You are on page 1of 21

Paper written in 2009 for an article collection on the diachronic typology of voice and valence-changing

categories edited by Leonid Kulikov and Seppo Kittilä

An exploration of the diachrony of Mapudungun


valency-changing operationsi

Fernando Zúñiga
University of Zurich

1. Introduction

Mapudungun is a language isolate currently spoken by an uncertain number —probably close


to 150,000— of the ethnic Mapuche population in Chile and Argentina. Several closely
related dialectal varieties of the language can be distinguished; this paper will focus on the
main Chilean variety (Central Mapudungun) and make explicit reference to dialectal
variation when necessary.
Mapudungun was first documented within the first century of the Spaniards’ arrival to the
region by Valdivia (1606), even though most studies dealing with descriptive issues, as well
as most text collection and dictionaries, have appeared in the 20th century. The other older
studies are Febrés (1765) and Havestadt (1777), and there is a brief but interesting account of
nominal and verbal morphology at the end of Falkner (1774). Comprehensive accounts of
Mapudungun grammar are found in Augusta (1903), Salas (2006), Zúñiga (2006) and Smeets
(1989, 2007). Since the northern variety documented by Valdivia is very similar to the
varieties spoken today, scholars do not usually distinguish between different chronological
periods for the language.
Mapudungun is an agglutinative head-marking language whose rich verbal morphology
indicates person and number of core arguments, inversion, aspect-related notions, spatial
categories, tense, evidentiality, and deontic modality, among others. There are no adnominal
markers of grammatical case for core arguments, and obligatory nominal number marking is
restricted to nonsingular with adjectives and to plural with (typically human) nouns;
independent pronouns regularly distinguish singular, dual and plural forms. Verb forms can
be categorized as either finite or nonfinite, the latter showing less productivity as to tense and
mood, but especially marking one person less than the former. Processes that reduce syntactic
valency are NP-incorporation, reflexivization, and passivization. Processes that increase
syntactic valency are different types of causativization, which introduce an argument
identified as the causer of the state of affairs, as well as different types of applicativization,
which introduce an argument that can bear a range of non-agentive participant roles.
The purpose of the present study is to explore possible and plausible sources of a number
of different markers signaling valency change in present-day Mapudungun based upon
internal reconstruction. Section 2 presents the fundamentals of Mapudungun morphosyntax
necessary for an adequate understanding of the inquiry. Section 3 proposes and discusses
some possible etymons for the causative suffixes -m and -l, the applicative suffixes -ñma
and -l, and the applicativizing serial roots tu- and ye-; the passive suffix -nge is tentatively
2

traced back to a copula, and the reflexive suffix -u and the causative suffix -fal are only
briefly mentioned since I have not been able to find any plausible etymons for them. Section
4 summarizes and briefly qualifies the findings.

2. Some fundamentals of Mapudungun morphosyntax

2.1 Argumenthood

Mapudungun clause structure is best analyzed as distinguishing the following core syntactic
functions, both of which can be instantiated by bare NPs: Focal Person (FP) and Satellite
Person (SP). The Focal Person controls the most explicit person marking on the verb (i.e., the
one distinguishing three persons and three numbers) and the reference of WH-words in
interrogative clauses; the Satellite Person controls the less explicit person marking on the
verb (typically distinguishing person, but not number). The markers are neutral in form, i.e.,
they do not encode the semantic role or the macrorole assigned to the argument they mark
(with the exception of the differential Undergoer marker -fi and the 3rd person Actor
marker -mew, cf. Examples 2-3 below). The choice between FP and SP is either referentially
determined (with interactions between 1st/2nd persons and 3rd persons, where the speech act
participant is invariably the FP) or pragmatically established (with interactions between two
3rd persons, where the most topical participant is the FP). Even though constituent order has
not been thoroughly studied yet, if both the Focal and the Satellite Persons are present as
lexical NPs, the typical order appears to be [FP V SP] (but note that [FP SP V] and [V SP FP]
are attested as well). In other words, the Satellite Person has to occupy a position adjacent to
the verb, and the Focal Person tends to precede the Satellite Person.
The following examples illustrate the indexing pattern just mentioned. In (1a, b) the
clause is intransitive and the only participant (chi kalku ‘the warlock’ and eymi ‘you(s.)’
respectively) is the FP; note that the 3rd person singular is zero-marked while the 2nd person
singular is marked as -mi on the verb (transparently related to material appearing on the
independent pronoun; the latter can be omitted from the clause).ii

(1) a. Aku-i-Ø chi kalku.


arrive.here-IND-3.FP ART warlock
‘The warlock arrived here.’
b. Aku-i-mi (eymi).
arrive.here-IND-2.SG.FP 2SG
‘You (SG) arrived here.’

In direct monotransitive clauses (2a) the 2nd person singular (Actor) is the FP and chi
kalku ‘the warlock’ (Undergoer) is the SP. Nonhuman SPs are normally zero-marked when
indefinite and/or nonhuman; the definite and human Undergoer triggers differential marking
(-fi) on direct verb forms. In inverse monotransitive clauses (2b) the 2nd person (Undergoer)
is the FP and chi kalku ‘the warlock’ (Actor) is the SP; the former is marked as -mi, just as in
(2a), and the latter is marked by -mew, unspecified for number; additionally, inverse verb
3

forms are characterized by an inverse suffix (here: -e), which occurs in the same slot as the
differential Undergoer marker -fi.

(2) a. Pe-fi-i-mi chi kalku.iii


see-3.SP-IND-2.SG.FP ART warlock
‘You (SG) saw the warlock.’
b. Pe-e-i-mi-mew chi kalku.
see-INV-IND-2.SG.FP-3.SP ART warlock
‘The warlock saw you (SG).’

While the Examples in (2) are clauses featuring an ambitransitive or labile verb (pe-
‘see’, which can be used intransitively or transitively), those in (3) are clauses with a
ditransitive verb (elu- ‘give’). Secondary SPs typically occur in the latter clause type (here:
kura ‘a/the stone’) and can appear as bare NPs, but do not trigger any kind of verbal
indexing.

(3) a. Elu-fi-i-mi kura chi kalku.


give-3.SP-IND-2.SG.FP stone ART warlock
‘You (SG) gave a stone to the warlock.’
b. Elu-e-i-mi-mew kura chi kalku.
give-INV-IND-2.SG.FP-3.SP stone ART warlock
‘The warlock gave you (SG) a stone.’

In sum, Focal Person markers follow a neutral pattern that does not distinguish the role of the
participants (here: -Ø ‘3.FP’ and -mi ‘2.SG.FP’) and Satellite Persons distinguish A’s (-mew
‘3.A.PSP’) from U’s (-Ø ~ -fi ‘3.U.PSP’).
Since the verbs of giving play an important role in the discussion of causatives and
applicatives, some additional remarks with respect to their morphosyntax is in order here.
Elu- is not the only Mapudungun verb translated as Spanish dar (English give); the other one
is the monotransitive verb wül-, which may be more adequately rendered as ‘hand
(something)’. With elu- (4a), the recipient is the Primary Satellite Person (PSP; kiñe wingka
‘a foreigner’) and the theme is the Secondary Satellite Person (SSP; ñi ruka ‘her house’); note
that passivization promotes the recipient, but not the theme, to Focal Person (4b) (see more
on passivization in Section 2.2). By contrast, with wül- (4c), the theme is the PSP and there is
no SSP; the recipient is an optional participant marked by the semantically empty
postposition mew, and passivization promotes the theme to FP (4d).

(4) a. Elu-fi-i-Ø ñi ruka kiñe wingka.


give-3.SP-IND-3.FP 3.PSR house one foreigner
‘She gave her house to a foreigner.’
b. Elu-nge-i-Ø ñi ruka kiñe wingka.
give-PASS-IND-3.FP 3.PSR house one foreigner
‘A foreigner was given her house.’
4

c. Wül-i-Ø ñi ruka kiñe wingka mew.


give-IND-3.FP 3.PSR house one foreigner PPOS
‘She gave (i.e. sold) her house to a foreigner.’ (Augusta 1916:256)
d. Wül-nge-i-Ø ñi ruka kiñe wingka mew.
give-PASS-IND-3.FP 3.PSR house one foreigner PPOS
‘Her house was given (i.e. sold) to a foreigner.’

2.2 Valency-reducing operations

As mentioned in the introduction, the three processes that reduce syntactic valency are NP-
incorporation, reflexivization and passivization. I will not address the former operation here;
the interested reader can consult Harmelink (1992) for further reference.
Reflexivization is marked on the verb by means of the suffix -u(w) (-w after vowels) and
signals that the Actor and the Undergoer are coreferential. Reflexive verb forms are used in
combination with welukon ‘each other’ for reciprocals:

(5) a. Langüm-i-Ø kiñe sañchu.


kill-IND-3.FP one pig
‘S/he killed a pig.’
b. Langüm-u-i-Ø.
kill-REFL-IND-3.FP
‘S/he killed herself/himself.’
c. Welukon langüm-u-i-ngu.
each.other kill-REFL-IND-3.DU.FPiv
‘They (DU) killed each other.’

Passivization is marked on the verb by means of the suffix -nge:

(6) a. Ngilla-n kiñe manshun.


buy-1.SG.IND.FP one ox
‘I bought an ox.’
b. Ngilla-nge-i-Ø kiñe manshun.
buy-PASS-IND-3.FP one ox
‘An ox was bought.’

In (6a), kiñe manshun ‘an ox’ is SP and does not trigger fi-marking on the verb; the Actor (1st
person) is FP. By contrast, in (6b) the Actor has been demoted (the Actor is normally absent
from passive clauses) and the Undergoer has been promoted to FP status.

2.3 Valency-increasing operations

2.3.1 Causatives
Golluscio (2007) is the most complete study on Mapudungun causatives to date, and I will
follow her analysis here; all the examples cited in this section are taken from that source
unless otherwise specified.v Causatives add to the clause an agentive argument, which can be
5

either FP or SP, and the original FP appears as either SP or FP in the derived clause.
Mapudungun features three causative derivations -m, -l and -fal that are, at least in principle,
in complementary distribution: the former applies only to intransitive verbs, while the other
two also apply to ambitransitives; -fal also applies to transitives. The following examples
illustrate the use of these markers with intransitive wadkü- ‘boil’, intransitive aye- ‘laugh’
and transitive ngütrüm- ‘call’:

(7) a. Wadkü-i-Ø ta ko.


boil-IND-3.FP PART water
‘The water boiled.’
b. Xwan ta püti ta wadkü-m-i-Ø ta ko.
J. PART PROG PART boil-CAUS1-IND-3.FP PART water
‘John is boiling water.’
c. Aye-i-Ø ta malen.
laugh-IND-3.FP PART girl
‘The girl laughed.’
d. Ta ñi chedki aye-l-e-i-Ø-mew
PART 3.PSR grandfather laugh-CAUS2-INV-IND-3.FP-3.SP
ta malen.
PART girl
‘The girl’s grandfather made her laugh.’
e. Ñi patron ngütrüm-e-i-Ø-mew.
3.PSR boss call-INV-IND-3.FP-3.SP
‘His boss called him.’
f. Ñi patron ngütrüm-fal-e-i-Ø-mew.
3.PSR boss call-CAUS3-INV-IND-3.FP-3SP
‘His boss had him called.’

Roughly, causative -m is essentially non-productive and is restricted to a closed class of


verbs where the causee is low in animacy and/or control, as wadkü- ‘boil’ in (7a, b). By
contrast, causative -l is productive and applies to a wide variety of verbs, of which aye-
‘laugh’ is but one example (7c, d); when it contrasts with -m (e.g. with verbs like lüf- ‘burn
(itr.)’ and tran- ‘fall’), it signals a causee high in animacy and/or control. Finally, -fal is
typically used with transitives (7e, f) and may have an ambiguous yield; in the following
examples, the Focal Person can be the causer or the causee:

(8) a. Dewma-fal-ün kiñe makuñ.


make-CAUS3-1.SG.IND.FP one poncho
‘I had a poncho made.’ or ‘I had to make a poncho.’ (personal field notes)
b. Iñche ngilla-fal-ün kamisa.
1SG buy-CAUS3-1.SG.IND.FP shirt
‘I had a shirt bought.’ or ‘I had to buy a shirt.’ (Smeets 1989:357)
6

2.3.2 Applicatives
Applicatives add to the clause a non-agentive argument, which can appear as either FP or SP
(depending on the rules governing inversion in Mapudungun mentioned at the beginning of
section 2.1, i.e., SAPs are preferred FPs when compared to 3rd persons, as are topics when
compared with non-topics, etc.). Different studies show different ways to treat them, but also
different numbers of them that have to be recognized. Augusta (1903:61) describes them as
follows: -l is said to simply increase transitivity but -ma ~ -(ü)ñma is characterized as
signaling that “the action of the verb is performed to the detriment of somebody or
something, or that it implies that one parts with a former possession” (my translation). Along
related but somewhat different lines, Salas (2006) says that the markers he calls
indirectizantes add an “indirect (ethical)” participant, and with some verbs the state of affairs
can be seen either as separating the theme from this additional participant (-ñma) or as
causing the former to approach the latter (-lel).vi Sometimes, these spatial displacement can
be taken to mean detriment or benefit of the additional participant, but “the detriment/benefit
contrast is secondary and accidental” (p. 123), because in many cases these notions do not
apply. The following examples from Salas (2006:122-123) illustrate the opposition
between -ñma and -lel according to the spatial parameter (9a, b) and the allegedly neutral
cases (9c, d; “allegedly” because a spatial meaning component might plausible be postulated
in 9d):

(9) a. Ngilla-ñma-fi-i-n Antonio ñi kawellu.


buy-APPL1-3.SP-IND-1.SG.FP A. 3.PSR horse
‘I bought Antonio’s horse from him.’
b. Ngilla-lel-fi-i-n Antonio ñi kawellu.
buy-APPL2-3.SP-IND-1.SG.FP A. 3.PSR horse
‘I bought Antonio’s horse for him.’
c. Miyaye rüngün-üñma-nge-ke-i-Ø ñi ütrar.
jimson.weed grind-APPL1-PASS-HAB-IND-3.FP 3.PSR pip
‘One grinds the jimson weed pips.’
d. Fey tuku-lel-nge-ke-i-Ø fürkü ko.
3SG put-APPL2-PASS-HAB-IND-3.FP cold water
‘One adds cold water to it.’

In (9a, b), the 1st person is the Actor and the FP; Antonio is the PSP and triggers fi-
marking on the verb, while ñi kawellu ‘his horse’ is the SSP and is not marked on the verb.
The non-applicative version of the verb (ngilla- ‘buy’) normally occurs with an animate
Actor FP (the buyer) and an Undergoer SP (the theme). In (a), Antonio is the seller and
therefore parts with his horse, while in (b) Antonio is the intended recipient of the horse and
the 1st person bought it on his behalf. The clauses in (9c, d) are habitual passives used in
order to give instructions as to how to perform some customary actions; the FP is ñi ütrar ‘its
pips’ in (c) and fürkü ko ‘cold water’ in (d), and miyaye ‘jimson weed’ and fey ‘it’ are the
PSPs respectively. Just like ngilla- ‘buy’ in (9a, b), the verb rüngü- ‘grind’ in (c) normally
takes an animate Actor FP and an inanimate Undergoer SP. Tuku- ‘put, set’ (d) is only
apparently different; even though the English gloss might suggest an underlying meaning ‘put
into’, the Mapudungun construction is structurally parallel to its Spanish translation (uno/se
7

le pone agua fría), where the originally verb takes an animate Actor SP (in Spanish, a
Subject) and an inanimate Undergoer FP (the theme; in Spanish, a Direct Object).
There are two other applicative constructions that have received comparatively little
attention in the literature, viz. those featuring so-called compounding serialization with the
verb roots tu- ‘take’ and ye- ‘carry’.vii Both seem to be considerably less productive than the
applicative derivations already mentioned, and there are cases where they do not show
predictable yields. The former appears with this applicativizing function typically with a
handful of verb roots that express motion and introduces an animate goal (10a), but also with
verbs whose SPs are arguably relatively unaffected (10b, c):

(10) a. José kon-tu-fi-i-Ø ñi chaw.


J. enter-take-3.SP-IND-3.FP 3.PSR father
‘José went to his father’s.’ (lit. ‘entered his father[’s house]’)
b. Ñi peñi aye-tu-fi-i-Ø chi wingka.
1SG.PSR brother laugh-take-3.SP-IND-3.FP ART foreigner
‘My brother laughed at the foreigner.’
c. Yafkü-tu-fi-i-Ø ñi lamngen.
get.angry-take-3.SP-IND-3.FP 3.PSR sister
‘He got angry at his sister.’

In contrast, serialization with ye- introduces comitatives (11a), topic of conversations and
various other kinds of participants (11b, c):

(11) a. Amu-ye-fi-i-Ø ñi wenüy.


go-carry-3.SP-IND-3.FP 3.PSR friend
‘S/he went with her/his friend.’
b. Ngüma-ye-fi-i-Ø.
weep-carry-3.SP-IND-3.FP
‘S/he wept for her.’
c. Patron-ye-nge-n.
boss-carry-PASS-1.SG.IND.FP
‘I am considered the boss.’

3. The origin and development of valency-changing operators

Unfortunately from a comparative-historical perspective, Mapudungun has no known


relatives and the variety of the language described by Valdivia (1606) differs from present-
day Mapudungun only slightly. To be sure, the northern dialect that was described in that
source showed some phonetic differences and some lexical items that seem to have been
specific of that variety, but not much reliable textual material in idiomatic Mapudungun was
published before Lenz (1895-1897) and Augusta (1910). Therefore, only internal
reconstruction is available to the modern linguist.
On the other hand, Mapudungun has a number of items that appear both verb-externally
and verb-internally (in the latter case sometimes phonetically reduced and displaying
8

allomorphy), so the hypothesis that some of the corresponding lexical items have become
grammaticalized to some degree without becoming obsolete as independent words is fairly
plausible. These include the demonstrative fey (which has become a 3rd person pronoun, but
is also most likely the source of the differential Undergoer marker -fi), the adjective ka
‘other’ (used as suffix in continuative verb forms) and the adjective we ‘young, new’ (used as
suffix in verb forms with the meaning ‘already’), but also a series of verbs like meke- ‘be
busy (doing something)’, müle- ‘be located’, miaw- ‘walk’, nie- ‘have’ and künu- ‘leave’, all
of which have aspectual functions (the last two also interact with the argument structure).
For example, non-stative verbs like lefün ‘run’ or rukan ‘build a house’ can appear
serialized with the root meke-: lef-meke-i ‘s/he is running’, ruka-meke-i ‘s/he building a
house’. By a similar token, the verb müle- seems to have been grammaticalized as a suffix
(-le after vowels and -küle elsewhere) that has a resultative function with some verbs
(kansha- ‘get tired’, kansha-le-i ‘s/he is tired’) and a progressive yield with others (lef-küle-i
‘s/he is running’) (cf. Zúñiga 2001 for more details on the dual nature of this marker).
Finally, consider the suffix -yaw (-kiyaw after glides and consonants) in verbs like küdaw-i
‘he worked’ vs. küdaw-kiyaw-i ‘s/he is going around working’, where the connection to
miaw- is still transparent. With respect to how old these constructions are, two arguments
suggest that meke- ‘be busy’ has been grammaticalized more recently than miaw- ‘walk’:
first, verb forms including meke- are not mentioned in sources written before the 20th
century, and second, unlike the suffixes -(kü)le and -(ki)yaw, meke- does not show
allomorphy, which is why I regard it here as an instance of compounding serialization rather
than as an example of derivational suffixation. (Cf. Section 3.4 for more on the latter issue.)
Let me now turn to the different formatives related to valency changes introduced in
Section 2. I will start of by addressing those affixes whose etymology is obscure and proceed
to those formatives where some internal reconstruction actually can be done.

3.1 Reflexive -u(w)

The origin of this suffix is unknown and there are no lexical items in the language that bear a
resemblance to it.

3.2 Causative -fal

This suffix might in principle be related to the verb root fal- ‘be worth’, but I think an
account that sees them as unrelated is preferable. The root fal- is clearly related to Spanish
valer ‘be worth, cost’; Valdivia (1606) mentions the existence of the suffix (p. 44) but does
not list vali- or valü- as verb in his word list (the variety he described had, like Febrés’,
voiced fricatives, hence v instead of f). Febrés (1765:655) mentions both, the suffix as -val
and the verb as vali- ~ valü-, but without noting any connection between the two. The
suffix -val means ‘commission, send, command, impute’ (Havestadt 1777: 107 also cites a
deontic/root possibility meaning) and valü- means ‘be worth, cost’. Augusta (1916 :43) gives
fali- with the meaning ‘be worth, be priced, be difficult’, but also some related forms like
causative falil- ‘appreciate’ and reflexive faliuwün kiñe che mew ‘use / take advantage of
someone’, the latter clearly a calque of Spanish valerse de una persona. Interestingly enough,
-fal shows two other uses: (i) it derives deverbal adjectives comparable to those in -able
9

~ -ible in English, e.g. illamfal ‘despicable’ (cf. illam- ‘despise’) and kimfal ‘intelligible’ (cf.
kim- ‘know, understand’), and (ii) it occurs as -u-fal-uw with the meaning of pretense or
simulation, and as -u-fal with the meaning of imputation:

(12) a. Chumgelu kam newe i-ufaluw-ke-la-y-mi?


why Q not.much eat-PRETENSE-HAB-NEG-IND-2.SG.FP
‘Why do you (SG) always pretend no to (want to) eat much?’
b. Weñe-ufal-e-n-mew.
steal-IMPUTATION-INV-1.SG.FP-3.SP
‘S/he says I stole [it] / I am accountable for the theft.’

Given both the documentation history and the difference in meaning between the suffix
and the full verb, I think it is safe to see -fal as older and unrelated to the borrowing of fali- ~
falü- for all varieties of Mapudungun.

3.3 Passive -nge

This formative is transparently related to the element nge, which has been either analyzed as
a copula, parallel to Spanish ser ‘be’, especially by older scholars, or seen as a (static
verbalizing) suffix, typically by more recent accounts (e.g. Salas 2006 and Zúñiga 2006).
Whether the “correct” synchronic analysis of ‘be a Mapuche’ is mapuche nge- or mapuche-
nge- is a question that cannot be addressed in detail in the present study, but suffice it to say
here that there is phonological and syntactic evidence suggesting that both a copular and a
suffixal nge might actually coexist in present-day Mapudungun, cf. Malvestiti (2006). For
instance, Smeets (1989:159-160) mentions the minimal pair in (13):

(13) a. nge-la-i-Ø chadi


be-NEG-IND-3.FP salt
‘there is no salt’
b. chadi-nge-la-i-Ø
salt be-NEG-IND-3.FP
‘it is not salty’

For our present purposes, I think it is reasonable to assume that there existed a verbal root
*nge- ‘be’, which became cliticized and eventually suffixed to NPs together with all its
morphology in some, but allegedly not all, cases. This very root also became grammaticalized
as passive morpheme and was integrated into the verbal complex with this function.

3.4 Applicativization with tu- and ye-

Even though these formatives may be seen as grammaticalized versions of the transitive
verbs tu- ‘take, get’ and ye- ‘carry, bring, take’, respectively, I will regard them as instances
of root serialization here.viii These constructions are comparatively unproductive, their yield
is often idiosyncratic, and the involved formatives have not developed phonologically
conditioned allomorphy like suffixes have. Also observe that tu- shows a wide range of
10

possible meanings as a non-applicative formative as well: it works as a de-nominal verbalizer


(e.g. mamüll-tu- ‘gather firewood’, from mamüll ‘(fire)wood’), but is also used to form the
iterative (e.g. rüngkü-i ‘s/he jumped’ vs. rüngkü-rüngkü-tu-i or rüngkü-rüngkü-nge-i ‘s/he
bounced, jumped repeatedly’) and very frequently simply means ‘back’ (e.g. aku-i ‘s/he
arrived here’ vs. aku-tu-i ‘s/he got back here’; also note the pair amu-i ‘s/he went’ vs. amu-tu-
i ‘s/he went away/back’). Finally, observe that there are ye-suffixes in addition to the serial
use of ye-: one marks plural S/O arguments (14a, b) and the other “completed action” (14c).
It does not seem likely that these other two -ye’s are related to the applicativizing root, but I
can say nothing more about the issue at this point.

(14) a. amu-ye-fal-i-in
go-PL-FAL-IND-1.PL.FP
‘we (PL) have to go in great numbers’
b. iñche ye-ye-pa-n
1.SG bring-PL-CIS-1.SG.IND.FP
‘I brought many [things]’ (Smeets 2007:272)
c. pe-(u)ye-n ñi pülata
find-COMPL-1.SG.IND.FP 1.SG.PSR money
‘I found my money’ (Salas 2006:141)

3.5 Low-animacy/control causative -m

Scholars have not been able to find any lexical items in the language that might plausibly
have given rise to this formative, and so its etymology remains unknown. A noteworthy
feature of -m, however, is that it is the only formative in the language that triggers a change in
the verb root that precedes it. Many suffixes show allomorphy themselves (e.g., ü-epenthesis
to avoid some consonant clusters at the morpheme boundary), as we will see in the case of
the causative -l shortly, but only before -m do the following changes occur: stem final -f
changes to -p: lef- ‘run’ vs. lep-üm- ‘make run’. Since -m is not productive the list of stems
that show this alternation is not very long, but it is clear that there is a general rule operating
here: the items include af- ‘end (itr.)’ / ap-üm- ‘end (tr.)’, traf- ‘gather (itr.)’ / trap-üm-
‘gather (tr.)’, chif- ‘overflow’ / chip-üm- ‘make overflow’, küf- ~ trüf- ‘choke (itr.)’ / küp-üm-
~ trüp-üm-‘choke (tr.)’, lüf- ‘burn (itr.), catch fire’ / lüp-üm- ‘burn (tr.), set ablaze’, ñif- ‘dry
(itr.)’ / ñip-üm- ‘dry (tr.)’ and wef- ‘appear’ / wep-üm- ‘make appear’, but also some cases
where the non-causative verb seems to no longer exist, like chüp-üm- ‘fill in excess’ (cf.
chüf-ma- ‘surround’), kop-üm- ‘roast, toast (tr.)’ (cf. kofeñ ‘roasted, toasted’) and wip-üm-
‘do in the correct way’ (cf. wif ‘correct, straight’). A similar alternation between g and k with
the two attested verb roots that show it is somewhat more problematic: for nag- ‘descend’,
Golluscio (2007:221) notes that nak-üm- ‘lower’ also appears as naw-üm-, although this is
probably caused by the variation in the root itself nag- ~ naw- ‘descend’. Since the causative
of lleg- ‘be born’ is lleg-üm- or llek-üm- ‘give birth’, it appears that these two alternations do
not have the same status.
Observe that the obstruents p /p/, t /t̟/, t /t/, tr /tʂ/, ch /ʧ/ and k /k/ are not licensed as
possible syllable codas in present-day Mapudungun. This fact might explain the alternation
between -l and -t in one of the direct and inverse nonfinite forms or verbal nouns used in
11

complement clauses, respectively: the direct verbal noun ending is (-fi)-el while its inverse
counterpart is -etew, the latter suggesting an original morphemic structure *-e-et-mew (INV-
NFIN-3.A.SP) that formally parallels (i) finite forms in that the mood and FP markers are
replaced by the nonfinite marker and (ii) direct nonfinite forms in that -fi and -e appear in
complementary distribution in the same templatic slot followed by the nonfinite suffix.
Consider the following examples of paired finite and nonfinite forms:

(15) a. ngilla-fi-i-Ø ngilla-e-i-Ø-mew


buy-3.SP-IND-3.FP buy-INV-IND-3.FP-3.SP
‘s/he bought it (e.g. the horse)’ ‘it (e.g. the horse) was bought by her/him’
b. ñi ngilla-fi-el ñi ngilla-etew
3.PSR buy-3.SP-NFIN 3.PSR buy-INV.NFIN
‘that s/he bought it’ ‘that it was bought by her/him’

These observations would make it tempting to reconstruct the nonfinite suffix as *-et and
postulate a general rule preventing obstruents from appearing in the coda, which would
explain not only the retention of t when a following vowel allows its resyllabification (as
in -e.tew; note that such resyllabification also applies, thanks to the deletion of the nasal in
mew, to the 1st person singular FP and 3rd person FP finite inverse forms as well: -e-n-mew
> -e.new and -e-i-Ø-mew > -e.yew) but also the change from p to f and k to g with low-control
causatives if we assume that at some point in the prehistory of the language the elements
expressing personal inflection and other categories were not suffixes but clitics or even
separate words and the original verbal roots for lef- and nag- were actually *lep and *nak. If
this assumption is correct, the roots like *lep originally appeared as *lef when not
causativized as *lepüm, and at some later point in the history of the language the personal
markers and other formatives were affixed to these roots, yielding present-day forms like lefi
‘he ran’ and lepümfi ‘he made him run’.
Nevertheless, there are some complications in addition to the variation with k/g.
Obstruents might have turned into the corresponding fricatives that were allowed to occur in
coda position, but neither the places nor the manners of articulation really match up for all
these alternations. The sounds f /f/ and g /ɰ/ do not belong to the same class: the former is a
fricative and patterns together with d /θ/ and s /s/, whereas the latter is a glide and patterns
together with y /j/ and w /w/. By the same token, g does not occur syllable-initially while f
does appear in that position, and the former alternates with w while f does not alternate with
any other sound. To be sure, both p and f are labial and both k and g are velar if we disregard
articulatory details (bilabial vs. labio-dental, velar vs. labio-velar). However, even though t
and l are both dento-alveolar, the corresponding fricative s would have been available and it
is not evident why the liquid l would have been chosen instead. On the other hand, the
segment s is not a particularly robust phoneme in the language, and it may well be the case
that older stages of Mapudungun did not have this phoneme.
Of course, it is also possible that there was not a single rule turning p, t and k into f, l and
g respectively; one rule could explain the p/f alternation and perhaps also the k/g alternation,
while another rule, possibly operating at a different point in time, would be responsible for
the t/l alternation. The conservative assumption would be to see in these alternations the
12

reflexes of some processes that were once operative but have become obscured by further
sound changes and/or some additional phonological processes that are no longer traceable.

3.6 Applicative -ñma

The full verb found in the present-day language that is superficially closest to -ñma is not a
likely etymon: ñümi- ~ ñümü- ‘pick, collect’. There is, however, some evidence that suggests
that this formative might be etymologically plurimorphemic. First, note the existence of the
rarely encountered and obsolescent “auto-benefactive” suffix -ñmu (Examples 16a and 16b
are from Smeets 1989:359-360; 16c is from Salas 2006:127):

(16) a. Treka-ñmu-a-n.
walk-AUTOBEN-FUT-1.SG.IND.FP
‘I am going to have a nice walk.’
b. Küdaw-ñmu-n.
work-AUTOBEN-1.SG.IND.FP
‘I worked for myself.’
c. I-ñmu-a-n kom tüfachi füdü.
eat-AUTOBEN-FUT-1.SG.IND.FP all this tinamouix
‘I will eat all this tinamou.’

Even though -ñma and -ñmu no longer seem to productively alternate as markers of allo- and
auto-affectedness respectively, the existence of reflexive -u(w) makes the interpretation in
terms of -ñm-a and -ñm-u at least an interesting hypothesis. Consider the following two
examples found in Havestadt (1777 :96), perhaps representing a stage where -ñma-u had not
yet evolved into -ñm-u:

(17) a. I-nma-u-n.
eat-APPL-REFL-1.SG.IND.FP
‘I ate it.’
(original: mihi illud in stomachum conjeci ‘I threw it into my stomach’)
b. Anü-m-nie-nma-u-n.x
sit.down-CAUS1-have-APPL-REFL-1.SG.IND.FP
‘I remember.’
(original: recordor, memini ‘I remember’)

Second, consider the allomorphy of -ñma. Depending on the author, this applicative
suffix has actually two or three possible forms: -üñma, -ñma and -ma.
Older treatments of -ñma are not particularly comprehensive or detailed. Valdivia (1606
:45) mentions -nma and gives dungu-ma-e-n (speak-APPL-INV-1.SG.IND.FP) ‘speak for me!’ as
only example. Febrés (1765:48) treats both applicatives -ñma and -l as if they were more or
less interchangeable and notes two allomorphs for the former, viz. -ñma or -nma, without
stating the allomorphy conditions; one of his two examples is tañi weñe-nma-nge-n mew ta
manshun ika (3.PSR steal-APPL-PASS-1.SG.IND.FP PART ox PL) ‘because they (IMPERS) stole the
oxen from me’. Havestadt (1777:95) mentions the applicative -ma ~ -nma (the latter being
13

the postvocalic allomorph) and gives very few examples: for our present purposes let me cite
only küpa-l-ma-a-e-y-mi (come-CAUS-APPL-INV-IND-2.SG.FP) ‘I will bring [it] for you’.xi
Augusta (1903) describes the distribution of the allomorphs as follows: -üñma after
consonants (as in langüm-üñma- ‘kill on somebody’, from langüm- ‘kill’, itself an irregular
low control causative derived from la- ‘die’), -ñma after vowels (as in weñe-ñma- ‘steal from
somebody’, from weñe- ‘steal’) and -ma after diphthongs (as in küdaw-ma- ‘work for
somebody’, from küdaw- ‘work’). Observe, however, that Augusta (1916) lists only a handful
of verbs as taking -üñma: these include kalpud- ‘pull out’, kim- ‘know’ and trolol- ‘hollow
out’. The form -ñma appears after vowels as predicted in Augusta’s grammar, but the
allomorph -ma does not only appear after diphthongs; it regularly appears after laterals (e.g.
lol- ‘become furrowed’, wifül- ‘throw (liquid)’, püra-kawell- ‘mount a horse’; actually,
trolol- is the only exception I found in the dictionary), nasals (e.g. ñiküm- ‘shelter (a house)
from the wind’, patron- ‘find a boss’, lawen- ‘use (something) as medicine’, trangliñ- ‘frost,
freeze’), fricatives (e.g. llipad- ‘(birds) fly towards (a prey)’, ñidüf- ‘sew’) and g (e.g. treg-
‘spark’). This distributional pattern is close, but not identical, to the one described by
Havestadt.
Salas (2006:121f) mentions the phonologically conditioned allomorphy one would expect
based upon the behavior of other verbal suffixes, viz. -ñma after vowels and -üñma
elsewhere, and a morphologically conditioned allomorphy that explains the ma-form after the
high control causative -lel (18a; Salas 2006:124). Also note that Salas postulates a
homonymous “participative” suffix -ma that signals participation or personal involvement in
the state of affairs on the part of the focal person (18b; Salas 2006:127):

(18) a. Küpa-lel-ma-nge-i-mixii kuram tami ñuke.


come-CAUS2-APPL1-PASS-IND-2.SG.FP egg 2.SG.PSR mother
‘They (IMPERS) brought eggs for your (SG) mother.’
b. Maw-ma-n.
rain-???-1.SG.IND.FP
‘I got rained on.’

Smeets (2007) proposes a similar analysis. She labels one -ñma “indirect object” and
describes it as changing “the patient into 1. a participant which is adversely affected by the
event [... and] 2. a participant that is indirectly affected by the event, viz. through possession
of or relationship with the patient” (p. 276). Her allomorphs are -üñma after consonants, -ñma
after vowels and -ma after diphthongs, l and sh. Smeets labels a second -ñma “experience”
and characterizes it as both occurring in a slot closer to the verb root and also indicating that
“the subject of an intransitive verb experiences the process or event denoted by the verb” (p.
301); -ñma is the postvocalic, -ma the postconsonantal allomorph. One of her examples of
each suffix is given in (19):

(19) a. I-ñma-nge-i-Ø kiñe sañchu ñi karukatu.


eat-APPL-PASS-IND-3.FP one pig 1.SG.PSR neighbor
‘They (IMPERS) ate a pig on my neighbor.’
14

b. Iñche aku-ñma-n kiñe küme dungu.


1.SG arrive-???-1.SG.IND.FP one good message
‘I received a nice message.’

Finally, observe that Harmelink (1996:231) recognizes only one -ñma and labels it
“indirect patient”; its allomorphs are postconsonantal -üñma and postvocalic -ñma. Two of
his examples are given in (20); Harmelink argues that the semantic yield of -ñma is neutral
between malefactive and benefactive and depends on verb semantics and/or context.

(20) a. Wülel-üñma-e-n-mew tañi fotüm.


hit-APPL-INV-1.SG.IND.FP-3.SP 1.SG.PSR son
‘S/he hit my son.’
b. Kellu-ñma-e-i-mi-mew tami ñawe.
help-APPL-INV-IND-2.SG.FP-3.SP 2.SG.PSR daughter
‘S/he helped your (SG) daughter.’

Middle-aged urban speakers of Chilean Mapudungun are not always sure as to (i)
whether a verb can take -ñma or not, (ii) the exact meaning such a form would have, and (iii)
whether the correct form is -üñma or -ma after consonants. Nevertheless, those older urban
speakers I consulted tended to confirm for me that most of the forms listed in Augusta’s 1916
dictionary were accurate. Consequently, albeit with the important caveat that there seems to
be some dialectal variation as well―the speakers whose forms most closely resembled the
ones given by Augusta were originally from the Araucanía region, whereas those speakers
from Los Lagos, Los Ríos and Bío-Bío tended to inconsistently produce forms that deviated
from Augusta’s―, I must come to the conclusion that Harmelink’s brief analysis does not
account for the ma-allomorphs at all, and that Salas’ suggestion that only his “participative”
(roughly, Smeets’ “experience”) shows -ma is probably mistaken. By the same token, the
conditions of occurrence of Smeets’ “indirect object” suffix are incomplete as well, since not
only diphthongs, l and sh are followed by -ma. Most importantly, I have not found conclusive
evidence of a different distribution of “indirect object” -ma and “experience” -ma, either with
respect to the slot(s) they occupy in the verbal complex―even though, of course, -ma can
and does have different syntactic yield with different predicates.
As a matter of fact, it is possible to distinguish between two -ñma suffixes based upon
both their syntactic import and their allomorphy rules. Whereas a transitive verb like kintu-
‘look for’ does not alter the syntactic status of the Focal Person when applicativized
with -ñma (21a), an intransitive like maw(ün)- ‘rain’ (Ex. 18b above, reproduced here as 21b)
does:

(21) a. kintu-ñma-fi-n mamüll


look.for-APPL-3.SP-1.SG.IND.FP wood
‘I looked for (fire)wood (in order to take it away from him/her)’
(adapted from Salas 2006:122)
b. maw-ma-n
rain-APPL-1.SG.IND.FP
‘I got rained on’ (Salas 2006:127)
15

Roughly, the FP in the (a) is the same as with the non-applicativized sentence kintu-n mamüll
‘I looked for (fire)wood’, and the introduced participant is the new PSP. By contrast, it is the
introduced participant that bears the FP role in (b); its intransitive counterpart is simply
mawün-i-Ø ‘it rained’, with a dummy 3rd person FP. The allomorphy rules for -ñma, on the
other hand, are the following. When applied to transitives, -ñma is the postvocalic allomorph;
-ma appears after g, liquids, nasals and some fricatives, and -üñma appears elsewhere. When
applied to intransitives, -ñma occurs after vowels and -ma elsewhere. Consult Zúñiga (i.p.)
for more details on the syntax and the allomorphy rules of -ñma.
Even assuming, as the account given here does, that the there is one original element
*-ñma, it is not trivial to trace its development back. If *-ma is the original form of the suffix
and it appears postconsonantally, then the same *-ma should appear after vowels as well (as
with the progressive construction with meke-, which has no allomorphs); this might
additionally suggest recent grammaticalization, which is rather unlikely given the
documentation history. Based upon the present-day make-up of other verbal suffixes, we
would expect an allomorphy of the type *-ma after vowels and *-üma or *-küma elsewhere
(parallel to causative -m ~ -üm and resultative/progressive -le ~ -küle, respectively), which is
not what we find. In addition, there is no epenthetic ñ segment elsewhere in the language. If,
by contrast, *-ñma is the original form of the suffix, then -ñma occurring after vowels and
either *-üñma or *-ma occurring elsewhere is not too unexpected: *-ñma would have been
reduced to *-ma after nasals (which accounts for approximately 40% of the forms listed by
Augusta), and this very variant would have been used when applied to other contexts.
Eventually, the more pervasive allomorphy pattern -CV... ~ -üCV... would have interfered
with the original distribution and led to the present-day variation.
Finally returning to the hypothesis formulated at the very beginning of this section: if the
original two related applicative formatives were *ñma (allo-affectedness) and *ñmu (auto-
affectedness) in terms of their segmental make-up and―admittedly a more speculative
hypothesis―*-ñ-m-a and *-ñ-m-u in terms of their internal structure, what did the different
components mean? The suffix *-m might be the low-animacy/control causative -m, of
unknown origin but introducing a non-agentive participant to the clause, and in that sense
compatible with both the semantics and, although somewhat more intricately, the syntax of
the applicative. The final suffixes could signify that the Agent is not the one affected by the
state of affairs (*-a, of unknown origin and unparalleled elsewhere in the language with a
similar function), or, alternatively, that he is (*-u, the reflexive). The suffix *-ñ would have
the effect of blocking the morphophonological rule that turns f into p before *-m, but it is not
evident what its meaning might be. In sum, it is possible, but not necessarily plausible, that
the complicated suffix -ñma and the obsolescent suffix -ñmu are the remains of a prehistoric
opposition that, perhaps due to both internal forces and contact with a language without a
comparable morphosyntactic feature, is now virtually lost.

3.7 Causative/applicative -l

Valdivia (1606 :45) introduces a causative with the dialectal variants -l for the southern
and -l-ka for the northern dialects, with two examples: ngolli- and lakutu- ‘get drunk’ vs.
ngolli-l-ka- and lakutu-l-ka- ‘make (someone) get drunk’. xiii Valdivia (1606 :41) further
mentions a suffix -l that turns intransitive verbs (“neutros”) into transitives (“activos”) and
16

gives examples built upon anün ‘sit down’ and küpan ‘come’, which are clearly instances of
causative -l as well. He does not seem to acknowledge the existence of -l as applicative.
Febrés (1765:48-49) mentions the applicative -l ~ -lel (as equivalent to -ñma; cf. Section
3.6) and gives some examples (22b includes the reflexive suffix). Note that -lel (a) and -l (b)
occur with the same verb without any apparent difference in meaning:

(22) a. Kintu-lel-fi-nge kawellu tami chaw.


look.for-APPL-3.SP-2.SG.IMPER horse 2.SG.PSR father
‘Look for your (SG) father’s horse!’
b. Kintu-l-u-fi-nge tami kawellu.
look.for-APPL-REFL-3.SP-2.SG.IMPER 2.SG.PSR horse
‘Look for your (SG) horse!’
c. Ketra-lel-nge-la-n mapu.
plow-APPL-PASS-NEG-1.SG.IND.FP land
‘They (IMPERS) did not plow my land.’

Havestadt (1777 :93-94) mentions -lel only briefly as a honorific device of sorts and gives
one example: fey pi-lel-e-n tami chaw (3.SG say-APPL-INV-1.SG.IND.FP 2.SG.PSR father) ‘please
tell your (SG) father in my name’ (original: dignetur, non gravetur dicere suo patri meis
verbis, meo nomine). Interestingly enough, Havestadt refers the reader to §289 (p.162-164) in
his study, where he treats several formality-related expressions derived from marimari ‘hello,
thank you’, e.g. the verb marimarin ~ marimarilün (alü marimari-l-en ~ marimari-lel-aen
‘(I) thank you very much’), implicitly saying that (i) -l can appear without recognizable
semantic yield (ii) -l and -lel are but variants of the same formative.
Augusta (1903:61-62) says -l is the postvocalic allomorph of the applicative (with -el
occurring elsewhere): katrü- ‘cut’ gives katrü-l-, küpal- ‘bring’ gives küpal-el- (recall that
küpal- is bimorphemic: küpa- ‘come’ + -l ‘causative’), nentu- ‘extract, get out’ gives nentu-l-
and recursively nentu-l-el-. Pi- ‘say, tell’ apparently takes -lel: fey pi-lel-en tami chaw ‘tell
(it) to your father’ (the same example as Havestadt’s, cf. above), and with kintu- ‘look for’
both -l and -lel are grammatical without noticeable change in meaning: kintu-l(el)-en tañi
chumpiru ‘look for my hat’. Incidentally note Augusta’s example are-l- ‘lend (something to
someone)’, which takes a FP, an animate PSP and an inanimate SSP, as opposed to are-nge-l-
m- ‘lend (something)’, which only takes a FP and an inanimate PSP. The latter verb includes
a passive, the applicative and a low-animacy/control causative: ‘cause (something) to be lent
(to someone)’; *are- does not exist with the meaning ‘lend (something)’. Similar, but not as
neatly built, is pe-nge-l- ‘show (something to someone)’, derived from pe- ‘see (something)’.
Salas’ (2006) account of this causative/applicative has already been addressed in Section
2 above, and he does not provide any additional insights as to etymology or allomorphy
conditions. Harmelink (1996:232-233) explicitly distinguishes the uses of -l with intransitives
(high animacy/control causative, e.g. tremo- ‘be/get healthy’ vs. tremo-l- ‘heal (tr.)’) from
those with transitives (applicative, e.g. küpal-el- ‘bring (something to someone)’), but
proposes a unified account of its yield as “transitivity increase”. The allomorphy conditions
Harmelink gives are the following: -l after vowels, -ül after w, -el elsewhere; here the
sequence -l-el is, contrary to what Salas says, not morphologically but actually
phonologically conditioned.
17

Smeets (1989) distinguishes the causative (pp. 395-399) from the “benefactive” (pp. 365-
367), not only with respect to meaning but also with respect to templatic position within the
verb complex. The former has allomorphs -l (after vowels) and -ül (after consonants), which
in turn “alternates with -el” with some verbs (e.g. küdaw- ‘work’, trupef- ‘get frightened’ and
leg- ‘hit (itr.)’). Moreover, some verbs only take -el (her list consists of af- ‘end’, aywiñ-
‘become a shadow’, kim- ‘know’, ariñ- ‘get burned’, mutrur- ‘bump against (something)’,
wim- ‘get used’, ad- ‘become beautiful’ and lingar- ~ fülang- ‘become white’). xiv By the
same token, i- ‘eat’ is said to take -lel. With some ambitransitive verbs, the Ø vs. -l
opposition does not introduce a causer to the clause but signals unintentional action instead:
llocho- ‘loosen / get loose’, ngüfo- ‘make / get wet’, and wacho- ‘break (itr./tr.)’. xv By
contrast, Smeets’ benefactive “changes the patient of an event into the beneficiary”: its
marker is -l ~ -lel after vowels and -el elsewhere. Unlike Salas, Smeets finds that the
applicative sequence -lel-ma is infrequent and the sequence -ñma-ñma ungrammatical. She
gives an example of the infrequent sequence, viz. anü-künu-lel-ma-fi-n (sit.down-leave-
APPL2-APPL1-3.SP-1.SG.IND.FP) ‘I left (it) for him’, as well as one of an “unreliable” instance
of the sequence -lel-el: anü-künu-lel-el-fi-n (sit.down-leave-APPL1-APPL1-3.SP-1.SG.IND.FP) ‘I
left (it) for him’. Finally, Smeets mentions that wülel- ‘hit (tr.)’ (cf. Ex. 18) might be derived
from wül- ‘hand, give’, and that both the latter verb and elu- ‘give’ do not appear with -l.
Golluscio (2007:209) says that the suffix -l “might come from the verb el[-] meaning ‘put
[in the world], create, array’”. As a matter of fact, it might prove useful to take the verbs wül-
‘hand, give’ and elu- ‘give’ (cf. Section 2) into account as well because of both their
incompatibility with -l and the contradictory analyses of the allomorphy conditions. The
present-day semantics of elu- makes it more attractive as a possible etymon of the
applicative―this grammaticalization path being attested in several other languages, e.g. in
Asia―, but both the broad semantic range and the form of el- make it at least as possible.
Both the semantics and the form of wül- would account for the allomorph -ül found by some
authors in certain contexts, even though the default epenthesis pattern might lead to a
postconsonantal allomorph -ül of -l anyway, parallel to the low-animacy/control causative -m
~ -üm.
What is the significance (and the bottom line) of all the details mentioned by the authors
cited? Let me start off by stating the allomorphy conditions in more detail, based not only
information from Augusta’s dictionary but also upon my informants’ responses to some of
the forms given by the other authors mentioned above. The verbs listed by Augusta (1916)
conform to the following general pattern: -l is used after vowels and -el elsewhere. The
exceptions are the following: i- ‘eat’, ina- ‘follow’, kullkü- ‘reserve (domestic animals)’, pe-
‘see’ and pi- ‘say’ take -lel instead of -l (Augusta does not mention kintu- ‘look for’ with
applicative morphology), and the verbs given in (23) take -ül:

(23) a. miaw- ‘walk’


ñawñaw- ‘meow’
ünew- ‘feel uncomfortable’
puw- ‘arrive there’
piaw- ??xvi
kamañkiaw- ‘go around serving as a shepherd’
küñatuyaw- ‘go around hand in hand’
18

b. mefür- ‘shake’ püllüf- ‘splinter’ trüker- ‘roll one’s eyes’


trünüy- ‘vibrate’ trupef- ‘get scared’ patrüg- ‘spread (tr.)’
nüwf- ‘hand by extending one’s arm’

The verbs in (23a) all end in aw, ew, iw or uw―the last two due to the perambulative
suffix -yaw ~ -kiyaw mentioned at the beginning of Section 3―, but note that there are some
verbs that take -el under the same phonological conditions: küdaw- ‘work’, fi-ñma-w- ‘get
distressed, become anxious’, rew- ‘be stirred’ and wew- ‘win (itr./tr.)’; also llow- ‘receive,
accept’ takes -el instead of -ül. Neither is it the case that intransitives and transitives would
behave differently, although the meaning of miaw-ül- is not the expected one (‘make
someone walk’): ‘take something along’.
The verbs in (23b), on the other hand, all end on a consonant or a glide but do not
take -el. Verbs that do take -el, like traf- ‘gather (itr.)’ and trag- ‘pass from one hand to
another’, contrast with verbs ending in the same non-vowels (labial and velar) but taking -ül
instead, like nüwf- ‘hand by extending one’s arm’ and patrüg- ‘spread (tr.)’; here it seems
that the segment ü in the root syllable triggers assimilation in the suffix, which would account
for püllüf-, trünüy-, mefür-, pütef-, and trüker-. However, there are exceptions: if trupef- takes
-ül in spite of not having ü in the root, why does luyüf- ‘become shiny’ take -el in spite of
being like nüwf-? Moreover, with pütef- ‘sprinkle chicha’, both pütef-el- and pütef-ül- are
attested with the same meaning, and of course all verbs listed with -ül in (23b) whose root-
final segment is f are found with -el as well, depending on the specific informant one asks.
Finally, there is one verb where both derivations appear with a difference in meaning: the
ambitransitive ütrüf- means either ‘lay down’ or ‘throw (something)’, hence ütrüf-el- ‘throw
(something to someone)’ and ütrüf-ül- ‘throw (something) away’. Some of my
informants―all of them urban Mapuche―have confirmed the latter opposition, others
identify ütrüf-ül- with the meaning ‘throw (something to someone)’ as the only possible
derivation.
Since I have not been able to find evidence supporting Smeets’ claim that there are two
different l-suffixes with different allomorphy rules and in two different slots of the verbal
template (even though -l(el) is both a causative and an applicative, cf. Zúñiga i.p.), I must
come to Harmelink’s conclusion with a revised allomorphy rule: there is only one suffix that
not only causativizes intransitives but also applicativizes transitives and is realized in
principle as -l after vowels, -ül after w and -el elsewhere. The exceptions to this general rule
are as follows: with some frequently used monosyllabic roots like i-, pe-, and pi-, but also
with some others like ina-, kullkü- and kintu-, the allomorph -lel is used (which is
homophonous with the causative-applicative sequence -l-el), possibly in order to avoid
potentially confusing forms due to l segments appearing in resultative/progressive and
subjunctive forms (e.g. i-le-i ‘s/he is eating’, i-li ‘if I eat’, both of which might be realized as
[ili]); with some roots including ü and ending in f or r, -el is realized as -ül, and, to the extent
that ütrüf-el- and ütrüf-ül- both exist with different meanings, this latter example would be
the only one suggesting a possible grammaticalization of wül- ‘hand (something)’. For all
other forms I regard either elu- ‘give’ or, perhaps, el- ‘put, create, array’ as probable etymon
and assume their origin to predate the first written documents on the language.
19

4. Conclusions

Since (i) the language does not have any known relatives, (ii) dialectal variation as to the
domain of valency-changing operations is, to my knowledge, negligible, and (iii) the variety
of the language described in the early 17th century is virtually the same we find today, I have
explored the origin and development of the markers involved based upon internal
reconstruction. I have proposed a number of etymons or briefly evaluated other authors’
proposals for those suffixes that seem to be related to verbs that are still in use in the
language: the passive suffix -nge is likely to be the grammaticalized version of the copula
ngen ‘be’ and causative/applicative -l may have originated in elun ‘give’ or elün ‘create, set’
but shows a rather complicated synchrony. Both reflexive -u(w) and causative -m might be of
verbal origin as well, but I do not know to which existing elements they might be related.
Finally, the applicative marker -ñma and its obsolescent counterpart -ñmu show the most
complex picture of all these suffixes, which may be due to an originally plurimorphemic
structure, possibly based upon relexive -u, causative -m and other elements *a and *ñ that
cannot be traced back to any element in a plausible way.

Abbreviations

APPL applicative NFIN nonfinite


ART article PART particle
AUTOBEN autobenefactive PASS passive
CAUS causative PL plural
CIS cislocative PPOS postposition
COMPL completive PSP primary satellite person
DU dual PSR possessor
FP focal person REFL reflexive
FUT future SG singular
IMPERS impersonal SP satellite person
IND indicative SSP secondary satellite person
INV inverse

Notes
i
The author is indebted to Seppo Kittilä and Leonid Kulikov for numerous and valuable comments on
previous versions of this article.
ii
The element -mi is bimorphemic, viz. -m ‘2’ + -i ‘SG’, but since morphological make-up is relevant here I
have chosen to give abbreviated glosses throughout the article. Most FP personal are bimorphemic; the notable
exception is -n ‘1.SG.IND’, which is also an exponent of the indicative mood.
iii
Mapudungun verbs are given here in their underlying form for convenience; some syllabification, elision
and assimilation rules apply, cf. Zúñiga (2006:105-106) for details.
iv
With human referents, the 3rd person Focal Person marker normally does distinguish number (-Ø
‘3.SG’, -Ø-ngu ‘3.DU’ and -Ø-ngün ‘3.PL’) under certain syntactic conditions that are immaterial here.
20

v
I have slightly adapted Golluscio’s orthography (incidentally note that her informants are speakers of the
Argentinean varieties of the language) in order to facilitate comparison; in particular, what I transcribe here as ü
is ï in the original. I have adapted some of her glosses as well, but I have followed her analysis of causatives to
the letter.
vi
I will return to the allomorphy of these applicatives in Section 3.
vii
Serialization with tu- ‘take’ has a number of other functions, e.g. telicization, the introduction of the
meanings ‘back’ and ‘again’, iterativization, etc., see section 3.4 for more details. Cf. Smeets (2008: 126f, 254f,
297f, 304f) for an analysis of -tu as suffix that appears in different verbal slot positions according to function.
viii
I thank Leonid Kulikov and Seppo Kittilä for insisting on the plausibility of the serialization analysis.
ix
Salas translates füdü as perdiz ‘partridge’, which is inaccurate. Cf. Zúñiga (2006: 344-345) for details.
x
Note a similar example with a fairly different meaning in Smeets (2007: 303): anü-ñma-nie-ñma-en ñi
makuñ (sit.down-APPL-have-APPL-2.SG→1.SG 1.SG.PSR coat) ‘you (sg) sat on my coat’.
xi
More recent studies that address the -ñma/-l opposition without enough detail for us to consider them in the
present context are Arnold (1997) and Rivano (1991).
xii
Salas’ analysis is incorrect here: küpa-l-el- is trimorphemic. The suffix -l causativizes the root küpa- ‘come’
(hence küpa-l- ‘bring’) while -el applicativizes the stem küpa-l- —hence küpa-l-el- ‘bring to/for (somebody)’.
See Section 3.7 for more details.
xiii
Only ngolli- means ‘get drunk’; lakutu- refers to a gift exchange ritual between namesakes; it is unlikely for
the latter meaning to have developed later and to have replaced ‘get drunk’―most probably, Valdivia
misunderstood the meaning of this verb.
xiv
Her example traf- ‘fit, be joined, gather (itr.)’ vs. trap-el- ‘tie’ is different because there is a low-
animacy/control causative formation with -m as well: trap-üm ‘gather (tr.)’. Since this is the only instance I am
aware of where root-final f becomes p before -l, I prefer to explain this alternation as an “anomalous” analogy to
or back-formation from trapüm-.
xv
Smeets (1989:397) gives katrü- ‘cut’ as belonging to this category, i.e. katrü-l- would mean ‘cut
unintentionally’. Nevertheless, according to my informants this non-applicative reading is ruled out, so I take
this to be an idiolectal peculiarity of the speech of Smeets’ informants.
xvi
The non-causative/non-applicative verb piaw- does not seem to exist anymore; piaw-ül- means ‘bother
(someone)’.

References

Arnold, J. 1997. "The inverse system in Mapudungun and other languages." Revista de
Lingüística Teórica y Aplicada 34:9-48.
Augusta, F. 1903. Gramática araucana. Valdivia: Imprenta Central J. Lampert.
Augusta, F. 1910. Lecturas araucanas (narraciones, costumbres, cuentos, canciones, etc.).
Valdivia: Imprenta de la Prefectura Apostólica.
Augusta, F. 1916 . Diccionario araucano. Mapuche-español, español-mapuche. Santiago:
Cerro Manquehue.
Falkner, T. 1774. A Description of Patagonia, and the Adjoining Parts of South America.
Hereford: C. Pugh.
Febrés, A. 1765. Arte de la Lengua General del Reyno de Chile. Lima: Calle de la
Encarnación.
Golluscio, L. 2007. "Morphological causatives and split intransitivity in Mapudungun."
International Journal of American Linguistics 73:209-238.
Harmelink, B. 1992. "La incorporación nominal en el mapudungun." Lenguas Modernas
19:129-138.
Harmelink, B. 1996. Manual de aprendizaje del idioma mapuche: aspectos morfológicos y
sintácticos. Temuco: Ediciones Universidad de la Frontera.
21

Havestadt, B. 1777 . Chilidúgú sive tractatus linguae chilensis. Leipzig: B.G. Teubner.
Lenz, R. 1895-1897. Estudios araucanos. Materiales para el estudio de la lengua, la
literatura i las costumbres de los indios araucanos [...]. Santiago: Imprenta Cervantes.
Malvestiti, M. 2006. "Polyvalence in Mapuzungun: Contributions from a Patagonian variety
of the language." In Lexical Categories and Root Classes in Amerindian Languages, X.
Lois and V. Vapnarsky (eds.), 189-209. Berne: Peter Lang.
Rivano, E. 1991. Topology and Dynamics of Interactions. With Special Reference to Spanish
and Mapudungu. Lund: Lund University Press.
Salas, A. 2006. El mapuche o araucano. Fonología, gramática y antología de cuentos.
Santiago: Centro de Estudios Públicos.
Smeets, I. 1989. A Mapuche grammar. Ph. D. dissertation, Rijksuniversitet te Leiden.
Smeets, I. 2007. A Grammar of Mapuche. Berlin / New York: Mouton de Gruyter.
Valdivia, L. 1606 . Arte y confesionario de la lengua de Chile. Leipzig: B.G. Teubner.
Zúñiga, F. 2001. "Dos progresivos y dos resultativos en el mapudungún." LIAMES 1:63-75.
Zúñiga, F. 2006. Mapudungun—el habla mapuche. Introducción a la lengua mapuche, con
notas comparativas y un CD. Santiago: Centro de Estudios Públicos.
Zúñiga, F. in preparation. Benefactive and malefactive applicativization in Mapudungun.

You might also like