Professional Documents
Culture Documents
AND
THE
THEORY
OF
INFORMATION
) (r,
'
ABSTRACT
of communication, in terms the is defined information of the concept of transmittediover a messages sets of of statistics has This no relation concept channel. communication to the logical concerned. of the messages content The concept defined closely sentences to analogous for of the of can be content of logical language in logical a manner a technical of information concept
In
the
technical
theory
in place that except probability of logical of "information concept with analogy, of "correlation" "confirmation" affinities or of one
probability a concept empirical The be related substituted. must has a similar logical transfer" both to the logical measures and to by another. measures of
"dependence" sentence
concerned "sentences"
the logical of all concepts by applying to not merely language as ordinarily understood
but
sentences" thesis.
the
or "questions"
in
of "information
analogous
concepts
indicates
basis for a as can serve that the logical concepts as the more fundamental. the with ordinary concept uses this of words by conclusion the
definitions
of
the to both
former,
and
are closer
to usage.
,,
iv)
page 1.
2.
Introduction
The technical theory of communication
1 18 to the information 76
115 144 170
and measurement 3.
4. 5. 6.
allied
Information Language
and information
References
188
1.
INTRODUCTION
our illustrated
six
It
pages
is
of information
it its'unit
no novelty,
seems, is
that for
"information" practical
a measurable
quantity;
purposes
the
page,
the
line,
the
chapter,
the perhaps,
paragraph is and as
in books, example,
for
contain
books
aim
Speculative
information, is not of
philosophy
and always specific is directories found even
the
a table Information
useful
excellence
beginning in
and under
"Today's
arrangements"
newspaper.
recently,
"information"
contexts
can be measured: in
I
A
sense.
Information,
it
would
has
of quality precision in
although-we still
may find
stands,
we cannot
afford
It
and criticise of information. in
is
the
purpose
of
this
thesis
known found
to
as the
set
out
a, theory It fields is
which
theory
a theory as widely
has
ready
separated
be ordinary
engineer.
theory It is
a scientific
the
theory
of relativity
scientific or
not
to be confirmed no "crucial
by experiment.
experiment" is no,
which, can demonstrate "classical" obsolete,; to goals The sort the theory except previously of theory
or false, replace
which
will as it
may suggest
cute routes.
only it
by more devious is
rather of the
"purely
mathematical" theory
theory of classes
theory
matrices
or the
(particularly)
3
theory analogies And perhaps to liken it to it is
of
not
stretching theory
a philosophical
like
Plato'stheory,
of forms.
The aik. "of theories of this' of second empirical framereformulations in time new stands class
is
not
the bt
or
systematisation of
facts works; of
provision
they
old'-ones,
which of
nos
may be expressible.
theory
by its
empirical
but
this
sort it
or falls
by the
it-provides
a convenience This
and expression.
that
theories
of the give
or do not
facts. of
Neither
him that
of this the
d$scription
of facts
making their
The other
side
invention
of the
concept in terms in
of force, of the
redescription space,
of dynamics ---
curtature
and so forth
does not
\4
itself involve an appeal out to facts, and could Thus for theory in principle much
separately.
example
framework
of the
of relativity
many years
equations. an n-dimensional put quite fruitful forward, in, spite physical
clothed
relativity in
Riemann's manifold.
when it ,
significant it had, of
and important
application. to, the led large" But minor to of the its to geometry
applications and it
figures about
surfaces:, of
physics.
spaces
positive of
finite" the
spaces, descriptive
power
mathematics
and physics
fruit
later
in
facilitating
of relativity
In other
of
planetary
motion
for
example,
accurate the
description invention
of
the
phenomena
preceded
be the
to
build. will
But usually in
which
is
while
a double, field If
Newton'
force also
to
the
of-planetary; development
a framowork
departments
of physical
'theory.
in terms in use. seems to abstract of
undefined its
rules of
proscribing its
no problem the
arise;
second,
concept
until
an interpretation
it for is the,
is
put
upon it.
that
As a point
these two
of
no doubt -introduction
desirable of
a concept At
should the
be
distinguished
'and kpt
distinct.
same
time,
it
here
that
the not
difference so great
between as it at
them is
first
very
often
appears.
for terms
this.
of which
concept
contexts,
undefined or use of on
defined
The chain
on which
hung either
of undefined
"primitive" ordinary
concepts language,
everyday to accept
pegs of as fixed
either concerned
appended afresh
be raised
the ----it
circumstances
to demonstrate independent
of the older
new concept
has as it claim
right
existence,
an equal
inhabitants.
Something
In connection originally some, of it their. with the
like
theory
this
of
happened
"matrices". of ordinary the
in mathematics,
Matrices numbers; order
defined
as arrays
properties -: to
so upset consider in in
accepted
became
fashionalbe
them
suitable of
definitions arithmetic of
elementary
that'of certain
the
commutativity could
matrices
be used
rotations
considered of a. class
matrices of
complex and it
elements was
was coextensive
quaternions;
later
shown
the that
algebra'of of in a class
complex of
matrices
such
a complicated connections
trace that
causal the
seen
the, basis
Russell;
and also
Eddington
by abstract possible
methods,
later One
matrix
interpretation.
is
tempted
that
real
numbers are
whose elements
caso,
it
may well
be
said the
that tail
the
"abstract"
procedure
practice
somewhat clouds It is
the
distinction to express,
two procedures.
more difficult
to is
it
is
that
even in
when a concept an interpreted may well that not the legitimare in Thus to say not the the two
mental particular
and'even, they
which in
that aro
contexts,. indifferent we shall a limited the concept B when point: have in the any one. need set
fields; to of set
be able
abstraction properties
a concept which
concept
the, present
presuppose
possibility definition
subsequent of it will
generalisation; be at best
we give
will in
Newton
terms
required
due to
seem inconsistent
with
this
suggests sorts of
that force. to
there It
are
at
least to
in later
establish (e. g.
there
phenomena of the
permitted
concept;
but in
possibility concept
Newton's do not
to force
come.
chose causes
to
conceive
"something"
gives air;
a straightforwardly alisations" had been which for'the reason been have but defined
exists
moment to
duo to the
contact
suppose
course
would simply
different; added
their first
poledefined
strength, in a narrow
or
....
"etc. this
context,
broader definition
definition, precludes
obvious
as soon as it in
is
said..
however, a form
can be put it
it'is
in which It is
will this.
command universal
agreement.
What is
more use
important is the
about
than it
its is
terminology
ordinary penumbra contexts the the if its either named". speaking, sometimes cases
language. of to
a word connected
like with
associations express
use
in
moral'or in
physical ordinary
scientific use it
normal that
be inclined or thatit
re-name is all
simply
re-name a good
be to concept
the'c'earch
dofinition
properly
called
by the
original
"To
name.
with the same example: as a sort and if attempt of to this limit is the
continue
"force" of that to
long
definition to that
least sorts
other
word,
must even in
' .. __ be it
confirmed
or strongly
disconfirmed
hypotheses
--
covering
77
these only
other possible
contingencies. definition
(cf. of
psychical which
forces).. is
The
"force"
consistent
with
this
requirement motion) it
is causes;
one in
terms
of
the motion
or a similar
i. e. Newton's,,.
could who has ''a
t
use
word
dynamics, feeling
forces, it is
studying
current
definition.
Such an account, and we must not Where there example, the is overlook an elaborate will
is
oversimplified, in the
Y
situation. for
formalism,
theorist
be driven is-no-doubt
of the
momentum of his
in some of the But of
mathematics:
more even "abstract" the technical have of
case
science class
"usages", people.
restricted
points the
are
important of
hero
only of
in
a principle
methodolygy In so far or
scientific
be admitted facts,
them, press
them
study
the
of
his
conceptual would
He must "logic"
logicians everyday
concepts
linguistic
This
that
he
flout
usage,
did
or of
not
change
from with
usage
an apprediatian will
a departure to
be self-stultifing.. and it
misunderstanding,
may be retrograde
in
It
that
is
it
loads
research
function of this
in wrong
of sort;
or unproductive
criticism it should
directions..
to be added,
an important mistakes
scientific though
correct
that
ibility
the
scientific
to avoid
critic
such
has himself
a special
respons-
mistakes.
Historical
necessity occurring of for some such example
examples
principle wherever it
in
support
sufficiently possible
of the
abundant, to. say that
are is
unwieldy
terminology --"the
levelled calculus
Newton
respect
notation
___- `- 1
e. g. at
ref.
37, Peirce 2, of
P" 39); in p.
at
(ref. his
7, work
'15);, P. in symbolic
'
C. S. (ref. theory
connection 519); at
logic his
in (ref.
with
physical
fundamentals
In the
been P.
field
of scientific
Berkeley
criticism
(eee e. g.
they
'ref.
have in
44, p.
particular
levelled'at
87 and.
to
be wise
after
event; without
procedure
situations be insisted
concerned. upon in
cases
these.
it
is'possible
to critic choices
indicate
the
kind
or scientific procedural
He must of this ask
of difficulty.
of current are usage retained
"What
features these
word? ",
and
"Which
features
in
the
formal
thoory?
";
a serious
appears in the or
either or
theory, alter
(4)
way,
last
the
formal
conform,
points
resort
these
a more chapter
detailed of this
principles;
"1
6)
asa
practical
essay
in
the
requisite
critical
of
science
concepts; of
probability.
of that. the
the
we can present
afford one;
for least
granted two
a, case
quite
interpretations
probability
cal: culus,
as : Jell uses
with ,
as a plethora word
over in connection
1 i
_
of the
induction,
"probable"
confirmation
all`theseproblems except
r a i
naturally
sacrosanct.
one of
tasks
adequacy
the the formal
of
current
conditions of of
definitions,
which
given
we should and
on the
wish to other less two
one hand
impose hand well a founded) may on
concept set
on the (albeit
similar for in
that
the
concepts we must
be linked,
be prepared
to
treat
the
relationship,
from
a logical
point
of vier,
as a reciprocal light
all
one. fields.
Ideally
we should
hope to
shed-some
Not
in
of
both
the
applications
of
the
"informin.
ation"-concept
are'of
a class.
Wo 'must., distinguish
particular
theory the
between
and the
(for
want
of better
terms)
the
"logical"
:
"technical"
theory,.
"logical'
content", by
as first
defined:
by'Carnap
other The
forward
by Hartley
one of
analogy
- which' has
between'
these,
an analogywhich
imperfectly Subsidiary'
of
the
those-of context
as applied
concept to
receiver of a full
the
development
the
logical
concepts will
11
certain
naturally
of the
not
requisite
in terms
logical
of
concepts
most
language call
as usually "unsigned
conceived, sentences"
we shall.
"questions".
Chapter
of with thiss relevant featuroc of
2'is
an introductory
theory. in
presentation
A comparison 3, and of
chapter
introduction-of
the-Logic
"questions"-
same concept
appears raised
chapter
for.
"information" at of least
with
the
physical their
"entropy"
and we must
adjudicate
claims.
Thertremains, point logical simple requiring, concepts mathematical systems a "message", languages. discussion. are defined models with of however, Both the one outstanding technical of relatively former for on what artificial in more and
in-terms reality,
the
Questions
application
general This In
contexts situation id at
are of
not
necessarily typical it of in is
course
the
case that
issue,
however,
too
assumed intrinsic
such
extensions and
application
difficulty,
particular
languages croatod at
In
(such for in
into
the (or
pattern, both)
"nessages" principle
languages
leant
this
of known techniques.
40) on tho by auch
connection of (ref.
"redundancy" Mandelbrot
English, both of
involve in
a presupposition, liable to
extremely case
in chapter
" etc.
attempted
chapter
It thesis which in
will
"language" it is
entere
this
information
carried, language in
language
logical
and secondly of it
that
is
language,
and we to
its
success
2.
is
necessary
because It of
largely (ref.
work
Shannon
an outline original of
presented, or or to for
be made nature
the
sources theorems
proofs special
detailed refs.
topics.
(See
especially
t#6).
applications "continuous" of
of the signals
theory as distinct
involve signals
(typically)
will however signals, is here less
telegraphy.
not and dealt than extend
theory continuous
somewhat justice.
summarily, Shannon's
perhaps of
information
"subject
a fidelity touched at
criterion" all.
(ref.
38 or 39,
ch"
5) is is
not also
7777
r 19
to
the
"theory for
of us
of in
Woodward certain
(ref.
46),
interest raises.
logical
We conclude between of
a brief
analogy
and the of
applidation
Maxwell's
demon in
thermodynamics
by }3rillouin
5' 6).
Certain
theory deferred.
7
results
into
connected
later
with
the
and
technical,
are
fit
more
naturally
chapters,
of certain
the
theory systems of
of
inform-
telegraphy: at of
a system is
following: with
a battery of multiple of
a number switch,
a number telegraph
different wires;
end of
a pair is at
receiving of
positions with
some set
such at
letters
voltmeter marked'in at
receiving
appropriate transmitting to
around
To each voltage, is
symbol which to
the is
turned
it;
voltage,
received
Zo
at
the meter,
sets, the
meter
needle
to
to point
over
to the.
this system
symbol.
We-imagine
messages
be sent
as sequences 2.11
be'a This maximum is
of the
available
symbols. reasons
such of
For various
speed only system time to at which
there
a device
so not of the
because
each the
telegraph
one another,
also limits
W4ys.
on the
power available
from they
the
battery.
These in various
have some
are-not-absolute;
Nevertheless any
may be extended
system will
practical
such limit,
imately quite second 2.12 (--exactly that
and it
will
be possible
to
specify
in
approx-
see later,
howevor,
not
alone
tell
messages is the
over
Equally
available
elementary
number the
transmission, switch
(non-equivalent)
positions
transmitting
end,
and the
corresponding
number of
significantly
different
2i
at are
the the
Lot the
us
position
indicate at the
be 27 switch
positions
transmitter meter
"signalbi 2.13
27 positions We call
it known by
needle
receiver.
and denote is well
this
"n".
that than
messages two
can
with
not
more
employs "off";
wo might of
standard of
Morse
practice of these
regarded levels.
time-sequences one elementary one time-interval and two and symbols, additional If
two
occupies throe;
time-interval, is allowed
two a
symbols,
Morse
as a code we have
symbols.
"dot"
"dash" letter-space
10
1110 00
Z'Z i.
whence
for
example
the
letter
"A"s
or
"dot-'lath",
becomes:
10111000
and the letter "Q", or "dash-dash-dot-dash", becomes 1110111010111000.
2.14
sentation code aural is that arises
The complication
of for course ready from
in
the
this
fact
method
that
manual
transmission case,
and
connection of the
modern
teleprintor.
time-intervals. are
Given
allotted usedifor
sequences
are
purposes. type,
2.15
This
is
clearly
as economical
a code,
of its
-as could
be designed.
Notice, however, that whereas in our
original
system
(with of the
27 signallind; alphabet in
levels)
we could timewe
teleprinter
system per
letter
Morse,
average,
a realistic over
estimate
of
messages
can be sent
a channel
account
levels,
or
(in
sense) the
To
n alternatives. would
symbols place;
second
mean making
choices
of each previous
to making a single doubling of
single
choice the rate
choice,
between r is n. n is n to
and this
nxnn
is
equivalent
n2 altctnatives; sgunring
to
alternatives
number short
saying
maximum
transmission
of mensagea is signalling
the nm
proportional levels,
of the number n of
(characteristic of log
the
identity
logarithm mm log
function): n.
We call
capacity
the
of
maximum rate
the channel,
of transmission
and write it
so defined
the
Cr
2.17 measures is to that of with this two It remains quantity. signalling
log
to fix
n.
on a set the it of units for
simplest
system
express
characteristics
systems unit"
a "binary of information
amount
involved
a choice
between
two alternatives.
(We shall
later
require
..
z4
that
the
second" to
we specify 2; for
a two-level
logarithms to
to base loge
base 10).
logarithms
From logg
bare is
measures unit
changed equivalent
about
Hartley capacities
in of
1928
(ref.
23)
proposed to its
"a measure
cliffernnt and
systems suggested
measure
effectively
we have
ation
capacity"
in
the in
systems, signals
somewhat
generalised
are that would in , not
to covor
to
cases where
It could
confined this to
discrete
without have of
restriction
place
a choice the
we should
have
a"continuum.
_I
i. ,
2S
would will
be unrealistic, always
voltages
together; signalling
a finite
"code"
employed
voltages the
nay
ecqipment signals
equipment, by
etc.; origin
of molecular
depending the
radiation
surrounding
it.
2.22 riyquist
(Sve. ref.
36).
Drawing to some extent proved Q .nee on the that work the of
(ref. in (
Hartley rereral.
information
z)
of a communication
limitations and This is on the is a result
realistic on the
"noise"
interference,
which
in
context 2.6).
by Shannon,
(section
z6
In-the different
work
of
Hartley
systems
presupposes of
efficient of
coding
a message
fundamental limit
an upper
what of
circumstances transmission.
possible means of in
that at
give
"information" sort of
because of
does is
us what
transmission at the
a message rate.
be regarded
as transmission
this
question of the of
we must different of
return codes.
work
Shannon, of
leading inform-
transmission
necessarily
transmitting
Let
us reconsider
the
telegraphic
systems
introduced ignore
the
above.
We can for
the
r,
the number of
of a system
symbols
per
per
and express
capacity
symbol
logt
assume that
Z7,
second let
is
constant that
and the
sane
for
all
oyotenis). say, of
us imagine of
a message English)
(consisting, is to
passage any
ordinary
coding
required
will
trans-
as the
0 to the
channel
rather
system,
on the that
assumption the
determine levels is
signalling channel
the
by this
21 i. e.
per
alphabetical i. e. what
requires
a capacity extent
they
a suitable information,
transmission different
compare
2.34 Before
a few must
is
provided let
by Shannon. us notice
a definition
to
give
it,
about
however,
the form
points
such
t 7
the
place, conceived
since
the
channel
as a maximum expect general cases, message that in bits it the less it or actual than can be of be
we should in ideal of
be a quantity 13 zk in choice of
system. in the
course, e. g.
must per
as C, and most of
symbol.
only on the
message
by a simple less
example. efficient
comparison embodies it
code, not
gives
alphabet
most designed
frequently to transmit.
the That
messages this
(English)
gives
a considerable
saving
in
the
case it is
messages
is
sufficiently that
obvious,
though
of of the the
so obvious
lead
application
in the case here, least with
could code.
dependence
on the are
used. "E",
a preponderance as regards
letters
"A"
compare
.3
2y
time of
very
with
of
the etc.,
preponderating but in
"Z" the
used; are
cases of If the
the
comparison on the
systems
either
depend
messages.
depend in
on the the
characteristics i. e. on their
symbols
message
frequencies.
2.36'
depends that rate which the (a) is of of it of is on the should the
But thirdly,
message it
if
the
information
to
rate
require
be possible
message" out
carried
transmit in-two in
form is'a
(though
statistical in
be-related paid to
correlations
as well the
as to
single
symbols.
Consequently
information
3n
rate form
should of the
in
way for it
either remains
and in
unchanpediby to per 2.37 information , simpler Shannon messages symbols. given whether notion calls take
coding. of the
may of
be necessary of symbols
account before
different
average
number
second
Hence rate of
xlxxkzk
rate i. e.
an "information by the
a set
specified
by the
messages the
before
measure class of
defined coding
a certain contemplated
operations,
above.
It
merit satisfies in
to
have these
information It is
distinguish
between systems.
communication as above,
there elements
is or is
information
some quantity
which
can
take
any in in our
a continuoue is frequently of
range
of'values. conventitn.
one of
example
system,
27 signalling is essentially
levels
arbitrarily If in this
feature of view
the the
appropriate system is
once
"diacrote" e. g. at
other number
usable
the
symbols of the
We shall
treat of
the
theory
of
be discussed
section
suppose
we have
a set
E of mutually about
P(E3) We look
exclusive
possiblo except
of the
events their
E1, E2,
.....
En
which
(j for -
to know nothing
1,2, ... n) of
probabilities
in some situation. involved
occurrence amount of
a measure
"choice"
in
selecting
the
event
of
which
actually
occurs,
we are with
or in
respect
other
to
words the
a measure occurrence
in
Calling
a measure
H(E),
we observe
d J
3Z
that These
it
will are
be a function subject 0 to
of
the
probabilities
P(EJ).
for
each
j;
exhaustive,
have
-.
-Since of is the
the
events
are
mutually
exclusive E8Ek of
the two
Boolean or
"intersection"
zero,
P(E jEk) o2, "join" is the what amounts Ei v Ek, the P(E Jv to the i. e. same thing, the
0 the
(j
'
k) of or Ek, the
probability of Ej
sum of
individual Ek)
P(Ek)
(j Yj
= k). of
the
"complement"
1 -P(E3).
is
a possible
set
of
requirements
should
be a continuous
function
of
the
P(ES). If all the probabilities increasing likely events P(E ) are function there equal, of is the more
number
33
choice,
is
or uncertainty,
increased). 3. If is the choice down should determining into two the occurrence choices, average
of the
broken H(E)
successive the
be the
sum of to of
measures set
appropriate consisting E+l these El, ... chosen whose the v two, Ei
such the
be the v E,
two
and
...
v En,
choice el
giving ; if will
been
this
probabilities is from
summation these of
using the :a
11(02) these of
case
according the
their
probabilities; thus
respectively:
requirement H(E)
+
+
P(El
P(E,
v ...
+1 v ...
v Ei)
H(el)
H(02). measure other for terms the
terra
right-hand first of
side
choice, xmamul
average
measure
choice.
3t
Shannon II(E)
proves
(ref. these
39,
p.
19)
that
the
satisfying
three
requirements
-K
P(Ei) This
log
constant.
result
subsequent It K it the
may be made unity logarithm; with choosing A convenient 2, H(E) of then choice
by a suitable fixing
value merely to
K together to
amounts
measured in cf.
amount probable
involved
alternatives:
the
use
unit
Thus
we have
TIt )
2.46
the maximum
i
the
of
(bits).
fixed,
If
value
number n of alternatives
11(E) i. e. 1062 n occurs P(Ei) (bits), value is when 1/n all
probJ.
are
equal, 2.
each in
H(E)
and is of unity
H(E)
and the
no choice
or uncertainty.
35
An important in the
property case of
of
the
independent
specified more
be put
E a set/of
El, F1, (i
... ... 1,
Em Fn. ...
set
F of
EF is
equivalent of of F. If
choosing
events
independent of
those involved
we find EF is
amount the
choice of
the
sum of
amounts the
choice
F separately.
Thus P(EiF
whence
H(EF)
J)
PCEjF j ) P(Ei)P(P j)
) logt
+ logt
P(FJ))
a-zZ
P(E1) +
-Z J
P(F
36
it
is
convenient
to
introduce
on relative of the
probability define
event involved
"amount -
choice
E" as follows:
H(g,
From the
E)
formula
P(E1FJ) j
1002 P(Fjt
Ei).
probability P(EiP
-there the
follows formula
H(E)
H(F,
E).
we have simply
H(P).
later when we dome to consider
;.These "noisy"
formulde channels.
are
of
this system
measure
to
we have
we introduce to examine
the
system.
system
parts
p.
4): -
. _...
37
(of (or
given
"capacity"), and
"decoder"),
We shall-'start ation , rate -of first source", at the which measure in the of "with it a view to
by
considering
the
"informof the
obtaining
a measure
"produces" 11 defined
problem
of
how to
statistical
"messages". An "information source", anything of in capable which finite the of discrete producing we may set. _, out of be
_,
considering, of symbols
is
some sort,
from source,
may be ruled
no interest. "random"
particular
wo shall or
"statistical",
"stochastic")
sources
of this
rather
sort,
than
however, in in
"determinate"
are not
ones.
necessarily the=selves, state of were
Distinctions
to but be taken rather with be --be
sources assumed if
knowledge known to
a source
producing in the
only
figures
expansion information
rate,
the
be calculated Alternatively,
.. _w. 3$,
be ' considered it
random
if
all
that of
were digits
known with
produced
(say
digit; be zero.
and
in
case
information apply.
Similar,
remarks
emit
recurrent
or periodic: 2.53 source_is, "-orf the produces. "; involves set 2.54 from it is the
assumed statistical
knowledge
of
by means'of set. of
parameters that it
sequence,
or
possible of auch
parameters is well to
assumptions
out` explicitly. - point The fundamental of view from of our of property present possible a single old of a "message", is that if
selected
Thus
be that symbols
selected If the
set not
of
possible
equally
we must each; of
and
since
mutually
exdlusive oust
probabilities is about in
be unity.
ensemble.. messages
..
31
this
2.541 of
sense.
More of generally, if a message speak consists of it as each N
N symbols, of
messages were
ensemble
generally
the of etc.
we must with
given
the
probability
associated
TZ-symbol messages, specification indirect. 2.55 we permit extend directions. assumption; probability Nevertheless to of to
a1thou ;h si:; ce there are aN such the may be a very large number, the probabilities in practice will
be
results i. e.
if to
positive rather
with the
remains
formulate limiting
required
processes
applications
can approximate
4o
"sufficiently attaches 2.551 of doubly
.....
long" to this
massages.
picture of
sjO,
our form
s
ensemble
infinite
53-2, ,
messages
s_l'
the
Sill
J2"
.....
with
specification an ensemble of is
of
parameters. if it is
called
by a. change of
sequences right.,
the that
such occupied
originally
by S-, k+x
stationary,
and the will or the
still
massages case
consists
still have
of the
the
same messages
as before
This positive that do not of the
whatever
the idea
number of
The basic
sequence we think
change
as the
we can equal
stationarity probabilities
change
Amongst
stationary
xwj=
ensembles
there
called but
is
an important
A full idea is
sub-class
definition a very that
of
ensembles
cannot one; of its
which
are
here,
be given an orgodic it is
simple
ensemble of the
one such
every
member
typical
ensemble
as a whole
as regards
statistical
41
properties.
The literature
on the
theory
of
ergodic
ensembles
covering reference. (ref. 15);
is
the
extensive;
case of
for
ensembles
definitions
of discrete (ref. without A full of/ 24+).
and basic
sequences or to is
theorems
'rechet
17) proofs
treatment 15-).
(ref. the
discussion sof
which
more is
general
theory (ref.
continuous
given
by H6pf
The socalled
"ergodic
of in
theorem",
to
the
can
effect
that
"almost
on to with
all"
members
an ergodic time, to
ensemble cribed
be relied tion,
convIlu-
proper
ensemble
as a whole,
was first
Birkhoff
by the others
(ref.
gonerGlised
incidentally' above
including use
(Notice in the
technical
formulation zerd').
to mean "except
for
a set
of total
probability
very
broadly, in terms
is
sense here to
a uniform indicate
will that
kind
restriction
implied.
2.561
different
dice,
A and B. with
of all possible
ensemble
sequences ,, similarly-that
of tosses all of
of die
A is
an ergodic sequences
ensemble, of tosses of
and
possible
4z
B,
provided normally
only
that
the to
of
tossing of i. e.
are this
examples theory, of
are
used
successive in the
instance).
as a combination positive
be two probabilities
numbers
whose with
associated by a,
A-be tosses
multiplied of
and those
B by b.
Then putting
we have
a new ensemble properties those a, zero) b; which e. g. Thus are only it and not if) is of
the but
weighting of propability
have
uniformly
statistical original of
those will
there typical
sequences If
2.562 to ergodic
our to
ensembles, somewhat,
terminology of
speak
statistical of those is
because
typical us
the
equate
`.
,3
in
a particular consideration
sequence of
from
for Sj
example, at
represent t ... in
by'E
i(t)"
occurrence the
of
symbol
a sequence, is at
where
Sn and t
occur (total Si in
sequences stationarity is
which
(and But if
assuming the
indepenalternat.
ensemble
we can
ively
represent
of the
this
probability
Sj in the
as the
limiting
under "almost
relative
considerall"
frequency ation,
symbol will
and the of
result
be the
sequences
the
ensemble.
2.564
Shannon generated such that further
For the
restricts
purposes
his
of
communication
to
theory
ensembles
by Markov the
processes, of more
processes on of
depends
a finite the
number, which
number, Such
symbols
immediately by giving
a process
probability
-ml backwards,
the preceding/symbols,
... ,S J1I will be
enumerated
writton
are
Sim
Jm_2
P(E m
and will relative of values
(t+n)
....
EUm-1 (t+n-i))
of such
a specification
process the
order
absolute
probability
E Jm (t+m) )
could
be tabulated,
are
since
this
and the
above relative
ratio by is summation.
mutually is
derivable; obtainable
and this
2.565
sequences sequences his claim is of
Shannon affirms
sufficiently practical
that
to In
this
cover order
class
all to
of random
messagesubstantiate
general interest.
he gives in
examples
of with
"synthetic" the
messages
constructed of ordinary
accordance
construct equal
symbols
XFOML p, XKHRJFFJUJ YDQPAA. PMZAACIBZLHJQD Next with their appropriate OCRO IILI EEI .
taking
letters
independently -
but
EU LL NB NESERYA TH
ALHEN 1TTPA OOBTTVA NAH BRL. IS', each letter is allowed to depend on
4s ,
the preceding one: ARE T INCTORE. ST BE S DEAMY ANDY
ON IE. ANTISOUTINYS L.CEIN D ILONASIVE TOBE SEACE CTISBE. At two preceding the next is
the
dependence -
on
letters
account:
EEY CRATICT' PROt IN 110 IST LAST W, PONDENOr OF CEE. The reoeLlblanco increases by carrying more striking at each the stage, to English OP DEMONSTURE3 OF ' THE REPTAGIN IS
text
clearly do so more
be made to Similar
process
examples
as the however
symbols to say
letters. below
to
extend
the as in a definition
"choice", provide
average
symbol
of
source",
representable is to
symhols; allowed
proceed infinitely
as the
long. the
us
consider
group
of
m successive
symbols these
starting symbols
at should
position be, in
t. order,
that S, is
Um
P(EJO(t)EJl(t+l) Since this is, independent P(jd', where for of p is-a , simple ill
.... of ....
Eim(t+m)). t we may write jm) function per symbol of In the this j. group Now it simply
numerical
the m+l
average
information we define
2. JO i1 ..... jal
P(Jpt
ill
...
Jm)
10g2 9(i01
,, and for the final definition lim. of
il,
...
JO
rate
average per
information symbol).
(bits
;. to ensure that
of the exists
in
ensemble
is
sufficient
"almost
that
always".
we shall larger Markov the ml;
passing that
i. e.
any
of is of
other the
when the of
order
convergence
H will
m greater between
the
essentially
statistical
thermodynaxaica
47'
with 2.9 of is in
concept
of
entropy. alternatively
(See
Shannon
H the such
"entropy as this
message". to the
committal since it is
general
wd later able to
consider word
channel a more
decir-
"information" usage,
restrictdd to equate
Technical
however,
tends
of
information of
rate
information see,
channel as a rule
"coding" 2.575 of is
of
a given ensemble
set
symbols
and the In
everywhere of the of n
choices here C of
para.
2.1+7)
equal, the
as above.
Thus
channel
maximum
information
the
permitted of as the
fundamental transmitter
speaks
"information coding
an effective not
a device
must
on single
symbols
'48
of
symbols; have
this
means of
that internal
oust
coding. finite;
assume
i. e.
transducers of the
(conditions govern
store), which of
symbols to
way in changes of
another
govern need
possible set of
symbols, of
be one-for-one). or
The irreversible.
may be roversibl exists ouput will another restore channel, and the
communication be reversible
should be its
the that
of
the
definition
of
information increase
rate the 7,
is
cannot
entropy loc.
theorem
transducer H is thus
must invariant
leave
unaltered.
under
operation.
2.577
fact
about
transmitter
in
what
is
known
as
the'-noisoless of rate
,ch'aniiel)..
4source'
C; ' there
exists-'a :
finite-state for
s transmitter -than:
transmit
a rate
1 is:. called is
message .1'. When the that it messago ' i`s' pssible redUndanoy which
a: channel
working.;
capacity means -
'. This
a tranuzitter"'iwhich
will,
froIa message:; a .!
to -its,
case
.yw,
of
w
interest,
ix J-...
tw"3
to
which as
we shall
rim
return, If
is
that
of
a natural
.. r..
such
English.
(as
be represented it is pertinent of
ask
A number the
this
rate at about
letters '--also,
any. English
be 'reversibly-
'-
,.
_'
S'
coded
into
about
half A special
its
Shannon the
(see
ref. of even
redundancy
would one-
a text its
be reversibly
coded
into
length.
Redundancy it is almost
in
is
not
necessarily
enSineering
practice
that
to its
full
capacity, consider
of the
when this
we come to
leads of the to
In Shannon's
rx capacity.. of the
definition This is
information second
channel
dfinition
The distortion on the simple messages recoding; to passing for through example
versa.
could further
the
receiver, is
concern takes
important a random
racoding
noisy the
channel, channel
a sequence
C `
s
mado up of sequonce a symbol given the symbols the S1, Sn; .. " T1, at ... the Tn,. output Each a time
and we are
(most P(E
im
(t+rn)Fk
Fkm(t+n))
for
and
the probability
simultaneous that the
of transmission
reception ensemble of of
See. sequence of jp
Tk of ... Qm joint Tk . We events process. set up an
8 im
sequence
assume EjF.
sequences
exact
transmitted basis.
P(Fk(t),
E(t))
for
k=
k'
otherwise,
and for
every
P(Ej(t),
k there
Fk(t))
is
a j'
such that
1 for j= j'
be possible
following to which
to set
definition a channel For
up a correspondence
and theorems may be used simplicity here to the for
at all.
the
The
extent
indicate
transmission.
of
exposition where
case
transmitted each
independent,.
and where
symbol
independent
of by
all
transmitted vice a
symbols versa).
except This
the
current that
one
(and
implication only
means (The
we need
consider to
generalisation
we aczume
we need
be given
the
the of
transmission
cimultaneouo
this
probabilities
alone,
the
rcievod
k 5P
syibolc P(EjFk)
(E. Fk) .
alone.
Thus
and
P (Fk)
2.583
is defined R(E1, F)
now the
information
rate
in
the
channel
P(
77 jkP
P(EjFk) (bits
This, function
(i) "noiseless it entropy is equal P(E4) in to
has the
It the the
following
is
properties:
the
transmitter, in turn
logt
HH(P) at
the
receiver.
--- ----
._--
S3
It
is
and all
equal j
to
zero i. e.
P(Fj)P(Fk) received
and k, of
independent
those
(iii)
If
wo write
"EF"
for
the
ensemble
of
joint
events
EjF.,
and consequently
If (EF)
for the . This "joint R(, F) entropy,
Z Z.
27
y
we find H(E) +
may be written
R(E, F)
where Hi(E) xH(EF) transmitted be considered destroyed message
iiF(r)
is the "entropy of the received", of and may information
to :the the
as a maasure noise in
amount
by the In
a similar
manner -
H(EF)
receivedimessage
transmitted".
2.534 greater entropy be careful entropies the other than of Notice H(E), the not signal. to treat that i. e. it the This any is possible to have H(F) the
about prompted
information to aale:
54
the
entropy
can be increased?
Rosghly
speaking,
we sae
that
entropy
(if
16 somewhat akin
it occurs) of is the
to
"randomness";
the by random the
and the
element in tho
increase
due to nessaGe
on that
noise
It alternatively inputs,
is
possible
to
represent
which noise
as interference to receive.
equal
formula only
information in
be rephrased
certain
special
cases.
In the
terms
of
the of
above noise,
informan
presence of the
coding
theorem when
noiseless is
random get
we cannot
expect
to
completely ----
transmission.
and this
is
still
remarkable
--
that
by appropriate
SS
coding
the
frequency
of
errors
can
be
made
as
small
as
desired, served
provided as regards
only rate
that of
suitable trancmioeion. 0 of of
conditions
are
ob-
The capacity defined sources the as the capable maximum of case ac before, is no simple value
is possible In
serving this
as inputs reduces to in C.
channel.
noiseless
but formula
except for
coding to attain
theorem
is
th
as close
as desired small
the
if
11 is
capable a frequency
than This
any
positive has
theorem the
channel per
(logt means
symbol) leas
H is contain
sequence
redundancy.
how much redundancy any the not given function all kinds channel. of of
redundancy redundancy
counteract equally
are
efficacious).
this of text
general a natural
principle language; by
is
obvious simple
because In
redundancy text
an unrodundant of reading.
each misprint
would
give
rise
of
the
a noisy
the cit.
eration us
theorem addition is
Le
assume
channel
an "observer" is
who can
see both
transmitted due)to
"correction a
the
receiver
to
correct
the
errors.
The information lost by HE). is over case: small the the it fraction channel has amount correction is possible of at the due to noise, according
to
the
above
theory, that
is
given this
Thus of
expect to
be supplied to be the
found
least
capacity.
briefly theory
the is of of that
To a large uously discrete are, the varying case signals by means two basic the
however, first
limit--processes of that
place, symbols
elementary
gives in the
telephony two to
with
dimensions, 16)).
optics of second
place, number
assumption per
a finite the
symbols
gives
transmission It
varying these
possible
separately. of the
we shall
first
theory of
transmission, whose
some quantity
dimensional) to consider
be possible sequences of of
analLibgously
transmission proceed,
sequences however,
symbols.
Wo shall
'
of is
e.
a quantity chapters
theory exposition
only is
present
hence
us
suppose
we are of
given
the x. x is
probability j+ Ax) of
the
entire small
range regions,
of
variation have
of
broken
stich with
we shall
a discrete the
as in so
para. that
we assume a finite
ensemble
number
members).
Thus
suppose
there
and that
xl, n). for ... xn;
their
the
respective
probabilities the formula
Applying the
we find Z P(x4)
63C 102
This
becomes
H=-Z
Now in of the intervals tcr11 becomes x the
P(x3)
ax
1092
-P(x J)
the
logt
size the
`x*
limit
and increase
number,
Cirst
----------
--
P(x) but than the second term expect infinite, continuous boeomen as the
dx This is
no more
we should
"fundamental
symbols" for to
becomes of
but
signals for if
measure
Ii useless signals,
between
different in
even of
considerably problem.
other
ways
relevance
is ono
to
ignore
tho
(ac
Ui' abovo).
This
giving
moans that
us
longer no can we
measure. only
regard
In in most
our formulae
caoon of
as
an absolute we are
interest, between
interested In rate
particular, in a noisy
that 2.583)
the
channel
(para.
and it
finite the II(E) the
can easily
in the limit,
be shown that
and will of the
this
have
rate
the
will
remain
if i. e. makes finite.
individual Noise,
adopted. of the
continuous
channel
us turn vary
our
attention with
to
the We
continuouoly
time.
might taken in
note in
here itcr
that
the
raust the
not
be
usual
mathematical f is "continuous
sense A contin-
which
functions". specified
uous of its
completely
over
any
finite
other
be to
any
application
domain.
Elootronic of mathoL1atics
ongineoring associated
anily
very the
is.
general
A mathematical
conditions, of
function
of time,
c an. under
superposition i. e.
functions,
functions sin
of
different the of f,
values as the
of
W and freequalunderThus
lp , qucncy" to 2
4) being
"angular (It is
sine-function ff is the
stood),
quantity
C9 is
it
to
speak of the
"frequency
is the
comvirtually
function; by giving
amplitude and
components.
amplitude
phase
are usually
given
together
in
the
form
of a"single
6/
number).
of
frequency even
and of the
spectrum" from
"Fourier the
transformation". of a complex
analysis pure
musical
sound
into
tones, in
application of "frequency
also
representation different
different bands"
not of
the
place
for
Fourier
2.65
of functions of of
Special
time
importance
whose spectrum are
attaches
covers
to the
only to as
class
a finite "band-
frequencies. functions,
referred
makes
the
application
straightforward. 53); -
As expressed
the
band
specified discrete
of time a function from 0 to W cycles per by giving its ordinates points spaced l/2V7
If
f(t) second at
is it
limited is
to completely of
a series apart.
seconds
..
_.:;. ________
-.;
_2
Thuo
in
the
case of to
of
band-limited
functions measures is
of
time
the we
generalisation do not per than Given points values all the need
information
consider
second",
because
are per
values the
second. set of
a discrete deduce
spaced of the
the
other by the
discrete
context
13,19)
terminology.
2W logons
per rate
average
and multiply in
Thus symbols;
only of of are
ence limited is
theory sequences
that
the of
chosen
a finite functions
of range
a number
a continuous
63
measures parallel of
for with
this the
for
the
rest-,
closely ensembles
We introduce
band-limited functions
functions, for
probability ordinates in para. functions Thus function of the an information source of the
concerned. we use
reasons of these
integration of the
density
summations xn) of is
discrete probability in
...
a sequence the by
n ordinates or
source, i given
entropy
information
rate
P(x1,
...
xn)
1062 p(xl,
dxl ...
.. 0 xn)
cxcn. (bits
.
par logon). inform-
for
relative
for
a noisy
theorem
no further of of the
concept
"coding"
known
particular noise
countering
many problems
associated
telegraphy
i
A --,
d-
to fact with
the
theory
of
continuously quantiitio2
signals, etc.
since ) associated
th.,
the
physical
essentially
continuous
quantities
2.69 band-limited applications, frequency any long case run band all functions since is
of
the in
theory
to
sharp
sicnals
bo represented as band-limited,
a :sufficient in a very
imFation
albeit
band.
2.7
The
theory-
of
rocent:
ion
In has
our
account on the
of
been
question for of
idoal
communication ".
system
n purpose
A question
a non-idoal
even best
receiver the
distribution hence
a new element
enters
that
need
be said
here,
however, to
can
be said
very
briefly,
and we confine
ourselves
genoral
observations.
2.72
-Lot
us
simplify
our
notation
al
little.
We shall
(j "E " the symbols - 1, ... n) for events use the transmission or of M symbols drawn from some of/sequences consisting
set, there which are we need just not further specify: of we this
n pousible
sequences
length.
1, ... n')
Simil arly
for of
we shall
use the
of whether one.
symbols
the
"FJ"
(j
of
events the
consisting
reception the
from
same need to
sot
We do not of our
time-behaviour
assumed:
successive at ta0.
sorting so that
assumed it is
case
where to
and we shall
reference
a limit-operation.
Let messages
reception
the
statistics
of of
in
of
the
any
channel
signal
Nov; the
in
problem
kaotiving
simply
P(Ej, F1}
for all j and k, i. e. the relative for probability each possible of the case of possible
a received
transmitted
sd quence.
sequences
-i E
If statistics P(Fk, thoorera B3), of are tha Bayes, (as is in most the is likely) of the P(EJ), by channel 2(rk) the and known result viz. torn
given
simply
inversion
P(E , F)
To apply
of the
--------needs one .
the
this
transmitted
forrtaulra
to
know
the
distribution
of the
signals,
distribution
receivedi
. rclative
oi6nalo,
to each
and the
transmitted
distribution
signal.
of received.
signals
This theory. if
fox^.nu1a is However, it in
almost
a triviality is thrown
we view
information
We can the in rate a noisy Fk) the due direct dovired to H(E) H(E) -
to
rocoivo
at rate
channel). at the
entropy is
receiver
But
reprocents of information
the
information Entropy
can
a eenae of
be
46 p.
49) Thus
not prior
"information" ignorance
"ignorance".
'!
_. 6 1
Il
of
what
willof
is
our
posterior that
ignorance
The fact
the
coding
is
value
leas
of
then
the
ideal
will
i. e. the
be reflected
the received
in
latter;
completely
determine
transmitted
in
turn
means that
the transmitted probabi; is 2 JO to itieo
all
that
signal (P(E for the of
can be said
is j, that Fly), vqrious this loot; Oet
at the
aro 21.,
receiver
such after transmitted
about and
there where
such
liticz
known)
po: sib1o of
only rest
probubil.
ware
and the
zero
entropy
2 74 .
be zero
and the
Thin
transmission
interpretation i: due to the it of
faultless.
of vloodaard of the coca (ref. Shannon, to some in indbviduel given is of 46), in that
non-optimum. involve: it is
which
from
practice to the
in
extent the
channel, of
signals in
account
we have
considered
the
entropy
of the
received
transmitted
signal to
signal
i; von some
as 11F(E). taken all over
strictly
speaking received
an average as well
porsible
signals
possible
trmmsmitted
ones.
A consistent,
though
to
some extent
,___
misleading .. Qasuro to of
be given
by
1002
a nevi= sort
relative of received
more in
reasonably
a particular although
this,
he does
some qualifications
ieasuxo
indocated'(loc. If what
is
at
receiver
some doinite alboit in an incorrect practice F, ) its gives is from. not ref. it. in 46 a device, general
calculate probablo to
distribution
most
mod.o . that
sequence to
readings
possible p. 60).
reconstruct if the
Thus
which .. signal,
automatically the
selects ac=t
information
maximuln "guesswork
obtainable. destroys
:aloodward information".
summarises;
by saying
The ideal
receiver
may in
i,.
. -....
,..
__
r_
____..
' .;
it destroy
.=,69
an irreversible
operation,
since
must
what
it
can of the
operation,
received
destroy the tute
"noise".
any more
But
than and of
it
it
must not,
need of
for
the i
ideal
from must
on this lity,
criterion distribution.
a probab
is not
possible : tabistics,
without except
compute from
knowledge in the
transmitter
ecn3e that
and channel
a receiver The theory hass somotines is at clearly . the
mi 01.t Which
P(FD).
boon
receiving,
a communication on a natural
and that
measurenent theory to
the apply
meaning
"channel"; of
some independent statistics". (loc. to cit. radar, here. of think the of The )-and. has Thus position the and
source
knowledge of the
theory of
by Woodward
processes
measurement,
e. g.
depended. is used
as a'transmitted 6igual"
signal" perturbed
obtained
as a "received
7o
bZynoise
in
the
"channel"s
but
this
model
only
sac=
-reasonable
in general
because
performed and in
the measurement
more any case accurataly it is
of distance
in atill other
can ba
way. -, than uncertain
by radar;
as a rule
to attach object's
theory-,
to the
concopt
o. apriori other
for only
probfeatures
of the
distance.
that of or
Certqin
"coc1ing" present the this
cormunicotion either
inotanco, in a drastically of
absent those
modified roceotion"
reasons for
bo preferred,
with in
a the
work
by Brillouin More to
physics (. ef.
Gabor
largely outside
continuous
and is
in
favour Lterely
entropy
not
information. entropy",
word
"nogontropy" with
and querts
"information"
with we
7(' :'
remember
that
the
"information
content"
of
a message
can also
to it.
be conceived
Sac ref. '11).
as the
"prior
ignorance"
with
respect
Brillouin a demon",
in charge
treats
in
who according
of a door
between
observe in such
moleculos
him,,
a way as to thus
between
system. "demon" of
the
principle
is
thus
not
violated. of Szilard's
rillouin's proof
a version
of
information.
In terms : horn can of firstly secure of is a simplified that, the model entropyto that that tho in of such a
system that of
proportional choices)
terms
binary
ho can order
obtain to of
molecules;
obtain tho
information by at least
ontropy
system
2.821
proof
4`
Lot
first. We suppose in order
us treat
that to the "roe"
the
second part
of the his
This
supply
own illumination
molecules.
lZ
the
exec
dituro where
of TSic
at the has
least
hd - kT chamber,
radiation
and the
roflocrod
signal
as tho increased,
background by at
fluctuations. least k, i. e.
ontropy of
that is is
one bit us
suppose into
by means
a sliding
us ascure
that
molecule, is the
hand has
"volume" gas
gas
volume decrease
entropy
the
entropy
can be calculated
from
(per kT pV =
molecule) Brillauin
as
k loge that
2. if one bit
2
concludes
information
is
taken all
units
o ontron; ,
,73
fron
of
Lw re11 Is paradox
enbrop7 to a 1av Thero in a cystew, Gas; of
is
alteration
e of
of the
entrop
Jaw of increase
increae.
r ninu3 of of the
information. entropy
may,
be a decrease of the
as in but
case when
cNanpie io
win; lo--=olecu1e
information
being
uce. up in
in. crcazO oxponsc of or
compensa icn.
in: roriiiatian a proportional
Sinila:
about
1y there
may be an
but only at the
a system; e of
incrca:
entrap y.
Thus
there
are
physical interpreted is
entropy in tho
basically
because symbols
different; in far
equally
be little our
concerned concern of is
priysical to
what
elucidato
connection
theory
xirtl=
to
communication
in the
with
next
a logical
chapter.
theory
of
information
be outlined
2. r3
In
its
more
direct
applications,
the
tccb
i a- theory
of infoxLlation
in
a remarkable in this
is
no sign
direction
Arc o before
new typal of
and have
s"3:3tcic. and
stimulRted
systems of
research
modulation deal
iepulso-code"
"delta"
have
o-'rod o groat
tr coSeveral or n.jrtial
on. the
methods, elimination
throihahold
of of
of
(i.
e.
elimination transmicsion
redkindancy)
becoming
speech
;; cep
noer?. nG possibilities
for
soma types
of
channel.
(gor
;; eneral
re
erencos
to
technical of
work
in
these
fields theory"
proceedings
the, "Coiunication
London 1953).
quite
Coiunication directly of to optical of spatial of work time has time systems the function also). been done
instead a twwobe
a function
` nossace" (which
tyypically colarso
may of
redundancy savings
appears could
be effected
has of
boon
suggested
(.ref.
33) . that is
the
simple to
lino-drawings
moasurable
signals
a eenvo-operated
drawing
nechanicm.
In physiology,
estimates
the with
of
partly
in
on the used
have be of
the
importance "storage
whose
capacity"
is
a subject
of controversy.
Coding techniques computation, for the assume where inversion importance methods of In involving
the
case
of
automatic (e. g.
random are in
sequences use.
matrices)
76
TO THE INFORMATION
CONCEPT
Having
seen
how the
concept
of
"entropy"
operates systems
to
of
information
in
engineering
examination in systems of
problems,
measures of abstract
a more
and related
concepts
In a logically
theory they these represent considerations some justification of of its the technical
ordered
should
account
of information
since
theory,
We shall
"logical
content"
Our aim is
of a sentence,
to major elucidate definitions rate of the
standing
of of
the
information
the
information
relations
between
sentences,
--
--
------_
. ya_
'i , .6.
77
;ti .'
etc.
as measures
in of
in
perspective theory,
major
definition rate in
information
a noisy
3.1
Logical
content
that in
a logical
proposition and
should
is
not
though
obvious modern
be employed.
The beginnings
in
a theory
of
the
"content"
of propositions
who says "If (ref. 45,
might
para. followa than the
be traced
5.14): from -
to
Wittgenstein,
the than
latter the
p then "
A tautology nothing.
from q and q p follows they are one and the same proposition. follows from all it propositions: says
however
does not
he hints
of probability. By the any numerical thing, measure term "logical vague says". content" but we ahall mean simple
of this
essentially
. 78
3.12
Before
we can
set
up a measure
of
quantity
and it
is
to distinguish
call "weak"
of scale,
objects
if the given
or entities
any two or is relation that or its of
of a certain
them it is
sat
possible
in
a certain
to are say
respect
which is
they to of" or
find the
an ancestor everyobject to In
in
entity of the
do so to a weak It
we shall
that simple
we can sense
speak of
unequivocally
a measure measure)
a one-dimensional
a metric. In (ref. 29, p. 34) "I there are of this on the maintain, of connection subject then, of we may quote "probability": in what follows, between Keynes that the
some pairs
probabilities
which no comparison is of magnitude that possible; we can say, nevertheless, of some that the one is pairs of relations of probability less, it greater is not and the other although to measure the difference possible between them; and that meaning type special a very of to a numerical can be given in case ..... comparison a of
members
79
magnitude".
Keynes is that
but
saying, only
in
there
that
is
there
in`general
a weak scale
in
may be a strong'-scale
special
(ref. ordering"
37 seens.
(For
such cases
see also
a weak the
one
scope
relation. this of
usually
which
Thus
we might
base
To do this of additional
relations
on age,
"anyone
a nonstrong
make all
depend
on the
of. time-order. Now from says" Wittgenstein's we have less been only than account one hint q if "how an from been'
of of
p follows to have
supposed of'Iaterial
referring in the
a relationship
of'Whitehead
and Russell;
do,
-----
in virtue
-- - -- ------
of its
definition
which
makes the
truth
or
falseho"od
of
a proposition
expressing
a material
implication
of the
statement
dependent
which
on the
the
truth
or falsehood
is
statements
between
relationship
rather
than
merely
on their
logical
'of
both
without
Dedticibility
by the
logical
form
in of the
of the
case
statements
of a formal for the
(sentences)
system system; in
concerned,
terms of
inference
by definition
sentences"
can stand
in
a deducibility
notation ref.
for
we shall "q is
pl,
(following deducible
... pl, pn)"; p2,
37h) Write,
D" 1
"g pn" " with
or more
is deducible is "q" to etc. be
p21 of
conjunction
This "p",
understood
as a syntactical
symbol,
names. our intention sentences of to base a measure a logical them. relation language The of the
deducibility
relationships is
between
a transitive
us only not in
of the of two
sentences,
that
TTTT Si
sentences p, q that one, is deducible from the that other. any
We. can
reasonably
postulate
measure
of logical
content
with it
the
relationships. if q is
than p)
we c'an'lay the
deducible
or then equal C(p)
p then
of C(q). q.
conttent
of ,
-
Or in'symbols:
It
followz
that i. e.
D(q; p)
interdeducible
sentences
have equal
content,
If,
and
D(p; refer
q) to
then these
C(p)
"
C(q). as
We shall
requirements
the
"deducibility For
criteria". some purposes of the then this is we should of these p) also like to i. e.
the
converse find
criteria, q),
C(p)
C(q) that
and D(p;
we shall
always the
practicable. deducibility
3.151
relation in
scale? that we might But similarly the equal, relation apart put this
might
C(p) would
= C(q) imply
whenever C(p)
unrelated.
a C(r) C(q)
unrelated, of q to r.
whateuor contents
end with
from
those
of tautologies
and contradictions.
8t
3.2
Logical
probability
3.21 ,
The first
explicit
measure
of
logical
put Popper of
,
forward
by Popper
in
37, bo measured
proposedi. that
"content"
de nies",
probability This
this
of
statement of
a measure it in
general
with
some detail.
to
flay something
about This
the notion
of
"logical
probability".
concept
of such of 37g)
has appeared
"truth
in many versions.
in a the Popper can be definition
Broaely, logical
it
is
language, of
shown without
recourse
separate this
language,
and that
a natural
and simple
following of the
basis
account,
for
the
theory
of probability.
a simple terminology omitting
In the
version and most
theory of
which
methods of the
Carnap
features to the
relevant ref.
22,29,37d,
37f)-
of the
relevant the
concepts "L-range"
what he calls
13
of
a sentence;
and this
concept
is
most
simply
defined
for
a sentential
calculus
which
with
a finite
call
number n of
pl, p2, """ pn"
elementary
sentences,
we shall
a note
dispose of the necessity order-to it is well to intercomments on notation, Our here on the conventions adopted. In
language with are not symbols of the logical them; but rather names for as such they which we deal, the language to be belonging in which considered as can (metalanguage, language), English, viz. syntax wo write in so far except quote marks etc. are unnecessary for by ordinary usage for other as they would. be called and
reasons.
The logical language, of logical however, is form of an expression by the evinced of placing the
constants, to it; which we use to refer for-logical constants symbols selves. Symbols names, i. e. in question. formulations suffixes, of these of in with
sometimes brackets
in the expression etc. thus we might say that serve as names for them-
integer
are the
constant language
usually with
addition, implicitly or
names
for
complete symbols
expressions
use
without
`'..
sentences
pl,
p2,
"""
pn
in
the
sense that
no deduci-
relations
between
them.
of a sentence any can be defined, a table F) for
'L-range"
in
terms
of
its
For
sentence (T or
can be prepared
truth-value
every
possible
combination
elementary
of truth-values
and its cases in
of its
'L-range which it may has
sentences; of
as the T. Bor
proportion
example,
in
the
case
of
the
sentence
Pi v p2 . pl F F F F T T T T Five out of
p3 p2, F P T T F F T T eight
p3
last
column
are
T,
of the
L-range
can, vary In
from
however,
to
the will
truth
concerned.
make this
and its
sentence
negation
together
form
a basic
pair.
A conjunction
' "-..
chain
containing
just
one member
of
oach
basic
pair
is
n elementary
2n Possible
universes.
about
the
calculus (--contradictory
sentences
be expressed chain the
to
zero --)
can
whose dinjunctive
are normal
of of
equivalent terms in
terms, the
disjunctive by the
divided to the
number
proportion,
consistent proportion
sentence;
as we in which
of
possible
universes
would
be true..
originally
the
notion
of
clearly
to the to
definition
the (equally)
cases" in the
mathematicians probability
Wittgenstein's (ref. 45 para.
development identical
truth-grounds"
and is
virtually
as "ratioof
_9
91
Carnap for in an exact a logical only numerical language slightly is only not
has
laid of
down a proposal logical probability which 9', appendix). for our purpose of the s$ntences Thus red", y
as that the
considered (ref.
differs
great where
cases
"families" as e. g.
a similar various
structure, colours of
are
"A is (like
e; ementry
sentence);
but
provision
the
family
quoted, allotted
greater
are
than
unity.
Thus the
exhaustive,
considered of 1/3
probabilities
each.
however
ignore
this
minor
complic-
for
same as L-range.
definition same calculus of occurrence of
A rival to the
of
--in
frequency sequence of
an event class.
an empirical
events
a certain
As usually
stated
this
definition
applies
to
the
probabil-
ities easily
of
events
(of to to
a given the
type);
but
it
equally
be applied e. g.
sentences euch
expressing as "The
occurrence,
a sentence
falle
heads".
,.
of probability radically
define distinct;
of prob-
which
many
are under
most
aom0 conditions
interesting
' yot
in
of the
applications
ability For
theory
there
ie. a close
relation
between only in
5)
them.
be interested
tatet
logical
discuss concept
(chapter
be brought
relation
There based
is
a mathematical
on probability language.
based one
on a symbolic
This
consists as if
in its
treating
probability in terms
a sentence
were of
of a given
its
probability
form
due to
logical elementary
when expressed
(hypothetical)
sentences.
To obtain
logical any foram we first number of
an expression
of the
requisite
to
express figures;
a as a binary i. e. in the
dedimal, form
desired
O. blb2
...
bN
where
each bj
is
0 or 1. last
or
Without
loss
of generality
we assume only
figure
unity,
to be 1 (where
cases
corresponding respectively).
with
contradictory
and tautological
formulae
Now taki:
6g N krpothetical
elementary
i:
sentences Pl where If the pl, p2, """ (P3 are to pN C we write ..... out the P. U) expression ..... )))
88
CP2 spaces
(PIT-1
b1 is "v".
write r
so on up to have L-range
so formed equal
the
sentential
calculus)
probability
For
we shall get
example
if
we take
aa
21/32
0.10101_
pl whose L-range
Proof. expression the for
(p2 v
(p3 v (p4'* .
may be calculated
Let p which sentence is p, "p(j)"
denote or
the
part
conjoined _1; i. e.
disjoined expression
elementary
the
Pi
with with figures prove (1) its the spaces
(Pi+1
filled
...
with Lot
..
(PIT-1
logical "b(j)"
) PN)..
connectives denote of the a. in last accordance N-J+l
decimal
expansion P(p(3))
relation
= 0. b()
j-N, provided
that for
this j-k1.
relation
holds
for
j-
k-i
We have P(PN) --
P (N) 14 -
a pN,, 0.1
and
bN
1.
Thus
O. b(N).
(2)
fixed or or k greater Pk-1 O. lb V P(k) k.
Let
than I.
P(p (k))
Now as is P(k_1)
O. b(k)
is
for
Pk-l' Is, of
some'
P(k) O. Ob(k) P(k) we
O. b(k-1) indPndent,
0.1
0.0b (k)
0. lb(k).
Thus It follows
N.
2 (P (k-1)) that
Since
0. b (k-1) 0. b(j)
and 0. b(1)
. all 3 from
our result
P(p(3))
P(1) "p
for
1 to
- a,
is
given
by the
case
ja
1.
3.3
3.31 measures
Popper's
measure
Logical
of content
probability, of we said a sontence; (para. it is 3.21),
the
"truth-possibility"
the proportion
of state-descriptions
or possible
universes
`.1n
consistent definition
with of
the content -
sentence
in
question. above, S
Popper's namely
as introduced 1. PW
measures of
"flasehood
possibility"; or possible
it
state-descriptions
universes
inconsistent
with
the true
sentence sentence
p. or tautology
and hence has I
A necessarily
11 is consistent with any
state-description
zero. sentence
At the
other
scale, it is
has content
all that
consistent
sentence
or contradiction
any sentence
sentence
(if
we ignore in para.
the
of
of 4. in
has groups,
which of
it the
negated; group,
those second.
inconsistent
ppk content 4,
of
two
and in
sentences
sentence of
can have
where language;
elementary
sentences
value
,_
is
attained
by any state-description.
If
we let
the'number the it is
language
increase,
sentence
remains
unaltered; has
that-any.,
elementary-sentence
content'A,
so on.
Hence it
finite consider
is
possible;
of
restriction
provided this` which ones; would
to
a
we
-number, only
elementary
sentences, Even
finite for
expressions of finite
as limiting in
language However
consist
anomaly
have
inconsistent
require to D(q; in p)
deducibility If
referred then
and D(p;
If
this
by
is
laid
down as a requirement
language, from since
then
it
is
contra-
vened cannot
the
infinite
a contradiction
be deduced
a state-description.
3.321 that
ition. n$ then though
It this
If
is
suggested
situation
a given it
can be circumvented
inequality as holding holde also it for in
we define
we can no longer
demonstrate
numerically.
92
This
point
is'of
more
importance-in
a functional
calculus,
which
3.4
briefly.
functional
calculus
so far
given
subject calculus
of a functional
This is
predicates. of logical
reference
h language
with
F1,
F21
sentences It is
required
that
shall
be independent
the the
so that with
for
practical
purposes with
a sentential additional
calculus point
existential
Mn elementary
concerns sentences. the
The only
of
universal
and
way of
considering describes
these a universe .
language
has exactly
n individuals
each of which
has a name
93
in are that the all language; the i. e. that in the the individuals universe. It al, a2, ... an
individuals
follows
we may write (x). Fjx .. and similarly . __ . F, al v, - F jag v ... vF jan can be considered of like
follows
Foal
F, a2 . .
...
Fat .
for
each
j;
sentences (truth-functions)
logical
the
sentences
in disjunctive
other
that
or
logical
and content
can be defined
language If
to the
zero.
increase content
without unity,
universal
sentences sentences
and existential
Kemeny on the
(ref.
26,27)
has of the
shown
how
applicability
measure
may be removed.
Firstly, the elementary to sentences the it is not necessary it is to assume nec-
independent; concept
only
essary
"state-description". just one member But not all of if each the maximal.
conjunction. independent,
elementary
sentences
conjunctions
will
be consistent.
In
this
case
a state-
94
description
is
any
consistent
maximal
conjunction.
Content
sentence of a
is
still
defined
with
an the
it.
proportion
of, state-descriptions
inconsistent
it
is
not
necessary the
to assume If
has a name in
with "number N (greater of
language.
n)
than
possible
universe, of models
terms in
proportion false. N;
question on the
measures
depend
number
universal
sentence
will
content
that that
the for
N to
increase regarding
however,
inequalities.
Finally,
quite calculus functional may arise. it will not this naturally with to higher
can be extended
e. g. a functional
languages, predicated,
two-place etc.
Additional in
be possible in effect
since
56 3.51
Additivity %
and
We noted
measured-in
a sentential
calculus
an olomontary
sentence
.A
9c
has content yi, and the might arise: conjunction is there of two has content such such that that 4.
a measuro and
elementary of two ,
sentences elementary
as units, sentences
measures
trice
as much as either,
alone, part
is
for
a moasuro which
i. e. which sentences C(p jpk) p is
is-, additiv
such that for
elementary
two distinct
and p. C(p j)
Since
we consider
all
elementary
sentences
vilue, of
this
is
equivalent
to
rgquiring
that
the
content be
a conjunction
eentencea, probability
Hence the
ohould
proportional
of such
to k.
(as I, -range)
simple expedient
a conjunction
is
suggested
measure,
as
opposed
the
linear
measure
so far.
(ref. view 1s of
It
by Bar-Hillel
has the
and Carnap
in sentential
the
base
any con k.
ction
/3lementary k of
sentences
llore
generally,
we might
squire
an add.itivity
condition
of
something
the
following
forms
.
C(pq) " which It this = are C(p) + C(q) for. -any (in two sentences to p and q
,, 9:-. ..
some sense
turns
logarithmic we use k
measure "independent" It is
condition
probability
a theorem
probability
`vo have
calculus
that
P(pq)
P(p)P(q).
If
wo use the
of
logarithmic
the measure
definition
for such
of content
independent
the
sentences
additivity is immediate.
There
in terms of a logical in is (but tha
is
no simple
the
way of expressing
conditions Absence nor of for interA either A in terms with
neither not
sufficient
necessary. is that
necessary)
condition or
q should
a tautology.,
condition numbers of
be expressed consistent
state-descriptions
respective
sentences. For some applications requirement some not. would of the content be linear
the
seem to and
The logarithmic
measures
are
equivalent, for
deal 5.
in
the also
any inequality
for these the other.
one measure
with
We shall in chapter
some applications
neasures
possibilities measures. is
are
however now
We shall topologically
soafar
introduced.
the the
linear
and of a with, we
but
identical measure
of the
ltself.
For
linear
i-C
the
logarithmic involving
measure logartthmic
a rather
some purposes,
the next to chapter}, content we shall though it of is
however,
(to
more fully
in a measure as between
allied
measures call
because perhaps
"content".
most
of this
of the
class
entropy
is
of
closely
a source, and
analogous
we shall such
to hannon's
call such
measure
measures
"entropy"
measures,
donot
a measure
"H(p)".
way to
average
aocur
of the in
sign-symmotry
non-symmetrical which this
measures
many ways
might
be done.
We shall
however
only , be concrnd
with
one,
namely metic
that
when"we
take
a weighted Here (1 -
arith-
mean of H(P)
measure. --
C(P) "-
C(-P).
1062
P(p).
This
being It is equal to 1. for it this for our
measure
It is P(p) the
P(p)
this
-,
but 0.
have
in does
chapter not it
4.
criteria which
certain
a2alogous
wo shall
then
consider.
3.6
Logical
correlation
There with
is
a large like
concepts
In
9g i] 1
most usual
applications
we deal
between
"correlation up,
numerical
"correlation in statistical
of
are set
and are
analysis.
We shall between
here
consider in
sentences"
. !
natural
extension
of
a consideration
of
measures
of
"content",
a sort of
since
mutual
a correlation
content of natural the
is
is
some respedts
concerned. of our
like
sentences extension
Another
investi-
gation
is
into
of
"confirmation's
(of
by another). ation in
consider
some measures
section
Let two
us
start
by
considering p and q in zc
sentences previous in
probabilities.
this
as in
"L-range", fact = b.
i
interpretation P(p) aa
It probability special
is than it is or
does not
of course pq is
follow ab,
that
the in the
of the case of
less loss
conjunction
except
than than
3.621 us to write
truth-function P(pvq) P(--p) P(p-q) P(-p-q)
numbers
1-a
awC
I-abfC
4
1o. ... p
ti
and so on.
i. e. given (whore p, of
We can also
"Pp (q)" q"): -
write
stands
down relative
for "the
probabilities,
probablility,
Pp(q) Pp(-q) P_, (q) P and the so on. usual pvl All (pq) these identities laws
(subject
existence
fractions
is
clear
that
if
we are
to
speak
of
a moasure q, it
of
"logical
correlation" in terms
between of the
paand
would
be simply
c.
It
propertiestb.
0<c<a,
ca ab in case
0<c<,
of
independence.
c>
ab
indicates (iii)
positive If
c=ab=a,.
correlation, then c-
c< ab - b;
ab
negative. and
a1
if
b-1
then
(iv) o ab=0.
(v) q entails p then (vi) If
If
a-
or if
bA0
then
p entails c tt b. If p entails
q then
c-a;
and
if
-q
then
c -. 0.
irr
(vii) c=a - b. If p is equivalent to q then
All
these
properties measure,
are but
intuitively
a correlation for
in
particular perhaps
we should
of independence of
to be repreronted with
of p and q themselves.
or c/ab, which for The
This. suggests
independence take has
ab
0 and. 1.
a
respectively.
second
of
these
been
the.
of dependence" is is
by Keynes
(ref.
29,
p.
151). finite;
being
as a, b p
of
and. c all
approximately both very
zero
same order
the find,
(i. e. with
equivalent small).
q but later
probabilities that
We shall
attaches
to
a measure
topologically
equivalent
which should
q.
scntencos correlation
p and
between
p and q",
we might
C(p,
require
) c1. this _ as the C(q, p). requirement";
We shall
refer
to
"Byrmetry
or more explicitly
as the
"sontence-symmetry
requirement",
.j
"""' lOt_ .
z
to
distinguish
this
of
from
tho
signis
symmetry fulfilled
mentioned we shall
requirement is
that
measure
"sentence-
are
symmetrical of
b ands c
The condition
in should b and vice Perhaps the terms
sentence-symmetry
the quantities a,
a measure is simply
remain versa.
unaltered
when a is'
substituted
simplest
examples relative
of
nonprobabilities
measures Pq(p)
When wo consider.
we shall
indicate
their
lack e. g.
of
sentence-
by our notation
by writing
also of the of
"Op(q)".
use It two in in may sentences. such
have
further
be that are
we wish
neasuro
related
independently
their
signs,
-p would If in
be counted addition
related
to q to the
same
as p.
implies
we have sentence
symmetry
q)
C(-p,
q)
C(p, in it and
-q) terms
C(-p, of a, b,
-q). c) for
a measure by testing a by
has the e by
above measures
has sign--syamotry.
'c,
i*, -.
/o.
3.642 call-"sign-perisymmetry". mo by Professor sentence alters popper). the measure more usual ' (This oasis what
-}
we shall to of a
Here' changng in
i. e.
f(x)
is x.
a numerical
Sign-pericymmetry as a spcial
sign-symmetry
case,
f (x) f (x)
X. -x
we ,tind
cases
with
the
measure
by writing
and then
aoe whether
of the have original. tho
the
result
(If
can be expressed
it.,, csn, the function
as
required measures
propeZty)", so far the if C (p. montioned, corroaponding q) and cab C(-p, -q) only
sir, m-poricynraetrical; f(%) q) This is -x. C(p, measure "positive" of Thus -q)
ab -c indicate
can thus or
whether in accord
"negative",
moacuroa
mathematical
statistics.
3.65
"product-nonnt" The meacuro which
Wo can actually
tochniqu results directly is due to
apply
to the
the
mathematical
of sentences,
case
Kemeny and
Oppenheim
(ref.
28).
. -
/o
The usua]A. application of this technique
is
to
a case like
connection population. to be
the
there
followings
is between
Wo wish
height of and
to
see,
say,
in
what
height the
n people
corresponding
weight of the
measurements height is
and similarly
this done,
fn-the
wo have
weight
measurements:
supposing
2 Xj Y
is computed, namely (1/n)
This is a suhl of
0 o.
of the measurements
Xjyj"
such that we count Positively
terms
the
or
and weight
and neratively it"to
are both
the if
above average,
remainder; height tends and
be positive if the It
increase if there in
no average
such
limiting
values
-1 of
root-mean-square product-moment is
values coefficient,
estimated
correlation,
__ dos
-a
3.651
If
this with
to
us associate
positive q.
the value
Then have to to
positive
the
mean we shall
associate
with
their
negative
truth
states
such that
xP(F) + x'P(-p)
9P(q) or xf y' Now tho _ -XI _yl product+ ) 3''P(-Q, a a =b omont
0
0
can be computed,
equal
to
4
This as the a scale and 1.
1-.
)b(1-b)
is fact more cmakes or ab, the less the
is
limits
former to
for -q.
p equivalent It is zero
independent.
lo 6
3.7
Confirmation
3.71
sentence-symmetrical, then we turn to
A concept
less "confirmation", since
of
"correlation"
is
customarily
commonly
seem to
"r
be inappropriate, with
virtually the
asy=etry application
most?: typical i. e.
"confirmation" Seneralisations. of
concept,
application
universal
sentence-symmetry
in
the L=
problematical.
Although:
in
is -
typical
in it
cases
the
evidence
which
related from
confirms
to the
a hypothesis
hypothesis -we
symmetrically be deducible
the
hypothesis
should apriori.
not
wish
to rule find
out in
measures
We shall feature
sentence-symmetrical in confirmation-
measures
theory.
to a certain
3.72
those It is just considered
A measure
is put
very
forward
closely
related
(ref. of
to
37e). x to
"tE(x,
y)",
read
as
clearly of
difference them
popper's .
wo use
sentences
instead
reduces
to
/07
E (p
q) ,
c-----
ab
It
sign. -symmetry
of of
thi:
measure
Popper
a measure
"degree
of confirmation
--------c+ab
c (1 + )
nor satisfy signthe follhas
This
measure
has neither It
sentonce-symiaetry is designed to
properties. of properties in
(where
translated (i)
a direct is
manner rsater
Cp(q)
or
less oi',
th=
zero
dopending q.
-1 0 If
as p supports,
is
indepandont
or undor11ineL
(ii) (iii) (iv)
<
Cp(p) <1
<1.
- P(-p) Cp(q)
Cp(p) a
a C(F)
C2
-'L
Lot
q have q)
a high ---and
content lot
--
so that q. power
E(p, p,
Cp (q)
increases with
q to
p,
and therofore
scientific
interest
/
(vii) greater as Pr(p) than, is (viii) (a) equal greater If If to, 0(p) or than, C(q) than to, p then 0. then Cr(p) C5(q) or less
less equal
q entails Cp(q)
(b)
incroaco toGether. (c) increase together.. (ix) If (a)
for
any given
q,
CP(q)
and
C(p)
for
any
given
p,
Cp(q)
and
P(q)
-q
is
consistent <0.
and
entails
p then
Cp(q)
q,
Cp(q)
and
Pep)
for
any
given
p,
p(q)
and
P(q)
will
that
the
definition
corresponds It of is
Popper's that
considered
to in
treat particular
to
inadequacy a definition
of
none (vi)
requirements:
the
main
and
. 10
3.712
0 (q)
bocause notice it that conflicts this
Popper
explicitly
rejects
a definition
=--0+
c-
ab
ab
(1 + c)
(iv). We might
with rejected
the
requirement is
definition
sentence-symmetrical,
the
grounds
for
its
rejection
have nothing
to do
:3ontcnce- , yrmtry. It
simpler measure
is. interesting
Cp(q) except c/ab
to notice
satisfies if --the i. e.
that
all
the
of
Popper's themselves
requirements are
(viii)Oe),
requirements; if the
reasonably
interpreted
numbers
if and Simplicity of
representing
C(p) is
limits
of ranges
asp equal would which thus does to
are
suitably
the
altered,
1/a. dropping
redefined
requirement important
ularly
measure it is
measure example
(pace
that there of
further between
difference sentential
the
co. copt
confirmation
and that
correlation. Popper, would of however, intimates for for This to is reasons because laws.
be unsatisfactory confirmation,
purposes to his
theory of to
external he wishes
requirements. be applicable
measure
universal
11
se have P(p) Thus =0 the and consequently c/ab other would
IZ
p is
such
a law, i. e.
P(pq) would
a 0; always of the
a-c-0.
together theory;
measures preclude
desired).
the
same time,
this in
still
invalidate are
considerations,
i. e. we can the where concepts these
as they
out
consider apply).
applications
do not
. 8
3.81 c/ab
"Information
tranofer"
If in the place of the "linear" measure
eve 'were
to
use
"logarithmic"
(topologically
equivalent)
one 1002
C
secure Popper's
To define
Pi "
of content as
previous All of
except appropriate
again
hold and in
with
addition infinity,
from
minus
when
are to
finite, the
to
infinity It
when is
a, zero
tend of
to
zero,
same'order.
independence. 3.82 concept in ghat the of the Analogy technical chaptor said with the of "information" as introduced with as to q, and
theory suggests
that
has
vie raust
a definition
chapter
by of
"P(BjPk)" the
the
transmission of symbol is
G, and
simultaneous rate
averaC, o information
channel
defined
R
This
P(F, Fk)
is
lo; ,
ovor the
F(EF,
P(E
) -
Mtn per
symbol) oymboln
j)P(Fk) the
expreesion
an averogo, of
all
transmitted
quantity
which involved
latter in
as the event
information
transfer
way,
of
the
communication
the
, ti
$.r
', fly.
C(P,
q)
c 1002^'
may be considered
when a sentence
as the
p is
information
and
which
is
tranoterred
q rocoived,
transmitted
a sentence
or vice
there
later. in
oxioto
a valid
analogy
however
of
this
kind
we shall
consider of "the
justified one
thu
speaking to ".
sentence
gives
respect
another"
information
transfer
yentance
3.9
3.91
Entropy-typo
If proceeu the relation of tho in
measure
as above in order between to in section 3.5 wo use
get
sentences
an analogue of rate
entropy
fundtiiental channel,
preceding
paragraph. We take a weighted of just truth-states introduced. avorage, of This over p and gives all q, of us the:
formula
R(p,
q)
Y(P<1) lOg2
PCP q) 2P(P)P(-q)
P(-P-q) P(-p)P(-q)
P(-p-q
)1092
--
Y? I
113.
The formula " p" the by first are "tp" wifficiently and vice verca to. =i, gjSn--oy=otry obvious simply and the properties if wo notice amounts second to and of this that replacing;
interchanging fourth.
and third
It
is
of
course
also of
In view
R(pI where H(p) -P) is tho If we have case if R(p, p and q)
etry
we find
p)
introduced p, q is is
11(p)
in para. or 3.54. contradictory the
neasure one of 0.
necessary also in
This
general
q are
independent.
3.92 fled
but the ne,
This in
aloo next
being
so, of the
jucti-
opoalcing,
of the
not
entropy
moroly
of a sentence,
In
transfer
between
sentences.
chapter, describing
however, this
we shall meaoux e.
introduce
an important
way of
3.94
In
this
chapter
wo have
introduced
four
theory. C (p)
iaeasures
Firstly, --
of importance
we have
in
the
connection
logarithmic we have
with
inforiaation
Secondly, this
measure
by an averaging
operation,
entropy-type
have two
measure H(p).
zn
of relations of
Than we:
betcteen dependence
corrospoizding the
measures
sentences:
1oGarithtic
coefficient
I/4
C(p, R(p,
q) q).
log2
c/ab;
and finally
the
"entropy
transfer"
We single of the But which their theory it could make analogy of to,
these
out
from in
the
others with,
because
and interest
connection in chapter
information also be
2.
mentioned. these
properties:
the
._j; .
4.
of in to
and that of
a logical kind. In
language this
introduce certain or
wo shall sign--,
measures
considerin
thorn ftndorstood
be measures in logical
as usually of ghat
wo might see) to
call
"unsigned This
shall
"questions*. be Soneraliscd
permits
measures
somewhat.
4,1
i norwri the "ig
A zigi-zymnc1rical
il of the sentence
neasuro
coxcecned.;
virtually
that in.,
it
has the
same value
for
a sentence
p as it
has for
lf
the is or
-p.
Now part
of
the
definition
of true
"sentence" or false;
something something
which which
symmetrical
measures
are
considered,
however,
to be giving setting
proceeding concerned would
with
away with
the
other;
up ontitie3
to is uce
a certain
such
property
and then
the property it
a way that
Procedurally, to think of
be bettor our
on which after
defined unnecessary
as themselves to specify
which
becomes the
of
measures. This involves were sotting represent call is up a now sort aLlbig; uously such like is an entity what not itself addition
which or its
can as it negation.
We can In a way
this
a "sentence becoxacs
f, orru" ; it
one when
completed
by the
way of
looking
sentences" question"
e. g.
something
possible plus
anstiwors
can be true
false,
and hence
a question
i7
is to sentence
answer
analogous
an unsigned
plus
sign.
4.13
convenient corresponding symmetry notation with
This
for
analogy
us: the
moreover
unsigned "? p".
suggests
sentence The sign-
inherent
a question equivalent in
that
"equivalent" careful
questions:
a more
definition
be given
below.
We shall also sometimes denote a question
by the
letter
"Q",
with
or without
a suffix.
4.2
moan not simply a
By an answer
"sign" (as "yes"
to
a question
or "no")
we shall rather
?p.
but
to
a statement.
Thus
p and -p
are
both
answers
Emphasis
because need not it
is
here
necessqry
to
only
a question
be a complete
statement
sometimes
less;
e. g. that giving
noun.
it
by simply
a proper
my name, which
Here the it is
necessary
out is
that
in
the
context
giving that it
have
my name is
all
equivalent
in
to making
general
my name;
the
and that
answers
of "false";
statements they
statements;
and so on).
We might
11S
"no" answer;
and
similar in
are
coded
and that
verbal
a form-
on of
frequently such
we need as those
which'have constitute an
exhauttive such
statements first,
being however, it
"answers".
d4finition we should
ordinary might
which
that
a question
(ii)
is
that
representable
the possible
as the
answers
set
to
of its
possible
answers,
a question
exhaustive,
that
answers
to
a to
are mutually
each of these
We shall
attempt
We want
to
syy that is
knowing
to a question is.
question here: is
needs demonstration
connected
!1q
above
that
an answer
to
must
question
by
name of not
possible what
equally is
principle the
answers answers is
that
possible
the
first
statements
"Luxembourg
Europe",
Asia"
as a set Ecuador,
equivalent
one asked.
It
might
be said
in
some cases
that
a question For
continent is
without
knowing
in the
answers
"In
are.
example,
is without or
what
a continent of the
continents is is
however,
irrelevant
the
word
"continent"
a generic
Europe,
Asia
etc.
is
logically
independent
ob
anyone knows it
------
to be.
.... >-s__-
hLo
(ii) to
for
the
set
illustrated
by the whiih
beating just
your
because all
indicated I,
l19 ical
cover
which it
the
" is
because
unimportant instead,
is
Honolulu?
" we should
be forced!
invent
the
supplementary which is
answer required
into
at all",
i. e. the set of
answer
possible
to make the
sot.
answers
an exhaustive
4.33
answers following continent or Asia, must also
(iii)
be mutually Suppose
To see that
exclusive, on being to
the
possible
the which Europe, that I had now Asia, I
examples is or given
reply
objection
be put or not e. g.
another
"Either answer, by
cannot
be a proper excluded
other
answer
answers. the
exclusive,
and this
things
.sr.
=., , ,
izr
by "completeness".
4.4
reasonably in accord
Our definition
with usage;
seems,
then,
to be
of course
and we cannot
It
consequencea
are
of
some interest.
First, in accordance let us notice with This that answers to our that definition, (with an answer can one
never proviso)
follows must
because
answers concerns
from
a question
one possible
4.42 If a question
is
prove
one possible
answer
a tautology. Thus consider p21 """ a question pn" If with these the n
possible are
answers
P1 their
answers
exhaustive, pl
disjunction '7 Pn
v p2 '7 `
case is the
the
answers "law
excluded
however
a question is clear
possible
answer,
that
disjunction
llfntY
zZ
will
reduce
to
the
single
statement
pl
itself.
Hence
pi
must be a tautology.
Moreover, since with of all tautologies answer with are will only logically be
equivalent, equivalent.
all
questions
a single the
We shall
speak
question
-one possible
4.43 theorem would
answer
as the
We might
"empty
note serve is to
question".
in passing that this of which
as a definition a statement
some question.
of the
theorem to
a tautology, follows
possible
This
from
the
reqqirement
that
answers
exclusive.
4.44 strain case of the just meaning Every of the question word has an answer. to cover We can the
"exhaustive" when it
one possible
answer,
specifies
that strain
the it
answer to is
cover
answers in
a null
could
possible answer.
question 4.45
Thus we find
logic
of questions
'-
differing If the
from is
the
logic with
of
compared to
the with
nothing
compare
question". turned,
no possible answer
which
results
Certain however,
of
the in
logic logic
the
To a certain with
calculus". to refer
as the
"statement-calculus"). 4.51 contains it first Thus for We shall say that a questio4 Q1
Q2 when from each another question is possible to deduce an answer to example " is and the question in of latitude the "In which
continent "What is
question
Ecuador's
highost.
because the
deduce
continent
in
case of the
....: ..
correct
answer,
but
in
the
case of any
12+-
the
This
other
use is
wo shall
consistent
use of the ?p is
4.53 of two
above,
i. e. in
saying
that
equivalent
the is
questions. two
This
is are
the
questions are in
together, questions ?p
answers Thus
answer binary
both
once. ?q the
J in
(which possible
we shall
the
answers-are of these
p-q,
that
none
? (p v q),
questions.
if
answers
their
join
Ql + Q2 answers,
be a question of
answer
be seen that
either any
question Q and
separately;
and that
question
the
empty question
is
equivalent
to
Q itself.
Every
.. -m-F-.
t2S=
question
contains
the
empty
question,
and the
empty
question
4.55 to this their the
contains
no. others.
Questions may be classified and it of is clear according that
number
of
their
will
content,
a question answers setting however, appropriate and natural subject lot than
Q2 only Ql.
Q2 dove
Banking
questions
up a "linear" that
interpretation. us consider
before case
we proceed of binary
que; tionil.
4.56 is twofold. In answers In the the is The importance first the second than the are place simplest place, those of binary questions with just two
relationship of that
questions
of
any
finite
i. e. the
order
into manner
can. be broken
of
down into
questions
binary
questions,
after or parts-
sequences of counsel
yes-no
asked witness
ciloss-examining
26_
cipanta: In logical
answer
to
a riddle. of of any
that join of
finite binary
order
questions. up to
questions number
answers, to the
questions
required is
represent smallest
N possible logt N.
answers
4.58
that should arise words oven questions
It
with in from
seems reasonablo
an infinity this the way. use of of Such special
to-postulate
possible questions answers typically
be reducible in practice
such
as "what".
"when",
"where",,
forth:
e. g.
"What. is is
the
length
of this to
In a
possible questions
as a rule
of binary is
represented
by the
of their
partial
j oina" .
4.6
possible it of first section measures becomes 3.15 of clear will
If
we turn
"content" that our
to the
in the
consideration
case; of
of
most
naturally on the
criteria" That
based iss.
_!t7
If'Ql
contains
a (Qj)
Q2 then
?C (Q2)
with
equality
when we have
number of
equivalence.
possible answers
Since
to
if
Ql
contains
Q2 the
Ql must
bQ greater
to Q2 it is
than
clear to
or equal
that number
to the
number of possible.
of content answers
answers
topologically to the
equivalent
the
would only
satisfy
the
criterion.
of metric,
on
by adding
present
an'1idditivity
purpo io ,
for
(A different below).
independence can
criterion"
If O(Ql
Q, _and + Q2)
Q2 are
independent, 0(Q1) +
then 0(Q2),
where
as before
the Join
of the
two
questions.
Now let re spey t ive ly. sense, the their If join they will
Q1, Q2 have. m, n possible,, are independent have mit possible is a function the in the
answers
answers. f(x)
content
number x of
answers,
additivity
I2$
requires4of
this f(mn)
function
that f(m)
it +
satisfy f(n)
for
any in, n.
This
leads
to
the x
function. the
Thus we shall
C(Q) - log to Q.
us to
to the
questions is
representation; a scale
this
of course
a matter
of choosing
If of information, bears
capacity capable
back to the
technical of of
a discrete
nuah. a definition to
In
a close
the
the
definition
care n of of
(section of
transmitting
a number
symbols
of
to
oqual
logt
time-duration
n of binary unite
the
channel
symbol.
capacity
is
equal
the a channel.
per
situation
a man at
the
receiving
He is signals
wo can cyrnbol
what the
the
possible
received
and prior
him
to
reception
the the e.
of a symbol
question cot of the our "Which
as asking If
himself
be? ".
an a logically operation of
regard of
as a postulate
then
this
question
are possible
has exactly
symbols,
as many possible
and the capacity
answers
of the
as there
zq
per
symbol
is
exactly
phe
of
this
question. that
possible
information
channel,
and is
symbols statistical obvious
equal
are
to the
equally redundancy
rate
achieved
in It
probable;. is zero. of
immediately
becomes or
that
the
concept is give
statistical also
redundancy, to questions; of
concept, will in
applicable
meaoure cppacity
analogous
but
information
rate.
Let
us now-assume
that
the
possible
them.
as logical chapter.
as in
the
4.71
Pl, of p29 pj". pp .. * Since
Thus lot
and let the
answers
probability
''P(p4)II are
answers
and mutually
exclusive
we have
P(pl
s
0
Vp2v
P(pl)
...
vpn)
+ P(p2) +
0 ..
P(Pn)
4.72
a measure
/3n::
of content
In order to
to
satisfy
our
our containment
ddditivity
criterion
as before.
a
apply of two
criterion it is
"independence"; are
questions
independent
answer
to of
the
first
is
independent to the
in
the
probability
every
answer
second.
answers
Thus given
(pl, ...
pp) said
independent
Call
j,
k).
criterion ad before., If in
may than
be expressed
in
the
addition
to
that the
these
it
criteria
we require of
when
measure measure
already as exactly
introduced, that of
A(Q)
This to may be proved Shannon's 2.44; requirement containment
areadily
P(p1) j
in
1062 P(Pj)"
closely analogous above in
a manner
"tuiiquenees where is it
theorem"
mentioned that
Shannon's of
approximately
conjunction
and additivity
131 ..
'
refer
to
measure consistently-
as the,
"H"
is
a in measure)
measure
obvious
analogy,
of
"entropy"
as introduced
by Shannon.
notice taken the in particular the sot that of is the
over
answers; measure
averaged in section
content in
could average
approached
the
possible
answers; one in
where the
the of
contentthe previous
sense of
measure
content
should
immediately nature
the not
be appropriate
to average
a non-additive
quantity.
4.75 anstiwor , the account probability, time, In averaging of i. e. each the answer content is of possible into It
taken factor.
a second
as a weighting
in
appropriate
here
to
say something
about
the
sort
t
132-
of
probability of Ir-range of
that (or
is
relevant,
whether sense) In
logical or
in
the
sense in the
empirical
sense the
relative
theory empirical
generally that
assumed to
messages
be sent
over
advance the
hand ones.
however,
sight
completely
obvious
that
it
is
legitimate as the
or weighting
appropriate factors.
4.76 we might
to use logical
probabilities
the
measure
which
results,
something
as followst
determines of
(or
section,
or
we mean the
by
the
consideration. equivalent
a question
a statement
to the the
disjunction
of the determined
in If
compartments definition
"L-range"
of probability
adopted,
)33
ft
ability relative
of
such
an answer of
will
by
the
number or,
compartment; of possible
relative out
universes such
operation, taken
an averaco in for
be cases, one.
however,
which example,
appropriate
Consider,
a questionnaire
issued Let
to
of
us suppose the
...
anzwero
and that
are which P(pl),
logical
P(pn).
of these
frenuonciec; depend,
these
answers
concerned: . Now in
lot ordor
us write to
then
an the P(pj) of
calculate
content we want
if
logical factors
content
answers
). Thus
received
for tho
weighting
logical
be the
the
avera6e
we have
expreeeion
11(pi) 4.78
a formula of. Wiener
logt
1'(p).
This
(rnf.
formula
4}3) in
has: affiliations
which ho refers
with
tar our
',
"P(pj)" as
as "apriori
probabilities"
our
"F(p,
)" is.
"aposteriori
the., not he
discrete; In both
does
howevr, the to
oauy in
a number
purpoooo
our
not
an "logical" calculated
probabilitioo in terms
probabilities
1ogi&al
form
of the
pj,
but
simply
as apriori
probabilities, in
regarded
an in
is
theory,
are and
howeverzi
to
apriori
probabilities
frequencies confines for given himself him never" that
assumed since
be relative further
in to
advance;
Shannon it
becomes probabilities
axiomatic
apriori
"almost "mixed"
differ, as the
no need
for
formulae
auch
questions 4.72,
are the
independent entropy of
77-7-r,
das`
their
Thus if
join
iaiequal
to
the
sum of their
entropies.
independent H(Q1)
we have + H(Q2)"
In
be less than than the
general,
sum of the
the
entropy
of the
though Q1, Q2 C
join
greater
will
either xQ1)
soparatgly. Q1
Thus + Q2) a
AQ + Ql)
H(Q1)
+ g(Q2)"
We have
Q`;
H(Q1)
H(Ql
case
+ Q2)
is that
if
in
and only
which
if
Q1 contains
ompty.
a particular
Q2 is
4.81_
ii(g1) As in + H(Q2) the theory over of
The quantity
H(Q.1 + Q2) communication,
which
is of
is
the
excess
of
special it
N7o write
and define
R(Q1,
It Qi measures (or Q2 on
Q2)
as it vice were versa).
H(Ql)
the
entropy
4.82
algebra section" however, defined noticed of the the to it the in joiin
Unlike
question-calculus parallel that-of
the
class-calculus
hays no concept
or Boolean
of "interof this, just be
"Join". to give
In the
spite.
entropy". related to
the
probability to
disjunction
-- ------
related
probability
136
of their
4.83 justified all easily
conjunction.
The name "intersection by reference from join: to the following entropy" may be
properties, of the
derived of the
comparable -
proportion
information
For
(i) (ii) is empty 0< If Ql
Ql and
< or
Q2
A(Q1)" iS either
R(Q2,0-1) independent,
" Q1 "
0.
containo
if
Q2
H(Q2)"
opocial caeo are of Q2 empty). as (for
the
entropies
defined
example) wix
then we have R(Q1, Q2) H(Q1) g(Q2) The extensive in 4.72 terms of probabilities is Q2) P(P. P(P Qg) 1092 Qk) (in _ for fQ2(Ql) H4l(Q2), R(ql, of Q2)
H41
0'
section
above)
R(Q1?
"
average;
and the
quantity
averaged
is
the
logarithmic
measure
of
introduced possible
above in answers
section joi3ht
the as
3.81.
The average'ia
question probabilities
to the
are
weighting
factors
answers.
before
ac vvs.
apply
probabilities
an wr ighting
4.831
The quantity
R(Q1,
Q2)
corrspondo
with
Shannon's
If
"information
we imagine
rate"
Ql as the
in
the
case of a noisy
"What "What the is is the the
symbol? " and Q2 res the (-" symbol? --), or vice und allocate versa,
measure in
symmetrical with
symbol
frequencies the
then symbol
R represents in which
per
ease
successive such
independent. in
however
consider
further
Returning notice to in
briefly briefly
to
the
eubjest this
of sort by
measures it
the
which problems. of
to
statistical correlation
non-metrical
of
3$ 1
which example
we shall of
give
(Thin
is
a classical
Galton, Lot
we wish of
to
measure
the
degree such
of hereditary as colour of
Eye-colours
n categories.
and their in the eldest
A number of observations
Sono, table of and the (matrix) results of
is
are
made on fathers
aummarisod P(E17k)* j parlance
numbers of In
finding
a father k. table".
eye-colour otatiotical
whose this
son has
eye-colour
a "contingency
By adding
of for get values the P(Ei) fathers, for. the
up rows we should
probabilities up of columns
get
a not
a similar
eons.
The usual
in
calculating
a measure
of correlation
P(Ej)P(Fk) assumption the old which the
would
(an of
be to
estimate
produce of the
of values on the
with
tgking section
value
An alternative
is
to
compute
the
(QlI
Q2) _. k
1'(E
) log k2
I3'7
which joint
if
desired
may be "normalised"
by dividing
by the
entropy
2 j, k
which
P(E jpk)
lies
logt
P(E jFk)
0 and 1. sense.
between in
for in
the of
the usual
complete
of the
dependence
is of uniquely prob-
sense by is
sons matrix
determined abilities by
(the or
a diagonal of rows
matrix,
interchange
and columns).
4.842
appears fathers name is symmetry for count fathers
"sentence-symmetry"
respect to (--the
as a
blue-eyed positively; to
children for
so would
a tendency
have
green-eyed
children.
4.847
cut of of intuitive information information sometimes in
This
meaning, nay from react which
moasurc
hau a relatively
that statistical Certain of the if the
and
back it
night
accrue units.
fron
non-normalised, a succession
measure,
binary
example,
of values
wore computed
for
r , tccos , ively
finer
subdivisions
+D 'J
of
eye-colour,
give, of
whether
applied
categories
also
in
detail.
4.844 class
by
, that could
Gabor
This
is
not
the
only
measure work.
of
its
be used in
(ref... 20)
statintical
discusses
A paper
D. and A.
the
closely
similar
neasure
2j, k P(EJFk) logt of P( )` logP(EIFk) PCI; )rCPk) and connected with
called
"coefficient
depndence"j,
a measure
called
the
"diversity" VIM
j 1062 2(EJ) "generalined entropy"
Elaowhoro
the
use
of
the
Z
has been proposed,
(P(E3))12(1062
222). Our
p(Ej))n,
"entropy" and "inter-
(rer.,
section satisfy
entropy`, both
however,
are the
only
moacures
which
our containment
and additivity
criteria.
4.85
Although
our
theory
of
gquestions".
resulting of the
measures, "sentential
to that
the
quantifiers logical
char-
no special
(4 t
attaches
to
thoir the
generalioation methods of
in Kemeny
this as in
be limited;
of Seneralisations infinities.
we continually
run up against
attaches measures
to H(Q) of
the
problem Q2)
directly in which
logical without
terms, the
language of
defined.,
mediate
use
probability are
is the
concepts. definable
sense probability
Since in
the logical
probability terms,
themselves
in a certain of the the
one. perhaps
use
tends
rationale
concepts,
irrelevant this in
terms
concepts
is
of course
of
derivable
the 1jo3i of Lot
from
the
it
two
so long
questions. then, content of its
as we have
the
us return, the
to of
the the
wo had simply
just of
dofinod the
terms
nunbor
possible
we have
C(Q) of binary
1002 n,
an approximation required to
smallest
number
questions
express
14z :1
in relative
an extensive content
binary measure
form.
a,
applicable namely
Ql have possible
p1, "..
pm and
q3 implies general
say inc. n,
izaply be xj.
pkj s Thus Pk
number number x, pk j is
possible content
answers of 0.2
Q2
iyn to
relative
j; the statement
and logt
define Q1 by Q2s
the
of
Q2
averaging thus
quantity to
possible
answer
cal which
members
answers
to Q2.
0
Thus we define
n Q1(R2) that logt x
We easily
prove
CQl(Q2)
Now let
<0
02)
us fact
suppose the
wo have of
a finite elementary
languago,
and that
in
number
1,43
sentences speak
in
the
language
is
I.
In
this
of the
riU". are
"universal
This is
question"
the question
of the
whom of the
which
we denote answers
all
(statements 2N;
describing
accordingly
question
others,
Since
from
a state--description
truth
for with
or falsity
any our question above
of any sentence
Q we can definition. calculate Let
of
the
1anguate.
in
Thus
CQ(U),
accordance answers
Q have
possible
pl I p2,
" ..
pn;
and let
yi with
bo the
state-
descriptions
concictent C (U)
In terms of the
"
express
of
this
we can
H(Q),
the
entropy H(R)
qucation P(Pj)
wo have P(Pj)
j22
N-
1062YJ "
The cecond a sun over term all is equal to CQ(U) since s Thus it, is in effect
state-descriptions.
we have
H(Q)
This might have
w
been
c(U)
used
cQ(u)
definition of H(Q).
as the
Notice
that
it
has3something
of the
form
of an intersection
144
Up to of
the the
we have relationship of
largely between
"engineering" chapter
logical
chapters in
We must
now examine
relationship
more
debil. is of some importance technical because, theory that concepts; "semantic the engineering strongly there
Our enquiry despite very is disclaimers, often virtually seem to writers imply
on the
95-98)
suggests the
context
a communication least to
channel
technical
theory
a significant
degree"
also
5.11
a semantic
theory.
Relative says Weaver, to. the there broad seem to subject of be problems
14g
Level
symbols
of
problem).
problems by are concerned
the identity, with in the interpretation compared --even when with the
or
of intended is only
and involved
speech"'. communicating However, in speaking of Shannon's "admittedly instance in the first which applies All, he goes on to say: a to problem problems "Part theory comes from of the significance levels that the fact of those signal of the Wand 0,
relatively
accuracies A. at level
A necessat Level C. larger B But levels a part and at of the that the analysis comes from the fact at that this level overlaps the other suspect. a significant
possibly more than one could naively Thus the theory A is, to of Level at least degree, levels theory B and C". of also a
5.12
problem" (and
Weaver's
particularly of
conception
the
of the
"semantic
problem")
"effectiveness
is
..
little a
-- _ -_
vague,
_ ____--
attempt
to
go into
detail.
I"
In the at present least concepts context a model of we can be more of the "semantic 3 and 4. a definition least in the exact, theory" But of since in we terms
have of
our
chapters for at
have
provided
restricted of this
the
technical
completely mathematical
except
as regards
first
try
to
the in
analogy
which
becomes
for of as breaks
however,
context such
a natural that is
contend account
analogy
misleading.
Let
us
imagine
a finite specifiable
though in
universes" For
a logical
language view of
calculus;
universes so initially of
be equally they
probable,
approximation)
further
into
equally
probable
`-
constituents.
In
such a
r47
situation
any
of
the
logical the
language senses
any
one of
an associated 4. For
most
analogous of section
technical genoralioatione,
their and
we have,
statement
M(P)
and for any question Q with
1062 -V(P)
posoibl answers p1, p2' "0 Pnt-
xCQ) In both
probability of 5.21 radically of of distinct, possible
-G the
sense
P(Pj)
" logical
the act
these
in
cases
the
symbol
of
denotes
in
truth-frequency
universos. Now there ways in are: two which into the this different, technical picture. and concept One channel the terms physical
inside
universe' of
transmission
a message
can be describod
of the
language
concerned.
consists
in
the
of communication of messages in
the universe, i. e.
ab_ ut
language. in turn.
We shall
two cases
148 l
system must
is
to
be
be a finitely concerned
logical
language of describing
be finite,
be incapable from
transmission consider
messages
an infinite of length, to
elementary a finite be
Since passage
no bound to the
specified,
limit
might
present
however,
as a matter made.
we cannot
has been
5.23 process,
sequence such
Let in
of
than,
some "physical" a
our finite
N symbols being language
produces
at
times from
symbols
chosen will
an elementary to include
Our logical
be taken
statements
(i
- 1,
jQ1,
...
...
M;
N)
the these
Si at time
tj.
logical inconsistent
two
relationships with
symbols
e. g. that
t occur unless at
is
(---
different such
the
same time
---);
relationships
express
There
apriori
may-also
knowledge
be constraints
of the process
of some sort
conceirned.
imposed by
what we might
call
"natural
laws",
e. g. Sl might
always
I4-q
be followed
we should
have
ti)
such the
D
apriori
P(S21 t j+1)
knaledge Quite any to
Ci ' 1 ...
be built into we shall the
N-1)
all of
language. with
generally, p in the
associate probability
language
P(p); to of the
knowledge of our
probabilities knowledge
be the prococo. us
extent
apriori 5.24 qk of
consider describe
the
completely
i. e. which
specify
complete
time-sequencs
of
symbols.
be a conjunction
each value of j;
of N statements
the total will number be MN'. and with
(assuming
no constraints)
The q,, are each (If qk the 5.25 message of there are qk
exclusive; can
them are
constraints
inconsistent
omitted).
We. ehell
content in
0(qk) terms
of of
a its
can be specified
logical
probability:
C(qk)
1062
Similarly, i. e.
_.
if
Q is
"What answers
the
question
15-0
k It. of is. tempting the to-refer This of the our to. this might, messages sequence; model,
an interpretation is only
there apriori of
one message ones. of the riori expectation 5.26 used the to The
weiphted
to
the
answer or
11(Q)
express of
by dividing concerned of
number
the
message if
similarly
content
second
a measure
associated
with
the-tJ).
.503
We now asserts
these
measures
are
a generalised communication
entropy is the per content is the
version theory.
symbol per (or
of
tho measures
employed
by message
Summing-up,
per second) (or per
we might
as defined second) whose of
say:
symbol
"What embody
" in message?
a language
logical
the
apriori
Our assertion,
at a number
of points.
sr
5.31
of a possible ensemble All that
the that
first our
let
us is
dispose for a
an infinite connection is in an is
this
a definition sense since the to former the,: limit such introduce would to the
all
under
passage to justify to
in
we should on our
have
model;
cases passage
applied,
limit
no difficulty.
Now let
us consider
has so fur
minimum implied
device however, which that laws",
i. e.
ensemble of it But
one member when known into (or with account. "stochastic") random processes This
postulates that
language
moans process. in
general;
he restricts
rsz
htmaolf to processes (a) (b) (e) (see 2.54 2.564) -
which
are
sections
doubly-infinite
The. order
in
in
time.
which these since until roetrictiona we cannot we have, at
countenance least,
passage
a stationary
ensemble.
ergoicity
implicc
theory
of
formulated convenient
(c), But
of infinite
(a), to be seam
of generality of finite
purpose of
would Markov in
development the
a theory to
restrictions
be introduced
above.
5.34
a constraint Thus for on the example
Each of theso
logical structure
restrictions
of our
represents
language.
statement at,
symbols the
commencing of at the
as equal
probability commencing
statement tj+l
same group
occurs
time
(provided
153
is falls
meaningful, within
i. e. the
that
the
time-shifted
signal Similarly
still the
allowable of
definigg by
property
a-Markov of
specifying
time-instants u (which
by more very
another 5.35
restriction
on the
property in
is
of
cspdial for
a certain of
sense,
statements
symbols
as calculated of
a particular requirement
an engineer frequency"
might
definition one. In
probability
favour
ensemble of
independently the of calculation zero The in theories etc. our model of (see of the
is
except
a set
measure
of
"writing-in" can, in
various the
follow
concepts in Our
references principle.
chapter assertion
no new questins
measures rieasures
agree is is
with thus
the
communan
scarcely
model it the
sufficiently of
general
models
a communication of Shannon.
system
requirements
If simply
we consider an information
ac
stern source,
of
connunication we find
again
can for Si the
nothing
express
new in principle.
two sorts of statement, the ta)"
If
our logical
'lp(Si,
lanuuaLm,
the at
sentence t,,
time
sentence (whore
recoption
of from
symbol different of
Sk at
time
finite
we still Let of
possible are
univorses. descriptions
Q be tho
answers the
and Q'
corresponding of received
proceedingJ=x
as before 2 k, l
we shall P(gkqj)
the
intersection
entropy
Q and Q'; and in exactly we can use this as a definition of channel. we might notice in
paocing,
appears
correspondence
between
"transmitted"
and "received"
messages.
,
the
technical by turn
inside
described
language, to be not
merely
we suitably
organise
language
knowledge conclusion
of
of the
communication
information
generality
tedbnical
5.39 however
theory.
This throws no light method of relating of the two theories
on the. theory
communication
with
regard
had by model
to Weaver's
to the consider
"Level
the
B";
because
of for only this the
vie have
symbols purposes
meanings system;
communication need
the
they events.
be considered enters
physical
"Meaning"
at
a higher
level, the
i. e. at
the
level
at which
statements events)
occurrence
of symbols
(physical
Let
us now turn
to the
system
the universe
capable the of
logical
language, about
statements
universe. to
case the
make no explicit
reference
form
statements that it
in contains
our
language: elenentaiy
'we'shall. sentences
simply p1,
and, logical
ofthem. us, imagine channel, of that, there. events over the at is the transmitting
end of
our
channel.
have
question,
the
logic
propertis'of information
channel
and -their
we can explicit
this details
question, of the
we must the
be much more of
choice
messages,
and the
concerned.
5.41 to consider
It soma asses
will
be convenient ospecially
for
ua first rolations
in which
simplo
hold.
(i
communication observer i. e. until has takes
Firot,
place over
lot
the
ua suppose
channel of
that
no
tho
until the
made coxipieto in
obsorvgtions to transmit
universe,
he is This whether
a position will it is
description. of thepJ,
way of
coding
such
a message
as an n--digit
ioprosent,
if and
in
order,
of the
the
truth-values
are
universe
taken n binary
a lo3ical (i. e.
symbol
of the'iuiiversal
with Shannon's
quostion").
measure to such in any
This
sense
is
in
clearly
which
in, accord
the latter,
an ensemble:
properties specificwe
'get .
complete
It should
is not
necessaxry, before
that
be complete
we can imagine
of ,
them proceeding
that or at the least the
course,
before
transmit
infoxnation
in '1.ncoiplete" or denial df
an
is suppose of
also that
in
the
following are
describable or their
negations, arbitrary;
order
made is
and that
as it
is
r1ade . it
If
the will
order
of
the
observto
be necessary
number
falsehood.
-, Given
involve of
(n/2)
ctoad occur,
that,
possible number
n say the
the in
transmitted). information
Wo can embodied in
addition
observations
that
the
channel
messages
transmit
been to wasted.
the
made.
information
'Wo have
such-and-such thir3'a
have
ditional
a necessary
who
systera language. in
capable Lot us
of
"oontonco say p,
expressed of
disjunctive
as a disjunction
state-descriptionz;
logical by saying
specified of the
language
whctlier
Given
n elementary and if
sentences these
2n state-descriptions,
we order
___
,,..
some way we can. code such . of that, jth the jth digit
binary
number, of absence
presence
the
statte-description
disjunctive..
normal
expansion. 5.431 , -binary digits the fora highest (and is the clearly This single content average not method of coding of involves: 2n
sentence that, is
rather
coding.
"optimum", to message
message-length Never-
adjusted it is beine of
according clear
probability. that
channol'canacits unconnected
a reason
details
5.44
from is to the above or
In
examples. apprbxinate agreement for i. e. all that the
fact
If
a general
the the logical
principle
content entropy,
emerges
of thorn and rcoiver) a message,
equal be prior
massage
must
(botwoen
transmitter of the
transmission
aGrocmont
on an ordered
need, be transmitted
are
of the
5.441 case, sent
answers.
Example where the successive answers to (i) digits a finite the message above of represents the messagg of such reprebinary can be a
successive
sequence
questions:
alternatively
as. a whole
Ito'
considered the (in 5.442 questions question. 'ing prior Join this of
aw a coded the
answer
to
a single of
namely
case
the
sequence
of
Saite
as a single-finite by suppos-
(between Q. finite, PA
We call
If
the
answers only
to
the raecsage-question
appropriate binary question equally 5"L. 5 we could answer in This i. e. with to to digits
A is
are
probable,
of
binary fact
logical
expected. of
lengths with
messages
would
be equal
logical
/61
5,5
and. "logical" does about .. the not the provide theory theories
This. method
of
of relating
the
"technical"
again
information, of
howovor, Weaver's in
a justification of ','Level of
speculations view of
restrictive
',nature be emphaeisld.
involved.
there
is of
in
ordinary'overydar port
no agroenent
this
us imagine
fas
the
following:
that it is
I want
about of that to
to toll
rain. of In
a friend
order td
something,
dispose and coding,
the
moment
probability both
and I exhaustive
understand
that exclusive
probable: content
that
my message unit.
5.52
stance involved is the in which no more circumstance been understood at simply I
Now there
could that in give one unit which between a particular saying it
is
my massage of
a form
should
infornation e. g. by
and in rain
"yea"
for
for
fine
(The actual
of course
matter,
events if going
if he it no But
inotanco, ",fin it
`-3
had just
It aukod might the
askod mo tho
aluo quoction
quoation
of but that
to
bo po3uibluj explicitly
frond" done
Saco,
provided
he understood interpret of
conucquontly of thane
atypical
conmunication. or
Aci a rule
we have
nay what
we want
Lay n ro
leas
fully
And this au it
involves wuro,
not
merely what
euying
point
should
that,
at
least
in
tbo
oxnmploo
cectionc
5.41-5.43,
there
nonuoncon
identification it in
coutont thuro in
even if
assumed that
"nancago-quontion",
wo should
lutvo
to certain xanplo, e.
statistical a Markov it in
unions,
gonoratod whothor
Now
wo nay uoriously
concidor
the
d3.
sequences processes.
of
messages
are
to
be regarded
as
this
question so long
as we consider
relatively
or words,
the
! structure"
specified
of a language
statistically. about longer
such as English
But sequences; `, nothing and the may run chapters. even and
be doubtful
all, less
1o iCRl count
Content,
thread-of through
an ardent a number of
discussion
paragraphs,
pagos,
5.6 another
with ferred in the to
is
relevant (ref.
It This
by Shannon
of English. 2.579). of
already
measuring;
English
consider
relatively
passages; a simple
statistical
approach
a passage
guesses,
and is
whether
he is
or wrong;
re4.
wrong, guesses number second -end in of is he is asked to guess again, and so on until; down "Guess until he the the the now has letter numbers
correctly. of guesses
letter",
process'ie The
passageis.
reached. of
numbers, suggests:
the-message. Decoding, it is suggested, twin" with in of the would the tont be possible subject; in every
if
one could
find
first
circumstances
idontical
"Guooe
exactly
before, tho
letter
subject could
original
message
be rocovered.
figures. :1hon the pcrforiaod in practice, large greater and, the to fir: the 3t half list of of This of this o porimont obtained twos, to in hac and be highly
ohoy3r3 it
redundant,. this
redundancy be at least
o:f English
au eutimatod
in
way appears
/6s
cis high
when 90%, as
long
passages,
are
used.
But
how
method in
onos which
straightforward long
twenty length are the
does not
consider a very logical as it at
work for
passages
pacoagdo
letters ao far
of English.
long: this,
iuppoco
is. aurely
concerned, of
guefliiea> of this
tout back.
loa8t
number of
lo-Gtore length,
The number
ppoaible of of
passages
a. ieuming is 2720;
an alphabet which is
letters. of 10`7;
pu n; agc. of
ever
were
add up to statistics
letters. long
letters
no mere slice
constitute is
a oufficiontly to speak of
course all,
nonsense
unless
we are
thinking
a sample
taken
0o bold
from
as
some theorotIicRl
to lay
enoenblo;
of
but
tho
who would
be
down a definition
charactristica
of a t-heorctical
the centuries the
ensemble
statistics language
of
such a sort?
of a lanuuaCe
lAoreover,
must change
over
appreciably
as the
changes,
and the
ensemble
could
1"
not
be considered
Bhannon', s ,"redundancy" of, statistics in '.Shannon's-. been -alone 5 62 structure'which easily detailed : will have it is not language specifiable in rules; element the most in terms -such :as of_letters. experiment, any But = clearly
the'test above,
influenced-to,,
by-letter-statistics.
English
'has
aMost
very
though
these
in. their
important that in
a natural of its
however,
the
become. closelyadaptwi"to
of
its
users;
from'their is
to the
mental
oztent
that
it
hardly can
The structure
be consideredof to those a
apart
------ all up to in is
--they that
been
greatat
possible might
the
ioctationa o: in
language
be best
doscribablo
psychological that
terms,
than
statistical; would
and
a. description
of language
---:
y".
TV
`lr'Y"^Tfi.
"TcT
+am
fe"^^'""o'-.
mss-. -. w-i.
e. _.
-.
-_"
-_""
1Ei
betaken-into content
will.
account to be given.
if
a satisfactory It is
will
definition that
natural
of a machine
languages
is
conceivable
handle
which
',
with
something
of, the
facility as coder
subject
"identical
while thaw .
possible at would.
remain
be imposoible,
machines which
a, quasi-human task of
kind
"cd, ini".
oioovor ion
ova Hunt
Still
d avr the by an
line
content
as indicated
oxpperiment
for artificial
of this,
sort
and logical
content;
defined
languagen. h'uppoio Shannon's rymbol letter since not, guess in in tho teat ubjoct woro making
his
; uo
ee oyribol of letter
by by
instead tho
previous
we feel, "12";
doing the
8o would content
be logical sentence.
ones,
connected
minimioing
of. the
But thorn
is
a hidden
fallacy
hro..
It
w p.:.,.
:. ,
16 3 8
be the
case
that of
the
mathematics
were
part
demonstration expect'
a "reductio
should
some
psychological
affecting
writer.
a mistake
would
"hunch" that right
point.
figuro -
', Or he might
would turn
simply
up, for
have a
no reason
(And he might
be customarily
The, study the our other study studies of the: linGuistic of logical
of
is of
no more its
than but
users: around
languages
bevolvecl
here ariTin
that
,o .e, ::rental at
what; to
exnectr; or
and if arithmetical
expoctationo
are , imply or
not that
automatically grammatical of
logical,
other
a ratheremore by terms
sketched 4). in
eevoxcal of the
Woodward
(ref'.
p)z.
: . 16q
of von at
in
of This all
games"
of
an attempt of the
take in the
message remove
computing other
redundancy.
however
difficulties
mentioned.
5.8 displays product claim results obtainable, to an attempt of the The theory. to treat but Mandelbrot (rof. 34)
-of language
as an evolutionary does although language with is not not certain are bridged.
alternately Hence
define of
content.
concerning concerned
C%
LANGUAGE'AND INFORMATION
task logical
involves or
something mathonatical
of
plane
to
that
in
the
theory information?
1`a theory
To answer is, the in
purports
necessary sense.
namely
know what
information look at
the
ordinary of the
To do'this,; everyday
we must
usage
word
in
contexts.
.,
it
is
wotth
danger quite
confusion activity is
"informing" Yet to
a'difforent of
"inform"
someone
something
(And there'-are
According grammar indifferently is to
similar
the
turns
of phrase
in most
languages).
our
simple-minded there is
metaphysics a sort or of
on which called
built,
commodity which
"information"
"knowledgoll
we can
i ., . --_-
,.
-..
. n., -.-,
.,.,.
,-
..; ff,. -. , . r ,,
t7!
transfer
from
place
to'placo
or
from
porson
to
person.
"Knowing"
a piece of
something
this
consists
in being
in
possession
someone
of
oleo
commodity,
and "informing"
is. like
6.11
giving
it
away to him.
But whatever into word is is that say that this "know" known used: it is "information" simple has is to true picture a peculiar may be, in logic
not For of
fit the
the
what word
when the
knows out at
something
implies to
(and really to
false all).
we have It is
he didn't tnntradictory
practically -
"incorrect the
"incorroct
appropriate ---
emphasis certainty.
know that
indicate
6.12
Plato that the
The notion
"commodity"
is
at
of
least
no old
will
as
not
theory
knowledgo
work.
me know"
that
course
accept
knowledge apply.
something
inborn
The Platonic
philosopher
tends
to
use
the word
"information"
for
everything
which
its not
7 f
r7z.
use
it
rather all
emphasise of
information, on this is
But
quarrel consider
score; the
wo shall is oven
whether of
describable
as a theory
information.
Lot
us be guided in this
by
the
use
of
pattern. between
particular noun
we should
notice verb
differences "inform".
"information"
and the
"Inform"
"tell". *+r else two that It represents
can be uccdhalmost
a relationship case. fact; someone There it
exactly
like
between tolls is
(typically) someone
people
no logical false,
restriction obvious, or
may be true,
contradictory are qualify ... " None, rather the or word "Ho however,
these
me in
neriou6neea out.
completely
173 e
It
is It
not is
quite not
the
act
of
informs
a book
os implied. is as a rule to without consider (is say contained "Ise informed speak
even
the
this
position
between.
me that
noun
"information"
impose
restrictive
connotation. The noun "information" the verb verb hau a correlate is it not is related not
But the
"inform",
good rather
English the
"He misinformed
so-and-so" in itself. we
"He misinformed of
complete
informed person's
thet act.
than
of
informing
74
the
victim.
(Though
of use).
course There
"informed" is a similar
usual
participial
use for
different formed direct
"misinformed",
contexts,
but
as when or fact
it
tends
saying
to be used in
that someone i. e. is
rather
misin-
about or
with
concerned.
When it
it has
is
used on its
reference
("You own
to
a closer
a particular
6.23
lines and we shall main point
be written
comments is to that
along
to
these
make. is
usage
mathematical course,
srggest
ordinary way.
language
"tedhnical by
scientists exact
relatively
with
a large former
become
additional replacing
the
completely
assimilated
almost
unavoidable
""
7_
that
its
must like
be chi to think
in that
certain ho has
more
the
speak "the
vagueness
distance It Circus
example, of
instead for
and
avoids
saying
Thams. special the for point case 6.25 associations are loss
arguments its
against
within
clearly bettor
accepted ---
everybody than of
Marble Castle
a better it is
reference relevant its habits. if the that amongst the word in the
Windsor
a now verbal If
study
do this meaning
-of
overlooked
unnecessary
created,
the
that
specialists
theta is that
themselves.
ample there opportunity already
And it
for is
would
be claimed
and certain the case
confusion, in
evidence of
confusion,
"information".
77m
; "'"-l
We are in
not
in present. of
all
the It
details is
a case in
that,
ones clear;
due to broad in
immediately is
in
the
cave'
hand. between
an important It other is
course, verbs
a noun,
most
suitably or
bolstered
quite
denote that
example) of of
object sense
of
"communication",
e. g.
Our point,
It is that the to even word.
however,
in =ore
logical.
without and We can never
roference without
and act
an act
information of
we shall history.
some period
We speak
it
just
as if
it
were
of but
we can't
no act there in
takes yet
"information"
We begin
"information" as: uue the some of is a highly
to soo that,
Platonic of
after
word. It
all,
the word .
to doopite
begins ,
the
attributes
"kuowledgell
differences.
6.3 be considered
Lot
us ask directly: Is it
as a commodity? from
sugar or
that person
i
can be transferred
to person like
place
to place
And if
boeawax?
how iQ it
created
eventually place, it
go to? is certainly
might call are need speak distinction_ so insub-
the
sort
of
thing
that People
around
habitually without
"roods-and. even
on occasion
admit
a thing economics
better
described
i..
6.31
that is I its give dietinguishs reproducibility. three but to if to the
is
one
thing Groceries,
in
from have
five two
and eat of
left; it but in
I have
the
can
stop
me from of .memory.
possession
a lapoo this
principle nature
no limit infinitely It
process;
information
copiable,
for
if
I give
someone
of a book, but
he has
only
the
a copy
is
of in
same book).
contained tho
reproduced original,
. -=e
information
as the
errors.
6.32
interesting the of and "physical" communication quantum is (but
These considerations
somewhat properties and in the For irrelevant) of information, fields example, or in more of
could
speculations both
lead
to
about
in
the
field
mechanics. reproducible; of
"entropy" whether
reproduction
a signal
involves
an entropy
179:
change, In the
in case the
the it
sense or not
or
in
Shannon's. in
without
a difficulty
'! commodity"
concept.
For Shannon,
only . , about however, in it. is the sense that this of
information
is a useful of
is
a "commodity"
speaking, in a channel,
way of
rate
information
which matter
the there
say having
that the
having measles,
of is
an infinitely
something its
disease;
the
virus,
reproducible. a "commodity",
virus
might
be considered possible to to
pack
measle But
virus although at
place
place.
neaslea, least it ,
certainly another,
one person
whether
a faintly
comic
Igo
However a commodity
there theory,
is
another that
namely must
information
source.
A commodity But it is
comes if laok it
latter) When I
is of
channel and
"noise". eay
out
window
"It's
raining",
It
havo I created
is difficult
a piece
to
of information?
ancwor of to give,
at
any
rata which I
kind
interprettheory. source
ation
the
technical
purposes
another
another reached by
rays
reacted the
through
detail; motions
of
information
it
wan raining
a. conceptual
confusion
as before. me : ay "It
Let is
us
consider
raining",
disposition that it is
someono of
same chain
events,
I81 'H j
nervous impule-es to and cortical this, as before, it could the serve activity but it nor is is and the not it the like, same version of this is can
describe
as a coded
wo were
traneraiosion )hysical
the
trarsmicdon is any
Technically,
the
problem
of
the
' courco"
the
concept
of
But if which is
this vilid.
of question
An information
we want
it
cannot
be answered. that
source i. o.
an such,
anything
anything
can bo condidered
as a signal,
anything
at all
It
is
abundantly in its
clear sense
that
if
we are
"information" or in to
as current equivalent
ordinary
usage,
sense,
we shall in
have mind.
linguistic
communication
be interpreted
as ruling
out
the
simpler
kinds
of
("pre-linguistic")
such
as pointing, similar
beckoning, kinds of
smiling,
and
as ruling
out
representations At the
languages even
pretend
be able
approximate
statements a secondary
about
information different
in concemi
and rather
us in
6.51 content good
a moment.
Some of as defined approximations as applicable criteria detailed and if of the properties language of the of arc logical undoubtedly
they the
were, sort in
scarcely But
wo have the
described. of
phenomenon definite" of it
case
concept
"makofr
justification At the
there
something io far
as in
information ...,.
is etc. "
sometimes
measured-in
lines,
In
saying
that
inforiiation
is
measurable
only
in
respect
of a language
we must be careful
to realise
153r
this technical
is
not
what
is
being
when, S29
theory,
fundamental in
speaking the
information in
using
sentence.
a use
secondary
between the
case
section
there of
relationship the
between logical
of
section
to
in which
the in
theory
ctatcments
channel. entropies
We found of
messages
only
between contents is of
and the
uhannon' of the to of
obtained 5.6,
in
irrelevant.
The confusion easy$ almost and in some cases that it even is this
is It
fatally might
difficult confusion,
be said
cyotematic
18
ambiguity much of
of the
the
word
is
the
source -
of'
interact
as a whole.
In obvious content defined 1arGp the for else lettors should that of at a distinction passages all in of terms to for natter since
the
simplest, needs to in
cases
it
English, of their
is
proportional
count or
letters
no absolute units).
reason
as the be in
fundamental to
proportion
length
be accepted'. by hardly
If of the form "The it such
anyone.
other in band'we consider oentencee. y of book
on the letter
position that
z is of will to
an S, ", of
becomes sentonces
obvious
content passage
u given in
contents passage is
len,
And it entropy
logical passage
content as defined
related
by Shannon. 4 Thus the remarks 5, can of Weaver, into a pas is quoted porDpetive. ago of
be put of
VIoaver English
content of
some other
message)
somothing
iss
of but the same kind with as entropy, allowance for calculable by similar process 115). of a methods; of
simply
an unspecified (ref. is 39 p.
content" of the
something or
messago
passage
the of act
there uid
answer: of concepts;
as a rough analogy.
a loading
that
"logical
probability"
(cf,
"L-range") application
or even approximate
numerical
statements.
by tho (and doducibility
as those
additivity' entropy
the
corroepondiug to the
uroa
information favour
alter
does
not fact
as it is
were a
there
can be conceivedl. and most concept concept to which importantly of for the information comparison technical
answer: the
the
logical
againnt
demonstrate
! 86
perverts
the
meaning
of
the
word
context
of the more is
of the
major
applications thaiaithe
of the convention
one,
technical that
derived
theory
nothing
needed
word is
the meaning
from the
correct
meaning
3iiailax concept
fact,
ccondary be made in
conputo
of
could
one can
wherovQr,
probabilities.
In
fact that
cugGestod whore it i
by what powoiblo
is
compute be found
probability application
will of
an approxiuato
information.
almost probability of the scale.
Given
the
all
definitions,
that in
in
needed
fact,
to is
this
nonvort
is
a
a triviality; into
content is
a transformation
(sufficiently transfor is
nearly)
probability.
And information
correlation,
ensemble there
or very
is
like
it.
Wharever, there
and hence
is
an
in this
a'quostion",
a content
sons: 6.63
too.
But only,
sense.
we nust
re-emphasise,
tempted,, in some
in
a secondary
context
whiabz is
close
to
th context
in which
the
187
is
relevant, the
to essence world
think of
that
wo are
really the
with
information, the is
makes
the
corrective lived
thinly which
We can
an apt
turn There
of
without.
may or there in
be one bit
but
a smile?
/36.
REFERENCES
General
reference
applications,
second of the above includes Reference by F. L'. Stumpers. below is indicated symposium
prepared this
1.
2.
Bar-Hillel
Bell E. T.
Y. and Carnap R.
"Development of
(Symp).
mathematics" 1940.
3. 4.
Birkhoff Birkhoff
G. D. G.
Proc, "Lattice
Nat.
Acad.
Sc.
17 (1931)
650. p.
theory"
1940.
5.
6. 7. 8. 9. 10. 11.
Brillouin
Brillouin Calder R.
L.
1.
semantics"
of probability" 1947-
Cawsey G. F. "Physical entropy and the entropy theory" of info rmation R. A. E. tedh. note
G.169 (1952).
12.
Cayley
A. Phil. mathematical
Trans. papers
collected
1$
Eddington A. S. and electrons" Tech. Fano R. M. of electronics, Feller Fellgett W. and its
"Relativity 1936.
theory
of
"Introduc#ion applications"
Phil. P. B. and Lin foot E. H. (1955). 931 247 A vol. no. ser.
Frechet !'Methodes des fonctions M. arbitraires. des evenements dans le cas Theorie en chaine d'un nombre fini d'etats 1938possibles" Gabor Gabor D. Phil. Mag. 41 (1950) p. 1161.
18. 19.
D. "Communication theory and cybernetics", Symposium on electronics and television, Milan 1954. D. and Gabor A. ser. A, 117 (1954) S. "Information J. p. Roy. 31theory" and the Statist. 1953. weighing of Soc.
Gabor
Goldman
J. R. V.
Biometrika
L. Bell
40 (1953)
System
Tech.
Ergebnisse Grenzgebiete,
Logic
13
(1948)
p.
16.
G.
Zr. Symb.
Logic P.
18 (1953) Phil.
p. Sc.
289. 19 1921.
(1940)
P. 763.
,1
31. 32.
Koopman B. Kolmogorov
0. A. N.
Ann.
Math.
ser.
2,41 of the
(1940) theory
p. of
269.
"Foundations
Eng. tr.
1950.
Loeb J.
Mandelbrot
(Symp).
B. (Symp). ' "Theory 1947. dissertation: of
0. Neumann J. von and Morgenstern behaviour" games and economic Riemann G. F. B. , Privat-docent
37. 37b.
Popper Popper
der
K. 'R.
56 (1947)
37c.
37d"
370. 37f. 38. 38a.
Popper K. R. p. 173.
Popper
Popper Popper Russell Shannon
'roc.
K. R.
K. K. R. R.
Mind 47 (1938)
Brit. Brit. W. Bell J. J. "The PhiL Phil. scientific
B. A. C. E.
1931.
System
Tech.
(1948)
Bell Zeits. f.
p.
50-
L.
E.
53 (1929)
29 (1945) extrapolation time-series"
840.
137. and 1949.
Math.
43. 44.
Wiener
N.
Wisdom J. 0. philosophy"
fL
45.
46.
logico-philosophicus"
and information to radar" 1953.