WHERE HAVE ALL THE VERBS GONE? REMARKS ON THE ORGANISATION OF LANGUAGES WITH CLOSED VERB CLASSES

Full text

1
11th Binnenial Rice Univerity Linguistics Symposium, 16-18 March 2006
Where have all the verbs gone? Remarks on the organisation of languages
with small, closed verb classes
Andrew Pawley
Australian National University
1. Questions
In most of the world’s languages verb roots comprise an open class. There are, however, some
languages that have a small, closed class of verb roots and, furthermore, lack processes for
deriving new, morphologically complex verb bases. Where have all the verbs gone in these
languages? Do languages with closed verb classes exhibit common principles of semantic and
grammatical organisation that distinguish them from open verb class languages? Is there a
correlation between particular kinds of ‘verbal’ concepts and particular kinds of complex
predicate structures? Are there any verbal concepts that are always expressed by a single verb
root? How do closed verb class systems arise?
2. Approach and background
We compare verbs and complex predicates in three unrelated languages – one with an open
verb class (English) and two with small, closed classes of verb roots that occur as independent
verbs and inflecting for subject person-and-number and tense, aspect and mood, one from
north central Australia and one from the Trans New Guinea family.
In northern Australia many languages have a closed class of inflecting verbs. Some
languages have as few as five to a dozen such verbs, others up to about 250. Jaminjung
represents a type in which most inflecting verbs function as independent verbs and have not
been strongly gramnmaticised (Schultze-Berndt 2000).
In New Guinea a number of languages of the Trans New Guinea family have between 60
and 150 inflecting verbs, and all or nearly all verb roots can occur as independent verbs. These
include the Chimbu-Wahgi languages centred in the Western Highlands and Chimbu
Provinces of Papua New Guinea and Kalam and its close relative, Kobon, spoken in the
Bismarck and Schrader Ranges, Madang Province. Kalam, which has about 130 verb roots, is
the best described. There is a large dictionary and this is probably close to an exhaustive list.
Jaminjung and Kalam both make up for their paucity of verb roots by having large, open
classes of complex predicates. Before considering the types and functions of CPs, let us look at
some properties of verb roots in the three languages that may be relevant to understanding how
CPs work.
2. Some verbs are more important than others. Notes on verb root
meanings, frequency and polysemy
Some possible universals emerge from a study of the meanings, frequency and polysemy of
verb roots in these three languages.
2.1 Frequency
Are some verb root meanings much more common than others in discourse? Do different
languages exhibit a similar frequency ranking? We can only give very rough answers to these
questions because frequently used verbs tend to be highly polysemous and none of the data

2
sources distinguish the different senses of verb roots. A necessary compromise is to assign a
single, ‘cover’ gloss to each verb root.
Jaminjung. The 10 most common verbs in Jaminjung make up 82.2 percent of all verb root
tokens in a sample of text. The most frequent 15 verbs account for 90.5 percent of tokens
(Schultze-Berndt 2000:415-417).
Table 1 Frequencies of productive inflecting verbs in Jaminjung
verb percent verb percent
1. -yu(nggu) BE 22.2 14. -minda EAT 1.5
2. -ijga GO 13.2 15. -wa BITE 1.4
3. -mili GET/HANDLE 7.8 16. -ina(ngga) CHOP 1.2
4. -yu(ngu) SAY/DO 7.1 17. -unga LEAVE 1.1
5. -arra PUT 6.9 18. -anJama BRING 1.0
6. -ruma COME 6.5 19. -irriga COOK 1.0
7. -ma HIT 5.9 20. -wardgiya THROW 1.0
8. -irdba FALL 5.2 21. -irna BURN 0.9
9. -ngawu SEE 4.5 22. -arrga APPROACH 0.8
10. -uga TAKE 2.9 23. -wardagarra FOLLOW 0.7
11. -ngarna GIVE 2.1 24. -(ma)linyma MAKE 0.6
12. -muwa HAVE 1.8 25. -inama KICK/STEP 0.5
13. -ijja POKE 1.7 26. -yungga TAKE AWAY 0.2
Kalam. In a corpus of about 14,000 words representing a variety of spoken texts, the 10 most
frequent verbs in Kalam make up 78.5 percent of verb root tokens. Fifteen verb roots make up
89.6 percent of such tokens and 30 roots account for 97.6 percent. However, all the most
common verbs are highly polysemous (see below) and these figures do not show the
frequencies of particular verb senses.
Table 2 The most frequently occurring verb roots in Kalam
percent percent
1. ag- SAY/SOUND 15.9 16. ñag- SHOOT 1.2
2. g- MAKE/HAPPEN 12.4 17. agi- HEAT S.TH. 0.9
3 md- STAY/BE 9.3 18. ym- PLANT 0.8
4. am- GO 9.3 19. yok- MOVE AWAY 0.73
5. n•- PERCEIVE 7.2 20. yap- DESCEND 0.65
6. ap- COME 7.0 21. tan- ASCEND 0.54
7. ñ•- CONSUME 5.3 22. kum- DIE/MALFUNCTION 0.53
8. ay- STABILISE 5.0 23. yom- SHOW 0.45
9. d- HOLD 4.9 24. su- BITE 0.42
10. pk- HIT 3.4 25. taw- STEP 0.35
11. tk- SEVER/PART 3.3 26. ju- WITHDRAW 0.25
12. ñ- TRANSFER 2.3 27. wok- EJECT FR. MOUTH 0.23
13. kn- LIE DOWN 2.1 28. ad- COOK IN OVEN 0.23
14. tag- TRAVEL 1.8 29. tb- CUT/CHOP 0.21
15.jak- STAND 1.5 30. sk- ENTER 0.18

3
English West (1953) gives the frequencies of English words and their main senses and
grammatical uses in a large corpus drawn from written sources. For main or lexical verbs (i.e.
excluding the auxiliaries be, have, will, used to, keep, etc. and other grammaticised uses of verbs
such as supporting do) the frequencies of the most common 30 verb roots in a corpus of
approximately 250,000 such tokens is as follows (with the qualifier that figures for be as a
copula are not available):6
Table 3. The most frequent 30 main verb roots in English (after West 1953)
verb percent verb percentverb percent
1. say 5.0 11. find 1.53 21. bring 1.12
2. make 3.84 12. think 1.43 22. turn 1.11
3. go 3.5 13. tell 1.42 23. hold 1.0
4. have 3.4 14. call 1.39 24. keep 0.97
5. see 2.81 15. seem 1.35 25. hear 0.92
6. take 2.8 16. become 1.29 26. give 0.87
7. do 2.4 17. show 1.28 27. get 0.87
8. know 2.25 18. begin 1.21 28. continue 0.85
9. ask 1.62 19. look 1.16 29. carry 0.85
10. work 1.59 20. stand 1.07 30. open 0.83
The most frequent 10 verb roots make up almost 30 percent of the tokens in West’s corpus.
The top 30 verb roots account for about 52 percent of the total, the top 50 make up about 66
percent and the top 100 about 80 percent.
2.2 Polysemy
There is a correlation between the frequency of a verb root and its degree of polysemy.
Jaminjung. Jaminjung’s 35 verb roots yield a total number of verb root senses of about 100
(my analysis, based on an interpretation of Schultze-Berndt’s data).. The 15 most frequent
verbs have some 70 specific senses between them (discounting ‘basic’ meanings). Of the 70
senses, 48 can occur in independent verbs, i.e. in the absence of a coverb. This is an average of
about 4.6 senses per verb including phrasal uses, and three senses per verb for independent
uses. Less common verbs have fewer senses.
Kalam. The Kalam dictionary distinguishes an average of about three senses per verb root.
Table 4 Number of senses attributed to the the most frequent 30 verbs in Kalam
no. senses no. senses
1. ag- SAY/SOUND 8 16. ñag- SHOOT 4
2. g- MAKE/HAPPEN 7 17. agi- HEAT S.TH. 3
3 md- STAY/BE 7 18. ym- PLANT 1
4. am- GO 6 19. yok- MOVE AWAY 4
5. n•- PERCEIVE 18 20. yap- DESCEND 5
6. ap- COME 5 21. tan- ASCEND 5
7. ñ•- CONSUME 7 22. kum- DIE/MALFUNCTION 8
8. ay- STABILISE 7 23. yom- SHOW 2
9. d- HOLD 10 24. su- BITE 6

4
10. pk- HIT 6 25. taw- STEP 7
11. tk- SEVER/PART 11 26. ju- WITHDRAW 4
12. ñ- TRANSFER 6 27. wok- EJECT FR. MOUTH 1
13. kn- LIE DOWN 5 28. ad- COOK IN OVEN 1
14. tag- TRAVEL 3 29. tb- CUT/CHOP 1
15. jak- STAND 4 30. sk- ENTER 2
The most frequent 10 verb roots average 8.7 senses each. The second most frequent 10 average
4.6 senses each. The third most frequent 10 average 3.7 senses each. The remaining verbs
(105) average 2.2 senses each.
English. The 10 most polysemous English verbs in The Macquarie total 552 numbered senses
– an average of 55 senses per root. The 200 most polysemous verbs total over 3000 senses, an
average of 15 per root.
Table 5 Number of main senses for the most polysemous English
verb roots in The Macquarie Dictionary of English
run 78 draw 33
take 73 call 32
strike 61 keep 32
break 60 raise 32
turn 55 rise 32
set 54 spring 32
pass 48 touch 32
roll 44 beat 31
open 40 cast 31
cut 39 sink 31
go 37 serve 30
drop 36 work 30
make 36 bear 29
play 36 blow 29
shoot 36 do 29
There is a strong correlation between frequency and extent of polysemy. Of the 30 most
frequent English verb roots, 23 have more than ten senses. The correlations have not yet been
fully charted.
Although Jaminjung and Kalam differ greatly from English in the size of the verb root stock,
all three languages show certain striking similarities in regard to the frequency and meaning of
particular concepts expressed by inflecting verbs.
(i) A 40/50 relation. In all three languages fewer than 40 verb roots make up more than 50
percent of the total verb root tokens in text. (The reference here is to lexical verbs, not to verbs
that have taken on grammatical functions, e.g. semantically bleached verbs whose sole or main
function is to carry inflections, or verbs that have become auxiliaries or markers of aspect.)
(ii) Frequency/polysemy correlation. High frequency verb roots tend to be much more
polysemous than less frequent verb roots.

5
(iii) Rank order of concepts. The basic meanings and frequency rankings of the 20 most
common verb roots in the three languages show a number of correspondences.
The commonalities summarised in (1)-(iii) may be proposed as possible universals.
2.4 Are there any universal verb root meanings?
Typically ‘verbal’ meanings can be defined cross-linguistically as those which are expressed
by verb roots in most language families. An alternative approach, following Hopper and
Thompson (1984), is to rank meanings in terms of how closely they correspond to dynamic
events opposed to durable states. A comparison of Jaminjung (and other Australian languages)
and Kalam (and other TNG languages) suggests the following hypothesis:
Almost all typically ‘verbal’ concepts are expressed phrasally in one language or another.
Are there any verbal concepts that are expressed by a verb root in every natural language?
Our sample of languages is too small to speak with confidence on this matter but it offers some
pointers. One might make a short list of candidates by comparing the sense units that are
common to Kalam, Jaminjung and other languages with small, closed verb clases.1\
Meanings expressed by a single verb root in both Kalam and Jaminjung. To be added.
3. Where have all the other verbs gone? Gone to CPs every one
Dixon (1991) offers a detailed taxonomy of the 900 commonest English verb roots based on
their meanings and syntactic behaviour. He recognises about 30 major types of verbal concepts
and examines how their semantic properties map onto syntactic relations.
Verbal concepts naturally divide into two sorts:
PRIMARY – those directly referring to some activity or state, i.e. verbs which can make
up a sentence by themselves with appropriate NPs filling the various semantic roles, e.g. I
HIT her, She SWAM across the river, He MUNCHED the apple, They WATCHED it.
These are lexical verbs in every language.
SECONDARY – those providing semantic modification of some other verb, with which
they are in syntactic or morphological construction, e.g… I MAY hit her, She TRIED to
swim across the river, We STOPPED him munching an apple, I LET them watch it.
(Dixon 1991:88)
These major categories have subclasses, given arbitrary labels A, B, etc. PRIMARY-A verbs
must have NPs (not complement clauses) in subject and object slots. The semantic types with
this property are:
MOTION, e.g. run, return, take, pull, throw, fall, spill
REST, e.g. sit, stay, put, hang, surround, hold
AFFECT, e.g. hit, punch, cut, sweep, cover, twist, burn
GIVING, e.g. give, lend, pay, present, donate, exchangeå
CORPOREAL, e.g. eat, taste, kiss, laugh, sleep, bleed
WEATHER, e.g. rain, snow, thunder, hail
COMPETITION, e.g. beat, win, attack, lose, compete
SOCIAL CONTRACT, e.g. appoint, govern, manage, join, marry
USING, e.g. use, employ, operate, wear, waste
OBEYING, e.g. obey, process, deal with, grant, perform
PRIMARY-B verb may have NPs filling subject and object slots but they also allow – as an
alternative – a complement clause to fill one of the slots, e.g. I understand my father, I
understand that he refused to sign the document,… [Types include]

6
ATTENTION, e.g. see, hear, notice, discover, watch
THINKING, e.g. think (of/about/over), imagine, assume; know, learn, understand,
realise; believe, suspect
DECIDING, e.g. decide (on), choose, resolve, elect
SPEAKING, e.g. shout, remark, propose, inform, tell, order, ask, promise
LIKING, e.g. like, love, hate, loathe, prefer, envy
ACTING, e.g. act, behave, copy, imitate, reproduce
HAPPENING, e.g. happen, take place, commit, experience, undergo
There are also two types that may have complement clauses in both A and O slots:
COMPARING, e.g. resemble, differ (from), compare, measure, cost
RELATING, e.g. depend on, relate to, imply, due
Various types of SECONDARY verbs are distinguished. Secondary A verbs have the
same subject as the verb they modify.
MODALS, e.g. will, can, should, must
SEMI-MODALS, e.g. be going to, be able to, have got to
BEGINNING, e.g. begin, start, finish, complete, continue
TRYING, e.g. try, attempt, succeed, fail, practise
HURRYING, e.g. hurry, hasten, dawdle
DARING, e.g. dare, venture
Secondary B verbs introduce an extra role, the Principal or the Timer, as subject of the main
verb.
WANTING, e.g. want, wish (for), hope (for), need, require, expect, intend
POSTPONING, e.g. postpone, delay, defer, avoid
Secondary C verbs introduce an added role, the Helper or Causer, as subject of the main clause.
MAKING, e.g. make, force, cause, tempt, let, allow, prevent
HELPING, e.g. help, aid, assist
Secondary D verbs add a role introduced by to:
SEEM: seem, appear, happen, look
MATTER: matter, count
Each of the major classes has various subclasses. For example, Dixon divides MOTION verbs
into seven subtypes, including mode of motion (run, walk, crawl, slide, wriggle). REST verbs
are divided into six subtypes including stance verbs (a. sit, stand, lie, b, stay, settle); and so on.
English-Kalam verb correspondences. About half of the 30 major semantically-motivated
classes of verbs recognised by Dixon for English have no semantic equivalents among the verb
roots of Kalam. There are no verb roots for the Primary classes he calls weather, social
contract, obeying, deciding, liking, comparing and relating. The same is true for all or almost
all of the Secondary classes helping, daring, hurrying, postponing, trying, seeming, mattering,
modals and semi-modals. For each of several other Primary classes, e.g. speaking, thinking,
attention, giving, making, happening and corporeal, Kalam has just a single verb root or a few
roots that are comparable in meaning (though it should be kept in mind that these are
polysemous). It has somewhat larger sets of verb roots for some semantic categories, e.g.
motion (along a path), posture and affecting.
English-Jaminjung correspondences. Even fewer correspondences. Details to be determined.

7
4. Complex predicates in Jaminjung
Jaminjung inflecting verbs and compound verbs
In Jaminjung every finite clause must have an inflecting verb. Inflecting verb roots are a
closed class with about 35 members, of which only 26 are productive. All productive
inflecting verbs can occur as simple (i.e. independent) verbs. Simple verbs make up 40
percent of all verbal predicates in text.
Jaminjung makes extensive use of phrasal predicates which Schultze-Berndt (2000) calls
‘compound verbs’. Compound verbs consist minimally of two words: a non-inflecting
‘coverb’ and an inflecting verb, which together form a semantic unit.
(1) [baramaj gani-yu] lambarra-ni.
promise 3sg:3sg-say.PAST + father.in.law-ERG
‘He promised it (i.e. to give his daughter), the father-in -law.’
Compound verbs form an open class of lexicalised phrasal predicates, with over 1000 recorded
members.
All 26 productive inflecting verbs participate in compound verbs, although seven occur
with many more coverbs than the rest. The semantic relationships between coverbs and
inflected verbs are for the most part transparent, given a fine-grained analysis of contextually
conditioned verb root senses. The kinds of meanings that inflected verbs contribute to such
compounds are largely lexical, although some grammaticisation is evident.
Jaminjung lacks serial verb constructions. It seems that the semantic range of
compound verbs is extended to cover certain event sequences, as in:
(2) Lanhginy-bina [lutrr ga-rdba-ny].
wood-ALL pierce 3sg-FALL-PAST
‘He fell onto a stick so that it poked him.’ (Schultze-Berndt 2000: 488)
The reliance on complex predicates (60 percent of verbal predicate tokens in text) is much
higher in Jaminjung than in Kalam (25 percent). This difference correlates with the fact that in
Kalam almost all senses of high frequency verb roots can be used independently whereas this
is not is the case in Jaminjung.
Generic verbs. “Jaminjung…[has] an overt system of event categorisation” (Schultze-Berndt
2000:211). The productive verb roots are termed ‘generic’ or ‘event classifying’ verbs
because all phrasal predicates can be grouped into a limited number of semantic classes
according to the generic verb they contain. As a generic verb is obligatory in every clause,
these verbs exhaustively carve up the semantic space covered by verbal predicates.
In other words, a Jaminjung … speaker has to select one of a small number of verbs
in every finite clause, and thereby ‘sorts’ all event expressions into a small number
of categories. …[It] has to be shown, of course that the choice of verb has a
semantic basis, rather than being randomly or lexically determined by a given
coverb. If this can be demonstrated one can ask what constitutes the conceptual
basis of event categorisation in this language. (Schultze-Berndt 2000:5)
Some of the productive inflecting verbs are highly polysemous. However, Schultze-
Berndt proposes an analysis of each verb in which most or all of its senses are subsumed under,

8
or are natural extensions of a single basic sense. These proposals underpin her claim that the
semantic relationship between inflecting verbs and coverbs is very largely systematic.
Coverbs. Jaminjung coverbs are treated by Schultze-Berndt as a major part of speech distinct
from nominals, verbs and minor word classes. The class of coverbs mainly covers semantic
domains that are associated, cross-linguistically, with verbs but it also covers some domains
associated with adverbs, adjectives, prepositions and nominals, e.g. sound emission, motor
pattern, posture, transfer, direction of gaze, colour, manner, and phase of event. A subset of
coverbs is sound symbolic.
Usually a coverb is compatible with just a small range of classifying verb partners. Over
20 classes of coverbs are distinguished according to the verbs they combine with. Semantic
relations between verbs and coverbs are very largely systematic. Seven generic verbs occur
with more than 20 coverbs: those glossed BE, FALL, GO, COME, TAKE, GET/HANDLE,
HIT, SAY/DO.
Jaminjung complex verbs as both constructions and lexical units. Each class of coverbs
together with its classifying verb(s) forms a minor construction type. The immediate
constituents of the construction are a pair of lexical classes, the coverb and inflecting verb,
both with highly restricted memberships. Each such construction generates a set of compound
verbs with a semantic dependency constraint in which a particular class of coverbs selects a
particular sense of the inflecting verb. At the same time each compound verb generated by the
construction is a conventional expression, the standard way of representing a type of event.
Schultze-Berndt writes as follows of the double role of Jaminjung complex verbs:
In the view maintained here,…canonical complex verbs have a dual nature;
they are licensed by a productive construction, and at the same time constitute
lexicalised collocations… By describing restrictions on complex verb
formations, one therefore simultaneously describes lexicalisation patterns
(after all the majority of predicates in Jaminjung are complex rather than
simple), and restrictions on the fillers of a construction. (Schultze-Berndt
2000:525)
It might be objected that most of the complex predicates of Jamnjung are too irregular in
their semantics to be called productive constructions. If a particular sense of an inflecting verb
occurs only when the verb collocates with a small set of coverbs aren’t the resulting complex
verbs idioms, and so best treated as purely lexical? The answer is that they are simply
behaving as high frequency verbs do in every language, i.e. having a number of senses that
apply in ‘restricted collocations’. Compare the different senses of the English verb pull that
are selected by different nouns in verb + direct object constructions, such as pull a punch (use
less than full power), pull a trick (perform deceptively), pull a muscle (strain by extending),
pull a horse (of a jockey, hold or check the horse when it is galloping) and pull a drive (in
golf, cricket, etc., hit a ball so that it flies to the left, in the case of a right-handed striker).
The growing body of corpus-based research on phrasal expressions confirm the
pervasiveness of such expressions in spoken and written English (for extensive references, see
Cowie 1998, Wray 2002). For example, almost half the verb + complement constructions in
English written discourse consist of restricted collocations (Cowie 1991).

9
5. Complex predicates in Kalam
In Kalam the unmarked word order is SOV. Locatives can precede or follow the verb. The
morphological and syntactical properties of verbs are highly distinctive. Verb roots are the
only part of speech to carry inflectional suffixes marking tense, aspect or mood, Subject
person-and-number, and switch reference. Coordinate-dependent verbs (also known as ‘medial
verbs’) are marked for switch reference, i.e. same or different subject relative to the following
verb in the sentence.
5.1 Verb adjunct constructions and quasi verb-adjunct constructions.
Verb adjunct constructions (VACS) denote a single conceptual event broken down into
generic and specific semantic components. The generic component is marked by the verb root,
which classifies the event as being of a certain general type. The adjunct gives more specific
information about the event – e.g. manner, concomitant, or effect.
Quasi verb-adjunct constructions differ from true verb-adjunct constructions in that in
the former the adjunct is an argument of verb whereas in the latter it is not. Quasi verb-
adjuncts often contribute specific information about the instrument, location, or agent of an
event.
Verb adjunct constructions are exemplified by (11)-(13). In these and later examples the
verb adjunct and verb roots appear in bold type and the complex predicate is enclosed in square
brackets.
(3) Sawan [guglum ag-ig] k-j-a-p.
Sawan snore say-SS.SIM sleep-PRES.PROG-3SG
‘Sawan is asleep, snoring.’ (Pawley and Bulmer 2003:135)
(4) Yakt omay ok [gub g-i] md-p-it.
bird two the perching.together do-SS.PRIOR stay-PERF-3DL
‘The two birds are perched together .’
(5) Kuypep ognap [kosi•d kn-b-ay].
water-rat sometimes stacked sleep-PF-3PL
‘Water-rats sometimes sleep piled one on another.’
(6) Yakt kotp kasek [gdey-bdey g-ya-k].
bird house quickly roughly.arranging do-3PL-PAST
‘They quickly built a roughly-constructed hide.’
Verb adjunct constructions closely resemble the compound verbs of Jaminjung,
(i) they consist minimally of an uninflectable root or derived form, the verb adjunct, plus a
verb root. The verb root is inflected except when the VAC functions as a non-final serial verb.
(ii) a given verb adjunct occurs only, or primarily as the partner of one or a very few verb
roots.
(iii) The verb marks the phrase as denoting an event as being of a certain general semantic
type and the verb adjunct that select that verb root specify particular subtypes. the verb carries
a general meaning and the verb adjunct more specific meaning. Some verbs in VACs are light
verbs, which contribute little to the meaning of the predicate but most have one of their usual
lexical senses.

10
(iv) Usually the adjunct and verb root are contiguous but certain kinds of material may
intervene, e.g. locative phrases, as in the final clause of (7), question words, as in (8), and
modifiers and negators.
(7) Tap tubtub-toktok wad g-i, [dad ms amn-a-k].
Things knick-knacks bag do-SS-PRIOR carrying outside go-3SG-PAST
‘Having put his personal effects in a string bag, he carried them outside.’
(8) Pa-skoy [si etp-nen ag-a-k]?
girl-small crying what-for say-3SG-PAST
‘Why did the girl cry?’
Are verb adjuncts a distinct part of speech?
Verb adjuncts are sharply distinct from verbs. Verb adjuncts cannot take verbal inflections
and cannot occur alone as the head of a predicate phrase. They differ from adjectives, which
can freely modify nouns. Rather more problematic is the comparison with adverbs and nouns.
Verb adjuncts resemble adverbs in being modifiers of verbs. In fact, Schultze-Berndt
(2000) treats adverbs in the north Australian language, Jaminjung, as a subclass of ‘coverbs’.
However, Kalam verb adjuncts differ from typical adverbs in several respects: (a) adding an
adjunct in some cases creates a complex predicate with a different argument structure from the
verb root alone, whereas adverbs do not affect the valency of the predicate phrase, (b) an
adverb can combine with many different verb roots whereas a verb adjunct can combine only
with a very restricted set, (c) adjuncts can take at least some adjectival modifiers but adverbs
cannot, (d) adverbs modify the verb qualitatively and in a general sense, e.g. for speed,
intensity or purposefulness, whereas verb adjuncts modify the verb in a different way,
typically specifying a particular kind of action, process or state that can be considered a
subtype of the broader category of event specified by the verb, and (e) the scope of verb
adjuncts is restricted to the verb, whereas adverbs may modify the whole VP.
Canonical verb adjuncts differ from nouns in that they are cannot be possessed or
quantified and cannot be questioned, topicalised or relativised. However, verb adjuncts
resemble nouns in that they can be modified by certain adjectives, such as yob ‘big, loud’
koªay ‘many’, tmey ‘bad, intensifier’, naban ‘very’ and in that they represent a propositional
participant of a performance verb such as ag- SAY and g- DO, i.e. the adjunct represents
something said or done. One has to say something and one has to do something. Thus in (4)
gub ‘perching together’ is the action performed, and in (8) si ‘crying’ is the sound that is
uttered even though neither verb adjunct has strong claims to be the direct object. In these
respects they resemble incorporated objects in many languages, some coverbs of Jaminjung
(Schultze-Berndt 2000:189-90) and, to some extent, nominals in Hindi complex predicates
which serve both as predicates and as arguments (Mohanan 1997).
Some classes of verb adjuncts and verb adjunct constructions
Verb adjuncts fall into a few large classes and many more minor classes according to their
meaning and the classifying verb they select. Examples follow from the four largest classes:
(a) adjuncts which select ag- SAY, forming constructions which denote kinds of sound or
sound-making.
(9) Some verb adjunct constructions containing ag- SAY
Verb adjuncts that select ag- almost all refer to kinds of noises, e.g.
asb ag- (begging say) keep asking for things to be given, beg constantly

11
bu ag- (exploding say) explode, burst
gigu ag- (rattling say) rattle, jingle
gu ag- (resounding say) resound, thud
guglak ag- (croaking say) croak
guglum ag- (snoring say) snore
gullg ag- (rumbling say) rumble, growl, as an empty stomach
paj ag- (taunting say) taunt or show pleasure to s.o., over their misfortune
(b) Adjuncts that selects a verb of locomotion, and forming constructions denoting manner,
direction or concomitant of the actor’s movement. The locomotion verbs that take the widest
range of verb adjuncts are am- ‘go’ and ap- ‘come, appear’. Some adjuncts also occur with
tag- ‘walk about, travel’, sa•d- ‘depart’, yap- ‘go down, fall’, tan- ‘go up, climb’, ju-
‘withdraw’, yok- ‘move away, be displaced’. Most verb adjuncts that occur with a verb of
locomotion fall into one or another of the semantic categories shown in (10).
(10) Some verb adjuncts occurring with am- GO and other verbs denoting locomotion
1. Manner of actor’s movement along a path:
kle•d am- (crawling go), ‘crawl’
mumlokd am- (rolling, tumbling go) ‘fall over, tumble down’
pug-tkd am- (running go) ‘run’
tawd am- (stepping go) ‘walk’
wald am- (sliding, slithering go) ‘slide, slither
wa•d am- (carried by water’ go) ‘float along’
w•d am- (flying go) ‘fly’
2. Direction of actor’s movement in relation to path or deictic centre:
gogeb-mageb am- (zigzagging go) ‘zigzag along’
koda• kodo• am- (hither and thither go) ‘go hither and thither’
pa•d am- (passing out of sight go) ‘pass from sight (in the distance)’
adkd am- (turning back, reversing go) ‘turn back’
skd am-d (entering go) ‘enter, go in
tkd am- (crossing over go) ‘cross over (a divide)’
3. Concomitant participant, manner of actor’s association with another participant:
dad am- (carrying go) ;carry s.th.’
po• d am- (guiding, leading go) ‘guide s.o., take s.o.. (to a place)’
sb am- (brandishing go) ‘go brandishing s.th. (usu. a weapon)’
kosi•d am- (stacked, carried on the back go) ‘go carrying s.th on the back’
wtsek am- (chasing go) ‘chase, pursue s.o.’.
Adjuncts of type 3 are valency-changing, making the verb adjunct phrase transitive and introducing
a patient or theme.
(c) A third class consists of adjuncts that select d- HOLD, forming constructions that denote
an event where (a) the experiencer is seized by sudden involuntary bodily or emotional
reaction, or (b) the actor comes to a sudden stop.
(11) Some verb adjunct phrases containing d- HOLD

12
(a) sudden involuntary reaction
glk d- (startled hold) be startled, taken by surprise
jep d- (trembling hold) tremble, shiver, quake
llmag d- (twitching hold) (of body part) twitch, have involuntary
movements
nab• d- (shame hold) be ashamed, embarrassed
ñawl d- (astonished hold) be astonished by s.o.’s accusation or anger
(b) semantically diverse
cci d- (crushing hold) crush in the hands
cebsogi d- (swoop.and.perching hold) swoop down and perch
jª d- (sharp.pointing hold) put a sharp point on s.th.
kobti-spak d- (breaching hold) cause a dam to burst
kopag d- (coiling hold) be coiled, twisted (as a rope)
pteªd d- (jumping.onto hold) land after leaping, jump onto s.th.
sb d- (trimming hold) trim s.th, cut off unwanted parts
(d) The verb that takes the largest number of verb adjuncts is g- DO, the most general activity
and process verb. Adjuncts that select g- DO generally denote either (i) the performance of an
intentional act by an animate actor, (ii) a dynamic event attributed to an inanimate effector, or
(iii) a state. Examples of verb adjuncts combining with g- are given in (12), with a division into
four broad semantic and grammatical categories. Note that some verb adjunct phrases belong
to more than one category, e.g. some adjunct phrases have both a transitive and a stative
reading. Although g- is glossed as ‘do’ throughout, in some contexts the gloss might equally
well be ‘make’ and in others ‘be’.
(12) Some verb adjuncts occurring with g- DO
(i) intransitive VACs denoting the performance of intentional act by animate actor
gsey-bsey g- (hurry-scurry do) hurry, be in a hurry or rush
gub g- (perching.together do) (of birds) perch together
jlan g- (nodding do) nod the head rhythmically
ñk g- (crouching do) crouch, duck
ññlo• ay g (pretending put do) tease, pretend to do
tob mogm g- (leg crossing do) cross the legs (tob is a noun)
wog-wati g- (work do) do work of any kind
(ii) transitive VACs denoting the performance of intentional acts by animate
actors
amñeb g- (feinting do) pretend to attack, feint
bl g- (abstaining do) abstain from an activity
gadal-badal g- (criss-cross do) lay things criss-cross or higgledy-piggledy
gdey-bdey g- (rumble-jumble do) do something roughly
saj g- (compensation do) pay compensation
smen g- (bridewealth do) give bridewealth
mkal g- (open-hinged do) open up something hinged
tug cckoy g- (by.hand turning do) turn s.th. round and round (by hand)
onim g- (on do) switch on (electric power)

13
(iii) intransitive VACs denoting acts by inanimate effectors
gti•-gto• g- (ding-dong do) make a lot of noise, make a din or racket
gley-wley g- (clitter-clatter do) rattle, clatter, as bones, metal objects
godey-bodey g- (revolving do) swing round, revolve like a propellor
kopay-mopay g- (blowing.fiercely, do) (of storm-winds) blow fiercely, rage
kulpl g- (vibrating.loudly do) vibrate loudly
lm g- (shooting do) (of plant suckers) shoot up, break the surface
pbo• g- (hot do) (of weather) be warm
takl g- (cold do) (of weather) be cold
wnwn g- (peeling do) (of e.g. bark) peel off, fall to bits
ygen g- (wind do) be windy, blow
(iv) intransitive VACs denoting states
gadal-badal g- (criss-crossed do) be higgledy-piggledy, criss-crossed
gdey-bdey g- (rumple-dumpled do) be in disarray
gogeb-mageb g- (twisted do) be twisted, crooked
gutgat g- (drenched do) be drenched and cold from rain or damp foliage
kolkol g- (tangled do) be tangled
jl g- (loose-fitting do) be loose, loose-fitting
jspk-maspk g (crunch-scrunched do) broken into small pieces, crushed to bits
g- also occurs with many quasi verb adjuncts, noun-like elements, to form constructions
which denote involuntary bodily and mental processes and states. Here the verb adjuncts
exhibit some of the properties of subject NPs, and the experiencer is clearly the direct object.
(
Some combinations of quasi verb adjuncts + g- DO denoting involuntarily bodily and
mental processes
(13) glgl g- (shaking do) tremble, shiver, have the shakes
jel g- (afraid do) be wary, afraid, nervous, apprehensive
jep-ptpt g- (shivering do) shiver, tremble, esp with fear
kaj-knm g- (wincing do) wince, shudder, on hearing a grating sound
kalaw g- (numb do) be numb, paralysed
km g- (bitter do) taste bitter, sour
nab• g (shame do) be ashamed, shy
ñekñek g- (hiccup do) have the hiccups
llmag g- (terrified do) be terrified, have one’s hair stand on end
slk g- (itching do) be itchy, bitter, hot (to taste)
tap g- (sickness do) be sick
wsb g- (sweat do) sweat, be sweaty
True verb adjuncts form an open class which freely accepts borrowings. Almost all such
borrowings (such as the verb adjuncts wasim ‘wash’, onim ‘switch on’, opim ‘open’, statim
‘start’ and pasim ‘close, secure’, from Tok Pisin) pair with g-.
Argument sharing in VACs
The combined arguments of certain VACs may exceed those of all of its constituent predicates. In
(14) we find a verb adjunct po•d ‘guiding, leading, accompanying (a person or animal)’ paired with
the verb ap- ‘come’.

14
(14) Yad b tud nup kotp-cn po•d ow-n-k.
I man white him house-our leading come-1S-PAST
‘I’ve brought the white man to our home.’
Together these two elements create a three-place predicate. The adjunct contributes two semantic
arguments: a guider (Actor, Subject) and a guidee (Patient, Object). The verb contributes two
arguments, a comer (Actor, Subject) and a destination (Object). The comer and the guider are
coreferential.
In (15) the adjunct dad ‘carrying (inanimate object)’ is paired with the verb n- ‘join (others on
a trip, catch up with someone who has left’.
(15) Cn np mdak ayn-bogs dad n-ng g-p-un.
we you(Obj) later iron-box carrying join-SS.FUT. do--PF-1P
‘We will join you later bringing the metal trunk.’
Together they create a three-place predicate. The adjunct dad contributes a carrier (Actor), Subject)
and a thing carried (Patient, Object). The verb n- contributes a joiner (Actor, Subject) and a joinee
(Patient, Object). The carrier and the joiner are coreferential. In (21) the thing carried ranks as the
Direct Object and the person joined as the Indirect Object.
5.2 Kalam serial verb constructions.
5.3.1 General characteristics
Serial verb constructions figure more prominently in Kalam than verb adjunct constructions.
Kalam SVCs provide an efficient, streamlined mechanism for depicting certain kinds of event
sequences or overlapping event clusters using a single clause structure. Their grammatical and
semantic structure has been described in some detail elsewhere (Lane 1991, Pawley 1987,
1991, 2004, in press, Pawley and Lane 1998) and Givon (1990, 1991) has demonstrated that
SVCs, in contrast to clauses ending in verbs of full or medium finiteness, almost always occur
under a single intonation contour.
The lexical nucleus of a SVC consists minimally of a verb (root) series, in which one or
more bare verb roots precedes an inflected verb root. In some kinds of SVCs other material
may intervene between verb roots but never a conjunction. Typically all the verbs share the
same subject and the argument structure of the complex predicate does not exceed that of its
constituent verbs. There is no grammatical limit to the number of verb roots that can occur in
sequence in a SVC. In practice the limit seems to be about nine. Sequences of six or seven are
fairly common..
(16) Am mon d ow-ngp-ay.
go wood get come-FUT-3PL
‘They will fetch the wood.’
(17) Basd skop …am kmn pak d ap ad ñb-elgp-al,…
grandfather distant go animal kill get come cook eat-PAST.HAB.3PL
‘Our distant ancestors… used to hunt [lit. go, kill, bring back, cook and eat] game
mammals,…’ (kmn are the larger marsupials and arboreal rodents, which men hunt)
Compact vs narrative SVCs. An important distinction is between SVCs of the compact the
narrative types. (van Staden and Reesink (2004), writing of languages of the New Guinea

15
area in general, make a similar distinction between what they call component serialisation and
narrative serialisation..)
5.2.2 Compact SVCs
These are conceptually and grammatically tight-knit, each component event being an essential
part of the whole. No non-verb material can occur within the verb series and the scope of
negation and modifiers is over the whole verb series. Examples are
(a) sequences of locomotion or transport + directional verbs, e.g. those whose meanings are
equivalent to English go in, go out, fall down, withdraw, and pass from sight,
(b) sequences that denote transporting or moving something along a path, e.g. functional
equivalents of bring, take, fetch, push, pull, drag,
(c) sequences of transitive + intransitive verb denoting an action that changes the state of the
affected object , e.g. functional equivalents of the transitive verbs rub off, clean, throw away,
open, close, split, shatter, extract,
(d) sequences of action verb + perception verb, denoting events of testing or experiencing, e.g.
functional equivalents of the transitive verbs taste, feel (by touching), and try on (clothing).
In comparative terms, the meanings conveyed by Kalam SVCs of the types (a)-(d) are
unremarkable. In virtually every language meanings of these sorts are represented by a simple
or complex verb.
They can be divided into many types on grammatical and semantic grounds. There is
space only for a few examples here.
(18) Some same-subject verb series denoting the actor’s movement along a path
am jak- (go reach) arrive at, reach (a far place)
ap jak- (come reach) arrive (hither)
ap tan jak- (come climb reach) rise to the top (as water in a container)
ju ap- (withdraw come) emerge, come out
ju am- (withdraw go) retreat, emerge and depart
ju yap- (withdraw descend) drop out of position, withdraw downwards
tan am- (ascend go) climb up
ap tan ap yap- (come ascend come descend) go back and forth, go up and down
tan ap tan ap yap- (ascend come ascend come descend) climb up and down
sak am- (break.through go) break through and escape (as water from a dam)
sa•d am- (pass.beyond go) go out of sight, pass beyond a distant point
ptk am- (take.fright go) flee, run away in fear
ap yap pk- (come descend hit) fall down
am pk- (go hit) lose consciousness and fall down, faint
pug ju pk- (break withdraw hit) come out and fall, as a tree in a storm
(19) Some verb series denoting testing events, with n•- PERCEIVE/KNOW
ay n•- (put perceive) try to fit s.th., try s.th. on (e.g. clothing)
ñn ay n•- (hand put perceive) feel inside s.th., grope
d n•- (touch perceive) feel s.th. by touching (deliberately)
ñb n•- (consume perceive) taste s.th.
pk n•- (hit perceive) feel s.th. by touching, nudge
pu•i n•- (pierce perceive) probe, test by probing
tb n•- (cut perceive) make a trial cut
wk n•- (burst perceive) test by cracking open

16
ag n•- (say perceive) ask, enquire, ask for, request
ap n•- (come perceive) visit s.o., come and see s.o.
piow n•- (search perceive) search and find, find what one is looking for
tag n•- (travel perceive) sightsee, travel and see
taw tag n•- (tread walk.about perceive) test (ground, branch etc.) by treading
Perception events such as seeing, hearing, thinking and dreaming , which do not require the
perceiver to perform a prior activity, are typically expressed by quasi verb-adjunct phrases.
Different-actor SVCs. In different-actor SVCs the final verb has a different actor from the
preceding verb(s). However, unlike true switch reference constructions, which are multi-clause
constructions, change of grammatical subject is not marked in the different-actor SVCs.
Instead, the last two verbs in the SVC are treated as a compound transitive verb, with the agent
of the transitive verb marked as subject. Change- of-actor SVCs usually express a cause-
effect sequence of sub-events.
(20) Some Kalam verb series with change of logical subject
ag yok- (say displace) send s.o. away, dismiss s.o.
d yok- (hold displace) throw s.th.
pk lug yok- (strike slide displace) brush or sweep s.th. away
pk cg- (strike adhere) stick s.th. on, cause s.th. to adhere
pk ju- (strike withdraw) knock s.th. out of position
pk sak- (strike break.off) knock (a piece) off
pk sug- (strike extinguished) put out (a fire)
pk wk- (strike shatter) knock s.th. to bits, shatter s.th.
pug sug- (blow extinguished) blow out (flame)
pu•i ask- (pierce be.open) prise s.th. open
pu•i lak- (pierce be.split) split s.th by wedging or levering
pu•i md- (press stay) stick s.th. in position
tb yok- (cut displace) cut s.th. away or out
sog ask- (pour free) pour until empty, drain s.th.
su ju- (bite withdraw) extract s.th. by biting, bite s.th.out
su pag yk- (bite broken be.open) bite s.th. (and break it) open
taw pag yok- (tread.on broken displace) break off by pushing with the foot
tb kluk yok- (cut gouge displace) gouge out (the centre of s.th.)
Narrative SVCs are syntactically and semantically more complex than compact SVCs.
Whereas compact SVCs are V-serialising, narrative SVCs are VP-serialising. The semantic
character of narrative SVC s is quite different from the word-like character of compact SVCs.
Narrative SVCs are designed to tell a story consisting of several stages. The semantic content
of narrative SVCs is shaped by discourse structure conventions, specifically the rules for
constructing well-formed reports of an episode, a sequence of related events.
A well-formed report of a deliberate, purposeful action, such as gathering firewood,
digging in the garden or planting a crop should mention not just the central action but a
sequence of associated happenings. You don’t just get firewood, or dig in the garden, or plant

17
crops you do such things for a purpose. There must be a resolution, an outcome. And in many
cases the actor has to perform some preparatory activity, even if this is only going from home
to the scene of the main action. Thus, the report normally consists of an orientation, a
complicating action and a resolution. The orientation is often minimal because the participants
and setting have already been established earlier in the discourse or by the external situation.
However, the complicating action may be quite elaborate. For this class of activities a well-
formed report should at least answer the questions.
1. Did the actor move to the scene of the pivotal action or was s/he already there?
2. What was the pivotal action? If there was a targeted objected, was this obtained?
3. Did the actor then move on to another place and if so, did s/he take along the targeted
object with him?
4. What was the outcome, e.g. was the search abandoned, what was done with the
transported object?
The major conceptual parts of a miminal well-formed report of a deliberate action event
sequence are summarised in (21):
(21) Formula for minimal well-formed report of a deliberate action episode
1 2 3 4
MOVE TO PIVOTAL TRANSPORT OF RESOLUTION
SCENE/STAY ACTION AFFECTED OBJECT
Such a report may be distributed over several clauses may be distributed over many clauses or
a few clauses. If certain conditions are met the report may be compressed into a single SVC,
with some or all of the four major conceptual elements or stages of the narrative represented
by a verb root or a verb series. Let us call the material representing a separate stage a phrase.
For a sequence to qualify as a narrative SVC there must be at least two phrases present, e.g. 1
+ 2, 2 + 4, 2 + 3, 3 + 4. If all four conceptual elements are not present in the SVC it does not
mean the missing elements have been left out of the event report as a whole. It merely means
that these elements appear in separate clauses.
(22) Formula for minimal well-formed report of a gathering event episode
1 2 3 4
MOVEMENT GATHERING TRANSPORT OF DISPOSAL
TO SCENE ACTIVITY GATHERED ITEMS
Some lexicalised verb series denoting gathering events.
(23)a.) am + [mon] puk + d ap + agi kn-
go +wood gather + hold come + ignite sleep
‘fetch firewood for the night’
b. am + [mon] puk + d ap + ay-
go +wood gather + hold come + put
‘fetch firewood and store it’
c. am + [as] + yg pk + d ap + ad ñ•-

18
go + small.mammals + dig hit + hold come + cook eat
‘go and dig for (small animals ), kill, bring, cook and them
d. am +[kmn] pk + d ap + ad ñ•-
go + game.mammal hit + hold come + cook eat
‘go and kill game and bring and cook and eat them
e. am + al•aw kab tk + d ap + ad ñ•-
go + pandanus nut pick + hold come + cook eat
‘go and pick pandanus nuts (and bring and cook and eat them
5.3 Frequency. In a text count of 800 verbal predicates, 75 percent consisted of simple verbs,
19 percent were serial verb constructions and only six percent of phrases were true verb
adjuncts. However, a qualifier must be added. Quasi verb-adjunct phrases were not counted
because these generally consist of a verb plus one or more arguments and as such are not
readily distinguishable from free expressions with the same structures. The total of all types of
verb-adjunct phrase will thus be more than six percent but it is unlikely to be as much as 10
percent.
6. Do Kalam and Jaminjung complex predicates do similar kinds of work?
6.1 Similarities. There is a strong family resemblance between the ‘compound verbs’ of
Jaminjung and some other north Australian languages and the ‘verb adjunct constructions’ of
Kalam and its relatives.
Both Kalam and Jaminjung use verb adjunct constructions/compound verbs to code
events whose components are neither discrete nor sequential. VACs can be subclassified into
fairly systematic semantic classes according to the meanings of the verb adjuncts and the
inflecting verb. Several of the semantic classes of VACs match quite closely across the two
languages, e.g. there are classes representing (a) mode of movement along a path (crawl, run,
jump, swim, fly), (b) internal (fixed base) movements (shiver, sway, shake, lean, stretch), (c)
noise-making events, where a particular kind of sound is produced : laugh, cry, squeal, croak,
thud, etc., (d) events where the actor performs a specific activity, e.g. working, weeding,
crushing, paying compensation, pretending to attack, and (e) events where the actor assumes
a particular posture. It is also noteworthy that in both languages the verb adjunct category
includes a number of expressive or ideophonic elements.
Schultze-Berndt (2000:551-52) comments cautiously on this resemblance, with specific
reference to the category that she calls ‘coverb’ and which I prefer to call ‘verb adjunct’. In
both Jaminjung and Kalam this ‘other category’ is a major part of speech in addition to nouns,
verbs, adjectives and adverbs. In both languages many members of this class correspond,
semantically, to prototypical verbs, but others correspond to nouns, adjectives or adverbs. The
coverb category in Jaminjung encompasses a somewhat wider range of semantic types than the
corresponding category in Kalam. This discrepancy need not invalidate ‘verb adjunct’ as a
candidate for a cross-linguistic grammatical category. It is notorious that languages vary in the
semantic range of the major part of speech categories that we confidently label ‘noun’, ‘verb’
and ‘adjective’.
Verb adjunct is a category with fuzzy boundaries both in Jaminjung, where there is
overlap between adjectival nouns and stative coverbs, and in Kalam, which shows a continuum
between ‘true’ verb adjuncts and ‘quasi’ verb-adjuncts. Quasi verb-adjuncts are nouns or
noun-like elements that in the same clause can serve as an argument of the verb and as a verb
adjunct in a complex predicate. This sort of mismatch between different layers of

19
grammatical and semantic structure is characteristic of complex predicates, a point made by a
number of contributors to Alsina et al. (1997). In particular, Mohanan (1997) speaks of one
type of Hindi NV complex predicates as having a complex argument structure but behaving
syntactically as a simple predicate.
6.2 Differences. Kalam uses a wider range of complex predicate types than Jaminjung to
represent types of events. The most important distinction in Kalam is between serial verb
constructions (SVCs) and verb adjunct constructions (VACs). There is a clear semantic basis
to the distinction. Prototypical SVCs encode a sequence of discrete events that make up a
conceptual whole, with the order of verb roots isomorphic with the order of conceptual events.
Verb-adjunct constructions are used to denote a different class of verbal meanings from those
represented by SVCs. If we operate with the principle that two semantic elements in a
construction count as separate events only if (a) they are represented by separate formal
elements, and (b) they refer to acts (processes, states) that are conceived of either occurring in
sequence or as being separate acts, prototypical VACs denote a single conceptual event which
is broken down into two primary semantic components standing in a generic-specific relation.
In the case of Jaminjung it is not possible to draw a neat line between a type of complex
predicate used to represent sequences of events and another type used to represent single
events that are decomposable into generic and specific components. Jaminjung lacks serial
verb constructions. It compensates by using VACs to represent certain kinds of complex
events which in Kalam would be expressed by SVCs, including (a) sequences of two discrete
events, (b) successive phases of a process performed by the same actor, e.g. causing,
beginning, ending, extending, and (c) discrete sub-events that overlap in time, e.g. rising up
noisily, or making a sound while collapsing.
Events where the actor expels a tangible bodily product, e.g. defecate, urinate, vomit, lay
an egg, are typically depicted in Jaminjung using a VAC. In Kalam, however, most such
events are expressed by a direct object + generic verb construction, where at least some of the
N + V sequences have claims to be considered a quasi VAC.
7. Comparisons with English
We have sene that Jaminjung and Kalam share certain characteristics apart from English in
their grammatical and semantic apparatus for talking about events. However, all three
languages show certain striking similarities in regard to the frequency and meaning of
particular concepts expressed by inflecting verbs.: The fact that three unrelated languages
exhibit strikingly similar patterns in the frequency of particular sorts of verbal meanings
suggests that some universal principles constraining language organisation and change are at
work. It follows that that event-classifying languages should not be regarded as especially
exotic. They merely exploit to an extreme degree features of the semantic organisation and
grammar of predicates that are present, but less developed, in many if not all open verb class
languages.
8. On the origins of small, closed verb classes
There are logical grounds for thinking that closed verb class systems where all or almost all the
verbs have grammaticalised (such as some languages of north-west Australia (McGregor
2002) have developed from systems with predominantly event-classifier verbs.
But how do languages come to have a small, closed class of verb roots in the first place?
A necessary condition is to have a set of high frequency verbs with general meanings. The

20
power of these verbs is that they are always available for forming periphrastic predicates,
which can then be in competition with synonymous simple verbs. However, as all languages
appear to have such verbs this condition is not sufficient. The evidence to hand suggests a
couple of more diagnostic preconditions, namely that a language should have a class of verb
adjuncts, and use some verb roots as event classifiers of these adjuncts. This is a tradeoff step,
because it allows a large class of verbal predicates to be formed using a small number of
generic verbs.
How might languages come by such a verb adjunct category? One route, it seems, is by
reclassifying some unmarked nouns as adjuncts. Haig (2002) notes a bundle of four structural
features that correlate with the development of N + V complex predicates in Iranian and Indo-
Aryan: (a) OV word order, (b) lack of morphological means for deriving verbs from nouns, (c)
zero-marking of direct objects, (d) lack of obligatory articles. To these might be added (e) the
restriction of certain nouns to co-occurrence with a very few verbs, initially on semantic
grounds (cognate object nouns are an example). All these features apply, more or less, to those
TNG languages which have defective nouns acting as quasi verb-adjuncts.
A second possible pathway – one that would allow for verb roots to be turned into verb
adjuncts – is by reanalysing certain existing serial verb constructions. Take an SVC of the sort
that Durie (1997:336) calls ‘manner serialisation’, in which the roots denote sub-events that
are (a) non-consecutive or temporally overlapping, and (b) where the first verb specifies the
specific manner in which the second verb is performed, denotes as in crawl come ‘come
crawling’, walk follow ‘follow on foot, walk after’. If the language already has some verb
adjunct constructions in which a class of uninflected roots (say, of the noun or adjective type)
are classified by a generic verb, the stage is set to reinterpret certain conventional manner
SVCs as verb-adjunct phrases, with the initial bare verb root in the series treated as just
another verb adjunct in partnership with a generic verb.
Possession of a sizeable verb adjunct category is, of course, no guarantee that a language
will develop a closed class of verb roots – certain Indo-European languages, especially in the
Indo-Iranian group, appear to have developed something like a class of verb adjuncts while
maintaining an open class of verb roots. Having a large verb-adjunct category merely takes a
language one step closer.
NOTES
1. It has long been recognised that most verb meanings lend themselves to phrasal expression
and that, as a consequence, languages can be constructed that make do with only a few verb
roots. When C.K. Odgen selected the lexicon of Basic English he included only 18 simple
verbs: be, come, do, get, give, go, have, keep, let, make, may, put, say, seem, send, take and will
(Ogden 1930). The list of 60 or so semantic primitives which Wierzbicka and Goddard posit as
the basis of their universal defining language (Goddard and Wierzbicka 1994, Wierzbicka
1996) contains about a dozen primitives that correspond to meanings typically expressed by
verbs in natural languages. There is just one primitive denoting speech (SAY), three denoting
existence or life (THERE IS, LIVE/ ALIVE and DIE), six ‘mental predicates’ (SEE, HEAR,
THINK, KNOW, WANT, FEEL) and three denoting other actions and events (DO, HAPPEN
(TO), MOVE). The list also includes some other primitives that are realised as verbs in some
languages, including NOT, CAN, BECAUSE, SAME, LIKE (resemble), OTHER, GOOD,
BAD, BIG, and SMALL. It is noteworthy that only three or four of the verbs in Ogden’s list
(be, do, go, say) have approximate counterparts in Wierzbicka and Goddard’s.

21
Kalam Pandanus language has only about nine verb root forms, which allowing for some
homophony probably represent 12 distinct morphemes. All the verb roots are used
independently but are also used extensively in combination with verb adjuncts. The verb
roots have the following basic meanings or macro-senses: MOVE (go, come, walk about,
climb, depart, arrive, jump, etc.), DO (do, act, make, work, cause, build, create, etc.),
HAPPEN (function, occur), STABILISE (put, place, plant, (take) form, become, be, live, stay,
sit, lie, sleep, etc.), BREAK (DESTABILISE IN FORM) (bend, fold, collapse, fracture, chip,
smash, etc.), CONTROL (hold, have, get, carry, stop, manipulate, etc.); FREE (untie, remove,
open, extract, cut off, avoid, make a hole, etc.), SOUND (speak, say, sing, cry, groan, rumble,
crackle, explode, etc.), HEAT (ignite, burn, cook, etc.), PERCEIVE (see, hear, feel, know,
think, etc.), TRANSFER (give, connect, buy, sell, exchange, etc.), CONTACT (hit, press,
poke, copulate, shoot, etc.).
2. Key to abbreviations used in glosses
D dual
DS different Subject (from
following verb)
DUR durative
FUT future
HORT hortative
I Instrument
IMM Immediate past
L Locative
lit. literally
NDR noun derivative suffix
Obj Object (case)
OPT optative
P plural
PF perfect (denotes today’s past,
present perfect or present-
iterative)
PAST remote past (yesterday or
earlier)
PAST.HAB past habitual
PRIOR prior or preceding (the event
denoted by following verb)
REC recent past
PROG present progressive
S singular
SIM simultaneous (with the event
denoted by following verb)
SS same Subject (as following
verb)
Subj Subject

22
REFERENCES (for extended version of paper)
Alsina, Alex, J. Bresnan and P. Sells (eds), 1997. Complex Predicates. Stanford: Center for the Study of
Language and Information.
Cowie, A.P., 1991. Multiword units in newspaper language. In S. Granger (ed.), Perspectives on the English
Lexicon: A Tribute to Jacques van Roey, 101-116. Louvain-la-Neuve: Cahiers de l’Institut de Linguistique
de Louvain.
Davies, John, 1981, Kobon . (Lingua Descriptive Series 3). Amsterdam: North-Holland.
_______, 1991. A New Approach to English Grammar, on Semantic Principles. Oxford: Clarendon.
Durie, Mark, 1997, “Grammatical structures in verb serialization.” In Alsina, Alex, J. Bresnan, and P. Sella, eds,
1997, Complex Predicates, 289-354. Stanford: Center for the Study of Language and Information,
Stanford University.
Farr, Cynthia, 1999. The interface between syntax and discourse in Korafe, a Papuan language of Papua New
Guinea. Canberra: Pacific Linguistics.
Foley, William, 1986. The Papuan languages of New Guinea. Cambridge: CUP.
Foley, William A. and Mike Olson, 1985, “Clausehood and verb serialization.” In J. Nichol (ed.), Grammar
Inside and Outside the Clause, 17-60. Cambridge: Cambridge University Press.
Foley, William and R.D. Van Valin, 1984, Functional Syntax and Universal Grammar. Cambridge, Cambridge
University Press.
Givón, Talmy, 1990, “Verb serialization in Tok Pisin and Kalam: a comparative study of temporal packaging.” In
J.M.W. Verhaar (ed.), In J. Verhaar (ed.) Melanesian Pidgin and Tok Pisin. Studies in Language
Companion Series 20. Amsterdam: Benjamins., 19-55.
1991. “Serial verbs and event cognition in Kalam: an empirical study of cultural relativity.” In Clare Lefebvre
(ed.)
Haig, Geoffrey, 2002. Complex predicates in Kurdish: argument sharing, argument incorporation, or what?
Sprachtypologie und Universalienforschung/Language typology and universals 55(1):25-48.
Heeschen, Volker, 1998. An ethnographic grammar of the Eipo language spoken in the central mountains of Irian
Jaya (West New Guinea). Berlin: Reimer.
2001. “Event-formulas: sentences as minimal narratives.” In A. Pawley, M. Ross and D. Tryon (eds), The boy
from Bundaberg: Studies in Melanesian linguistics in honour of Tom Dutton, pp. 155-173. Canberra: Pacific
Linguistics.
Hopper, Paul and Sandra Thompson, 1984. The discourse basis for lexical categories in universal grammar.
Language 60(4):703-752.
Labov, William, 1973. “The transformation of experience in narrative syntax.” Language in the inner city .(ch. 9)
Philadelphia: University of Pennsylvania Press.
Lane, Jonathan. 1991. Kalam serial verb constructions. MA thesis, Dept. of Anthropology, University of
Auckland. (Revised version in press with Pacific Linguistics.)
Lang, Adrianne, 1975. The Semantics of Classificatory Verbs in Enga (and other Papua New Guinea
Languages). Canberra: Pacific Linguistics.

23
Majnep, Ian Saem and Ralph Bulmer, 1990, Kalam Hunting Traditions, vols 1-6. University of Auckland Dept.
of Anthropology Working Papers, nos 85-90.
n.d., Kalam Hunting Traditions, vols 7-12. Printout. Department of Linguistics, Research School of Pacific
and Asian Studies, Australian National University.
McGregor, William, 1990. Functional Grammar of Gooniyandi. Amsterdam: Benjamins.
______, 2002. Verb Classification in Australian Languages. Berlin: Mouton de Gruyter.53.
Mithun, Marianne, 1984. The evolution of noun incorporation. Language 60(4):847-894..
Mohanan, Tara, 1997. Multidimensionality of representation: NV complex predicates in Hindi. In Alsina et al.,
431-471.
Nash, David, 1982. Warlpiri verb roots and preverbs. Work Papers of SIL-AAB (Summer Institute of Linguistics,
Australian Aborigines Branch), Series A, vol. 6:165-215
1986. Topics in Warlpiri Grammar. New York: Garland Publishing.
Ogden, C.K., 1930. Basic English. A General Introduction with Rules and Grammar. London: Kegan Paul.
Pawley, Andrew, 1987, “Encoding events in Kalam and English: different logics for reporting experience.” In R.
Tomlin (ed.), Coherence and Grounding in Discourse, 329-360. Amsterdam: Benjamins.
1991. “Saying things in Kalam: Reflections on the nature of language and translation.” In A. Pawley (ed.),
Man and a Half : papers in Pacific anthropology and ethnobiology in honour of Ralph Bulmer, 432-444.
Auckland: Polynesian Society.
2004, Episodic event reports in Kalam: case studies from serial verb and multiclause constructions. Paper
presented at conference on Event Representations in Mind and Language, University of Oregon, Sept.
2004.
forthcoming, Compact versus narrative serial verb constructions in Kalam. In Gunter Senft,.ed. Volume on
serial verb constructions from 9-ICAL. Canberra, Pacific Linguistics .
Pawley, Andrew and Ralph Bulmer, 2003. A dictionary of Kalam with ethnographic notes. Prinout, pp. 750.
Department of Linguistics, Research School of Pacific Studies, Australian National University.
Pawley, Andrew and Jonathan Lane, 1998, In Anna Siewierska and Jaj Jung Song (eds), Case Typology and
Grammar, 201-227. Typological Studies in Language 38. Amsterdam/Philadelphia: Benjamins
Rappaport-Hovav, Malka and Beth Levin, 1996, Two types of derived accomplishment. Proceedings of the First
LFG Workshop. Grenoble, August 1996.
Schultze-Berndt, Eva, 2000. Simple and complex verbs in Jamingung. Nijmegen: MPI Series in Psycholinguistics.
Talmy, Leonard, 1985. “Lexicalization patterns: Semantic structure in lexical forms.” In T. Shopen (ed.),
Language Typology and Syntactic Description: vol. 3, Grammatical Categorioes and the Lexicon.
Cambridge: Cambridge University Press.
2000. Towards a cognitive semantics. (2 vols) Cambridge MA: MIT Press.
van Staden, Miriam and Gunter Senft, forthcoming . Serial verb constructions in a linguistic area. In Gunter
Senft,.ed. Volume on serial verb constructions from 9-ICAL. Canberra, Pacific Linguistics . Canberra:
Pacific Linguistics.
Wierzbicka, Anna, 1996. Semantics: Primes and Universals. Oxford/New York: Oxford University Press.
Wray, Alison, 2002. Formulaic Language and the Lexicon. Cambridge: Cambridge University Press.