-Greek world was relatively limited. The
Lycians of southern Asia Minor used a nonacrophonic numerical notation system
in the late fifth and fourth centuries bc that is probably an epichoric variant of the
acrophonic system, although their language was not Greek (see the following discussion). The enormous cultural debt of Lycia to classical Greece is beyond doubt,
and its geographic and temporal proximity strengthens this hypothesis. Also likely
is the possibility that the South Arabian numerals, which arose in the fifth century
bc, derive from the acrophonic system. The South Arabian numerals are cumulativeadditive, base-10 with a sub-base of 5, and use acrophonic numeral-signs. However, more evidence of cultural contact is desirable before this hypothesis can be
proven.
Acrophonic numerals are found on inscriptions on stone, lead, and silver as
well as on potsherds; they may also have been used on wood or other perishable
materials, though evidence is lacking. Of the thousands of Greek papyri from

104

Numerical Notation

the fourth century bc onward, only a handful from Saqqara contain acrophonic
numerals (Turner 1975). Inscriptions on stone use acrophonic numerals far more
frequently, including accounts, inventories, lists, regulations, treaties, and boundary markers. As well, graffiti or other marks on pottery often indicate quantities
for commercial purposes. The acrophonic numerals expressed measures of volume
or distance, quantities of goods, or monetary values.
What is notable is the wide range of purposes for which acrophonic numerals
were not used, even compared to other cumulative-additive systems used in the
Mediterranean in antiquity. Firstly, the numerals could only be used to express
cardinal numbers; ordinal numbers were expressed using lexical numerals or,
when available, alphabetic numerals (Tod 1911: 128). Similarly, with the exception of monetary amounts, there was no acrophonic numeral expression for fractions. The Greeks never expressed dates in acrophonic numerals. The practice of
expressing the age of the deceased at death on funerary inscriptions, a source of
much information on other numerical notation systems, was not customary in
Greece. The custom of dating using regnal years did not arise until the Alexandrine period. Documents in connected prose (decrees, for instance) rarely contain acrophonic numerals, except to indicate the price of executing the inscription (Threatte 1980: 112).
There is no evidence that the acrophonic numerals were used direcly for arithmetic or accounting. For these purposes, as with the Roman and Etruscan systems,
the Greek acrophonic system was complemented by the use of the pebble-board
abacus, in which several grooves were labeled with the appropriate acrophonic
numerals. Lang (1957) has established that many of the mathematical errors made
by Herodotos demonstrate his use of the abacus to perform calculations, with
which certain types of errors (especially in multiplication and division) can occur
easily. All of the thirteen examples of abaci (and fragments thereof ) known from
classical Greece have the row values inscribed with acrophonic numerals (Lang
1957: 275276). Most notable among these abaci is the remarkably well-preserved
Salamis tablet, which probably dates from the fifth century bc (Menninger
1969: 299303). The numerals on it range from T (one talent) to X (1/8 obol); the
monetary values of the numeral-signs suggest that it was used for practical commercial computations.
The decline of the acrophonic system is thoroughly entwined with the fate of
the Athenian state as a Greek power. As Athens ceased to be a dominant power
in Mediterranean affairs, acrophonic numerals were used less often; by the third
century bc, they had been supplanted by the alphabetic numerals for most purposes throughout large parts of the Hellenistic world, including Ptolemaic Egypt
and Seleucid Persia. Only in Athens and the surrounding areas did the acrophonic
system continue to flourish. There are only a handful of known first-century bc

Italic Systems

105

examples from Athens (Threatte 1980: 113). By this time, Greece was firmly under
Roman control. Yet there is no evidence that the acrophonic system was replaced
by Roman numerals except, as one might expect, in southern Italy, where Latinspeaking populations dominated. However, the use of acrophonic numerals did
continue in one very limited domain stichometry, or the enumeration of lines of
verse in classical texts (Tod 191112: 129130). This practice continued as late as the
third century ad with the writings of the Neoplatonist philosopher Iamblichus.
Such late examples are analogous to the use of Roman numerals in the modern
West, in contexts in which it is very useful to have two separate numerical notation systems for paginating introductory sections versus the body of a work, or
for distinguishing volume numbers from page numbers in certain texts.

Lycian
Lycia was a small state of southern Asia Minor in the middle of the first millennium bc, centered around the city of Xanthus. The Lycians spoke an incompletely
understood Indo-European language related to the earlier Luwian language, which
was spoken in the Neo-Hittite kingdoms of Asia Minor. Lycia occupied an intermediate position between the Greek and Persian spheres of influence, and was
intimately involved in interregional commerce and conflict. The Lycian alphabet,
which was developed around 500 bc, is an epichoric variant of the Greek script,
like many others used in the Greek peninsula and western Asia Minor, except that
the language of the inscriptions was not a Greek dialect. A few hundred instances
of the Lycian script have survived, mostly from inscriptions on stone and on coins;
they are written almost exclusively from left to right and date from the fifth and
fourth centuries bc.
The Lycian numerical notation system is still very poorly understood. The
Lycian system, like the Greek acrophonic, Etruscan, and Roman systems, is cumulative-additive with a base of 10 and a sub-base of 5. However, the exact values of the
numeral-signs are still in debate. The signs of this system are shown in Table 4.8
(cf. Shafer 1950, Bryce 1976).
There is also a sign, 2, that probably represents , although Shafer (1950: 260)
argues that it may represent an additional one-half of any numeral-sign that precedes it; 2 would be 15 and <2 would be 7 according to this theory. The
numeral-signs for 50 and 100 are found only on a few inscriptions. The value 50
is assigned to primarily by default; its value is certainly between 10 and 100 (it
is found after 1 but before ). I follow Frei (1976: 15) in assigning it the value of
50. We can be fairly certain about the value of the sign for 100, because it is found
in the Lycian portion of a trilingual Greek-Lycian-Aramaic inscription found at
Letoon and dating from 358 bc (Frei 1976: 1315). Shafer (1950: 258259) suggests,

Numerical Notation

106
Table 4.8. Lycian numerals
1

<
127 = 1<11

10

50

100

based on one inscription, that the Lycians may have used the subtractive principle
to express the number 4 as a<. However, in other inscriptions, 4 is expressed as aaaa.
The Lycian numerals are very likely a previously unidentified variant of the
Greek cumulative-additive systems. Lycian numerals arise in the early fifth century bc at the time of the peak use of the Greek acrophonic numerals in Athens
and throughout the Hellenic world. Both systems were cumulative-additive, had
a base of 10 with a sub-base of 5, and were used in the fifth century bc in the
Aegean region. While the Lycian numerals are not acrophonic, this is true of many
of the epichoric numerical notation systems of the classical Greek world. Alternatively, the Lycian system may have been based on an Aramaic model, with the
sign for 100 (1) being in fact multiplicative (1 100) rather than constituting a
single numeral-sign (Frei 1976, 1977). No numbers higher than 120 are expressed
in any Lycian inscriptions, so we do not know whether, for instance, the Lycian
numeral-phrase for 200 was additive (1\ 1) or multiplicative (11). Because none
of the Semitic systems had separate signs for 50, but all of them had signs for 20,
we would need to modify the value of the Lycian to 20, which is consistent with
the numeral-phrases known from inscriptions. There is little similarity, however,
between the numeral-signs of Lycian and either Phoenician or Aramaic, and the
Lycian system uses a sign for 5, which is very rare in Aramaic. As well, Lycian,
like the acrophonic numerals, is written from left to right, whereas the Levantine
systems are all written from right to left.
The Lycian numerical notation system apparently did not diffuse outside Lycia.
Although Shafer (1950) argues that the similarities between the Roman and Lycian
numerals are sufficient to indicate the derivation of the former from the latter, this
likeness is no greater than that between Lycian and the Greek acrophonic system.
Furthermore, while there is some similarity between the Lycian numeral-signs and
other systems of the Italic phylogeny (especially South Arabian), these similarities do not correspond to any plausible circumstances of cultural contact. In Asia
Minor, scripts such as Phrygian and Lydian, both of which are closely related to
Lycian and were used in the fifth and fourth centuries bc, used numerical notation
based on the Phoenician-Aramaic model rather than on the Greek or Lycian.
Lycian numerals are found primarily in a single context sepulchral epitaphs
indicating monetary amounts, normally including a numeral-phrase preceded
by the word ada, now considered to be a monetary unit (Bryce 1976: 175). The

Italic Systems

107

monetary values may have stipulated a penalty to be paid should the tomb in
question be violated (Shafer 1950), or they may indicate fees paid in advance by
the family for a tomb site (Bryce 1976). The only nonfunereal context in which
Lycian numerals are used is the trilingual inscription found at Letoon, a public
legal regulation (Frei 1976). As Lycian numerals are not found on coins or on
financial inscriptions, they are quite distinct from the numerals of the rest of Asia
Minor and the Aegean.
As the Lycians became increasingly caught up in imperial conflicts between the
Persians and the Greeks, their script was used increasingly infrequently. By 300 bc,
the Lycian script had assimilated to the Greek, and its numerical notation ceased
to be used, replaced by the Greek alphabetic numerals.

South Arabian
The Old South Arabian scripts are of a very ancient origin, first appearing around
the turn of the first millennium bc in the southern part of the Arabian peninsula
(modern Yemen). They are consonantal and are characterized by large, well-formed
letters and by their extremely varied direction of writing (left-to-right, right-to-left,
or boustrophedon, depending on the inscription). They were used to write South
Semitic languages such as Minaean, Sabaean, Qatabanian, and Hadramauti. During their early history, these scripts did not possess any numerical notation system,
but following the rise of the kingdoms of Minaea and Saba in the fifth century bc,
numerical notation began to be used in monumental inscriptions. The numeralsigns used are shown in Table 4.9, including both left-to-right and right-to-left sign
forms, where appropriate (Halvy 1873: 511512; Hommel 1893: 8).
The system is cumulative-additive, with a base of 10 and a sub-base of 5, and is
written in whichever script direction is used in the inscription as a whole. The sign
for 1 is, as in all Italic systems, purely iconic. The signs for 5, 10, 100, and 1000,
however, are acrophonic; each is simply the first letter of the appropriate South
Arabian lexical numeral (Beeston 1984: 8). The sign for 50 is non-acrophonic, but
is simply a halved version of the sign for 100. In one inscription (Biella 1982: 531),
the sign X is apparently used with the numerical value 4, possibly in imitation
of the Nabataean system (Chapter 3). There are no signs for 500 or 5000 known
from any South Arabian inscriptions; these numbers were written with five signs
for 100 and 1000, respectively (Biella 1982: 1, 265). Normally, numeral-phrases
were placed between large hatched bars to avoid confusing numeral-phrases with
words, given the use of the acrophonic principle (Halvy 1875: 78). Large sets of
unit-signs were not divided into smaller groups, which creates a legibility issue
because the sub-base of 5 is not used throughout the system; one inscription from
Sirwah denotes 12,000 using twelve signs for 1000 (Ifrah 1998: 187).

Numerical Notation

108

Table 4.9. South Arabian numerals


1

10

50

1
!
@
#
Right to left 1

3697 = \%%%$$$$$$#@@@@!11\ (L-R)


\11@@@@\ (R-L)

Left to right

100

1000

In some inscriptions, the South Arabian numerals used an unusual technique


of implied multiplication that resembles positional notation. Most often, this was
done when expressing values greater than 10,000, by placing signs to the left of
a sign for 1000 (when reading from left to right), which were implicitly taken to
represent multiples of 1000. For instance, one inscription (R3943/2) has \@@@%\
for 31,000, in which the three @ signs have the value of 10,000 instead of 10, while
% retains its ordinary value of 1000 (Biella 1982: 349; Ifrah 1998: 187). Apparently
this technique was also sometimes used for multiples of 100; Halvy (1875: 79)
notes an inscription that has \111\ instead of \$$$\ for 300. We know that the
multiplied value is correct because of contextual information and because South
Arabian inscriptions commonly list the appropriate lexical numeral beside the
numeral-phrase. Ifrah sees in this use of implied multiplication what might be
called the germ of our place-value notation (Ifrah 1985: 232). However, without
contextual information, such numeral-phrases would simply be confusing and
ambiguous, as there is no sign for zero. These formations are very rarely attested
throughout the systems history.
The South Arabian system is derived from the Greek acrophonic system (Ifrah
1998: 186; Fevrier 1948: 579). Both systems are decimal and cumulative-additive,
and both have a sub-base of 5. Additionally, both systems use the acrophonic principle, a feature that is otherwise uncommon during this period. The numeral-signs
are not similar to those of any other system, but this is unsurprising, since the
system is acrophonic. In the fifth century bc, when the South Arabian numerals
were first used, the Minaeans and Sabaeans were actively engaged in trade with
the Greeks at a time when the acrophonic numerals were the only ones the Greeks
were using for monetary and metrological purposes. While one would expect the
South Arabian scripts to have numerical notation systems similar to those used in
North Semitic scripts at the time (Aramaic or Phoenician), this is not borne out
by comparing the systems. There is no sign for 20 in the South Arabian system,
while the Aramaic and Phoenician systems did not normally use signs for 5 and
50. There is no evidence that the South Arabian numerals ever diffused outside
the Arabian peninsula. The Geez script used for the Ethiopic languages, which is

Italic Systems

109

derived from a South Arabian model, used numerals based on the Greek alphabetic system (Chapter 5).
Although some South Arabian cursive inscriptions on wood have been found,
these contain no numerals. The system just described is documented only in monumental contexts. The functions of the numerals included recording details of
sacrifices or offerings to gods, quantities of booty obtained, numbers of military
troops, and information on construction projects such as monuments and irrigation systems. The South Arabians did not use an enumerated dating system, nor
do South Arabian coins contain numerical notation of any kind.
By the second century bc, instances of the South Arabian numerals were normally preceded by the appropriate lexical numeral written out in full. While this
aids modern scholars in their interpretation, doing so also removed any incentive
to continue to use the system. By the first century bc, although the South Arabian
scripts continued to be used, the numerical notation system had become extinct,
and it was not replaced until the seventh century ad, when the Islamic conquest
brought alphabetic and later positional numerals to southern Arabia.

Roman
Despite the importance and continued use of Roman numerals, the early history
of the system was very poorly understood until Keysers (1988) study. There are
several different classical variants of the Roman numerals, while the sign forms
and structure of the Roman numerals used today are medieval in origin. The
Roman alphabet was developed on an Etruscan model around 600 bc at a time
when much of Italy was under Etruscan political domination; it was written from
left to right, as it is today. Like its Etruscan precursor, the Roman numerical notation system has a base of 10, with a sub-base of 5, and is essentially cumulativeadditive; unlike it, the Roman numerals are written from left to right and are
sometimes used subtractively. The great variety of numeral-signs used throughout
two millennia of its history contrasts strongly with the highly static quality of the
equally long-lived Babylonian cuneiform and Egyptian hieroglyphic numerals.
Table 4.10 presents the numeral-signs used during the republican period (Ifrah
1985: 132; Keyser 1988).
These numerals were cumulative-additive in structure in most inscriptions.
Only the signs for 1 and 10 remain unchanged throughout the entire history of
the system. The inverted V sign for 5, Q, is found only in early contexts, and
is evidence of the systems indebtedness to its Etruscan ancestor, but P is used
exclusively thereafter. The signs for 50 in Table 4.10 are roughly in chronological
order; the inverted arrow forms are earliest, with the inverted T forms prevalent
until about ad 25, and L most common thereafter (Gordon and Gordon 1957: 181).

Numerical Notation

110

Table 4.10. Roman numerals (republican period)


1

10

50

100

Q
P\

\ V
U
T
\
S

500

X
W

1000 5000 10,000 50,000 100,000

,
Y\\
_
\

|
19,494 = YYYYUUUUTRRRRaaaa
19,494 = YUXRUaP

.\

~\

There is no sign for 100 in any early Roman inscription although there surely must
have been one to complete the series. The V form for 100 is extremely rare; Ifrah
(1998: 188) lists only a single inscription where it is found. Keyser indicates that
the first Roman C = 100 whose date is secure is from 186 bc, but he postulates
a third-century bc origin for the symbol as a reduction of the Etruscan ;, even
though there is no example of any sign for 100 at this early date (Keyser 1988:
542). The number 500 is expressed using X in all early contexts, with assimilation
to the alphabetic D occurring around the transition to empire. The familiar M =
1000 used from the Middle Ages to the present occurs only in one classical Latin
inscription as part of a numeral-phrase, although it is also found in various places
where M is simply an abbreviation for mille or milia (Gordon and Gordon 1957:
181182; Gordon 1983: 45). The signs for 5000, 10,000, 50,000, and 100,000 are
rarely encountered, though they are all attested as early as the third century bc.
Adding arcs on either side of the most common sign for 1000, Y, indicates successive powers of 10, while the right half of the appropriate base-10 sign represents
the quinary component.
In a very limited set of texts, is used to represent 500,000 (Mommsen 1965
[1909]: 788791; Gordon 1983: 45). The sign is probably derived from alphabetic
Q and is thus an abbreviation of quingenta milia. Its use was limited to the later
Republic, and it was certainly not familiar a century later to Pliny the Elder, who,
in his Natural History, wrote Non erat apud antiquos numerus ultra centum
milia or Among the ancients there was no numeral larger than 100,000 (Natural History 33.47.133). Alongside the Roman numerals, the Romans had a duodecimal fractional system based on the as of twelve unciae (Menninger 1969: 158162;
Cagnat 1964: 33).

Italic Systems

111

There was very little regional variation in the signs or the structure of the classical Roman numerals (due, no doubt, to the centralization of the empire). In
North African inscriptions, however, the signs for the sub-base were sometimes
not used, for example, IIIII for 5 and XXXXXX for 60 (Cagnat 1964: 3031).
Similarly, the signs for C and L were often written cursively, even in inscriptions
on stone, to distinguish them from the corresponding letters (Salama 1999). This
practice of cursively writing Roman numerals on stone was also attested in Spanish inscriptions of the fifth and sixth centuries ad, borrowing the numeral forms
used on much earlier Roman paypri written in Egypt (Mallon 1948).
Starting in the late republican period, the subtractive principle was occasionally
used for writing multiples of 4 or 9 (or rarely 8) of powers of 10. As would later
become the rule with modern Roman numerals, placing a lower-valued power of
10 to the left of a higher numeral-sign indicated subtraction of the former from
the latter (IX for 9, XL for 40, but never using sub-base signs as the subtrahend
i.e., VC is unacceptable for 95). This reduced the length of numeral-phrases four
or five numeral-signs were replaced by two. Similarly, in the Augustan period
and into the early empire, the use of the subtractive XIIX and XXIIX for 18 and
28 were common, and IIX and XXC for 8 and 80 are also attested (Gordon and
Gordon 1957: 176181). Addition was used almost exclusively in the early republican period, and is the more usual form even in later classical inscriptions (Sandys
1919: 5556).4 Subtractive numerals are more common where a numeral is at the
end of a line of an inscription, allowing the engraver to avoid crowding many
numeral-signs into a limited space (Cajori 1928: 31). They may be more common
in informal texts than in formal inscriptions (Gordon and Gordon 1957: 180181;
Cagnat 1964: 3031). Despite Guitels (1975: 202203) denigration of subtractive
notation because it lacks the simplicity of the pure additive principle, it is a very
economical way of structuring numeral-signs. The Latin lexical numerals use the
subtractive principle for 18 and 19 (duodeviginti, undeviginti), perhaps explaining
the origin of this practice. Note, however, that while duodeviginti and undeviginti
are subtractive, novem (VIIII/IX), quatuordecim (XIIII/XIV), nonaginta et novem
(LXXXXVIIII/XCIX), and all other Latin lexical numerals are not.
Around the same time, the multiplicative principle began to be employed to
write very large numbers. Even as early as the third century bc, the Roman republic had become a large centralized state, and the need to express large numbers was
acute, yet the highest numeral-sign was 100,000. At times, this led to extremely
cumbersome numeral-phrases, such as the inscription on the famed Columna
4

In the tradition of modern Roman numeral hour-numbers on clocks, 4 is normally


denoted additively (IIII), while 9 is denoted subtractively (IX), possibly because IIII
aesthetically balances the left and right sides of the clock face (Hering 1939: 319).

112

Numerical Notation

rostrata of the consul Gaius Duilius, originally erected in Rome in 260 bc, which
celebrated a naval victory over Carthage in which over two million aes worth of
loot was plundered. The column is inscribed with at least twenty-two signs for
100,000, and possibly as many as thirty-two, as the inscription is fragmentary
(Menninger 1969: 4344). Although it was recut, probably in the Augustan period
when other means of expressing this number were available, the original structure
of the expression was retained (Stenhouse 2005: 5960). One is struck, in looking
at this inscription, at the sheer enormity of the numeral-phrase, and thus by the
impressive amount of booty obtained, and in fact this may have had something to
do with the reason it was written at all. Gordon and Gordon (1957: 180181) suggest
that one of the factors working against the widespread acceptance of subtractive
notation was the desire of public officials to impress and indulge. This conspicuous consumption of numerals is cross-culturally frequent, for example, in the
Narmer macehead (Chapter 2) and other royal commemorative inscriptions.
Starting in the late republican period, a horizontal bar (vinculum or virgula)
placed above a numeral-phrase or some portion thereof indicated that the number
under the bar should be multiplied by 1000 (Smith 1926: 7678; Gordon and
Gordon 1957; Cagnat 1964: 3132; Gordon 1983). This principle was first used in
the Lex de Gallia Cisalpina written between 49 and 42 bc (Gordon 1983: 47). For
most numbers, doing so did not improve conciseness; both gedddda\SROOO and
....YSROOO for 191,063 require twelve symbols. The main advantage is
that one need no longer remember so many numeral-signs or invent new ones; the
signs for 1, 5, 10, 50, and 100 are sufficient to express any number up to 500,000,
whereas eleven different signs would be needed under the purely additive system.
One slight complexity is that starting around the same time, barred numerals
were often used to distinguish ordinal from cardinal numbers, as in abbreviations
such as aaVIR for duumvir (Gordon and Gordon 1957: 166176). In one irregular
inscription from Pompeii from the first century ad, the numeral-phrase LXXXX
apparently indicated 90,000 using a regular sign for 1000 multiplicatively, a practice not encountered again until the Middle Ages (Smith 1926: 7).
Starting around the early Imperial period, three vertical bars enclosing a
numeral-phrase on the top and sides signified multiplication by 100,000. Thus,
instead of the thirty-two signs for 100,000 found on the Columna rostrata, one
would need only to write mdddaan. This technique was first used in the late first
century bc, according to Gordon (1983: 47), but was employed rather sparsely
until the second century ad.5 In this way, any number less than 500 million could
5

According to Suetonius (Galba, 5), the emperor Tiberius willfully read the will of Livia
in such a way as to read a three-barred numeral min as i in order to reduce by a factor
of 100 the amount of inheritance owed to the future emperor, Galba (Cagnat 1964: 32).

Italic Systems

113

Table 4.11. Roman numerals (multiplicative)


1

Regular signs
Multiplicative
(1000)

10

50

100

500

1000

1000

5000

10,000

50,000

100,000

500,000

1,000,000

100,000 500,000 1,000,000 5,000,000 10,000,000 50,000,000 100,000,000

Multiplicative
(100,000)

man
mbn mdn
men
35,863,120 = mgggebaaan edaaa URR

mgn

mjn

mkn

be expressed with just the lowest numeral-signs plus two types of bar to express
multiplication. This revised system is still a decimal system with a sub-base of 5;
however, instead of being purely cumulative-additive, it is a hybrid system using
cumulative-additive structuring for numbers up to 1000 and multiplicativeadditive thereafter. The entire system (up to 100,000,000) as used in the Imperial
period is shown in Table 4.11.
Gordon (1983: 47) claims that the largest number expressed using this hybrid
cumulative and multiplicative system is 35,863,120, though an inscription at Ostia
from ad 36 apparently indicates 100 million as mkn (Menninger 1969: 245). Most of
the higher signs are attested only rarely. Curiously, Guitel (1975: 215) regards Roman
multiplicative notation as an evolutionary dead end, because, she argues, they no
longer needed to develop a more efficient positional system. The teleology of this
argument is immediately apparent, as it regards this development only with respect
to its failure to lead to a superior system. The Romans themselves likely perceived
it as a means of improving conciseness, while reducing the number of signs one
needed to memorize. Although the use of the 1000 bar continued among some
post-Roman scribes, the use of the 100,000 box did not outlast the empire.
While the origin of the Roman numerals is a common topic of inquiry, unfortunately, as Cajori (1928: 31) noted, the imagination of historians has been unusually active in this field. Fortunately, Keysers (1988) panoptic essay on the origin
of the Roman numerals, which examines a variety of theories, ranging from the
sixth-century theories of the grammarian Priscian to the twentieth-century theories of modern classicists, has firmly settled the issue. The popular belief that the
Roman numerals originated as alphabetic signs is false. While the modern Roman
numeral-signs are also letters, the signs I, V, and X mean 1, 5, and 10, rather than
U, Q, and D, for unus, quinque, and decem. C is the first letter of centum, but this
is coincidental, since C is a reduction of the older Etruscan sign ; or VaU (Keyser

Numerical Notation

114

Table 4.12. Etruscan and Roman numerals

Etruscan
Roman

10

50

100

1
1

Q
Q\P\

R
R

\:
U\\;
\\T U

500

1000

\
\X

\\\/
Y\\\

1988: 542). The signs for 50 and 500 were not associated with the letters L and D
until the late Republic, and M was not used for 1000 until the Middle Ages. It
is fortuitous that the older sign for 1000 (Y) could be easily transformed into an
M. Similarly, theories employing the pictographic principle (for instance, I = 1
from a single finger; V = 5 from an outstretched hand, and X = 10 from two hands
together) were proffered by many early modern antiquarians, and later by classicists such as Sandys (1919), but, while they are imaginative, there is no evidence
to support them. Until recently, the most widely accepted theory was that of
Theodor Mommsen (1965 [1909]), who argued that the signs for 50 (), 100 (U),
and 1000 (Y) were taken from letters of the Chalcidic Greek alphabet that were
not needed to transliterate the Latin language: chi, theta, and phi, respectively,
which somewhat resemble the numeral-signs. Unfortunately, this attractive theory
has several flaws: the sign for 100 does not really resemble the Chalcidic theta;
these lost letters were sometimes used in Etruscan and Roman inscriptions; and
special pleading is required to derive the origin of X = 500. While it was once
plausible, it is no longer a parsimonious theory.
In fact, the Roman numerals up to 1000 developed through direct diffusion
from the Etruscans. The astonishing similarity between the Etruscan and archaic
Roman numeral-signs, as shown in Table 4.12, ought to be enough to prove a relationship between the two systems. The Etruscan numerals have temporal priority
over the Roman numerals, which do not appear until well into the fifth century
bc, and are not frequently encountered until the third century bc. In fact, the
similarities between the two systems are so great that one could treat them as a
single numerical notation system; they are identically structured, and many of
their numeral-signs are similar or identical. I treat them separately because the two
systems are written in opposite directions and because the Roman system used
signs for much larger powers at an early date.
The Roman numerals were used in a broader range of contexts than any other
cumulative-additive system. In its earliest forms, they were used on coins, on pottery, and on inscriptions on stone. Dates, monetary values, and measures were
all frequently expressed in the Roman numerals. The Roman numerals could be
employed to express both cardinal and ordinal values. Their use in administration
and literature was widespread from the republican period onward. In texts, Roman
numerals were used to enumerate page and line numbers. Accounts, inventories,

Italic Systems

115

and legal documents also occasionally provide us evidence of their use in commercial and institutional contexts.
While Roman numerals certainly were used in the contexts of arithmetic and
calculation, there is minimal evidence that they were ever used for calculation.
Glautier (1972) discusses the Roman account records, which he characterizes as
primitive from an accounting standpoint because of the lack of positionality. A
similar point is raised by Meuret (1996) in his discussion of the Lamasba tablet, an irrigation regulation from North Africa during the reign of Elagabalus
(218222 ad), which contains a multiplication table to enable quick calculation
of water supplies, thus overcoming the computational deficiencies of the system.
Maher and Makowski (2001) demonstrate persuasively, however, that the assumption that Roman mathematics was poor because of Roman numerals cannot be
correct, given the complexity of the arithmetical calculations attested in Latin
texts. Nevertheless, their stronger claim that Roman mathematicians actually
used written numerals to do arithmetic, in particular for calculations involving
fractions remains unproven.
Roman numerals were used for computational functions if not directly for
computation, then certainly to mark the rows on the abacus. While few Roman
abaci survive, Taisbak claims that the Romans did all their calculations with them,
and even that the notation of Roman numerals originates from the abacus reckoning (Taisbak 1965: 158). This finding is contradicted by the derivation of the
Roman numerals from the Etruscan system and ultimately from an older tallying system. Because cumulative systems use one-to-one correspondence intraexponentially, just as one counter equals one multiple of a power on the abacus,
the Roman techniques of numeration (Roman numerals) and computation (the
abacus) complement one another. This correlation is confirmed by the quinary
(base-5) component of the abacus (there are rows not only for the powers of 10
but also for their halvings). No row on the abacus ever would have contained more
than four counters, which would have facilitated reading and working with them.
However, because the original Etruscan system probably emerged from a system
of tallying, it is more likely that the structure of the abacus emerged out of the
structure of the numerals than vice versa.
Despite the enormous influence of Roman civilization on Europe, North Africa,
and the Middle East, and despite the extraordinary chronological duration of the
Roman numerals (almost 2,500 years), they produced relatively few descendants.
While other systems in use over similar periods, such as the Brhm numerals and
Greek alphabetic numerals, changed their form greatly as they spread across time
and space, the Roman numerals of antiquity spread largely unmodified throughout Western Europe and other areas where the Roman alphabet was used. While
the Indian and Greek systems spread throughout many different scripts, changing

116

Numerical Notation

the forms of signs as they were transmitted, Roman numerals were infrequently
adopted by users of other scripts.
Of the few descendants of the Roman numerals, I have already discussed the
hybrid multiplicative-additive system used occasionally from the first century
bc onward. In the medieval period, Roman numerals were essentially the same
as classical ones, though with slight differences in form and structure. In Arabinfluenced Spain, certain variant Roman numeral systems were used starting in
the tenth century ad. Around the same time in northern Europe, certain types
of medieval calendars contained unusual Roman numerals. Finally, as the Roman
numerals came increasingly under assault from the rival Western system, certain
positional variants of the numerals were occasionally used, combining features of
both systems.

Computation with Roman Numerals


The computational efficiency of the Roman numerals is a common subject in the
history of mathematics. The consensus of these arguments, with which I am in
general agreement, is that the Roman numerals are poorly suited to performing
arithmetical calculations. This accords with the historical finding that people used
abaci or finger computation to actually manipulate numbers, with the Roman
numerals being used only to express the result. Yet the rejection of the efficiency
of the Roman numerals for computation has been used to draw two unwarranted
implications.
Firstly, it is unlikely that the use of the Roman or other cumulative-additive
numerals has any simple or unilinear correlation with developmental stages of
psychology, as Murray (1978), Hallpike (1979), and Dehaene (1997) have suggested, either as the cause or effect of a less abstract way of thinking about number.
Although many ancient numerical notation systems are cumulative-additive, this
is not evidence that the Roman number concept is less abstract than the modern
one. Other evidence, such as the use of abaci and finger reckoning, refutes any
simple correlation between the cultural evolution of numerical notation and cognition. There may be some correlation, but it must be demonstrated, not assumed
from the inefficiency of the Roman numerals for a task for which they were never
intended.
Secondly, some scholars claim that the Roman numerals prevented the Romans
from developing other useful institutions or techniques. Glautier (1972) asserts
that the Romans did not develop an efficient accounting system due to the lack of
a suitable numerical notation system, while Williams (1995) bemoans the limitations of Roman numerals for doing the complex calculations with fractions needed
for alloying coinage. Yet the Romans evidently had sufficient accounting techniques

Italic Systems

117

to administer their empire, and while double-entry bookkeeping is made easier by


a ciphered-positional numerical notation system, it must have functional equivalents, or else no society lacking such numerals could administer a large political
entity. The inferiority of Roman numerals is used by Guitel (1975) to explain why
the Romans were poor mathematicians as compared to their Greek subjects. However, both Greek and Roman mathematicians relied considerably on Babylonian
knowledge, and educated Romans knew the Greek alphabetic numerals. Maher
and Makowski (2001) show that the Romans were, in fact, quite good at arithmetical computation. If, in fact, the Romans were poorer mathematicians than
the Greeks, evidence other than their numerals must be sought.
In contrast, a small body of research within the subdiscipline of the history of
mathematics holds that the Roman numerals are not less efficient for computation
than the Western numerals. At least four modern scholars claim to show how the
Roman numerals could have been used in written calculations without the aid of an
abacus or similar technology (Anderson 1956, Krenkel 1969, Detlefsen et al. 1975,
Kennedy 1981). These analyses, apparently derived independently of one another,
differ in the exact technique used in performing calculations, but all conclude that
even if the Romans never used their numerals in such a fashion, the Roman numerals are in fact amenable to computational functions. While this is superficially
true, I regard this argument as highly spurious. The proposed techniques are often
more complicated than they are presented to be, and certainly more complicated
than the standard arithmetical techniques using Western numerals. For instance,
the technique proposed by Detlefsen and colleagues (1975) involves a complex
transformational grammar that is far removed from the knowledge systems of
Roman or medieval scholars. Additionally, none of these studies establishes that
the Roman numerals are equally or more efficient for computation than Western
or other numerals. A systems minimal utility for a function is hardly proof that
it was or should have been used for it. These studies presume that it is natural for
numerals to be manipulated for computation, even in the absence of historical
evidence. Anderson (1956: 145) suggests that any reader, once he discovers how
simple the operations are, will be inclined to imagine that some Roman engineers
and surveyors, in building their great projects, did occasionally do their computations very much in the way described below, even though they left no records of
their work. Detlefsen and colleagues (1975: 147) go so far as to blame the Romans
for not recognizing the computational potential of their numerals.
While the possibility that the Roman numerals were used in this way cannot
be ruled out, the argument ex silentio is highly implausible. We have persuasive
evidence that ancient and medieval scholars used the abacus, finger computation,
and other techniques (Lang 1957, Taisbak 1965, Murray 1978). The best way to
counteract the denigration of the Roman numerals is not to show that they are

118

Numerical Notation

mildly (or even greatly) useful for a function for which they were never known
to have been used. The issue is not simply a mathematical game to see whether a
system can serve some arithmetical function. By understanding the functions for
which the Roman numerals were not used, we may better understand the circumstances of their eventual replacement.

Medieval Additive
As early as the second century ad, but most prominently in the fifth and sixth centuries, Roman numerals inscribed on stone inscriptions frequently took on a cursive
quality, and some signs became ligatured together (Gordon 1983: 46). A special sign
for 6, , which is nothing more than a cursively written and ligatured vi or ui, occurs
on many inscriptions from late antiquity (Lassre 2005: 5759). On sixth-century
Byzantine imperial coins, was as common as VI (Wroth 1966: cx). The use of
this sign probably ceased in the eighth century (Bischoff 1990: 176). Similarly, in
late antique and early medieval inscriptions, 40, 60, and 500 could be written cursively, even in inscriptions on stone. The Visigothic script used between the seventh
and thirteenth centuries was particularly characterized by these and other ligatured
Roman numerals, particularly for 40 (XL) (Schapiro 1942, Bischoff 1990). These
sporadic shifts toward ligatured notation would later play some role in the paleographic modification of Arabic signs into what would become the Western numerals
(Lemay 1982: 393; see also Chapter 6, of this volume). Yet these ligatures did not
represent an inexorable diachronic trend toward ciphering in Roman numerals, and
the majority of early medieval Roman numerals remained cumulative-additive.
After the fall of the western Roman Empire, literate knowledge was distributed
rather sparsely for instance, among Western monks, Byzantine bureaucrats, and
Middle Eastern scholars. The great early medieval Mediterranean polities the
Byzantine Empire and the early Muslim caliphates mainly used cipheredadditive numerical notation systems such as the Greek and Arabic alphabetic
numerals (Chapter 5). In Western Europe during the Middle Ages, the Roman
numerals were the only ones in common use. Knowledge of other systems was
restricted to peripheral regions such as Spain and southern Italy, and to a tiny
well-educated elite. Even at the height of the Carolingian Renaissance (around
800 ad), arithmetic was the province of a learned few, and was acquired late in the
scholars education (Murray 1978). Still, while the need for large-scale bureaucracy and the corresponding need to express large numbers had declined since the
height of the Roman Empire, Roman numerals were still frequently encountered,
and even expanded in the range of functions they served.
The medieval Roman numeral-signs differ from the classical ones in several
aspects. Rather than being written solely in majuscule characters, Roman numerals

Italic Systems

119

were frequently written in the lowercase script on perishable materials. When


written in minuscule form, placing a stroke through the last numeral-sign of a
numeral-phrase indicated that one-half was to be subtracted from the represented
value. As a measure against fraud, the last i in cursive numeral-phrases was often
extended into a j, preventing anyone from adding additional signs to the end
of the phrase (Menninger 1969: 285). This practice originated around ad 900 in
South Italian manuscripts, although it may have had an antecedent in the upward
rather than downward extension of phrase-final cursive i into I in a few classical
texts (Smith 1926: 74). Alphabetic forms of the numerals for 500 (D) and 1000
(M) replaced the earlier X and _, marking the end point of a long process of
alphabetization. This would have had the advantage of some mnemonic convenience; even if most of the numeral letters did not correspond to the numbers they
represented, at least literate learners of the system would not need to learn an
entirely new set of signs. Alphabetization of the numeral-signs also enabled one
to use them for numerical riddles, particularly chronograms, in which the total
value of the Roman numerals in a line of verse expressed the date of an event
described in that verse (Menninger 1969: 281). The multiplicative vinculum bar
for 1000 used in classical antiquity continued to be used under a new name, the
titulus, but the three-sided box symbol for multiplying by 100,000 was no longer
used (Menninger 1969: 281). Large numbers were very rarely needed, and even
the need for the titulus was limited. Subtractive forms were used more frequently
in the Middle Ages, though purely additive forms (e.g., IIII) were still common.
Finally, starting in the tenth century, numeral-phrases for ordinal numbers were
often adverbialized using superscript endings such as -o and -mo for instance,
Mmo CCmo Lmo IVto, incorporating aspects of the medieval Latin lexical numerals
into numeral-phrases (Smith 1926: 75).
Shipley (1902) suggests that the changing use of Roman numerals between classical antiquity and the ninth century ad led to transcription errors in medieval
manuscripts. Comparing the fifth-century ad Codex Puteanus, containing sections
of the works of Livy, and the ninth-century ad Codex Reginensis, a copy of the
former, the analysis of copying errors reveals much about the Roman numerals used at the time of copying. Where the classical Roman form for 1000 was
_, the medieval scribe was more accustomed to using Z, and thus _ was often
transcribed as R. Where the classical manuscripts contained X for 500, medieval
scribes used W, and thus often omitted the X symbols entirely on the theory that
the horizontal stroke indicated that the scribe had crossed out an error. Finally,
because the subtractive form XL for 40 was used in medieval times as opposed to
RRRR, instances of RRRR were abbreviated to RRR to correspond to correct
medieval numeral-phrases. Shipleys analysis confirms the increased use of both
subtractive structuring and alphabetic Roman numeral-signs.

120

Numerical Notation

The range of functions for which Roman numerals were used expanded considerably in the Middle Ages. Astronomical texts, which in antiquity were almost
exclusively written using Greek numerals, often employed Roman numerals in the
medieval era. As mentioned earlier, the alphabetizing of the signs for 500 and 1000
to D and M allowed the creation of number-riddles such as chronograms in which
the numerical value of the Roman numerals in a phrase expressed the date of an
event. Evidence for their use in legal documents and account records increases
greatly, though this may be a function of the differential survival of perishable
materials from later periods.

Medieval Modifications
The replacement of the Roman numerals by the Western system was neither easy
nor uncontested. Instead, between the twelfth and seventeenth centuries, there
was great controversy throughout Western Europe regarding which system to
use, with cultural, sociopolitical, and practical considerations being invoked in
favor of one system or the other. Nor was the situation simply a choice between
two options. While the general trend was toward the Western system in the great
majority of contexts, as knowledge of the ciphered-positional system spread into
Western Europe, a few individual writers made idiosyncratic modifications to the
Roman numerals in response to the interloping newcomer. While none of these
modifications was adopted on a wider scale, they can enlighten us about the circumstances under which the Roman numerals were replaced.
The most complete positionalized version of the Roman numerals is one of the
earliest. Around 1130, the mathematician H. Ocreatus, a student of Adelard of
Bath, invented a positional numerical system using the Roman numeral-phrases
for 1 through 9 (I, II, ... IX) along with a special sign (O or t, called cifra) to
indicate an empty position (Smith and Karpinski 1911: 55; Burnett 1996, 2002c,
2006). This system is attested only in one thirteenth-century manuscript, a collection of arithmetical texts (Cashel, G.P.A. Bolton Library, Medieval MS 1), of
which Ocreatuss Helcep Sarracenicum (Saracen Calculation) is only one part. Positions were separated using a dot to avoid confusion; thus, 1089 was expressed
as I.O.VIII.IX.6 This system, which blends the Roman cumulative-additive and
Arabic ciphered-positional systems, is cumulative-positional and base-10 with a
sub-base of 5. While Murray (1978: 167) characterizes Ocreatuss system as clumsy,
it should be noted that it is fully positional, far more so than later compromises
6

Smith (1926: 72) argues that this practice was attested in a few classical inscriptions as
well, such as XII.L.D for 1,250,500 (!), but I know of no inscriptions where this was
actually done.

Italic Systems

121

Table 4.13. Medieval multiplicative Roman numerals


Year

Numeral-Phrase

Roman with
Multiplication

Western Source

122025

IIDCCCXIIII

Menninger 1969: 285

IX.XX.XVI

aaWUUUROP
URUPO
ZUUSPOOO
ZZSROOO
SRRRPOOO

2814

1231

196

Smith 1926: 6

1258

Steele 1922: xvii

2073

Crosby 1997: 208

88

Guitel 1975: 225

aaaUSRRROOO

3183

Cajori 1928: 33

USOR
ZWP
UU
UWSR
ZWROP
ggg
baWUUROOO

159

Preston 1994

1505

Menninger 1969: 285

200
460

Cajori 1928: 34

1514

Smith 1926: 7

300000

Menninger 1969: 285

6713

Menninger 1969: 283

gddaaaUWSPO

123456

Cajori 1928: 33

1258

Mij lviii

1340

IILXXIII

1388

143738

IIIIxx et huit
M XX
III C IIII III
VIIXXXIX

1505

IVcV

1392

1514

II
IIIIC.LX

1514

XVCXIV

1550

CCCM

1554
1771

vi vii xiii
c m
c
i xxiij iiij lvj

made between the Roman and Arabic systems. Nevertheless, this system was not
taken up more widely in medieval scribal or mathematical texts. The speculation of Busch (2004) that a difficulty with seafaring travel times in the fifteenthcentury Icelandic Landnmabk can be resolved by interpreting VI (6) as V I
(e.g., 51) is highly dubious.
While the use of barred numerals to indicate multiplication by 1000 was of
early origin and continued throughout the Middle Ages, new multiplicative forms
began to be used starting in the twelfth and thirteenth centuries. Some examples
of such numeral-phrases are listed in Table 4.13, along with the transcriptions of
the appropriate number both in the classical Roman system (including the use
of subtractive and multiplicative forms, where appropriate) and in the Western
system.
These numeral-phrases primarily express multiplication by 100 or 1000 by juxtaposing C or M either immediately beside the appropriate multiplier, above it,
or in superscript, and sometimes interposed with a dot. Often, the number 20
occupies a special role as a multiplier; such phrases are almost all from France
and result from assimilation to the partly vigesimal structure of the French lexical
numerals and/or to the monetary system (Preston 1994). These numeral-phrases
have a multiplicative component, but they are not positional the value of the

122

Numerical Notation

Table 4.14. Partially positional Roman numerals


MCCCC8II
CC2
ICC00
15IIII
15X5
MDZ4
MCCCC4XVII
IV0II

1482
202
1200
1504
1515
1624
1447
1502

numeral-signs does not change due to their position, but rather due only to their
juxtaposition with another sign. It might be thought that multiplicative forms
were adopted in order to write numerals more concisely or with a smaller set of
numeral-signs as was the case with the initial use of multiplicative forms in classical Rome. However, as seen by comparing the numeral-phrases in Table 4.7 to
their equivalents in standard Roman numerals, there is no such benefit.
In addition to these multiplicative forms, we find many cases where the signs of
the Roman system were intermingled with the positional principle of the newer
Western system, as well as its actual numeral-signs. The earliest known example is from the late twelfth-century Microcosmographia, in which the only known
copy (from the thirteenth or fourteenth century) is dated mclxxviii (1178) but in
another place as mc87 (an error for 1178) (Williams 1934: 107). Preston (1994)
describes a number of such mixed uses in Durandus of Saint-Pourains Commentary on the Sentences of Peter Lombard of 1336, such as xxx3 for 33 and xl7 for 47. In
one example from an English almanac from 1386, Western numerals both precede
and follow Roman numerals in a numeral phrase for 52,220, written as 52mcc20
(Halliwell 1839: 116). Menninger (1969: 287288) provides examples of such
admixtures starting in the late fifteenth century, as indicated in Table 4.14.
In these partly ciphered-positional numeral-phrases, conciseness is greatly
increased over their Roman counterparts. It is unknown whether this was being
done consciously as a compromise between the two systems, as a misunderstanding of the Western system, or as numerical playfulness. In at least one case, the
blending is obviously erroneous; in an astronomical table, the scribe wrote MCC6
for 1269, then crossed out the entire phrase and wrote 1269 in Western numerals
(Steele 1922: xvii). Such combinations are not necessarily advantageous; the idiosyncratic nature of these numeral formations almost certainly decreased their
comprehensibility. None were used frequently or consistently enough to create a
true variant system. These hybrid formations no longer appear after about 1650.
The roughly contemporaneous expansion of Roman multiplicative notation
and the introduction of hybrid forms employing ciphering and positionality in

Italic Systems

123

late medieval European scribal traditions meant that any number could be written in several ways. The wide range of textual genres containing such variants,
and their geographical and chronological breadth, confirm that this variety was
not isolated, but neither was it standardized. In any event, a parallel process was
ongoing from the twelfth century onward that would eventually lead to the nearly
complete replacement of Roman numerals by Western ones throughout Europe
and eventually throughout the world.

Replacement and Persistence


Although positional numeration was first introduced into Western Europe in the
late tenth century by Gerbert of Aurillac (later Pope Sylvester II), it was rarely used
before the publication in 1202 of Liber Abaci by Leonardo of Pisa, also known
as Fibonacci. This mathematical text sparked an important debate between two
camps: the abacists, those who preferred computation with the medieval abacus,
and the algorithmists, who preferred pen-and-paper calculations using the Western ciphered-positional numerals. The history of this debate is well documented,
as it involved many important commercial families, renowned mathematicians
and clergymen, and even state authorities (Menninger 1969: 422445; Evans 1977;
Murray 1978: 163175). Issues of computational efficiency were often addressed.
In his dictionary of 1530, the English lexicographer-priest John Palsgrave included
the sentence, I shall reken it syxe tymes by aulgorisme or you can caste it ones by
counters as a sample sentence for the verb to reckon (Palsgrave 1530: 337).7 However, these debates did not compare the Roman and Western numerals, but rather
two techniques of computation: pen-and-paper arithmetic versus abacus calculation. The use of these techniques was correlated with specific numerical notation
systems, but evaluations of computational efficiency require more information
about how numbers were manipulated.
Yet the debate was not only about efficiency. In 1299, the Arte del Cambio or
moneychangers guild of Florence prohibited the use of Western numerals in its
registers, a prohibition that was maintained for at least twenty years (Struik 1968).
The reason given in the document is to prevent fraud, which was encouraged by
the numerals ease of falsification, but Struik rightly notes that socioeconomic
explanations for this prohibition are just as persuasive, taking into account conflict between conservative (aristocratic) and progressive (mercantile) factions. In
1348, booksellers at the University of Padua were required to list prices in their
7

This statement must be considered in light of the fact that a) it is a sample sentence
rather than part of an evaluation of the two methods and b) Palsgraves book was paginated throughout using Roman numerals.

124

Numerical Notation

inventories not by ciphers but by plain letters (Berggren 2002: 361). As late as
1494, the Frankfurt Brgermeisterbuch (a mayoral book of edicts) ordered reckoners not to compute using Western numerals (Menninger 1969: 426427). One can
certainly imagine the consternation of merchants and bookkeepers upon learning
that one can simply add zeroes to the end of a numeral ad infinitum to multiply
its value by ten each time, but the role of xenophobia and traditionalism in all
of these cases should not be underestimated. The Western numerals may have
been rejected in part because of practical considerations, but they were, after all,
a foreign and newfangled invention that involved the unusual principle of positionality and that (because of their novelty) could be used to conceal information
or deceive others.
The rarity of paper in the earlier Middle Ages may also have contributed to the
continued use of the Roman numerals (Smith and Ginsburg 1937: 29). Until penand-paper calculation became feasible, the Roman numeralfriendly abacus was
the computational technology of choice, whereas the switch to pen and paper
made the conciseness of Western numerals more attractive. This attractiveness
increased with the introduction of printing presses, where long strings of movabletype Roman numerals would have been unwieldy in comparison to Western
numerals. Yet despite Eisensteins contention that, [t]he use of Arabic numbers for pagination suggests how the most inconspicuous innovation could have
weighty consequences in this case, more accurate indexing, anotation, and crossreferencing resulted (1979: 1056), most incunabula contained no Western numerals. Even where the utility of Western numerals for computation was recognized,
the Roman numerals were not always abandoned entirely, as in Italian metrological documents studied by Travaini (1998), in which computations were undertaken using Western numerals, but totals were then written in Roman numerals.
William Cecil (Lord Burghley), Englands Lord High Treasurer from 1572 to 1598
and a chief advisor of Elizabeth I, would frequently transcribe economic documents from Western back into Roman numerals, and was decidedly uncomfortable with the new system (Stone 1949: 31).
Despite such resistance, the Western numerals had taken hold by around 1300
in the Italian city-states. Elsewhere in Western Europe, particularly in Germany
and England, the Roman numerals predominated until the late fifteenth century or even later. Jenkinson (1926) finds limited evidence for the use of Western
numerals in English archives before 1500, and notes that Roman numerals were
forbidden from use in state accounts only in the nineteenth century. Barradas
de Carvalhos (1957) study of fifteenth- and sixteenth-century Portuguese texts
showed that the Roman numerals were replaced there only around 1500. Several
important sixteenth-century arithmetical texts continued to use Roman numerals in
preference to Western ones. Jacob Kbels (14701533) Rechenbiechlin, published in

Italic Systems

125

Figure 4.2. Tables of variant Roman numerals from the Mysticae numerorum significationis
(158384) of Petrus Bungus. Source: Smith 1908: 380383.

Augsburg in 1514, an important commercial arithmetic book intended for mercantile instruction, uses Roman numerals throughout and strongly advocates the use
of the counting board, although Kbel frequently used multiplicative phrasing for
hundreds and thousands, and his text is the source of the unorthodox 260/400
Roman numeral fraction in Table 4.13 (Smith 1908: 100107). However, by the
seventeenth century, the battle was essentially finished, and the Roman numerals ceased to be used in most contexts. Figure 4.2, from the Mysticae numerorum
significationis of Petrus Bungus (d. 1601), first published in 158384, depicts a wide
variety of archaic and curious Roman numeral phrases, but is essentially a text on
number mysticism and symbolism rather than a manual for practical use (Smith
1908: 380383).
Why, after over a millennium of essentially unchallenged use in Western Europe,
should the Roman numerals have ceased to be used for most purposes? The traditional answer given that Roman numerals were less efficient for computation
is correct but incomplete. The Roman numerals were used alongside the Greek

126

Numerical Notation

alphabetic numerals for well over a thousand years, despite the great increase in
conciseness that could have been achieved by abandoning the Roman system. In
medieval Western Europe, debates over the use of Roman numerals lasted into
the seventeenth century, with the relative usefulness of the Roman numerals for
calculation being greatly increased by the accompanying use of counting boards.
However, two developments were fundamental in rendering the Roman system obsolete in the West. Firstly, the development of the printing press and the
consequent rise in literacy after 1450 correlates very well with the rapid adoption
of Western numerals throughout Europe (particularly in northern regions). The
newly literate middle classes of Western Europe were learning to read and calculate, unconstrained by centuries of traditional use of the Roman numerals. At
the same time, the rise of mercantile capitalism in the Renaissance changed how
numbers were viewed and used (cf. Swetz 1987). The functional needs of Western
society to represent number changed dramatically between 1300 and 1700, and
the computational functions that are better served by Western than by Roman
numerals increased in importance. Nevertheless, Crosbys (1997: 41) insistence
that Roman numerals were adequate for the weekly market and for local tax collection, but not for anything grander is misplaced if intended as an explanation
for their replacement. In fact, as Crosby himself notes, Roman numerals persisted
for centuries in the record books of institutions as grand as the British Exchequer
(Crosby 1997: 115116). Double-entry bookkeeping was facilitated to some degree
by the transition to Western numerals, but it was by no means a requirement.
The introduction of a more fully mercantile and urban economy in Western
Europe did, however, expose more people, particularly members of the newly wealthy
and increasingly literate middle class, to a new form of numeration. Rather than
examining this change in notation simply by comparing the two notations, it is more
productive to examine the question from a social perspective. If Roman numerals
were truly so inferior, how did they persist for two millennia without being replaced?
Why did it take six centuries even after the introduction of Western numerals for this
shift to take place? The most likely scenario is that the new users of writing and written numeration, unconstrained by past practices, provided a critical mass of users for
the Western numerals that facilitated certain mercantile functions.
Even today, the Roman numerals have not disappeared. The use of the Roman
system on clock faces, in the enumeration of kings and popes, on many dated
inscriptions, and to mark the copyright dates on films is rather archaic, but the
cultural importance of its symbolic connotations of antiquity, tradition, and prestige likely guarantee its future survival. In media where archaism is desired, Roman
numerals are often preferred; this is the case among the Cornish church sundials
from 16701850 studied by Burge (1994), for instance. An inscription on a gateway at Harvard uses archaic Roman numeral-signs, noting the colleges 264 years of

Italic Systems

127

existence in 1900 as ANN. DOM. _WUUUU. COLL. HARV. UUTROOOO (McPharlin 1942: 18). Pot (1999) notes the existence of several twentieth-century postage
stamps in which additive Roman numerals rather than the ordinary subtractive ones
are employed for instance, XXXXI instead of XLI on a 1990 German issue. Presumably these variants are not errors per se, but are mobilized for aesthetic reasons,
just as on Roman numeral clock faces, 4 is normally written IIII instead of IV, even
though 9 is IX (Hering 1939). The fifteenth-century accountant Luca Pacioli advocated the use of Roman numerals for the year-dates in accounting books for aesthetic
reasons (although Western everywhere else), recommending [u]se the ancient letters in making this entry, if only for the sake of beauty (Crosby 1997: 222).
At the other end of the class and prestige spectrum, Roman numerals were
often retained in peasant or rural traditions and trades long after they had
ceased to be used in formal texts and administration. Gmr (1917) describes
a wide variety of wooden tally sticks of Swiss provenience, dating from the
sixteenth to the nineteenth centuries, containing Roman numeral-phrases. The
same principle applies to the much earlier calendar-numerals used on Northern European calendrical documents (see the following discussion). With
the inexorable spread of industrialization and capitalist commercial practices
throughout Europe, the use of wooden tallies containing Roman numerals has
essentially disappeared.
When used to paginate the introductions of books, or to enumerate items in
sublists or subsections, Roman numerals continue to serve the very practical function of providing an alternate form of enumeration wherever two different series
of things must be listed. I would not expect that any industrial society could function solely with a cumulative-additive system such as the Roman numerals. Conversely, however, I would not expect them to be replaced entirely. The failure of
over five hundred years worth of predictions of the Roman numerals imminent
demise, and complaints regarding their utility, suggests that functional considerations, while important, do not tell the entire tale with respect to the history of
numerical notation systems.

Arabico-Hispanic Variants
Arabic and Western European knowledge systems interacted intimately on the
Iberian peninsula between the tenth and sixteenth centuries the period of the
Reconquista and somewhat beyond. At this time, the Roman numerals were
well known in Spain and Portugal, as in the rest of Europe. Arab mathematicians and astronomers had, since around 800 ad, used a ciphered-positional
system much like the one used today (Chapter 6), along with older cipheredadditive abjad numerals related to those of the Greeks (Chapter 5). Medieval

128

Numerical Notation

Spain thus presents us with a remarkable case where three coexistent systems
each had a different structuring principle (cf. Labarta and Barcel 1988). But
rather than a simple case of the two Arabic systems replacing the Roman numerals, medieval Spanish astronomers, bookkeepers, and scribes experimented and
mixed principles, incorporating ciphering and positionality into the basic structure of the Roman numerals in innovative ways. It was, as described by Berggren (2002: 358), a promiscuous blend of systems, a mlange that persisted for
several hundred years.
In the tenth and eleventh centuries ad, Spanish astronomers wrote extensive
astronomical texts using primarily the Roman numerals. Even the use of the subtractive principle was very infrequent in Spain at this time. Despite the use of
Arabic positional numerals in the region for some time previous and familiarity
with works such as the Arithmetic of Al-Khwarizimi, Spanish astronomers did not
adopt the system directly (Lemay 1977: 458). Instead, the Roman numerals were
modified to increase their conciseness. Instead of using the cumulative principle
to express the numbers 3 through 9 (III, IIII, V, etc.), they used new acrophonic
symbols based on the first letters of the Latin numeral words: 3 = t = tres; 4 = q =
quatuor; 5 = Q = quinque; 6 = s = sex; 7 = S = septem; 8 = o = octo; 9 = N = novem.
Of these, only the symbols for 4, 8, and 9 were common, probably because they
are the longest numeral-phrases below 10 in Roman numerals (IIII, VIII, VIIII).
In addition, a special sign for 40 was employed: , a cursive ligatured version
of XL, and the only exception to the abandonment of subtractive forms (Lemay
1977: 459). Bischoff (1990: 176) reports that this sign for 40 is also found in some
Visigothic manuscripts, suggesting considerable antiquity for the sign on the Iberian peninsula. These changes altered the classical Roman numerals into a partly
ciphered-additive system. However, these modifications were not accompanied
by the adoption of positional notation, and the system still had a mixed base
of 5 and 10 and was purely additive. These developments increased the conciseness of numeral-phrases considerably. The number 99, which would have been
LXXXXVIIII, could now be expressed as LN. This system was used in mathematical and astronomical texts until about the mid twelfth century, at which time
the Western numerals took hold in astronomy, as in the rest of Western Europe.
Yet while the Arabic positional numerals had firmly established themselves and
were later transformed into the Western numeral-signs familiar to us, the Roman
numerals did not cease to be used on the Iberian peninsula, though their use
became increasingly limited. Labarta and Barcel discuss two curious offshoot
Roman numeral systems found in early modern Spanish documents, as shown in
Table 4.15 (Labarta and Barcel 1988: 3234).
Both systems are cumulative-additive, decimal systems with a sub-base of 5,
unlike those of the earlier astronomical texts. Unlike most Roman numerals,

Italic Systems

129

Table 4.15. Arabico-Hispanic numerals


1

10

50

100

numeral-phrases are written from right to left (highest values at the right), which
is curious because even in the Arabic script, which has a right-to-left direction, the
numerical notation system has a left-to-right direction. The first system is found
in a few late sixteenth- and early seventeenth-century documents to indicate monetary quantities, and its numeral-signs are somewhat similar to letters of the Arabic script. The second system is found only in a single Inquisition document from
1576 (Labarta and Barcel 1988: 34).8
From the sixteenth century onward, Western numerals became increasingly
common throughout the Iberian peninsula, but certain anomalies remained in the
Roman numerals of the early modern period. Subtractive numerals remained rare
in comparison to the rest of Europe. A symbol known as the caldron was sometimes used; it meant 1000 but was used multiplicatively, preceded by unit-signs,
not additively (Cajori 1922, Lowe 1943). The caldron was most often shaped like
a U, sometimes with one or two diagonal bars across it (). Thus, a manuscript
from the Ponce de Leon papers from 1501 has the year dated as JDJ (1 1000 500
1), reminiscent of the use of M for that purpose in other parts of medieval Western
Europe (Lowe 1943: 9). Cajori (1922) hypothesizes that the caldron originated
in Italy in the fifteenth century, possibly as a modification of the older Y for
1000 that was not fully replaced by M until the modern period. Curiously, even
after the Roman numerals had been abandoned entirely, it continued to be used
multiplicatively with Western numerals, as in a contract written in Mexico City in
1649 in which a sum of 7291 pesos is expressed both as 7U291 and VIIUCCXCI
(Cajori 1922: 201).

Calendar Numerals
In the late Middle Ages, unusual numerals were used in some Northern European
documents and inscriptions pertaining to calendrical calculations. Known as calendar numerals, runic numerals, or peasant numerals, they are a variant of the
Roman numerals. I reject the term runic because few runic inscriptions contain
8

However, see Chapter 10 for a discussion of a potential descendant of this system in the
form of Berber numerals used in North Africa.

Numerical Notation

130

Table 4.16. Calendar numerals


1

a b c d e f g
A B C D E F G
1 2 3 4 5 5 5
1 2

10 11 12 13 14 15 16 17 18 19 20

h
H
5
3

i j k l m
I J K L M
5 6 6 6 6
4
1 2 3

n o p q r
N O P Q R
6 7 7 7 7
4
1 2 3

s t
S 8
7 8
4

them.9 Likewise, the term peasant numerals tells us who may (or may not) have
been using them, but lacks the precise functional association of calendar numerals. Some examples of these signs are shown in Table 4.16 (cf. Kroman 1974: 121;
Ifrah 1985: 146147).
These systems are all cumulative-additive and have a base of 10 with a sub-base
of 5. Units are marked by strokes or dots; fives are marked by angled or curved
lines or loops to create U or V shapes, and tens are marked by transecting
the vertical line perpendicularly, creating a cross or X. Although these numeralsigns are often joined together into single figures resembling digits by using a
vertical stroke, the system is not ciphered. There are no signs for 50, 100, or
higher values, and I am not aware of calendar numerals being used for numbers
higher than 31 (the number of days in the longest months). Other than their
unusual numeral-signs, the calendar numerals are identical to ordinary Roman
numerals. While they were used in rural traditions that suggest to some that
they were part of the large cultural substratum of tallying and notching in medieval and early modern Europe (Menninger 1969: 249251; Ifrah 1985: 146147),
the calendar numerals are not tallies in this sense. A tally is used ordinally
one places marks as necessary on some material (wood, paper, stone, etc.) in
sequence in order to keep a running total, rather than, as with numerical notation systems, marking an already totaled value. The calendar numerals are simply a variant of Roman numerals in which numeral-phrases are written vertically
and attached to a line.
By the late fifteenth century, the Western numerals were fairly well known
in Germany and were becoming much more common in England and Scandinavia. Calendar numerals were used in a very delimited set of contexts, namely

The medieval antiquarian tradition of manuscripts discussing runic writing contains minimal evidence of any specifically Runic numerical system. The clophruna (Old Norse
klapprnir) system of notating numbers by dots (a = 1 = .; b = 2 = ..; c = 3 = ) discussed
by Derolez (1954: 134) is of obscure origin and is not clearly related to any other numerical notation.

Italic Systems

131

on documents of wood, stone, or horn designed to assist the largely nonliterate


populace in determining the dates of festivals, especially Easter. Throughout the
medieval period, the Metonic cycle of nineteen years, after which the moons
phases recur on the same day of the year, was used as a rough-and-ready guide to
calculate the date of Easter. The function of the calendar numerals was solely to
denote the various years of this cycle on stylized perpetual calendars, known as
runic calendars in German- and Norse-speaking areas and as clog almanacs in
England. Because of this specialized function, there was almost never any reason
to express numbers higher than 19. While these texts were used for computation,
the numerals were not used directly for arithmetic. One merely needed to line up
the appropriate days and years to get the correct value. Calendar numerals were,
however, reasonably compact and easily understandable by anyone who knew the
Roman numerals.
The calendar numerals lie at the heart of one of the major pseudoscientific controversies of New World archaeology the so-called Kensington Rune
Stone of Minnesota, which purportedly contains a Viking inscription left in
1362 by Norse explorers who had traveled westward from Vinland. At issue
are the calendar or runic numerals used in the inscription, particularly the
numbers 14, 22, and 1362, expressed as ad, bb, and acfb, respectively (Hagen
1950; Nielsen 1986: 51). These numeral-phrases are ciphered-positional, like our
Western numerals but unlike the cumulative-additive calendar numerals (in
which 14 would be written as n). It is improbable that a Norse explorer in
Vinland would have been familiar with the Western numerals in 1362, which
were known only by the very educated in northern Europe at the time (Struik
1964: 167). Even if this knowledge is presumed, we need to explain why the
savant substituted the cumulative-additive runic numeral-signs for the appropriate Western figures. While some fourteenth-century runic inscriptions used
Western numerals (Nielsen 1986), calendar numerals were never used in such
a positional fashion. There is no evidence for the use of calendar numerals
except in calendrical texts, which the Kensington stone is not. Ohman, the
Swedish-American discoverer of the stone, was likely familiar with the calendar numerals (easily found in any book on runology) but not with the proper
structure of the system. Similar inscriptions with runic numerals, such as that
on No Mans Land, an island south of Marthas Vineyard, Massachusetts, are
undeniable forgeries (Marstrand 1949).
The calendar numerals were a brief and local phenomenon in northern Europe.
As the Western numerals came increasingly to be used for calendrical purposes,
their use waned. By 1643, when Ole Worm wrote his Fasti Danici describing the
runic numerals used in calendar tables, it was primarily as a curiosity and to aid
the transition to newer methods of calculation (Worm 1643).

132

Numerical Notation

Summary
The Italic numerical notation systems developed in the early sixth century bc with
the invention of the Etruscan and Greek acrophonic numerals, based on a previously existing but still poorly understood system of tallies, and spread eastward
from Italy during the period of classical Greek preeminence in the Mediterranean.
However, of all the Italic systems, only the Roman numerals had any extensive use
in the Christian era, as the Greeks had switched to the ciphered-additive alphabetic
numerals (Chapter 5) by the Hellenistic period. The remarkable persistence of the
Roman system and the swift decline of other systems are best explained by the
changing political fortunes of their users. The use of the earlier Italic systems was
mainly limited to inscriptions and commercial marks, though the Roman numerals
were later used for an enormous variety of functions in different social contexts.
All the Italic numerical notation systems are cumulative-additive and decimal,
with a sub-base of 5, and virtually all use a single vertical stroke for the units. Some
unusual structural features occasionally emerge in the higher powers of 10, such
as the use of implied multiplication in South Arabian and the hybrid multiplicative structure of later Roman numerals. Although the epichoric numerals of Argos,
Nemea, and Epidaurus use dots rather than vertical strokes for units, they are related
to the Greek acrophonic numerals and must be considered part of this phylogeny.
The cultural history of some Italic systems is intermingled with those of the
Hieroglyphic (Chapter 2) and Levantine (Chapter 3) phylogenies, making the
construction of accurate cultural phylogenies more difficult, particularly because
the systems of all three are cumulative-additive, decimal, and used in the eastern Mediterranean. Often the three families can be distinguished on structural
grounds: the Hieroglyphic systems all lack a quinary component, while the Levantine systems all have special signs for 20 and are multiplicative-additive above
100. This structural distinction can be confirmed independently by examining
known patterns of historical contact. Despite the jumbled state of our present
knowledge, we can identify phylogenies, not only because of similarities in their
systems structure, but also as a result of the attested cultural connections among
the societies in which they were developed. Nevertheless, to insist too strongly on
pure cultural phylogenies following a model of speciation would be erroneous.
In comparison to either the Alphabetic (Chapter 5) or South Asian (Chapter 6)
systems, however, the diversity of Italic systems is far less than might be expected,
given the preeminence and impact of the Roman Empire on European social life. The
Roman numerals gave rise to relatively few descendant systems when they were borrowed, they were copied rather than modified. We would do well to remember that
the Roman numerals are simply one numerical notation system among the worlds
diverse systems, despite their importance within the Western consciousness.

chapter 5

Alphabetic Systems

The systems discussed in the previous three chapters are primarily cumulative,
repeating signs within each power of the base to indicate addition. In contrast,
the next two families the Alphabetic and South Asian systems consist mainly
of ciphered systems, which use, at most, a single sign for any power to indicate
its different multiples: 1 through 9, 10 through 90, 100 through 900, and so on,
in the case of decimal systems. Ciphered numeral-phrases are thus much shorter
than cumulative ones, but require their users to be familiar with many more signs.
Alphabetic numerical notation systems generally use phonetic script-signs, in a
specified order, to express numerical values, and thus mitigate the effort needed to
memorize both script-signs and numeral-signs. Despite the name, the scripts in
question are not always alphabets; some, such as the Hebrew and early Arabic, are
abjads or consonantaries, expressing primarily consonantal phonemes, and one,
the Ethiopic Geez script, is an alphasyllabary or abugida, expressing consonant +
vowel clusters.1
Alphabetic systems were used as far north as England, Germany, and Russia
and as far south as Ethiopia, and throughout Africa and the Middle East from
Morocco eastward to Iran. Their history spans over two thousand years, from the
development of the Greek numerals around 600 bc to the present, but in some
1

For more extensive discussion of script typology, which is not warranted here, see
Daniels and Bright (1996).

133

134

Numerical Notation

cases important historical questions remain unresolved. While they are mostly ciphered-additive, they are not structurally identical. We can learn much more from
these structural differences than from the paleographic curiosities of the signs of
various systems. I hope in this chapter to illuminate areas of study where our
knowledge is less than perfect in order to draw attention to the need for further
specialized research. Table 5.1 shows the most common variants of the major alphabetic systems.

Greek Alphabetic
In Chapter 4, I discussed the cumulative-additive Greek acrophonic numerals,
which were given their name because the letters used are the first letters of the
appropriate Greek numeral words. Another Greek system, much more common, is sometimes called the Ionic or Milesian system due to its origin in
western Asia Minor, but is most commonly called the alphabetic system. While
the Greek alphabet was based on a Phoenician model, probably in the ninth or
eighth century bc, none of the earliest Greek inscriptions contains numerical
notation; thus, the debates on the time of the origin of the alphabet are not relevant to the origin of alphabetic numerals (cf. McCarter 1975, Swiggers 1996).
The first examples of the alphabetic numerals date to the sixth century bc and
are written using the letters of the archaic Greek script used in Ionia and the
Ionian cities of Caria, such as Miletus. Table 5.2 has representative examples of
these early signs.
The system was purely ciphered-additive and decimal, and was usually written from left to right, though right-to-left and boustrophedon (alternating direction) inscriptions are not unknown. The numeral-signs are archaic variants of the
twenty-four familiar Greek letters, plus three special signs called episemons: vau
or digamma (6), qoppa (90), and san or sampi (900), which were added to reach
a full complement of twenty-seven signs for all the values from 1 through 900,
allowing one to write any natural number less than 1000.2 Vau and qoppa were
occasionally used phonetically in the Ionic script, with the values of [v] and [k],
while san appears to have been borrowed from Phoenician sd [ts], and was possibly used in archaic Greek with a similar phonetic value (Swiggers 1996: 265266).
There are very few examples of alphabetic numerals from this early period, and
all of them express values under 1000, so we do not know whether higher values
could be represented.
2

There was no general name for the three signs in classical Greek; the term episemon originally referred only to the sign for 6, but following its etymology, some classicists use the
term in a more generic sense (J. Kalvesmaki, personal communication; Foat 1905).

135

70

60

50

40

30

20

10

a
b
g
d
e
v
z\
h
q
i
k
l
m
n
x
o

Greek

Coptic

Ethiopic

.a.
.b.
.c.
.d.
.e.
.f.
.g.
.h.
.i.
.j.
.k.
.l.
.m.
.n.
.o.
.p.

Gothic Hebrew

Table 5.1. Alphabetic numerical notation systems

g
h
i
j
k
l
m
n
o
p

Syriac

a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p

Arabic

a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p

Fez

a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p

a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p
a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p

a
b
c
d
e
f
g
h
i
j
k
l
m
n
o
p

(continued )

a
b
c
d
e
f
g
h
i
k
l
m
n
o
p
q

Armenian Georgian Glagolitic Cyrillic Latin

136

6000

5000

4000

3000

2000

1000

900

800

700

600

500

400

300

200

100

90

80

p
,
r
s
t
u
f
c
y
w
.
/a
/b
/g
/d
/e
/v

Greek

Coptic

Ethiopic

.q.
.r.
.s.
.t.
.u.
.v.
.w.
.x.
.y.
.z.
.{.

Gothic Hebrew

Table 5.1. Alphabetic numerical notation systems (continued )

q
r
s
t
u
v
sv
tv
uv
vv
svv
/a
/b
/c
/d
/e
/f

Syriac

q
r
s
t
u
v
w
x
y
z
:
;

Arabic

q
r
s
t
u
v
w
x
y
z
,

Fez

q
r
s
t
u
v
w
x
y
z
!
@
#
$
%
^
&

q
r
s
t
u
v
w
x
y
z
1
2
3
4
5
6
7
q
r
s
t
u
v
w
x
y
z
/
1
2

q
r
s
t
u
v
w
x
y
z
{

r
s
t
u
x
y
z

Armenian Georgian Glagolitic Cyrillic Latin

137

1,000,000

100,000

90,000

80,000

70,000

60,000

50,000

40,000

30,000

20,000

10,000

9000

8000

7000

z
h
q
i
p

/z
/h
/q
a
b
g
d
e
V

Greek

Coptic

Ethiopic

Gothic Hebrew

-f
-g
-h
-i
-j
-s

/g
/h
/i
-a
-b
-c
-d
-e

Syriac

Arabic

Fez

*
(
)

8
9
0
-

Armenian Georgian Glagolitic Cyrillic Latin

Numerical Notation

138

Table 5.2. Greek alphabetic numerals (archaic)


1

a
10s
j
100s
s
562 = wob

1s

b
k
t

c
l
u

d
m
v

e
n
w

f
o
x

g
p
y

h
q
z

i
r
,

In the classical and Hellenistic periods, the familiar Greek alphabet supplanted
the archaic regional (or epichoric) variants, and the alphabetic numerals developed along with them, retaining their order and numerical values but assuming
their modern (majuscule) forms. In addition, starting in the middle of the fifth
century bc, two new techniques were used to express higher values. For multiples
of 1000, a small slanting mark (known as a hasta) was placed to the left and below
a sign for 1 to 9 to indicate that its value should be multiplied by 1000; thus, G
means 3, but /G means 3000 (Threatte 1980: 115). Values above 10,000 are rarely
encountered except in mathematical works, and individual writers used different
methods to do so. The most common method, used by Aristarchus, involved placing a small alphabetic numeral-phrase (less than 10,000) above a large M character (= myriades) to indicate multiplication by 10,000 (Heath 1921: 3941).3 Thus,
3,000,000 would be expressed with only two signs, as y. This allowed any number
less than 100 million to be easily expressed.4 The system thus appeared as shown
in Table 5.3, using the Athenian letters.
The system is thus ciphered-additive for values under 1000, and thereafter is
multiplicative-additive at two different levels: firstly, through the use of a hasta
to indicate multiplication by 1000, and then through the use of an M to indicate
multiplication by 10,000. The alphabetic numerals were generally written in descending order, with the highest values on the left. Numbers between 11 and 19
were often written with the 10-sign (I) following the unit-sign, however, to correspond with the way in which the ancient Greek lexical numerals were formed:
3

Heath also discusses techniques such as that of Herons Geometrica, where two dots
placed over a sign indicate multiplication by 10,000; that of Apollonius, using tetrads,
turning the system into a mixed base-10/10,000 ciphered-additive system; and that of
Nicholas Rhabdas, a fourteenth-century scholar who used Herons technique, except
that additional pairs of dots above a number indicated successive powers of 10,000.
None of these systems was ever widely used.
For the ancient Greeks, ten thousand times ten thousand 100 million, the first uncountable number in the alphabetic system was of symbolic significance, as in Revelation 5:11: and the number of them was ten thousand times ten thousand.

Alphabetic Systems

139

Table 5.3. Greek alphabetic numerals (classical)


1

Aa
Ii

Bb
Kk

Gg
Ll

Dd
Mm

Ee
Nn

Vv
Xx

Zz
Oo

Hh
Pp

Qq
<,

1000s

Rr
/A

Ss
/B

Tt
/G

Uu
/D

Cc
/E

Ff
/V

Yy
/Z

Ww
/H

>.
/Q

10,000s

1s
10s
100s

562 = CXB

hendeka, dodeka, treis kai deka, tettares kai deka, and so on. For instance, Threatte
(1980: 114) provides a number of examples from Attica where GI, ZI, HI, and
QI appear for 13, 17, 18, and 19. Starting in the Roman period, the signs became
more rigidly fixed in highest-to-lowest order. There was a mild taboo against the
use of theta (Q) for 9 in some texts because it is the first letter of death;
circumlocutions were used instead, such as writing the number lexically, or as an
additive combination of alphabetic numerals such as E (5 + 4) or AH (1 + 8)
(Smith 1926: 69; Thomas 1977).
Because alphabetic numerals could easily be confused with written words, classical Greek alphabetic numerals were sometimes distinguished from the rest of the
text with special signs, most commonly a horizontal stroke above the numeralphrase, but occasionally with dots placed to either side of it. One of the problems
in identifying earlier Greek alphabetic numerals is the lack of such marks, meaning that any single letter could be an alphabetic numeral or a non-numerical label.
Even in later periods, numerals frequently appear without any indicator mark
whatsoever (Threatte 1980: 115).
In most Greek monumental inscriptions, the only fractions used are acrophonic signs for fractions of different monetary units (Tod 1950: 134). In mathematical
and literary texts, an entirely different system was used in which small accents
or strokes placed above and to the right of a numeral indicated unit-fractions
(Thomas 1962: 43). Special signs existed for 1/2 (< and U) and 2/3 (), in addition to standard unit-fractions (Thomas 1962: 45). From the second century
ad onward, the requirement of using only unit-fractions was lifted, and fractions
were expressed with both numerators and denominators using alphabetic numerals. Finally, a special system for fractions was used in astronomy, combining the
alphabetic numerals with sexagesimal structures borrowed from the Babylonians.
There is no evidence for an early (eighth century bc or earlier) origin of the
numerals, although prominent classicists such as Larfeld (190207) endorsed this

140

Numerical Notation

theory, arguing for descent from the Phoenician alphabetic system, because the
Greek alphabet was borrowed from a Phoenician ancestor and because many Semitic scripts have alphabetic numerals (Brunschwig and Lloyd 2000: 388). Yet no
Semitic consonantal numerical notation systems existed before the second century
bc, and they were based on Greek rather than the other way around (Gow 1883).
None of the very earliest Greek alphabetic inscriptions contains numerals. Conversely, however, the once-popular theory of a very late origin (late fourth or even
third century bc) cannot be sustained in light of evidence from earlier periods
(Gow 1883).
The first epigraphic evidence for alphabetic numerals comes from a vase, dating
to around 575 bc, found at Corinth, which contains the inscription SYM g, which
Johnston reads as mixed batch of 7 (Johnston 1973: 186). There is good evidence
from Attica and Corinth for the systems use on mercantile vases in the late sixth
and early fifth centuries bc (Hackl 1909). Yet it is unlikely that the numerals actually developed in either of these localities. Rather, the numerals probably developed in western Asia Minor, in the regions of Ionia and Caria, especially in the
cities of Miletus5 and Halicarnassus, where several early instances of the numerals
have been found (Heath 1921: 3233). Another early example is found in the tunnel of Eupalinos on the Ionian island of Samos, constructed around 550 bce, in
which distances are noted using letters that are probably alphabetic numerals (as
opposed to non-numerical or ordinal labels) (Verdan 2007: 1213). All the early
examples of the alphabetic numerals, even those found outside Asia Minor, are
written using the Ionic script, which was used in Ionia, Caria, and various Ionic
colonies throughout the Mediterranean. This reflects the predominance of Ionia
in regional and international commerce during the sixth and early fifth centuries
bc. While vau and qoppa were used phonetically in the early Ionic script, san was
not, and so had lost its place between pi and qoppa in the numerical order (Jeffery
1990: 327). Later, it was reincorporated into the script for the purpose of providing
a twenty-seventh sign for the numerals to function, but was placed at the end of
the system, with the numerical value 900.
The principle of the Greek alphabetic numerals was borrowed directly from the
Egyptian demotic numerals in the early sixth century bc, only using alphabetic
signs as numeral-signs (Chrisomalis 2003; see also Zaslavsky 2003). Boyer (1944:
159) regarded the similarity between the two systems as indicative of a historical
connection, but his paper was not primarily oriented toward such an argument.
Because the alphabetic numerals were the first to use phonetic signs as numeralsigns, there will be no paleographic similarity between the Greek numeral-signs
5

Miletus, from whence the adjective Milesian, was the most important Ionian city in
Caria, the region of Asia Minor immediately to the south of Ionia proper.

Alphabetic Systems

141

and any other system. Yet the alphabetic numerals are structurally similar to the
demotic numerals. They are both ciphered-additive, base-10 systems. While it has
yet to be established whether the alphabetic numerals used multiplicative notation
at an early date, both systems are multiplicative-additive above 10,000. Unlike the
demotic numerals, the alphabetic system is also multiplicative for the thousands.
The Greeks could simply have continued the series above 1000 using 10 through
90 and 100 through 900 (/I = 10,000; /K = 20,000; etc.), so the use of two-stage
multiplication may be a clue to the alphabetic numerals history. It would be
reasonable for the Greeks to adopt the multiplicative principle at the same level
as in the demotic numerals, namely 10,000. However, it would not have been feasible to find nine extra signs for the values 1000 through 9000. Consequently, the
inventor(s) of the alphabetic numerals may have had the idea of using multiplication for the thousands values as well as the ten thousands. The only remaining
problem is to explain why the Greeks, recognizing this irregularity, did not then
abandon the higher multiplicative series.
Furthermore, Greek arithmetical techniques for dealing with fractions strongly
resemble the Egyptian unit-fraction (1/x) tradition of computation (Knorr 1982,
Fowler 1999a). Both systems used unit-fractions formed by placing a small mark
above numeral-signs to indicate the appropriate unit-fraction. As well, both used
non-unit-fractions for specific values such as 1/2 and 2/3. Historians of mathematics are unanimous that the Greeks borrowed the unit-fraction technique from the
Egyptians, and I see no reason to doubt that the Greek use of special signs for 1/2
and 2/3 is also a result of Egyptian influence.
Turning to the historical evidence, the demotic numerals were the predominant ones in use in Egypt (especially Lower Egypt) in the early sixth century bc,
when Greeks were just starting to encounter Egyptians in large numbers for the
purposes of international trade. Most notable among the Greek traders in Egypt
were Ionian colonists from Miletus, who set up an important emporion (port of
trade) at Naukratis in the western Nile delta in the seventh century bc. Naukratis
quickly became the central locus for cultural contact between Greece and Egypt, a
position that it held until the Ptolemaic era. Inscriptions in the Ionic Greek script
dating as early as 650 bc have been found at Naukratis (Heath 1921: 33). It should
be noted, however, that no known inscriptions from Naukratis contain alphabetic
numerals, and there are later (fourth-century bc) inscriptions with acrophonic numerals (Gardner 1888). Since the earliest examples of the alphabetic numerals are
from containers for commercial goods, the context of the systems development
was probably in mercantile activity.
The only alternative to the hypothesis of Egyptian origin is that the Ionians
independently developed a ciphered-additive, decimal numerical notation system
within a few decades of coming into contact with Egyptians in large numbers,

142

Numerical Notation

founding a colony at Naukratis, and being exposed to the demotic numerals used
throughout Lower Egypt. These connections are too significant to be coincidental.
This should not be taken as a denial of the Greeks inventiveness, however, because
the alphabetic numerals have several distinctive properties. Firstly, while some of
the demotic numeral-signs use the cumulative principle, the alphabetic numerals
use purely ciphered signs. Secondly, as mentioned earlier, the alphabetic numerals
use the multiplicative principle for 1000 through 9000, obviating the need for
nine more signs for those values, as one could simply write a hasta before a unitsign. Finally, correlating numeral-signs with the ordered set of alphabetic signs
meant that anyone who knew the order of the alphabet could determine the signs
values as long as the episemons were taken into account. The often-mentioned
weakness of the alphabetic numerals, that too many signs needed to be learned,
is thus illusory. In learning to read and write, Western pupils must learn twentysix alphabetic signs (in their proper order) plus ten digits in order, making thirtysix total signs in two separate series, while the ancient Greeks needed to learn only
twenty-seven alphabetic signs and two auxiliary signs (/ and :), and needed only
twenty-nine total signs in one series.
After a period of Ionian cultural dominance between 575 and 475 bc, when
alphabetic numerals were commonly found, alphabetic numerals are found only
rarely in a period starting in 475 bc and lasting around 150 years (Johnston 1979:
27). During this period, the height of Greek civilization, Athens came to the forefront as an Aegean power, while Ionias power waned after the Milesian-led Ionian
revolt of 499 to 494 bc against Achaemenid Persia. The system did not disappear
entirely; at Halicarnassus, there is evidence of its continued use (Keil 1894; Heath
1921: 3133). A curious inscription from Athens (IG I2 760) from the middle of
the fifth century bc contains a long series of alphabetic numerals, written with
Ionic letters (Tod 1950: 137). There is, as well, good reason to believe that some of
the graffiti on amphorae from the Athenian Agora had weight and volume marks
notated in alphabetic numerals (Lawall 2000). In general, however, acrophonic
numerals were used in most Greek texts during this period.
The resurgence of the alphabetic numerals in Greece around 325 bc corresponds
precisely with the rise of the Ptolemies in Egypt. In this renewed period, some
of the earliest instances of the numerals come from Egypt. Figure 5.1 depicts a
portion of a Greco-Egyptian astronomical text including a calendar for the Saite
nome (Hibeh Papyrus i 27) dating to around 300 bc, one of these very early instances (Grenfell and Hunt 1906, Fowler and Turner 1983). Ordinary (unmarked)
Greek letters stand for whole numbers, while 1/x unit-fractions are represented
by signs modified with long oblique strokes. This is the earliest attested example of
alphabetic numerals used in a scientific context. Coins dating to 266 bc indicating
the regnal year of Ptolemy II Soter are, to my knowledge, the first coins bearing

Alphabetic Systems

143

Figure 5.1. Hibeh papyrus i 27, with Greek alphabetic numerals indistinguishable from
letters, but with unit-fractions clearly distinguished through the addition of long upward
slanting strokes. Source: Grenfell and Hunt 1906: Plate VIII.

any ciphered-additive numerals (Tod 1950: 138). It is interesting that while these
examples come from Egypt, we have no record of the alphabetic numerals use in
Egypt during the interlude of the fifth and fourth centuries bc. The evidence, at
present, is simply too scanty to conclude what stimulus caused the rejuvenation
of the alphabetic numerals.
From the third century bc onward, the alphabetic numerals were preferred over
the acrophonic numerals throughout most of the Greek-speaking world, with only
Athens retaining the acrophonic system until around 50 bc (Threatte 1980: 117).
While the acrophonic numerals used in different city-states varied quite widely,
the alphabetic numerals had no regional variants. As such, they could be used as
an effective instrument of cross-cultural communication and trade among diverse
regions (Dow 1952: 23). Whereas Greece before Alexander was highly fragmented,
rendering the development of a universal Greek numerical notation system unlikely, Alexandrine and especially Roman Greece provided a suitable environment
for the development of a single pan-Hellenic notation. That the system was a very
concise way to represent numbers, and that it relied on alphabetic symbols that
had become invariant throughout Greece by this period in history, cannot have
hurt this process.

144

Numerical Notation

In the sixth and fifth centuries bc, the alphabetic numerals were no more than
a system for labeling mercantile containers. All the early instances of the systems
use are from marked vases and potsherds. Even then, most numerals on vases are
acrophonic or other cumulative-additive Greek numerals, not alphabetic ones. In
these very early contexts, the alphabetic numerals, like the acrophonic ones, were
used for cardinal quantities, particularly of money, weights and measures, and
discrete quantities of commodities, the sorts of numerical expressions likely to
be found in inventories and decrees. From the third century bc onward, though,
when the alphabetic system became the predominant one throughout the Greek
world, the numerals were used in a much wider range of contexts. In contrast to
the acrophonic numerals, which are found solely on ceramic vessels and stone,
alphabetic numerals are found, in addition, in manuscripts of various sorts as well
as on coins. As described by Tod (1950: 130134) and Threatte (1980: 115116), the
functions for the alphabetic numerals include:
a.
b.
c.
d.
e.
f.

cardinal quantities of commodities, persons;


phrases indicating lengths of time in days, months, and years;
monetary values (denarii, drachmas, and obols);
weights, measures, and distances;
ordinal numerical adjectives and adverbs;
ordinal dates, for example, to indicate a specific year in the tenure of an archon.

Before the Hellenistic period, we have no evidence that the alphabetic numerals
were used for arithmetic or mathematics. Early writers such as Herodotus instead
used the acrophonic numerals to record the results of computations performed
with a pebble-board or abacus (Lang 1957). This situation changed once the alphabetic numerals began to be used more widely. Most surviving Greek mathematical
manuscripts use the alphabetic numerals, which were used by all Greek mathematicians beginning with Archimedes and Apollinius (third century bc). Users
of alphabetic numerals had a wide variety of computational techniques available.
Schrlig (2001) suggests that while the earliest abaci are most conducive to working with acrophonic numerals, certain types of abaci were well suited to working with alphabetic numerals also. Furthermore, the Greeks developed a complex
system of computing on the fingers up to 9999, reminiscent of the alphabetic
systems structural shift above 10,000 (Williams and Williams 1995). To facilitate
multiplication and division, which were difficult to undertake using the abacus or
finger computing, the Greeks used multiplication tables written with alphabetic
numerals (Dilke 1987: 16; Sesiano 2002). Finally, Greek arithmetic was sometimes
done in a manner analogous to modern Western practice, by writing down numbers and manipulating numeral-signs directly; the primary difference between the

Alphabetic Systems

145

two techniques is that in the Greek case, because the numerals are nonpositional,
there was no need to line up numbers in columns (Smyly 1905).
One would think, as Boyer (1944: 160) comments, that the adoption of the
alphabetic numerals by such prominent Greek mathematicians would curb the
criticism of modern scholars that the system was dysfunctional for mathematics.
Yet many modern scholars denigrate the alphabetic numerals in comparison not
only to ciphered-positional systems but also to cumulative-additive systems such
as the acrophonic numerals (cf. Boyer 1944: 160166). In particular, the need to
learn many signs and the lack of resemblance between numeral-signs for common
multiples of different powers (e.g., 5, 50, and 500) are perceived as serious weaknesses. In the only instance where a modern scholar actually attempted to learn
and use the numerals, however, the system fared very well. The classicist Paul Tannery found that calculation with alphabetic numerals took little more effort than
with Western numerals, with which Tannery, despite all his efforts, surely had far
greater experience and familiarity (Boyer 1944: 160161). In any event, only a small
fraction of the Greek texts containing alphabetic numerals serve mathematical
functions. Like virtually all numerical notation systems, numerals are primarily
for representation and only secondarily for computation.
Throughout their history, the Greek alphabetic numerals were used primarily
in Greek-speaking areas or in regions under the control of Greek speakers. During
its early history, the system was used in the eastern Mediterranean (particularly the
Aegean), Ptolemaic Egypt, and Seleucid Persia. Cursive varieties of the alphabetic
numerals on Greco-Egyptian papyri show a great deal of paleographic variability
(Foat 1902, 1905). A large number of variant multiplicative signs for 1000 and
10,000 were used in these papyri, most notably for 10,000, which was the
normal form starting in the second century ad (Brashear 1985). While alphabetic
numerals were not used for administration during the height of Roman power,
from the fourth century ad onward, they were used as the primary numerals of
administration, law, literature, and mathematics in the Eastern Roman Empire.
Whenever and wherever the Greek alphabet was used in the Middle Ages, the alphabetic numerals followed. Additionally, the Greek alphabetic numerals were the
most common system used in Arabic papyri for several centuries after the Islamic
conquest for recording the results of financial transactions (Grohmann 1952: 89).
Most of the descendants of the Greek alphabetic numerals substituted the
letters of their own scripts for the Greek signs. In the late second century bc,
the Israelites developed the Hebrew alphabetic numerals along a Greek model.
In the mid fourth century ad, the Goths developed alphabetic numerals along with
their Greek-influenced script, while in regions of Africa under Greek influence,
the Coptic script of Egypt and the Geez script used in Ethiopia both developed
alphabetic numerical notation systems based on a Greek model. In Armenia and

146

Numerical Notation

Georgia, right on the border of the Eastern Roman Empire, scripts and alphabetic
numeral systems developed in the fifth century ad at around the time they were
Christianized. In the sixth century ad, the Syriac script abandoned an earlier
cumulative-additive numerical notation system (Chapter 3) in favor of one based
on the Greek. Around the same time, the Arabic abjad numerals used following
the Islamic conquest of the Middle East were derived at least in part from the
Greek alphabetic system. The Glagolitic and Cyrillic numerals, developed in Slavic
regions in the late ninth century ad, under the auspices of the missionary-related
script development of Cyril and Methodius, closely resemble their Greek ancestor.
Finally, knowledge of the Greek alphabetic numerals in Western Europe led to the
development of a set of Latin alphabetic numerals in the twelfth century ad.
It is remotely possible that the Greek numerals are ancestral to the Brhm numerals (Chapter 6), which were used from the late fourth century bc onward in
India, and which eventually gave rise to Western numerals. The Brhm numerical
notation system is ciphered-additive and decimal, and used a variety of the multiplicative principle. The chronology of its invention, corresponding almost exactly
with the Alexandrine conquests and journeys in India, is also suggestive. However,
the Brhm system is more similar in structure and numeral-signs to the Egyptian
demotic numerals than to the Greek alphabetic numerals.
The large number of descendants of the Greek alphabetic numerals is due in
part to their longevity. This cannot be a full explanation, however; other systems,
such as the Egyptian hieroglyphs, were used over a much longer period, yet generated few direct descendants. Other long-lived systems, such as the Roman numerals, spread very widely over large parts of the world due to Roman imperial power,
but they were often accepted by colonized or subordinate societies unchanged,
and did not replace indigenous systems entirely. Because the Greek system was
alphabetic, cultures borrowing the principle of alphabetic numeration tended to
modify the signs to fit their own scripts (whether alphabets or consonantaries)
rather than adopting the Greek alphabetic numerals directly, and also made minor
structural changes to the system.
In the early Middle Ages, when the Eastern Roman Empires fortunes were
prosperous, the numerals were widely used throughout Greece, the Balkans,
Egypt, the Levant, and Asia Minor, and were incorporated into the learning of all
European mathematicians. For instance, they were known to the English scholar
Bede, who described them in his De temporum ratione (The Reckoning of Time)
in the early eighth century ad (Wallis 1999). Many Western European manuscripts
include numerical glossaries, for instance, a tenth- to eleventh-century codex
(Vatican Library, Codex Urbinas Latinus 290) that lists (among other information
concerning non-Latin alphabets) the Greek lexical numerals, letter names,
and alphabetic numerals along with their translations into Roman numerals

Alphabetic Systems

147

(Derolez 1954: 106109). Such tables would have aided medieval scribes in understanding texts, but the Greek alphabetic numerals were not actively used except
in rare instances. For instance, in one ninth-century Latin manuscript, the Greek
numerals 1 (A) to 21 (KA) stand for the letters A through X in a cryptogram to
indicate that it was written by Irish writers at the Welsh court of Merfyn Frych
(824844) (Derolez 1954: 9799). Such cryptographic uses confirm rather than
refute the obscurity of the system, however. In the Western European debate between the abacists and algorithmicists, the Greek alphabetic numerals, used by
all the great mathematical minds of antiquity, did not rate a mention.
The eventual fate of the Greek numerals was directly tied to the fortunes of the
Byzantine Empire. Byzantine mathematicians and astronomers used alphabetic
numerals exclusively until the twelfth century, and numerals such as regnal years
were often stamped on Byzantine coins. By 1300, however, the geographical extent
of the numerals use had become very limited, and mathematicians were using
Western numerals under the influence of Arab learning in Spain and Italy. In the
Byzantine Empire, mathematicians used Arabic positional numerals in marginal
notes on Euclids Elements in the twelfth century (Wilson 1981). The first major
Byzantine mathematician to recommend the switch to the Arabic numerals was
Maximus Planudes (c. 12601310) (Schub 1932). In some Byzantine astronomical
texts, Arabic numerals were used to write year numbers while Greek alphabetic
numerals were used for all other functions; even then, the Greek equivalent of the
Arabic numeral-phrase was often written in the margin of the page (Neugebauer
1960: 5). In a few late Byzantine mathematical texts, combinations of Greek alphabetic numerals and Arabic positional numerals with a zero occurred, such as
(400 + 50 + 0) instead of for 450 (Neugebauer 1960: 5). The ultimate extension
of this principle was the use by some writers of only the first nine Greek letters
(A through ) to indicate the units, then adding a dot, a circle, or a special sign, ,
to indicate a zero position (Schub 1932: 59; Menninger 1969: 273274).
In 1453, with the fall of Constantinople, the Greek numerals ceased to be used
administratively. Nevertheless, they continued to be used thereafter for restricted
purposes, such as paginating the introductions to scholarly texts and enumerating
ordinal lists, just as the Roman numerals were used in Western Europe. Starting in the seventeenth century, the sign for 90 (qoppa), little used in alphabetic
contexts, underwent a palaeographic alteration, so that a new, roughly Z-shaped
sign served only as a numeral, while the older Q-shaped sign could serve both
alphabetic and numerical functions, but was regarded as a different sign (Everson
1998). Alphabetic numerals are still used today in many Greek Bibles to enumerate
chapter and verse numbers, in certain legal contexts, and in pagination. Given the
prestige associated with these functions, it is unlikely that the alphabetic numerals
will cease to be used entirely.

148

Numerical Notation

Coptic
The Coptic language is the last descendant of the Egyptian languages. It was
written using the Greek alphabet until the fourth century ad, when a distinctive Coptic alphabet originated based on a Greek model, but using six additional
characters taken from the demotic script to express uniquely Egyptian phonemes.
Unlike the earlier Egyptian scripts, it is written from left to right and has signs for
vowels. The adoption of the Coptic script was accompanied by the introduction of
a numerical notation system based on the model of the Greek alphabetic numerals
(Megally 1991, Messiha 1994). The Coptic numeral-signs are shown in Table 5.4.
The numerals, like the script itself, were written from left to right, and were
ciphered-additive and decimal. The numeral-signs closely resemble the Greek
uncial signs used between the fourth and ninth centuries ad. In addition, as in the
Greek alphabetic numerals, a horizontal stroke above the numeral-phrase indicates
that it is a numeral rather than a word, and a slanted subscript stroke under a unitsign (the Greek hasta) indicates multiplication by 1000 (Megally 1991: 1821). There
is no known sign or multiplier for 10,000 or higher values for these numerals.
The classical age of Coptic lasted from the fourth to the tenth centuries ad,
during which time the script and numerals were used extensively, surviving the
seventh-century ad Muslim conquest of Egypt. There may have been a geographical division in the frequency of their use, with northern Egyptian scribes using
them frequently, while southern writers tended to write out numbers using lexical
numerals (Till 1961: 80). While the Coptic numerals were generally used in formal
manuscripts, the ordinary cursive Greek alphabetic numerals were used for calculation and administration, possibly because the Coptic numerals, being uncials
without tails, were less practical for rapid writing (Megally 1991: 1821).
It is unclear whether the Ethiopic numerals (used to write the Geez language from
the fourth century ad onward) were based directly on the Greek alphabetic numerals
or derived through a Coptic intermediary. While the Ethiopic system is generally
said to derive directly from Greek, the Coptic uncial letter-signs are similar enough
to the Greek to render such a determination premature, and given the geographic
proximity of Egypt and Ethiopia, this possibility cannot be dismissed outright.
While the primary function of Coptic numerals has always been religious, in
the context of the Coptic Church, their administrative and arithmetical functions
should not be discounted. Despite the control of the population of Egypt by a succession of foreign powers, and despite the decline of Coptic as a spoken language
from the twelfth century onward, the use of Coptic numerals continued as late as
the fourteenth century. They are still used today in Coptic Christian liturgical texts
for pagination and stichometry, although for most ordinary purposes, either Western or Arabic positional numerals are preferred by those familiar with Coptic.

Alphabetic Systems

149

Table 5.4. Coptic numerals


1

1s
10s
100s
1000s
6085 =

ZIMM Numerals
In Egypt in the tenth century ad, when the Coptic script was being replaced
by Arabic for most administrative purposes, a different Coptic cursive numerical
notation system developed, known as numerals of the Epakt (named after the
computation used to resolve the discrepancy between the solar and lunar calendars) (Messiha 1994: 26) or, more generally, as urf al-zimm account book; register numerals. Often simply labeled Coptic numerals in modern scholarship,
it seems prudent to distinguish them from their uncial antecedents, particularly
since the zimm numerals are often found alongside Arabic or Hebrew texts, not
ones in the Coptic language or the Coptic alphabet. Their long-standing use in a
wide variety of contexts renders their importance far greater than the restrictedfunction Coptic uncials. While doubtless many users of the zimm numerals were
Coptic Christians from Egypt, this was certainly not an ironclad rule, and Egyptian Jews and Muslims writing in Arabic script frequently used Coptic numerals
instead of the Arabic abjad or Hebrew alphabetic numerals. The zimm system is
shown in Table 5.5.
This system is structurally identical to the classical Coptic system, but the numeral-signs are cursive minuscule letters rather than uncial ones. Many of these
signs bear little or no resemblance to the classical signs. Some of them are probably taken from the signs of the Arabic abjad numerals, which I will describe later,
while others are of indigenous development. This system also uses two stages of
multiplication (at 1000 and 10,000), like the Greek alphabetic system. A horizontal stroke and two dots placed below a sign indicated multiplication by 10,000
(Sesiano 1989: 64). A fifteenth-century multiplication table includes instructions
on writing numerals of the Epakt, and indicates that this sign for 10,000 could
be used in conjunction with any of the twenty-seven letters, thus allowing any
number less than ten million to be expressed (Sesiano 1989: 5455). It may be
that the two-stage multiplicative principle at 1000 and 10,000 existed even in

Numerical Notation

150
Table 5.5. Zimm numerals
1

1s
10s
100s
1000s
7104 =

the earlier uncial numerals, and that we simply have no paleographic evidence to
confirm this.
The paleographic relation between the first and second Coptic systems remains
unclear; although some of the signs are related, more are not, and in fact many
are modifications of Greek minuscule letters rather than Coptic uncials. These
numerals were often used in early bilingual Arabic-Coptic documents, suggesting
that the Arabs were making concessions to local administrators. This situation is
quite extraordinary, because many Egyptian Arabs by this time were employing
the ciphered-positional Arabic numerals used today. This suggests that the advantages of ciphered-positional systems over ciphered-additive ones may not have
been perceived to be important. In addition to Arabs, many users of the zimm
system were Egyptian Jews who used the system alongside their own alphabetic
numerals (Goitein 196788: II, 178). Their primary functions were commercial,
including contracts, bills of sale, payments, and accounts. There is no evidence
for or against the position that numerals were manipulated for arithmetic; this
surely would not have been done on permanent media, even perishable ones like
papyrus.
Although zimm numerals originated in commercial contexts and are rare in attested astronomical and mathematical texts, Lemay (1982: 384) has found one early
text dating to ad 938 (MS Garullah Efendi 1508, Sleymanie Library, Istanbul)
in which the numbers in the text are written lexically, the Arabic abjad numerals
used in astronomical tables, and the zimm numerals used in pagination. King
(1999: 7678) discusses an astrolabe made in Cairo in ad 128283 with an Arabic
inscription but zimm numerals. At this period, particularly in Egypt, the Coptic
calendar was used for various purposes by Muslims, potentially explaining this
artifacts unusual blend of representational systems. Zimm numerals were also
used very widely for pagination and foliation of Arabic texts written by adherents
of all three Abrahamic faiths, throughout Egypt and more broadly throughout
the Maghreb (Troupeau 1974). Mingana (1936: II, 183184) reports that an Arabic

Alphabetic Systems

151

letter of bequest from 1586 relating to a Syriac Christian manuscript (Mingana


Syriac 617) is dated in Coptic numerals (presumably zimm rather than uncial),
suggesting that even at this late date, their use was widespread and not restricted
to Coptic Christians and Egyptian Muslims.
Although Coptic began to become obsolescent as a language in the twelfth century, and Coptic Christians made the transition to the Arabic language and script,
the zimm numerals continued to be used widely in economic documents otherwise written in Arabic, and for divinatory purposes. Their obscurity lent itself to
cryptographic uses; a letter sent in 1139 from the Egyptian Jewish trader Halfn ben
Nethanel to the poet Isaac ibn Ezra uses zimm numerals as a code for Hebrew characters that in turn denote Arabic words (Goitein 1971: II, 303). Their opacity and rarity could also be useful in avoiding fraud, for instance, on a thirteenth-century check
in which the amount was written in Arabic words and zimm numerals, presumably
because their obscurity made it less likely that they would be altered fraudulently
(Reif 2002: 26).6 In the fourteenth-century Muqadimmah of Ibn Khaldn, he mentions the use of zimm ciphers, the same as those used for numerals by government
officials and accountants in the contemporary Maghrib alongside the Arabic abjad
numerals and the ghubar positional numerals (Chapter 6) in the zirajah technique
of numerical divination (1958: I, 239).
The zimm numerals played a considerable role in the development of the socalled Fez numerals used in Spain and parts of North Africa from the twelfth century onward (see the following discussion). Both the Greek alphabetic numerals
and the Arabic abjad numerals were also involved in the development of the Fez
numerals, however, and the paleographic lines of descent among these systems are
complex. The problem is made more complex by the fact that the later users of
zimm numerals were otherwise writing Arabic texts, not Coptic ones. Moreover,
the Fez numerals are first attested in Spain, not in Africa. However, the similarities
of some of the sign-forms for particular numerals suggest some relationship.
The zimm numerals were used for several centuries, and survived the conquest
of Egypt by the Ottoman Empire in 1517. Egyptian Coptic children in the sixteenth century learned Arabic before Coptic in kuttb schools, and their practical
education emphasized their assimilation to Ottoman practices, but they continued to learn and use zimm numerals for bookkeeping and accounting, thereby
acquiring a reputation for competent and secure financial practices associated with
6

The use of highly ligatured Roman numerals in late medieval Europe, and of special
complex numeral-signs in Chinese banking, are additional examples of the practice of
using unusual numerical systems to avoid fraud, and are somewhat paralleled by the
modern requirement of writing out numerals lexically in addition to using Western
numerals on checks.

152

Numerical Notation

what was by this time a highly cryptic numerical code (Hamilton 2006: 38). They
ceased to be used only in the seventeenth century, after which time Arabic positional numerals were used for these functions (Ritter 1936; Messiha 1994: 26).

Ethiopic
The Ethiopic script developed in the late third century ad, primarily on the model
of the Minaeo-Sabaean consonantary used in South Arabia, but also influenced
by the Greek and Coptic alphabets used to the north. It was used (and continues
to be used) for writing various languages of Ethiopia, especially Geez, the liturgical language of the Ethiopian church, and modern Amharic. While the earliest
Ethiopic script has no signs for vowels, by the fourth century ad, the script was
an alphasyllabary, in which each individual sign represents a consonant + vowel.
The direction of writing is always left to right. While the Ethiopic script is based
on South Arabian writing, its numerical notation is strictly Greek and Coptic in
origin. The earliest Ethiopic numerical notation, such as that found on third- and
fourth-century inscriptions from the kingdom of Aksum (on which the signs for
70 and 90 are not found) is shown in Table 5.6 (Drewes 1962; Ifrah 1998: 247).
The earliest inscription with Ethiopic numerals is on the late third-century Anza
stela erected by Bazat, negus (local king) of Agabo, in which the highest number
expressed is 20,620 (loaves of bread) (Drewes 1962: 6567). The modern Ethiopic
script uses a system derived from these early inscriptions, structurally unchanged
but slightly modified, as shown in Table 5.7 (Fossey 1948: 99; Haile 1996: 574).
The Ethiopic numeral-signs are not letters of the Ethiopic script, but rather
are derived from the Greek or Coptic letters. Even though the signs have no nonnumerical meaning in Ethiopic, the signs normally have marks both above and
below them to indicate that their value is numerical. This practice was universal
only from the fifteenth century onward, and it is not found at all in the Aksum
inscriptions (Ifrah 1998: 246247). In addition to these signs, modern Amharic
texts use a sign for 1000 ( ), which lacks marks above and below it (Bender et al.
1976: Table I). It may be that the minimal demand for writing very large numbers
led to the abandonment of the higher multiplicative formations and the introduction of an indigenous 1000-sign.
The system has a hybrid structure: it is ciphered-additive for the units and tens
but multiplicative-additive for numbers over 100. For 10,000, a multiplicative
sign consisting of two ligatured 100-signs was used, as shown in Table 5.7. Thus,
because there are special signs for 100 and 10,000, the Ethiopic system has a mixed
base of 10 and 100, with the base-10 formations governed by eighteen ciphered
characters for 1 through 9 and 10 through 90 and the hundreds and ten thousands
governed by multiplicative power-signs that combine with the ciphered signs.

Alphabetic Systems

153

Table 5.6. Ethiopic numerals (Aksumite period)


1

1s
10s
100s
1000s

etc.

10,000s

etc.

20,620 =

This flexibility allowed the Ethiopic system to express very large numbers with
a very limited set of numeral-signs. In theory, it could be extended infinitely by
adding additional signs for 10,000 as needed, although such large numbers rarely
occurred in practice (Guitel 1975: 272273).
The development of the Ethiopic system occurred under cultural contact and
Christianization by Egyptian and Syrian missionaries. While some scholars argue
that borrowing of the signs must have been from the Greek uncial script (Bender
et al. 1976: 124; Ifrah 1998: 246), Haile (1996: 574) raises the possibility that this
transmission might have taken place by means of a Coptic intermediary. Egyptian
missionaries were active in Ethiopia throughout the fourth century ad, which
might tip the scales slightly in favor of a Coptic origin. Saint Frumentius, generally held to be the first major converter of the Aksumites, was a Syrian by birth,
trained by Greeks, but his missionary work was based in Alexandria and focused
on establishing connections between the Aksumites and the Egyptian Copts. The
Table 5.7. Ethiopic numerals (modern)
1

1s
10s
100s
1000s
10,000s
647,035 =
100,000,000 =

(60 + 4) 10,000 + 70 100 + 30 + 5


(10,000 10,000)

154

Numerical Notation

South Arabian numerals were already extinct by this time and had no influence
on the Ethiopic system.
The use of base-100 for the multiplicative component of the system is the Ethiopic systems most notable feature. No other alphabetic system begins using the
multiplicative principle until 1000 or 10,000. This innovation had a clear antecedent in the Greek and Coptic use of multiplication at 10,000, but it eschews the
extra nine signs for 100 through 900 needed for both Greek and Coptic numerals.
Instead, it employs the entire set of eighteen ciphered signs, then the same set again
beside the 100-sign, then the same set beside the 10,000-sign. While the Ethiopic
numerals were based on the Greek or Coptic systems, the Ethiopic script was not.
Of all the descendants of the Greek alphabetic numerals, the Ethiopic system is the
only one to use non-phonetic signs as numeral-signs. While this makes it more cumbersome, since one needs to learn all the script-signs as well as twenty numeral-signs,
there is also no impetus to use all the borrowed numeral-signs. In systems that assign
numerical values to an ordered series of script-signs, it is natural that one would
assign values to all the signs, rather than stopping at some arbitrary point. In this
case, however, the Aksumites borrowed the first eighteen symbols of the alphabet
and then, rather than adopting another nine signs for 100 through 900, which were
meaningless to them, took only the sign for 100 and used the multiplicative principle
thereafter, thereby reducing the number of new signs they needed to learn.
Many Aksumite inscriptions on coins (indicating regnal years) and stone monuments (indicating cardinal and ordinal quantities of various kinds) contain Ethiopic numerals. After the fall of the kingdom of Aksum and the Islamic conquest,
the Ethiopic script was used only rarely, and over time became an esoteric script
known only to scholars and Ethiopian Orthodox priests. It was used in the mathematical, numerological, and astronomical texts of medieval Ethiopia, both for
writing whole numbers and for writing fractions in a special sexagesimal notation
based on Greek traditions (Neugebauer 1979; see the following discussion). In Amharic texts, Ethiopic numerals were used for a wider variety of functions from the
fifteenth century to the present day; for instance, they were used in the personal
correspondence of Amharic elites in the nineteenth century (Pankhurst 1985). The
New Testament printed in Amharic in 1852 uses the Ethiopic numerals throughout
for page, chapter, and verse numbers (Novum Testamentum in linguam amharicam
1852). Today, the numerals are still occasionally used for writing dates, but have
largely been supplanted by the Western numerals (Bender et al. 1976: 124).

Gothic
The Gothic alphabet was developed around 350 ad by Wulfila, a bishop who translated the Bible into his native language. Gothic was an East Germanic language

Alphabetic Systems

155

Table 5.8. Gothic numerals


1

a
b
10s
j
k
100s s
t
665 = .xoe.

1s

c
l
u

d
m
v

e
n
w

f
o
x

g
p
y

h
q
z

i
r
{

spoken by the Germanic tribes who migrated throughout Europe in the latter
years of the Western Roman Empire (Ebbinghaus 1996: 290). The script was
alphabetic and written from left to right. Along with the script, an alphabetic
numerical notation system was employed, as indicated in Table 5.8 (Braune and
Ebbinghaus 1966: 10).
Like the Greek alphabetic numerals, the Gothic numerals were usually distinguished from the rest of the text either through dots to either side of the numeralphrase or by placing a horizontal stroke above the phrase (Braune and Ebbinghaus
1966: 10). The Gothic numerals never expressed quantities higher than 1000, and
thus there is no evidence of the use of the multiplicative principle. Larger numbers
were always written out in full (Menninger 1969: 260). The system is therefore
ciphered-additive and decimal throughout. The numeral-signs are related to the
Greek uncial letters. Of the episemons, f (6) was the sixth letter of the Gothic alphabet, and had the phonetic value [kw]. The other two episemons, qoppa (r) and
san ({), had no phonetic value and were simply used to fill out the full complement
of twenty-seven signs using Greek models (Fairbanks and Magoun 1940: 319). The
possibility has been considered, but now largely rejected, that the Gothic script
owes its ancestry at least in part to either the Latin alphabet or the Germanic runes
(Ebbinghaus 1996: 290291). However, since neither of these scripts uses alphabetic numerals, the Gothic numerals must be of Greek origin.
The Gothic alphabet is attested in only a limited set of documents, mostly
translations of parts of the New Testament, but also in a small number of secular texts. Most numerals in Gothic texts indicate chapter and verse numbers in
Bibles. Additionally, they were used within the text to indicate numerical values,
while in Greek Bibles such numbers were always written out in full (Menninger
1969: 260). There is no evidence that the Goths ever did arithmetic or mathematics using these numerals. The sign for 900 is only attested in the Codex
Vindobonensis 795, an antiquarian text of the ninth century, by which time the
script had already fallen out of use (Fairbanks and Magoun 1940, Ebbinghaus
1978).

156

Numerical Notation

Very little of what surely must have been written in Gothic has survived.
Most surviving texts date to the sixth century ad, although the assignment of a
fourth- century ad origin to the numerals is undisputed. It is unclear exactly when
in the seventh or early eighth century the Gothic language died out, but around
that time the script and numerals ceased to be used, and were replaced by the Roman numerals that were coming to be used throughout Western Europe.

Hebrew Alphabetic
The earliest Hebrew scripts began to diverge from the earlier Phoenician consonantary in the ninth or tenth century bc. Then as now, Hebrew consisted of twenty-two consonantal signs, written from right to left, and placed in a customary
order. Early Hebrew inscriptions used a variant of the Egyptian hieratic numerals
(Chapter 2). Somewhat later, particularly in the fifth and fourth centuries bc,
some Hebrew speakers used the Aramaic numerals (Chapter 3) for administrative
and commercial purposes. Only at a much later date, probably in the second century bc, did a uniquely Hebrew set of numerals develop. This system is indicated
in Table 5.9.
The first twenty-two signs, indicating 1 through 400, are the letters of the Hasmonean Hebrew script circa 125 bc, at the time of the writing of the Dead Sea
Scrolls and also when the numerals were first used (Goerwitz 1996: 488). The
system is decimal and ciphered-additive, and written from right to left. Values
between 500 and 900 were represented using the sign for 400 in conjunction
with one or more signs for the lower hundreds (i.e., 500 = 400 + 100, 600 = 400 +
200 ... 900 = 400 + 400 + 100). This structural irregularity exists because there
are too few letters in the Hebrew consonantary to fill out the twenty-seven signs
needed to extend the system to 900. The number 400 occupies a special structural
role, but is not a base of the system, as its powers (16,000, 6,400,000, etc.) do not
receive any special treatment.
While the very earliest Hebrew inscriptions contain no signs for numbers above
1000, the need to do so quickly arose, as the numerals began to be used for dating
on grave inscriptions using the Hebrew calendar. For multiples of 1000, a mark
either a small curved stroke to the left of a numeral-sign or two dots placed above it
could be used to indicate multiplication by 1000; thus, would signify 9000
and 90,000. This feature is similar to, but distinct from, the Greek alphabetic
system, which adds a stroke above or below a numeral to indicate multiplication by
1000, but begins again at 10,000 by placing the multiplicand above the sign Z. Thus,
while the Greek system could express any numeral up to ten million, as opposed to
one million for the Hebrew numerals, the Hebrew system is arguably easier to use
because it has only one value at which the multiplicative principle is employed.

Alphabetic Systems

157

Table 5.9. Hebrew alphabetic numerals (Hasmonean)


1

1s
10s
100s
369 =

The Hebrew alphabetic numerals were derived from the Greek alphabetic numerals between 125 and 100 bc for use on coins inscribed with the Hasmonean
script (Gandz 1933: 76; Millard 1995: 192). Theories of a much earlier (ninth- to
seventh-century bc) independent origin are no longer credible (Smith and Karpinski
1911: 33; Gandz 1933: 7576; Schanzlin 1934; Zabilka 1968: 176178). A bilingual
ostracon from the Palestinian village of Khirbet el-Qm and dating to 277 bc uses
both Aramaic and Greek alphabetic numerals to denote a financial transaction
(Geraty 1975). Although it is not a Hebrew text, the village was in Judah and hundreds of contemporary Hebrew inscriptions have been found there. If the Hebrew
alphabetic numerals had been in use at that time, they likely would have been used
instead of the Aramaic numerals in a situation where the Greek inscription a few
lines below used a similar system.
The first safely dated instance in which the use of Hebrew alphabetic numerals
is certain is on coins from the reign of the Hasmonean king Alexander Janneus
(103 to 76 bc), some of which were stamped with Greek script and alphabetic
numerals, others with the Hebrew script and alphabetic numerals, in both cases
using alphabetic signs (Naveh 1968). Avigad (1975) suggests that a clay seal reading
Jonathan high priest Jerusalem M) might refer to the fortieth (m = 40) year of
the Hasmonean kings, and thus dates it to 103 bc, early in the reign of Alexander
Janneus, whose Hebrew name was Jonathan. That these early Hebrew alphabetic
numerals would both be found in the context of the same man is strong circumstantial evidence.
Despite the special formation of Hebrew numerals from 500 to 900, the similarities between the Greek and Hebrew numerical notation systems are striking.
The two systems share not only a similar structure (decimal and ciphered-additive)
but also the alphabetic principle for forming the numeral-signs. The notion that
the Hebrew numerals were independently developed can no longer seriously be sustained, despite the agnostic attitude of some scholars, including Ifrah (1998: 239). That
Hasmonean coins were struck in both languages and using both systems provides
specific contextual evidence that the Hasmonean kings adopted the technique

158

Numerical Notation

from the Greeks. The cultural influence of the Ptolemaic and Seleucid kingdoms
in the Levant at this time was enormous; Greek alphabetic numerals were used on
coins from the Phoenician cities of Sidon, Tyre, Byblos, and Akon from the mid
third century bc onward (Millard 1995: 193). At the same time, the Hebrew use
of 400 as a stepping-stone for representing the higher hundreds is an important
innovation, as it did not require that Hebrew speakers learn and adopt additional
nonphonetic signs.
Despite the development of the Hebrew numerals, most Semitic peoples of the
ancient Levant (Nabataeans, Aramaeans, Palmyrans) never used alphabetic numerals, but continued to use their own hybrid cumulative-additive/multiplicativeadditive systems (Chapter 3), which coexisted with the Hebrew alphabetic system for several centuries, and were replaced only over a long period. Until the
seventh century ad, inscriptions on Jewish graves throughout the Mediterranean
region were often written, not with the Hebrew numerals, but rather in the Greek
alphabetic numerals (Ifrah 1998: 238239). Gradually, the Hebrew numerals replaced the Greek and Levantine systems, until they were firmly established as a
distinctive system peculiar to the Jewish populations of Europe, North Africa,
and the Levant. Also, in the sixth century ad, the Hebrew numerals were partly
or wholly used by the creators of the Syriac alphabetic numerals (to be discussed),
which are also ciphered-additive and decimal and have the same break at 400 as
the Hebrew system.
The modern Hebrew alphabetic numerical notation system has the same structure as the ancient system, but uses modern script-signs, as shown in Table 5.10.
Around the beginning of the tenth century, five additional Hebrew characters, the
signs used for kof, mem, nun, pe, and tsade in word-final position, were sometimes
used to complete the sign set for 500 through 900. These signs occur in some
Masoretic commentaries on the Old Testament, but were never the regular forms
used (Gandz 1933: 96102). The older formations using additive combinations of
hundreds-signs are the usual means of representation. The newer forms probably
were not widely accepted because the word-finality of these signs is inconsistent
with the principle of the numerical notation system that the highest values should
come first in a numeral-phrase (Gandz 1933: 98). To put a word-final letter for
500 through 900 at the head of a numeral-phrase would be inconsistent with its
original purpose; to put it at the end of the numeral-phrase would be inconsistent
with the rule of decreasing ordering of the powers.
Also in the tenth century, Hebrew scholars became aware of positional numerals, and occasionally experimented using combinations of the alphabetic numerals
and the positional principle. In a Masoretic poem by Saadia Gaon (ad 882942),
numbers are written quasi-positionally in two columns; the rightmost represents
the thousands, while the leftmost column represents the ones, so that 42,377 was

Alphabetic Systems

159

Table 5.10. Hebrew alphabetic numerals (modern)


1

1s
10s
100s
OR

, i.e., (7 + 70 + 300) + ((2 + 40) 1000) (Gandz 1970: 487488).


written as
Some medieval Hebrew writers simply used the first nine alphabetic symbols in
place of the ordinary Western or Arabic signs, supplementing them with a circle for
zero, which made the system ciphered-positional (Gandz 1933: 110). This practice
is first attested in the mid twelfth-century Sefer ha-Mispar of Abraham ibn Ezra
(Burnett 2000c: 14). The ordinary Hebrew numerals were always used in the regular
text of such works, with the positional variants used only for mathematical expressions
(Schub 1932). This technique was not commonly used, however, and later Hebrew
mathematicians and astronomers simply used Western or Arabic numerals.
One of the more important functions for which the Hebrew numerals have
been used historically is gematria, the art of number-magic (Ifrah 1998: 250256).
Because every letter of the Hebrew script has a numerical value, every Hebrew
word has a numerical value equal to the sum of its letters values. Among medieval
and early modern scholars, this practice was commonly employed for interpreting
passages from the Talmud and the Midrash and for finding symbolic associations
among words that share the same numerical value. For instance, two of the terms
associated with the Messiah, shema
seed and menakhem
consoler,
have the same numerical value (8 + 40 + 90 = 40 + 8 +50 + 40 = 138). A related
practice is the construction of chronograms, in which a specified set of words has
a numerical value equal to the date of an event (e.g., a persons death) to which
the words refer. Because these practices (also used with the Arabic abjad, described
below) can only be done with a system for correlating phonetic signs with numerical values, they probably contributed to the retention of the alphabetic numerical
notation systems long after ciphered-positional systems had been adopted.
While the Western numerals are used in modern Israel for most purposes, the
alphabetic system is regularly used for dates using the traditional Jewish calendar,
especially in religious texts and on graves. In 1999, the Israeli Supreme Court ruled
that gravestones in orthodox Jewish cemeteries could henceforth record numbers
using Western numerals rather than the Hebrew alphabetic system, whose use had
previously been mandatory in that context (Copans 1999). It is unclear whether
this ruling will have any long-term effect on the use of the Hebrew numerals for

160

Numerical Notation

dating. However, the fact that such a ruling needed to be made at all shows the
continued health and vibrancy of Hebrew alphabetic numerals, albeit in a limited
set of religious contexts. The Hebrew alphabetic numerical notation system is
now over two millennia old, and is one of the oldest systems in continuous and
regular use.

Syriac Alphabetic
In Chapter 3, I described the numerals used alongside the Old Syriac script between about 50 and 500 ad. Around the fifth century ad, this script diverged into
two forms: an eastern variety, Nestorian, used by the Christians of Persia, and a
western variety, the Serto script used by the Jacobite Christians of Syria (Daniels 1996). This split was precipitated by the expulsion of the Nestorian Christians from the Byzantine city of Edessa and their subsequent migration into the
Sassanian Empire (Duval 1881: vii). Soon thereafter, both the Nestorian and Serto
scripts began to use an alphabetic numeral system akin to those used elsewhere in
the Middle East. The basic signs of this system (sign-forms are those of the Serto
script) are shown in Table 5.11.
The system is decimal and ciphered-additive, and written from right to left.
As in many alphabetic numeral systems, numerals were sometimes distinguished
from letters by placing a horizontal stroke above a numeral-phrase, but often no
special mark was present. Like the Hebrew alphabetic numerals, values from 500
to 900 were usually written using the signs for the lower hundreds with the sign
for 400 in various additive combinations. Alternately, the upper hundreds were
occasionally expressed multiplicatively, by placing a small dot above the signs for
50 through 90 to indicate multiplication by 10 (Duval 1881: 15; Noldeke 1904:
316317). For values above 1000, a slanted stroke placed beneath a unit-sign indicates multiplication by 1000, while a horizontal stroke placed beneath a sign indicates multiplication by 10,000 (Hoffman 1858: 67). In this way, any number
below ten million could be expressed. Even higher values could be expressed by
placing two small strokes beneath a sign to indicate multiplication by ten million,
and placing one small stroke above and one small stroke beneath a sign indicated
multiplication by ten billion; however, these techniques were used extremely rarely
and not followed uniformly (Duval 1881: 1516). As in several other alphabetic systems, placing an oblique stroke above a numeral-sign indicated the appropriate
(1/x) unit fraction (Duval 1881: 16).
The Syriac alphabetic numerical notation system was probably invented in the
sixth century ad, and gradually replaced the Old Syriac system over the next two
hundred years (Duval 1881: 15). Independent invention of this system can be ruled
out, given its strong similarity to others used in the region. Two likely possibilities

Alphabetic Systems

161

Table 5.11. Syriac alphabetic numerals

1s
10s
100s

a
j
s

b
k
t

c
l
u

d
m
v

e
n
sv
w\
/e
-e

f
o
tv
x
/f
-f

g
p
uv
y
/g
-g

h
q
vv
\z
/h
-h

i
r
svv
,
/i
-i

OR

/a
10,000s -a
369 = iou
1000s

/b
-b

/c
-c

/d
-d

are that it was modeled on the Greek alphabetic numerals prevalent in the Byzantine Empire or else on the Hebrew alphabetic numerals. The Hebrew and Syriac
numerals are the only two systems in which 400 occupies a special structuring
role, in that the higher hundreds are expressed using additive combinations of
400 and the lower hundreds. The ordering of the Syriac numerals follows the letter-order shared by the Syriac and Hebrew scripts, not that of Greek. Finally, the
Syriac system was written from right to left like the Hebrew system, but unlike
the Greek. On the other hand, the Syrians were closely affiliated with Eastern
(Greek) Christianity, and many Syrians lived under Byzantine rule. The hypothesis of Greek ancestry is supported by the shared feature of the two systems that
the multiplicative principle was used at two different powers of the base, 1000
and 10,000, whereas the Hebrew numerals did so only at 1000. The inventor(s) of
the Syriac numerals probably would have been familiar with both the Greek and
Hebrew numerals, so it is possible that features of both systems were blended in
the Syriac system.
The Syriac scripts were never used as the official script of any polity, and thus
Syriac numerals are rarely found on stone monuments or coins. However, their
use in religious texts is extremely prevalent from the time of their invention to
the present day, a situation afforded by the relative separation of the Jacobites and
Nestorians from both Roman and Greek Christianity. Most Syriac liturgical texts
are dated and paginated using the numerals, making it easy to examine paleographic changes. There is no evidence that the Syriac numerals were ever used in
mathematical texts, nor, in contrast to the Hebrew numerals, were they used for
letter-magic or numerology.
Although the heyday of the Syriac scripts came and went before ad 1000, both
the Nestorian and Serto scripts survive to this day, the former in Iraq, Turkey,
and Iran among a small number of Nestorian Christians, the latter in Lebanon

162

Numerical Notation

among the Maronite Christians of that country. Both scripts retain their distinctive numerical notation systems, albeit restricted in the contexts of their use to
the same liturgical functions for which they have been employed for nearly 1,500
years. Arabic or Western numerals serve most other purposes. Nevertheless, there
is no reason to think that the Syriac numerals are about to disappear, particularly
given the special status accorded to Maronite Christianity in Lebanons 1990 constitution.

Arabic Abjad
In pre-Islamic times, Arabic speakers used a variety of the Nabataean script and
numerals (Chapter 3). While the classical Arabic script is directly descended from
this ancestor, the Nabataean hybrid cumulative-additive/multiplicative-additive
numerals were abandoned in favor of a ciphered-additive system based on the
Arabic script-signs. The basic signs of this system are shown in Table 5.12 (Saidan
1996: 332).
The system is decimal and ciphered-additive below 1000 and, like the Arabic
script, is written from right to left with the signs in descending order. The numeralsigns shown are the unligatured signs of the Arabic consonantary; in numeralphrases, signs were ligatured to one another as appropriate for the letters in question. Often, a horizontal stroke was placed above a numeral-phrase to distinguish
it from an ordinary word. Curiously, the signs are not valued according to the
normal Arabic letter-order, but rather according to the letter-order of the Hebrew
and Syriac scripts, which was also the order used early in the Arabic scripts history.
The first three signs of this order (alif, ba, jim) gave the system its most common
name, h.isa-b abjad.7 Because the Arabic script has twenty-eight basic consonantal
signs, the remaining sign, ghayn (;), was assigned the numerical value of 1000.
Ghayn was not used as a single unit-sign, but as a multiplier in numeral-phrases
by placing another sign before it. In this way, any number up to and including one
million could be written. The system is thus multiplicative-additive above 1000.
The values assigned to six of the abjad numerals were different among users of
the Arabic script in the Maghreb (North Africa and Spain). This ordering developed somewhat later than that used elsewhere, probably in the ninth century ad.
Other than the different values assigned to the six signs in Table 5.13, the system is
structurally identical to the regular abjad numerals.
It is possible that the abjad numerical system is of pre-Islamic origin, and that
it spread from the north. However, it is more likely to have originated around
650 ad, at or shortly after the time of the early Islamic conquests in Syria, Egypt,
7

The first three letters in the modern Arabic script are alif, ba, and ta.

Alphabetic Systems

163

Table 5.12. Arabic abjad numerals


1

b
k
t

c
l
u

d
m
v

e
n
w

f
o
x

g
p
y

h
q
z

i
r
:

1000

a
j
s
;

7642 =

b m x;g (7 1000 + 600 + 40 + 2)

1s
10s
100s

and Mesopotamia. Under Byzantine rule, this region used the Greek alphabetic
numerals for administrative and commercial functions; furthermore, both the
Syriac and Hebrew alphabetic numerals were used in their respective scripts. The
abjad numeral-signs have exactly the same order as the corresponding Syriac and
Hebrew characters up to 400 (above which point the other two systems are structurally irregular). The Arabic script adopted a different letter-order starting in the
eighth century ad, but the older order was retained for the numerals. The Greek
system follows an order similar to that of the Syriac and Hebrew numerals up
to 80, but diverges thereafter by putting sampi (equivalent to the Hebrew tsade)
at the end of the system, rather than in the middle. Thus, either the Hebrew or
Syriac systems inspired the development of the Arabic abjad numerals (Guitel
1975: 276278; Ifrah 1998: 243). Table 5.14 illustrates how the Arabic order is directly parallel to the Hebrew and Syriac, while the Greek numerals diverge from
them starting at 90.
Table 5.13. Arabic abjad numerals (Eastern vs. Maghreb)
Sign

o
r
u
z
:
;

Letter-name

Eastern Value

Maghreb Value

sin

60

300

sad

90

60

shin

300

1000

dad

800

90

900

800

ghayn

1000

900

Numerical Notation

164

Table 5.14. Arabic, Greek, Hebrew, and Syriac numeral-signs and letters
Arabic
60
70
80
90
100
200
300
400
500
600
700
800
900

o
p
q
r
s
t
u
v
w
x
y
z
:

Greek
sin
ayin
fa
sad
qaf
ra
shin
ta
tha
kha
dhal
dad
a

X
O
P
<
R
S
T
U
F
C
Y
W
>\

Hebrew

Syriac

xi

samekh

omicron

ayin

pi

pe

qoppa

tsade

rho

quf

sigma

resh

tau

shin

upsilon

tav

phi
chi
psi
omega
sampi

o
p
q
r
s
t
u
v
w
x
y
z
,

semkat
e
pe
sode
quf
rish
shin
taw

However, the Arabic abjad was rarely used in the same texts as either the Hebrew or the Syriac numerals. In contrast, Greek alphabetic numerals are found in
Arabic documents from the seventh to the ninth centuries ad. In ad 706, Caliph
Walid I dictated that, although his Greek financial administrators in Damascus
should no longer use the Greek alphabet, they could continue to use the Greek
alphabetical numerals (Menninger 1969: 410). In an eighth-century ad Arabic
tax record, numbers are expressed in both Greek alphabetic numerals and abjad
numerals (Cajori 1928: 29). Because many of the regions conquered by the Arabs
even those such as Syria and Palestine in which the Syriac or Hebrew numerals
were found were under Greek rule, Greek numerals were the normal system
used for administration, on coins, and in inscriptions. It is unlikely that the Greek
system played no role in the development of the abjad numerals.
The phonetic values of the final six Arabic characters do not correspond to the
Greek, however. Furthermore, the Greek system has only twenty-seven rather
than twenty-eight signs (lacking a sign for 1000). While the Arabs were aware that
the Greek system had signs for the higher hundreds, and may thus have attached

Alphabetic Systems

165

numerical values to the remaining signs in their own script, the use of a special sign
for 1000 is unique to the Arabic system among all four systems under consideration. More likely, the Arabic system was based on the Semitic letter-order but employed the structural advantages of a system, such as the Greek, with a full complement of numeral-signs. This feature would have been particularly important, since
the administrative needs of the new Islamic caliphate were growing rapidly.
By the late eighth century ad, the Arabic abjad numerals had spread throughout the Middle East and into the Maghreb, including southern Spain. They were
used on administrative documents, in literary and scientific texts, on dated coins,
and on monuments, though not for the most part on ephemeral media such as
ostraca. In areas in which an existing administrative apparatus was retained from
the Byzantine Empire (such as Egypt), the Greek and Coptic alphabetic numerals
were used much more frequently on administrative and financial documents than
were the abjad numerals (Grohmann 1952: 89). While the systems are structurally
similar to the abjad numerals, it would have taken some effort for Arabic writers
to learn an entirely new set of twenty-seven signs, so its failure to be used more
widely is somewhat surprising. Issues of identity and ethnicity may have played a
significant role in determining the scope of their use. For instance, Ifrah describes
a ninth-century ad Christian manuscript written in Arabic but in which the verses
are numbered in Greek (Ifrah 1998: 243). In this case, the writer was a Christian
who associated himself with Greek Christianity through the alphabetic numerals
even though he wrote in the Arabic language and script. Abjad numerals were
rarely if ever used directly for arithmetical calculation. Instead, a system of finger
computing borrowed from the Byzantine Greeks, and known as h.isab al-uqd
(finger-joint arithmetic), was employed (Saidan 1974: 367368), as well as computation on dust-covered boards.
Despite their use over a wide area, the abjad numerals did not give rise to a
large number of descendants. In part, this must be due to the remarkable stability
of the Arabic script itself. The Coptic zimm numerals used in Egypt from the
tenth century ad onward are an interesting blend of Greek and Arabic influences,
although relatively few of the numeral-signs are evidently based on those of the
abjad, and it is unclear to what extent the Arabic abjad affected the structure of
the system. A similar situation probably arose in Morocco, where Fez numerals,
incorporating elements of the Arabic abjad as well as Greco-Coptic alphabetic
numerals, were used until very recently (Colin 1933). Finally, like Greek and Hebrew scholars, Arabic astronomers used a very unusual system for writing fractions, which combines a ciphered-additive system with a base-60 (sexagesimal)
positional notation.
Shortly after they were invented, the abjad numerals began to be supplanted
by the well-known ciphered-positional Arabic system (Chapter 6) borrowed from

166

Numerical Notation

the Hindus, and known in Arabic as al-h.isb al-hind (Indian arithmetic). The
Islamic conquest of enormous territories to the east brought the Arabic and Indian
spheres of influence into close contact by the mid seventh century. As Islam spread
eastward throughout the eighth century ad as far as the Indus River, the Indian
style of numeration began to diffuse westward and supplant the Arabic abjad,
which itself was still a novelty in western regions such as North Africa. This replacement was hastened by the arrival in 773 ad of Hindu astronomers and astronomical knowledge at the court of Caliph al-Mansur in Baghdad, which formed
the basis for the early ninth-century ad writings of the famed Arab mathematician
al-Khwrizm, who popularized the Arabic positional system (Menninger 1969:
410411). By the late ninth century, the positional system was being used in administrative and financial documents, and by the late tenth century, on inscriptions (Grohmann 1952: 89). The latest Arabic papyrus in which abjad numerals
express a date is from 517 ah, or 1123/4 ad (Destombes 1987: 131).
Despite the introduction of positional notation, abjad numerals survived in a
variety of contexts. Astronomers continued to use ordinary abjad numerals much
later than other writers, probably because they had also adapted them for use in
the quasi-positional sexagesimal fractions to be described later. Until the thirteenth
century all astronomical and astrological texts used abjad numerals alone, and for
several centuries thereafter, abjad and positional numerals were often used side
by side in any given text (Lemay 1982: 385386). Abjad numerals were commonly
used on most Arabic astrolabes (both for marking gradations on the instrument
and for recording dates of construction) until the sixteenth century (Destombes
1987). Mauritanian mathematicians did not adopt Arabic positional numerals
until the end of the eighteenth century (Rebstock 1990). That positional numerals would be adopted more rapidly in nontechnical manuscripts than in scientific
ones is seemingly paradoxical, but only because it is customarily assumed that
positional numerals would be accepted owing to their efficiency for arithmetic. In
fact, however, Arabic arithmetic was largely conducted using dust-boards, flat surfaces covered with a layer of fine dust on which calculations could be made using
any numerical system, but of course no record of this would survive; in contrast,
permanent records were often made in more traditional and authoritative numerical systems like the abjad numerals (Lemay 1982: 383384). We do not know to
what extent pre-modern Arabic astronomers may have performed computations
using Arabic positional numerals.
The abjad numerals survive to the present for very limited cryptographic, literary, and magical functions, such as chronograms (described earlier for Hebrew
numerals), known in Arabic as h.isb al-djummal (Colin 1971: 468). Chronograms
using the abjad numerals were common throughout the Middle Ages, particularly in Persia and Islamized parts of India (Ahmad 1973). Babajanov and Szuppe

Alphabetic Systems

167

(2002) have recently published an unusual Persian cryptographic inscription in


Arabic script dating to 1790 from the site of Chr Bakr near Bukhara, Uzbekistan.
It is notable not only for its late date, but also because it records the computational
work performed in translating the name whose numerical value is being taken
(into abjad numerals, then into Arabic figures, then performing arithmetical calculations [using Arabic positional numerals] to double the entire abjad value to
give the number associated with the unnamed writer).
Finally, the abjad numerals were retained for pagination of prefaces and tables
of contents of books, in a manner similar to the Western conventional use of the
Roman numerals (Colin 1960: 97). Abjad numerals survived particularly well in
the Maghreb, and continue to be used there for chronograms, cryptographic correspondence, and functions related to magic and divination. Several abjad-derived
systems were used by nineteenth-century Ottoman administrators for cryptographic purposes (Chapter 10) (Decourdemanche 1899). Similarly, Monteil (1951)
describes a text from Mali that discusses many cryptographic systems derived from
the abjad numerals that were used in North Africa in the mid twentieth century.8
Most modern Arabic grammars mention the existence of abjad numerals and note
the numerical values of the various letters.

Astronomical Fractions
In addition to the ciphered-additive decimal systems peculiar to their civilizations,
ancient and medieval astronomers from various linguistic backgrounds and script
traditions used distinct systems for fractions, combining their ordinary alphabetic
numerals with a base of 60 and the positional principle, borrowed from Babylonian
astronomy. This ingenious system of astronomical fractions represents a curious
digression in the history of numerical notation. From about 2100 bc to 0 ad, Babylonian astronomers and mathematicians used a cumulative-positional numerical
notation system with a base of 60 and a sub-base of 10 (see Chapter 7). Numbers
smaller than 60 were expressed through cumulative combinations of cuneiform
signs for 10 (l) and 1 (k), while the positional principle was used to express multiples of powers of 60 (60, 3600, 216,000, ...). By the third century bc Greeks firmly
controlled most of the lands formerly under Babylonian rule, under the potent Seleucid kingdom that came into existence after the Alexandrine conquests. At some
point between the third and first centuries bc, the Babylonian positional notation
and the sexagesimal base were married to the Greek ciphered-additive numerals and used thereafter by Greek astronomers (Jones 1999: 9). The first Greek
8

Curiously, one of these systems is known as el-Yunani (Ionian!), suggesting that its users
were aware of the Greek origin of such notations.

Numerical Notation

168

Table 5.15. Astronomical numerals (Greek)


1

a
b
10s
i
k
/afie k ie
1s

1515

g
l

d
m

e
n

+ 20 ( 1/60) + 15 ( 1/3600) = 1515.3375

mathematical astronomer to use sexagesimal notation was Hypsicles in the second


century bc (Fowler 1999). The first major text in which this new system appeared
was the Syntaxis of Ptolemy, written in the second century ad (Heath 1921: 4445).
In place of the cumulative Babylonian signs in each position, fourteen of the
Greek alphabetic numerals were used (the units 1 through 9 and the decades 10
through 50) to write any number from 1 through 59, as shown in Table 5.15. Unlike the Babylonian system, however, the Greek sexagesimal system was never used
for expressing integers. Numbers greater than 60 were always written with the
ordinary alphabetic numerals. The sexagesimal numerals correspond to Greek astronomy, in which, as today, the circle is divided into 360 degrees, each degree into
60 minutes, each minute into 60 seconds, and so on. Thus, in Theon of Alexandrias
(fourth-century ad) commentary on Ptolemys Syntaxis, the numeral-phrase /afie
k ie expresses 1515 (/afie) degrees, 20 (k) minutes, and 15 (ie) seconds (Thomas
1962: 5051). The degrees value uses the decimal alphabetic numerals,9 while the
latter two positions are sexagesimal. Sexagesimal fractions could be used to express
any fractional value; successive positions represent 1/60, 1/602, 1/603, and so on, to
express as small a value as desired.
Within each sexagesimal position, decimal ciphered-additive numeral-phrases
represent the value from 1 to 59. However, the primary base of the system the
one involved in the positional aspect of the system is 60; thus, 10 (the base of
the ordinary alphabetic numerals) becomes the sub-base of the sexagesimal fractions. The systems base and sub-base thus follow the Babylonian system, but the
intraexponential principle used in forming the signs of each position is ciphered
rather than cumulative. This system is thus both ciphered and positional, but
it is not identical to ordinary ciphered-positional systems such as the Western
numerals. Rather, because it has a sub-base, it is intraexponentially cipheredadditive rather than simply ciphered. It is thus a (ciphered-additive)-positional
system.
9

If this value were expressed sexagesimally, we would expect it to be written as ke ie, or


(25 60) + 15.

Alphabetic Systems

169

The astronomical numerals used a special sign as a placeholder to indicate an


empty position in order to avoid ambiguity. In some late (fourth- to first-century
bc) Babylonian texts, a placeholder sign that served some of the functions of zero
was used (see Chapter 7). Greek astronomers from the first century ad onward
took up this practice, originally using a sign (, which in later manuscripts was
written as ) (Irani 1955; Jones 1999: 6162). The latter sign is sometimes held
to represent omicron, the first letter of the Greek word ouden, nothing, with a
stroke added above to distinguish it from the appropriate letter (Ifrah 1998: 549).
Yet it is unlikely that the Greeks would have chosen a sign that already had a
numerical value (o\= omicron = 70) in the alphabetic system (Neugebauer 1957:
14; Jones 1999: 61). The zero-sign was probably a paleographic outgrowth of the
earlier form, which was, like the Babylonian placeholder sign, purely ideographic.
As far as can be discerned, however, it is completely unrelated to the later Hindu
zero, which emerged in the sixth century ad in India.
The principle of sexagesimal fractions could, in theory, be combined with any
alphabetic numeral system, replacing the Greek signs for 1 through 59 with those
of any other system. The Arabs, who inherited the bulk of Greek astronomical
knowledge when they took control of Mesopotamia in the mid seventh century
ad, used the signs of the Arabic abjad rather than the Greek alphabetic numeral-signs. Unlike the abjad numerals, the Arabic sexagesimal fractions are written
from left to right, following the Greek practice, and the Arabic symbol for zero
(), or later () is derived from the Greek symbol (Irani 1955). Similarly, medieval
Hebrew astronomers adopted sexagesimal fractions, as in the fourteenth-century
astronomical writings of Levi Ben Gerson (Ifrah 1998: 158). From the Arabic translations of Greek astronomical texts, sexagesimal positional numeration reached
the Byzantine Greeks and, by ad 1000, Western Europe (Neugebauer 1960,
Berggren 2002). Medieval European astronomers used sexagesimal notation for
fractions thereafter, and in the Alphonsine tables edited in Paris in 1327, even
whole numbers were written sexagesimally, the only case I know of where this
was done (Berggren 2002: 364). Medieval Ethiopian astronomers used sexagesimal fractions in their own computations and in translations of Greek and Arabic
documents (Neugebauer 1979). The use of the Greek variant waned after the end
of the Byzantine Empire in the mid fifteenth century, after which Western numerals were used for most purposes. The Arabic version of the system survived
even longer; Irani (1955: 3) lists many texts from the sixteenth and seventeenth
centuries and one from as late as 1788, although I suspect that this latter text is deliberately archaic. Astronomical numerals often occur in the same texts as pure decimal
systems (either ciphered-additive ones such as alphabetic numerals, or cipheredpositional ones such as Arabic or Western positional numerals). In effect, sexagesimal notation is not so much a distinct numerical notation system as it is a technique

170

Numerical Notation

that can be used with any numerical notation system as a means of combining
numeral-phrases for 1 through 59 into a positional notation system for fractions.
The sexagesimal fractions were used only by astronomers and mathematicians
working with astronomical problems and writing in manuscripts. Even so, when
writing non-astronomical material, or when paginating and dating astronomical texts, they used other numerical notation systems. Yet, on astronomical instruments and in astronomical manuscripts, sexagesimal notation alongside alphabetic numerals was employed continuously for well over a millennium. That
sexagesimal notation was used solely for astronomy suggests that the demands of
the discipline led to its retention. The division of the circle into 360 degrees (with
subdivisions of 60 minutes per degree and 60 seconds per minute) is very useful,
since 60 has a large number of divisors.10 Faddegon (1932) showed that this feature
enables quick and easy multiplication and division using sexagesimal fractions.
The utility of sexagesimal fractions must therefore be evaluated in relation to computations involving this specific metrological system. The tenth-century Persian
mathematician al-Biruni reported, however, that because it was inconvenient to
multiply using a 60 60 multiplication table, Islamic astronomers would convert sexagesimal fractions into decimal numbers, multiply them, and then convert
them back into sexagesimal notation (Berggren 2002: 362).
Although sexagesimal fractions are no longer used, modern astronomers still
use the sexagesimal division of the circle, and anyone who can read a digital
clock uses a kind of sexagesimal numeration. While we no longer mix additive
and positional principles in notating time and angles, astronomers continue to
restrict themselves to values under 60 for the division of the sky into segments,
just as everyone is able to realize that thirty minutes pass between 1:50 and 2:20.
These vestiges come to us, via Greek and Arabic sexagesimal fractions, from
the Babylonian custom of numbering by 60. In this way, a peculiar custom
of numeration, useful for astronomy but not much else, influenced how humans perceive and structure time throughout most of the world today. However,
these do not represent a sexagesimal numerical notation system, but simply a sexagesimal division of various metrological units (for angles and for time) that are
then represented with decimal numerals. The notation 11:05 does not mean
665 minutes (11 60 + 5), but simply 11 hours and 5 minutes. That we continue
to measure time and angles in this way is a fascinating issue in the history of
astronomy and of timekeeping.

10

In number theory, 60 is a highly composite number, defined as a natural number that


has more divisors than all the numbers below it (Wells 1986: 127128). The number 60
has twelve divisors: {1, 2, 3, 4, 5, 6, 10, 12, 15, 20, 30, 60}.

Alphabetic Systems

171

Table 5.16. Fez numerals


1

a A
10s
j J
100s s
766 = foy
1s

b B c
d D e
f
k
l L m
n N o
t T u
v V w W x X

g G h H i I
p
q Q r
y
z
, <

Fez Numerals
In the western extremity of the Muslim world, first briefly in medieval Spain, then
around the city of Fez in modern Morocco starting in about the sixteenth century,
an alphabetic numeral system was used, distinct from both the Arabic and Greek
systems. This system was known as isb al-qalam al-Fs (Fez signs) or rm signs
(Roman, the name given to the Byzantine Greeks) (Guergour 1997: 68). The
numeral-signs, including paleographic variants where appropriate, are indicated
in Table 5.16 (Colin 1933: 199201).11
The system is decimal and ciphered-additive, and is written from right to left
with the highest values on the right. The twenty-seven signs are thus sufficient to
express any number smaller than 1000. The resemblance between the signs for
6 and 7 and the modern Western numerals is likely a coincidence. For higher
numbers, a stroke placed to the left of or beneath any of the twenty-seven signs
indicates that its value should be multiplied by 1000. In some documents, two
strokes placed underneath a figure indicated multiplication by one million (1000
1000) (Guergour 1997: 69). The Fez system is thus a hybrid multiplicative-additive
system for values above 1000. At least in later periods, there were signs for specific
fractions, and for specific monetary values (Sanchez Perez 1935).
The earliest version of the Fez numerals was used among the Mozarabs (Arabic
Christians) of Toledo, Spain, in the twelfth and thirteenth centuries (Colin 1933:
204). Levi Della Vidas (1934) study of these documents, which includes a table of
these numeral-signs, confirms that they are essentially identical to the later ones
except that they are written from left to right. The question then arises how these
numerals reached Spain in the twelfth century. Three possible ancestors the Arabic
abjad (Maghreb variant), the Greek minuscule alphabetic numerals, and the Coptic zimm numerals are depicted alongside the Fez numerals in Table 5.17.

11

See also Sanchez Perez (1935) and Guergour (1997) for different paleographic variants of
this system.

Numerical Notation

172

Table 5.17. Arabic abjad, Greek, zimm, and Fez numerals


Arabic Greek Zimm Fez Numerals
1
2
3
4
5
6
7
8
9
10
20
30
40
50

a
b
c
d
e
f
g
h
i
j
k
l
m
n

A
b\
G
D
E
V
z\
E
Q
I
k
l
m
n

a A
b B
c
d D
e
f
g G
h H
i I
j J
k
l L
m
n N

Arabic Greek Zimm Fez Numerals


60
70
80
90
100
200
300
400
500
600
700
800
900

r
p
q
z
s
t
o
v
w
x
y
:
;

x
o
p
<
r
s
t
u
f
c
y
w
>\

o
p
q Q
r
s
t T
u
v V
w W
x X
y
z
, <

All four systems are written cursively and have an enormous amount of variation. It is possible that the Mozarabs numerals are paleographic variants of the
Greek alphabetic numerals, and thus came to the Arabs of Spain via direct diffusion from the Byzantines (Levi Della Vida 1934: 283). The attribution of these
signs as rm (Byzantine) would tend to confirm this origin. However, many of the
numeral-signs bear no resemblance to the Greek alphabet. Colin suggests, rather,
that the Fez numerals (and their Spanish antecedent) were borrowed, not directly
from the Greek alphabetic numerals used in the Byzantine Empire, but by way of
the Egypto-Coptic zimm numerals (Colin 1933: 213). The paleographic similarities
between several of the Coptic and Fez numeral-signs (e.g., 8, 80, and 500) suggest
that some connection must exist between the two. Yet while the Fez numerals are
multiplicative at only one level (1000), both of these candidates for its origin are
multiplicative at both 1000 and 10,000. An alternate ancestor is the Arabic abjad
system used for number-magic and astronomy in the Maghreb at that time. Some
of the paleographic resemblances between the Fez numerals and the zimm numerals (e.g., the signs for 7, 30, and 90) can be explained by both systems connection
to the appropriate letters of the Arabic abjad. Furthermore, unlike the other two

Alphabetic Systems

173

systems, the Arabic abjad is multiplicative only in combination with 1000. Most
likely, the Fez numerals are an unusual blend of the Greek, zimm, and Arabic
alphabetic systems adopted among a very unusual group of users, highly educated
Arabized Christians living in Muslim-dominated southern Spain.
The Fez numerals did not last long in Spain; I know of no texts from the fourteenth
century or later in which they are used. They were described by the Moroccan mathematician Ibn al-Banna in the early fourteenth century, and later by the great historian
Ibn Khaldun in the late fourteenth century, both of whom lived and worked at Fez
(Colin 1933: 206; Guergour 1997: 68). Their use in Morocco began in earnest only in
the sixteenth century, however, after the expulsion of the Moors from Spain in 1492.
They were used frequently throughout the sixteenth and seventeenth centuries in accounting and other commercial contexts, after which time they began to be replaced
by the Arabic positional numerals. While they were known to scholars, their use was
always mercantile and legal, never mathematical or astronomical (Guergour 1997: 74).
The Fez numerals were still used in the early twentieth century for indicating monetary
values in wills and in related legal documents (Colin 1933, Sanchez Perez 1935). Because
the meaning of the numerals was known to only a few learned notaries by this time,
the system had become a cryptographic notation to prevent fraudulent modifications
and forgeries, and generally to restrict access to information (Colin 1933: 195). Political
changes in post-colonial Morocco have ended this systems use.

Armenian
Before the introduction of Christianity, there was no native Armenian script, and
the Babylonian, Greek, and Old Persian scripts were used for literary purposes. The
Armenian adoption of Christianity in the early fourth century ad was followed
by enormous influence from the Greek-speaking world. In the early fifth century ad (probably in 406 or 407), the Armenian scholar-monk Mesrop Mashtots
(c. 360440) developed the first uniquely Armenian script, an alphabet of thirtysix letters, in order to translate the Bible from Greek into Armenian (Sanjian 1996:
356).12 At the same time, the letters of the alphabet were assigned numerical values
as shown in Table 5.18.
These signs are the erkatagir iron-forged letters preferred from the fifth through
thirteenth centuries ad, and still used for epigraphic inscriptions (Thomson 1989;
Sanjian 1996: 357). In the tenth century, cursive letters known as bolorgir began to
be used, and are the standard forms used in modern Armenian writing. The Armenian system is ciphered-additive and decimal, and is written from left to right.
12

The modern Armenian script has thirty-eight letters, the last two of which (o and f)
were introduced in the medieval period and have no numerical value.

Numerical Notation

174

Table 5.18. Armenian erkatagir numerals


1

A
10s
J
100s S
1000s 2
346 = UMF

1s

B
K
T
3

C
L
U
4

D
M
V
5

E
N
W
6

F
O
X
7

G
P
Y
8

H
Q
Z
9

I
R
1
0

Because the ancient Armenian alphabet had thirty-six letters, it had enough signs
to express the complete series from 1000 to 9000 as well as all the units, tens, and
hundreds. The system could thus denote any number less than 10,000. However,
unlike many ciphered-additive alphabetic systems, the Armenian system does not
use multiplication to express higher values, which were written in full using lexical
numerals.
Very little epigraphic or paleographic evidence survives from the earliest centuries of the systems use. The Armenian numerals were probably developed on the
model of the Greek alphabetic numerals, just as the Armenian script itself was derived from the Greek. Many other scripts have been suggested as possible ancestors
of the Armenian script, based on resemblances in the shapes of certain characters
(Gamkrelidze 1994: 37), while there are few resemblances to the Greek alphabet.
However, of these likely ancestors, only the Greek alphabet used appropriate alphabetic numerals. Thus, regardless of the origins of the script-signs, the principle of
alphabetic numeration was certainly borrowed from Greece. It is unclear whether
the Armenian alphabetic numerals were developed by Mesrop Mashtots himself (or
his assistants) in the early fifth century ad, or whether they were produced later in
the century. Figure 5.2 is a monumental grave inscription from the temple of Garni
east of Yerevan, which commemorates a ninth-century Armenian Catholicos, also
named Mashtots; the first three signs are numerals (300 + 40 + 6), indicating his
death-year to be 346 according to the Armenian calendar, or 897 ad.
Although a connection is sometimes asserted to exist between the Armenian
and Georgian alphabetic numerals, the evidence for this is too tenuous to suggest
any definite link. The primary similarities between the two are that they were used
in the same region and had distinct signs for 1000 through 9000. The only system
that is derived from the Armenian alphabetic numerals is the variant Armenian
system developed in the seventh century ad by Anania Shirakatsi. The Armenian
numerals did not spread beyond the limited area around Lake Van where Armenian was spoken, nor do they appear to have inspired the creation of any foreign

Alphabetic Systems

175

Figure 5.2. Grave inscription of the Armenian Catholicos Mashtots (897 ad); the first
three visible signs of the inscription are 300 + 40 + 6, the year of his death. Courtesy Gabriel
Kepeklian.

systems. After the development of the minuscule Armenian script, these signs
were also used numerically in the same manner.
Ciphered-positional numerals the Arabic system used by the neighboring
Seljuk Turks were first used in Armenia in the twelfth century (Shaw 193839:
368). Yet Armenian writers retained the alphabetic numerals for most ordinary
purposes long afterward. Only in the mid seventeenth century, when Armenia had
been firmly under Ottoman control for some time, did ciphered-positional numerals (Arabic, then later Western) replace the alphabetic system. Wingate (1930)
has published an unusual, undated, and unsolved magic square arithmetical
puzzle using both Arabic positional and Armenian alphabetic numerals, part of
a scroll contained within a Armenian family amulet designed to be worn upon
the person. The Armenian alphabetic system is still sometimes used for numbering chapters of the New Testament, although page and verse numbers are most
often written using Western numerals. Otherwise, the numerals used in modern
Armenia are the standard Western numerals.

Shirakatsis Notation
The Armenian astronomer, geographer, and mathematician Anania Shirakatsi13
was born ca. ad 595600 and was most likely a monk in the Armenian Church
(Hewsen 1968: 34). While little-known today outside his native country, Shirakatsis contribution to Armenian learning is unparalleled, particularly his synthesis
13

Also known as Ananiah Shiragooni, or Ananias of Shirak.

Numerical Notation

176

Table 5.19. Armenian numerals: Shirakatsis notation

1s
10

A
j

D
s

G
2

100

1000

of Persian, Arabic, Greek, and other scientific knowledge. In addition to these accomplishments, Shirakatsi developed a very interesting numerical notation system
in his Book of Arithmetic (Tuabantiwn), which is a collection of arithmetical
tables designed for the instruction of pupils. The basic form of this system uses
twelve signs, as shown in Table 5.19 (Shaw 193839: 270).
The individual signs are identical to those used for the appropriate numbers
in the traditional Armenian system. However, Shirakatsi showed how these signs
could be combined to express numbers through multiplication as well as addition.
In this system, a unit-sign followed by one of the three power-signs (for 10, 100,
or 1000) indicates that the values of the two should be multiplied; these pairs of
signs were combined into numeral-phrases through addition. Instead of writing
9642 as 0XMB (9000 + 600 + 40 + 2), as in the traditional Armenian alphabetic
numerals, Shirakatsi would write the same number as I2FSDjB (9 1000 + 6
100 + 4 10 + 2). Thus, where the traditional Armenian system is ciphered-additive, Shirakatsis system is multiplicative-additive.
Any numeral-phrase can be written more compactly with the traditional alphabetic numerals than with Shirakatsis variant so why would Shirakatsi advocate
its use? Firstly, it requires knowing fewer symbols (twelve versus thirty-six) in
order to express any number less than 10,000. More importantly, numbers greater
than 10,000 could be expressed using multiplicative combinations of two or three
signs. To do so, however, one needs the entire repertoire of Armenian numerals from 1 through 9000, as described earlier. For numbers from 10,000 through
90,000, Shirakatsi juxtaposed the signs for 10 through 90 with the sign for 1000.
Similarly, the numeral-phrases for 100,000 through 900,000 combined the signs
for 100 through 900 with the sign for 1000. Alternatively, the hundred thousands
could be expressed using unit-signs followed by a 100-sign and then a 1000-sign.
Thus, one could write 460,000 as VO2 (400 + 60) 1000 or DSO2 ((4
100) + 60) 1000. This system is no longer a purely decimal system, but has a
mixed base of 10 and 1000. For values below 1000, it is purely multiplicativeadditive, but above 1000, the multiplicand that is juxtaposed with the sign for
1000 (2) is not a single sign, but rather a ciphered-additive numeral-phrase. In
the Book of Arithmetic, numbers up to the ten millions are expressed relatively
compactly (Abgarian 1962: 46; Hewsen 1968: 42).

Alphabetic Systems

177

Shaw (193839: 369) believes that this system was not developed by Shirakatsi in
the seventh century, but was a commonly used variant system, of which Shirakatsis writings are the only surviving remnant. I do not believe there is any reason to
regard the system as anything other than the creation of Shirakatsi himself, since
its structure is never found in Greek, Syriac, Hebrew, or any other alphabetic system. Shirakatsi may have borrowed the notion of multiplicative structuring from
one of two foreign sources. The numerals developed by the fifth century ad Indian
mathematician ryabhata (Chapter 6) were multiplicative-additive; it is possible that Shirakatsi, a mathematician with extensive knowledge of foreign writers,
knew of ryabhatas numerals and emulated them. Similarly, it is vaguely plausible
that Shirakatsi knew of the classical Chinese multiplicative-additive numerical notation system (Chapter 8). Neither hypothesis has any direct evidentiary support.
Shirakatsis system is thus a structurally innovative local variant of the Armenian numerals designed to facilitate the representation of large numbers of the sort
that would be needed for his astronomical and mathematical calculations. There
is no evidence that his system was adopted by any later writers, or that it had any
effect on the development of other numerical notation systems throughout the
world. Instead, we should view this system as the creative invention of a single
individual, used only within his lifetime.

Georgian
Like the Armenians, the Georgians developed a script and numerical notation
system modeled after the Greek alphabet shortly after they converted to Christianity. While the creation of this first Georgian alphabet is often attributed in
folklore to King Parnavaz in the third century bc, there is no direct evidence of
Georgian writing until the fifth century ad, at which time the asomtavruli or majuscule script began to be used (Holisky 1996). More familiar to modern scholars,
however, are the mxedruli characters developed in the eleventh century ad, which
continue to be used to write the Georgian language today. The numerals associated with this script are shown in Table 5.20 (Holisky 1996: 366).
The system is decimal and ciphered-additive and, like the Georgian script,
is written from left to right. Like the Armenian script, the Georgian script had
enough letters to serve for all numerical values up to 9000. Some later inscriptions
even include a special sign for 10,000 (-). There is no evidence that the Georgian
alphabetic numerals were ever used to express larger numbers than this, either
through multiplication or through additional signs. Presumably, such numbers
were always written out in full using lexical numerals.
There is an undeniable structural similarity between the Georgian and Armenian systems, which both, unlike the Greek alphabetic numerals, have enough

Numerical Notation

178

Table 5.20. Georgian numerals


1

a
10s
j
100s
s
1000s 2
4808 = 5zh
1s

b
k
t
3

c
l
u
4

d
m
v
5

e
n
w
6

f
o
x
7

g
p
y
8

h
q
z
9

i
r
1
0

additional letters to represent the values from 1000 through 9000. However, while
the Georgian and Armenian scripts both use thirty-six signs for 1 through 9000,
the letter-order of the two scripts is vastly different. The Georgian letter-order
was modeled very closely on the Greek, with additional signs added as necessary
at the end of the series, while the uniquely Armenian signs in that script were interspersed randomly within the original Greek letter-order. It is unlikely that the
Georgian numerals would be modeled on the Armenian numerals but retain the
Greek letter-order for their values (Gamkrelidze 1994: 77). There may have been
some mutual influence between the two numerical notation systems, given certain
similarities in the sign-forms, but the direction of this influence remains unclear
(Gamkrelidze 1994: 8182). For now, the hypothesis of direct diffusion from the
Greek alphabetic numerals, while using distinctly Georgian signs and using distinct signs for the thousands, is most strongly supported.
The Georgian numerals were used in literary and religious texts throughout the
medieval period, particularly for pagination, dating, and stichometry, as well as in
monumental inscriptions. Their regular use ended in the sixteenth century, when
Georgia came under Ottoman control, after which Arabic positional numerals
were used for administrative and commercial purposes, although the alphabetic
numerals may have been retained for religious functions. However, Paolini and
Irbachs 1629 Georgian-Italian dictionary, the first book printed in Georgian, does
not contain any mention of the alphabetic numerals alongside its list of Georgian
letters. Since the eighteenth century, when Georgia fell under the Russian sphere
of influence, the Western numerals have been those normally used for all purposes
in written Georgian.

Glagolitic
The Glagolitic script was probably developed between 860 and 870 by the brothers Cyril and Methodius, who, while on a mission to the Moravian Slavs of what
is today modern Serbia, Croatia, and Macedonia, created an alphabet for liturgical

Alphabetic Systems

179

Table 5.21. Glagolitic numerals


1

a
j
100s s
708 = yh

1s

10s

b
k
t

c
l
u

d
m
v

e
n
W

f
o
x

g
p
y

h
q
z

i
r
/

writings in the language now known as Old Church Slavonic (Schenker 1996:
166167). There may have been a pre-Christian script in the region, which might
explain why many of the Glagolitic letters have no correlation with the Greek
alphabet, but no pre-Christian numerical notation existed (Cubberley 1996). The
numeral-signs of the Glagolitic numerical notation system are shown in Table 5.21
(Vaillant 1948, Gardiner 1984).
As with the Greek and many other systems, Glagolitic numerals were frequently
distinguished from words in texts by placing dots to either side of a numeralphrase or by placing a mark of some sort above it (Vaillant 1948: 24; Schenker
1996: 182). In addition to these twenty-seven signs, additional signs for 1000, 1,
and 2000, 2, were purportedly used in some texts. The system is ciphered-additive
and decimal, and is always written from left to right. However, for the numbers 11
through 19, the ordinary sign order is reversed (e.g., bj instead of jb for 12),
which reflects the Slavic lexical numerals for the teens (Schenker 1996: 182).
Yet none of the surviving Glagolitic manuscripts apparently use numerals higher than 1000 (Gardiner 1984: 15; Lunt 2001: 28). Schenker (1996: 182) contends
that the Glagolitic thousands were expressed by placing a small diagonal or curved
stroke (like the Greek hasta) to the left of a numeral-sign to indicate that its value
should be multiplied by 1000; if so, Glagolitic is a hybrid multiplicative-additive
system above 1000. Gamkrelidze (1994: 3940) and others contend that, because
the earliest Glagolitic script had thirty-six characters, the last nine letters of the
alphabet (most of which were later dropped from the script) originally had the
values 1000 through 9000. The issue remains unresolved and contentious.
The Greek alphabetic numerals were the sole and direct ancestor of the Glagolitic
numerals. The similarities in structure between the Greek and Glagolitic systems
are considerable, and Cyril, Methodius, and their followers were Greeks. Among
other possible ancestors, the Gothic numerals were long defunct by the ninth
century ad, and the Cyrillic numerals were not invented until later in the century.
Nickels (1973) suggestion that the Glagolitic numerals may have originated from
the tamgas (clan identifiers) used by Turks and Iranians in southern Russia as early
as the first century is problematic. While a variety of scripts, such as the Latin,

180

Numerical Notation

Greek, Samaritan, and Hebrew, may have been used as the model for one or more
of the letters of the alphabet, many other signs have no obvious correlates in other
scripts (Schenker 1996: 168172). Regardless, the Glagolitic letters must have been
assigned their numerical values under the influence of Greek Christianity.
Manuscripts were written in Glagolitic throughout the medieval period in the
region of modern Croatia, Serbia, Slovakia, and even into the Czech Republic
and Poland. Yet, even during the Middle Ages, Catholic or Western-influenced
areas began to prefer the Roman numerals to the Glagolitic, while areas under
Bulgarian or Serbian control tended to adopt the Cyrillic numerals and script. By
the fifteenth century, almost all the Slavs had adopted either Roman or Cyrillic
numerals. Only in Croatia, particularly along the Adriatic coast (Dalmatia), where
they were retained for the Croatian Roman Catholic liturgy (Cubberley 1996:
350), did the Glagolitic script and numerals flourish. They were also used in monumental inscriptions in Croatia from the eleventh century onward, a context not
seen elsewhere. Yet, even in Croatia, the Glagolitic script and numerals declined
greatly in use after the Ottoman conquests of the sixteenth century, and were used
only rarely from the seventeenth century onward (mostly in religious texts). It is
not clear whether the Glagolitic numerals survived as long as the Glagolitic script,
which persisted until the beginning of the twentieth century in the Quarner archipelago in northwestern Croatia.

Cyrillic
Like Glagolitic, the Cyrillic script was developed under the guidance of the missionaries Cyril and Methodius. It is quite likely that Cyrillic was developed in ad 890
900, after the deaths of Cyril and Methodius, by Cyrils followers and disciples in
Bulgaria, who then named the script after their deceased mentor. Cyrillic was originally used for writing the Old Church Slavonic language, but later was adopted for
writing a variety of Slavic languages, most notably Russian. An alphabetic numerical
notation system14 was developed around the same time. The Cyrillic numeral-signs
are shown in Table 5.22 (Gardiner 1984: 1617; Cubberley 1996: 348).
The system is ciphered-additive and decimal, and is normally written from left
to right. For the numbers 11 through 19, the ordinary sign order was often reversed
(e.g., bj instead of jb for 12), which reflects the structure of Slavic lexical numerals
(Vaillant 1948: 24). Numeral-phrases were often distinguished from ordinary letters by placing a bar or other mark (titlo) above the phrase, and sometimes also by
14

While some scholars call this system Slavonic, I use the term Cyrillic to prevent
confusion with the Glagolitic system.

Alphabetic Systems

181

Table 5.22. Cyrillic numerals


1

A
10s
J
100s S
708 = yh

1s

b
k
t

c
l
u

d
m
/

e
n
w

f
o
x

g
p
y

h
q
z

i
r
0

placing dots on either side of the signs (Lunt 2001: 28). Placing a small stroke
to the left of a number indicated that its value should be multiplied by 1000
(Vaillant 1948: 24; Schenker 1996: 182). The Cyrillic numerical notation system is
thus a hybrid: purely ciphered-additive below 1000 and multiplicative-additive for
higher powers. In some cases, higher Cyrillic numerals were expressed by using the
signs preceded by an unusual sign, [, to indicate multiplication by 1000 (Gardiner
1984: 15). Apparently in some cases the multiplier-stroke could be repeated two
or three times to indicate multiples of 1,000,000 and 1,000,000,000, respectively
(Berdnikov and Lapko 1999: 16). In the earliest phase of the scripts history, there
were a number of ideograms for powers of 10 from 10,000 up to 1,000,000,000,
but these rare notations are poorly studied, and the range of their use is unknown
(Berdnikov and Lapko 1999: 17).
While there are only twenty-seven signs listed in Table 5.22, there are more than
twenty-seven signs in all varieties of the Cyrillic script; modern Russian Cyrillic
uses thirty-two letters, and earlier Cyrillic scripts used a number of older signs
that have now fallen into disuse. The signs that are assigned numerical values in
Cyrillic are those that are directly derived from Greek, including the otherwise
rarely used signs for xi (o), psi (y), and theta (i). Yet numerical values were
never assigned to the commonly used but non-Greek characters (Gardiner 1984:
1415). Thus, the Cyrillic numerical values do not correspond to the customary
order of letters, remaining faithful to the Greek order instead.
The Cyrillic script and numerals originated around 890 ad, at which time Slavs
and Greeks who had been influenced by Cyril and Methodius were extremely
active in the Christianization of the Slavs in the region of modern Bulgaria and
Serbia. That this missionary work was undertaken under the auspices of the Byzantine Empire confirms what is clear from the paleographic evidence that the
sole external influence on the Cyrillic script and numerals was the Greek uncial
alphabet used at the time. The non-Greek signs for additional consonantal Slavic
phonemes were never assigned numerical values, further confirming the Greek
origin of the Cyrillic numerical notation system.

182

Numerical Notation

The Cyrillic numerical notation system spread to Kievan Rus in the tenth century. The earliest printed books in Cyrillic script, those printed in Krakw from
1491 onward, were paginated using Cyrillic rather than Roman or Western numerals (Zimmer et al. 1983). The Balkans fell under Ottoman influence in the fifteenth
century, after the fall of Constantinople, and the alphabetic numerals largely ceased
to be used there by around 1500. In Russia, the Cyrillic numerals were used much
longer. The late sixteenth-century English Slavist, Christopher Borough, authored
a Russian copy of the pseudo-Aristotelian Secret of Secrets (Bodleian MS Laud
misc. 45 (SC 500)), which was paginated in Western numerals but used Cyrillic
numerals and Arabic positional numerals elsewhere in the text (Pennington 1967:
681682). Western numerals were known in Russia in the seventeenth century; a
seventeenth-century sundial from Mangazeia (in Siberia, near the Arctic coast)
is numbered using Western numerals, showing that they were known even in
the Russian hinterland (Ryan 1991: 375). Yet the popular Schitanie udobnoe ready
reckoner arithmetic text of 1682, intended for merchants, contained only Cyrillic
numerals (Okenfuss 1973: 329). For most purposes, the schety or Russian beadabacus, which had been used in Moscow since the eleventh century, was perfectly
adequate for the computational needs of pre-Petrine Russia (Ryan 1991: 373374;
Simonov 1993).
Not until the reforms of Peter the Great around 1700, and the introduction
of technical training in mathematics by scholars from Western Europe, were the
Western positional numerals introduced into Russia on a widespread basis (Hans
195960, Fedosov 1995). Magnitskiis Arifmetika of 1703 used both systems side by
side (Ryan 1991: 373). In the same year, logarithm tables published for the use of
students at the Moscow School of Mathematics and Navigation were printed in
Western numerals only. In 1710, however, Peter deferred to the clergy in decreeing
that church texts were to be printed using the traditional Cyrillic numerals (Cracraft 2003: 103). Throughout the eighteenth century, aristocrats and literate officials
would have needed to be familiar with Roman, Western, and Cyrillic numerals in
order to read the full range of textual genres used in Russia (Billington 1968: 209).
As late as 1918, Tsaritsa Alexandra (Alix of Hesse) was learning Cyrillic numerals
and using them to paginate her final diary, demonstrating that their use was still
relevant, if increasingly formal, in the late tsarist period (Kozlov and Khrustalev
1997: 23). Many texts were paginated using the older system even as Western
numerals increasingly were employed for a wider variety of functions. Unlike the
transition in Western Europe, where the Western numerals took centuries to be
adopted fully, the change from the Cyrillic to the Western numerals took place
relatively smoothly and rapidly, however. Today, the Cyrillic numerals are occasionally used in modern Church Slavonic texts (especially for numbering chapters and
verses in Bibles), but never in ordinary Cyrillic writing (Gasparov 2001: 1718).

Alphabetic Systems

183

Latin
From the fifth through the twelfth centuries, knowledge of the Greek exact sciences in Western Europe was relatively limited. Starting in the eleventh century,
as Arabic translations of Greek astronomical works began to be retranslated into
Latin, Western European scholars became increasingly aware of the Greek numerals. In a handful of texts, an attempt was made to convert the Greek alphabetic
numerals or Arabic abjad numerals for use with the Latin alphabet, as shown in
Table 5.23 (cf. Lemay 2000, Burnett 2000c).
The system is ciphered-additive and written from left to right, and simply employs twenty-two letters of the Latin alphabet as it existed at the time to represent
1 through 9, 10 through 90, and 100 through 400 (e.g., xlh = 328). Additionally,
in the Dresden Almagest of 1121, z, , and represent 500 through 700 (Burnett
2000c: 61). There is, however, no known way to write 800 and 900, and there is
no known multiplicative technique to express numbers higher than 1000. In one
of the texts containing such an alphabetic system (MS Cambrai Bibliothque Municipale 930, a copy of Hermann of Carinthias Astronomia), even the sign for 400
is not used (Lemay 2000: 378379). Yet, because the texts containing the Latin
alphabetical numerals are astronomical treatises, in which numbers higher than
the 360 degrees of the circle are rarely needed, this was not a serious weakness of
the system, which was perfectly adequate for such values. While minuscule letters
were used most of the time, in one text, majuscule letters for K, M, and N, and
occasionally for G, R, Q, and L were used, in order to prevent confusion with
similar-appearing Western numerals (Burnett 2000a: 82).
Burnett (2000c: 6162) briefly discusses a few tenth- to twelfth-century astrolabes and manuscripts with Latin letters serving as numerals, but these are direct
transliterations of Arabic or Hebrew numerals that follow the Semitic alphabetic
order rather than the Latin one. The earliest documents containing the Latin alphabetic numerals in Table 5.23 date to 1121 and 1127 and were copied in Antioch
in the Crusader States, probably in association with the work of Stephen of Pisa,
an early twelfth-century translator of Arabic scientific texts. These contain no key,
suggesting that the system would already have been understood by its intended
audience (Burnett 2000a: 76). The copy of Hermann of Carinthias Astronomia
similarly contains no key, and simply uses Latin alphabetic numerals to annotate
astronomical diagrams (Lemay 2000). A later manuscript (MS London, British
Library, Harley 5402), dated 1160 and written in a mixture of Latin and Italian,
does, however, have a key for translating Roman numerals into the Latin alphabetic system (Burnett 2000a: 76). All of these manuscripts use Roman numerals
in the text and for nonastronomical purposes such as column labels, and some of
them use positional numerals as well. The Liber Mamonis uses Roman numerals

Numerical Notation

184

Table 5.23. Latin alphabetic numerals

1s

1
a

2
b

3
c

4
d

5
e

6
f

7
g

8
h

9
i

10s

100s

for single digits and some low compounds, Latin alphabetic numerals for many
two- and three-digit compound numbers, and represents larger numbers using
Arabic positional numerals (not Western numerals) (Burnett 2000c: 6465).
The Latin alphabetic numerals represent a brief and abortive attempt to adapt
the Greek and Arabic systems to the Latin alphabet for the sole purpose of translating astronomical documents efficiently. They did not, however, lead to a consistent or long-lived tradition of Latin alphabetic numeration. By the time the
Latin alphabetic system was developed, ciphered-positional numeration was already being widely adopted by Western European mathematicians and scholars,
rendering the alphabetic system obsolete. I know of no thirteenth-century or
later documents that use the system described here. However, two late thirteenthcentury manuscripts probably written in Flanders use the letters A through I in
place of 1 through 9, in conjunction with 0, in a ciphered-positional alphabetic
system written from right to left (e.g., B0G = 702), parallel to but probably independent from ibn Ezras use of Hebrew alphabetic numerals in the same way,
discussed earlier (Burnett 2000b: 200; 2000c: 6263).
In some medieval Latin texts from Western Europe, a set of unusual lettersymbols were associated with numerical values in a sporadic and nonsystematic
way, of which one example is shown in Table 5.24 (Cappelli 1901: 413421).
In some respects, this notation resembles the ciphered-additive Latin notation.
These letter-symbols occur in very different contexts, however in the same texts
as Roman numerals rather than in translations of Greek or Arabic astronomical
texts. The only letters that were not assigned values in this system were I/J, U/V,
X, L, C, D, and M, for obvious reasons these already had numerical values in
the Roman numeral system. Like the corresponding Roman numerals, placing a
bar above any of these abbreviations indicated multiplication by 1000; e.g., =
11,000. Unlike the Latin alphabetic numerals or the Roman numerals, however,
these abbreviations could not be combined to form numeral-phrases that is, one
Table 5.24. Medieval Latin numeral abbreviations
A
B
E
F G
H K
N O P
Q
R S T
Y
Z
500 300 250 40 400 200 151 90 11 400 500 80 70 160 150 2000

Alphabetic Systems

185

could not write HN for 290 and be understood. Some numbers are represented
twice (G and P both equal 400); many multiples of powers of 10 (30, 60, 600)
are not represented; and strange numbers such as 151 are assigned letters. In other
manuscripts (e.g., the tenth-century De inventione litterarum in Strasbourg, Bibliothque Nationale et Universitaire MS. 326), the numerical associations of the
letters are completely different than the ones in Table 5.24 (Derolez 1954: 332335).
In fact, this form of notation is not related to either the Greek or Latin alphabetic
numerals, but instead is a complex but unstructured quasi-numerical abbreviatory
system.

Summary
Alphabetic numerical notation systems originated with the Greeks in the sixth
century bc, who combined the structure of the Egyptian demotic system with
the idea of using phonetic signs as numeral-signs. The political and ecclesiastical authority of Greek speakers, coupled with the brevity and adaptability of
ciphered-additive numeration, led to the development of other alphabetic systems
using numeral-signs specific to each script. This phylogeny expanded tremendously
between the fourth and seventh centuries ad (the time of greatest Eastern Roman/
Byzantine power), with eight new systems arising during this period. Yet most
systems of this phylogeny had died out, or had at least been greatly reduced in the
contexts of their use, by the sixteenth century ad, during which time the Arabic
and Western positional numerals replaced them. Many alphabetic systems are still
used today, but only in limited contexts such as liturgical texts, numbered lists,
and divinatory magic.
There is no one feature common to all the alphabetic systems. Most are cipheredadditive and decimal, with or without the use of multiplicative structuring for the
higher powers. However, the Armenian notation of Shirakatsi is multiplicativeadditive and sometimes uses a base of 1000, while the astronomical fractions are
positional and involve a sexagesimal base. The number of signs used, the degree
to which multiplication is used, and the correspondence of numeral-signs with
script-signs are all variable.
One of the great advantages of alphabetic systems is that, if the signs are ordered using a local script, one need not learn both a set of script-signs and a set
of numeral-signs; one merely superimposes the decimal structure of the numerals
onto the script, thereby lessening the mnemonic burden on both new learners and
experienced users. The Greek alphabetic, Coptic, Gothic, Hebrew, Syriac, Armenian, Georgian, Glagolitic, and Latin systems all take advantage of this feature.
However, the values assigned to Arabic and Cyrillic letters do not correspond to
the customary letter-order, thus reducing this benefit. The Fez numerals and the

186

Numerical Notation

zimm numerals are blended alphabetic systems, combining the numeral-signs of


two or more existing alphabetic systems to create a new system that does not correspond with the sign-order of any script. Finally, in the Ethiopic system, the users
of one script adopted the ordered numeral-signs of another (in this case, the Greek
alphabet) rather than adopting both the script and numerals.
Despite the advantage of combining phonetic and numerical representation
systems, alphabetic numerical notation systems require many more signs than
the cumulative-additive systems of Chapters 2 through 4. Even the Ethiopic system, which is multiplicative above 100, uses nineteen separate signs, more than
any cumulative-additive system, and the Armenian and Georgian systems require
thirty-six signs. Decimal ciphered-additive systems require nine signs for each
power: twenty-seven signs to express all numbers up to 1000. In the case of the
Hebrew and Syriac systems, whose scripts had only twenty-two signs, numerals
above 400 were expressed through cumulative combinations of hundred-signs.
While this solves the problem of having only twenty-two phonetic signs, it makes
numeral-phrases longer and more complex.
As it is inconvenient to develop nine new signs for each higher power of 10,
many alphabetic systems are ciphered-additive for lower powers but use multiplicative-additive structuring above some specific point, a feature originally borrowed by the Greeks from the Egyptian demotic numerals. The Gothic, Armenian, Georgian, and Latin systems do not use multiplication at all, and express
only numbers below 1000 (Gothic) or 10,000 (Armenian and Georgian). The
Ethiopic system is multiplicative above 100, a feature that can exist only because
the signs of the system are not Ethiopic script-signs. A large plurality of systems
Cyrillic, Hebrew, Fez numerals, Coptic, Arabic abjad, and possibly Glagolitic
use multiplication above 1000, a natural way to proceed in systems with twentyseven ordinary signs. Three other systems the Greek, the Syriac, and zimm
numerals are multiplicative at both 1000 and 10,000; that is, after 8000
(8 1000) and 9000 (9 1000), one uses a new sign for 10,000 (1 10,000)
rather than (10 1000) as in systems that are only multiplicative at 1000. Interestingly, only the obscure Armenian system developed by Shirakatsi takes the step of
making the entire system multiplicative-additive. Because multiplicative-additive
numeral-phrases are usually longer than ciphered-additive ones, doing so may not
have been appealing. While there were occasional efforts to positionalize Greek,
Hebrew, and Latin alphabetic numerals by simply using the first nine letters along
with a zero, these attempts were sporadic and short-lived.
The longevity of the alphabetic systems is remarkable. Seven alphabetic numerical notation systems were regularly used for a millennium or longer (Greek, Coptic, Ethiopic, Arabic, Hebrew, Syriac, Armenian, and Georgian). In cases such
as the Greek system, this can be explained by the political importance of the

Alphabetic Systems

187

systems users, but in others, such as Hebrew and Armenian, the systems users
have largely been politically and culturally marginalized. That these systems could
survive in such circumstances and where, in many cases, ciphered-positional systems were available, requires explanation. Structural features may partly explain it:
alphabetic systems alphabeticity means that one need not learn a set of numerals in addition to a script, and ciphered-additive systems are more concise than
ciphered-positional systems for any Western numeral-phrase containing zeroes,
the corresponding ciphered-additive numeral-phrase will be shorter.
A more satisfying explanation, however, is that alphabetic numerical notation
systems, like scripts, can be important markers of cultural identity. In many cases
(e.g., Coptic, Gothic, Armenian, Georgian, Glagolitic, and Cyrillic), a group of
people developed a unique set of alphabetic numerals and developed their own
script at the same time. The point of alphabetic numerals is not to be comprehensible translinguistically, but rather for each system to serve for one script alone.
Under these circumstances, an alphabetic numeral system becomes an integral
part of a script, and thus marks ethnic identity. Even when these systems ceased to
be used regularly, many of them continued to be used in restricted functions, particularly in the domain of religion (e.g., Hebrew, Syriac, Coptic, Greek, Cyrillic).
Of course, it must not be forgotten that ultimately, in the face of massive globalization over the past five centuries, Western and Arabic positional numerals have
become the earliest and standard systems learned and used by almost everyone.
The future of alphabetic numerals seems likely to be one of increasing vestigiality
and obsolescence.

chapter 6

South Asian Systems

The South Asian numerical notation systems include all systems that derive from
the Brhm numerals used on the Indian subcontinent, including Western numerals. With the possible exception of China, South Asian numerical notation systems
are predominant throughout the entire world today. While most of the modern
South Asian systems are ciphered-positional, the earlier systems were cipheredadditive or multiplicative-additive. An important evolutionary development was
the shift from ciphered-additive systems, such as the early Brhm numerals, to ciphered-positional systems. Despite attempts to postulate the origin of the important ciphered-positional structure elsewhere (Greece, China, or Mesopotamia),
this development came out of South Asia.

Brhm
The Brhm script came to prominence in the mid third century bc, during the
reign of the Mauryan emperor Aoka, although inscribed potsherds from the site
of Anuradhapura in Sri Lanka have been dated (controversially) to 400 bc (Coningham et al. 1996). Brhm script was probably derived from a Semitic prototype
(Aramaic, South Semitic, or Phoenician), although many South Asian scholars
still support the theory that the script was indigenously developed (Salomon 1996:
378379). Along with the slightly earlier Kharoh script used in the northwestern regions of India, it was the first script used in India after the collapse of the
188

South Asian Systems

189

Table 6.1. Brhm numerals

1s
10s
100s
1000s

A
J
S
1

B
K
T
2

C
L
U
3

D
M
V
4

e
N
w
5

F
O

G
P

H
Q

I
R

Harappan civilization.1 Both scripts are alphasyllabaries (scripts in which each


sign has a consonantal base that is modified to indicate which vowel sound is associated with it), but structural differences between the two suggest that their origins
are different. This is supported by the fact that the Kharoh numerals (Chapter
3) are a hybrid cumulative-additive/multiplicative-additive system very much like
the Aramaic system, while the Brhm numerals are quite different in principle.
The basic Brhm numerals are shown in Table 6.1 (Bhler 1896: Plate IX; Datta
and Singh 1962 [1935]: Tables III-IX; Salomon 1998: 58).
The signs shown in Table 6.1 are those found on Katrapa coins (second to fourth
century ad); however, there are enormous paleographic variations among Brhm
inscriptions. Numeral-phrases are written from left to right, proceeding from higher
to lower powers; thus, 289 might be written as TQI. The signs for 1 through 3
are cumulative, with horizontal strokes indicating units, but otherwise, the Brhm
system is ciphered-additive up to 100, using separate signs for each of the units
and the tens. In later inscriptions, even the cumulative signs become ligatured or
distorted; for instance, the seventh/eighth-century ad grants of the Gang dynasty
contain + and = for 2 and 3 (Datta and Singh 1962 [1935]: Table V].
Above 100, the systems structure becomes more complex. The numeral-signs in
Table 6.1 are not, as they appear, simple multiplicative formations that juxtapose
a unit-sign with a power-sign for either the hundreds or the thousands; otherwise,
one would expect 3 for 2000 rather than 2. Instead, these numeral-signs for 100
through 300 and 1000 through 3000 are quasi-multiplicative. Still, the graphic
similarity of the various signs for the hundreds and thousands to the corresponding units is significant. On an inscription from Nana Ghat (first century bc), the
numbers 400, 700, 1000, 4000, 6000, 10,000, and 20,000 are written in a straightforward multiplicative fashion as @, #, $, %, ^, &, and *, and thus combine unit-signs with signs for 100 (!) and 1000 ($) (Indraji 1876, 1877). Slightly
different but equally straightforward signs are found on slightly later inscriptions
1

Most Western epigraphers and archaeologists accept that, aside from the Harappan
script, there was no pre-Mauryan writing in India (Salomon 1995, 1998).

190

Numerical Notation

Figure 6.1. Rubbing of an inscription from Nsik, Maharashtra state, dating to 120 ad,
bearing a variety of Brahmi numerals. Source: Senart 19056: 82.

from Nsik, as in Figure 6.1, dated 42 aka (ad 120) (Senart 19056: 82; Sircar 1964:
164167). The Vktaka grants (fifth century ad), one of the latest texts containing signs for the thousands, denote 8000 as x, a ligature of the signs for 1000 (y)
and 8 (z) that occur on the same grants (Datta and Singh 1962 [1935]: Tables IV,
IX). Despite paleographic changes, the basic structure of the Brhm numerals was
always ciphered-additive below 100 and multiplicative-additive at both 100 and
1000. No special sign for 10,000 was used; 10,000 and 20,000 are written as
1000 10 and 1000 20 rather than 10,000 and 10,000 2. In the Nana Ghat
inscriptions, 24,400 is written as *%@ (1000 20 + 1000 4 + 100 4).
The Brhm numerals appear on some of the earliest Aokan inscriptions,
dating to the middle of the third century bc, but not in the early Sri Lankan
writings. These early inscriptions contain only a few signs (for 1, 2, 4, 6, 50, and
200), but already the hybrid cumulative-additive/multiplicative-additive structure
of the system was in place.2 Most of the signs are recognizably ancestral to later
ones, such as the more complete sets of numerals found at Nana Ghat and at
Nsik Cave. While there is no paleographic evidence of Brhm numerals prior to 300 bc, Datta and Singh (1962 [1935]: 37) claim that, because the Aokan
2

This opinion contradicts that of Guitel (1975: 605), who sees the Aoka numerals as being non-multiplicative on the basis that the sign for 200 does not sufficiently resemble a
ligatured multiplicative 100 2.

South Asian Systems

191

inscriptions are found all over India, the Brhm system must have been developed
much earlier than the paleographic evidence would indicate, perhaps between
1000 and 600 bc. This is a spurious use of the discredited age-area method
(determining the age of features by their geographical distribution). In the early
Mauryan Empire, an enormous region was quickly encapsulated within a single
polity, so it is unsurprising that Mauryan administrative inscriptions are widely
distributed. While it was certainly plausible for nineteenth-century Indologists to
hope to find earlier paleographic evidence for the numerals, such hopes now seem
remote. I agree with Salomon (1996, 1998) and many other Indologists that a midthird-century origin for the Brhm numerals and script is probable.
The question of the ultimate origin of the Brhm numerals specifically,
whether or not they constitute a case of independent invention, and if not, on
which ancestor(s) they were modeled is unresolved, and is made more complex
by the politicization of the matter. Previous scholars have emphasized the paleographic comparison of individual signs. I believe that the consideration of the
systems structural features and historical context of origin supplemented by
paleography, where appropriate will be a more fruitful approach.
One set of theories regarding the origin of the Brhm numerals derives them from
existing representational systems used in South Asia. Borrowing from the letters of
the Brhm script to create an alphabetic numeral-system, while once a popular
theory, is not really sustainable (Prinsep 1838, Woepcke 1863, Indraji 1876, Datta and
Singh 1962 [1935], Gokhale 1966, Verma 1971). While a few Brhm numeral-signs
resemble phonetic signs, if one accepts certain paleographic transformations, these
correspond neither to the standard Brhm letter-order nor acrophonically to any
languages lexical numerals. Renou and Filliozat (1953) note that in texts containing
both purported letter-numerals and the corresponding signs used phonetically, the
two varieties are quite different. The derivation of the Brhm numerals from the
Kharoh letters is even more improbable, and has not been seriously proposed for
over a century (Bayley 1882). Ifrah (1998) proposes but discards the theory that the
Brhm numerals derive from the Kharoh numerals, which can easily be dismissed
by noting the temporal priority of the former. Finally, a more recent set of theories
derives the Brhm numerals from those of the Indus Valley civilization (Sen 1971,
Kak 1994), but there are no examples of any writing from India between the latest
Harappan inscriptions (around 1700 bc) and the first Brhm inscriptions (around
250 bc), and only limited and conflicting evidence for the nature of the Harappan
numerical notation system (Parpola 1994; cf. Chapter 10).
If not derived from any South Asian system, the Brhm numerals could have
developed independently. Woodruff (1994 [1909]: 5360) speculated that both the
Chinese and Brhm numerals derived from a hypothetical ancient set of cumulative tally signs for 1 through 9, which would then have spread to both China and

192

Numerical Notation

India. Kaye (1919) argued that the Brhm numerals developed independently during Aokas time, with their structural features representing three different stages
of development, but inexplicably then argues against Indian creativity. Ifrah (1998:
390), arguing that there are universal constants caused by the fundamental rules
of history and paleography, postulates nonattested cumulative signs for 1 through
9 which later became abbreviated and ligatured into the Brhm system. Salomons
position is more sensibly agnostic; recognizing the problems involved with many
other theories, he simply notes that numerical signs are sometimes cursive reductions of collocations of counting strokes, citing the hieratic and demotic systems
as examples (Salomon 1998: 60).
Finally, a number of theories argue for a foreign origin of the Brhm numerals.
Falk (1993: 175176), noting structural and paleographic resemblances between
the Brhm and the earliest Chinese (Chapter 8) numerical notations, argues for
a Chinese origin. However, there is little evidence of contact between the two
regions at this period, and the only paleographic similarity between the systems
is the common use of horizontal strokes for 1, 2, and 3. It has occasionally been
proposed that the Greek alphabetic numerals inspired the Brhm numerals, given
their appearance following the Alexandrine period, the strong trade ties with the
Greco-Iranian kingdoms of Parthia and Bactria, and the structural similarities
between the two systems. However, the evidence for the alphabeticity of the
Brhm numerals is weak at best (see the previous discussion), and there is no
paleographic correspondence between the Greek and Brhm numerals.
It is most plausible that the Brhm numerals are derived from the Egyptian hieratic or demotic numerals. Burnell (1968 [1874]) argued for a demotic origin, while
Bhlers (1963 [1895]) much more prominent analysis argued for a hieratic origin.
The three systems are structurally similar: they are all decimal, hybrid cipheredadditive/multiplicative-additive systems, and represent 200, 300, 2000, and 3000
by adding quasi-multiplicative strokes to the signs for 100 or 1000. There are resemblances in around one-third of the sign-forms, and very close resemblances for a few,
such as 9 (hieratic = i; demotic = i; early Brhm = i) (Bhler 1963 [1895]: 115119;
Salomon 1995, 1998). While there was not tremendous Egypto-Indic cultural contact, Ptolemaic traders reached as far as the city of Muziris (modern Cranganore) on
the Malabar Coast, and Aoka is known to have sent Buddhist missionaries to Alexandria (Basham 1980: 187). Of the two Egyptian systems, I believe the demotic to be
a more likely ancestor, because in the Ptolemaic period the use of hieratic numerals
was very limited. Thus, although the demotic and Brhm systems differ in both the
power at which multiplication is used and the direction of writing, I believe that a
demotic origin should be adopted as a working hypothesis.
The Brhm numerals spread throughout the Indian subcontinent during the
Mauryan period. Only in the northwest, where Kharoh numerals predominated,

South Asian Systems

193

did the Brhm numerals fail to penetrate until around the fourth century ad.
They were used primarily for writing dates on stone inscriptions and copper land
grants. Thus, a full set of numeral-signs up to at least 1000 is attested, and the
numeral-signs can be assigned exact dates. Other functions for which Brhm
numerals were used include stichometry and the recording of financial transactions. While it is interesting to speculate on the use of Brhm numerals on other
materials than stone and copper, the Indian climate and geography are unsuitable for the survival of perishable materials. In Central Asia, manuscripts in the
Tocharian language were written using a variant of Brhm script and numerals
from the sixth to ninth centuries, and in this drier climate, Brhm numerals are
attested to have been used on wood tablets, palm leaves, and paper (Sander 1968).
There is no surviving evidence that the Brhm numerals were used for arithmetic or accounting; the rather substantial body of medieval Indian mathematical
works, sometimes attributed to very early dates, use either lexical numerals only,
or employ one of the alphasyllabic notations to be described here.
After the Kharoh script died out in the fourth century ad, variants of the
Brhm numerals were the only ones used in India until the late sixth or early
seventh century. They spread not only throughout the Indian subcontinent, but
also into Central and Southeast Asia, regions that were heavily influenced by India
during this period. In some Central Asian manuscripts, numeral-phrases were
written from top to bottom rather than from left to right (Renou and Filliozat
1953: 702). There was enormous variation in the shapes of the numeral-signs from
location to location, suggesting that, as with the Indian scripts, no pressure existed
toward the formation of interregional standards. The primary regional division,
between northern and southern systems, began as early as the second century ad,
and these two basic variants diverged further in later centuries.
The end of the traditional Brhm numerals and the later local additive variants
was a gradual process, instigated not by external influences but by the invention of
ciphered-positional notation beginning in the late sixth or early seventh century
ad. Over the next couple of centuries, the older ciphered-additive forms became
increasingly rare, and by the ninth century ad, the Brhm numerals had been replaced by the modern ciphered-positional system throughout India and Southeast
Asia. Only on the southern tip of India and in Sri Lanka were additive systems
retained (though in an altered form) until significantly later.

From Addition to Position


No numerical notation systems have been as widely studied and discussed as the
ciphered-positional systems that originated in India, primarily because, through
the intermediary of the Arabs, these systems are ancestral to Western numerals.

194

Numerical Notation

While it would be teleological to portray the history of numerals in a linear fashion leading to our own system, the present near-universality of the cipheredpositional, decimal structure originating in India and spreading westward throughout the Islamic and Western spheres requires explanation.
Around ad 600, a change began in the writing of dates in the Brhmderived scripts of India and Southeast Asia. Instead of writing smaller numbers with
ciphered-additive notation and larger numbers with multiplicative-additive notation, all numbers were written using paleographic variations of the nine Brhm
numeral-signs and a dot to indicate zero a purely ciphered-positional system.
The spread of the older additive systems throughout South and Southeast Asia
between the third century bc and the seventh century ad was followed by a second wave of diffusion of the positional principle and zero (seventh century ad
onward), wherein the additive systems changed into positional ones. The change
is actually a very simple one. The unit-signs were retained, but the power-signs
were replaced with a single sign for zero. This process is confirmed by comparing
the signs for 1 through 9 in the nonpositional (Brhm-derived) systems with the
very similar unit-signs used with a zero in later periods.
This was not the first zero; the Babylonians (Chapter 7) and the Maya (Chapter 9)
had already invented the positional principle and zero-signs well before this time.
Some scholars have claimed that the Babylonian zero diffused eastward to India
just as it diffused westward to Greece (Fvrier 1948: 585; Menninger 1969: 3989).
Yet the Babylonian and Maya systems were both cumulative-positional, and used
a sub-base in addition to a base (Babylonian 10 and 60, Maya 5 and 20). Similarly,
while Greek and Arab astronomers used positional fractions with a zero, these
had a sexagesimal (Babylonian-derived) base and never used positional notation
for whole numbers. Thus, there is no apparent historical relation between these
systems and the later Indian one. The Indian positional system first combined
ciphering, a zero, and a single, decimal base.
It is frequently claimed that the earliest example of ciphered-positional numerals is found on the Sankheda or Mankani copper plate bearing the date 346 in the
Kalacuri era, which translates to 595 ad (Bhler 1896: 78; Smith and Karpinski
1911: 46; Das 1927a: 118; Kak 1990: 199). This plate is a donation charter of Dadda
III, used to certify a land grant. When discussing any land grant, the issue of a
later forgery always arises, as attempts to claim land by producing such evidence
were common in India (as elsewhere). Since much of the paleographic evidence
for early positional numerals comes from such land grants, we must be cautious
to avoid dating inscriptions simply by the date as inscribed, and also take paleography and historical context into account. We must also remember that texts containing positional notation that are transcriptions or translations of earlier works
must not be assigned an early date based simply on their putative earlier authors.

South Asian Systems

195

While we need not go so far as Kaye (1919: 346), who claimed that all positional
numerals in India prior to the ninth century ad were forgeries, most Indologists
are very wary of the Sankheda plate (Salomon 1998: 61). The fact that it is ninety
years older than any other positional numeral-phrase suggests that we need to
question carefully any sixth-century ad evidence for ciphered-positional numerals
in India.
The earliest surviving and unquestionable examples of ciphered-positional
numerals with a zero derive, not from India itself, but from Southeast Asia, in
Khmer, Old Malay, and Cham inscriptions from the late seventh century ad. A
calendrical inscription found at Sambor (in Cambodia) and written in a mixture
of Old Khmer and Sanskrit is dated 605 in the aka dating system, or 683 ad; the
zero appears as a small dot (Coeds 1931: 327). As the Sambor inscription is a calendrical passage rather than a land grant or financial document, it is unlikely to be
a forgery (Diller 1996: 126). Similar inscriptions from the Old Malay kingdom of
Sriwijaya have been found at Palembang and at Kotakapur on the nearby island of
Bangka dating to 683, 684, and 686 ad, or 605, 606, and 608 aka, respectively; in
these cases, the zero was written as a circle rather than as a closed dot (Diller 1995).
It is intriguing that the Old Khmer and Old Malay inscriptions appear in the same
year (Diller 1995: 66). Nevertheless, there is no reason to believe, as Kaye (1907)
did, that the existence of these inscriptions must mean that ciphered-positional
numerals actually originated in Southeast Asia and diffused from there to India.
The existence of intermediate additive-positional forms from the sixth and early
seventh centuries ad, which I will discuss later, coupled with the probability that
some of the earlier copper grant plates are authentic, make it likely that the invention of ciphered-positional numerals occurred around ad 600.
However, there is compelling evidence for something akin to the positional
principle and zero in certain earlier literary and religious texts. In texts using the
bhutasa khya or word-numeral system, special cryptic words for one through nine
could be combined in a sort of positional fashion with a word for zero in order to
represent dates lexically in a fashion quite different from that of ordinary Sanskrit
number-words (Datta and Singh 1962 [1935]: 5363; Mukherjee 1977). The earliest attested text to use this system is the Yavanajtaka of Sphujidhava (ad 269),
a Sanskrit version of a Greek astronomical text (Yano 2006: 15). In it, the words
moon and earth mean one, eye and twin two, and so on, with sky and dot
meaning zero, suggesting a sort of positionality of words if not of graphemes,
while also allowing multiple words to represent the number, a key to producing
numeral-phrases that are unambiguous and yet flexible enough to fit into strict
Sanskrit verse (Plofker 2007: 395). The bhutasa khya system is thus suggestive
of positionality but does not constitute a system of graphic numerical signs, nor
should its use be taken to imply the widespread use of decimal positional numerals

196

Numerical Notation

in Indian manuscripts. Notably, the regular name for Indian numeral symbols,
anka mark, is also a numerical word for nine in this system, which suggests
that there were originally only nine numerical graphemes that is, that the zero,
and place-value, were latecomers to Indian numerical notation even though they
existed conceptually in the word-numerals (Clark 1929: 229230).
The earliest Sanksrit word for the zero-sign, nya-bindu (literally, void-dot),
is first used in Subhandus Vasavadatta, written around 600 ad (Sen 1971: 175;
Salomon 1998: 63). This evidence suggests a correspondence between the early use
of numeral-words and the structurally identical later use of the ten numeral-signs.
Thus, the literary concept of a zero-space in Hindu thought, the use of chronograms, and the term nya-bindu in the fifth and sixth centuries ad may have prefigured the eventual development of ciphered-positional numerals. Within the interlinked tradition of Indian religious and mathematical thought, the invention of
the zero is as much a metaphysical concept as it is an arithmetical tool, if not more
so. Almost all the attested early ciphered-positional numerals are nonarithmetical,
and are simply used to register dates and other numbers on inscriptions and copper plates. It would be erroneous to assume that the positional numerals originally
had an arithmetical function, and then to use this assumption to hypothesize an
ancient mathematical tradition of ciphered-positional numerals.
There is no evidence that true ciphered-positional numerals were used prior
to the middle of the sixth century ad, and any claim prior to the middle of the
seventh century ad requires careful examination. The Indian climate and topography are not particularly suitable for the survival of materials other than stone
and metal, and we certainly do not have as much evidence as we would like.
Nevertheless, there is plenty of inscriptional evidence for continued use of the old
additive numerals from the sixth through the eighth centuries ad, they decline
significantly in frequency only in the ninth century. It is unlikely that all the evidence for ciphered-positional numerals was lost where so much survives for the
additive system though not impossible, if positional numerals were used only
on perishable media at first.
Further evidence for the chronology of the shift comes from inscriptions dating from the late sixth to the middle of the eighth century ad from the Orissa
region, which are written with unusual mixed structures combining the features
of the older additive and newer positional notations (Datta and Singh 1962 [1935];
Acharya 1993; Salomon 1998: 6263). The earliest of these is from the Urlam copper plates of the Eastern Gang king Hastivarman, dated to 578 ad, in which the
Gang era year 80 is written as the additive sign for 80 followed by a zero, but
this date is questionable (Salomon 1998: 62). Acharya (1993) describes many Orissan inscriptions dating from 635 to 690 ad in which dates such as 137 are written
as 100 3 7 rather than 137. This series of dates leads directly into the first fully

South Asian Systems

197

positional date found in India, on the Siddhantam grant of Devendravarman


(195 Gang = 693 ad), just ten years after the Southeast Asian examples mentioned
earlier (Salomon 1998: 62).3 Datta and Singh (1962 [1935]: 52) mention additional
eighth-century ad examples combining additive and positional notation, which
they characterize as representing the gradual forgetting of the older system. Their
argument rests on the claim that these are quite late examples of additive notation
and that the positional principle was well established by this time. In fact, it is
far more likely that they represent incomplete attempts to incorporate the novel
positional principle into inscriptions. These mixed numeral-phrases confirm the
hypothesis of an early seventh-century origin of positional numerical notation in
India.
In the eighth century, the positional system gained significant ground, and was
preferred by 800 ad. Around this time, the spread of scientific knowledge from
India to China (primarily through the medium of Buddhist scholarship) led to
awareness of the ciphered-positional numerals in China. In the Kaiyuan period
(713 to 741 ad), the Indian astronomer Qutan Xida (Gautama Siddhartha) translated an Indian calendar into Chinese, using positional numerals (with a dot for
zero), and commented on their ease of use (Gupta 1983: 24). In the ninth century
ad, the additive system became much scarcer. Salomon (1998: 62) notes a striking
late example from the Ahar stone inscription in north central India, which is a
composite record of documents of different dates; those up to 865 ad are all dated
using additive numerals, and those from 867 ad using positional numerals, providing precise information on the date of replacement. The plate of Vinyakapal
(931 ad) is an extremely late northern (Nagari) inscription containing the additive
system (Singh 1991: 170). By the tenth century ad, only the far south of India
(Tamil and Malayalam-speaking areas) consistently used additive systems, but
even there, the old system was replaced by a purely multiplicative-additive structure. Of all the descendants of the Brhm system, only the Sinhalese numerical
notation system preserved the old structure until comparatively recently.

Modern South Asian


By the end of the ninth century ad, the transformation of Brhm numerals into
modern ciphered-positional forms with a zero was complete. The structural evolution of the Indian systems ended at this point, although paleographic developments in the numeral-signs continued. It is well beyond the scope of this work
3

I am uncertain what to make of Mukherjees (1993) assertion that the copper-plate inscription of Devakhadga expresses the date 73 in the Harsha era (starting 606 ad) using
positional numerals, which would thus be dated to 678 ad.

Numerical Notation

198

Table 6.2. North Indian numerical notation systems


Script

Bengali

Devanagari
Gujarati
Marathi
Oriya
Punjabi
Nepali

to present the paleographic data concerning the development of Indian numeralsigns from 800 ad to the present day (cf. Salomon 1998; Ifrah 1998: 367385 for
more complete analyses of this issue). Nevertheless, a look at some of the more
important variations on this common pattern of ciphered-positional decimal systems is warranted. Many of these systems (or very close descendants thereof ) have
been employed for well over one thousand years and continue to be used. Most
major South Asian languages have their own alphasyllabaries and numeral-signs.
Their numerical notation systems are structurally identical to one another and to
Western numerals. Today, all these indigenous systems are in competition with
Western numerals, especially for commercial and scientific purposes. In religious
and formal contexts, the traditional numerals are still frequently preferred.

North India
The ancestor of the northern Indian numerical notation systems is the Brhm
system used in the Gupta Empire, which ruled most of northern India from the
Indus to the Ganges from the fourth to sixth centuries ad, and also significantly
influenced central India. These earliest Gupta Brhm numerals were nonpositional, but the idea of positionality and the zero sign spread quickly through the
systems of the region. The most common modern varieties of this subgroup are
the Bengali, Devanagari, Gujarati, Marathi, Oriya, Nepali, and Punjabi; they are
thus used in a swath across Pakistan, northern and central India, Nepal, and Bangladesh. The northern Indian systems are also directly ancestral to both the modern Arabic and Maghribi numerals associated with the Arabic script, and thus,
indirectly, to Western numerals. The similarities between Western numeral-signs
and many of the north Indian numerals, especially those for 0, 2, and 3, are quite
evident in Table 6.2.

South Asian Systems

199

Table 6.3. Central Asian numerical notation systems


Script

Tibetan
Mongolian

Central Asia
The Gupta script also gave rise to a small number of scripts in the Himalayas and
Central Asia, of which the most important are the Mongolian and Tibetan. The
Tocharian script had used a variant of the Brhm additive numerals from the
sixth to eighth centuries ad, but the Tocharian language and script died out before the introduction of positionality. Tibetan writing and numeration developed
in the ninth century, and Mongolian numerals developed from Tibetan in the
thirteenth century.4 These systems are related to the northern Indian systems. The
classical Mongolian numerals were usually written from top to bottom in vertical
columns, but the forms listed here are those used when they were written from left
to right. These systems are shown in Table 6.3.

South India
The scripts of the southern half of the Indian peninsula diverged from those of the
north as early as the second century ad. There are five modern scripts in this subgroup: Telugu and Kannada, two closely related scripts of east central India, along
with Tamil, Malayalam, and Sinhalese. Of these five, the Telugu and Kannada numerals shifted from addition to position in the seventh and eighth centuries ad,
and are thus structurally identical to those of the northern systems (Syamalamma
1992: 51). The numerals of the Grantha script (sixth to twelfth centuries), which is
ancestral to modern Tamil, Malayalam, and Sinhalese, did not switch principle.
The Telugu and Kannada numerals are shown in Table 6.4, while the other three
systems are described later in this chapter.

Southeast Asia
Far from being a cultural backwater or simple recipient of positional notation,
South Asian societies used ciphered-positional numerals very early. Scripts such
4

Despite Ifrahs assertion (1998: 382) that each of the Agnean, Kutchean, and Khotanese
scripts of Chinese Turkestan would have used a set of ten positional numerals, I know of
no evidence that this was the case.

Numerical Notation

200

Table 6.4. South Indian numerical notation systems


Script

Telugu
Kannada

as Kawi (the ancient script of Java) and Cham (used in Vietnam until the thirteenth century) originally used hybrid additive numerical notation systems on the
Brhm model, but these transformed into ciphered-additive positional systems
in the seventh century. The modern descendants of these systems include Khmer,
Thai, Burmese, Lao, Balinese, and Javanese. Of these, Balinese and Javanese are
closely related to one another but paleographically distant from any other South
Asian systems. They use Javanese letters to represent certain numbers, while retaining older signs derived from Kawi for the others (0, 4, 5, and 6). The Southeast
Asian systems are shown in Table 6.5.

Tamil
The Tamil script used in the far southeast of India and parts of Sri Lanka is derived
ultimately from Bhattiprolu, the southern variety of the Brhm script that developed in the first or second century ad, and immediately from the Grantha script
of the sixth through twelfth centuries, which is ancestral to Tamil, Malayalam,
and Sinhalese. The Tamil script is alphasyllabic and similar to other Brhm-based
scripts, but has unique features, such as the ability to represent consonant clusters as
a sequence of individual consonant signs. Similarly, the Tamil numerical notation
system is rather different from those of other Brhm-derived scripts. The Tamil
numeral-signs are shown in Table 6.6 (Pihan 1860: 113119; Guitel 1975: 614615).
Table 6.5. Southeast Asian numerical notation systems
Script
Khmer
Thai
Burmese
Lao
Balinese
Javanese

South Asian Systems

201

Table 6.6. Tamil numerals


1

A
B C
10
100
J
6408 = FLDKH

Units

D
K

G
L

1000

The numeral-signs are ultimately derived from Brhm, and are thus related
to all the systems of India and Southeast Asia. Following the Indian pattern,
numeral-phrases are written and read from left to right, but are structurally neither
ciphered-additive, like those of the Brhm numerals, nor ciphered-positional, like
those of most of the Indian systems. Rather, the traditional Tamil system is multiplicative-additive and decimal. There is no power multiplier for the ones. Tamil
has no signs for 10,000 or higher powers of 10; large numbers were expressed by
placing an appropriate numeral-phrase before the sign for 1000, then multiplying.
Thus, 800,000 would be written as HKL (8 100 1000). There is no ambiguity
in this phrases meaning, because phrases are always read strictly from left to right.5
This is the only instance where a lower power sign may precede a higher one.
The Tamil numerals acquired their distinct structure in the medieval era, although
it is not clear exactly when the divergence arose. The change from a hybrid cipheredadditive/multiplicative-additive to a purely multiplicative-additive structure is easily accomplished; because Brhm numerals are multiplicative above 100, all that
is required is that the nine individual signs for the decades 10 through 90 be
replaced by a single sign for 10. At this early period, we know them largely from
inscriptions on stone, although we cannot exclude the possibility that they were
used in other contexts. The numeral-signs are derived from those of the Grantha
script, and are closely related to others of southern India. It has sometimes been
claimed that the Tamil numerals are a uniquely Dravidian invention using letters
of the alphabet, and, indeed, there are resemblances between the numeral-signs
for 1 through 9 and nine Tamil phonetic signs (Burnell 1968 [1874]: 68). Nevertheless, since these resemblances can be found only by comparing the modern paleographic forms of the numbers and letters, this argument cannot be offered as a
theory of their origin. Rather, the similarity is probably due to a later assimilation
of the numeral-signs to the phonetic signs.
Only Tamil and Malayalam, of all the South Asian systems, altered the Brhm
ciphered-additive/multiplicative-additive system to a purely multiplicative-additive
5

Curiously, this system is structurally identical to the Armenian alphabetic notation of Anania
Shirakatsi (Chapter 5), but it would be an error to make too much of this resemblance.

202

Numerical Notation

one. (Sinhalese retained the Brhm structure, while all other Brhm-derived systems became ciphered-positional.) The Chinese classical numerals (Chapter 8)
are multiplicative-additive, so contact with Chinese Buddhists might have stimulated the development of the unique notations characteristic of areas of southern
India that were also Buddhist, or, more likely, made their retention more appealing. However, there is no paleographic similarity between the Chinese and
Tamil numerals, making diffusion from China improbable. Moreover, none of
this evidence explains why the Tamils retained their system even after they ceased
to practice Buddhism. Another possibility is that the users of the Tamil and
Malayalam systems did not switch to positional notation as an effort to maintain
their cultural distinctness from the north.
Late in the systems history, an abbreviated form of the Tamil numerals developed that, for some numbers, adds an element of positional notation by omitting
the power-signs for 10, 100, and 1000. For instance, Pihan notes that while 21
was traditionally written BJA (2 10 + 1), it could also be written BA, abbreviating the phrase without any loss of information (Pihan 1860: 117). Such
numeral-phrases are purely ciphered-positional. While this presents no problem
for numerals that lack any empty positions, a zero sign is needed in other cases;
however, no zero appears in any pretwentieth-century Tamil numeral-phrases.
Sometimes, rather than using a sign for zero, Tamil writers used the power-signs
for 10, 100, and 1000 to eliminate ambiguity. In one case, 2205 is written as BBKE
(2, 2, 100, 5), which indicates that the second 2 is to be understood as a hundreds
value rather than as a tens value, and that therefore the first 2 must be understood
as 2000 (Guitel 1975: 614615). Such phrases combine multiplicative-additive and
ciphered-positional notation. These mixed multiplicative and positional phrases
are no longer used, and appear to be a transitional product of the colonial period, when contact with the West began in earnest. Apparently, in the nineteenth
century some Tamil astronomers were computing using a mixed decimal and sexagesimal computation system by manipulating pebbles or shells on a flat surface
in a manner reminiscent of Greek or Babylonian techniques, but there was no
corresponding numerical notation (Neugebauer 1952). Today, some formal Tamil
writings use the traditional numerals, while for most commercial and informal
purposes an ordinary zero sign is used, making the system ciphered-positional.
Most literate Tamils today are familiar with and use the Western numerals.

Malayalam
The Malayalam script, like Tamil, is derived from the Grantha script of southern
India. It is used to write the Dravidian language of the same name used in Kerala
at Indias southwestern tip. It first emerged as a distinct script around 700 ad,

South Asian Systems

203

Table 6.7. Malayalam numerals


1

M N
10
W
6408 = RYPXT

Units

P
X

S
Y

100

1000

although its letters and numeral-signs are closely related to those of the other
Brhm-derived scripts. The Malayalam numerical notation system, like Tamil
and Sinhalese, was unaffected by the cultural and political influence that rendered the northern Indian numerical notation systems structurally identical and
paleographically similar. The traditional Malayalam numeral-signs are indicated
in Table 6.7 (Pihan 1860: 122125; Ifrah 1998: 335).
The similarities between the Tamil and Malayalam systems are striking. Both
systems are decimal and multiplicative-additive, and written from left to right.
There are many paleographic similarities between the numeral-signs of the two
systems, thus refuting the claim that the Tamil numerals are phonetic in origin. As
in Tamil, there is no power-sign for the units, and the 1 is understood in any numeral-phrase with a units value. Furthermore, Malayalam numbers above 10,000
are also expressed using multiplicative combinations of the sign for 1000 with
those for 10 and 100, as necessary. None of these similarities is particularly surprising, given the close cultural and geographic proximity of these two Dravidian
peoples. The only structural difference between Tamil and Malayalam numeration
is that Malayalam numeral-phrases were never expressed using the hybrid additive and positional notation that was occasionally used later in the Tamil systems
history.
There are few distinctly Malayalam inscriptions that date before ad 1000,
by which time it had already acquired its multiplicative-additive structure. The
numeral-signs are derived from those of the Grantha script, showing that the Malayalam numerals are native to South Asia, although, as with Tamil, we cannot exclude
the possibility of some influence from Buddhist China. The fact that Buddhism was
maintained longer in the south than in northern India may be a partial explanation for the difference in structure. A millennium of trade with and domination
by other peoples of South Asia, most of whom used ciphered-positional notation,
did not affect the structure of the Malayalam system. Malayalam writers employed
this system regularly until the middle of the nineteenth century, at which time
European contact introduced the zero and the idea of positionality. A new sign
for zero was introduced (V), which, when combined with the nine regular unitsigns, produced a regular ciphered-positional system. Today, the older Malayalam

Numerical Notation

204

Table 6.8. Sinhalese numerals


1

a
b c
10s j
k
l
100 s
1000 t
3684 = ctfsqd

1s

d
m

e
n

f
o

g
p

h
q

i
r

system is used rarely if at all (largely by those who need to understand old texts),
and is quickly becoming a historical curiosity.

Sinhalese
The Sinhalese (or Singhalese) script developed from the model of the southern
Brhm scripts for use among the speakers of the Sinhalese language, and was
influenced by the Grantha script that is ancestral to the Tamil and Malayalam
scripts used for writing the Dravidian languages of southern India and northern
Sri Lanka (although Sinhalese is an Indo-European language). It is an alphasyllabary, written from left to right, and is used today in Sri Lanka and the Maldives.
The traditional Sinhalese numeral-signs are indicated in Table 6.8 (Pihan 1860:
140141; Ifrah 1998: 332).
Sinhalese has distinct signs for each of the units, each of the decades, 100, and
1000. The numeral-signs for the units resemble many of those used in other South
Asian numerical notation systems, and many of the signs also resemble, but are
not derived from, the phonetic signs of the Sinhalese script. Numeral-phrases
are written from left to right. The system is ciphered-additive below 100, but it
is multiplicative for the hundreds and thousands, combining the unit-signs with
the appropriate power-signs. It is unknown how 10,000 and higher numbers were
written, though Pihan (1860: 141) speculates that it may have been through multiplicative forms such as those used in Tamil and Malayalam.
The Sinhalese numerals are thus structurally identical to the old Brhm system. The Sinhalese did not adopt the positional system when the peoples of India (except Tamil and Malayalam speakers) and Southeast Asia did, between the
seventh and ninth centuries ad. Sinhalese inscriptions and texts used this system
throughout the medieval period, seemingly unaffected by the shift to positionality occurring in other Indian numerical notation systems. Pihan (1860) shows no
awareness of any structural changes in the Sinhalese numerals in use at the time

South Asian Systems

205

he was writing; although his knowledge of the numerals was limited, there is no
reason to believe that they were in significant decline in the mid nineteenth century. Modern Sinhalese writings normally use the Western numerals, although the
traditional numerals are retained for certain formal purposes.

Indian Alphasyllabic
The primary numerical notation systems of India were ciphered-additive before
the seventh century ad and ciphered-positional afterward, with only a few systems (Tamil, Sinhalese, Malayalam) remaining additive after that point. The numeral-signs of these systems are abstract and do not resemble closely the letters
of the Brhm script or its descendants. However, starting around 500 ad, Indian
astronomers and astrologers began to use a very different principle for representing numbers: assigning numerical values to the phonetic signs of various Indian
alphasyllabic scripts. The basic principle of the Indian alphasyllabaries is that a set
of consonant-signs are combined with a set of diacritic marks that indicate vowels
to produce a set of signs for CV syllables; unmarked consonant-signs denote the
syllable with the inherent vowel a.6 These numerical notation systems, known
collectively as varnasankhya systems, were considered to be distinct from the normal Indian systems that had abstract numeral-signs (Ifrah 1998: 483). The three
systems that I will now discuss ryabhatas numerals, katapaydi numerals, and
aksharapall numerals, represent an important side branch of the South Asian systems. Although used only by a limited group of initiates, they are very important
for understanding Indian astronomy, astrology, poetry, and numerology, as well as
serving important mnemonic functions by linking words and numbers.

ryabhatas Numerals
While it is sometimes claimed that the Indian grammarian Pnini used alphasyllabic numerals in the seventh century bc (Datta and Singh 1962 [1935]), this is a
highly dubious proposition given the lack of attested writing in India at that time.
The first attested alphasyllabic numerals appear about 510 ad, near the end of the
dominance of the Guptas over India. The system was very probably invented by
the mathematician and astronomer ryabhata, in whose works (later named the
ryabhatiya by his disciples) it first appears. ryabhata, who lived in Kusumapura
in modern Bihar, was renowned among Indian scholars of the Gupta empire and
later centuries, was known to Muslim scholars as Arjabhad, and was likely the figure
6

The number of consonant-signs and vowel diacritics varies from script to script, and
there are also signs for V syllables (isolated vowels) and CCV syllables.

Numerical Notation

206

Table 6.9. ryabhatas numerals

&



%

&

G

ka
1

kha
2

ga
3

gha
4

a
5

ki
100

khi
200

gi
300

ghi
400

i
500

<

<

E

;

D

A

ca
6

cha
7

ja
8

jha
9

a
10

ci
600

chi
700

ji
800

jhi
900

i
1000

?

@

4

6

B

wa
11

wha
12

sa
13

sha
14

ta
15

wi
1100

whi
1200

si
1300

shi
1400

ti
1500

8

5

K

C

3

ta
16

tha
17

da
18

dha
19

na
20

ti
1600

thi
1700

di
1800

dhi
1900

ni
2000





!

F



pa
21

pha
22

ba
23

bha
24

ma
25

pi
2100

phi
2200

bi
2300

bhi
2400

mi
2500

'

'

7

#

ya
30

ra
40

la
50

va
60

yi
3000

ri
4000

li
5000

vi
6000

(



H



a
70

va
80

sa
90

ha
100

i
7000

vi
8000

si
9000

hi
10,000

named Ardubarius in the seventh-century Byzantine text, Chronicon Paschale. His


work focused on astronomy (he is regarded as the first great Indian astronomer)
but also contains much pure mathematics, in addition to the cosmological hypotheses from which the exact sciences in ancient India cannot be divorced. His
numerals are concise and readable with little training, and while not infinitely
extendable, are capable of expressing very high numbers. Table 6.9 shows the signs
of ryabhatas numerals, using modern Nagari signs for convenience (Fleet 1911a;
Guitel 1975: 582583).
Due to the structure of Indian alphasyllabaries, while there are many hundreds of
possible syllables, to learn the signs of the system one need only learn the two sets
of signs, which then can be combined with one another. The thirty-three unmarked
signs, in their assigned order and divided into groups on the basis of similar phonetics, assume the numerical values 125, 3090, and 100, as shown in the leftmost five
columns of Table 6.9. The ingenious principle involved in this system is that changing

South Asian Systems

207

Table 6.10. Order of power diacritics

















ka
100 (1)

ki
102

ku
104

kri
106

kli
108

ke
1010

kai
1012

ko
1014

kau
1016

the vowel attached to one of the basic signs alters its numerical value. When combined with the vowel i, the signs take on the numerical values 1002500, 30009000,
and 10,000, as shown in the rightmost five columns. While this means that there are
two signs with the value 100 ha () and ki () this has little potential to cause
confusion. Each successive vowel diacritic multiplies the value of the sign by 100 with
respect to its predecessor, as shown in Table 6.10 (indicating only the combinations
of k + vowels). Using these signs in combination, any number up to 1018 could be
expressed, and ryabhatas system does not exhaust the available diacritics.
Numeral-phrases were written with the lowest powers on the left, which reflects
the order of powers of the Sanskrit lexical numerals, but which is opposed to the
Indian numerical notation systems, in which the highest power was on the left. No
sign for zero was needed, and none used. The signs for 11 through 19 and 21 through
25 were not strictly necessary; 15 could be written as GA instead of B without any
ambiguity, but the latter was more concise. These extraneous signs normally were
not used in numbers such as 85, which was written as G (5 + 80) rather than B(
(15 + 70). In some cases, these rules were violated (we do not know why), so that in
one astronomical table, 106 is written as 16 + 90 and 37 as 16 + 21 (Guitel 1975: 587).
Table 6.11 indicates several numeral-phrases written alphasyllabically.
Table 6.11. Alphasyllabic numeral-phrases
Alphasyllabic
Representation Transcription and Sign-values
kha
va
62
%
2
60
ta
ha
ta
OR
116
8or8 16
100
16
ra
chu
70,040
7E
40
70,000
kha
ya
ghi
li
765,432
%'&#<( 2
30
400
5000
ma
vi
nu
jri
98,206,025
 3;H
25
6000 200,000 8,000,000
phu
ghe
40,000,220,000 &
220,000 40,000,000,000
Value

ki
100

cu
u
60,000
700,000
sri
90,000,000

208

Numerical Notation

The best way to conceive of this system is as a base-100, or centesimal, multiplicative-additive system with a decimal and ciphered sub-base. Unlike most multiplicative-additive systems, however, there can be up to two unit-signs within each
power of 100, each of which combines with its own power-sign. For instance, in
the representation of 9800 in Table 6.11, the signs for 8 (;) and 90 (H) each combine separately with the diacritic sign for 100 ( ). The system is slightly irregular below 100 in that the basic thirty-three signs include signs for 11 through 25. ryabhata
recognized that the system was centesimal rather than decimal, as he distinguished
the set of centesimal powers, or varga classes, from the intermediate powers, or
avarga (Das 1927a: 110; Jha 1988: 80). ryabhatas system is not positional, since
placing an unmodified consonant-sign in the middle of a numeral-phrase would
render it meaningless.
If ryabhata was unfamiliar with positional numeration, then he may have developed alphasyllabic numerals because of the insufficiency of the Brhm cipheredadditive system for writing large numbers, a task that could be done very concisely
using his own system. However, there is no evidence that the calculations that
ryabhata undertook were done directly with these numerals. Instead, their function was primarily to efficiently express very large numbers in verse, in the manner in which traditional Indian mathematics was written (Clark 1929: 232233).
The versification of mathematical expressions, while seemingly constraining and
complicating the expression of numbers, assisted the user mnemonically while
retaining the brevity of other numerical systems.
It is possible that ryabhata was familiar with the Greek alphabetic numerals in
addition to the Brhm numerals. ryabhatas work was inspired in part by Greek
astronomical writings, and Fleet (1911a), among others, has argued that both ryabhatas astronomy and his numerals are derived from Greek sources. However,
even if he borrowed the general idea of using script-signs as numerals from the
Greeks and there is no definite evidence either way this does not tell us very
much, because the two systems are radically different (Das 1927a: 111114). Even
if he did know the Greek numerals, they played little role in the invention of his
own system, whose idiosyncratic features, such as a base of 100, are not found in
other systems. Another possibility is that even though his system was not positional and lacked a sign for zero, ryabhata had a complete knowledge of cipheredpositional numeration when inventing his alphasyllabic system (Ganguli 1927). As
I discussed earlier, the concept of nya or emptiness existed in the fifth century
ad and may have prefigured the use of positional numerals in India, but no good
evidence survives for an actual ciphered-positional numeration system prior to the
seventh century ad, long after ryabhatas death. Ifrahs (1998: 450451) statement
that the use and adoption of ryabhatas system caused the Indian discoveries of
the place-value system and zero, which took place before ryabhatas time, to be

South Asian Systems

209

irretrievably lost to history is typical of the confused anti-empiricism of recent research on this system. The nonpositionality and relative complexity of ryabhatas
system argue against his having been familiar with positionality.
While ryabhatas numerals were known to Indian astronomers and mathematicians long after his death, they were used solely in the context of commentaries
on his work. The systems lack of facility for arithmetic, coupled with the difficulty
in pronouncing the resulting numeral-phrases in Sanskrit, led to its abandonment
even among most of ryabhatas proponents (Jha 1988: 8586; Yano 2006: 149).
It was replaced, in part, by the regular numerical notation systems of India, but it
also gave rise to a variety of successor systems for correlating phonetic signs with
numerical values, most notably the katapaydi system. While these successors were
not as unusual as ryabhatas system, they were far more successful.

Katapaydi Numerals
When later scholars experimented with alphasyllabic numeration starting in the
ninth century ad, they immediately saw that an alphasyllabary could also be
turned into a ciphered-positional system. Known as katapaydi, the signs of this
system are shown in Table 6.12 (Fleet 1911b; Datta and Singh 1962 [1935]: 70).
In this system, each V and CV syllable is given a value from 0 to 9. Unlike
ryabhatas system, changing the vowel of the syllable does not change its numerical value, so that ka = ki = ku = 1. Two of the signs (a and na) take on the
value of zero, as did isolated vowel-signs (those representing a V syllable alone,
without any consonantal component), which did not have a numerical value in
ryabhatas system. CCV syllables do not have their own numerical values, but
are considered to have the value of the consonant to the left of the vowel, so that
tva = va = 4 and ntya = ya = 1. As a result, any sequence of syllables can be assigned a numerical value, and any number has a wide variety of possible phonetic
transcriptions. Katapaydi numerals were read with the lowest power on the left,
as in ryabhatas numerals. Thus, the word bhavati or F 8 is had the numerical
value 644. The name katapaydi itself is taken from the four syllables (ka, ta, pa, ya)
that are assigned the value 1 in this system. Although it is unusual in that each
digit from 0 to 9 has several alphasyllabic values that represent it, structurally this
system is an ordinary ciphered-positional and decimal system.
The earliest example of the katapaydi numerals is from the Grahacranibandhana
by the astronomer Haridatta, written in ad 683 (Sarma 1999: 274). Datta and Singh
(1962 [1935]: 71) place its invention around ad 500 and claim that it must have
been known to ryabhata himself, but there is no textual evidence to support this
assertion. Haridatta was a direct intellectual descendant of ryabhata, and used
his predecessors system as the basis for his own. Nevertheless, as the katapaydi

Numerical Notation

210

Table 6.12. Katapaydi numerals

&

ka
1

kha
2

ga
3

gha
4

ua
5

<

ca
6

cha
7

ja
8

jha
9

a
0

wa
1

wha
2

sa
3

sha
4

ta
5

ta
6

tha
7

da
8

dha
9

na
0

pa
1

pha
2

ba
3

bha
4

ma
5

'

ya
1

ra
2

la
3

va
4

a
5

va
6

sa
7

ha
8

is ciphered-positional, like the general Indian positional numerals in ascendance


at the time, the existing positional numerals must certainly have influenced him.
Thus, this system is very likely a blend of ryabhatas numerals and the ordinary
Indian positional numerals. Aside from being able to express any number, it gave
every word a numerical value, and gave every number many corresponding words.
This would have allowed for the construction of various mnemonic devices to aid
scholars and students, and would have served a prosodic function (for astronomical texts were written in Sanskrit verse, which had strict metrical rules). The katapaydi numerals were also important in the Hindu traditions of number-magic,
divination, and chronograms in which the sum of the numerical values of the
signs of a word or verse produced a meaningful date.
The katapaydi numerals, as well as related systems that are identical except
for the use of local script-signs and the assigning of different digit-values to various signs, were used continuously throughout much of India for many centuries.

South Asian Systems

211

Several variants of the katapaydi developed, most of which changed a few numerical values or eliminated the values of certain categories of signs, such as the
isolated vowels (Datta and Singh 1962 [1935]: 7172). Some of these systems were
unique to one writer, while others were used in specific regions over a longer
period. Renou and Filliozat (1953: 708) claim that their use in paginating loose
manuscripts served a cryptographic function in that the pages of the text, once
jumbled, could be placed back in order only by initiates of the system. Katapaydi
numeration survived much more extensively in southern India, particularly in the
province of Kerala curiously, where the additive Tamil and Malayalam numerals were only recently and incompletely replaced by positional systems. Northern
Indian katapaydi numerals are rare, although Sarma (1999) discusses a sixteenthor seventeenth-century astrolabe labeled in the northern Devanagari script that
uses them, perhaps in imitation of the Arabic abjad numerals. They were still used
in astrological manuscripts and horoscopes in South India even in the late nineteenth century (Burnell 1968 [1874]: 7980).

Aksharapall
A third variety of alphasyllabic numerals, sometimes confused with the cipheredpositional katapaydi in the scholarly literature, is known as aksharapall numeration (after akshara, the word for the CV syllable-clusters that comprise the basic
unit of the Indian alphasyllabaries). Whereas ryabhatas system was multiplicative-additive, and the katapaydi system was ciphered-positional, the aksharapall
systems are ciphered-additive and decimal, assigning the numerical values 19,
1090, and sometimes also the low hundreds (but never as high as 1000, to my
knowledge) to a set of phonetic signs. It was used very widely for paginating
books, and was written in the margins from top to bottom with the highest
power at the top.
Unlike the first two alphasyllabic numerical notation systems, there was never
a single regular system for correlating signs with numerical values in the aksharapall. Datta and Singhs (1962 [1935]: 73) search through old manuscripts revealed
no fewer than three signs for 1, twelve different signs for 4, and nine signs for 60.
This is less complex than it seems, because within specific traditions, there were set
sequences of signs that would be understood by anyone working within them. In
some instances, parts of these sequences may be comprehensible; for instance, in
Nepali manuscripts from the eleventh to the fourteenth centuries, the numbers 1
through 3 were represented by the syllables e, dvi, and tri, which correspond to the
Nepali lexical numerals (Burnell 1968 [1874]: 66). In other cases, the signs used appear to have been assigned almost randomly. Datta and Singh (1962 [1935]: 73) list
many signs for which they cannot even attach a plausible syllabic value. Dialect

212

Numerical Notation

differences in pronunciation or paleographic variation among scripts may account


for the irregularity of the aksharapall systems.
The aksharapall numerals are ciphered-additive, but the lack of a regular correlation between the signs and numerical values suggests no obvious origin. They
may have developed directly from the Brhm ciphered-additive numerals, with
only the use of phonetic rather than abstract symbols to distinguish them. This
matter is made even more complex by the fact that many modern scholars still
maintain that the origin of the Brhm numeral-signs was as a modification of
phonetic signs (as mentioned earlier). Indian scholars considered the aksharapall
to be part of the varnasankhya tradition of alphasyllabic numeration, so I believe
they are related to the other alphasyllabic Indian systems, although possibly with
some influence from the Brhm system. The primary use of aksharapall numerals
was for the pagination of manuscripts. This may explain why there was apparently no need for any such system to express numbers in the high hundreds and
thousands.
Aksharapall numerals had the greatest and most consistent level of use of any
of the alphasyllabic numerals of India. They were used with great frequency in the
manuscripts of the Jains until the sixteenth century, although it is not clear why
this system would appeal specifically to Jains (Datta and Singh 1962 [1935]: 74).
They also survived for a very long time in Nepal (Burnell 1968 [1874]: 65). Temple
(1891) goes into great detail concerning a ciphered-additive numerical notation
system used for arithmetic by Hindu astrologers in Burma in the late nineteenth
century. While he does not indicate whether the signs were alphasyllabic, nor does
he use the name aksharapall (or any other name) to describe this notation, no
other ciphered-additive notation is likely to have been used in Burma at that time.
Aksharapall numerals appear to have thrived along the Malabar Coast; they were
still common enough in Malayalam-speaking regions in the middle of the nineteenth century to be included in some grammars (Bendall 1896). These groups are
all relatively distant from the central political and religious movements of India, so
their survival may reflect the marginal status of such places in Indian history. Aksharapall systems continue to be used occasionally throughout India, Bangladesh,
Nepal, Tibet, Burma, Cambodia, Thailand, and Java (Ifrah 1998: 484).
The tradition of Indian alphasyllabic numeration, while lasting well over a millennium and playing a significant role in Indian astronomy, astrology, and mathematics, did not influence numeration practices outside of South and Southeast
Asia. The different varnasankhya systems have in common a suitability for flexibly
representing numerical phrases in verse, but in mathematical traditions that did
not involve versification, alphasyllabic numeration was undesirable. Moreover, the
Arabic and Western scripts were unsuitable for modification to suit the unusual
structure of the alphasyllabic numerals, and both the Arabic abjad and the Greek

South Asian Systems

213

alphabet had their own alphabetic numerical traditions. Thus, while the Indian
tradition of ciphered-positional numerals spread fairly readily into the Arab world,
alphasyllabic numeration remained a strictly South Asian phenomenon.

Arabic Positional
The Arabic script is written from right to left, and is basically consonantal, though
with some representation of vowel sounds. The earliest Arabic speakers used the
hybrid cumulative-additive/multiplicative-additive Nabataean numerals (Chapter 3);
this was replaced after the Islamic conquest of Greek-speaking regions by an
alphabetic system (the abjad numerals) akin to the Greek, Hebrew, and Syriac
systems (Chapter 5). Shortly after the introduction of abjad numerals, however,
users of the Arabic script became aware of Indian ciphered-positional numerals,
and developed their own system on this basis, whose modern numeral-signs are
shown in Table 6.13.
This system is written with the higher powers on the left, rather than from right
to left following the direction of the Arabic script. Thus, 26,049 would be written
. Burnett (2000b) discusses this phenomenon in light of the fact that
Western numerals are read from left to right but in this case in accordance with
the direction of the script, noting that this may have led to some confusion when
the Arabic numerals were borrowed into the West as to the direction in which the
numerals should be read. In the modern Arabic numeral-signs, there are alternate
signs used for 4 ( ) and 5 ( ), and the modern zero sign is written with a dot instead of with a circle because the circle was already assigned the value of five.
There are undeniable paleographic resemblances between the Arabic positional
numeral-signs and those used in medieval north India. Table 6.14 compares the
Arabic positional numerals found in eleventh-century mathematical and astronomical treatises with the inscription found at Gwalior, India, dated to 876 ad,
containing the Nagari numerals used in medieval India. These signs are very similar, and it is thus safe to assume that the Arabic numerals have an Indian origin.
In some cases, as for 2, 3, 7, 8, and 9, the Nagari numeral-sign became rotated or
inverted, which may have resulted from the scribal practice of writing from top
to bottom, then rotating the manuscript to read it (Ifrah 1998: 532533). The fact
that medieval and modern Arabic scholars are unanimous in attributing an Indian
origin to these signs, and call them isb al-hindi (Indian numerals), abundantly
confirms the paleographic evidence.
The social context of the transmission appears to have been limited to the exact
sciences initially, specifically to astronomy. In 662 ad, a Syrian Christian bishop,
Severus Sebokht, noted the Hindu proficiency in astronomy, commenting that
as for their skilful methods of calculation and their computing which belies

Numerical Notation

214

Table 6.13. Arabic positional numerals


0

description, they use only nine figures (Nau 1910). The meaning of this statement is unclear, as he does not mention the zero, but it is likely that Sebokht was
referring to ciphered-positional numerals; if so, the Arab-speaking world probably
would have had some such knowledge as well, possibly through Persian intermediaries (Kunitzsch 2003: 3). In ad 773, an Indian embassy visited the court of the
Abbasid caliph al-Mansur in Baghdad, among whose members was an astronomer
who brought with him a copy of a Hindu work of astronomy, which was translated into Arabic (Folkerts 2001: 15).
Within fifty years of this episode, the mathematician Muhammad ibn Ms
al-Khwrizm wrote his Arithmetic (c. 825 ad) using ciphered-positional numerals
extensively, prompting later mathematicians and astronomers to follow his lead in
replacing the old ciphered-additive abjad numerals with the new positional system.
While al-Khwrizms work does not survive in its original Arabic (the earliest surviving manuscript is a twelfth-century Latin translation), al-Khwrizm knew of
positional numeration and advocated its simplicity and functionality. We do not
know specifically, however, what numeral-signs he used, and, indeed, almost no
contemporary texts containing positional numerals survive. The earliest direct paleographic evidence for positional Arabic numerals comes from an Egyptian papyrus
(PERF 789) with a numeral 260 at the bottom, but only if that numeral is the date
260 a.h., or 873/4 ad, a point that Kunitzsch (2003: 5) disputes. The Trkh of alYaqb of 889 mentions the sign for zero as being a small circle, without describing
the system further (Kunitzsch 2003: 4). The manuscript MS Paris, BNF ar. 2457 by
the astronomer al-Sijzi, written between 969 and 972 ad, provides secure evidence
for the mid tenth century, but this is a considerable gap in our direct evidence
from the numeral-forms (Folkerts 2001: 14; Kunitzsch 2003: 56). Nevertheless, the
textual non-paleographic evidence demonstrates that some Arabs were surely using
them at least by al-Khwrizms time, and possibly as early as ad 775.
Table 6.14. Early Arabic and Nagari positional numerals
0

Arabic
Nagari

South Asian Systems

215

From their origins in the late eighth and early ninth centuries ad, the numerals spread throughout the Islamic world, though not without resistance or
confusion. Many conservative scribes and bookkeepers resisted the new numerals in favor of older calculation on the fingers and with numeral-words.
In his Kitb al-muallimn, the ninth-century scholar al-Jhiz recommended
finger reckoning above the isb al-hindi because it needed neither speech nor
writing, a position echoed a century later by the historian al-Sl in his Adab
al-kuttb (Kunitzsch 2003: 45). Whether this functional explanation is complete, or whether other, ideological considerations came into play, is unknown.
Lemay (1977: 440444) questions the extent to which the Indian numerals
were known to the Arabs before the tenth century, and shows that there was
confusion among some Arabic thinkers over how they worked. While positional numerals began to dominate in both mathematical and nonmathematical
contexts starting in the eleventh century, astrological texts remained far more
conservative, retaining the abjad numerals solely until the fourteenth century
(Lemay 1982: 385386).
Contrary to the diffusion of most numerical notation systems, scientific functions rather than commerce or religion provided a significant impetus for the
transmission of the positional numerals from India westward to the Arabs. It is
nonetheless generally the case that medieval Arabic arithmetic did not distinguish
commercial, astronomical, divinatory, and other arithmetical practices unambiguously. The Arabs borrowed not only the Indian numerals, but also a host of computational techniques and devices, including the dust-board, a flat tablet strewn
with sand into which figures could be written for undertaking computations (Bag
1990: 290293; Folkerts 2001: 14). Other techniques available included a complex
Greek-derived system of finger reckoning and the use of counters or shells; accordingly, the use of written pen-and-paper arithmetic was apparently not part of the
initial practice of Indian-derived numeration.
The earliest major Arabic arithmetical text to advocate Indian numeration
instead was the Kitb al-fusl f al-hisb al-hind of al-Uqldis the Euclidean,
written in Damascus in ad 952/3, the earliest extant copy of which dates to ad
1186 (Saidan 1966; Burnett 2006: 16). Yet al-Uqldis recommended concealing the
Indian origin of the technique, using the first nine letters of the Greek alphabet
or the Arabic abjad instead of the numeral-signs for 1 through 9 (Saidan 1966:
478479). Many of the earliest Arabic arithmetic texts had names such as kitb
al-takht book of the board, confirming the association of the new numerals with
dust-board computation (Smith and Mourad 1927). Yet once the isb al-hindi
had been firmly established by the eleventh and twelfth centuries, Arabic mathematical texts began to advocate doing computations with paper and ink instead
of the dust-board, which of necessity involved rubbing out intermediate steps of

216

Numerical Notation

computations and thus could lead to error. Clearly, Arab mathematicians sentiments toward the innovation were complex both recognizing it as highly useful
in comparison to earlier techniques, yet desiring to improve it to suit their own
purposes and prejudices.
Despite their prominence, the Arabic positional numerals have given rise to few
structurally distinct descendants. Some of the cryptographic systems of the
Ottoman Empire (Chapter 10) may have been inspired by the Arabic positional
system, although most of these systems are additive and may be more closely related to the abjad numerals. A number of sub-Saharan African ciphered-positional systems developed in the twentieth century (Chapter 10) were created by
inventors who knew the Arabic and/or Western positional numerals.
While the debt of Western numerals to Arabic positional notation is undeniable, they are directly related to the ciphered-positional Maghribi numerals
(known sometimes as ghubar numerals, and in medieval Latin as figure toletane
Toledan figures) described in the following section, and are thus at best phylogenetic cousins of the signs commonly used in the modern Islamic world.
Burnett (2000c) notes, however, that routes of transmission were complex, and
the standard Arabic numerals were described in several Western European texts,
through the medium of the Crusader state at Antioch in the twelfth-century Latin
translation Liber Mamonis, and through the city of Pisa in the Latin translation of
the Hebrew works of Abraham ibn Ezra (see Chapter 5). These Eastern figures
(known in Latin as figure indice Indian figures) were, if never widespread, at
least known to Western mathematicians, but by the early thirteenth century, the
migration of Toledan translators to northern Italy marked the decline of Indian
figures in Western Europe (Burnett 2002a).
The Arabic numerals enjoy a degree of currency and use in the modern world
second only to the Western numerals. They are used regularly in most contexts
throughout all regions that employ the Arabic script, and are thus found regularly
from Morocco to Indonesia. While global commerce and the effects of mass media have introduced Western numerals into the Arabic-speaking world, and most
literate users of the Arabic script know them, it is unlikely that this will have any
long-term effect on the use of Arabic numerals.

Maghribi (GHUBAR) Numerals


A set of ciphered-positional numerals quite distinct from the regular Arabic system
was used in North Africa and southern Spain during the medieval era and sporadically thereafter. These numerals, sometimes known in Arabic as isb al-ghubar
dust-reckoning and commonly known in the scholarly literature as ghubar dust,
sand numerals, are particularly important for this study because they are the

South Asian Systems

217

Table 6.15. Maghribi numerals


0

Wc

Y e Z

[g

\ h

immediate ancestor of the Western numerals.7 While the paleographic forms of


the Maghribi numeral-signs vary, representative examples are indicated in Table 6.15
(Gandz 1931, Souissi 1971, Labarta and Barcel 1988).
In Chapter 5, I showed that North Africa and Spain were quite distinct from
the rest of the Arabic world, both in their use of a different ordering of the abjad
numerals and in their use of special Fez numerals. Likewise, comparing the
Maghribi numerals to the standard Arabic positional numerals (either the medieval or modern forms), while they are both decimal, ciphered-positional numerical notation systems written with the highest powers on the left, the two systems
differ paleographically.
Potential early examples of the Maghribi numerals come from two documents
dated from 874 and 888 ad, respectively, in texts from the Maghreb (Gandz 1931:
394). It is perhaps notable that the first textual example of the Maghribi numerals comes only one year after the appearance of the first regular Arabic positional
numerals known in Egypt, but this may simply reflect accidents of survival and
discovery. If these ninth-century texts actually contain Maghribi numerals, this
would accord well with their tenth-century diffusion into the Latin manuscript
tradition in Spain. However, Kunitzsch (2003: 11) argues, contrarily, that the earliest unambiguous Western Arabic/Maghribi numeral-signs in Arabic texts are from
MS Florence, Or. 152, dating to the middle of the thirteenth century (!). A good
part of this discrepancy arises from the fact that it is not a simple matter to distinguish the Maghribi numerals from the Eastern forms in early texts; clarifying the
origin of the divergence will help to resolve the problem.
Smith and Karpinski (1911: 98), Das (1927b: 359), and Datta and Singh (1962
[1935]) argue that the Maghribi numerals are closer to the original Indian forms,
and thus represent an earlier transmission, than the later Eastern Arab forms. They
claim from this that the Maghribi numerals were the ones used by al-Khwrizm
and other early mathematicians; however, such a conclusion is overly speculative. Another theory, popular from about 1915 until 1935, holds that the Maghribi
numerals (and hence Western numerals) came from India to Spain via NeoPythagoreans in Byzantium, while the standard Arabic numerals came from India via the caliphate of Baghdad (Carra de Vaux 1917; Cajori 1919; Gandz 1931:
7

Following Kunitzsch (2003: 10), who argues persuasively that the term ghubar has
been misapplied, I use the label Maghribi, reflecting the systems geographical origin.

218

Numerical Notation

395; Miller 1933; Lattin 1933: 184185). Yet there is no evidence for ciphered-positional numerals in Byzantine Greece prior to the twelfth century (Wilson 1981).
The rejectionist view is further refuted by the fact that medieval Arabs, Western
Europeans, and Byzantines were in accord that the numerals were of Indian origin
(Lemay 1982: 382). It is best now regarded as an ethnocentric relic of the misguided notion that Indians and Arabs were uncreative.
The term ghubar, with its unusual meaning of dust or sand, has prompted
some comment as to the function of the number. Das (1927b: 358) and Gandz (1931)
assert that the ghubar tradition represented a sort of Arabic abacus, but Kunitzsch
(2003) shows that texts discussing the system refer to a takht board on which writings were made and from which items could be erased that is, boards covered
with dust or sand that were used as calculating boards by drawing figures on them.
Some of the variation between the Arabic positional and Maghribi numerals may
be explained by their use on differing media, the former in permanent media and
the latter for arithmetical calculations on sand-boards. Their forms, thus fixed by
the separation of contexts, might have become entrenched through centuries of use
in disparate parts of the Islamic world. However, no actual dust-numerals in that
medium survive, of course. Kunitzsch (2003: 910) argues that despite the terms
isb al-hindi and isb al-ghubar being references to different media and computational techniques, they do not necessarily imply two distinct sets of graphemes
associated with each. While the term ghubar seems to have originated in Tunisia
and is associated with Maghribi scholars such as ibn Khaldun (whose fourteenthcentury Muqaddimah mentions only ghubar, abjad, and zimm [Coptic] numerals, not hindi) (Lemay 1982: 384387), the distinction in sign-forms is thus better
understood as a geographical rather than a functional one.
In fact, the ordinary Arabic numerals and the Maghribi numerals were quite
similar until the twelfth century; their numeral-signs for almost all values are similar enough to be explained as graphic variations of a common system of Indian
derivation (the medieval Nagari ciphered-positional system). In a tenth-century
manuscript written by the Persian astronomer Sijzi, the form of numerals used is
intermediate between the Arabic and Maghribi forms (Mazaheri 1974). Maghribi
numerals are thus a subset of the larger class of Indian-derived positional, decimal
numerals (Lemay 1977: 437), both of which stand in contrast to the abjad numerals described in Chapter 5. This does not mean that the transmission of positional
numeration was a singular event, but it does suggest that it was not a matter of
two distinct waves of diffusion, one into the Maghreb and the other into the rest
of the Arab world, but rather a significantly more complex series of episodes that
resulted in two parallel systems.
While the Maghribi numerals began as a paleographic variant of the Indian
numerals, they eventually took on a distinct cultural meaning among the scribes,

South Asian Systems

219

astronomers, and mathematicians of the western Islamic world. This is partly because of the relative independence of polities such as the caliphate of Cordoba
from the Baghdad-based Abbasid caliphate. The traditionalism of the Maghrebi
and Andalusians may partly explain why the numerals persisted even after the
rest of the Islamic world had adopted the signs now used throughout the modern
world (Ifrah 1998: 539). They were still regularly used in Spain and North Africa
in the fifteenth and sixteenth centuries, and sporadically thereafter (Labarta and
Barcel 1988).
The Maghribi numerals would be little more than a paleographic curiosity, merely
one of many ways of writing Arabic numerals, if not for the fact that through them,
Western Europe adopted ciphered-positional numerals. For the past seventy-five
years, all major scholars have agreed that the resemblances between Maghribi and
Western numerals, coupled with the first appearance of the latter system in medieval Spain, demonstrate this origin. Yet even as the Western numerals developed
along their own trajectory within the Christian European context, the Maghribi
numerals survived for almost a full millennium. Ifrah (1998: 535) provides examples
of arithmetical texts written using the system from as late as the eighteenth century,
and suggests that the system may have survived into the nineteenth century before
being completely replaced by the standard Arabic numerals.

Western Numerals
From their origin as a foreign and suspicious novelty during the medieval period,
the ten Western numerals, structured by the use of the positional principle, have
become so familiar that it is easy for the nonspecialist to forget that there are
other numerical notation systems. The ubiquity and universality of the Western
numerals make understanding their origin and diffusion all the more important.
Unfortunately, no monograph has dealt systematically with the topic since Hill
(1915), whose work is rather outdated as a result of advances in paleography.
The first example of Western numerals is generally held to be the Codex Vigilanus, written in 976 in the monastery of Albelda near the town of Logroo in
northern Spain, in which the numerals are described (in Latin) as Indian figures
(Hill 1915: 29; Burnett 2002b: 241). The nine units are listed, in descending order,
but no zero-sign, probably because the signs were intended for use with a counting
board. These signs are shown in Table 6.16.
These figures are very similar to the Maghribi numerals shown in Table 6.11,
and in fact there is no reason to consider them a separate system, except that they
are used in a Latin and Christian text from northern Spain rather than in an Arabic one from Andalusia. Toledo was a major center for the transmission of Arabic
knowledge to the Christian West in the tenth and eleventh centuries, and Lemay

Numerical Notation

220

Table 6.16. Western numerals (Codex Vigilanus, ad 976)


1

(1977: 444445) believes that later scholars became aware of ciphered-positional


numerals through reading Toledan texts.
These numerals found their way into slightly more widespread usage through
the writings of Gerbert of Aurillac (c. 9451003), who was to become Pope Sylvester II in 1000. Gerbert traveled extensively and studied arithmetic in Islamic
Spain in 967, at which time he almost certainly learned the Arabic numerals
(Folkerts 2003: 12). Thereafter, he helped renew interest in computing boards
in his Regulae de numerorum abaci rationibus (c. 980). Later authors such as his
biographer Richer credited Gerbert with having introduced the use of the formae
or notae, the nine numeral signs (excluding zero) on counting boards, replacing a
large number of tokens placed in any column with a single token bearing one of
these signs (Berggren 2002: 356357; Folkerts 2003: 2).8 No zero-sign was needed
because counting boards are positional by their nature, without the need for a
placeholder, although over time, a symbol was added, first (for terminus) and
later the familiar circle (Berggren 2002: 358).
These marked tokens, called apices, were used as a teaching tool by medieval mathematicians, known as abacists, between the tenth and twelfth centuries
(Evans 1977; Lemay 1977; Gibson and Newton 1995). Around thirty-five treatises
on calculation with the Gerbert abacus survive from this period (Folkerts 2003: 3).
Beaujouan (1947) has demonstrated that the apparent rotation of numeral-signs
written in many tenth- through twelfth-century texts is explainable by the fact
that the numerals on apices could be oriented in any direction when the tokens
were placed on a board. Once the apices computational technique was no longer
used, the numeral-signs were no longer rotated when written on paper. Yet this
early period of Western numeral use was relatively restricted, and did not spread
further than a limited group of mathematicians and astronomers. All of the
eleventh-century manuscripts containing Western numerals, and the vast majority of the twelfth-century ones, are from the abacist tradition and are didactic in
nature, describing the use of the counting board rather than using the numerals
for performing calculations (Hill 1915: 2931; Burnett 2006).
8

A remaining puzzle is why the Roman numerals for 1 through 9 were never employed on
the counters; although Burnett (1997: 11) argues that this would have been impractical, I
do not see any reason why even a long numeral-phrase like VIIII for 9 would have been
too long to have been used in this fashion.

South Asian Systems

221

The spread of Western numerals into the tradition of manuscript writing (in
both mathematical and other texts) began in earnest in 1202, at which time the
mathematician Leonardo of Pisa, better known as Fibonacci, promoted their use
in his Liber Abaci (Book of the Abacus). Despite its name, the purpose of Fibonaccis text was to promote not the use of the abacus, but rather the use of written
numerals for computation, with nine unit-signs and a zero-sign (k), following
Arabic practices advocated by scholars such as al-Uqldis and al-Khwrizm. Later
thirteenth-century texts following in Fibonaccis wake, such as Alexander de Villa
Deis Carmen de algorismo and John de Sacroboscos Algorismus vulgaris, used the
term algorismus, a corruption of the name al-Khwrizm, to refer to this new art
(Burnett 2006: 19). While he did not use the term himself, Fibonacci was thus the
forerunner of the algorithmists who, in direct conflict with the abacists, promoted
the use of written numerals for computation rather than the use of counting
boards (cf. Evans 1977, Murray 1978). This technique is the precursor to modern
computational techniques with pen and paper.
Despite the unquestionable importance of Gerbert, Fibonacci, and other mathematicians in introducing ciphered-positional numerals to the West and promoting their use, their eventual adoption is not a vindication of a great man theory
of history. Among the names given to the zero-sign in a late twelfth-century Latin
manuscript (Cambridge, Trinity College R.15.16, Fol. Av) was chimaera, suggesting that it was assimilated only with difficulty into the conceptual system of
medieval mathematical thought. Western numerals were not initially given the
same conceptual status as letters of the alphabet, or even as Roman numerals.
They were, instead, seen as characteres, signs to be made on physical artifacts, and
only gradually assimilated into texts as written signs (Burnett 2006: 29). The diffusion of the Western numerals from Andalusia and North Africa to the West occurred slowly, numerous times and by several different routes, some of which were
more fruitful than others (Gibson and Newton 1995: 316). In fact, Greek and Italian mathematical manuscripts contain the standard Arabic Eastern positional
numerals, not ghubar-derived ones, through the twelfth century (Burnett 2002b).
A late twelfth-century Latin manuscript from Bavaria contrasts the Toledan
figures and the Indian figures (Burnett 2002b: 241). Only in the thirteenth
century, when many of the important Toledan astronomers moved to northern
Italy, were the standard Arabic forms fully abandoned. Contact between the Arab
and Western cultural spheres followed several paths in the Middle Ages: through
Spain, to be sure, but also through Norman Sicily, along main trade routes from
African cities such as Tunis and Tripoli to Venice and Genoa, through the Crusader states such as Antioch, and through Byzantine Arabs (Burnett 2006).
Far from being an instantaneous adoption, then, the Western numerals were
used only by a small number of Western European scholars in the Middle Ages.

222

Numerical Notation

The ordinary populace of Western Europe used Roman numerals, if any, while
Eastern Orthodox regions used alphabetic systems such as the Greek or Cyrillic
alphabetic numerals. I have already discussed the various transitional and blended
versions of Western and Roman numerals used from the twelfth through seventeenth centuries, and the various medieval prohibitions enacted against the
use of Western numerals in Florence, Padua, and Frankfurt (Chapter 4). Most
notable among these are the Visigothic Roman/Western blended systems of
medieval Spain, where the forms of the Roman numerals affected the writing
of similar-appearing Western numerals (Lemay 1982). Also as mentioned earlier,
mixed Greek-Western and Hebrew-Western numerical structures were used in
some late medieval mathematical and astronomical documents (Chapter 5). In
astronomical texts, numbers up to 360 (i.e., degrees of the circle) continued to
be written in alphabetic numerals (usually Greek, sometimes Hebrew or Arabic),
while higher numbers were written in Western numerals (Burnett 2006: 20).
Whether we regard these blends as hidebound efforts to retain fragments of an
older notation or as progressive attempts to innovate, they illustrate the often
haphazard manner in which the Western numerals came to be introduced into
European scholarly circles.
Table 6.17 demonstrates the slow transmission of Western numerals throughout Europe, including both their first occurrence in each region and the period
in which they became more commonly known. In general, Latin and scholarly (particularly mathematical and astronomical) uses of the numerals preceded
their vernacular and commercial use by several centuries (Murray 1978: 193194).
Most of the earliest examples of the numerals in any given region are found in
mathematical treatises and texts designed specifically to explain the new numerals.
Where Western numerals were used in nonscientific contexts, they often served
cryptographic or secretive functions. One of the earliest such instances comes
from the legal documents of a notary from Perugia dating from 1184 to 1206,
in which the numerals indicated the numbers of lines of documents (Burnett
2006: 20). Similarly, the early thirteenth-century Genoese notary Lanfranco used
Western numerals only in the margins of his private documents to make records
of payments made to him by clients, while retaining Roman numerals otherwise
(Krueger 1977).9 Western numerals were used, however, as assembly marks on
timbers of the roof of Salisbury Cathedral in the 1220s; we do not know why this
was done, and there is no obvious reason why they would have been preferred over
Roman numerals (Tatton-Brown and Miles 2003). Only when the audience for
9

The quasi-alphabetic Fez numerals used by notaries in the Maghrib (Chapter 5) served
similar functions, but in that case the nonpositional system was the obscure one and the
positional system the more commonplace.

South Asian Systems

223

Table 6.17. Early Western numerals in Europe


Location
Spain

Italy

France

England

Germany/
Austria
Greece
Scandinavia

Iceland

First Attested Example


976: Codex Vigilanus (Hill 1915:
29)
c. 10501075: Pandulf of Capuas
De Calculatione
(Gibson and Newton 1995)
mid to late 11th century: abacus
treatises (Hill 1915: 29)
c. 1130: Adelard of Baths translation of al-Khwrizm
13th century: Wells Cathedral (first
epigraphic use)
(Morley 1947: 81)
1143: translation of al-Khwrizm
into Latin at Vienna
(Menninger 1969: 411)
12th century: commentaries on
Euclids Elements (Wilson 1981)
c. 12751300: Valdemars yearbook (Kroman 1974: 120)
c. 13101330: Haukr Erlendssons translation of Carmen de
algorismo (Benedict 1914: 17;
Bekken 2001)

Portugal

1415: Livro da Virtuosa


Bemfeitoria (Barrados de
Carvalho 1957: 124125)

Russia

17th century: sundials


(Ryan 1991)

Common Use
1490:a dating pages in texts
c. 1325: banking records and
account books in major cities
(Struik 1968; Menninger
1969: 428)
c. 1400: dating, accounting, etc.
15251550: archival records,
accounting books (Jenkinson
1926)
16001630: probate inventories
(Wardley and White 2003)
15251550 (Smith and
Karpinski 1911: 133)
c. 1400: Ottoman conquest of
most Greek-speaking areas
c. 1550 (books, manuscripts,
records)
c. 1550 (books, manuscripts,
records)
14901510: travelogues, scientific documents (Barrados de
Carvalho 1957: 125)
16th17th century: general use
(de Oliveira Marques 1996)
early 18th century (reforms of
Peter the Great)

Arabic documents from Spain used the ghubar numerals extensively from the tenth century ad onward; this date refers only to their common use in Christian Spain.

the numerals expanded from monks and astronomers did Western numerals begin
to replace Roman numerals more generally, however.
The common use of Western numerals in Europe was surely aided by the
transmission of double-entry bookkeeping from Italy in the fourteenth and fifteenth centuries; this technique greatly benefits from ciphered-positional numerical notation (Jenkinson 1926: 267). Yet the Italian merchant Francesco Datini
(c. 13351410), an early adopter of double-entry bookkeeping, switched his

224

Numerical Notation

accounts to Western numerals in 1366 but adopted double-entry accounts only in


1383, suggesting that the numerals may in fact have been the cause rather than the
effect (Crosby 1997: 205206). By 1600, however, Roman numerals were essentially
absent in most accounting traditions.
Continental Europeans started to date coins with Western numerals in the fifteenth century, the first being a Swiss coin from 1424; Austria followed in 1456,
and France, Germany, and the Low Countries in the final quarter of the fifteenth
century (Hill 1915: 94105).10 The earliest Western numerals on English coins are
those of an issue of Henry VIII (undated, but with the regnal name Henric 8)
from 152644 (Wardley and White 2003: 1516). The increasing complexity of the
alloying of coinage, particularly the need for complex divisions involving fractions, between the thirteenth and seventeenth centuries made the adoption of
Western numerals by goldsmiths and mint workers highly useful (Williams 1995).
Once coins began to be minted and records kept using the new numerals, their
spread to a large segment of the populace was inevitable.
The rise in frequency of the Western numerals corresponds well with the birth
of printing in the middle of the fifteenth century. The rise of literacy after the invention of the printing press, and the consequent expansion of the use of numeration to a broader range of people, correlated with a new willingness on the part
of the middle class to use the new invention for a variety of purposes, including
bookkeeping, inscriptions on coins and seals, foliation, and stichometry. Throop
(2004) emphasizes that the integration of Western numerals into Western typography was both a technical and a conceptual issue; typographers had to determine
not only how to distinguish 0 / o / O or 1 / l (a problem not yet fully resolved today)
but also how to produce numerals that accorded with the aesthetic canons of the
rest of the typeface, a task first accomplished by Claude Garamond (14801561).
Doing so further helped in the standardization of numeral-forms, which had been
extremely flexible prior to the fifteenth century (Hill 1915). Printed books were
first paginated in Western numerals in 1470: an edition of Chrysostomuss Homiliae in Rome (McPharlin 1942: 2021) and Werner Rolewincks Sermo in festo
praesentationis in Cologne (Archibald 1921: 423). Bibles began being printed using
Western numerals in the mid sixteenth century (Williams 1997).
In all of these cases, micro-scale processes and effects relating to users social status and proximity to major centers of learning affected the exact dates at which the
transition from Roman to Western numerals occurred. Wardley and White (2003)
have demonstrated that even within a single country (England) and considering
10

A copper coin of Norman Sicily dated to 533 ah (1138 ad) is the earliest positionally
dated coin in Europe, but it is inscribed and dated using the Arabic script and numerals
(Hill 1915: 16; Menninger 1969: 439).

South Asian Systems

225

only a single document type (probate inventories), regional differences in the


dates of adoption of Western numerals were as much as a century in Newcastle
and some regions near the Thames, Roman numerals were still used commonly
as late as 1700. Nevertheless, the general trend in all regions was always in favor
of Western over Roman notation. By the eighteenth century, Roman numerals in
Western Europe served primarily archaic and formal functions, ending two millennia of their effective domination. It is remarkable that a system that had not
yet been established in its heartland in the fifteenth century could almost entirely
replace not only the Roman numerals but also many other ancient systems in less
than three centuries.
The spread of the Western numerals throughout the world, and their eventual
replacement of large numbers of indigenous numerical notation systems, occurred,
however, only when European countries had become politically powerful, as the
modern world-system began to form with the Western European nations at the
core. The replacement of non-Western numerical notation systems began en masse
in the sixteenth century.11 Within a few decades of Europeans reaching the New
World, the Aztec and Maya systems had become obsolete and the Inka khipu
greatly restricted in scope. At around the same time, the ciphered-additive systems
of Eastern Europe and the Caucasus (Cyrillic, Glagolitic, Armenian, and Georgian) began to be replaced by Western numerals or Arabic positional numerals.
The modern era of colonialism brought about the replacement of further systems starting in the nineteenth century. The Hebrew, Coptic, and Syriac alphabetic
numerals all continue to be used for religious and formal purposes, but Western
numerals are used in most other contexts. The indigenous numerical notation
systems of South and East Asia have not been completely replaced, but they too
have been supplanted for many purposes by the Western numerals. While there
is no functional reason for the replacement of one ciphered-positional system by
another, the dominance of the European nations, coupled with the desire to have
a single, universally intelligible symbol system, have made the Western numerals
an attractive option. Even in places like Japan and Thailand, which were never
under direct political control by a European power, the Western numerals are
usually preferred.
By the late nineteenth century, the history of the Western numerals had been
sufficiently obscured that considerable disagreement arose regarding their purported origins. While the correct notion that they originated in India and were
11

Curiously, the sixteenth century also marked the development of the attribution Arabic
to the Western numerals, whereas earlier they had always been seen (correctly) as an
Indian invention (Clark 1929: 217). It is possible that this occurred due to growing
awareness of Arabic learning among early modern scholars.

226

Numerical Notation

transmitted through an Arabic intermediary was always prominent, Kaye (1919)


and Carra de Vaux (1917), among others, espoused the theory that their origin was
among Greek Neo-Pythagoreans, who transmitted the knowledge of positionality
to the Persians and ultimately to India. This imperialistic theory has no redeeming
virtues and no evidence in its favor; there is no evidence that the Neo-Pythagoreans
knew of place value. This theory rests chiefly on the misguided notion that India
has produced nothing of real scientific value.
At the same time, the spread of Western numerals in the nineteenth and twentieth centuries has spawned a host of descendants among North American and
African peoples, such as the Iupiaq, Cherokee, Oberi Okaime, and Mende systems (Chapter 10). Many of these are structurally different from the Western numerals, and are not simply the ordinary system recast with new numeral-signs.
Modern octal, binary, and hexadecimal numbers used for electronics make use of
the Western numeral-signs and ciphered-positional structure, merely substituting
different numerical bases. As well, while not structurally significant, graphic variations of many of the numeral-signs are quite common (e.g., ! vs. #, $ vs. %, & vs.
*, ( vs. )). Despite the near-universality of the Western numerals, the history of
numerical notation is thus far from complete.

Summary
We end in the modern era with a host of local numerical notation systems and two
(the Western and Arabic) that spread enormously on the heels of political conquests, but we should not forget the origins of these systems in the South Asian regional tradition originating with Brhm. The common feature of the South Asian
systems is the set of nine Brhm unit-signs which persist, though greatly altered,
in the surviving numerical notation systems and a decimal structure. Only the
alphasyllabic systems use distinct signs, the letters of the Indian alphasyllabaries.
It is common practice to end studies of numerical notation with the Western
numerals (e.g., Guitel 1975, Ifrah 1998). Yet to do so portrays the spread of Western numerals throughout the world as the inevitable replacement of worse with
better systems, in continuous progress from primitive beginnings to the perfection
of the Western decimal positional system, an achievement that can never be surpassed. Most surviving modern systems are ciphered-positional, which indicates
that they are useful, but this does not demonstrate that they are the inevitable conclusion of a teleological historical process. If technology truly spread only through
the diffusion of what is functional and the replacement of what is not, one would
expect that structurally identical systems should expand with equal rapidity and
geographical reach. The South Asian phylogeny, with so many decimal cipheredpositional systems surviving and in use, provides a good testing ground for this

South Asian Systems

227

theory. Because only the Arabic and Western numerals (and, to a lesser extent,
Nagari) have spread extensively, their diffusion must be due mainly to sociopolitical factors. Furthermore, the geographic distribution of the surviving nonpositional systems is no less than that of many positional ones. Why would the additive Tamil system be as widespread as, say, the Khmer system, if functionality is of
supreme importance? Do we really expect Tibetan ciphered-positional numerals
to survive and Chinese multiplicative-additive ones to decline?
I do not mean to suggest that functionality has nothing to do with the spread
of numerical notation systems, especially ones such as the Arabic and Western numerals that have been used extensively for accounting, arithmetic, and mathematics. Yet to proclaim the Western numerals spread as the triumph of functionality
and reason over illogic and unwieldiness is to ignore the many ciphered-positional
systems that have failed to spread or failed to survive. While, owing to the political might of nations that use them, the Western numerals are very important,
they are merely one branch of one phylogeny. In placing them in the middle of
my study, I choose to emphasize that their present triumph is neither inevitable
nor eternal.

chapter 7

Mesopotamian Systems

Numerical notation first developed in Mesopotamia around 3500 bc, contemporaneously with or slightly earlier than its development in Egypt. Scholars interested in the diffusion of Babylonian astronomy and mathematics to the Greeks
have long studied Mesopotamian numeration (Neugebauer 1957, van der Waerden
1963). Yet to depict the Mesopotamian phylogeny of numerical notation systems
as an archetypal case for the evolution of numerals, or to use it as the basis for a
universal evolutionary pattern, is dangerous. While Mesopotamian mathematics
is important for understanding later Greek developments (and, in turn, modern
Western mathematics), Mesopotamian numeration is nearly a historical dead end.
Although their history spans three millennia, the Mesopotamian numerals did not
spread geographically far beyond their point of origin, and did not survive when
placed under pressure from the numerical notation systems of later inhabitants of
the region.
The main signs of the Mesopotamian numerical notation systems are shown in
Table 7.1. There are several ways to classify them, depending on which features we
emphasize. Looking at the numeral-signs alone, the systems divide rather neatly
into archaic systems, used prior to 2000 bc and written using curviform symbols
made with a round stylus, and later cuneiform systems, written using wedgeshaped symbols. Both were written almost exclusively on tablets using a stylus to
impress signs onto wet clay. A second important distinction is between systems
that are primarily decimal and those that are primarily sexagesimal, or base-60.
228

229

Babylonian
positional

Old Persian

Hittite

f
f
f

g
g
g

AssyroBabylonian

Mari

B
B
Z

A
A
S

10

Sumerian

Cuneiform
Systems

Proto-Elamite
decimal

Bisexagesimal 2

Bisexagesimal

Sexagesimal

System
Archaic
Systems

C
C
P

60

f
i
w

100

Table 7.1. Mesopotamian numerical notation systems

M
Q

120

600

fm

gi

1000

N
R

1200

3600

7200

10,000

36,000

230

Numerical Notation

Mesopotamia is the only region of the world where sexagesimal numerical notation is attested.1 Finally, comparing the interexponential structures of the systems,
we can distinguish between additive systems, which include most of the systems,
and positional systems, of which the only true example is the Babylonian positional system.

Proto-cuneiform
Around 3200 bc or perhaps slightly earlier, the antecedent of the later Sumerian
script arose at the city of Uruk in southern Mesopotamia, during what is now
known as the Uruk IV period.2 This proto-script, which was probably read in Sumerian, lacked any means of expressing phonetic sounds. By the Uruk III period
(c. 3000 bc), it had spread from Uruk (the primary Mesopotamian city at the time)
to the north, to Jemdet Nasr, Khafaji, and Tell Uqair. The texts of this period of
Mesopotamian history do not represent a true literate tradition but rather a protohistoric system of bookkeeping and administration. In total, about 5,600 clay tablets have been recovered that record this script, known as proto-cuneiform. Around
sixty of the twelve hundred proto-cuneiform signs can be assigned numerical or
metrological values (Nissen, Damerow, and Englund 1993: 25).
Although Falkenstein (1936), who wrote the first comprehensive description of
the Uruk tablets, thought the proto-cuneiform texts from Uruk were both decimal
and sexagesimal, Friberg (197879, 1984) determined that there was no protocuneiform decimal system. In the 1980s, using computer-aided analysis of the
entire corpus of texts, Nissen, Damerow, and Englund (1993) established that as
many as fifteen distinct systems (of which five were particularly common) were
used at Uruk, each used for enumerating a specific category of discrete objects
or metrological quantity.3 By examining the maximum number of times each
numeral-sign is repeated, they determined the relative values between signs in
any given system (as if we were to infer that the Roman numeral V represents 5
by noting that I is repeated four times at most). This technique works because
cumulative-additive systems order powers within each numeral-phrase from

However, Price and Pospisil (1966) claim that the Kapauku of Papua New Guinea derived their sexagesimal lexical numerals from the comparable Babylonian numerical notation.
Over the past twenty years of research, the chronology of protohistoric Mesopotamia
has been shifted backward; older sources tend to regard the Uruk IV period as representing the early third rather than the late fourth millennium bc.
My discussion of the systems (including their functions) is derived almost entirely from
the work of Nissen, Damerow, and Englund (1993: 2529).

Mesopotamian Systems

231

highest to lowest and regularly replace lower power-signs with higher ones wherever possible. A difficulty is that a given numeral-sign may be found in several of
the proto-cuneiform systems, but its value often varies from system to system.
Thus, B is equal to 10 A in some systems, but to 6 A in others.
Yet while we can identify the numerical ratio between the values of any two
signs within a system, we often cannot identify the specific quantity represented
by any one sign. For systems used for counting discrete objects, it is easy to identify the basic sign for 1, since fractions of humans do not normally occur in texts,
but for systems that measure area or capacity, we can never ascertain with certainty
which sign (if any) has the basic value of one unit. Following Nissen, Damerow,
and Englund (1993), I present the values for these metrological systems as ratios,
since we can only tell the value of a sign relative to the other signs of the system.
Despite having different numeral-signs and different numerical values, all the
proto-cuneiform numerical systems have much in common. All are cumulativeadditive, although some individual numeral-signs are formed multiplicatively
e.g., D (600) = C (60) B\(10). Groups of identical signs were sometimes sorted
into two or three rows for easy reading, but this is not a universal rule, and some
tablets contain long strings of signs. Numerals were most often grouped with signs
arranged from highest to lowest (although there are some rare exceptions, which
may be scribal errors). A single numeral-phrase, together with one or more ideograms, was enclosed in a box in a section of the text.4
The proto-cuneiform texts are mainly accounting documents, often written on
both sides the obverse with a series of amounts of commodities, the reverse
with a single total. The greater-than-expected prevalence of round or nearly
round numerals in proto-cuneiform texts allows the identification of hypothetical
problems that were used as training exercises and thus were not actual economic
texts (Friberg 1998). Learning the various signs, the ratios among them, and how
to construct texts would have required considerable scribal training, including
school mathematics associated with the temple economy at Uruk (Robson 2007:
6364). For instance, the late fourth-millennium tablet W20044,20 is an exercise in calculating the area of an irregular quadrilateral field (Robson 2008: 30).
Throughout the history of the various systems use, only a very small portion of
the populace would have had access to the training necessary to master the protocuneiform notations.
4

Note that the proto-cuneiform script was written vertically in columns reading from top
to bottom, but I follow Assyriological convention (and that used by Nissen, Damerow,
and Englund) in showing the signs rotated ninety degrees counterclockwise and thus
read horizontally from left to right. This convention reflects a similar change in the
direction of writing cuneiform signs around the middle of the third millennium bc.

Numerical Notation

232

Table 7.2. Sexagesimal numerals


36,000
a

Sexagesimal (S)

3600
=10 E

600
=6 D

60

1/2

=10 C =6 B

=10 A

( =6 &

=10 ^

Sexagesimal (S)
a

10

=2

The letters in parentheses in this table and the following ones are those assigned to each system by
Nissen, Damerow, and Englund (1993) in their research.

Sexagesimal Systems
The two sexagesimal systems shown in Table 7.2 alternate between factors of 6
and 10, and were the first and easiest to be deciphered because their structure is
identical to that of the later Sumerian numerals. The main sexagesimal system
(S) is employed in slightly less than half the Uruk texts (Damerow 1996: 292). It
was used to enumerate most discrete objects: humans, animals, finished products,
tools, and containers, which explains its frequency of use. The subsidiary S system
was used to enumerate a much smaller category of discrete objects, such as dead
animals and jars of some liquids.

Bisexagesimal Systems
The two bisexagesimal systems shown in Table 7.3 are so named because an additional factor of 2 is interpolated among the factors of 6 and 10 used in the sexagesimal systems. While the regular bisexagesimal (B) system is identical to the regular
sexagesimal (S) system up to 60, it has new signs for the values of 120 (60 2),
1200 (120 10), and 7200 (1200 6). It enumerated discrete numbers of grain
products, cheese, and fresh fish, and is the second most common system found in
the archaic texts. The function of the identically structured but much less common B* system is unclear, but it may have indicated discrete quantities of some
kind of fish. Both systems appear to have been part of a rationing system, one for
which a number-sign between 60 and 600 may have been useful.

Table 7.3. Bisexagesimal numerals


7200
Bisexagesimal (B)
Bisexagesimal (B*)

1200
=6

120

60

10

=10 M

=2

C =6 B

=10 Q

=2

P =6 Z =10 S

=10 A =2

1/2

Mesopotamian Systems

233

Table 7.4. GAN2 numerals

=6

=10

=3

=6

=10?

GAN2 System
The GAN2 system shown in Table 7.4 is used to represent area measures. While
its signs are the same or similar to those of the common sexagesimal system (S),
the GAN2 signs values differ from those of the main system. For instance, where F
means 36,000 in sexagesimal numerals and is thus 3,600 times greater than B (10),
in the GAN2 system it is only ten times greater. This similarity probably has something to do with the use of round-ended writing styli in all the proto-cuneiform
numerical systems.

EN System
This uncommon notation system, shown in Table 7.5, is known from only
twenty-six texts, and may have represented weight measures. All but one of
the tablets from Uruk on which the EN system was used were found at a single
location, suggesting that whatever its function, it must have been very restricted
in use.

E Systems
This relatively common group of numerical systems, shown in Table 7.6, denoted
various capacity measures of grain. While its signs are similar to those of the sexagesimal systems, their order and the ratios between successive signs are quite
different. For instance, while the ratio between A and B is 10 in the sexagesimal,
bisexagesimal, and EN systems, it is only 6 in the E systems. The regular system
enumerated capacity measures of barley, the system for germinated barley for
brewing beer, and the * system for barley groats.

U4 System
This rather unusual numerical notation system becomes less so considering that
its function is for recording time and calendrical units. By combining a single
Table 7.5. EN numerals

=10

=2

=2

Numerical Notation

234
Table 7.6. E numerals
System

D =10 C =3

System

=3

System *

E =10 B =6 A =5 J
=10 =6 =5 =5
=10 =6 =5

ideographic sign with numerical signs for 1 and 10, all the major divisions of the
year could be expressed easily.

The Origin of Proto-cuneiform Numerals


The most popular theory on the origins of Mesopotamian numeration is that it
emerged from a system of clay tokens used for accounting in preliterate times.
Throughout Mesopotamia and even further abroad, small clay objects of various
shapes and sizes have been found in strata dating between 9000 and 2000 bc.
Oppenheim (1959) assigned an administrative function to a hollow clay ball, or
bulla, found at Nuzi inscribed with a brief cuneiform text enumerating forty-eight
animals, and containing within it forty-eight small stone counters. Amiet (1966)
showed that this technique was used much earlier than previously thought (since
at least 3000 bc) and that the bullae were double documents through which
transfers of goods such as livestock could be conducted while minimizing the
risk of fraud or error. A literate official could see the quantity of goods from the
inscription on the outside, but if there was any doubt, the bulla could be broken
open and the clay or stone tokens inside counted to match them up with the
actual quantity received.
More recently, Denise Schmandt-Besserat (1984, 1987, 1992) has shown that
the clay tokens are of even greater antiquity, and are ancestral to both the protocuneiform numerals and the proto-cuneiform script. She holds that the tokens
represent a stage of concrete counting, fusing quantity (the number of tokens)
and quality (different shapes representing different commodities), but do not represent abstract numbers (Schmandt-Besserat 1984: 55). A number of late fourthmillennium bc bullae, especially from Susa in modern Iran, are impressed with
Table 7.7. U4 numerals
\\+
*
10 months /
1 year

=10
=12

,
1 month

=3

10 days

=10

.
1 day

Mesopotamian Systems

235

signs resembling later archaic numerals and contain the correct total of tokens,
suggesting that the systems are connected (Nissen, Damerow, and Englund 1993:
127129). Furthermore, there are some similarities between the three-dimensional
tokens and the proto-cuneiform ideograms, suggesting that the tokens developed
into writing through the recognition that, if the total of a transaction is written on
clay, one need not actually use the clay tokens but need only record their values.
Schmandt-Besserats conclusions have been received with some skepticism (see
especially Lieberman 1980, Zimansky 1993).5 Firstly, the scope in time and space
of the token system is far greater than that of the proto-cuneiform numerals;
it is implausible that such a widespread phenomenon represents a uniform system. Moreover, in Schmandt-Besserats study of tokens from the Uruk-Jemdet
Nasr period (c. 3000 bc), around two-thirds of the tokens come from Susa in
Iran, while only 10 percent come from the very thoroughly excavated site at Uruk
(Lieberman 1980: 353). This suggests that the token system is unlikely to have
given rise to numerals and writing at Uruk. Finally, some of the most common
tokens are correlated with proto-cuneiform signs for rare objects such as nails and
days of labor, whereas given the accounting function established for the tokens,
we would expect livestock, people, and grain to be the most common tokens, as is
the case in proto-cuneiform texts (Zimansky 1993: 316).
This discrepancy points to a further problem. The archaic numeral systems
always place a numeral-phrase in front of an ideographic sign; 16 + sheep =
16 sheep, and so on. While the proto-cuneiform numerals partly fuse quantity
and quality, because different systems represent different commodities, they do
not do so completely, because one always needs a further sign to indicate exactly
what is being counted. With the tokens, however, there is no separation of numerals and the objects being counted; to show sixteen sheep, one simply uses sixteen
tokens for sheep. Thus, there is no correspondence between the archaic numeralsigns and the shapes of tokens. The use of tokens sealed within bullae appears to
have been an accounting technology that pre-dated, but then later coexisted with,
the proto-cuneiform numerals. While some early proto-cuneiform numerals are
found on clay bullae, this is insufficient evidence that tokens led to numerals.
Conversely, numerical signs resembling the proto-cuneiform ones have been
found, not on bullae, but on ordinary clay tablets in late preliterate contexts
at Uruk as well as at Jebel Aruda, Susa, and elsewhere (Nissen, Damerow, and
Englund 1993: 127130; see especially Figures 113, 114). These tablets have numerical signs only (no ideograms), and disobey the ordinary rule that once a certain
5

I cannot hope to address her claim that the tokens are ancestral to the proto-cuneiform
script, and will restrict myself to the similarities and differences between the token system and the proto-cuneiform numerals.

236

Numerical Notation

number of lower-valued signs have been written, they are replaced with a single
higher-valued sign. For instance, one tablet from Jebel Aruda contains twenty-two
B signs, among others (Nissen, Damerow, and Englund 1993: 130). In any of the
later systems, twenty-two signs would have to be replaced by a smaller number of
higher-valued signs. Because these inscriptions are found in late preliterate contexts and are similar but not identical to the proto-cuneiform numerals, they are
immediately ancestral to them and date from a period when the system was still
being developed, weakening the hypothesis that they derived from tokens.
While we do not know the language in which proto-cuneiform numerals were
read, the Sumerian lexical numeral system is mainly sexagesimal, and furthermore,
10 is a sub-base in the lexical numerals just as it is in the proto-cuneiform numerals (Powell 1971, 1972a, 1972b). On this basis, Powell (1972b: 172) has correctly
discerned that the presence of a sexagesimal system of notation in the archaic
texts from Uruk and Jemdet Nasr constitute [sic] the best indeed irrefutable
evidence that Sumerian is the language of those texts. Yet, as Hyrup (2006:
7982) points out, proto-cuneiform notation was not meant to record any spoken
language, but was an artificial recording system designed for limited purposes, and
only very gradually became assimilated to Sumerian.6
The multiplicity of proto-cuneiform numeral systems and bases is not lexical in
origin, but is likely based on Sumerian metrological systems in the Uruk period,
for which we have minimal nontextual evidence. We do, however, have substantial textual evidence for the metrological systems of the Early Dynastic and later
periods. The ratios between various signs in the proto-cuneiform numeral systems
dealing with measures of capacity, area, and weight are similar to the ratios found
in later Sumerian metrological systems. This supports the contention that the odd
ratios of some of the older systems are due to unattested metrological systems
that continued into better-documented periods. The increasing administrative
demands associated with the rise of the Uruk city-state in the late fourth millennium bc created a new need for record keeping, metrology, and accounting, of
which the numerals and the clay bullae are two distinct consequences.

Cognitive Consequences of Proto-cuneiform Numeration


The analysis of the proto-cuneiform numerals has also led researchers to speculate on
the possible cognitive correlates of the use of multiple numerical notation systems.
Damerow (1996) has argued that the material record from the archaic period in
6

Hyrup attributes this purpose as being the administration of a multiethnic society of


slaveholders and immigrant slaves, which we need not accept to recognize that the general principle is true.

Mesopotamian Systems

237

Mesopotamia directly reflects the numerical abilities of Mesopotamians, and represents a universal stage of concrete numeracy that precedes the modern abstract
number concept. In this respect, his argument is similar to that of Hallpike (1979),
who applies the insights of Piaget, Vygotsky, and others from developmental psychology to draw a parallel between individual cognitive development and the evolution of thought in societies. Damerow claims that the peoples of early Mesopotamia
could not conceive of abstract numbers, but were only capable of concrete counting
(Damerow 1996: 275297). Because multiple proto-cuneiform numerical notation
systems were used for representing different objects, and a single sign could have
different relative values in different systems, he argues that Mesopotamian scribes
could conceive of 8 sheep or 8 jars of oil but not simply 8. Taken to its logical
conclusion, this would imply that users of the proto-cuneiform numerals could see
nothing in common between eight sheep and eight jars of oil. I cannot see how this
can be the case; if so, it would be impossible to make any connection between eight
sheep and eight marks on a clay tablet, and numeration would be impossible.
Furthermore, in order for these context-dependent numerals to represent a
stage of archaic arithmetic in the evolution of numeration, as Damerow (1996:
296) suggests, we would expect similar systems to be present in other civilizations.
Yet nothing of the sort can be found in Shang, Predynastic Egyptian, or Zapotec
inscriptions, the other early and independently invented systems. This is not to say
that there are no cognitive consequences of the use of a dozen or more numerical
notation systems, but whatever they are, they will not be universally applicable to
every society.
Finally, we have no idea how many of these systems would have been known to
any individual official, and no evidence from the archaic period as to how numerals were manipulated and used arithmetically. We simply have values and totals,
which do not tell us very much about how people were actually thinking about
number.7 Even if individuals used many systems, this does not prove concreteness of
thought. In contrast to the proto-cuneiform numerical notation, there was a single
perfectly ordinary set of Sumerian lexical numerals (Powell 1971).8 Someone capable of abstract thought might well use multiple systems of numerical notation to
7

Liveranis (1983) intriguing conclusion that a fragmentary Uruk IV-period clay tablet
indented with holes may have served as a counting board has not been confirmed and
must remain tentative unless further finds are made.
Schmandt-Besserat (1984, 1992) has made much of the parallel between the many protocuneiform numerical notation systems and the use of numeral classifiers in Japanese,
the Mayan languages, and others, where the set of numerals is modified depending on
the class of object being counted. Numeral classifiers are not a feature of Sumerian. If
taken to its logical conclusion, this would imply that the modern Japanese do not have
a concept of abstract number.

238

Numerical Notation

prevent confusion as to the type of thing being counted, or to correspond to metrological systems. There is no qualitative difference between the Uruk systems and
the modern use of Roman numerals to distinguish the foreword of a book from its
main text, or the use of hexadecimal numerals for computing purposes. Ironically,
one of the principles behind the new mathematics movement in North America in
the 1960s was the claim that teaching students to calculate using numerical systems
of different bases would improve their understanding of abstract number concepts. I
believe the Uruk scribes had an abstract number concept, but realized that abstract
written numerals were not the most efficient solution to the problems they were
facing. The theoretical importance attributed to the proto-cuneiform numerals as
evidence of an evolutionary stage of cognition is entirely unwarranted.

Convergence and Decline


While they are an interesting early example of numerical notation, the protocuneiform numerals did not diffuse extensively or last for an extended period.
There are no significant resemblances between the proto-cuneiform numerals and
the Egyptian hieroglyphic numerals (Chapter 2), which may precede the protocuneiform systems in any case. Their only descendants were the proto-Elamite
systems used from about 3000 bc at the site of Susa and elsewhere in modern
Iran, and the later Sumerian numerals. The start of the Early Dynastic period
in Mesopotamian history marked a turning point in the history of its numerals.
Beginning around 2900 bc, there was a marked decline in the frequency of almost
all the proto-cuneiform numerical systems, while the sexagesimal system rapidly
assumed the functions of the other systems. While the system for measuring area
(GAN2) continued to be used as late as the Fara period (c. 2500 bc), it was in
decline and considered archaic by that point (Nissen, Englund, and Damerow
1993: 137138). While each metrological system had its own numerical notation
system in the archaic period, eventually officials decided it was better to express all
numbers, regardless of function, using a single notation.
There are three plausible explanations for this convergence. The simplest is that
the use of so many systems for so many different functions was cumbersome for
administration, potentially confusing, and open to abuse. This may simply be a
modern prejudice attributable to the Western use of only one set of numerals.
While 200 years is a short time in the context of world history, it is a long time
for a truly inefficient set of systems to persist. Secondly, while the archaic texts
were used at only a very few locales (mainly at Uruk), the later numerals were
used throughout Mesopotamia. If the Early Dynastic period marks the first era
when Mesopotamian numerals were employed for long-distance communication,
the use of a single system to facilitate communication among many individuals

Mesopotamian Systems

239

would be advantageous. Finally, changes in Sumerian metrological systems may


have reduced the usefulness of the proto-cuneiform systems by eliminating the fit
between metrology and numeration.

Proto-Elamite
Around 3100 bc, a ideographic writing system developed in southern and western
Iran, the region known as Elam in later Mesopotamian sources. This script, now
known as proto-Elamite, is attested in over 1,500 texts, mainly from the major
urban center of the region, Susa; most date from the Susa III period around 3000 bc.
A few other proto-Elamite texts have been found at Tepe Yahya and elsewhere in
modern Iran. It is a linear script, read from right to left and in lines proceeding
from top to bottom. The language it was intended to represent cannot be identified, but proto-Elamite numerals can be read. While the proto-Elamite ideograms
are different from those of early Mesopotamia, the proto-Elamite numerals are
very similar to the proto-cuneiform systems.
As with the proto-cuneiform numerals, confusion over the nature and number
of proto-Elamite numeral systems has delayed their correct decipherment until
recently. Brice (196263) provides a useful summary of several early twentiethcentury efforts to decipher the proto-Elamite numerals, all of which assume a
single decimal and cumulative-additive numerical notation system. An adequate
decipherment of the proto-Elamite numerals has been achieved recently through
the mathematical analysis of the corpus of proto-Elamite texts by Robert Englund
and Peter Damerow (Damerow and Englund 1989, Englund 1996). Damerow and
Englund realized that, as with the proto-cuneiform numerals, not only were there
multiple proto-Elamite numerical notation systems, but the relative values of
individual numeral-signs vary from system to system. There are five major protoElamite systems: three for counting discrete objects, another for capacity measurements, and another for area measurements (Englund 1996: 162).
The proto-Elamite numerical notation systems for counting discrete objects
are shown in Table 7.8 (Englund 1996: 162; cf. Potts 1999: 78).9 The three systems
are identical for 1 and 10, and the sexagesimal and bisexagesimal systems are further similar for 60. The sexagesimal system, like the proto-cuneiform sexagesimal
numerals, is not a pure base-60 system; instead, each successive number alternates
by factors of 10 and 6; that is to say, it has a sub-base of 10. In the bisexagesimal
system, the value 120 comes after 60 (a factor of 2).
9

As with the proto-cuneiform numerals, I have represented the numerals as they would
be read horizontally (following Assyriological convention; cf. Damerow and Englund
1989) rather than vertically (cf. Englund 1996).

Numerical Notation

240

Table 7.8. Proto-Elamite numerals (discrete objects)


1

10 60 100 120 600 1000 1200 3600 10000 Function


Inanimate
E
B C
D
objects

Sexagesimal

Bisexagesimal

A B

Decimal

A B

Grain
products

Animate
objects

The main signs of the systems for measuring capacity and area are shown in
Table 7.9. Because they are not used for discrete objects, they are represented in
terms of the ratios between values, not as discrete numerical values. These two systems are very similar (though not identical) to the E and GAN2 proto-cuneiform
systems, so, following Damerow and Englund, I have used those labels.
The striking resemblances between the proto-cuneiform and proto-Elamite
numerals make it certain that the latter were modeled on the former (Potts 1999:
7677). In fact, while the respective scripts are entirely dissimilar, it is a matter of personal preference whether we regard the proto-cuneiform and protoElamite numerals as distinct sets of systems or as two regional variants of a single tradition. Because the first texts from Uruk date to the thirty-third century
bc, while those found at Susa date to the late thirty-first century bc, the protoElamite ones cannot have been ancestral to those at Uruk, and must have diffused
from west to east in the context of interregional trade. Given the importance of
the Uruk city-state in the late fourth millennium bc, it is unsurprising that the
numerals would spread to Susa, the other major polity at that time. The main
difference between the two sets of numerical notation systems is the existence of
a decimal system for counting discrete quantities of animals and humans in protoElamite. It may be that the language of the writers of the proto-Elamite texts had
decimal lexical numerals, whereas we know that Sumerian numerals are primarily
sexagesimal.

Table 7.9. Proto-Elamite metrological numerals


Capacity
(E)
Area
(GAN2)

X =6 D =10 C =3 E =10

=6

=5

E =10?

=3

=6

J
K
A

=2

Mesopotamian Systems

241

The proto-Elamite numerals did not spread beyond Susa and a few other sites
in modern Iran. Brices (1963) tentative identification of similarities between the
proto-Elamite and Linear A (Minoan) numerals cannot be taken seriously as
indicative of a historical connection, given the geographical and temporal distance between the two. The proto-Elamite numerals ceased to be used around
2900 bc, following the decline of Susa as a major urban polity in the early part
of the third millennium bc and the subsequent rise of the various Mesopotamian
city-states. The numerals in the Old Elamite cuneiform texts, which are roughly
contemporaneous with the Old Akkadian texts in Mesopotamia, are derived from
later Mesopotamian systems rather than from proto-Elamite (Potts 1999: 79). The
proto-Elamite numerals are best seen as a brief florescence within a single citystate, rather than as part of a longer tradition.

Sumerian
The only system among the multitude of proto-cuneiform systems to survive
into the Early Dynastic period (2900 to 2350 bc) was the sexagesimal (or, more
accurately, the decimal-sexagesimal) system. While it was originally used only for
counting discrete objects, it began to be used for all numerical functions as the
older metrological systems were abandoned. At the beginning of the Early Dynastic, significant changes were taking place in the script of the region. The older
ideographic and curviform proto-cuneiform symbol system slowly transformed
into a writing system that used wedge-shaped (cuneiform) signs and expressed
phonetic as well as conceptual information. From this, we can tell that Sumerian
was the language in which the script was read. Yet, despite these alterations to
the script, the numeral-signs remained identical to the archaic sexagesimal ones.
One important change occurred around the twenty-seventh century bc, when the
numerals, like the entire script, underwent a ninety-degree rotation, so that they
were written and read horizontally from left to right rather than vertically from
top to bottom. The Sumerian numerals are shown in Table 7.10 (Nissen, Damerow,
and Englund 1993: 28).
These six numeral-signs were combined to make a cumulative-additive numerical notation system. Normally, groups of four or more signs were arranged in two
rows to facilitate rapid reading. Because this system has signs for both 60 and 3600
(= 602), it has a sexagesimal component. In a purely sexagesimal system, one would
need to repeat each sign up to fifty-nine times, which is impractical, but the Sumerian system also has sub-base signs for 10, 60 10 (600), and 3600 10 (36,000). The
latter two signs are multiplicative combinations of the small circle for 10 with the
sexagesimal signs for 60 and 3600. This decimal sub-base is similar, but not identical, to the use of the sub-base of 5 in the Roman numerals. While the figures of the

Numerical Notation

242

Table 7.10. Sumerian archaic numerals


1

10

G
B
Horizontal
A
B
14,254 = EEE\DDD
DD

Vertical

60

\A

3600

36,000

H
I
E
F
C
D E
F
CCCC\BBB\AA
CCC\
AA

(3 3600) + (5 600) + (7 60)

19 = BB

600

+ (3 10) + (4 1)

(20 LAL 1)

Roman sub-base (V, L, D) could occur only once in any numeral-phrase because
5, 50, and 500 are half of 10, 100, and 1000, respectively, the decimal signs in the
archaic Mesopotamian numerals could be repeated up to five times. The sign for
60 is a large version of the sign for 1, just as the sign for 3600 is a large version
of the sign for 10. Because the big 1 is 60 times greater than the regular 1, but
the big 10 (3600) is 360 times greater than its counterpart, I cannot agree with
Liebermans (1980: 343) suggestion that these signs represent the use of size-value,
which then evolved into place-value, over time. This feature simply derives from
the fact that two styli, one twice as large as the other, were used to impress numerical signs on clay tablets (Powell 1972a: 1112).
The Sumerian numerals provide the first evidence for the use of subtractive
notation to express certain numbers, especially those that end in 8 or 9 in the
Western numerals, using a Sumerian ideogram that corresponded with the phonetic value LAL. Thus, instead of writing 19 as one sign for 10 plus nine signs for 1,
it could be written as 20 1, as seen in Table 7.10. A sign or signs placed inside the
LAL sign indicated an amount to be subtracted from the signs preceding it. This
technique was used at Fara (ancient uruppak), perhaps as early as 2650 bc (Jestin
1937: Pl. LXXXIV). There is no evidence of subtractive lexical numerals in Sumerian
comparable to the Latin duodeviginti and undeviginti; the Sumerian words for 18
and 19 are etymologically 10 + 5 + 3 and 10 + 5 + 4, respectively (Powell 1971: 47).
Rather, this innovation had its origin strictly in numerical notation and the desire
to express numbers more concisely.
As in the archaic period, Early Dynastic numerals are found overwhelmingly in
economic or administrative texts, or in scribal exercises related to these functions.
In the archaic period, there was no indication how calculations were being done
(though calculations must have been made). At Fara, however, Sumerian tables
of squares, geometrical and arithmetical exercises, and other arithmetical aids

Mesopotamian Systems

243

Table 7.11. Sumerian cuneiform numerals


1

10

60

600

3600

36,000

216,000

k o

nm

LAL:

have all been found (Powell 1976, Hyrup 1982). Nevertheless, the use of numerals for representation, especially in administrative contexts, greatly exceeds the
frequency of their use for computation. There is nothing indicating the direct use
of Sumerian numerals for computation (by lining up columns, etc.), as in Greek
and Western arithmetic. Damerow (1996: 236237) laments the fact that, despite
the wealth of Early Dynastic economic records, we have no idea how multiplication was performed; he suggests that it must have been through a nonpermanent
means, such as counting boards, finger reckoning, or mental calculation.
By 2500 bc, the transition from the older Sumerian script to cuneiform signs
had been completed, except for the numerals. Beginning in the Presargonic period
(c. 26002350 bc), the older curviform numerals began to be replaced with a
set cuneiform numeral-signs, while remaining virtually unchanged structurally
(Powell 1972a: 13). This had the advantage of requiring only one stylus for all writing, whether lexical or numerical. While this trend appears to have been initiated
by the Sumerians themselves, it was hastened considerably, starting around 2350
bc, by the rise of Akkadian hegemony over Mesopotamia. These new numeralsigns are shown in Table 7.11 (Powell 1971: 244).
The signs for 1 and 60, which had previously been semicircular and horizontal,
became vertical wedges. The earliest cuneiform numeral-sign for 60 was written as
a big 1, just as it had been in the curviform numerals, but because the two signs
were made with the same stylus, the size difference was always minimal, and soon
the two signs became identical (Powell 1972a: 13). This feature does not mean that
the system used the concept of place-value, although it may have played a role
in the invention of the later sexagesimal positional system (Powell 1972a: 1314).
The old round sign for 10 was replaced by a Winkelhaken or corner wedge, made
by impressing the stylus onto the clay tablet perpendicularly, while the large round
sign for 3600 was represented visually by four (or occasionally five) wedges placed
in a rough circle. In other respects the writing of 600 and 36,000 as 60 10 and
3600 10, and the basic cumulative-additive structure the cuneiform numerals
were identical to the curviform ones. The phrase used for 216,000, not attested
in the earlier numerals, is a combination of the sign for 3600 and the ideogram

244

Numerical Notation

GAL big, and is quite rare (Powell 1972a: 7). Powell also describes an even more
complex phrase for 12,960,000 (216,000 60), argal unutaga, big everything
which hand cannot touch. For such lexical phrases, we need to ask at what point
a phrase ceases to become part of a numerical notation system. The subtractive
ideogram LAL is used in this system, as in the archaic one, but it is depicted using
two cuneiform wedges.
The replacement of the curviform by cuneiform numerals was a gradual process. While the older system required additional styli to write numerals, and its
numeral-signs generally took up more space, it stood out more clearly in a text
of cuneiform characters, making totaling easier (Powell 1972a: 12). Additionally,
a norm developed by which the two sets of numerals could be used side by side
to indicate different functions. Possibly the older numerals indicated quantities
directly counted, possibly using clay counters, while the cuneiform numerals
enumerated quantities of objects not actually present to be counted (Lieberman
1980: 344345). Alternately, Damerow (1996: 238) notes that some Early Dynastic
economic texts from Girsu use the older numerals for amounts of grain and the
cuneiform numerals for amounts of animals, and hypothesizes that this may have
been done to avoid confusing the two different categories when taking sums. This
use of multiple numerical notation systems is analogous to the modern use of
Roman and Western numerals side by side. The round Sumerian numerals had
been abandoned by the Ur III period (when Sumerian rulers regained control of
Mesopotamia), and are not attested later than 2050 bc (Powell 1972a: 13).
The Akkadian conquest, the most important political event of third-millennium
bc Mesopotamia, had a minimal effect on numeration. The Akkadian kings and
officials (c. 23502150 bc) were content to use the cuneiform and even the archaic
numerals for most of the same purposes for which they had been used in the Early
Dynastic period. More change in the numerals is visible in the Neo-Sumerian
Ur III period (2150 to 2000 bc), during which the archaic numerals disappeared
entirely. One slight modification that was tried in some Akkadian texts was to
write multiples of 60 using units followed by the Akkadian lexical numeral for
60, u-i (:), using multiplicative notation (Labat 1952: 244247). Thus,
instead of writing 120 as ff, it would be written as ff:. This is a much more
cumbersome representation, and probably was used in part to distinguish 120 (ff)
from 2 (ff). For the higher decades 70, 80, and 90 the regular Sumerian forms
were always used by the Akkadians (fg, fgg, and fggg, respectively). Regardless,
many Akkadian inscriptions where u-i could have been used are written in the
ordinary Sumerian fashion.
The Sumerian cuneiform system is ancestral to all the later systems of Mesopotamia. The Semitic cuneiform decimal systems (Eblaite and Assyro-Babylonian)
were directly derived from a Sumerian ancestor. The decimal structure of these

Mesopotamian Systems

245

systems reflected the lexical numerals of the Semitic languages of its users. While
Thureau-Dangin (1939: 107) believed this tradition to have been developed in the
Old Akkadian period (starting c. 2350 bc), it is now clear from the library at Ebla
that it developed as early as 2500 bc (Pettinato 1981). The sexagesimal cumulativepositional system used in Babylonian mathematics and astronomy was also
modeled on the Sumerian cuneiform system. It may have arisen in the Ur III
period, and was used by the twentieth century bc at the very latest (cf. Powell
1976, Whiting 1984).
The Sumerian cuneiform system continued to be used for most purposes until
the Old Babylonian period (c. 20001595 bc). Around that time, the Assyro-Babylonian decimal system began to be used for most administrative, commercial, and
literary functions, while the sexagesimal positional system was used for mathematics and astronomy once again perpetuating the tradition of using multiple numerical notation systems for multiple purposes. Several Old Babylonian
tablets provide translations from the old Sumerian additive numerals to the new
positional system (Nissen, Damerow, and Englund 1993: 146147), indicating
either a need to learn the new positional system or, alternatively, that the older
cuneiform system was already being forgotten. By the fifteenth century bc, it had
disappeared from regular use. However, a peculiar vestige of the base-60 Sumerian system survived in certain late inscriptions, particularly those indicating the
sizes of buildings (De Odorico 1995: 4). One such example is the Nameninschrift
of the Assyrian king Sargon II (722705 bc), which describes the dimensions of
the fortress at Khorsabad as 16,283 cubits, the numeral of my name, notated
using Sumerian numerals rather than the Assyro-Babylonian system that would
be expected (Fouts 1994: 207). In this case, the notation served literally to indicate
a numeral corresponding to the royal name of Sargon by correlating the signs for
3600, 600, and 60 with the phonetic values ar, nr, and , which could not have
been done using the Assyro-Babylonian decimal system. This inscription has been
compared to later Greek, Hebrew, and Arabic gematria using alphabetic numerals
(Chapter 5), but it seems to be an earlier and independent development.

Eblaite
The inhabitants of the city-state of Ebla (in the western part of modern Syria)
spoke a West Semitic language but were strongly influenced by Sumerian culture.
A great library of thousands of Eblaite cuneiform texts dating certainly to the
period prior to 2350 bc (the Akkadian conquest), and possibly as early as 2500
bc, provide us with ample evidence regarding the numerals used by the Eblaites
(Pettinato 1981). This system is almost identical to that used by the Babylonians
some centuries later, and reflects the shared Semitic language and culture of these

Numerical Notation

246
Table 7.12. Eblaite numerals
1

10

60

100

1000

10,000

mi-at
li-im
r-ba
A
B
C
f
g
f
AA\r-ba AA li-im AAA mi-at CBBAA
AA\\\\\\\ AAA

100,000
ma-i-at OR
ma-i-hu

10,000

1000

100 60 10 10 1 1 = 24,682

two groups in contrast to those of the Sumerians. As indicated in Table 7.12, the
Eblaite numerical notation system consisted of two sets of numeral-signs for numbers below 100, one curviform and the other cuneiform (corresponding with the
Sumerian archaic and cuneiform systems), but only one set of expressions for the
powers above 100 (Pettinato 1981: 183184).
The Eblaite system is cumulative-additive for values less than 100, and multiplicative-additive above that point. The signs for 1, 10, and 60 are ideographic
signs identical to those used in the two Sumerian sets of numerals. The two sets of
numeral-signs served quite separate functions: the curviform numerals were used
for basic enumeration and counting discrete objects, while the cuneiform numerals were used only for capacity measures such as the mina and gubar, as well as for
regnal years of kings (Pettinato 1981: 183184). The sign for 60 was used to express
the tens values in numbers between 60 and 99; its presence, a holdover from
Sumerian, is the major irregularity in an otherwise perfectly decimal system. The
signs for numbers above 100 are in fact the Eblaite lexical numerals and were
combined multiplicatively with the unit-signs as necessary.
Because it is decimal and multiplicative-additive above 100, this system required
only one ideographic sign (the crescent or vertical wedge) for the higher powers;
however, the repetition of intraexponential signs for the units, coupled with the
use of complex two- and three-syllable words, meant that numerals were fairly
long and cumbersome. To reduce this length, two features were often used. Firstly,
just as in the Sumerian system, subtractive numerals were sometimes used for
certain numbers to eliminate the need to write seven, eight, or nine unit-signs
by placing the subtrahend after the syllable lal or l. Secondly, the words mi-at
for 100 and li-im for 1000 were often shortened to the single syllables mi and li,
respectively. Thus, in one text, 7879 (expressing a number of gubar measures of
barley) is written in cuneiform numerals as 7 li 8 mi 60 10 10 l-1 (7 1000 + 8
100 + 60 + 10 + 10 1) (Pettinato 1981: 134). Such syllabic abbreviations are reminiscent of the Greek acrophonic numeral-signs (Chapter 4).

Mesopotamian Systems

247

Most Eblaite texts served economic or metrological functions. It is not clear


whether the Eblaite numerical notation was ancestral to the later Assyro-Babylonian system or whether the latter developed out of the Sumerian cuneiform
system in parallel to the Eblaite system. Because the two systems are very similar
in structure (even including their common use of multiplicative structuring above
100 with abbreviated lexical numerals), the possibility that the earlier Eblaite notation was borrowed by the Babylonians is a good working hypothesis. The Eblaite
system did not persist past about 2300 bc, after which point Ebla came under
Akkadian, and later Amorite, control.

Assyro-Babylonian Common
Because historians of mathematics are especially interested in the origins of our
base-60 units of time and the division of the circle, enormous attention has been
paid to the Babylonian positional numerals the cumulative-positional, base-60
system used for astronomy and mathematics. The far more common decimal and
additive numerals (which I call the Assyro-Babylonian common system), which
were used for most economic, monumental, and literary purposes throughout
Mesopotamia, are almost forgotten. This system came into common use in the
Old Babylonian period (starting c. 2000 bc), a position it would maintain for
over 1,500 years. The numeral-signs of this system are shown in Table 7.13 (De
Odorico 1995: 4).
The system is cumulative-additive below 100, multiplicative-additive above 100,
and is always written from left to right. For the most part, it is purely decimal.
The units were expressed cumulatively, except that 9 could be written using three
overlaid vertical strokes ($) as an alternative to writing it with nine strokes (9)
(De Odorico 1995: 4n). The tens values were usually expressed decimally using
one through nine Winkelhaken corner wedges for 10. The vertical wedge for 60 is
identical to that for 1, but unlike the Sumerian system, f was not normally used to
represent 60 alone, which would have created ambiguity, but only in combination
with signs for 10 and 1 to write numbers from 70 to 99. As in the Akkadian variety
of the Sumerian cuneiform system, the lexical numeral u-i (:) was used to
indicate 60 or multiples thereof; however, this phonetic form was not used to write
70, 80, or 90, and even for 60 its use declined over time. Therefore, there were as
many as three different numeral-signs for 60, one decimal () and two sexagesimal (: or f). Above 100, the multiplicative principle was used quite freely
and could be combined in various ways to express very high numbers. The sign for
100 was the syllabic sign ME, an abbreviation of the Babylonian word for 100,
meat, while that for 1000 was a multiplicative combination of the signs for 10 and
100. For instance, a scribe from the period of Sargon II wrote 305,412 as shown in

Numerical Notation

248

Table 7.13. Assyro-Babylonian common numerals


1

10

60

305,412 =

or

100

1000

gi

3i5gi4ia2
((3 100 + 5) 1000) + (4 100) + 10 + 2

Table 7.13 (Ifrah 1998: 139). In theory, this system could be extended as far as one
wished by juxtaposing signs for 100 and 1000 repeatedly, even though there was no
sign for zero. Furthermore, unlike the Sumerian system, in which the signs for 1
and 60 were identical, this system presented no ambiguities to the reader.
While this system was sometimes called Akkadian (Thureau-Dangin 1939), it
was rarely used during the period of Akkadian control of Mesopotamia and began
to predominate only during the Old Babylonian period. It originated in response
to the increased power of Semitic peoples in Mesopotamia in the latter part of the
third millennium bc: Akkadians, to be sure, but also Eblaites, Babylonians, and
others. Its structure reflects the decimal lexical numerals of the Semitic languages
rather than Sumerian lexical numerals, although the continued use of a special
sign for 60 gives testament to its descent from the Sumerian numerals. All of the
administrative, commercial, literary, and religious texts of the Babylonians and the
Assyrians were written using this set of numerals.
Perhaps the greatest significance of the Assyro-Babylonian common system is
the large number of descendant systems it produced. Earliest among these is the
system used at the city-state of Mari around 1800 bc, which blends features of this
system and the Babylonian positional system. In the middle of the second millennium bc, both the Ugaritic and the Hittite cuneiform scripts began using numerals based on the Assyro-Babylonian ones, in the context of Mesopotamian trade
with these polities. The Ugaritic texts written between the fifteenth and twelfth
centuries bc at Ugarit on the Mediterranean coast use the cuneiform ideograms
for 1 (f) and 10 (g); however, numerals in Ugaritic were normally written lexically
(Gordon 1965: 42). This was likely borrowed from the Assyro-Babylonian system,
but we have no idea whether higher numbers could be expressed through numerical notation. The Old Persian cuneiform numerical notation system, developed in
the sixth century bc (by which time Mesopotamia was under Persian rule), also
derived from the Assyro-Babylonian system rather than from any of the numerous
other systems then used in the region. Finally, as I argued in Chapter 3, the earliest
Levantine systems (Phoenician and Aramaic) developed around 800 bc as a blend
of Egyptian hieroglyphic (or perhaps Hittite) and Assyro-Babylonian influences.

Mesopotamian Systems

249

Table 7.14. Graphic changes in numeral-phrases


4
Sumerian
Babylonian

p
4

q
7

s
8

40

r t
9 d

The Assyro-Babylonian additive system flourished despite enormous political


changes. It was the system used for administration and commerce by both the
Babylonians and the Assyrians until the Persian conquest of Babylon in 539 bc.
Afterward, it began to be supplanted by the Old Persian cuneiform system and,
more importantly, by the Aramaic system that then became the principal Mesopotamian administrative and commercial system. Both these systems were indebted
greatly to their Assyro-Babylonian ancestor. It is unclear when the Assyro-Babylonian system disappeared entirely, but it was used at least to a limited extent
throughout the period of Achaemenid rule (539332 bc). The latest cuneiform
tablets date to the first century ad.

Babylonian Positional
The Babylonian positional numeral system is assigned such great importance
by many historians of mathematics that one could easily get the impression
that it was the only noteworthy form of Mesopotamian numeration. Despite
Neugebauers (1957: 17) warning that the positional numerals are a relatively minor
part of the body of Babylonian numerals, these sexagesimal positional numerals,
used for mathematics, have been assigned priority over much more widespread
systems (Sumerian and Assyro-Babylonian). In fact, positional numerals were
used in only a limited set of mathematical and astronomical contexts and over a
much shorter period, serving primarily as a means of easing certain computations
(Robson 2008: 756).
The system uses only two basic numeral-signs, the vertical wedge f for 1 and
the corner-wedge or Winkelhaken g for 10, to write any number between 1 and
59. Thus, small numeral-phrases were usually identical to those of the Sumerian
cuneiform system. Nevertheless, certain graphic changes (shown in Table 7.14)
were made to the numeral-phrases for 4, 7, 8, 9, and 40, so that, instead of grouping signs in at most two rows of up to five signs, three rows of no more than three
signs were used. This shift eliminated any phrases that placed four or five signs side
by side, and may have increased the systems legibility (Powell 1972a: 16).
Unlike the earlier cumulative-additive Mesopotamian systems, this system was
cumulative-positional, combining the two basic signs in multiple positions to

250

Numerical Notation

express powers of 60. It was thus a base-60 system with a sub-base of 10. It had an
additive structure within each power, because of the way that 10-signs and 1-signs
combine together, but a positional structure among different powers. Just as in
the Sumerian and Assyro-Babylonian systems, subtractive notation was frequently
used to write numbers such as 9 (10 lal 1) and 19 (20 lal 1) (Thureau-Dangin 1939:
106). According to the rules of the system, 4,252,914 would be written as a9
d4 b a4 = (19 603) + (44 602) + (20 60) + 14. In addition to expressing
integers, positional numerals could be used to express fractional powers of 60:
1/60, 1/3600, 1/216,000, and so on.
During the Old Babylonian period, the positional numerals did not have any
sign for zero to indicate an empty position within a numeral-phrase, nor was there
any way to distinguish an integer from a fraction (i.e., there was no sexagesimal
point). However, many texts list numbers in columns in which the positional
values of all the numbers are lined up with one another, making misinterpretation
less likely (Powell 1976: 421). When numbers are embedded in the middle of a
text or occur alone, the lack of a zero leads to ambiguity; there is no way, except
through contextual information, to determine which positional value expressed
which power, and thus a single numeral-phrase could have an infinite number of
readings. The simple phrase a2 3 could mean 723 (12 60 + 3), 43,380 (12
3600 + 3 60), 12.05 (12 + 3/60), and so on, depending on which positional values
we assume are indicated. When the empty position was both preceded and followed by numerals, this difficulty was sometimes obviated by using a large empty
space to indicate the empty position (Neugebauer 1957: 20). Thus, 1 b (80) could
be distinguished from 1
b (3620). Yet this technique was not used universally, and in some texts what looks to be a large space does not bear any numerical significance. Unless numeral-signs were arranged in columns, there was no
way during the Old Babylonian period to distinguish numbers where the empty
position came at the end or beginning of the numeral-phrase. Nevertheless, by
organizing numbers in columns, and through commonsense interpretations of
texts, Babylonian mathematicians would not have experienced insurmountable
difficulties in reading numbers despite these ambiguities.
The Babylonian positional notation probably developed, in fact, in the twentyfirst century bc, during the Ur III (Neo-Sumerian) period. The late Sumerian
system of weight units is purely sexagesimal and notated in a way that could be
ancestral to positional notation (Powell 1972a: 14). Powell (1976: 420) also found
positional numerals on several early texts, which led him to assert that the development of positional numerals occurred in the twenty-first century bc at the very
latest. Robson (2007: 7879) discusses twenty-first-century texts from the cities of
Umma and Girsu that clearly depict sexagesimal place-value numerals. The development of the notation may have resulted from Ur III administrative reforms,

Mesopotamian Systems

251

which followed from managing much larger amounts of goods than had previously been the case (Powell 1976: 422; Hyrup 1985: 9). Nissen, Damerow, and
Englund (1993: 142), however, remain agnostic regarding Ur III positional numerals, because most texts can be dated only paleographically, and the numerals do
not show much variation throughout time. I am unconvinced by Whitings (1984)
assertion that the positional numerals developed as early as the Old Akkadian
period (i.e., the twenty-fourth or twenty-third century bc).
Most of the early texts containing the positional numerals are mathematical
texts of the Old Babylonian tradition, and thus date between 2000 and 1600 bc,
with the majority from the latter part of that period (Powell 1976: 419). These
range from simple multiplication tables and arithmetical exercises to complex
problems that can legitimately be called algebra. Education in the positional notation and in pure mathematics was a significant component of scribal education, although of course only a small proportion of the total populace would
have had access to such training. Figure 7.1 is a small square clay tablet on which
a squaring exercise has been written in Babylonian positional numerals; the two
numbers on the left are each 2, 30 (150), and their product (22,500) is depicted
on the right as 6, 15 the final position, whose value would be 0, must be understood from context (Nemet-Nejat 2002: 262263). Many arithmetical exercises
and texts for translating numerals into the new positional system date from the
Old Babylonian period, indicating the existence of a vigorous process for teaching the system to scribes (Nissen, Damerow, and Englund 1993: 142147). Yet,
because nonmathematical texts did not contain positional numerals, scribes who
did not write mathematical texts probably would not have been familiar with the
positional numerals. The converse is not the case, however; mathematical texts
containing positional numerals are often dated in Assyro-Babylonian numerals,
showing that mathematicians also knew the common system (Neugebauer 1957:
17). The numerals were rarely attested to have been used to perform arithmetical
calculations directly, as opposed to writing down results. However, it is possible that calculations were made on clay scratch pads that could be moistened
and rewritten to record results, after erasing the preliminary work (Powell 1976:
420421). It is equally probable that some sort of counting board or wax writing
tablet was used, on which the intermediate steps of calculations were not preserved (Hyrup 2002). We do not know much about the precise steps by which
arithmetical computations were performed.
After the end of the Old Babylonian period around 1600 bc, attested texts
containing sexagesimal positional numerals are extremely sparse for over a millennium. The degree to which there was an actual decline in the systems use is
unclear, however. The small number of postOld Babylonian texts, in combination with the relatively small number of place-value mathematical texts regardless

252

Numerical Notation

Figure 7.1. An Old Babylonian clay tablet (YBC 7294); the numerals at left each signify
150 (2 60 + 3 10) and the product, 22,500, is at right. Courtesy Yale Babylonian
Collection.

of period, may suffice to explain the decline, or it may have actually waned in use.
Robson (2007: 154156) has published a small number of mathematical texts from
late second-millennium bc Babylonia that denote numbers using this notation.
In Neo-Assyrian texts such as lnam iur ankia, an omen text dating to 712 bc,
a technique called downwards upwards, upwards downwards, playfully altered
sexagesimal numerals by swapping positional registers, so that, for instance,
140 (2 60 + 20) became 22 (20 + 2 1) (Robson 2008: 148).
The positional numerals reemerged during the Seleucid period (the beginning
of which is dated from the Alexandrine conquest of 332 bc) in new contexts and
with several structural differences from the Old Babylonian system. First, subtractive expressions such as 20 lal 1 for 19 were no longer used in Seleucid texts
(Neugebauer 1957: 5). Second, while in the Old Babylonian period the positional
system was used exclusively in mathematical texts, by the Seleucid period it was
used in astronomical texts as well (Neugebauer 1957: 14).
The most important change was the introduction, in certain circumstances, of
a sign serving some of the functions of zero, usually written as 0 or %, to fill in an
empty position within a numeral-phrase. This sign could be used at the beginning
of a numeral-phrase to indicate that the ones place was empty (i.e., to distinguish
a fraction from an integer) or in a medial position to prevent misreading 3620

Mesopotamian Systems

253

as 80, as in the earlier example, but never phrase-finally (Neugebauer 1957: 20).
Neugebauer (1941) emphasized that the primary role of the zero-sign was as much
epigraphical as it was mathematical. He demonstrated that in a small number of
texts, a zero-sign was inserted, apparently superfluously, in numeral-phrases preceded by an amount in tens and followed by an amount in units. This was done in
order to preclude misreading b 7 (20 60 + 7, or 1207) as b7 (27); by writing
the former as b07, the latter interpretation is prohibited. In such numeralphrases, the zero-sign does not indicate an empty position, but simply separates two consecutive positions (Neugebauer 1941: 213). In fact, the zero-sign was
originally used to separate sentences, which may indicate that it originally merely
indicated separation or space. It is unlikely that the Babylonians conceived of zero
as an abstract number. No cuneiform positional text contains the bare numeralphrase 0; it always occurs in phrases along with other signs. Thus, 0 was not
equivalent to 0 in the same way that b was equivalent to 20. The abstract concept
of zero accompanied by a special sign for that concept developed independently
among the Greeks and Indians, but probably never among the Babylonians.
Despite its use of the positional principle, the system was restricted to an
extremely small group of Babylonian mathematicians and astronomers in the Old
Babylonian and Seleucid periods. It does not appear to have been known by merchants or most administrators, and certainly did not diffuse to other peoples of the
Middle East, such as the Aramaeans, Phoenicians, and Persians. There is no reason
to believe, as was formerly held by some, that the Babylonian positional numerals
survived long enough to be ancestral to the Indian positional numerals (Fvrier
1948: 585; Menninger 1969: 398399). The only direct descendant of the Babylonian positional numerals is the sexagesimal Greek positional system used by classical mathematicians and astronomers to represent fractions. Around the second
century bc, the Greeks combined the Babylonian system with their alphabetic
numerals to produce a positional, base-60 numerical notation system (Chapter 5).
Neugebauer (1975: 590) states that the use of a sexagesimal division of the circle
into sixty parts by Eratosthenes (ca. 250 bc) is the earliest evidence of this borrowing, although Eratosthenes did not use the sexagesimal fractions. This Greek system was used only in mathematical and astronomical texts, and only for fractions.
The quasi-positional cuneiform system used in a few texts in the city-state of Mari
also appears to derive in part from the Babylonian positional system.
While the Seleucid astronomical texts are important from the perspective of the
history of science, they represent the work of a limited group of scholars whose
knowledge was being surpassed by Greek mathematics and astronomy even in the
fourth century bc. The Greek alphabetic numerals were those used for everyday
purposes as well as for mathematics, and so, slowly, the Babylonian system fell
into decay. The last example of positional cuneiform numerals dates from the first

Numerical Notation

254
Table 7.15. Mari numerals
1

10

100

1000

10,000

century ad (Powell 1972: 6a). That this system, the first positional system ever
and one much lauded by modern scholars, should be replaced, after having nearly
been abandoned once before by its own inventors, suggests that the advantages of
positional systems do not correlate closely with their survival.

Mari
The city of Mari, located on the Euphrates River at the border of modern Syria
and Iraq, was an independent city-state between the twentieth and eighteenth
centuries bc. During this time, it was engaged in extensive trade relations with
Canaan and Babylonia. A large number of cuneiform tablets have been recovered
from Mari, mainly dating to the eighteenth century bc, upon which a very unusual numerical notation system has been found. This system is shown in Table 7.15
(Durand 1987; Ifrah 1998: 142146).
Below 100, the system is purely cumulative-additive. For the units, it is identical to the Assyro-Babylonian system, but in the tens, there is no special sign for
60; rather, the higher decades are written as , , , and , respectively.
For the hundreds, the multiplicative ideogram i (ME) was often used (preceded
by unit-signs for multiples). However, some texts omit the i sign entirely, turning the system into a quasi-positional one. Thus, 476 is written in one inscription
as p\\6 (Soubeyran 1984: 34). The sign for 1000 is a syllabic representation of
LI-IM, while the sign for 10,000 is created by superposing the signs for 10 and
1000. For powers above 1000, the system is multiplicative-additive; there are no
instances where higher power signs are omitted. The omission of the multiplicative sign for 100 is very interesting, but it does not represent a fully positional
system. If it did, we would expect 476 to be written as p\7\6 (4 7 6), not as
p\\6 (4 70 6). The Mari system resembles the experiments with positionality in India in the seventh century ad, before the positional principle was fully
adopted (Chapter 6).
Most of the administrative and commercial texts from Mari use the AssyroBabylonian common system, while mathematical texts mostly use the Babylonian positional numerals. One text (M7857) uses both systems side by side (Robson 2008: 130). We thus know that the Mari scribes understood these systems

Mesopotamian Systems

255

Table 7.16. Hittite cuneiform numerals


1

10

60

f or : i

7169=

100

1000

10,000

\<

7;\fi:9
(7

1000 +

1 100 + 60

+ 9)

perfectly well. What may have happened, at least in the case of mathematical texts,
is that the aberrant system was used unofficially (perhaps for calculation) and
then retranscribed for official purposes using the sexagesimal positional numerals
(Soubeyran 1984: 34). It is possible that the decimal structure of the Mari numerals was borrowed from the Assyro-Babylonian ones, and that the idea of using
positional notation for the hundreds was taken from the Babylonian positional
numerals. This system should be considered as an aberrant and short-lived experiment with positionality, as the conquest of Mari by Hammurabi in 1755 bc ended
its use.

Hittite Cuneiform
In Chapter 2, I discussed the Hittite hieroglyphic system, which was probably
borrowed from the Egyptian hieroglyphic system or the Linear B (Mycenaean)
system. A separate Hittite script, written in cuneiform characters and related to
the various Mesopotamian scripts, was used at the Hittite capital of Hattusha
between the seventeenth and thirteenth centuries bc. The numerals used alongside this script, shown in Table 7.16, are similar to those of the Assyro-Babylonian
common system (Rster and Neu 1989: Table 7).
The system is decimal and cumulative-additive below 100, while the multiplicative
principle is used for higher powers. Numeral-phrases are written from left to right.
To write 60 or multiples of 60, the Akkadian loanword u-i (:) was employed,
just as it could be in other cuneiform systems. Yet, when writing numeral-phrases
between 70 and 99, a simple vertical wedge represented 60 (Rster and Neu 1989:
271). The sign for 100 is simply the ME syllable of Assyro-Babylonian numerals borrowed into Hittite, while the complex and apparently lexical expressions for 1000
and 10,000 appear to be indigenous. Thus, 7169 was written as shown in Table 7.16
(Rster and Neu 1989: Table 7). There is no evidence for the use of subtractive notation in the Hittite cuneiform numerical notation system.

Numerical Notation

256

Table 7.17. Old Persian cuneiform numerals


1

10

100

604=

6wp

Because the royal archives at Hattusha are our main source for Hittite cuneiform inscriptions, we do not yet know a great deal about the range of functions for
which Hittite numerals were used. We can be quite certain that the numerals were
borrowed from the Assyro-Babylonian common system, which was widely used at
the time the Hittite numerals are first attested, given the two systems close similarity in structure and numeral-signs. There does not appear to be any connection
between the Hittite cuneiform and the Hittite hieroglyphic numerals, which are
entirely different in structure. With the collapse of Hittite power in the thirteenth
century bc, the cuneiform numerals ceased to be used.

Old Persian
The Old Persian script was invented early in the domination of the Achaemenid
Empire over Mesopotamia, probably near the beginning of the reign of Darius I
(522 to 486 bc) (Testen 1996). It is an alphasyllabary; thus, while its letter-signs
are cuneiform, it represents a distinct break from the older Assyrian and Babylonian scripts. The Old Persian numerals, shown in Table 7.17, closely resemble the
Assyro-Babylonian common numerals (Testen 1996: 136).
The system is decimal and cumulative-additive, with numeral-phrases written
from left to right. The sign for 100 combines multiplicatively with preceding unitsigns in the one attested text containing a large numeral-phrase. No known Old
Persian texts have numerals of 1000 or higher. Whereas the Assyro-Babylonian
units could be grouped in sets of two or three, in up to three rows, Old Persian
units and tens were arranged in at most two rows, with odd units represented at
twice the size of paired ones. There is no trace of sexagesimal notation in Old
Persian; 60 is expressed with six signs for 10 and 600 as 6 100.
Although the Old Persian script has traditionally been regarded as strictly
a display-oriented script used for prestige purposes related to the Achaemenid
monarchy, the recent translation and interpretation of an Old Persian administrative record from the Persepolis Fortification Archive has cast some doubt
on this notion (Stolper and Tavernier 2007). Figure 7.2 shows an Old Persian

Mesopotamian Systems

257

Figure 7.2. Old Persian tablet from the Persepolis Fortification Archive (Fort. 1208101,
obverse), which begins with the number 604, the only evidence for multiplicative notation
for the Old Persian hundreds. Courtesy of Matthew Stolper.

cuneiform text (Fort. 1208101, obverse) originally excavated in 1933 from Persepolis but published only in 2007, which begins with the numeral 604, probably
denoting a quantity of a dry capacity measure (Stolper and Tavernier 2007: 1213).
This numeral is the highest found to date in any Old Persian inscription; it is one
of only two to contain the sign for 100, and the only one to denote higher hundreds, thus attesting the hybrid multiplicative structuring. As an isolated document among many thousands of Elamite and Aramaic economic and other texts
found at Persepolis, Fort. 1208101 does not prove widespread Old Persian literacy
or numeracy in the Achaemenid period, but it does provide tantalizing evidence
that the scripts use was not as limited-purpose as was once believed.
The invention of the Old Persian numerical notation system occurred in a context of cultural contact between Persians and Assyrians/Babylonians in the late
sixth century. There are a number of Babylonian/Persian bilingual inscriptions,
and the two systems existed side by side at that time (as evidenced by the thousands of cuneiform documents at Persepolis). The absence of any attested trace of
sexagesimal notation is a unique feature that clearly differentiates this system from
its Assyro-Babylonian ancestor, however. While Fort. 1208101 shows that Old

258

Numerical Notation

Persian numerals (and indeed, the script) were used administratively in at least
one instance, the Aramaic numerals (Chapter 3) were the system used much more
commonly for international communication and commerce throughout Persia,
and the Assyro-Babylonian numerals were used alongside cuneiform inscriptions
in the Elamite language. The Old Persian system did not survive the Alexandrine
conquest of Persia.

Summary
The commonality among the Mesopotamian numerical notation systems is their
use of cumulative notation with signs for 1 and 10. In other respects, there is considerable variation among these systems, whether due to linguistic (Sumerian vs.
Semitic) or functional (administrative vs. mathematical) factors. The survival of
the sexagesimal base over a period of nearly 3,500 years is testament to the Babylonians preservation of Sumerian traditions, but most of the numerals used after
2500 bc were primarily decimal. The mathematical functions of the various systems, while interesting to historians of mathematics, are minimal in comparison
to their administrative and literary functions.
The conquest of Mesopotamia by the Achaemenids and later the Seleucids sounded the death knell for native Mesopotamian traditions, after which
Aramaic and then Greek numerals were used for most purposes. Even the positional numerals, the hallmark of Babylonian arithmetical achievement, quickly
disappeared under conditions of cultural and political domination. The AssyroBabylonian common system was borrowed and modified in regions of the Middle
East where Mesopotamian influence was strong, but the history of Mesopotamian
numerals is largely linear rather than branching, with each system giving rise to
its successor but not giving rise to many systems outside Mesopotamia. Multiple
systems were often employed at the same time within Mesopotamia, the use of
which was determined by their contexts in ways that remain unclear. With the
sole exception of the Levantine systems, which derive from both Mesopotamian
and Hieroglyphic ancestors, the Mesopotamian systems did not give rise to a large
number of descendants either within Mesopotamia or without.

chapter 8

East Asian Systems

The East Asian numerical notation systems, like the regions scripts, reflect the
pervasive importance of Chinese civilization over the past three millennia. The
classical Chinese system used from the Qin dynasty (221206 bc) to the present
day, and which spread throughout the region, is foremost in duration and significance. Still, the history of East Asian numeration is one neither of total Chinese
hegemony nor of complete stasis, and these systems are more diverse in structure
than any of the other phylogenies I have investigated. Attested historical connections and similarities in their numeral-signs allow us to identify their common
ancestry. Table 8.1 indicates the most common numeral-signs of the East Asian
systems.

Shang and Zhou


Chinese mythical histories record that the Yellow Emperor directed the scholar
Li Shou to create mathematics and the abacus (suan pan) in the twenty-seventh
century bc (Li and Du 1987: 1); however, the first well-attested phonetic writing
and numeration in East Asia dates to the latter part of the Shang Dynasty
(ca. 15231028 bc). The most common Shang inscriptions are records of royal
divinations written on bones and tortoise carapaces, called oracle bone inscriptions ( jiaguwen) by modern scholars, most of which were found at Anyang in
Henan province. These brief texts, which date from 1300 to 1050 bc, often contain
259

260

1 2 3 4 5 6 7 8 9 0 /

! @ # $ % ^ & * ( )

Kitan

Jurchin

Chinese
positional

10,000

1000

Chinese
commercial

100

10

Late rodnumerals

Rod-numerals

Chinese
classical

Shang /
Zhou

Table 8.1. East Asian numerical notation systems

East Asian Systems

261

Table 8.2. Shang numerals


1
1
10
100
1000
10,000
4539 =

f g
q <
:
(
~
)
*
>
o

h
>

i
?

j
$

k
%

n
&

numerical indications of tribute received, animals hunted, numbers of sacrificial


victims, or counts of days, months, or other miscellaneous quantities relating to divinations (Takashima 1985: 45). The attested numeral-signs used on oracle-bone inscriptions are shown in Table 8.2 (Needham 1959: Tables 22, 23; Djamouri 1994: 39).
The Shang numerical notation system combines the nine unit-signs with signs
for the powers of 10; it is thus multiplicative-additive and decimal. The numeralsigns for 1 through 4 are cumulative combinations of horizontal strokes, while the
signs for 10 through 40, only slightly less obviously, are ligatured combinations
of vertical strokes. The numbers 20, 30, and 40 were never expressed using multiplicative expressions involving the signs for 2, 3, and 4. As in most multiplicativeadditive systems, there was no sign for zero; if a particular power was not needed,
no sign indicated its absence in the numeral-phrase. These are perfectly regular
combinations of the sign for 10 and various unit-strokes. While Needham (1959:
1314) makes the case that the Shang numerals contain place-value components
because of the regular highest-to-lowest ordering of the powers, the signs for 10
and its powers cannot be omitted, and so it is multiplicative, not positional.
For the tens between 50 and 90, the unit-sign was placed below the sign for 10,
which was normally a vertical stroke but could apparently be a cross when writing
60. For the hundreds, the unit-sign was placed above the power-sign, while for
the thousands and ten thousands, the relevant unit-sign was superimposed upon
the power-sign. There was also a symbol for and which was placed between the
hundreds and tens, or the tens and ones (Martzloff 1997: 182).
The signs for 1 through 4 are simple ideograms, but otherwise most of the
symbols have semantic or phonetic correspondences with non-numerical words.

262

Numerical Notation

The sign for 1000 is identical to the Shang character for man, probably due to a
homophonic correspondence (Djamouri 1994: 1516; Martzloff 1997: 180181). By
contrast, the use of identical graphs for scorpion and 10,000 may result from
the association of an immensely large number with swarms of colonies of newborn
scorpions (Martzloff 1997: 181). The highest Shang number expressed is 30,000
(Martzloff 1997: 182). Numeral-phrases were written in vertical columns read from
top to bottom, with the highest power at the top.
In almost all the oracle-bone inscriptions, numeral-phrases are not found alone,
but are accompanied by a character for the object being quantified. On this basis,
Djamouri (1994: 33) regards Shang numeral-phrases as determinatives of nounphrases, and argues that each sign was read as a single morpheme in the ancient
Chinese language. Each numeral-sign corresponds to a single Chinese morpheme,
an atypical correspondence between language and numerals that leads Djamouri
to regard the Shang numerical notation system as a purely linguistic rather than
a graphic phenomenon. This feature, which it shares with the Chinese classical
system, raises the issue of whether we ought to consider such quasi-lexical formulations to be real numerical notation systems.
Like the Shang script, the Shang numerical notation system was independently
invented. Needhams (1959: 149) tentative suggestion of stimulus diffusion from
Babylonia rests on the dubious notion that the Shang numerals use place-value.
Moreover, the correspondence of numeral-sign and numeral-word suggests that the
Shang numerals have a linguistic origin. If the signs originated to represent Old
Chinese morphemes, this further confirms their indigenous development (Djamouri
1994: 1819). Some of the Neolithic marks on pottery and tortoise shells, such as
those dating from 66006200 bc found at Jiahu in Henan province, resemble
numeral-signs, which could extend this systems history back several millennia further in China (Li et al. 2003). Yet there are no numeral-phrases only single signs
among these marks, and in any case there is no way to be sure that their meaning
remained constant. The signs probably have a mixed abstract and phonetic origin;
more important than phonetic correspondences may be the fact that most of them
are graphically quite simple as compared to the other Shang characters. The function of the Shang numerals is quite clear, however, in the context of royal divinatory inscriptions.1 There is nothing resembling a Shang accounting text or commercial inventory parallel to those found in Mesopotamia or Egypt.
After the collapse of the Shang Dynasty, large parts of what is now China were
controled by the Zhou Dynasty, first from its western capital at Hao (1027770 bc)
1

Zhang and Liu (198182) go still further and argue that the oracle bones mark the beginning of the long-standing tradition of bagua milfoil divination in the pattern later
exemplified in the Yijing.

East Asian Systems

263

Table 8.3. Zhou numerals


1

10

j
r

l
r

Unattested

n
r

p
r

100

1000
10,000

and then, after the failure of the Western Zhou state, by a more decentralized
Eastern Zhou polity centered at Luoyang (770256 bc). The Zhou kingdoms continued to employ the script and numerals of the Shang. In the early Zhou period,
oracle-bone inscriptions continued to be written, but from the tenth to the third
centuries bc, Zhou numerals were often stamped on bronze vessels and coins
(Needham 1959: 5). The increasing complexity of Chinese society over this long
period brought the numerals into use for a much wider range of functions than
is documented to have previously been the case. While the Zhou numerals are
structurally identical to the earlier Shang ones, except that the sign for 10 could
also combine multiplicatively with the unit-signs for 2 through 4, the numeralsigns began to exhibit great graphic variability. Pihan (1860: 10) provides a comprehensive chart showing the various numeral-signs used between the sixth and
second centuries bc, containing, for instance, no fewer than thirty-eight different signs for 10,000. Despite this extraordinary paleographic variability, the signs
shown in Table 8.3 were the ones most commonly found on coins and bronzes until
the third century bc (cf. Needham 1959: Tables 22, 23). The power-signs immediately
ancestral to the classical Chinese ones were among the variants used in the late Zhou
period. These signs, shown in Table 8.4, differ greatly from those in Table 8.3.
By the late Zhou, multiplier-signs were no longer superposed onto or ligatured
with the power-signs; instead, numerals began to be written more regularly with
unit-signs preceding power-signs from top to bottom. As well, while the older
signs for 20, 30, and 40 were retained, more typically the signs for 2, 3, and 4 were
placed next to the sign for 10, just as with the rest of the tens. The system was still
Table 8.4. Late Zhou power-signs
10

100

1000

10,000

264

Numerical Notation

multiplicative-additive, only less opaquely so than previously. Just as there is no


sharp break in the forms of signs between the Shang and Zhou systems, neither is
there a distinct break between the Eastern Zhou numerals and those of the Qin
Dynasty; rather, the former gradually transformed into the latter. Yet, given the
rather important changes in Chinese writing that took place after the unification
of the country in 221 bc, I have chosen that point of demarcation to separate the
earlier numerical notation system from the classical Chinese system.

Chinese Counting-Rod Numerals


Before turning to the classical Chinese system, however, I will address a system
that developed alongside the written numerals of the Warring States period (476
221 bc). This system, known in Mandarin as suan zi, is both a numerical notation
system and a computational technology, translated in English as counting-rods.
Short rods known as chou or suan were used to compute on flat surfaces. While
these rods were often made of bamboo, they could also be made of bone, wood,
paper, horn, iron, ivory, or jade (Lam 1987: 369). Although it is poorly known in
the West, counting-rod calculation was the primary computational technology
used in East Asia before the sixteenth century, when the bead-abacus (Mandarin
suan pan, Japanese soroban) began to supplant it. Yet rod-numerals were not simply computational aids, but could also be written using vertical and horizontal
lines to represent the computing rods, as shown in Table 8.5.
The system is quite simple to learn and use; vertical and horizontal lines are sufficient to write any number. For the units position, vertical strokes signified 1 and
horizontal strokes 5; combinations of vertical and horizontal strokes indicated the
value. Conversely, for the tens, the values of the individual strokes were reversed,
so that horizontal strokes meant 1 and vertical strokes 5. Each successive position
was modeled alternately on the ones and the tens; positions in which the sign for
1 is vertical (ones, hundreds) are called zong, while those in which it is horizontal
are called heng (Needham 1959: 89). Thus, Chinese scholars learned the following
rhyme (Li and Du 1987: 10):
Units are vertical, tens are horizontal,
Hundreds stand, thousands lie down;
Thus thousands and tens look the same,
Ten thousands and hundreds look alike.

In the earliest rod-numerals (fourth century bc to third century ad), the use
of zong and heng numerals as appropriate to their position was not strict, so that
horizontal strokes could be used for ones and vertical strokes for tens. However,
the system had stabilized by the end of the Han Dynasty. No zero-sign was used at

East Asian Systems

265

Table 8.5. Early rod-numerals (Needham 1959: Table 23)

1s
10s
100s
1000s

J
A
J
A

K
B
K
B

L
C
L
C

M
D
M
D

N
E
N
E

O
F
O
F

P
G
P
G

Q
H
Q
H

R
I
R
I

762
7008
905,920
6.49

G
E

K
Q
O

this early date. Most authors presume that the numeral-signs were lined up strictly
by position, leaving blank spaces as appropriate, obviating the need for a zero;
however, Martzloff (1997: 187) notes that there is limited evidence for such spacing
in written rod-numerals (as opposed to physical counting-rods).
The rod-numerals constitute a cumulative-positional system with a base of 10
and a sub-base of 5. While it is possible to regard each sign such as R for 9 as a
single sign, thus making this system ciphered-positional, the systems true structure
is best reflected by classifying it as intraexponentially cumulative, which allows us to
recognize how the sign is constituted and to note its sub-base. While the numerals
6 through 9 are written using compounds of 5 and 1 through 4, the sign for 5 alone
is always five strokes; if a horizontal stroke were used for 5 in a zong position, there
might be more risk of confusion with the horizontal stroke for 1 in the next-highest
(heng) position. Because the direction of the strokes alternates with each successive
position, the rod-numerals are irregularly positional,2 since a sign takes its meaning
from both its position and its horizontal or vertical orientation. To put the sign G in
the tens position indicates 70, but to put it in the ones or hundreds position would
have violated the systems structure, except during the earliest phase of its history.
The rod-numeral system was infinitely extendable by using these two alternating sets of numeral-signs in successively higher positions. Decimal fractions could
be written by designating one of the places as the units position, with the places to
the right of that one representing 0.1, 0.01, and so on (Volkov 1994: 81). In numeralphrases containing both whole and fractional positions, the ones position could be
2

Martzloff (1997: 205) coins the term dispositional to reflect this irregularity.

266

Numerical Notation

identified by the presence of a character beneath it to indicate what sort of thing


was being counted (Libbrecht 1973: 73). Where numbers were arranged strictly by
columns, however, it was not necessary to include this extra sign. In addition, as
early as the Han Dynasty, negative numbers could be written, either by using different-colored rods (red for positive numbers, black for negative numbers) or by
placing an extra rod diagonally across the last nonzero digit of the numeral (Lam
1986: 188).
The earliest physical rods to be unearthed are several found at Fenghuangshan in
Hubei province, which date to the reign of Wen Di (179157 bc) (Mei 1983: 59). Textual and epigraphic evidence shows, however, that the rod-numerals were developed
much earlier. Coins from the Warring States period frequently contain rod-numerals (Needham 1959: 5). Similarly, Warring States earthenware bearing rod-numeral
signs has been found in Dengfeng County (Li and Du 1987: 8), so the system can
hardly have been developed much later than 400 bc. Yet its acceptance was not automatic. The Daodejing (Tao Te Ching), written in the early third century bc, advises
that [g]ood mathematicians do not use counting-rods, confirming that the system
was in use at that time, while also showing that it was not yet universally accepted
(Needham 1959: 7071). Yet by the time of the writing of Sunzi suan jing (The Mathematical Classic of Master Sun) in the fourth or early fifth century ad, counting-rods
were presumed to be the only foundation for arithmetical learning (Dauben 2007:
295). Counting-rods were not simply arithmetical tools, but served also as divinatory
instruments, money, and even to hold food (Martzloff 1997: 210).
While the rod-numerals originated as a means of computation, the late Zhou
numerals also may have influenced their development. While the systems differ
structurally, their signs are similar; the Zhou sign for 1 is a horizontal line and the
sign for 10 a vertical line with a dot. Because the early rod-numerals did not have
a regular orientation, a horizontal rod could indicate 1 and a vertical rod 10. Given
that the inventor(s) of the rod-numerals were probably literate, they would have
been familiar with the Zhou signs and may have borrowed them. The rod-numerals
cumulative-positional structure and quinary sub-base allow a limited number of
rods to express any number, though in practice, the use of physical rods would
have limited the number of positions that could be managed easily. Although the
rod-numerals are identical in structure to the Greco-Roman abacus (which predates the rod-numerals by at least two centuries), I attribute this similarity solely
to the two technologies common function.
Lam Lay-Yong (1986, 1987, 1988) hypothesizes that the rod-numerals were
ancestral to the Hindu positional numerals, because the rod-numerals are positional and decimal, and because there was considerable cultural contact between
China and India in the sixth century ad, when positionality developed in India.
Because the rod-numerals were used in computation and commerce, she asserts

East Asian Systems

267

that it is inconceivable that the Indians would not have learned of this system from
the Chinese, and, since it is so practical, they obviously would have borrowed it
(Lam 1988: 104). Yet the Indian positional numeral-signs are those of the earlier
Brhm numerals, not of the rod-numerals, and the rod-numerals have no zerosign (whereas the Indian system does). Moreover, the rod-numerals have a quinary
sub-base that the Indian numerals lack, and the rod-numerals are intraexponentially cumulative, whereas the Indian positional numerals are ciphered. No Indian
texts of the period mention rod-numerals or any other Chinese numeration.
In the sixth or seventh century ad, the numerals and rod-computation were
introduced into Japan at a time when Chinese cultural, religious, and political
influence in Japan was enormous. The original rods were long, thin, round, made
of bamboo, and called chikusaku; they were, however, quickly replaced by shorter
square rods known as sangi (Smith and Mikami 1914: 23). There is no evidence
of their use outside China, Japan, and Korea. The last coins to use rod-numerals
are the five chu coins of the Liang Dynasty (502557 ad), but these numerals are
highly irregular (de Lacouperie 1883: 316317). They continued to be written in
Chinese texts and used directly for computation.
In the twelfth and thirteenth centuries (the late Song Dynasty), the rod-numerals as written in texts though not their computing-rod counterparts transformed in three significant ways. Although this was a time of considerable political turmoil in China, due to invasions by groups such as the Jurchin and Mongols,
it was also a time of considerable scientific achievement. Table 8.6 indicates the
system as it was used at that time (Needham 1959: Table 22; Libbrecht 1973: 68).
First, new signs for 4, 5, and 9 were introduced, while the original (cumulative)
signs were retained. Because the only signs to change were those in which four
or five cumulative strokes had previously been required, this was probably done
to simplify the signs, though it meant that the written system differed from that
used with physical rods. These changes made the system less cumulative than it
previously had been, so that it approached a ciphered-positional structure. Second,
written numeral-phrases sometimes were condensed into single glyphs, compressing
the individual signs together so that they formed a monogram. Needham (1959: 9)
attributes this development to the requirements of the new technology of printing
books. Finally, a circle was introduced as a sign for zero. The first text known to use
a zero-sign is the Shu shu jiu zhang (Mathematical Treatise in Nine Sections) of Qin
Jiushao, published in 1247 (Libbrecht 1973: 69).3 Needham (1959: 10) suggests that
the idea of a circle for zero may have been an endogenous development, based on
the philosophical diagrams of twelfth-century Neo-Confucian scholars. I agree with
3

As I will discuss later, this text is also the first to use a circular sign for zero in conjunction with the classical numerals.

Numerical Notation

268
Table 8.6. Late rod-numerals

Zong
(1, 100, ...)
Heng
(10, 1000, ...)

Old Style
5804

E Q

M
S
D
S

N
T
E
U

R
V
I
W

0
0

New Style

Martzloff (1997: 207), however, that this development was more likely related to the
Indian zero, which had passed to China along with the transmission of Buddhism
in the eighth century ad. We may never know, however, whether the exact route of
transmission was through Southeast Asia, Tibet, or India proper.
The rod-numerals were linked directly with arithmetical computation from the
time of their invention. While they began as a system involving the physical manipulation of rods, Chinese mathematicians quickly adopted them for writing results of computations. The earliest strictly mathematical Chinese text, the Jiuzhang
Suanshu (Nine Chapters on the Mathematical Art), which dates no later than the
first century ad and summarizes the learning of earlier centuries, uses them extensively (Lam 1987: 367368; Li and Du 1987: 3337; Volkov 1994: 81). Thereafter the
rod-numerals were a central part of the computation techniques used alongside
most Chinese mathematical and astronomical texts until the sixteenth century.4
Most Chinese characters having to do with computation use the bamboo radical
because of its association with bamboo computing rods (Needham 1959: 72). Tong
(1999) asserts that overreliance on concrete, context-situated counting rods and
rod-numerals acted as a stumbling-block preventing the development of propositional mathematics in the Song Dynasty. Yet the modifications to the system, including the addition of a zero-sign, suggest that the rod-numerals, as an infinitely
extendable notation using the principle of place-value, could be used as objects
of arithmetical thought independently of their materiality. Dauben (2007: 191)
contends rightly, in my view that the perfectly regular and decimal character
of counting-rod arithmetic greatly facilitated the extraction of roots and advanced
work with linear equations.
4

We may never know the true extent of their use, since many printers considered the rodnumerals, with their vertical lines, to be insufficiently literary, and replaced them with
the classical numerals (Needham 1959: 8).

East Asian Systems

269

The introduction of the bead-abacus (suan pan) in the fourteenth or fifteenth


century brought this novelty into direct competition with the rod-numerals. The
earliest surviving suan pan dates from the sixteenth century, but the text Duixiang siyan zazi of 1337 indisputably depicts one (Martzloff 1997: 213215). Textual
sources indicate that the suan pan was perceived as being more efficient for computational purposes than the rod-numerals. The divorcing of rod-numerals from
the physical manipulation of rods made their use in written form rather archaic.
At first, the suan pan was an instrument for popular arithmetic, while computing
rods were retained by mathematicians and the elite (Jami 1998: 4). Throughout the
Ming Dynasty (13681644), they were used increasingly rarely in Chinese books,
and they had become a historical curiosity by 1600 (Cheng 1925: 493).5 None of
the many seventeenth- and eighteenth-century European scholars who mentions
the abacus also notes the rod-numerals (Needham 1959: 80), and in fact they first
came to the attention of the West in Biots (1839) antiquarian treatment. However,
rod-numerals were used in Japan for some time after they had been abandoned in
China, and were used actively through the nineteenth century in traditional Japanese mathematics (Menninger 1969: 368369; Martzloff 1997: 210211). A new
Chinese computing technique developed in the seventeenth century in which
computing rods were inscribed with classical numerals, probably under the influence of the system of numbered rods developed by the English mathematician
John Napier (Needham 1959: 72). This technique (similar to a slide rule) need be
given no attention here, since it is not a numerical notation system but simply a
computing technology that uses the Chinese classical numerals.
Two relatively recent numerical notation systems may be derived at least in part
from the rod-numerals. The Chinese commercial numerals employed in Hong
Kong and other regions since the sixteenth century use many of the rod-numeral
signs, combined with the multiplicative-additive structure of the classical Chinese
numerals (see the following discussion). It is less likely but still possible that some
knowledge of the rod-numerals and/or the classical Chinese numerals among
the inhabitants of the Ryukyu Islands south of Japan led to the development of the
cumulative-additive sho-chu-ma numerals (Chapter 10).
The manipulation of rod-numerals on boards appears to have been nearly as
important to ancient and medieval Chinese scientific and commercial calculation
as the bead-abacus would later be. Their origin and persistence must have had a
great deal to do with their efficiency for computational functions. However, this
supports rather than refutes my thesis that the history of numerical notation systems should be divorced from their use as mathematical tools. The rod-numerals
5

Wang Ling (1955: 91) reports that Chinese logarithmic tables were still written with rodnumerals in the early twentieth century, but I cannot substantiate this assertion.

270

Numerical Notation

and the classical Chinese numerals coexisted for nearly 2,000 years, and yet the
former had no noticeable impact on the latter. If there truly existed a unilinear
trend for positional systems to supplant additive ones, we would expect the rodnumerals to replace the multiplicative-additive classical numerals entirely, or at
least to facilitate their transformation into a ciphered-positional system.

Chinese Classical
The basic numerals associated with the Chinese script are perhaps the most stable
symbol system currently in use; the numeral-signs of the Qin Dynasty (221206 bc)
are practically identical to those used in modern Chinese literature. While there are
structural differences between that system and the way the numerals are normally
used today, ancient numeral-phrases are still easy to read. The basic numeral-signs
are shown in Table 8.7a, and a selection of numeral-phrases in Table 8.7b.
In traditional writing, numerals, like the script, were arranged in columns from
top to bottom, with the highest powers at the top. In modern writing, numerals are normally written in rows from left to right, although right-to-left writing
is not unknown, in which case right-to-left numeration is employed. The basic
system is multiplicative-additive; numbers are written by combining the signs for
1 through 9 with the appropriate signs for the powers of 10 to indicate their multiplication, and then taking the sum of these pairs of signs. There is no power-sign
for the units; the unit-signs for 1 through 9 stand alone. When writing 11 through
19, the unit-sign attached to 10 is always omitted, although in numbers such as
214 the unit-sign for the tens is often included. Prior to the Tang Dynasty, it was
optional to put the unit-sign 1 in front of 10, 100, 1000, and 10,000 in numeralphrases, but including the unit-sign 1 later became standard (Martzloff 1997: 185).
However, in modern Chinese, the powers of 10 alone can be written without the
unit-sign. When the multiplier of a power is zero, both the unit-sign and the
power-sign are always omitted. There is no zero-sign in the classical system,
although there is in modern Chinese numerals (to be described later).
In addition to these standard signs, there are three nonstandard signs used for
20, 30, and 40, which have their origins in the Shang/Zhou cumulative signs for
the lower decades (Needham 1959: 13). These signs were most often used in literary
contexts, for paginating certain texts, and when denoting days of the month. In
the fifth-century mathematical manuscripts found at Dunhuang in Central Asia,
however, they were also used for mathematical purposes (Martzloff 1997: 185).
They are still used occasionally, although the sign for 40 is very rare because it is
not needed to enumerate days of the month. It was always acceptable (and now is
preferred in most contexts) to use the standard multiplicative combinations of the
unit-signs 2 through 4 and the power-sign for 10.

East Asian Systems

271

Table 8.7a. Classical Chinese numerals


1

or

yi

er

san

si

wu

liu

qi

ba

jiu

10

100

1000

10,000

100,000,000

or

or or

shi

bai

qian

wan

yi

20

30

40

nian

xi

sa

or wan wan

Unlike Western numerals, which are grouped in chunks of three digits, Chinese
numerals are grouped in sets of four, using the character wan (10,000, or, if you
will, 1,0000) (Mickel 1981: 83). Any number from 10,000 to 100,000,000 could
be written by placing a multiplicative numeral-phrase from 1 to 9999 before the
sign for 10,000. The system did not stop there, however. By the first century ad,
multiples of 100,000,000 could be written by placing a multiplicative numeralphrase in front of two signs for 10,000 or by using a sign for 100,000,000, either
or (Martzloff 1997: 183). Another technique for expressing large powers of 10,
which developed early in the history of Chinese numeration, involved a complex
Table 8.7b. Chinese numeral-phrases
15

10

118

RU
100

74,002

100

10

10,000 4

1000

b7d2c4ba9
100

4,703,600,854

1,072,419

10

7 10,000 2 1000 4

100 10

4a73b6ad8b5a4
4

10

7 100 mil. 3 100

10 10,000 8 100

10

Numerical Notation

272

Table 8.8. Chinese higher power-signs

Sign

Phonetic Value

Lower Series
xia deng

Middle Series
zhong deng

Upper Series
shang deng

wan

104

104

104

yi

105

108

108

zhao

106

1016

1016

jing

107

1024

1032

gai

108

1032

1064

zi

109

1040

10128

rang

1010

1048

10256

gou

1011

1056

10512

jian

1012

1064

101024

zheng

1013

1072

102048

zai

1014

1080

104096

system of power-signs that was assigned three different series of values, as shown
in Table 8.8 (Needham 1959: 87; Martzloff 1997: 99).6 These power-signs combine
multiplicatively with the nine basic unit signs, and thus extend the basic system.
Needham (1959: 87) asserts that these signs first appeared in the Shu shu ji yi
(Notes on Traditions of Arithmetic Methods) of Xu Yue, and dates this text to
around 190 ad, but it may be a sixth-century forgery (Dauben 2007: 300). In any
event, these techniques are well attested from the fifth century ad onward, and
demonstrate a keen interest in extending the range of numeration far beyond that
needed for any practical purpose. While this system may seem hopelessly complex
and ambiguous, this confusion is identical to that resulting from the different values assigned to billion and trillion in American and European usage. In the lower
series, each exponent is one greater than the one before it; in the middle series,
each exponent (excepting wan) is eight greater than the one before it; and in the
upper series, each exponent is double the one before it. The first sign in all three
series is the standard sign for 10,000, and the second sign (yi) is one of the basic
signs for 100 million (thus corresponding to the middle and upper series, but not
to the lower one). Martzloff (1997: 9799) holds that these systems may have been
6

The middle series (zhong deng) values in Martzloff and Needham differ somewhat; I use
Martzloffs values in Table 8.8.

East Asian Systems

273

borrowed from similar Sanskrit systems transmitted at the time of the introduction of Buddhism into China in the middle of the first millennium ad. None of
these systems was ever in common use.
The Chinese numerals began to take their modern form starting in the third
century bc, developing directly out of the numerals used in the Warring States
period. With the spread of a unified administrative apparatus under the Qin and
Han Dynasties, they spread throughout areas under direct and indirect imperial
control. The unification of China led to many efforts to standardize the forms
of Chinese script and numeral-signs, although this was not accomplished to any
significant extent until late in the Han Dynasty. Figure 8.1 depicts a Han Dynasty
administrative calendar from 94 bc, found at Dunhuang (Gansu province), with
numerals used to enumerate months and days (Chavannes 1913: Plate XV).
Even as the signs of the system were being codified, however, Chinese writers
began to use calligraphic variants and other modifications of the basic system for
specific functions. These variants used different numeral-signs (ranging from mild
paleographic variations to radically different signs), but their structure is identical
to that of the basic system (decimal and multiplicative-additive). These variants
were strictly functional, not regional.
Perhaps the most important of these are the accountants numerals (da xie shu
mu zi), which developed in the first century bc (Needham 1959: 5, Table 22). Structurally, they are identical to the classical numerals, but while the classical numeralsigns are quite simple, the accountants numerals were intentionally made very
complex; thus they were considered more elegant and less susceptible to falsification. The signs are homophones of the phonetic values of the appropriate Chinese
words, so they bear no graphic resemblance to the basic signs. Hopkinss (1916)
analysis of their origin as phonetic variations of the standard numerals is dated but
quite thorough. Despite their name, they were used not only for accounting but
also, for instance, on thirteenth-century coins (de Lacouperie 1883: 318319) and
even in a mathematical text, the Tongwen suanzi qianban of 1614 (Martzloff 1997:
184185). Today, they are still used occasionally on checks, banknotes, coins, and
contracts in order to prevent falsification.
Another highly complex variant of the classical numerals are the shang fang da
zhuan, a variant set of numeral-signs that developed in the Han Dynasty (Pihan
1860: 13; Perny 1873: 113). These numerals are highly stylized linear versions of the
standard numeral-signs that were designed to be used on seals, and are still sometimes used for that purpose today. These signs are shown in Table 8.9.
The diffusion of the Chinese classical numerals was associated with the spread of
Chinese political influence throughout East Asia. In the late second century bc, the
Chinese numerals were employed in tributary regions such as the Gansu corridor in
Central Asia, the Vietnamese states, and the colony of Lelang (modern Pyongyang,

274

Numerical Notation

Figure 8.1. A Han Dynasty monthly


calendrical document from the archive
excavated from Dunhuang. The columns
of the register at the bottom left begin with
the numerals 23 and 22, with 20 indicated in
each phrase using the ligatured sign for 20.
Source: Chavannes 1913: Plate XV.

East Asian Systems

275

Table 8.9. Shang fang da zhuan numerals


1

10

100

1000

10000

North Korea). The Chinese numerals were borrowed directly (without any transformation) by the Japanese as part of the kanji characters starting in the third century ad.
The Korean hangul script developed in the fifteenth century has no corresponding
numerical notation system, but Koreans often used the Chinese classical system. The
numeral-signs associated with the chu nm script of the state of Annam (in modern
Vietnam) are simply the basic Chinese signs with additional phonetic notation; the
basic Chinese system was also known and used in the region (Pihan 1860: 2021).
The numerals associated with the scripts of non-Chinese peoples of China, such
as Tangut (Kychanov 1996) and Miao (Enwall 1994: 86), are also derived from the
basic Chinese system, although sometimes with considerable paleographic modification. None of these systems is structurally distinct from the basic Chinese numerals. Starting in the tenth century, China began several centuries of intensive contact
with its neighbors to the north and west; warfare with these nomadic groups and
the conquest of China in turn by the Kitan and Jurchin led to the development of
Chinese-inspired numerical notation systems among these two groups, which are
structurally distinct and described later.
The classical Chinese numerals were nonpositional and used no zero-sign for
over a millennium after their development. The positional principle was known
in China, however, through the cumulative-positional rod-numerals that had
been used since 400 bc. Moreover, Chinese mathematicians became aware of
Indian ciphered-positional numerals in the eighth century ad. Qutan Xida,7 an
Indo-Chinese Buddhist astronomer working at the Tang capital at Changan, first
reported the use of nine unit-signs with a dot for zero in his great astronomical compendium, Kaiyuan zhanjing, written between 718 and 729 ad (Needham
1959: 12; Guitel 1975: 630631). This transmission reflects the enormous scientific
contact that accompanied the introduction of Buddhism into China in the eighth
century ad. Yet the knowledge of ciphered-positional numerals had no attested
impact on traditional Chinese numeration for many centuries.
7

This name is the Sinicization of the authors original name, Gautama Siddharta.

276

Numerical Notation

In the mid thirteenth century, a period of scientific vigor during the late Song
Dynasty, the first zero-signs appeared alongside classical Chinese numerals in
mathematical texts. The first such text was the Shu shu jiu zhang of 1247, the same
document in which the zero-sign is first found with rod-numerals (Libbrecht
1973: 69). This modification allowed a circular zero-sign to be used whenever one
of the decimal powers in the middle of a numeral-phrase was empty. In theory, this
allowed Chinese mathematicians to use only the unit-signs from 1 through 9 in
conjunction with the 0 to express any number thus making the system cipheredpositional. Yet, during the Song Dynasty zero was used only to fill in empty medial
positions, while retaining the power-signs, so that where 12,001 would be written in the classical style as , it is written as in the
Shu shu jiu zhang, a less concise form that provides no other obvious advantage.
Whether this resumption of the use of the zero-sign was a result of the continuation of its eighth-century use, or a reintroduction from India or the Middle
East, is unknown. Starting in the late sixteenth and early seventeenth centuries,
when Chinese mathematicians of the Ming Dynasty were in extensive communication with the West through the intermediary of Jesuit missionaries, this form
of ciphered-positional Chinese notation was employed more regularly (Martzloff
1997: 185). Tables of logarithms appeared at this time, using the nine traditional
unit-signs and a circle for zero in a way identical to the use of the Western signs
0 through 9 (Menninger 1969: 461).
Before the sixteenth century, zero was employed only in mathematical and scientific texts. In the late Ming Dynasty, it began to be used more widely, but rather
than using the circular sign for zero found in the Song texts, a character, ling
() raindrop, which had been used to designate remainders in division, was
assigned the meaning zero. The first text in which it featured prominently is
Cheng Daweis Suan fa tong zong (Systematic Treatise on Arithmetic) of 159293,
which is also the first text to describe the Chinese commercial numerals or ma zi,
and additionally contains the first complete description of the bead-abacus or suan
pan (Needham 1959: 16, 7578; Li and Du 1987: 185187). In this and other early
texts, ling was used in exactly the same way as the circle-sign had been used previously, with one ling sign for every missing power, so that 30,008 would be written
as (3 10000 0 0 0 8). While the ling sign introduced an element
of positionality into the system, it was not fully positional, since the power-signs
were retained, and ling was used only in medial positions. Chinese writers soon
realized that they could omit all but one ling when multiple consecutive powers
are empty, so that one could write 30,008 simply as (3 10000 0 8). The
classical Chinese system normally uses ling in this manner today.
In modern China, any given number can be expressed in no less than six distinct ways, the choice of which depends greatly on context. Four of these forms are
variants of the classical system. For literary and prestige purposes, the pure classical

East Asian Systems

277

Table 8.10. Modern Chinese expressions for 20,406


Classical

Classical with ling (zero)

2d4b6 2d[4b[6

Ciphered-positional

Western

20406

20,406

Chinese numerals (without any sign for zero) are used, thus representing continuity of the system from the Qin Dynasty to the present. In most ordinary prose
writing, some sign for zero is usually introduced in the medial positions, while
retaining the power-signs. The use of ling has even spread to spoken Chinese, so
that the preferred way to say 203 is not simply er bai san but rather er bai ling san.
Where conciseness or clarity is desired, and in most scientific contexts, the nine
unit-signs along with a sign for zero are used positionally, as in the seventeenthcentury logarithm tables. In contexts where there is concern with forgery, the
accountants numerals can be used. The final two options are to use the commercial or Hangzhou numerals, which I will describe later, or Western numerals.
The Chinese classical numerals are ancestral to several numerical notation systems.
The ciphered-positional variant Chinese numerals used in modern mathematics are,
of course, one such descendant, as are, more indirectly, the Hangzhou numerals. The
structurally distinct numerical notation systems used by the non-Chinese Kitan and
Jurchin during the period in which they were in contact with (and later dominant
over) the Chinese are also largely derivative of the Chinese classical numerals. Finally,
two more obscure systems, the sho-chu-ma numerals used on wooden tallies on the
Ryukyu Islands, and the Pahawh Hmong system developed recently for use among the
Hmong of Laos, may also be derived in part from the Chinese system (Chapter 10).
Western numerals, while known in China from the seventeenth century, were
not widely used until the beginning of the twentieth century; the Shuxue wenda
of 1901, an arithmetical primer for use in elementary schools, was one of the earliest such texts (Martzloff 1997: 3536). In most scientific and technical contexts in
China today, however, Western numerals are preferred. Because the ciphered-positional Chinese numerals with the circle for zero had been used for several centuries
prior, this shift was strictly social and political, unrelated to structural considerations. Mao Zedong was amenable (at least initially) to the replacement of Chinese
numerals by Western numerals, as indicated in a 1956 speech that was later suppressed (DeFrancis 1984: 262263). Nevertheless, the replacement of Chinese with
Western numerals has not been uninterrupted or uncontested. Some institutions
reacted sharply to this trend, and anti-Western sentiment led to the replacement
of Western numerals by the corresponding Chinese numerals in certain academic
publications (DeFrancis 1984: 274275). Western numerals are well known to all
reasonably educated people in China. In Japan and South Korea, the dominance
of Western numerals is considerably greater than it is in China. Nevertheless, the
Chinese numerals continue to be known and taught in these countries, though

278

Numerical Notation

they are acquiring an archaic flavor. In China itself, however, the use of local numerals shows no sign of sharp decline, and there is every reason to believe Chinese
numeration will persist into the foreseeable future.

Chinese Commercial
The Chinese commercial numerals (often known as Hangzhou numerals)8 arose
in the sixteenth century. The numeral-signs of the system are shown in Table 8.11
(Needham 1959: Table 22).
Comparing these signs to those in Table 8.6, we see that all of the unit-signs,
save that for 5, closely resemble the late forms of the rod-numerals used during
the Ming Dynasty, although they have been borrowed haphazardly from the zong
and heng forms. The unit-signs for 1, 2, and 3 use vertical rather than horizontal
lines, showing that they are unrelated to the classical Chinese numerals. Hopkins
(1916: 318) explains the aberrant form of 5 as a form of the character wu, which is a
homophone of the Mandarin numeral word for five. The most common versions of
the power-signs for 10 through 10,000 are obvious variants of the classical systems
power-signs. The circular sign for zero was in use in both the rod-numerals and
the classical system. This evidence strongly suggests that the commercial numerals
originated as a blend of the late rod-numerals and the Chinese classical numerals.
The system is multiplicative-additive, with the zero used only to fill in empty
medial positions, never at the end of numeral-phrases. Unlike the Chinese classical
numerals, commercial numeral-phrases place the signs in two rows, with the unit
multipliers of the various powers on the top row and the power-signs, zero-signs,
and the signs for the ones position on the bottom row (Pihan 1860: 6). Numeralphrases were thus read in a zigzag fashion, starting at the top left, proceeding from
top to bottom and then diagonally up and to the right.
This basic system was made more complex by a large number of irregularities.
When the number being expressed was a simple multiple of a power of 10 (e.g.,
50, 800, 2000), the multiplier usually was placed to the left of the power-sign (as it
would be in the classical system) rather than above it (Perny 1873: 101). When the
number 10 occurred alone or in numbers such as 610 and 2010, the unit-sign 1
was always omitted, and the unit-sign could optionally be omitted when the sign
for 10 was combined additively with unit-signs, as in numbers such as 18 and
212. Moreover, the special classical Chinese numeral-signs for 20 (,) and 30 (.)
could be used in the commercial numerals where appropriate (Hopkins 1916: 319).
When there were two consecutive zero-signs in a numeral-phrase, they could be
placed one atop the other rather than side by side in the bottom row, as would
8

Other names for this system include ma zi, Suzhou numerals, and hua ma.

East Asian Systems

279

Table 8.11. Chinese commercial numerals


1

10

100

1000
or

w y

or

or

10,000 0

be normal. Finally, the standard classical unit-signs for 1 through 3 (horizontal


rather than vertical strokes) are sometimes used in the units position at the end of
numeral-phrases, though they cannot be used as multipliers in conjunction with
power-signs (Hopkins 1916: 319). The combination of all these irregularities and
options means that almost any number may be expressed in several valid ways.
Table 8.12 depicts a selection of numeral-phrases as written in this system.
We do not know exactly when the commercial numerals were invented, but
the earliest printed text that describes them is the Suan fa tong zong, published
in 1593 (Needham 1959: 5). Because they were not used for prestige purposes in
literature or mathematics, for example but were restricted to a limited set of
commercial contexts (invoices, bills, signs for prices, and so on), sixteenth-century
or earlier evidence of their use may not have survived. The rod-numerals, from
which the commercial numerals are partly derived, were obsolescent by 1600, so
it is unlikely that they would have been used as the basis for a new system as late
Table 8.12. Chinese commercial numeral-phrases

40,709

26

162

917

3008
5000

S\\ G
z0x0t
K
vF OR ,F
JF
JF
xaK OR xag
tJ
t
xvG OR xvG
L0
y0H
sy

10,000
2
10
1

6 OR
6

100 10
9
1
100
3

1000

20

2
9

10
0

1000 0

100

OR

100

10

280

Numerical Notation

as 1593. Yet early texts that mention them associate their invention and use with
the great commercial city of Suzhou (in Jiangsu province). As this city came to
prominence only in the sixteenth century, if the attribution of their invention to
Suzhou is correct, a presixteenth-century origin is unlikely.
As is suggested by their name, the commercial numerals were (and are) used solely
in commercial contexts. They are still used even today in some regions on bills, invoices, and signs in shops and markets (primarily to indicate prices), though their
use is waning in favor of regular Chinese numerals or Western numerals. They are
most common in regions where Cantonese is spoken, including Hong Kong.

Kitan
The Kitan (or Khitan) were an Altaic-speaking people who ruled Manchuria and
other parts of northern China between 907 and 1125 ad, a period now known as
the Liao Dynasty (Kara 2005: 7). While there was no Kitan writing before their
conquest of Manchuria, two scripts were developed shortly thereafter, the large
script and the small script, both based largely on Chinese, and possibly also
under the influence of the Central Asian Uygur script. Neither Kitan writing
system is fully deciphered, because the Kitan language is only poorly known, but
the meanings of the Kitan numeral-signs are understood. The numerals of the
large script are identical to the classical Chinese numerals, but the small script,
purportedly developed by the Kitan scholar Diela during the visit of an Uyghur
delegation to the Kitan court in 924 or 925 ad, had a distinct numerical notation
system. The attested signs of this system are shown in Table 8.13 (Kara 1996: 233).
While the Kitan numeral-signs have a vaguely Siniform appearance, they are
entirely dissimilar to the corresponding Chinese numerals, and must be of indigenous origin. Numeral-phrases are multiplicative-additive and are read vertically
in rows from top to bottom and then right to left across the page, as in traditional
Chinese writing. A slight ciphered element is introduced into the system by the
existence of distinct characters for 20 and 30; this practice is probably derived
from the analogous Chinese signs, and , although the Kitan signs are not
cumulative. It is not known how (if at all) numbers higher than 1000 were written.
Numeral-signs could also serve as phonograms; for instance, the symbol for 5 (tau)
was employed homophonically in the word tao-li hare (Kara 2005: 13).
Because Kitan writing is so poorly understood, it is difficult to know the total
scope of contexts in which the numerals were used. Kitan texts are known from
epitaphs on royal tombs, a text on a bronze mirror, some other stone monuments, and inscriptions on seals and vessels (Kara 2005: 9). Most texts were probably historical records of events, in which numerals are used primarily for dating.
The Kitan script and numerals did not outlast the period of Kitan independence,

East Asian Systems

281

Table 8.13. Kitan numerals

1
10
100
473 =

1
0
/

2
,

3
.

4
/
7
0
3

which ended in 1125 at the hands of the Jurchin. In 1191, the use of the Kitan script
was forbidden by Chinese imperial order, after which time no further Kitan texts
are attested (Kara 1996: 231).

Jurchin
The Jurchin (also Jurchi or Jurchen) were the rulers of what is now known as the Jin
Dynasty in the northern part of China (11151234), and one of the groups constituting the Manchu who conquered China in the seventeenth century. The Jurchin
script, which consists of logograms and syllabograms in addition to a set of numeral-signs, is attested from inscriptions and texts from the twelfth through fifteenth
centuries, but may have developed somewhat earlier (Kiyose 1977). The Jurchin
numerals are shown in Table 8.14 (Grube 1896: 3435; Kiyose 1977: 1323).
Jurchin numerals, like the script, were written in vertical columns read from
top to bottom, with the highest-valued powers at the top. The system is primarily
decimal; although the distinct numeral-signs for 11 through 19 suggest a vigesimal
component, it is a product of the fact that the Jurchin lexical numerals have distinct
words for 11 through 19 that are not connected to those for 1 through 9, but the
irregularity extends no further than the teens (Kiyose 1977: 133). Because the Manchu language, in contrast to Jurchin, had no such words, Jurchin numeral-phrases
could also be written using the sign for 10 additively with the signs for 1 through 9.
For writing numbers from 20 to 99, unit-signs from 1 through 9 sometimes were
combined with the power-sign for 10 as in the classical Chinese system, so the
Jurchin numerals appear to be multiplicative-additive. Yet there were also ciphered,
nonmultiplicative Jurchin numeral-signs for 20 through 90. For numbers above

Numerical Notation

282
Table 8.14. Jurchin numerals
1

HPX

XZH

LODQ

&

GXZLQ

XQD

QLQJX

QDGDQ

11

12

13

14

15

16

17

10

DNXQ

X\XQ

XZD

18

19

20

DPR LUKRQ JRUKRQ

GXUKRQ WRERKRQ QLOKXQ

GDUKRQ QL\XKXQ RQL\RKRQ RULQ

30

40

50

60

VXVDL

JXLQ WHKL

70

80

90

100

QLQMKX

QDGDQMX MKDNXQMKX X\XQMX WDQJX

1000

10,000

PLQJDQ

WXPDQ

100, the multiplicative principle was always employed. Thus, the Jurchin system
is structurally closer to hybrid ciphered-additive/multiplicative-additive systems,
such as the Ethiopic numerals (Chapter 5) and Sinhalese numerals (Chapter 6),
than it is to Chinese. In the Sino-Jurchin texts from the Ming Dynasty published
by Grube (1896), which date roughly to the period 14501525, only the unit-signs 1
through 9 and the power signs 10, 100, 1000, and 10,000 were used.
A Jurchin large script was introduced in 1120 by Wanyan Xiyin, and was based
on the Kitan script with significant Chinese influences; the script was officially introduced in 1145 by Emperor Xizong, with a number of small script characters
added (Kara 1996: 235). The Jurchin numerals are found on many monuments of
the Jin Dynasty and some manuscript fragments. The writings that survive are
historical and literary in nature, and the numerals in them are mainly dates. Our
best evidence for them comes from the Ming Dynasty (13681644), when Chinese
translators produced a bilingual glossary and translated documents, in which the
numeral-signs just described are found (Kara 1996: 235). The earlier signs differ
paleographically but not structurally from the Ming ones.
Although the Jurchin did not control large regions of China for very long, the
Jurchin script survived for several centuries. It was used on a Ming inscription of
1413, suggesting that it was not simply a historical curiosity, but was being preserved because it was being used (at least by some people). It continued to be used
until at least 1525, at which time Ming translators were still working with Jurchin
documents. The Jurchin were one of the major constituent groups of the Manchu who conquered China in the seventeenth century (in fact, the ethnonyms
Jurchin and Manchu may refer to a single group), but by this time they used

East Asian Systems

283

either the classical Chinese numerals or the ciphered-positional, Indian-derived


Mongolian numerals.

Summary
Chinese numerals are central to the history of the East Asian systems. Today, the
classical Chinese numerals (along with positional variants) occupy a role parallel to
the supremacy of the Roman numerals in Europe prior to 1500, despite the increasing use of Western numerals for science, technology, and commerce. This systems
continued strength (at least in China) suggests that it will continue to thrive, especially in nontechnical prose writing. We must also take into account the strong
cultural preference for Chinese symbol systems when analyzing the present state of
the Chinese numerals; functional considerations alone cannot account for it. The
increasing rarity of the Chinese numerals in Japan and Korea represents not only
the functional rejection of an inefficient system, but also resistance to a Chinese
cultural feature in favor of the more international Western numerals.
The Chinese numerical notation system as used today is enormously variable
in structure, and employs a host of representational techniques. On the surface,
this appears hopelessly nonfunctional, and we might question why such a system
would survive. I think that its quasi-lexical nature the fact that Chinese numerals act as both ideographic script-signs and graphic numeral-signs renders this
variability both comprehensible and rational. If Western numerals incorporated
archaisms such as score, or accounted for the fact that 1400 can be one thousand
four hundred but is more commonly fourteen hundred, similar eccentricities might
arise. The Chinese classical numerals are well suited to being read because they account for the irregularities in spoken Mandarin. The basic multiplicative-additive
structure of the system permits all sorts of structural manipulations, such as the
occasional use of positionality or ciphered signs for the lower decades, without
creating any ambiguity. The systems flexibility and its correspondence with language are thus advantages rather than hindrances.
The comparison of this phylogeny to the ones I have discussed previously is
quite instructive. In Chapters 2 through 7, most systems employed a single common structural principle. By contrast, the East Asian systems display considerable structural variety: cumulative-positional (rod-numerals), ciphered-additive
(Jurchin), ciphered-positional (Chinese positional variant), and multiplicativeadditive (Shang/Zhou, Chinese classical and commercial, Kitan). Yet there can
hardly be any doubt that these systems comprise a cultural phylogeny. The historical connections among systems are well established, and the similarities in the
numeral-signs are quite strong. If we were to rely on structural qualities alone, we
would be at a loss to describe their cultural history.

chapter 9

Mesoamerican Systems

Mesoamerica was the homeland to a distinct family of numerical notations with


two separate subtraditions. Mesoamerican written numerals were first developed in
the lowlands (Yucatan, Belize, Honduras, Guatemala) by 400 bc at the latest, while
a major set of variants developed around ad 1000 in the central Mexican highlands.
These two interrelated traditions are associated most closely with the Maya and
Aztec civilizations, respectively.1 Both these traditions flourished until the Spanish conquests of the sixteenth century. In past research, Mesoamerican numerals
have provided clear New World examples of independent invention of features of
numerical notation systems such as additive notation (Guitel 1958) and the zero
(Kroeber 1948: 468472). Along with calendrical signs, Mesoamerican numerals
were the earliest aspect of the regions representational systems to be deciphered,
and thus are among the best understood, but misinterpretations of the data persist.
The numeral-signs of the major Mesoamerican systems are shown in Table 9.1.

Bar-and-Dot
The bar-and-dot numerals were the most commonly used system in lowland
Mesoamerica, both on stone monuments (400 bc910 ad) and the four surviving
Maya bark-paper codices (10001500 ad). This system has long been an object of
1

I treat other, unrelated New World inventions, such as the Inka khipu and the Cherokee
numerals, in Chapter 10.

284

Mesoamerican Systems

285

Table 9.1. Mesoamerican numerical notation systems


System

Bar-and-dot (monumental)

V e
V E
V
T
U

Bar-and-dot (codices)
Aztec
Texcocan line-and-dot

20

400

8000

2
1
X v Y x y
V

3\9
0\]

study (Bowditch 1910, Morley 1915). While used in all the lowland Mesoamerican
polities, it is most commonly associated with the Maya. Bar-and-dot numerals are
ubiquitous in most lowland Mesoamerican texts, reflecting both the strong interest in dating and calendrics and the practice of incorporating numerical values
into the names of deities and individuals.
The numbers from 1 to 19 are written by combining a dot sign for 1 and a bar sign
for 5 additively. When the bars are vertical, as is most common on stone inscriptions, they are usually placed to the right of the dots, but they are placed below the
dots when the bars are horizontal, as in the codices and a few monumental inscriptions, particularly early ones (Table 9.2). Short numeral-phrases such as these were
normally combined with another glyph indicating the thing being enumerated,
most often time periods. While the primary and original function of the signs was
numerical, some bar-and-dot numerals could also be used syllabically in the Maya
script (though not, as far as we know, in any of the other Mesoamerican scripts). For
instance, four dots could mean 4, but could also indicate near or partial homonyms
of Classic Maya kan four, such as kan sky, height and the first part of kanhan
haughty (Macri and Looper 2003: 262). Mesoamerican hieroglyphic writing on
stone was a very ornate art, and numerals could be altered or ornamented in various
ways that can make reading a numerical value difficult. Ornamental crescents were
often employed in order to fill in a numeral that would otherwise have an empty
space, and these can easily be confused with dots (Thompson 1971: 130). Similarly,
decorative lines were sometimes added to bars for aesthetic purposes, which can
make it difficult to distinguish one from two bars.
In addition, in Maya monumental inscriptions and also (with a different form)
in Maya codices, a sign for 20 was also occasionally used, producing a base-20
cumulative-additive system with a sub-base of 5. In addition to the numerical value
of 20, it is also a glyph meaning moon or lunar month (Lounsbury 1978: 764).2
2

Closs (1978: 691) notes that the central dot in the latter of these signs is found only on
inscriptions where the glyph has the numerical value 20, thus distinguishing it from the
more generic moon, where the dot is missing.

Numerical Notation

286

Table 9.2. Bar-and-dot numerals

18: vertical vs. horizontal orientation

13 vs. 11 (with ornamental crescents)

20/moon
(monumental)

20/moon
(codices)

20/moon (?)
(Isthmian)

3\ 9

0\]

0/completion
(monumental)

0/completion
(codices)

The sign for 20 can occur on its own or in conjunction with bar-and-dot numerals from 1 to 19, thus representing numbers as high as 39. However, it is never
repeated in a numeral-phrase; that is, one would not write 60 with three 20-signs.
The accompanying bar-and-dot numerals could be placed above, below, or to
either side of a 20-glyph (Kelley 1976: 23). The 20-sign was mainly used to indicate
intervals between dates between twenty and thirty-nine days, thus avoiding the use
of combinations of uinals (periods of twenty days) and kins (one day) (Thompson
1971: 139). Very rarely, it was used in expressions for longer time intervals, such
as the irregularly constructed date on Stela 5 at the Maya site of Pixoy, indicating
a quantity of 20 tuns (periods of 360 days) (Closs 1978). In a few instances, to be
discussed, the 20-glyph was used for noncalendrical counts as well.
A sign for zero also accompanied the bar-and-dot numerals. There is considerable paleographic variation in the signs used, but a shell sign was commonly
used in the codices, while different signs were used in monumental writing. The
Mesoamerican zero-sign is not completely synonymous with its Western counterpart; normally it served as a placeholder within dates, with the rough meaning of
completion of a given cycle of time. While this raises the issue of whether we
should regard this sign as meaning zero at all (Thompson 1971: 137), which I will
discuss later in greater detail, I do not see any reason to deny the Maya their zero.
The Maya zero-sign is definitely numerical in function; it is found in the same
contexts as the regular bar-and-dot numerals, but not normally elsewhere, and so
the meaning zero is quite appropriate. While the Maya probably did not have a
concept of zero as a whole number, as is present in Western mathematics, neither
did Seleucid Babylonian astronomers (Chapter 7), for whom the zero-sign served
as the marker of an empty medial position.
While bar-and-dot numeration is most closely associated with the Classic period (ca. ad 150-900), it developed centuries earlier, in the latter part of the Middle

Mesoamerican Systems

287

Formative period (1000400 bc). While Macri and Looper (2003: 4) insist that
writing on perishable materials must have preceded the tradition of carved writing
on monuments and portable objects, there is no iron law of script development
that requires this to be true. Bar-and-dot numerals are among the first identifiable
signs of Mesoamerican writing systems, and occur in all three of the major Formative script traditions: Isthmian (epi-Olmec), Zapotec, and Maya. All three of these
scripts survived into the Classic period, but the Maya inscriptions are by far the
most numerous and significant. Understanding the early history of bar-and-dot
numeration, however, does not give clear priority to any of the three.
The poorly understood Isthmian (or epi-Olmec) script, known from a handful
of inscriptions starting in the Late Formative period (400 bc to 50 ad), is associated with the latter stages of the Olmec civilization along Mexicos Gulf Coast.
Recent claims would give Olmec writing a much longer and complex history,
however. The Cascajal block, which appears to date to the early first millennium
bc and to be associated with the Olmecs, does not contain bar-and-dot numerals
(or any other apparent numerals), but as it is a unique artifact, its relevance to
the history of Mesoamerican numeration and writing remains unclear (Rodriguez
Martinez et al. 2006). There is apparently no connection between the Cascajal
block and Isthmian or indeed any other Mesoamerican script. The San Andrs
cylinder seal, found near the Olmec center of La Venta and dating to the seventh
century bc, is asserted by its discoverers to contain the personal name King Ajaw 3
using stylized dots (Pohl, Pope, and Nagy 2002).
Isthmian writing itself is best known from Late Formative texts from the first
century bc to the second century ad, a period when Olmec political fortunes
were on the wane. In some of these texts, bar-and-dot numerals were arranged
for the first time in vertical columns to indicate periods of time and specific dates
in the famous Long Count system, the calendar expressing dates as a series
of numerals indicating five time periods. We can thus assign absolute dates to
these artifacts from the calendrical evidence alone. The earliest of these is Stela
2 (actually a fragment of a wall panel) from Chiapa de Corzo, dating to 36 bc;
similarly, Stela C from Tres Zapotes has a Long Count date corresponding to 31 bc
(Marcus 1976: 4953).3 The longer second-century Isthmian inscriptions, such as
the La Mojarra stela 1 (156 ad) and the Tuxtla statuette (162 ad), contain bar-anddot numeral-phrases including Long Count dates (Justeson and Kaufman 1993).
The Long Count date 8.5.16.9.9 (corresponding to July 16, 156 ad) occurs on the
La Mojarra stela, with standard bar-and-dot numerals arranged vertically in a single column. The stela also contains a sign identified by its decipherers as meaning
3

As with all such dates, there is a small possibility that they were inscribed at a date later
than the one given textually.

288

Numerical Notation

20 or moon, parallel to the later Maya practice (Justeson and Kaufman 1993).
However, aside from the numerals, much about the decipherment of Isthmian is
hotly contested, with Justeson and Kaufman (1993) arguing that their decipherment is well under way, but Houston (2004b: 297298) arguing that because of
the contextualized nature of the system, the script may be indecipherable.
The Zapotec civilization of the Valley of Oaxaca in southern Mexico began to
rise to ascendance in the late Middle Formative period, and developed a script
tradition quite distinct from Isthmian. The earliest well-attested Zapotec numeral
is found upon Monument 3 from San Jos Mogote (600500 bc) in the Valley of
Oaxaca, where the day-name 1 Earthquake is written with a stylized dot (Marcus
1976: 4445). If the Isthmian San Andrs cylinder seal is misdated or non-numerical,
then this inscription is the earliest attested instance of bar-and-dot numeration.
Stela 12 from Monte Albn provides the first example of a combined bar-and-dot
phrase for 8, apparently indicating a day of the Zapotec month (Marcus 1976:
4546). Both of these are Middle Formative and definitively Zapotec rather than
Olmec, and their early date points to Oaxaca as a potential region of origination
for the system. Colville (1985: 796), however, is agnostic as to whether the barand-dot system was invented by the Olmec or the Zapotec, since in his analysis,
both used vigesimal lexical numerals with a quinary component, a structure common also to the bar-and-dot numerals.4
While it was once widely held that the Maya tradition was a latecomer in the
history of Mesoamerican writing, there is some evidence that the Maya tradition
may have emerged alongside the Isthmian and Zapotec scripts, perhaps as early
as 400 bce. Monument 1 from El Portn is an extremely eroded stela with one
partially readable column of glyphs that may be ancestral to later Maya glyphs,
including numerals. Although the inscription has no date, archaeological evidence
places it in the late Middle Formative (Harris and Stearns 1997: 122). This raises the
possibility that the three Mesoamerican traditions were essentially contemporaneous, and may have emerged in a context of economic and diplomatic exchange.
Yet there is a distinct scarcity of Maya or Maya-ancestral texts from the Middle
and Late Formative. Starting in the second and third centuries ad, the Isthmian
and Zapotec traditions began to wane, and Maya inscriptions predominate in the
Mesoamerican lowlands thereafter. The first certain Maya inscription that uses
bar-and-dot numerals is Stela 29 from Tikal, which dates to 292 ad (Lounsbury
1978: 809); however, Stela 5 from Abaj Takalik, which dates to 126 ad, may also be
an early Maya inscription (Closs 1986: 327).
4

Colville accepts the idea that the Olmecs spoke a Mixe-Zoquean language, whose modern speakers have numerals of this structure; this is not a universally accepted conclusion,
and in any case the numerals almost certainly changed in the intervening millennia!

Mesoamerican Systems

289

This sequence of inscriptions demonstrates conclusively that the bar-and-dot


tradition was an independent Mesoamerican invention with a complex history
that remains only partially understood. While Seidenberg (1986) and others have
attempted to postulate diffusion from Babylonia as the source of Mesoamerican
mathematics and numeration, this is highly improbable on the basis of the evidence just discussed. Aside from the use of place value (itself a contested issue,
as I will show), there is no similarity between Babylonian and Maya systems;
they use different bases, different directions of writing, different media, and serve
extraordinarily different sets of functions. The early origin of bar-and-dot numeration alongside the Middle Formative Mesoamerican scripts, the quinary-vigesimal
structure of the system, and the general increase in the frequency and complexity
of numeral expressions over time all point to its indigenous development.
There are many thousands of identified Maya inscriptions from the Classic
period (150900 ad), the vast majority of which contain at least some bar-and-dot
numerals. There is no identifiable regional variation in the form or ornamentation
of the numerals within the Maya sphere of influence. Very early in the Classic
period, the bar-and-dot numerals spread through highland Mexico; the Mixtecs
used bar-and-dot notation to write numbers up to 13 (Caso 1965: 955). It is not
clear whether they borrowed them from the Zapotecs (who also lived in the Oaxaca region, and continued to use bar-and-dot numeration throughout the Classic
period) or from the more influential Maya. Bar-and-dot numerals were used occasionally at the highland city-state of Teotihuacn; on about a dozen inscriptions,
numbers smaller than 13 were written with bars and dots (Langley 1986: 139142).
There is some evidence that the bar-and-dot numerals survived in central Mexico
until the Spanish conquest. In Mixtecan-Pueblan texts such as Codex FejervaryMayer and Codex Cospi, sets of bars and dots arranged vertically or horizontally
could represent counted bundles of offerings (Love 1994: 61). Although it is an
irregular formation by the normal rules of the system, a set of four bars from the
Mixtec Codex Selden may represent a quantity of twenty bundles (Boone 2000:
43). However, the Aztecs never used bar-and-dot numerals, instead relying on
their own additive base-20 numerals. These developments will be discussed in
greater detail later in this chapter.
After the collapse of classic Maya civilization in the tenth century ad, attested
examples of bar-and-dot numerals become increasingly rare. The latest Maya
monumental inscription dates to 909 ad (Closs 1986: 317). Many regions where
bar-and-dot numerals had previously been used, such as Oaxaca and the Valley of
Mexico, abandoned the old system in favor of the central Mexican dot-numerals
(see the following discussion). Bar-and-dot numerals were retained during the
Postclassic period (tenth to sixteenth centuries) in Guatemala and Yucatan, where
they were used on bark-paper codices until the Spanish conquest (Urcid Serrano

Numerical Notation

290

Table 9.3. Maya period glyphs

kin
1 day

uinal
20 kins

tun
18 uinals

katun
20 tuns

baktun
20 katuns

2001: 3). The last text on which bar-and-dot numerals occur is one of the books
of Chilam Balam, in which an annotated description of the system is dated 1793
(Thompson 1971: 130). Yet the system essentially had ceased to be used by 1600
and was replaced by Roman or Western numerals.

Maya Calendrics and Positional


Bar-and-Dot Numeration
In all three of the Formative period Mesoamerican script traditions, numerals were
written with dots placed above horizontal bars and not directly linked to any other
glyph (although we can tell that the quantities enumerated were periods of time).
This technique would later be the standard practice in the Maya codices. However,
in the Classic period, most Maya monumental numerals were written with vertical
bars with dots to their left, and were linked to glyphs for various periods of time.
A bar-and-dot numeral-phrase from 0 to 19 would be combined with one of five
glyphs for time periods kin (one day), uinal (one month of twenty kins), tun (one
year of eighteen uinals), katun (twenty tuns), and baktun (twenty katuns) by placing the numeral to the left of the period-glyph.5 Each successive period is twenty
times the previous one, except for the tun of eighteen uinals, which comprises a sum
of 360 days, corresponding roughly to the solar year.6 Some of the more commonly
used period glyphs are shown in Table 9.3 (cf. Closs 1986: 304305).
To express a specific fixed Long Count date, five numeral-glyph combinations were required (one for each period, written from longest to shortest).
These were normally written in pairs of columns, most commonly but not
5

The terms katun and baktun mean, literally, 20 tuns and 400 tuns. The latter term is
in fact a coinage of Mayanists; there is no evidence that this word was associated with
the glyph in question in ancient times. There are several extremely rare glyphs for longer
periods, again with coined names: pictun (8000 tuns), calabtun (160,000 tuns), and kinchiltun (3,200,000 tuns), which presume a purely vigesimal progression of dates (Closs
1986: 303).
It appears, however, that the Yucatecan and Cakchiquel Maya may have had a purely
vigesimal year of twenty months of twenty days, though their numerical notation does
not reflect this fact (Satterthwaite 1947: 89).

Mesoamerican Systems

291

Table 9.4. Initial Series date (9.14.10.0.2) with period glyphs


(9 baktuns)
(10 tuns)
(12 kins)

i%
j#
l!

n$
9@

(14 katuns)
(0 uinals)

exclusively read from left to right and top to bottom. These expressed the
amount of time between the starting point of the Maya calendar (corresponding to the date August 10, 3113 bc, in the widely accepted Goodman-MartinezThompson correlation with the Gregorian calendar) and any other date. In
addition, the amount of time between any two days could be expressed by a
Distance Number, such as 12 tuns, 0 uinals, 4 kins. Modern scholars use a
convention whereby time values are expressed by writing the five numerical
coefficients separated by points; thus, the date shown in Table 9.4 would be
written as 9.14.10.0.12.
For both Long Count dates and Distance Numbers, if the coefficient of a time
period was zero (e.g., 0 uinals in Table 9.4), the Maya included both a zero coefficient and a period glyph for that value, even though it was not logically necessary
to do so in order to interpret the phrase correctly. While it is not known exactly
why the Maya did this, it was probably for aesthetic reasons. Only occasionally
in Distance Numbers (though never in Long Count dates), a period with a coefficient of zero was suppressed (Thompson 1971: 139).
In a few texts, period-glyphs were omitted entirely, and dates were written
simply by placing the five coefficients in a single vertical column. As mentioned
already, the technique was present in the Isthmian and Zapotec inscriptions by
the first century bc, and continued to be used by the Preclassic Maya (Marcus
1976: 4957). Although it was largely abandoned thereafter, Stela 1 at Pestac
contains a date (9.11.12.9.0) written in this format, which corresponds to 665 ad
(Closs 1986: 326327). Most other Maya inscriptions include all the periodglyphs, although sometimes the glyph for the last position (kins) was omitted
(Closs 1986: 308). Our best evidence for the omission of period-glyphs comes,
however, from the Dresden Codex, a Postclassic text that was probably written in the early thirteenth century, though it may be a copy of a much earlier
document (Marcus 1976: 35).7 It is the most astronomically sophisticated of the
surviving Maya texts, and contains more vertical columns of numbers than any
7

One set of five numbers without period glyphs is found on the fifteenth-century Madrid
Codex that may qualify, but otherwise no other codices have them.

Numerical Notation

292

Table 9.5. Initial Series date without period glyphs

I
N
J
]
L

9 baktuns
14 katuns
10 tuns
0 uinals
12 kins

other. Table 9.5 shows the Long Count date 9.14.10.0.12 as it would be written
in this manner.
This system requires that all the relevant numerical coefficients be included,
even for periods for which there is a zero coefficient, to ensure that the correct
quantity of time is counted. The bottom value always represents kins, the second
from the bottom uinals, and so on, preventing any misreadings. Because these
units of time are arranged in a mainly vigesimal sequence each higher value
is equal to twenty of the next lower value, except the tun of eighteen uinals
Mayanists today agree that this system of writing dates is a base-20 cumulativepositional numerical notation system with a sub-base of 5 (Kelley 1976, Marcus
1976, Lounsbury 1978).
If so, then when the Maya wrote number columns such as the one in Table 9.5,
each position must have represented a particular component of a single number. Positional numerical notation systems do this by having each successive position represent the next higher power of a base. Thus, when I write the number 1942, I mean a
single count of some quantity, of which there are 1942, consisting of one thousand,
nine hundreds, four tens, and two ones. In the Maya case, where the lowest unit
expressed is kins, it is quite natural to assume that it counts kins. It is easy to translate
the five periods into counts of days and then to take the sum, as in Table 9.6.
However, if the period-glyphs were meant to be inferred when reading these
columns, then such numerals ought to be read as five separate values, each no
greater than 19, just as they would be if the glyphs were included. How, then, can
we tell whether the interpretation in Table 9.6 is one that the Maya themselves
made, or whether they simply read in the missing period-glyphs? How do we
decide whether the correct interpretation is 1,400,412 kins or 9 baktuns, 14 katuns,
10 tuns, 0 uinals, 12 kins?
If bar-and-dot numerals were used for large quantities of things other than
time, we would have clear instances where the higher positions represent powers

Mesoamerican Systems

293

Table 9.6. Positional Maya count of days

I
N
J
]
L

9 144,000 days

1,296,000 days

14 7200 days

100,800 days

10 360 days

3600 days

0 20 days

0 days

12 1 day

12 days
= 1,400,412 days

of a base, rather than large calendrical periods. Yet no Mesoamerican texts use
positional bar-and-dot numerals to count noncalendrical amounts. One is
struck, upon comparing Maya inscriptions to those of any other civilization, by
the virtual absence of phrases indicating large quantities of captives taken in war,
goods paid in tribute, wealth owned by individuals, or any other noncalendrical
quantity. In the rare instances where the Maya wrote numbers of other quantities above 19, sometimes they used additive techniques, such as the moon-glyph
for 20, which is used in counts of 20 and 21 captives, but this does not allow the
writing of very large numbers (S. Houston, personal communication). In other
cases, it is possible that multiplicative techniques were used. Houston (1997) suggests that on a mural from Bonampak, a bar numeral for 5 was combined with a
glyph, pi, which may have meant unit of 8000 cacao beans, producing a quasinumerical expression for a count of 40,000 cacao beans (Houston 1997). The
Yucatecan, Cholan, and Tzeltalan languages all use numeral classifiers linguistic
particles that obligatorily follow lexical numerals and indicate the thing being
counted (Berlin 1968; Bricker 1992: 7173). Macri (2000) contends on this basis
that the period-glyphs kin, uinal, tun, katun, and baktun, as well as any other
glyphs that follow numbers, should best be interpreted as numeral classifiers.8 If
this interpretation is correct, then by analogy with the calendrical system, the
Maya likely expressed large noncalendrical quantities by combining bar-and-dot
numerals with a sign for a metrological unit that may have been a multiple of
some smaller unit but was not expressed in terms of that unit (just as 1 tun = 360
kins but was not expressed in such terms). These different means of writing larger
8

For Macri, this also explains why the early epi-Olmec and Zapotec calendrical inscriptions written in Mixe-Zoquean languages do not use period glyphs but simply series
of bar-and-dot numerals.

294

Numerical Notation

numbers, and the lack of noncalendrical positional vertical number columns,


cast doubt on the entire existence of Maya positional numerals.
In Chapter 8, I discussed the transformation of the Chinese multiplicativeadditive system into a ciphered-positional one by adding a zero-sign and deleting the power-signs for 10, 100, 1000, and so on, so that (7
1000 + (0) + 4 10 + 9) becomes 2(7049). There are similarities between
this transformation and the removal of the Maya period-glyphs, but in the Chinese case, the removed power-signs are numerical (representing the powers of 10),
whereas in the Maya case the period-glyphs are calendrical. The assumption that
the fourth position of the Maya numerals means 7200 is wrong. Even so, it
could still have been read as 7200 days. This is undemonstrated, however, and I
do not consider it likely. When the period-glyphs are present, as they are in most
of the inscriptions on stone, Mayanists do not consider the calendrical system to
be a positional one, and do not treat dates as a sum of days. Why, then, should the
removal of these period-glyphs be anything more than an abbreviatory convenience? One year is equal to 365 (or 366) days, but this does not mean that if I write
the date 2005/06/14 I really mean a sum of days equal to 2005 years, 6 months,
and 14 days, and certainly I do not calculate such a sum in my head.
A neglected tradition in the study of Maya calendrics and numeration recognizes that Long Count dates (with or without period-glyphs) are capable of
being read positionally or nonpositionally. Teeple (1931) and Thompson (1971)
claimed that such dates should be considered as a count of tuns (years), in which
the final two places (uinals and kins) represented two separate fractions of years.
Satterthwaite (1947) held that they should be read as two separate counts, one of
years (the first three positions), the other of days (the last two). Closs (1977), the
most recent scholar to deal seriously with this issue, holds that there were in fact
three counts: a tun count, comprising, first, a positional numeral indicating 1, 20,
and 400 tuns; second, a nonpositional bar-and-dot numeral indicating uinals;
and third, a nonpositional bar-and-dot numeral indicating kins. All agree that the
highest three periods (baktuns, katuns, and tuns) were read and understood by the
Maya as a single count of tuns. Moreover, they claim that the Long Count dates
were understood in this way, whether or not the period-glyphs were present.
These readings are made on the basis of several lines of evidence. Separating
the higher values, which are purely vigesimal, from the lower ones, in which the
18 uinals = 1 tun irregularity occurs, renders the system more readable, given the
purely vigesimal structure of the Maya lexical numerals. It also helps explain a
number of texts where the glyphs for the tun and its multiples are distinguished (by
color or ornamentation) from the other two (Closs 1977: 2223). In the irregular
Tun-Ahau statement from Xcalumkin, a Long Count date is expressed simply as
9 baktuns, 16 katuns, 2 tuns without uinal or kin values, further suggesting that

Mesoamerican Systems

295

tuns (and multiples thereof ) occupy a special role in the Maya calendrical system
(Closs 1983). Finally, the glyphs for katun and baktun often show an affiliation
to the basic tun sign. Yet there is no reason to think that the Maya wrote glyphs
for the baktun and katun but then simply ignored them in reading. As an analogy,
the English words decade, century, and millennium etymologically refer to
tens, hundreds, and thousands of years, but 9 millennia, 4 centuries, 3 decades,
6 years is read and understood differently from 9436 years even though both
phrases refer to the same time value. I agree fully with Closs that the kin, uinal,
and tun counts were read separately, but believe that he has not gone far enough,
and regard the Maya Long Count dates as five separate nonpositional counts of
five different time periods.
The assumption that the numerals in the Dresden Codex must have been positional is linked with the belief that positional notation was highly useful for doing
calendrical calculations. Since the Maya did do these calculations, and since these
numbers look like positional notation, it is natural to infer that they were, even
though Classic Maya dates were normally written as five different periods rather
than as a single sum of days. When Mayanists interpret Mayan chronology, they
must translate Maya dates into a single number of days in order to the correlate
Maya and Western calendars (e.g., the Goodman-Martinez-Thompson correlation
establishes the beginning of the Maya calendar as Julian day number 584,283). Yet,
however the Maya may have read these columns of numbers, there is no evidence
that they ever calculated with them. The Dresden Codex is a repository of calendrical
data, including what appear to be multiplication tables, but there are no calculations
on paper. There is specific ethnohistorical evidence concerning Maya computation,
from Landas Relacin de las cosas de Yucatan, which suggests that the sixteenthcentury Maya did not calculate directly using bar-and-dot numerals:
Their count is by fives up to twenty, and by twenties up to one hundred and by
hundreds up to four hundred, and by four hundreds up to eight thousand; and
they used this method of counting very often in the cacao trading. They have other
very long counts and they extend them in infinitum, counting the number 8000
twenty times, which makes 160,000; then again this 160,000 by twenty, and so on
multiplying by 20, until they reach a number which cannot be counted. They make
their counts on the ground or on something smooth. (Tozzer 1941: 98)

Computation was done on some sort of flat surface, suggesting that some
sort of physical counting board was employed. Some Mayanists have turned
their attention to what sort of physical counters the Maya might have used and
whether the bars and dots used as Maya numerals had physical correlates in rods
and beans, or some other such markers (Thompson 1941: 4243; Tozzer 1941: 98;

296

Numerical Notation

Satterthwaite 1947: 3031; Fulton 1979: 171). Sol Tax, working among the Maya of the
Guatemala highlands at Panajachel in the 1930s, found that they computed using
beans or stones in groups of five and twenty, supporting the idea that the ancient
Maya may have done similarly (Thompson 1941: 42). Counting boards are often
positional in structure, and some use special counters or markers for empty
positions signs that resemble a zero. On this basis, some suggest that numerals
were written positionally in a purely vigesimal fashion for noncalendrical purposes
that is, with the third and fourth positions having the values of 400 and 8000
in emulation of the mode of computation (Marcus 1976: 39; Lounsbury 1978:
764). Yet the host of speculations on the use of bar-and-dot numerals directly in
calculation, without an intermediary computational device, is useless (Sanchez
1961, Bidwell 1967, Anderson 1971, Lambert et al. 1980, Mhlisch 1985). While,
as Anderson (1971: 63) states, it is not unreasonable to suggest that some attempt
to use the numerals directly in computations might have occurred, this pastime
tells us much more about the ingenuity of modern scholars than it does about
the actual practices of Maya mathematics. Just as the Romans and Greeks had a
place-value abacus but no positional numerical notation system, the presence of a
Maya abacus-like device does not presuppose that they had positional numerals.
The columns of an abacus work just as well if they indicate distinct units of baktuns, katuns, tuns, uinals, and kins as they do if they represent the power-values
144,000, 7200, 360, 20, and 1. The manipulation of counters is identical, but the
reading of the results is very different.
Unfortunately, the great bulk of Maya codices is now lost to us forever due
to the tragic destruction of manuscripts on Spanish orders in the early colonial
period. It is far too easy to create hypotheses concerning lost positional inscriptions when huge quantities of evidence have literally gone up in smoke. Yet the
surviving evidence does not support the hypothesis that the number columns in the
Dresden Codex should be interpreted as sums of days, and thus as a cumulativepositional numerical notation system. The most parsimonious explanation is that
the omission of period-glyphs was abbreviatory but did not entail a radical rereading of the numerical coefficients.
In his analysis of Maya arithmetic, Fulton noted that it is possible to have a
strictly positional notation, not altogether different from our present one, without any zero whatsoever (1979: 171). Positionality requires some way of avoiding
ambiguity between 749 and 7049, but this may be simply an empty space. Inverting this insight, I believe that the Maya bar-and-dot system had a zero, but did
not use the positional principle. This is not to say that the Maya zero or completion-sign was nonfunctional. While it was retained for aesthetic purposes in places
where it was not strictly needed (when period-glyphs were present), the zero
was needed whenever the period-glyphs were omitted and there was an empty

Mesoamerican Systems

297

period. But the purpose of a Maya zero in a number such as 1.0.4 does not appear
to be to indicate that the first number should be multiplied by 360, but rather
simply to indicate that the middle position is empty, and thus the 1 should be read
as 1 tun rather than 1 uinal. While something like positionality is used to distinguish
different units of time, there was no Maya positional numerical notation system.
Once we abandon the notion that the presence of place-value is an eternal standard
of utility in numeration, we can see that the bar-and-dot system was highly useful for recording dates even though it was not, strictly speaking, positional. The
main bar-and-dot system is cumulative-additive, and when cumulative-additive
numeral-phrases were combined to express large time periods, what results is a quasipositional calendrical notation, but not a true positional numerical notation.

Maya Head-Variant Numerals


In place of the bar-and-dot numerals, the Maya occasionally used a set of complex
glyphs for the numbers 0 through 19, many of which correspond to the heads of
Maya deities.9 These head-variant glyphs are far more variable in form than are the
very regular bar-and-dot numerals. Each head-variant replaced the corresponding
bar-and-dot numeral-phrase in an expression for a Maya date. An example of each
of the signs is shown in Table 9.7 (redrawn from Thompson 1971: Figure 2425).
Because the highest number expressed using head-variant numerals is 19, there
is, strictly speaking, no base to this system. However, because they replace barand-dot numerals, head-variant glyphs are associated with the five calendrical
coefficients (baktun, katun, tun, uinal, kin), and thus assume elements of a vigesimal structure. The head-variant numerals from 1 through 12 are written with elementary signs. The signs for 14 through 19 are additive combinations of a bared
jawbone element that represents 10 and the upper head of the sign for the appropriate unit. There are two signs for 13; the more common one (13a) is an additive
combination of the bare jawbone for 10 and the head-glyph for 3, while the other
(13b) is a distinct glyph for some sort of monster, and possibly holds some lunar
significance as well (Macri 1985: 74). Because individual signs are not repeated
to signify their addition, the head-variant numerals have more in common with
ciphered than they do with cumulative numerical notation systems, but since they
never exceed 19, they cannot be said to be either additive or positional.
Although the head-variants for 1 through 12 are elementary signs, the system
does not have a base of 12, as stated by Kuttner (1986). The relevant subunit of
the head-variant numerals is not 12 but 10, since the signs for 14 through 19 (and
9

For an analysis of the specific deities and other symbolism associated with each glyph,
see Thompson (1971): 131137; Macri (1985); Stross (1985).

298

Numerical Notation

Table 9.7. Maya head-variant glyphs

sometimes also 13) are expressed additively using 10 (Macri 1985: 75). No other
Mesoamerican numerical notation system uses a decimal sub-base. The origin of
this feature probably lies with the lexical numerals of the Mayan language family, which uses decimal structuring to express the numerals from 13 to 19, but has
rather opaque formations for 11 and 12 (in Classic Maya, buluc and lahca), just as
the English eleven and twelve do not show any clear relation to ten (Lounsbury
1978: 762). Additionally, Macri (1985: 48) suggests that it may have been important to have thirteen simple signs to correspond to the thirteen deities used to
name days in the Maya sacred calendar.
The head-variant numerals are relatively common on Maya inscriptions, though
less common than the bar-and-dot numerals. They also appear occasionally in the
Dresden Codex, though not in the other Postclassic codices (Thompson 1971:
131). Macri (1985: 55) hypothesizes that they may have had a Preclassic origin,
but no pre-Maya inscription uses them. Macri (1985: 48), pointing to phonetic
correspondences between the head-variant signs and the Eastern Maya lexical

Mesoamerican Systems

299

numerals, suggests an Eastern Maya origin for the system, but Stross (1985) points
out that many of the same correspondences exist in the Mixe-Zoquean language
family, to which the Olmec language may have belonged. Yet none of the Isthmian inscriptions contains head-variant numerals, and many centuries lie between
the decline of the Olmec civilization and the appearance of head-variant glyphs.
The head variants are extremely different graphically and structurally from the
bar-and-dot numerals, and cannot have emerged directly from the bar-and-dot
tradition. We had best think of them as a complex set of metaphors by which the
numerical symbolism of deities was used as a code for numerical information, not
as a numerical notation system in their own right.
Given the destruction of so many Maya codices, as well as the imperfect state
of Maya archaeology and hieroglyphic decipherment, it is difficult to say when
the head-variant numerals ceased to be used. Since the Dresden Codex is the only
surviving Postclassic codex to use them, and then only occasionally, it is possible
that they declined in use during the Postclassic period.

Mexican Dot-Numerals
During the Postclassic period (tenth to sixteenth centuries), many of the peoples
of central Mexico began using a system of dots to represent small integers in their
pictographic manuscript tradition. Since this means of representation lacks a base
and relies only on one-to-one correspondence, strictly speaking it does not constitute a numerical notation system, but it deserves some mention here. In their
early history, the Mixtec and Teotihuacni used bar-and-dot numerals, borrowed
from the Maya or the Zapotecs, but after the tenth century ad, the system fell
into disuse (Caso 1965: 955; Langley 1986: 143). While bar-and-dot numerals were
occasionally used in a few later Mixtec codices, apparently for archaic or sacred
reasons, they were largely replaced by a system whereby dots alone were used for
the numbers 1 through 19, representing day-numbers and other objects (Colville
1985: 839841). The peoples of Oaxaca, the Valley of Mexico, and the Gulf Coast
used this system until the time of the Spanish conquest. A numeral-phrase was
composed of a series of dots in a single row. To facilitate reading and to save space,
larger numbers were often grouped in segments of three to five units, sometimes
connected by lines, and sometimes changing direction (e.g., horizontal to vertical) in the middle of a numeral-phrase. Numbers above 20 were never expressed
in this system.
Given that the central Mexican calendar is part of a Mesoamerican calendrical
tradition, and given the common use of dots for units in both the Maya and dotonly systems, I think it is plausible that between the tenth and twelfth centuries
ad, the use of bars for 5 was gradually abandoned, although the reason behind

300

Numerical Notation

this change is not clear. The influence of Toltec culture, which was becoming
predominant in Mesoamerica at this time, has been cited as the cause of this shift
(Caso 1965: 955). Yet this argument begs the question of why the Toltecs did not
adopt bar-and-dot numerals. Dot-only numerals are not known from anywhere
in Mesoamerica prior to the tenth century ad, so it is unlikely that there was such
a tradition prior to that point. Thus, unless the use of dots for units developed
independently in the two different parts of Mesoamerica, the dot-numerals must
be descended from the bar-and-dot system.
The dot-numerals were ancestral to the later Aztec numerals, a base-20 cumulative-additive system. Because the Aztecs, like the Maya and Mixtecs, used dots for
units, but because, unlike the bar-and-dot numerals, the Aztec system has no quinary component, the dot-numerals are a likely intermediary between the lowland
and highland Mesoamerican systems. Both the dot-numerals and the Aztec numerals use up to nineteen dots for units, the difference being that with the Aztec
numerals, the dots were more regularly grouped in fives, and higher numbers were
written using different signs for the powers of 20. It is generally believed that the
Aztecs inherited their tradition of manuscript writing from the Mixtecs (Colville
1985: 839). Dot-numerals continued to be used in Aztec manuscripts even after
the development of the cumulative-additive numerals in the fourteenth century.
By the time of the Spanish conquest, the Aztec numerals had supplanted the dotnumerals in some areas outside their tributary area, and were used in many of the
post-Conquest Mixtec codices (Terraciano 2001).

Aztec
The name Aztec applies most precisely to the Nahuatl-speaking inhabitants of
the region immediately surrounding the ancient city of Tenochtitlan (modern
Mexico City), who controlled a substantial tributary system in central Mexico between the fourteenth and sixteenth centuries. More generally, the term often refers
to the various Uto-Aztecanspeaking peoples of central Mexico who were under
Nahua rule during this period. The Aztec tributary network, which embraced
numerous small states, produced a large number of manuscripts, using a combination of ideographic and phonographic signs. The considerable debate concerning
whether this Aztec manuscript tradition constituted true writing or simply served
as a mnemonic aid is irrelevant to the study of Aztec numeration. The Aztecs most
definitely possessed a vigesimal numerical notation system, whose signs are shown
in Table 9.8.
The sign for 1 is the dot that was commonly used for units throughout Mesoamerica. The signs for the vigesimal powers are depictions of objects: for 20, a flag
(pantli); for 400, a feather (tzontli, literally hairs); and for 8000, a bag used to

Mesoamerican Systems

301

Table 9.8. Aztec numerals


1

20

400

8000

X v

Y x :

yyy

xxxxx vvvvv
xxx
vvvvv
vvv

ttttt
tttt

(3 8000) +

(8 400) +

(9 1)

27,469 =

(13 20) +

hold copal incense (xiquipilli) (Harvey 1982: 190). These signs were combined in
a cumulative-additive fashion, written in horizontal rows with the highest powers
on the left. Although the Aztec numerals, unlike the Maya bar-and-dot numerals,
did not use a sign for 5, groups of more than five identical signs were arranged in
sets of five for easier reading. Groups of five signs were sometimes joined to one
another with a horizontal line underneath the set.
The purely vigesimal structure of the Aztec numerical notation system and the
shapes of its numeral-signs are quite different from those of the lowland Mesoamerican bar-and-dot system. Instead, the Mexican dot numerals are the most
likely ancestor of the Aztec system. It is plausible that the Aztecs originally used
dots alone, but then, as the administrative needs of their tributary system grew,
invented new numeral-signs for 20 and its powers. As far as can be discerned, the
inventors and early users of the Aztec system were not influenced directly by the
lowland Mesoamerican systems of the Maya. The Mexican dot numerals do not
constitute a numerical notation system according to my definition, because they
lack a base, meaning that the Aztec system was invented relatively independently.
The most important function of the Aztec numerals was to record the results of
economic transactions, such as amounts of cacao beans, grain, clothing, and other
goods received from different regions of their tributary system (Payne and Closs
1986: 226230). Numerals were also used in Aztec annals and historical documents, such as the record of the massacre of 20,000 prisoners in the Codex Telleriano-Remensis (Boone 2000: 43). Sometimes, when recording amounts of goods,
individual numeral-signs were attached to an equal number of pictographic signs
for goods. Accordingly, one might record 1200 balls of incense not as the numeral
1200 followed by a picture of an incense ball, but rather using three balls of incense, each of which would be placed immediately underneath a sign for 400.
The use of Aztec numerals to record large quantities of tribute and individuals
stands in sharp contrast to the Maya bar-and-dot numerals, which were almost

302

Numerical Notation

wholly calendrical in function. The Aztecs denoted their thirteen months using
series of dots in rows, just as the Mixtecs did, but when they did so, they did not
group dots regularly in groups of five, and thus this represents a continuation of the
dot-numerals in Aztec manuscripts (Boone 2000: 4344). Normally, the Aztecs did
not record dates or other calendrical information using the larger numeral-signs. In
a single text, the Vatican Codex, large periods of time seem to have been expressed
using cumulative-additive combinations of different signs, the largest of which represents 5206 years with thirteen signs that probably represent 400 (the third sign in
Table 9.8), above which six dots were written (Payne and Closs 1986: 234235).
After the Spanish conquest, the Aztec numerical notation system continued
to be used in various colonial documents. In fact, its use spread well beyond the
Valley of Mexico, as Nahuatl increasingly became a lingua franca used by indigenous highland Mesoamericans. For instance, Aztec numerals are common in the
Mixtec Codex Sierra, a mid-sixteenth-century account book that uses Western,
Roman, and Aztec numerals side by side (Terraciano 2001: 4045).10 In a few postconquest manuscripts, fractions could be depicted by segments of 1/4, 1/2, and 3/4
of a dot, low multiples of five by filling in quarters of the pantli flag sign, and 100,
200, and 300 using segments of the tzontli sign for 400 (Vaillant 1950: 202).
A few post-conquest Aztec codices use multiplicative rather than strictly additive numerical notation. Guitel (1958; 1975: 177) was the first to point out that
one of the often-reprinted examples of Aztec numbers depicts a basket of cacao
beans from which four signs for 400 emerge, above which a pantli or flag for 20 is
placed. This numeral-phrase represents a total amount of 32,000 cacao beans multiplicatively, as 20 baskets of 1600 beans each, rather than additively, as 4 xiquipilli
of 8000. In a circumstance where cacao beans come in baskets of 1600 beans, however, it is important to denote that there are 20 baskets of 1600 each, not simply
32,000 beans. This does not certify that placing the numeral-phrases for 20 and
1600 together means 32,000. However, Guitel was not aware of another text, a
Texcocan document now known as the Codex Kingsborough, where multiplicative notation was used extensively (Paso y Troncoso 1912, Harvey 1982). I will treat
this structurally distinct variant of the standard Aztec system later.
As disease, warfare, and acculturation diminished the strength of Aztec traditions, the old numerals ceased to be used. I do not know of any documents from
later than 1600 that use Aztec numerals. After this point, Roman and especially
Western numerals were employed throughout highland Mexico.
10

Boone (2000: 254) indicates that Oaxacan texts do not contain signs for 400 or 8000; at
least in the case of the tzontli sign for 400, she is incorrect, as this is found in the Codex
Sierra (cf. Terraciano 2001: Figure 2.16). Yet the year-date, written as 1563 in Western
numerals, is not transliterated in Aztec numerals but rather in Mixtec lexical numerals.

Mesoamerican Systems

303

Texcocan Line-and-Dot
The city of Texcoco in the province of Tepetlaoztoc was one of the most powerful
cities in the Valley of Mexico both before and after the Spanish conquest. While
many sixteenth-century colonial documents continued to use the Aztec numerals
just described, a handful of Texcocan documents contain a quite different system,
which I will call the Texcocan line-and-dot system. These documents have been
studied extensively by Herbert Harvey and Barbara Williams and are identified
collectively as the Tepetlaoztoc Group (Harvey and Williams 1980, 1981, 1986;
Harvey 1982; Williams and Harvey 1988, 1997). The numeral-signs of this system
are shown in Table 9.9 (Harvey and Williams 1980: 500).
This system is cumulative-additive, with a base of 20 and a sub-base of 5. The
sign for 5 consists of five unit-strokes joined together by a curved line, so it is
perhaps just a matter of personal preference whether we see it as a separate numeral-sign. A similar technique was occasionally used to group sets of five dots
for 20 into a single unit of 100. Perhaps the most unusual feature of this system is
that, whereas other Mesoamerican numerical notation systems used a dot for the
units, here a vertical stroke denoted the units, while the dot took a value of 20.
Numeral-phrases were written in a variety of directions, but were always arranged
in a single line from highest to lowest sign (Harvey 1982: 191).
This form of notation has been found in only three texts, all of which were written
in the vicinity of Texcoco in the 1540s. Two of these, the Cdice Vergara and the
Cdice de Santa Mara Asuncin, were cadastral records written around 1545 to
enumerate individuals and their land holdings. These two are in fact so similar
that they may have been parts of the same manuscript at one point, or at least were
drawn at the same time (Williams and Harvey 1997: 2). The third, the Oztoticpac
Lands Map, was written around 1540, and is also a record of lands, though as a
map rather than a census record (Cline 1966). The primary function of the numerals in all of these cases was to record land measures. Pictographic signs expressing
fractional linear units sometimes accompanied the numerals, but their meanings
are still unclear (Williams and Harvey 1997: 26). Several numeral-phrases on the
Oztoticpac map were used to count sums of days, showing that this system was
not restricted to one domain.
The Cdice de Santa Mara Asuncin used a modified form of this system
to express numbers positionally rather than additively. In studying this text,
Harvey and Williams (1980) showed that line-and-dot numerals occurred in two
different sections, but served very different functions. In one section, known
as milcocoli, line-and-dot numerals were used in the regular manner, written
along the edges of maps of plots of land owned by different individuals to indicate their lengths. In another section, known as tlahuelmatli, the plots of land

Numerical Notation

304

Table 9.9. Texcocan line-and-dot numerals


1

20

100

from the milcocoli section were redrawn as rectangles (regardless of their original shape). This section also contained numerals indicating the areal measurement of each individuals land holdings. Comparing the milcocoli values, which
indicated the lengths of the sides of plots, and the tlahuelmatli values, which
recorded their total area, Harvey and Williams showed that a form of positional
notation was used to record land areas in the tlahuelmatli section using a set of
three distinct registers within a rectangular depiction of a plot of land (Harvey
and Williams 1986: 242). In the top right corner, dot-and-line numerals indicated values from 1 to 19 in a small protuberance. On the bottom line of the
rectangle, units and groups of five indicated multiples of 20 units. No dots were
ever used in either of these two registers. When dots were found, they occurred
with or without units in the center of the rectangle. Strangely, this third register
also counted multiples of 20 (i.e., lines equal 20 and dots equal 400). No plots
of land show values both on the bottom line and in the center. When the twenties register and the units register were added together, a total area value was
reached. Harvey and Williams found that in 71 percent of the land plots they
examined, the tlahuelmatli value was within 10 percent of the projected area for
that plot based on the milcocoli measures (1980: 501). While this may not seem
remarkably accurate, the plots were often very erratic in shape, so that calculating area was not simply a matter of multiplying length by width. Where there
is no value in the third (central) register, a corn glyph, or cintli, is drawn at the
top of the rectangle (Harvey and Williams 1980: 501). This sign may have been
to indicate that the third register is empty, and thus may have served one of the
functions of a zero-sign.
These numbers can be read as a base-20 cumulative-positional numerical notation system with a sub-base of 5. Unlike Western numerals, in which the positions
are arranged in a straight horizontal line, the Texcocan system uses three registers,
the last two of which have an identical positional multiplier. However, the cintli
glyph is not used to indicate empty positions, but rather provides information as
to where to find the twenties power (on the bottom line, rather than in the center
of the rectangle), and thus is conceptually distinct from the Western zero (and,
indeed, from other zeroes such as the Babylonian zero). While I think that the
correlation established by Harvey and Williams demonstrates that the tlahuelmatli
value represents an area value, I am not fully convinced that it is meant to be read
as a single number; it may instead represent two values, one of which represents a

Mesoamerican Systems

305

Figure 9.1. Numerical phrase from the Codex Kingsborough enumerating the population of Tepetlaoztoc at 27,765 (3 8000 + 9 400 + 8 20 + 5). Source: Paso y Troncoso
1912: 218v.

larger area value that is twenty times another value. I do not know how this issue
could be resolved at present.
A unique Texcocan document from 1555, the Codex Kingsborough, also uses
something like the line-and-dot numerals (Paso y Troncoso 1912). It was a record
prepared as part of a legal plea made to the Spanish encomendero of the region,
denoting the massive amount of tribute paid to Spanish officials by the inhabitants of the Tepetlaoztoc region in an effort to convince colonial officials that the
populace was overworked; extensive description in Spanish confirms the numerical values (Harvey 1982: 193). Curiously, this text combines Aztec numerals and
Texcocan line-and-dot notation. Lines and chunked groups of five lines indicate
1 and 5, respectively. To write larger numbers, dots organized in lines of five were
placed beside the signs for 20, 400, and 8000. The dots were placed in a single row,
with the signs for 20 and 400 above them and the 8000 sign below them. Thus,
where the regular Aztec numerals use these three signs cumulatively, the Kingsborough numerals are written using just one of each sign, next to which units from 1
to 19 were expressed with dots. Figure 9.1 depicts the numeral-phrase 27,765, indicating the population of the district at the time, but replacing the standard Aztec
sign for 8000 with a head above a sack (Paso y Troncoso 1912: 218v).
Whereas the basic line-and-dot system is cumulative-additive, and the tlahuelmatli system is cumulative-positional, this system is multiplicative-additive.
While the dots look like the 20 dots of the line-and-dot system, they each stand
for 1 in this system. The total value of the numeral-phrase is taken by multiplying
the dots for units by the values of the power signs and taking the sum. To add
to the complexity of this situation, in some cases the flag glyph for 20 could be
omitted, retaining only the dots (Paso y Troncoso 1912: 261r, 238v, etc.). In these
cases, we have the elements of a cumulative-positional system, since the value of
the twenties power is determined by its position in the numeral-phrase through
implied multiplication. Finally, in a couple of numeral-phrases, lines are placed to
the left of dots, as where a number is written as II, which might be read from

306

Numerical Notation

right to left as 42 (Paso y Troncoso 1912: 274v). The erratic nature of the system
suggests that whoever wrote it was extremely inventive and was in the process of
experimenting with different means of representation.
The most important question regarding the line-and-dot numerals, their positional variant in the tlahuelmatli records, and their multiplicative variant in the
Codex Kingsborough, is whether they existed before the Conquest, or if their
development was stimulated by contact with the Spanish. Neither the Western or
Roman numerals are cumulative-positional or multiplicative-additive, and neither
uses a base of 20, so the Texcocan systems are structurally distinct from those of
the Europeans. Thus, it would be premature to conclude that Spanish contact
brought about the development of these systems. It would be a mistake to attach
much importance to the use of a vertical stroke for 1 (parallel with both Western
and Roman numerals), given the ubiquity of this notation worldwide. Harvey
and Williams (1980: 503) argue that, while the tlahuelmatli numerals are positional
and have something like a zero, the use of different registers around a rectangle is
quite different from Western positionality, and the zero does not serve the same
functions as the Western zero. On this basis, they regard these systems as a native
invention.
I agree that the Texcocan numerical notation systems are so different from
Western and Roman numerals that the Spanish could not have introduced
them. Nevertheless, these may be instances of stimulus diffusion, which the
Texcocan scribes developed with an awareness of Western and/or Roman numerals but without adopting the form and structure of those systems. That the
Texcocan systems occur in only a handful of documents in a single region in
the generation immediately after the Conquest and cease to be used after only
two decades suggests that this was not a system of great antiquity. I believe that
the multiplicative (Kingsborough) and positional (tlahuelmatli) variants may
well have been stimulated within the rapidly changing social and intellectual
environment of the early colonial period, while the cumulative-additive lineand-dot numerals probably existed in the pre-Conquest period. After 1545, epidemic disease greatly diminished the indigenous population of the region, and
it appears that the Texcocan numerals ceased to be used after the middle of the
sixteenth century.

Other Systems
Because our understanding of Mesoamerican numerals is imperfect, a number
of Mesoamericanists have developed theories regarding other forms of written
numeration. I think it quite likely that more numerical information has been
recorded than we are currently able to read in the Maya, Zapotec, Teotihuacni,

Mesoamerican Systems

307

and Aztec texts. Even if these hypotheses turn out to be incorrect, some elements
of them may be salvaged in the reconstruction of as-yet unknown numerical notation systems.
In the 1950s, Howard Leigh postulated that in addition to bar-and-dot numerals, some Zapotec inscriptions contained encoded astronomical data using a different set of glyphs (Urcid Serrano 2001: 4950, 54). In addition to bars and dots,
this supposed system had over twenty unique signs, including elements of base-10,
base-13, and base-20 notation, culminating in a special sign for 1,186,380 (3 3
3 13 13 13 20). I am unconvinced that such a system actually existed in
the form asserted, but the Zapotecs may have encoded numerical information in
some of these glyphs, though not in the way Leigh imagined.
An unusual cumulative-additive bar-and-dot numerical notation system may
have existed at Teotihuacn, a system in which the bars did not have a fixed value
but could mean 5, 10, or 30, depending on their configuration (Langley 1986: 141).
The nature of the script of Teotihuacn is still controversial, though it is increasingly thought that there was a complex pictographic script of the type used later in
highland Mexico (Taube 2000). However, because Teotihuacn never used phonetic writing, and because, unlike the Aztec situation, there is no body of colonial
documents to explain the numerals, there is no way to confirm the values of any
potential numeral-signs.
Penrose (1984) asserts that in the almanac portions of the Dresden, Madrid, and
Paris codices, the Maya used cryptoquantum numerations to represent an encoded quantity of days separately from the bar-and-dot or head-variant numerals.
He argues that the Maya represented hidden counts of large numbers by assigning
numerical values to special signs indicating the days of the Sacred Round 260day calendar, and then by manipulating them through multiplication. Mayanists
do not appear to be aware of Penroses research, and his conclusions must be
viewed as highly speculative and even pseudoarchaeological. The manipulations
necessary to extract meaningful numerical information from these signs are probably no more than numerological play.
An unusual form of numerical notation is employed on the Codex Mariano
Jimenez, a sixteenth-century post-Conquest manuscript from Otlazpan (in the
province of Atotonilco). It is cumulative-additive and uses dots for units, horizontal lines for twenties, and horizontally oriented tzontli (feather) glyphs for
400, with fractions of 400 depicted by showing partially denuded tzontli signs.
Although treated by Harvey and Williams (1986: 251253) as simply a variation
on the Texcocan system described earlier, the differences between the two systems
suggest that they are quite distinct. If more documents using this sort of notation
are found, we would have yet another post-Conquest regional variant of the Aztec
numerals.

308

Numerical Notation

Summary
The two features common to all the Mesoamerican numerical notation systems
is that they have a vigesimal base and that they are all cumulative rather than
ciphered. The Maya head-variant glyphs, a sort of ciphered symbolic code that
expresses only units up to 19, constitute a partial exception to this rule. The
presence of a quinary element is quite common, as is the use of dots for units,
but neither of these features is found in all the systems. Like the East Asian
phylogeny (Chapter 8), Mesoamerican numerical notation systems use a variety
of basic principles, and our primary evidence for their commonality is historical
rather than structural.
When the bar-and-dot numerals were the only part of the Maya script to be
deciphered, it must have seemed remarkable to be able to extract calendrical information from such otherwise inscrutable documents. Yet we have a less complete
understanding of the cultural history of the Mesoamerican numerical notation
systems than we do of most Old World families. As our reading of Maya and
Aztec writings becomes more sophisticated, it is to be hoped that we will come to
a clearer understanding of their numerical notation.

chapter 10

Miscellaneous Systems

Around twenty systems do not fit neatly into the phylogenetic classification presented in Chapters 2 through 9. A few, such as the Inka khipu numerals, the Indus
(Harappan) numerals, and the enigmatic Bambara and Naxi numerals, apparently
arose independently of any other system, but gave rise to no descendant systems.
Others are cryptographic or limited-purpose systems used in the medieval or early
modern manuscript traditions of Europe and the Middle East. The majority of this
chapter, however, deals with systems that emerged in colonial settings under the
influence of the Western or Arabic ciphered-positional numerals, in conjunction
with the development of indigenous scripts. Most of these systems were developed
in sub-Saharan Africa, but Asian (Pahawh Hmong, Varang Kshiti) and North
American (Cherokee, Iupiaq) indigenous groups have also developed their own
numerical notation systems. Finally, a few systems are probably members of other
phylogenies, but their exact affiliations remain inscrutable enough that no definite
conclusions can be reached.

Inka
The Inka civilization was an enormous state on the Pacific coast of South America that reached its pinnacle between 1438 and 1532. While writing is often (and
mistakenly) seen as a sign of civilization, or at least as a necessity for large-scale
bureaucracy, the pre-colonial Inka state operated in the apparent absence of any
309

310

Numerical Notation

writing system capable of expressing phonetic values. Instead, the primary means
of encoding information was a system of knotted cords of different colors, known
as khipus,1 whose main purpose was to record numerical information to aid in the
administration of the Inka state. About 500 to 600 Inka khipus survive, although
accurate provenances cannot be established for most of them (Urton 1998: 410).
As first established by Locke (1912), khipus encode information using a decimal
positional numerical system of knots, and around two-thirds of all attested khipus
encode information in this readily understood fashion. Around one-third, however, do not follow this structure; they remain completely undeciphered, and may
well have encoded non-numerical information (Urton 1997, Quilter and Urton
2002).
A khipu is a set of colored cotton or wool cords consisting of a main cord (ranging from 10 to 20 cm up to several meters in length) from which multiple cords
are suspended. These numeral-bearing cords are subdivided into pendant cords,
which hang directly down from the main cord when it is held horizontally and
stretched taut; top cords, which hang from the main cord but are tied so as to lay
on the opposite side of the pendant cords; and subsidiary cords, which hang from
a pendant cord, top cord, or another subsidiary cord rather than the main cord
(Ascher and Ascher 1980: 1517). The designation that pendant cords hang down
and top cords hang up is an artifice; while they naturally hang on opposite sides
of the main cord, we do not know how they would have faced. In numerical khipus, pendant, top, and subsidiary cords may contain a numeral-phrase or, more
rarely, two. The system used to encode information is cumulative-positional with
a base of 10. In each position, the value of that power of 10 is encoded using one to
nine knots or loops. There is no sign for zero; instead, a space was left on the cord
in an empty position. The units position is the one farthest from the main cord
(its loose end), while the highest power is found closest to the main cord. While a
khipu theoretically could express any number (because the system is positional),
in practice, five-digit numbers are the largest recorded, and these are quite rare
(Ascher and Ascher 1972: 291). Despite this obvious numerical structure, khipus
are often erroneously conflated with unstructured systems that use one knot for
one object (cf. Ifrah 1998: 70). Khipus contain a numerical notation system (a
positional one, in fact) and thus must be compared to written numerals rather
than to simple tallies.
Three different sorts of knots encoded numeral-phrases, as seen in Figure 10.1.
To encode a value in the tens, hundreds, or higher powers, the khipu maker would
tie an appropriate number of single knots in a line. For the ones power, however, two
1

The spellings Inka and khipu currently enjoy favor with Andeanists over the older
Inca and quipu.

Miscellaneous Systems

Single knot
10s, 100s, etc.

Long knot
Units: 2-9

311

Figure-8 knot
Units: 1

Figure 10.1. Khipu knots.

different types of knot were used. For all the units except 1, the cord was looped
around itself an appropriate number of times for the number being expressed;
the long knot shown in Figure 10.1 represents 4. Because a long knot cannot
be made with fewer than two loops, a value of one in the units position required
the use of a different knot, a figure-8. The use of different knots might appear to
take away from the purely positional nature of the system. Yet, because there is
no zero-sign, this technique greatly reduced the chance of misreading a cord. If a
cord contained six single knots followed by two single knots, it could not be read
as 62 but only as 620 (or possibly 6200). The use of long or figure-8 knots in the
units position makes it much easier to tell which is the units position, and thus to
identify the subsequent positions.
Figure 10.2 depicts an unattested but plausible khipu. The main cord lies horizontally, with the pendant cords (P1 through P4) hanging down and the top cord
(T1) facing up, and with subsidiary cords (S1 through S3) hanging off both pendant and top cords. On this cord, only a single value would have a figure-8 knot
(the 1 in the units position on P4); the other units values (3 on P2, 6 on S1, 2 on
P3, 6 on T1, and 6 on S3) would be made with long knots, and all the tens and
hundreds figures with single knots. As is sometimes the case in attested khipus,
the top cord value (776) is equal to the sum of the pendant cords (360 + 23 + 102 +
291), while the value on the top cords subsidiary (S3 = 26) is the sum of the subsidiaries of the pendant cords (20 + 6).
Although we can read the numerical values on khipus, their origin and early
history remain unclear. A set of twelve cotton strings twisted around sticks excavated at the late pre-ceramic pyramid complex of Caral, Peru (c. 26002000 bce)
has been claimed by its excavator to be an early form of khipu; however, this
claim is unsubstantiated, and full data on the artifact remain unpublished (Mann
2005). Bennett (1963: 616) notes that some Mochica vessels (Early Intermediate
period, c. 200600 ad) bear markings that are suggestive of khipus. The first wellsubstantiated evidence for khipu use comes from Middle Horizon sites (c. 600
1000 ad) associated with the Wari civilization in coastal Peru (Conklin 1982).

Numerical Notation

312

T1
776

P1
360

P2
23

S3
26

S1
6

P3
102

Figure 10.2. Khipu structure.

P4
291

S2
20

Miscellaneous Systems

313

These khipus cannot be deciphered numerically because of their deteriorated condition (although they may have used nondecimal bases), and they use color in very
different ways than the Inka khipus, but are nonetheless clearly of the same basic
type. Most surviving khipus were collected haphazardly; prior to 2001, only two
archaeological discoveries of khipus had adequate proveniences (Urton 2001: 131).
The khipu system may have developed out of an earlier knot-based system using
simple one-to-one correspondence, because knot tallies of this sort are widely distributed in the Circum-Pacific region (Birket-Smith 1966). There is no evidence
of any connection between the khipu notation and any other numerical notation
system, and thus it is definitive that the Andes was home to an independent development of the place-value principle.
Khipus were a vital part of the Inka record-keeping system; they were employed
in this capacity for censuses, tributary records, and similar administrative functions. Jacobsen (1964), noting the frequency with which the top cord equals the
sum of the pendant cords, suggests reasonably (but unconfirmably) that such khipus may have been part of a double-entry accounting system. The decimal base
of the khipu notation system corresponds to the decimal divisions of society by
which the state was administered. Some khipus contained calendrical rather than
administrative information (Ascher and Ascher 1989, Urton 2001). For instance,
khipu UR6 from Laguna de los Cndores contains a series of cords with values of
20 to 22 followed by cords with values of 8 or 9, and the sum total of these cords
is 730 (365 2), strongly suggesting that it may have been a biennial calendar
(Urton 2001: 138143). Most surviving khipus with good provenience have been
recovered from mortuary contexts. The Inka probably placed khipus in the graves
of khipukamayuqs (khipu makers and users). It is unclear whether this implies that
some of them should be read as tomb texts, because at present we are unable to
extract non-numerical information from them (Urton 2001: 34).
Much ink has been spilled recently about whether khipus constituted something more than a numerical notation system, approximating the functions of a
writing system. Ethnohistorical data suggests that khipus recorded genealogical,
historical, and literary information, which raises the question of what code was
used to do so (Bennett 1963: 618). Gary Urton (1997, 1998, 2001) has argued forcefully that many khipus contain syntactic and semantic information far exceeding
their numerical functions. He contends that purely numerical readings that translate khipu texts as Western numerals inevitably mask, and eliminate from analysis, any values and meanings that may have been attached to these numbers by the
Quechua-speaking bureaucrats of the Inka empire who recorded the information
(Urton 1997: 2). He argues against the idea that a khipu could have been interpreted
only by its maker or those trained in an idiosyncratic private code (Urton 1998: 412).
The khipus must have recorded some non-numerical information; a list of pure

314

Numerical Notation

numbers is practically useless. In some way, at least the nature of what was being
counted must have been recorded somehow. The most likely possibility is that this
was done with color; the 1609 Comentarios of Garcilaso de la Vega (15391616)
inform us that colored cords were used to record different commodities (Bennett
1963: 617). Yet many khipus use multiple colors of cord, and there exists no reliable
means of reading the type of items counted.
Recent scholarship has established that at least some khipus encoded toponymic information, associating particular records with the places to which they
refer, which helps us to clarify how information was communicated within the
Inka administrative hierarchy (Urton 2005, Urton and Brezine 2005). A cache of
twenty-one khipus excavated from a single urn in the palace of Puruchuco (northeast of Lima, Peru) revealed many whose introductory cords begin with arrangements of three figure-8 knots (which normally represent the numerical value 1),
suggesting that this served as an identifier with which any reader could associate
the numerical data. Some of the khipus in this cache encoded identical or nearly
identical information, suggesting that copies might be kept at the site of a khipus
manufacture, with other copies distributed to the capital, Cuzco, or elsewhere.
Urton has also identified three-term number sets that occur on some of the Puruchuco khipus that do not fit into the numerical structure of the remainder of the
record, and suggests that these are labels, perhaps an ayllu (kinship group) with
which the khipu was associated (Urton 2005: 162163).
The khipus encode at least as much information as the proto-cuneiform
accounting signs of Mesopotamia (Chapter 7), which identify only items being
counted and the quantity of each item, but which similarly served as a state-oriented bookkeeping system of credits and debits (Urton 2005: 164). Since the
proto-cuneiform system is regarded as proto-writing, it is reasonable to attribute
the same status to the Inka recording system (Salomon 2004). It is possible that
the khipu system, over time, might have developed into a system for representing
speech (though doing so would be more difficult for a knot-based notation than
for a system based on inked or impressed signs). A single khipu cannot be at the
same time both a record of numbers and of things being enumerated and a fully
developed system for recording history and literature. Yet the roughly one-third
of khipus that do not follow an ordinary decimal and positional structure may
well have been non-numerical. It is equally possible that some numerical khipus
recorded ideas or speech through some sort of code, but without a key, we cannot
definitively conclude that this was the case.
Urtons (1997: 179) speculation that there might have been two pre-colonial
khipu systems (one for recording quantity and another for recording narrative) is
useful but at present unconfirmed. While post-Conquest chroniclers state explicitly that the khipus carried only numerical meanings, Urton postulates that the

Miscellaneous Systems

315

early colonial Spanish, in order to undermine traditional patterns of knowledge,


rapidly transformed the khipu system from a full-fledged writing system into a
purely numerical and non-narrative recording instrument (Urton 1998: 410411).
I admit that the Spanish may have wished to denigrate Inka knowledge, and
also that there is an enormous issue of translation between indigenous concepts
and what is claimed in early Spanish chronicles. Yet it would have been much
simpler to replace the khipu system with European administrative techniques
than to attempt such an alteration of its function. Moreover, analogies with the
mathematical practices of modern Quechua speakers will not help us to interpret
centuries-old khipus unless continuity between pre-colonial and modern ways of
thinking can be demonstrated. We know that the pre-colonial khipu system was
at minimum a number + noun information system, of which only the numerical
component can usually be determined, but we do not know more than this with
any certainty. The earliest Mesopotamian civilization did not require phonetic
writing, nor did that of the Yoruba, to mention only two highly complex but nonliterate sets of polities. To infer a writing system out of nothing but an assumption
that such a system would have been necessary is grossly anti-empirical.
Regardless, khipus alone cannot have been used for performing arithmetical calculations. Khipus are even less amenable to physical manipulation than are written
numerals (which can be lined up and crossed out). We do know, from sixteenthcentury documents, that the khipukamayuq were responsible not only for making
and reading the khipus but also for calculating the results, and that they did so
using a set of stone tokens (Urton 1998, Fossa 2000). While no archaeological evidence has confirmed the existence of such a system, there is limited documentary
evidence for an Inka abacus in the Nueva cornica y buen gobierno, a document
written between 1583 and 1613 by Don Felipe Guaman Poma de Ayala (ca. 1534
1615), a descendant of an Inka princess who was an important chronicler of life in
late sixteenth-century Peru and a critic of Spanish rule (Wassn 1931; Urton 1997:
201208). In one corner of a page depicting a khipukamayuq at work, there is a grid
of five rows by four columns, in each square of which is found a number of circles:
five dots in the first column, three in the second, two in the third, and a single dot
in the fourth. Moreover, some of the dots have been filled in, while others remain
empty. Unfortunately, while the commentary that accompanies this picture notes
that the Inka reckoners used computing boards, there is no description of how this
system worked. Wassns (1931: 198199) effort to infer this information assigns the
rows values of the powers of 10 (starting with 1 at the bottom) and the values 1, 5,
15, and 30 to the columns (which were multiplied by the row-value), but he does so
unconvincingly, solely on structural grounds. Nevertheless, it is unlikely that this
board is a result of diffusion from Spain, since no comparable board was used in the
sixteenth century anywhere in Europe (Wassn 1931: 204).

316

Numerical Notation

After the Spanish conquest in 1532, khipus continued to be used for the same
administrative functions as they had been previously, and the Spanish, through
Andeans who could read their values, used the data recorded on them (Loza 1998,
Fossa 2000). Brokaw (2002) discusses the cognitive shift required of Quechua
speakers in the sixteenth century with the transition from khipu notation to
European literate conventions, as demonstrated through Guaman Pomas Nueva
cornica. This shift involved changes in how texts were organized, written, and
read, and also forced pre-existing Quechua ideas about numbering and counting into conflict with Western textual conventions (e.g., regarding pagination).
Also in the sixteenth century, the mestizo chronicler Blas Valera (15451597), who
advocated for Quechua as a Christian liturgical language in addition to Latin,
developed a system of forty syllabic knots to be used on so-called royal khipus,
which reflected Valeras theories about Quechua as a worthy language and his
conviction that one could not rank societies based on the quality of their writing systems (Hyland 2003: 129135). While Valeras system has occasionally been
regarded as a pre-colonial invention for which he took credit, thus making the
khipus at least partly a phonetic writing system, it is substantially more likely that
the royal khipus were a colonial invention that applied the notion of phoneticism
to the existing pre-colonial system.
The widespread use of khipus was curtailed in the 1580s, when they were declared
to be idolatrous and the Spanish colonial administrators decreed that they should be
destroyed. Yet in that same decade Mercedarian friars began using khipus to encode
information about Christian life, using the principles outlined by Valera (Hyland 2003:
136137). While it was once thought that the use of khipus had essentially ceased by the
sixteenth century, it is now evident that their use for secular administration in the colonial period continued. Moreover, in local accounting contexts apart from state control,
khipus have continued to be used by animal herders in parts of Peru and Bolivia for
recording quantities of livestock up to the present day (Bennett 1963: 618619; Ifrah
1998: 6970; Urton 1998: 410; Salomon 2004). Nineteenth-century khipus found by
the explorer Charles Wiener in Paramonga have systems of knots and bundles quite
different from the pre-colonial khipu, but confirm that the practice continued in varying forms well after the colonial period (Hyland 2003). These were not simply tallying
systems, however, but were cumulative-positional and decimal, and thus constitute a
survival of the Inka numerical notation system.
A significant part of khipu studies today, then, and of Inka ethnomathematics in
general, relies on ethnographic work with the descendants of the Inka, for example, modern Quechua and Aymara (Quilter and Urton 2002, Salomon 2004).
The modern episteme of numbers rests on different principles than Western
arithmetic, in particular placing great emphasis on even numbers as complete
and odd numbers as incomplete or even dangerous (Urton 1997; Brokaw 2002:
281287). However, between the sixteenth and twenty-first centuries, significant

Miscellaneous Systems

317

changes almost certainly occurred in these ideas, much as the foundations of sixteenth-century European mathematics bear only a passing resemblance to modern
practices. Further ethnographic, ethnohistorical, and archaeological data promises
to help resolve some of the remaining mysteries concerning the khipu records,
and the establishment of the Khipu Database Project will help facilitate computer analysis of khipus in museums and collections worldwide (Khipu Database
Project 2004). A complete decipherment of the khipus as they were used in premodern contexts, however, may well be impossible.

Bambara
One of the most peculiar African numerical notation systems was used by the
Bambara of Mali in religious and divinatory contexts (Ganay 1950). Although
details of the systems history are sketchy, we have a fair idea of the numeralsigns and the structure of the system. The Bambara numeral-signs are shown in
Table 10.1.
The Bambara system is structurally irregular; while it is additive, it alternates
between cumulative and ciphered notation, and while it is mainly decimal, it has
vigesimal components. For instance, 1 to 19 are written primarily with vertical
cumulative unit-strokes. The value of a set of vertical strokes is doubled if a horizontal line is crossed through it (effectively dividing the number into two registers,
one above and one below the line). For odd numbers, an additional half-stroke
can be placed at either end of the phrase, sometimes vertically and other times
at an angle. Each of the tens from 20 to 170 has its own sign, which makes the
system ciphered at this point. The signs for 180 and 190 are additive combinations
of 100 + 80 and 100 + 90, respectively. To add a number of units from 1 to 9 to
one of these ciphered signs, an appropriate number of strokes are attached to the
sign for the multiple of 10 (or dots, when adding units to 60, 160, or 170). This
means of representation is decimal each decade has its own sign to which up
to nine unit-signs were attached. Yet, because there are signs for 110, 120, and so
on, it is unlike the Greek ciphered-additive alphabetic numerals (in which 100 is
followed by 200, 300, and so on). Moreover, some of the decade-signs are similar
enough to the ones preceding them (40 vs. 50, 100 vs. 110, 140 vs. 150, 160 vs. 170)
to suggest an additional trace of a vigesimal base. For numbers higher than 200,
the cumulative principle is again employed by repeating the sign for 100 (another
decimal component) as many times as required in a vertical column, with any
needed additional signs placed at the top of the column. Figure 10.3 shows some
higher numeral-phrases (as reproduced from Ganay 1950: 300).2
2

Large numeral-phrases for 1935 and 4000 are also listed, but are highly irregular, and I
cannot determine what principle has been used to determine their value.

Numerical Notation

318

Table 10.1. Bambara numeral-signs


1

aa

aaa

aaaa

aaaaa

10

aaaaaa

aaaaaaa

bbbb

bbbbc

bbbbb

11

12

13

14

15

bbbbbc

bbbbbb

bbbbbbc

bbbbbbb

ybbbbbbb

16

17

18

19

20

bbbbbbbb

bbbbbbbbz

bbbbbbbbb

ybbbbbbbb

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

The Bambara numerical notation system was used primarily in ritual contexts,
especially those pertaining to divination using numbers (Ganay 1950: 298). Little
is known of its origin, period of use, or decline. It shows no resemblance to any
of the systems that would have been known by Bambara, who had considerable
contact with the Muslim world. While the ciphered-additive Arabic abjad numerals commonly used for divination in the Maghreb are the most likely ancestor,

220

489

230

240

Figure 10.3. Bambara numeral-phrases.

Miscellaneous Systems

319

the Bambara system is quite different in most respects its frequent use of the
cumulative principle, the presence of a vigesimal component, and its numeralsigns. I have no idea whether this system continues to be used, though I suspect
that it does not.

Berber
The Berbers, or Imazighen, live in North Africa and speak a set of closely related
Afro-Asiatic languages. For most of their history, the Berbers have been a marginal
people living on the periphery of larger polities (Carthage, Rome, and various
Muslim states), but they have nonetheless retained considerable cultural independence. The Berbers developed a consonantal script on the model of that used
in Punic Carthage possibly as early as the sixth century bc, which was in continuous use until at least the third century ad; the Tifinigh script (still used by the
modern Tuareg for love letters, domestic ornamentation, and games) is descended
from it (OConnor 1996). There is no numerical notation system associated with
either the classical Berber script or its modern descendant. Nonetheless, a distinct
numerical notation system was used by traders in the Berber city of Ghadames (on
the border of Algeria and Libya) in the nineteenth century, and appears to be in use
still (Rohlfs 1872, Vycichl 1952, Aghali-Zakara 1993). Vycichl (1952: 8182) presents
the system as described by two separate authors, Hanoteau and Si Mohammed
Serif, while Rohlfs presents a third system; I reproduce all three in Table 10.2.
The system is cumulative-additive and written from right to left, with the decimal powers repeated up to four times and the halved powers only once in any
numeral-phrase. Sometimes, groups of signs could be placed in two rows to save
space (Rohlfs 1872). In addition to these signs, Hanoteau claims that a horizontal
line stood for the fraction 1/4, and that this sign could be grouped vertically to
indicate 1/2 and 3/4 (Vycichl 1952: 81). The two sets of numeral-signs are identical,
except for the signs for 500 and 1000. It is possible that both of these systems were
actually used, either in different contexts or at different times. However, it is more
likely that an error of interpretation created the discrepancy, because Hanoteaus
1000-sign is essentially identical to Serif s 500-sign.
The question of the Berber systems ancestor (if any) is still open. It is possible
that it was an entirely independent development. The similarities between certain
numerical signs and letters of the Berber consonantary (r with 10, f with 500,
and s with 1000) are interesting, but they do not correspond to the Berber lexical
numerals in any obvious way. The Phoenician/Punic numerical notation system is
quite different in its structure, lacking a sign for 5, and employing a special sign for
20 and a hybrid multiplicative-additive structure above 100. The use of | for 1 and >
for 5 is superficially similar to the Roman system; Ghadames was an important

Numerical Notation

320
Table 10.2. Berber numerals
1

1
1
Rohlfs
1
1/4: E 1/2: J 3/4: O

Hanoteau

Si Mohammed Serif

10

50

100

500

1000

44 = 88
488 = 111

trading post (Cydamus) under imperial Roman control, and there are Roman
numerals on some of the Latin inscriptions found there. However, the systems
are written in different directions and have different signs for the higher values.
Vycichl (1952: 83) suggests that the system derives from the South Arabian numerals. The Berber script may be somehow indebted to the South Arabian (OConnor
1996: 112). If Hanoteaus list of signs is correct, the Berber system, like the South
Arabian, lacks a sign for 500; furthermore, both systems use O for 10. However,
the South Arabian system ceased to be used in the first century bc and was never
used in Africa, so to accept this theory requires that we believe in a two-thousandyear unattested history for this system. The system having the most promise as an
ancestor is the Arabico-Hispanic variant Roman numerals (Chapter 4) used in a
Spanish Inquisition document of 1576 (Labarta and Barcel 1988: 34). This system
employed |, V, and O for 1, 5, and 10, was written from right to left, and was used
in the same general region as the Berber system. Though three centuries is still a
chronological gap that needs to be resolved, it is not nearly so great as the enormous leaps that need to be inferred to hypothesize alternate paths of diffusion.
Ultimately, more data are needed for this system to be assigned unambiguously
to any phylogeny.
The Berber system was used in the nineteenth century for indicating the prices
of trade goods. Rohlfs (1872) learned about this system as a traveler in the Ghadames region, but only ascertained the meanings of the signs through great effort
and negotiation. He thus believed that the system was semi-cryptographic, restricting the flow of information concerning prices to a limited group of Berber traders
in order to give them an advantage over Arab traders. The system is not especially
difficult to decipher, however, and so I am unconvinced that this purpose was very
important. Aghali-Zakara (1993: 151153) reports that several numerical notation
systems are still used in the region of Ghadames; one of these is the system just
described; another simply repeats the sign for 10; and a third, inexplicably and
surely incorrectly, is seen as having no signs for the powers of ten but only for the

Miscellaneous Systems

321

Table 10.3. Oberi Okaime numerals


1

10

11

12

13

14

15

16

17

18

19

@ # $ % ^
1938 = 4^* (4 400 + 16 20 + 18)

&

sub-bases, 5, 50, 500, and 5000. None of these systems is widely used, but they do
appear to be in present use among at least some Tuareg.

Oberi Okaime
In the late 1920s, a syncretic indigenous-Christian religious movement known as
Oberi Okaime (or Obri Vkaim) arose among the Ibibio-Efik, speakers of a set
of related dialects of the Niger-Congo language family in southeastern Nigeria.
By 1931, the divinely inspired leaders of this movement had developed an alphabet
(written from left to right) and a set of numeral symbols (Adams 1947; Hau 1961).
The script represented an arcane revealed liturgical language of the sect, but was
not used to write Ibibio. The Oberi Okaime numeral-signs are shown in Table 10.3
(Hau 1961: 295).
The system is ciphered-positional and vigesimal; it is the only known cipheredpositional base-20 system with no sub-base, with the partial exception of the Maya
head-variant glyphs. The vigesimal structure of the system is based on the similarly
vigesimal Ibibio lexical numerals (Abasiattai 1989: 505506). Numeral-phrases are
written from left to right with the highest powers on the left. The inventors of
the Oberi Okaime numerals were educated in Christian missionary schools in the
1920s, where they became literate in English and learned Western numerals. While
none of the numeral-signs resemble the corresponding Western numerals except
for 0, the script and its numerals were strongly influenced by Western traditions
of writing (Dalby 1968: 160161). Haus (1967) highly dubious suggestion that the
Oberi Okaime script derives directly from Minoan Linear A, used thousands of
kilometers away and over three millennia previously, cannot possibly apply to the
numerals. The numerals were used in a relatively small number of liturgical texts
and personal letters among the members of the Oberi Okaime sect. The system
was still used by some individuals when Kathleen Hau corresponded with its leaders in 1961. In 1986, Sunday school classes were begun in order to revive the Oberi
Okaime liturgical language along with the numerals and script, but this appears

Numerical Notation

322

Table 10.4. Bamum numerals (original)


1

10

100

1000

10000

76 =
or

to have been unsuccessful (Abasiattai 1989: 506). Western and sometimes Arabic
positional numerals are used in the region today.

Bamum
The Bamum live in part of southwestern Cameroon near the border with
Nigeria. In the late nineteenth or early twentieth century,3 Sultan Ibrahim Njoya
(ca. 18751933), a Bamum ruler, took it upon himself to develop a script for his
people. Njoya, aided heavily by an assistant, Nji Mama, developed the script
through several stages, starting with a large logosyllabary and gradually reducing
the number of signs into an eventual syllabary of eighty characters (Tuchscherer
2005: 479). From its inception, Bamum writing made use of numerical notation.
The earliest Bamum numerals are shown in Table 10.4 (Dugast and Jeffreys 1950: 6).
This system is decimal and multiplicative-additive, with numeral-phrases written from left to right. Curiously, the power-sign for the units could either precede
or follow the unit-sign (Dugast and Jeffreys 1950: 30). The unit-signs for 7, 8, 9,
and 10 were not at this stage fully ideographic, but instead were constructed of
two graphic parts, each of which represented a syllable in the two-syllable Bamum
words corresponding to those numbers (Dugast and Jeffreys 1950: 98). At this
point in the systems history, we could well consider it to be a set of lexical numerals. This is the same problem we encountered with the Shang/Zhou and Chinese
classical systems (Chapter 8), which, not coincidentally, also are multiplicativeadditive and associated with logosyllabic scripts in which some characters (including numeral-signs) are ideograms.
Around 1921, Njoya supervised a transformation of the script into a form
known as mfmf, which altered the numerals from multiplicative-additive to
3

Dugast and Jeffreys (1950: 4) place its invention in 1895 or 1896, although it may have
been as late as the turn of the century.

Miscellaneous Systems

323

Table 10.5. Bamum numerals (mfmf)


1

ciphered-positional by removing the power-signs (Dugast and Jeffreys 1950: 30).


The old sign for 10 took over the role of zero; numeral-phrases were written
from left to right with digits for 0 through 9, like Western and Arabic positional
numerals, the Bamum systems primary rivals. The mfmf numerals are shown in
Table 10.5 (Dugast and Jeffreys 1950: 31).
During its heyday in the first three decades of the twentieth century, the Bamum
numerals were used quite widely, primarily due to Njoyas political clout, and were
employed on legal documents, census records, histories, and personal letters, both
handwritten and printed. Njoya was deposed in 1931 and died two years later, after
which time the Bamum script and numerals rapidly fell into obsolescence. Today,
the script is preserved to a small extent as a source of ethnic pride among some
Bamum, but there are very few surviving users (Tuchscherer 2005: 479). Nevertheless, because we are able to trace their rapid transformation from an additive to a
positional structure, the Bamum numerical notation systems are more than just a
historical curiosity and tell us a great deal about the way that numerical systems
change.

Mende
Around 1917, an Islamic scholar named Mohamed Turay developed a syllabic script
known as Kikakui to represent graphically the Mende language spoken in southern Sierra Leone, almost certainly influenced by the Vai script of Liberia developed in the previous century (Tuchscherer 2005: 478). A few years later, Kisimi
Kamara, Turays grand-nephew and student, expanded and revised the script in a
second stage. While Western numerals were always used alongside the Vai script,
the inventors of the Mende script developed a distinct set of numerical signs to
accompany the syllabary, although it is unknown which of Turay or Kamara was
responsible for the innovation. The Mende numeral-signs are shown in Table 10.6
(Tuchscherer 1996: 7175).
The system is decimal and multiplicative-additive, and numeral-phrases are
constructed with the highest powers on the right. Because the system is multiplicative-additive, no sign for zero is needed or used. Unit-signs are placed above the
corresponding power-signs, and so numeral-phrases are read from top to bottom
and from right to left. There are two signs for 10. The first, 10 (+) in Table 10.6,
combines additively with the units for 1 through 9 in order to write 11 through 19,

Numerical Notation

324

Table 10.6. Mende (Kikakui) numerals


1

10 (+)

10 ()

100

1000

10,000

100,000 1,000,000

m
n
o
P
d
k
\\\\ba
hjm
\F
io
\\e\e\e\e\e\e
e\j\m\n\o\p\

14
128
60,009

5,555,555

while the other, 10 (), is a multiplicative power-sign for 10 that combines with
the unit-signs for 2 through 9, or with 10 alone by placing a dot rather than a
sign for 1 above it (Tuchscherer 1996: 7172). The higher power-signs use vertical
strokes to indicate repeated multiplication by 10; the number of strokes represents
the exponent of 10 corresponding to the number. This feature is quite distinct
from the cumulative principle, which always refers to repeated addition of similar
symbols, and is unique to the Mende system. In theory, the system could have
been extended infinitely without using the positional principle, although there are
practical limits to how many vertical-strokes could be read easily.
Some scholars once thought that the numeral-signs for 1 through 10 derived
acrophonically from the Kikakui signs for the first syllables of the numeral words
for 1 through 10 (Tuchscherer 1996: 130132). While the syllabic values and the
numeral-signs correspond, Tuchscherer (1996: 140142) has demonstrated that the
Mende numeral-signs (at least those for 1 through 5) are also similar to certain
signs (and variants) of the Arabic positional numeral-signs. From this, he argues
that the Arabic numerals inspired the signs of the Kikakui syllabary for the first
syllables of number words. While the similarities are not striking enough to prove
the case conclusively, I am reasonably convinced that the Arabic positional numerals influenced the development of the Mende system. Yet the Mende numerals are
multiplicative-additive, not ciphered-positional (like the Arabic positional system)
or ciphered-additive (like the Arabic abjad-based system). The only other multiplicative-additive system used in West Africa is the earliest Bamum system, but it

Miscellaneous Systems

325

is a long way from Sierra Leone to Cameroon, and by the time the Mende system
was developed in 1921, the Bamum had switched to ciphered-positional numerals.
Moreover, the use of two different signs for 10 (one additive, one multiplicative)
and the use of repeated strokes to indicate exponents are features that are not
attested elsewhere. Thus, the structure of the Mende system should be regarded as
largely indigenous. Curiously, the modern Mende lexical numerals are not decimal but vigesimal. While this might suggest that the base of the Mende numerical
notation was borrowed from the Arabic numerals, in the nineteenth century the
Mende had decimal lexical numerals (Tuchscherer 1996: 148150). If this system
survived (in even a vestigial form) into the first decades of the twentieth century,
it, rather than a foreign numerical notation system, could have inspired the decimal base of the system.
The Mende numerals were used for a wide variety of functions, and were taught
in schools throughout the 1920s and 1930s. Some individuals used the system for
accounting and record keeping, but it is not clear whether the numerals themselves were used directly for arithmetic (Tuchscherer 1996: 69). Dalby reports that
the syllabary was used by some weavers and carpenters for recording measurements, which would presumably also require numerals (Dalby 1967: 21). Both
the Kikakui syllabary and the numerals continue to be used for some purposes,
including correspondence, record keeping, religious writings, and legal documents
(Tuchscherer 2005: 478).

Sub-Saharan Decimal-Positional
In addition to the African systems just described, which are structurally distinct
from their ancestors, several of the indigenous scripts of sub-Saharan Africa have
decimal and ciphered-positional numerical notation systems, and are thus structurally identical to their Western or Arabic ancestors. While these systems are of
less interest from a structural point of view, they are noteworthy from a historical
perspective. I list these systems in Table 10.7.
The Bagam syllabary was invented early in the twentieth century in western Cameroon and used briefly by the Eghap (known in scholarly literature as the Bagam) of
that region (Tuchscherer 1999). The only text to preserve Bagam writing and numerals is a recently discovered 1917 description of the system by a British colonial military officer, Captain L. W. G. Malcolm. The numerals probably were borrowed from
the Bamum system rather than from the Western numerals, based on some graphic
resemblances between the Bamum and Bagam sign sets. The Bagam numerals do
not include a sign for zero, but do include a sign for 10. It is thus unclear whether it
was a ciphered-positional system or how (if at all) it expressed larger numbers. In the
early part of the century, the Bamum system was still multiplicative-additive, which

Numerical Notation

326

Table 10.7. Decimal systems of sub-Saharan Africa


1

Manding

q
L

A
V

r
M

B
W

s
N

C
X

t
O

D
Y

u
P

E
Z

V
Q

F
,

w
R

G
.

x
S

H
/

y
T

I
<

Osmaniya

<

>

Wolof

_ =

Bagam
Bt
Fula (Dita)
Fula (Adama Ba)
Kpelle

10

z
U

J
>

suggests that the Bagam system may also have had this structure. The Bagam script
and numerals are now extinct, and recent ethnographic investigations in the region
have revealed no knowledge of the numerals even among elderly Bagam (Konrad
Tuchscherer, personal communication).
The Bt numerals were invented in late 1957 or early 1958 by Frdric BrulyBouabr, a native Bt from the western part of Ivory Coast, to accompany a
syllabary of over 400 characters that he had invented a year earlier (Monod 1958).
Bruly-Bouabr, who was fully literate in French, did not use Western models in
developing his script-signs, as can be seen from the abstract nature of the numerals.
However, the use of a dot for zero shows at least some influence from the Western
numerals (or perhaps the Arabic numerals, although it is not clear whether BrulyBouabr knew Arabic at all). The unusual sign for 10 may have been used multiplicatively or additively in conjunction with the unit-signs. There is evidence of a
quinary component to the Bt system in the fact that the signs for 6 through 10
are inverted forms of the signs for 1 through 5, with the exception of the extra dot
atop the sign for 5 (Monod 1958: 437). Bruly-Bouabrs efforts to have this system
accepted among the Bt met with minimal success. I do not know whether it is
still used at present.
Two alphabets invented for the Fula of Mali have accompanying cipheredpositional decimal numerical notation systems. The first of these, known as Dita,
was developed by Oumar Dembl between 1958 and 1966; in keeping with his
being a woodworker, his signs have a linear character (Dalby 1969: 168173).
Dembl attended a Koranic school and spoke French, so the structure of the
system was based on either Western or Arabic numerals. The second system,
invented by Adama Ba, a Fula Muslim literate in French, before 1964, is identical

Miscellaneous Systems

327

in structure, but its signs are more curvilinear and perhaps show some influence
from Western numerals (Dalby 1969: 173174). Neither system was ever used
except by its inventor.
The Kpelle numerals were developed in the 1930s by Gbili, a paramount chief of
the Kpelle in central Liberia, in conjunction with an indigenous syllabary (Stone
1990). Its numeral-signs include a sign for 10 but none for zero, so it is not clear
how, if at all, higher numbers were written. Both Arabic and Western numerals
were known in the region, and the Kpelle script was developed on the basis of
the Vai script, which used Western numerals. Although the Kpelle signs vaguely
resemble both Arabic and Western numerals, no definite historical ancestry can
be assigned to them. The script was used traditionally for tax records as well as for
official communication among chiefs, and was restricted to a small segment of the
populace. Today, most Kpelle use Western numerals, although the indigenous system continues to be used for personal correspondence by a few individuals (Stone
1990: 141; Tuchscherer 2005: 478).
A set of numerals was developed around 1950 by Souleymane Kant, an educated trader who was literate in both French and Arabic, in conjunction with an
alphabet known as Nko (Dalby 1969: 162165). It was designed for use among the
many peoples whose dialects fall under the label Manding, most notably Mandinka, and was intended to provide a means of communication accessible without
the need for formal schooling. The numerals are ciphered-positional and decimal,
and perhaps are related graphically to the Western numerals; however, numeralphrases are written with the highest power on the right. Texts written in this script
apparently included treatises on calculation, suggesting that the numerals may
have been used for arithmetic (Dalby 1969: 163). Nko continues to be used today,
and probably has tens of thousands of users.
Around 1920, an alphabetic non-Arabic script known as Osmaniya (also known
as far soomaali and cismaanya) was developed by Ismaan Yuusuf Kenadiid, brother
of the sultan of Obbia, as an alternative to Arabic for writing the Somali language
(Lewis 1958: 140142). The Osmaniya decimal ciphered-positional numerals, like
the script, were written from left to right. The fact that the script fully represented vowel sounds and was written from left to right shows influence from the
Latin alphabet, so it is possible that the numerals were mainly of Western rather
than Arabic origin, but the Osmaniya numeral-signs resemble neither Western
nor Arabic positional numerals. While Osmaniya was declared an official script in
Somalia starting in 1961, a Latin-derived orthography was adopted in 1972, after
which Osmaniya was used far less regularly.
Assane Faye developed a Wolof script around 1961 that has a set of cipheredpositional numerals (Dalby 1969: 165168). Faye, who was literate in both French
and Arabic, presumably drew more influence from the Western numerals in

328

Numerical Notation

creating this system, whose signs more closely resemble Western than Arabic
numerals. Numeral-phrases were written from left to right. Curiously, Faye also
assigned numerical values to nineteen of the letters of his script (19, 1090, 100)
in imitation of the ciphered-additive Arabic abjad system (Dalby 1969: 167168).
Neither the script nor the numerals survives today; most Wolof use either Arabic
or Western numerals.

Miscellaneous West African


While some African numerical notation systems (e.g., Mende, Bamum, Oberi
Okaime) are structurally distinct from the Western and Arabic numerals, these
systems probably would not have developed without contact with the West.
Most of the prominent pre-colonial West African states, including the Yoruba
and Benin civilizations, did not use numerical notation per se, although they
were, like most other West African societies, quite numerate and capable of
complex calculations using the cowrie currency ubiquitous to the region. At the
same time, however, there is suggestive evidence that some pre-colonial West
Africans occasionally used numerical notation. Unfortunately, we have only a
handful of ethnographic details pertaining to the peoples of West Africa in the
twentieth century concerning systems that may be considerably older. While we
should not assume that these systems are of entirely indigenous origin, given
extensive pre-colonial contact with Muslim traders from the north, neither
should we discount the possibility. Historians of mathematics interested in African capabilities have not discussed these systems, doubtless because they were
unaware of them (Zaslavsky 1973, Gerdes 1994). Because they are not attached
to phonetic scripts, they have not been compared to other numerical notation
systems and are often grouped inappropriately with unstructured tallying signs.
I expect that a more thorough search of the ethnographic literature (especially
from the early twentieth century) would reveal additional numerical notation
systems.
A. S. Judd (1917), reporting on the state of education in Nigeria, reported that
the Munshi (the Tiv, speakers of a Niger-Congo language in central Nigeria)
employed a numerical notation system. This system, which has a thin line representing the units, a circle the tens, and a broad line made by the thumb representing a score, was apparently used when drawing in sand or earth (Judd 1917: 5).
Presuming that Judds description is accurate, this system was most likely cumulative-additive with a base of 20 and a sub-base of 10.
The tradition of graphic symbolism practiced by the Dogon of Mali in rock
paintings and sand drawings includes numerical signs that can be combined with
one another. In one system, straight lines represent units and circles represent 5;

Miscellaneous Systems

329

a drawing of a man with four circles (each representing one of the limbs with five
digits) joined with a cross means 22 (Griaule and Dieterlen 1951: 1112; Flam 1976:
37). Another represents a period of sixty years by three rods of decreasing size, each
with the value of 20 (Griaule and Dieterlen 1951: 28). There may not have been a
regular system of correspondences between numbers and signs. In the context of
reckoning and calculation, cowries representing 1, 5, 10, 20, 40, and 80 apparently
were used (Calame-Griaule 1986: 232). The exact technique employed is unknown,
however, and this may not have constituted a numerical notation system.
While most systems of tally sticks use only one-to-one correspondence (thus
lacking a base), Lagercrantz (1973: 572) reports that among the Ganda and Djaga
of Uganda and Tanzania, tally sticks are also used in which units are marked by
small notches, 10 by a larger notch and 100 by an even larger notch. It is not clear
whether this system recorded cardinal numbers, or whether it is simply a series
of marks equal to the number being counted, of which the tenth is large and the
hundredth larger still. Another tallying system, possibly of more modern origin,
was used on riverboats along the Ogowe River in Gabon in the twentieth century.
When refueling steamships, a stroke on a piece of paper was written for every
ten loads, and a cross for every hundred loads (Lagercrantz 1970: 52). Again, it is
entirely possible that this system was not used to indicate cardinal numbers, but was
simply an ordinal tally. The conceptual distinction between a system used only to
mark items as they are counted, and one used to indicate whole sums after counting should not be underestimated; nevertheless, it is quite plausible that at least
some of these African tallying systems did, eventually, transform into cumulativeadditive numerical notation systems. Regardless, even tallying systems that use
specific abstract signs for powers of a base instead of one-to-one correspondence
represent a considerable conceptual advance.

Cypriot Tallies
Buxton (1920: 190) describes an otherwise undocumented numerical notation system used by nonliterate Greek speakers in Cyprus:
The numbers are continually used as follows: a perpendicular stands for a unit, five is
sometimes indicated by a cross and sometimes by a circle, ten either by a circle, by a
theta, or by a cross inside a circle, twenty by a cross inside a circle, where that symbol
has not already been utilized previously; if it has, there seems to be no alternative.
Fifty is written by a loop on top of a perpendicular, and a hundred by two fifties. It
will be seen that two of these symbols are not dissimilar to Arabic numerals, namely,
the circle and the symbol for fifty. The Arabic symbol for five is, however, not circular, and it is possible that the two signs are connected, but the value of the looped
line is in Arabic nine, not fifty. ... At Enkomi a man scores at cards in this way. He

330

Numerical Notation

chalks down units up to four, then he rubs them out and writes a circle, adds units
to ten when he erases them, and draws a line through the circle, draws units up to
fourteen then adds a circle; at twenty he erases the added nine and draws another
line through the theta, which thus becomes a circle with a cross through it.

This evidence indicates that although numeral-phrases are constructed sequentially as tallies, rather than being written as a single sum, because intermediate values are erased and replaced with higher values, ultimately the result is a cumulativeadditive numeral-phrase with a base of 10, a sub-base of 5, and a special sign for 20.
This sort of notation is qualitatively different from simple one-to-one correspondence,
or tallying in which intermediate marks are not erased but simply continue on (e.g.,
XXVII vs. IIIIVIIIIXIIIIVIIIIXIIIIVII as two notations of 27). Other than this one
brief description, however, we have no information on the origin, history, or use of the
Cypriot system. While there are parallels between this system and the decimal cumulative-additive system used in ancient Cyprus (Chapter 2), there is no reason to think
that this system is anything but a locally developed technique, one that is evidently
idiosyncratic given the multiple numeral-signs used for 5, 10, and 20.

Indus
The writing system of the Harappan civilization, centered in the Indus River
valley, is one of the great remaining mysteries in the field of script decipherment. It
was used from around 2500 bc to 1900 bc on several thousand very short inscriptions (averaging five signs per text), and was written primarily from left to right
(Parpola 1996). Unfortunately, there is no reliable basis for deciphering the script,
because the language it represents is unknown (though sometimes asserted to be a
Dravidian language) and there are no bilingual inscriptions. The situation is even
more grave than for scripts such as Linear A, where there are many easily readable
numeral-phrases and associated ideograms (see Chapter 2). Many dubious interpretations of Indus numeration have been proposed (e.g., Subbarayappa 1996). We have
barely enough evidence to confirm the existence of a numerical notation system in
the ancient Indus Valley, much less determine its origin, history, or function.
There have been several earnest attempts to decipher the Indus numerals,
mostly relying on the very frequent occurrence of groupings of vertical strokes
on the inscriptions. Table 10.8 shows these numerals as well as the frequency with
which they are encountered in the texts (Fairservis 1992: 62).4
4

Fairservis (1992: 183) provides no count of single and double short strokes because these
are also assigned grammatical functions (as genitive and locative case markers, respectively) in his decipherment.

Miscellaneous Systems

331

Table 10.8. Short and long Indus strokes and frequencies


1
Short strokes

Long strokes

10

111
111

1111
\111

1111
1111

11111
\1111

11111
11111

38

70

11

111

1111

111
\11

151

70

38

aa

aaa

aaaa

aaaaa

aaaaaa aaaaaaa

149

365

314

64

22

These signs probably represent low numbers in a cumulative fashion; the short
strokes are grouped into sets of three, four, or five, just as the signs of most other
cumulative systems. The longer ungrouped vertical strokes occur only in the early
Indus inscriptions; during its mature phase, the shorter strokes were used exclusively (Parpola 1994: 82). Because these sets of strokes are paired interchangeably
with non-numerical graphemes (e.g., the fish sign + is attested in combination
with three, four, six, and seven strokes), we are relatively confident that they were
numeral-signs (Parpola 1994: 81). Yet Ross (1938) long ago pointed out that some
groupings of vertical strokes pair noninterchangeably with other signs, which suggests that they may have had phonetic or grammatical values (Ross 1938; Fairservis
1992: 12). This is parallel to the frequent use of numeral-signs phonetically in
Chinese writing, and resembles abbreviations such as K-9 for canine in English.
One enigmatic symbol consisting of three rows of four vertical strokes occurs
frequently, but never in the same contexts as other putative numerals; Fairservis
(1992: 71) argues that it should be read as rain, which may or may not be correct,
but is far more likely than 12. The Indus texts are so short and devoid of contextual information that we must be very careful not to read too much numerical
information into them.
This interpretive framework for the Indus numerals does little to establish
whether this system had a base and used an interexponential principle to write
larger numbers. Fairservis notes that there is a sharp drop-off in frequency after
seven for both the long and short vertical strokes, and that in fact there are no
attested instances of eight or more long strokes. From this, he concludes that the
Indus numerals were probably octal or base-8 (Fairservis 1992: 6162). Perplexingly,
however, he then proceeds to assert that there are pictographic signs for 8, 9, 10,
and 11 that were simultaneously numerical and calendrical, indicating the eighth
through eleventh months of the conjectural Harappan calendar, because these four
signs, along with vertical strokes for 1 through 7, are found in association with a
sign that he thinks represents month (Fairservis 1992: 65). This theory has not been
widely adopted by scholars of the Indus script (cf. Pettersson 1999: 103).

332

Numerical Notation

Figure 10.4. Inscription on artifact DK-7535 from


Mohenjo-daro.

Our best evidence for a legitimate Indus numerical notation system comes from
nine inscribed potsherds and copper and bronze tools found at Mohenjo-daro, Canhujo-daro, and Kalibangan, inscribed with sets of vertical strokes and crescents/
hooks (and sometimes other script-signs). These two signs are sometimes found in
combination on the seal inscriptions, but never in large quantities and never clearly
separated from the rest of the text. Pettersson (1999) adds that, in addition to vertical
strokes and crescents, a distinction needs to be made between vertically and horizontally oriented strokes. One object, a chisel or axe blade (DK-7535)5 from Mohenjodaro, contains all three signs, as shown in Figure 10.4 (Parpola 1994: 108).
While I am reasonably convinced that this inscription and similar ones on other
Harappan tools are numerical in function, there is no agreement as to the specific
structure and value of the signs. Fairservis (1992: 6769) has constructed a convoluted argument whereby the vertical strokes (standing for units) can serve either an
additive or multiplicative role in the numeral-phrase depending on whether they
follow or precede the crescent sign(s). Pettersson (1999: 102103) points out, however, that there is no case where a crescent sign is both preceded by and followed by
vertical strokes. Fairservis (1992) and Pettersson (1999) argue that because none of
these nine objects contains more than seven of any sign, the Indus numerals must
be octal rather than decimal. Yet Parpola (1994: 82) argues that the crescents probably represent 10 rather than 8. Either of these interpretations of the system would
mean that the Indus numerical notation system was cumulative-additive.
At present, there is insufficient evidence to decide whether the crescent-sign had
a value of 8 or 10. Nine numeral-phrases is a very limited corpus from which to
conclude that, since no sign is repeated more than seven times, the numerical base

There is some confusion over the identification of this object, which is assigned different
artifact numbers by Parpola (1994) and Pettersson (1999).

Miscellaneous Systems

333

must be 8. The limited linguistic reconstructions regarding Proto-Dravidian (even


granting the controversial hypothesis that the Harappan language was a Dravidian tongue) are ambiguous and tenuous, but on balance best support a decimal
interpretation, since Proto-Dravidian lexical numerals for ten and hundred have
been reconstructed (Parpola 1994: 169). The linear measures of the Harappans
appear to have been decimal (Sarton 1936a), and the system of weights is partly
decimal and partly binary (Parpola 1994: 169; Pettersson 1999: 106). None of the
Harappan weights bear any inscription, numerical or otherwise (Pettersson 1999:
91). Petterssons (1999) attempt to correlate the numerical signs on the metal tools
with their weights showed only that no metrological interpretation of their meaning (either decimal or octal) was likely to be correct.
The Indus numeral-signs are entirely unlike the Sumerian numerals (Chapter 7)
used in Mesopotamia at the time of the invention of the Indus script. It is interesting that the Egyptian hieroglyphic numeral-signs for 1 and 10 are | and ,
respectively, but, even if the Indus crescent-sign represented 10, this similarity
could very easily have arisen by chance. While there are vague similarities in the
metrological systems of Egypt and the Indus Valley, there was minimal cultural
contact between the two regions (Petruso 1981). It is probably best to assume that
the Indus numerals were independently invented.
The fact that the Indus script is completely undeciphered, coupled with the limited number of surviving numeral-phrases, makes it nearly impossible to identify
the function(s) for which they were used. There is no evidence of the numerals
use for accounting or administration, which is abundant for other undeciphered
scripts, such as Linear A and Proto-Elamite. The wide variety of materials on which
numerals are found (clay seals, potsherds, metal tools) suggests that they were used
widely among literate Harappans, but even this hypothesis requires caution. The
Harappan civilization declined precipitously after 1900 bc, although it may have
survived in certain regions for a century or two longer. There is no evidence that
the Indus numerals had any influence on the Brhm numerals (Chapter 6), which
arose almost 1,500 years later.

Naxi
Naxi (also known as Nakhi and Moso) is a Tibeto-Burman language spoken
by approximately 250,000 people in the northwestern part of Yunnan province
in southwestern China. Naxi is written in three indigenous scripts: dongba (or
tomba), a pictographic notation system, and two syllabaries (geba and malimasa),
in addition to a more recent Latin-based orthography. The dongba script is highly
idiosyncratic, consisting of 1,5002,000 largely pictographic signs with some phonetic components, although there is not a regular correspondence of signs with

334

Numerical Notation

either words or phonemes. It is used primarily as a mnemonic aid or promptbook to assist priests in reciting memorized texts (Bockman 1989: 155). It was
reputedly invented in the twelfth century ad (Coulmas 1996: 353). The earliest datable dongba texts, however, are from the middle of the eighteenth century (Bockman 1989: 153). The dongba script is actively used by some Naxi priests, and has
even been part of modern literacy programs. Nevertheless, because dongba texts
are pictographic and rely on oral and mental knowledge to draw meaning from
them, their interpretation by Western scholars is incomplete and poor.
In at least some dongba texts, numerical notation was used alongside the
script. In the Nichols manuscript first made available to Western scholarship by
F. H. Nichols in 1904, three repeated South Asian swastika-like signs precede
six vertical strokes. These are interpreted by Rock (1937: 236) as representing 100
and 10, respectively, producing a sum of 360, indicating the 360 yu-ma deities
of the Naxi. Groups of three or more signs are clustered in rows of three signs,
where appropriate. In other dongba manuscripts, a simple cross rather than a swastika represents 100. This is a cumulative-additive numerical notation system with
a base of 10, leaving open the question of how the number 1 was represented.
Bockman (1989: 1952) suggests that the dongba signs X, +, and were numerical,
but does not assign them specific values. In other dongba manuscripts, however,
vertical strokes or hooked vertical strokes mean 1, and X or + means 10; in one
very clear instance, Hs. Or. Sim. 279 / R. 1912, Blatt 9r 10r, eighteen consecutive
panels depict gods, each enumerated using this system (Janert and Janert 1993:
2753). Like many cumulative-additive systems, the units 6 through 9 are represented in two or more rows of three to five strokes (3 + 3, 4 + 3, 4 + 4, 3 + 3 + 3); 5
is depicted with a single row of five strokes, but 15 is depicted with X (10) followed
by 5 indicated in two lines of three and two strokes respectively.
The variability in signs suggests that base-structured numerical notation was
used only irregularly or idiosyncratically in the dongba texts, with local or even
individual scribal tradition determining which signs represented which numbers.
Against this position, however, a wide variety of dongba texts contain numeralsigns, and all appear to be cumulative-additive and decimal (i.e., signs are repeated,
but no sign is repeated more than nine times). The origin of this numerical notation system, and its relation to any other system, remain obscure. It is possible that
it was independently invented, as no other cumulative-additive systems were ever
used alongside scripts of either East Asian or South Asian origin.

Varang Kshiti
In the twentieth century, several scripts were developed for the various Munda
languages of central and eastern India, of which Sorang Sompeng, Ol