Review of Verb Classification in Australian Languages
Date: Sun, 22 Dec 2002 12:06:42 -0500
From: Claire Bowern
Subject: McGregor (2002): Verb Classification in Australian languages
McGregor, William B. (2002) Verb Classification in Australian Languages.
Mouton de Gruyter, xxvi+531pp, hardback ISBN 3-11-017141-4.
Claire Bowern: Harvard University and Centre for Research on Language
Change, Australian National University.
William McGregor has written the first major account of verbal
classification constructions in Australian languages. Many languages of
Northern Australia have a common construction whereby an uninflecting
preverb is paired with an inflecting verb, yielding a complex predicate.
The preverb carries most of the lexical information about the predicate,
while the inflecting verb acts as a host for tense and agreement
morphology. McGregor gives a detailed discussion and analysis of these
constructions in the languages of the Western Kimberley (particularly in
the languages Gooniyandi and Nyulnyul) and relates them to other languages
in Australia and beyond. He considers how the construction could have
arisen and touches briefly on other related verbal phenomena. The book also
contains a summary of basic information about Australian languages and a
guide to orthography.
The potential audience for this book is quite broad. Anyone interested in
typology and verbal classification, light verb constructions, etc, would
benefit from McG's arguments and examples. The presentation presumes no
specialist knowledge of Australian languages and is very friendly towards
those with no experience in these languages.
SUMMARY OF CHAPTERS
Chapter 1 is a summary of the program and an introduction to McGregor's
ideas of verb classification. McG. distinguishes between the notions of
superclassification and subclassification. In nominal superclassification a
noun is assigned to one of a small number of classes; gender systems are a
good example of this type of classification. Nominal subclassification
systems involve dividing the noun into several types. McG's example is
'trains', which can be subclassified in English as 'steam train', 'diesel
train', 'passenger train', and so on.
Verbal classification systems also show both types. Verbal
superclassification involves assigning the event to one of a number of
defined and fairly discrete categories.
McG, following Dixon 1982 and others, distinguishes a cline of
classification systems between 'class system' and 'category' systems,
according to parameters of boundedness of the marker, number of categories,
exhaustivity of the system, disjoint classes vs overlap, and obligatoriness
of use. McGregor uses the analogy of types of library classification
systems to illustrate the differences.
Chapter 2 presents the Gooniyandi verb classifier system in detail.
Gooniyandi is a non-Pama-Nyungan language spoken in the North-West semi
arid region of Western Australia. In Gooniyandi all verbs must appear with
a classifier. The categories are largely disjoint there are not many verbs
which can appear with more than one classifier. Comparison is made with
Gooniyandi's only relation, Bunuba.
In Chapter 3 McGregor surveys compound verb constructions (CVCs) in several
different languages. CVCs comprise a usually uninflecting preverb and an
inflecting verb root. The preverb and the inflecting verb form distinct
word classes. The inflecting can almost always be used on its own, without
a preverb. The following examples from Bardi illustrate a CVC and an
inflecting verb used alone.
1. inflecting root used alone:
nganamboogal 'I hit him.'
root: -boo- 'hit'
2. examples of CVCs
garr nganamboogal 'I rubbed him.'
Preverb: garr 'rub'
Inflecting verb: -boo- 'hit'
entire predicate means: 'to rub (something)'.
roowil innyagal 'He was walking.'
Preverb: roowil 'walk'
Inflecting verb: -nya- 'pick up, catch'
Entire predicate: 'to walk'
A large portion of chapter 3 is concerned with the Nyulnyulan languages and
their CVCs. McG examines the frequency of various inflecting verbs, their
'collocational potentials' (the most common semantic denominator that
events which use the same inflecting verb denote).
Chapter 4 also deals with cross-linguistic comparison, in this case with
the types of category system themselves. McG surveys the degree of
grammaticalisation of the constructions and the verbs used as inflecting
verbs. Half this chapter is a detailed comparison of the systems of
Gooniyandi and Nyulnyul, which represent in some ways opposite ends of the
grammaticalisation continuum. Chapter 5 draws comparison between
conjugation systems in Australian languages. Chapter 7 is also typological:
it discusses grammatical phenomena related to CVC constructions, such as
In chapter 6 McG summarises previous treatments of verb superclassifying
constructions. Previous analyses include semantic bleaching (a 'light verb'
analysis), a classifier analysis, and treatments as fusions (in LFG or
construction grammar) or complex predicates. More on this below in the
Chapter 8 discusses a possible origin of the CVC constructions in
Australian languages. McG argues that ideophones are the most likely source
of such constructions. He further raises a suggestion that Pama-Nyungan
verb conjugation classes could have arisen from CVCs.
The discourse uses of verb classification are discussed in chapter 9, which
is mostly a case study of their use in one Gooniyandi text. Chapter 10
presents conclusions and directions for further research.
I have several criticisms of this book (mostly about chapter 6 and the
parts that concern Bardi), but I do not want the following criticisms to
cloud what I see as a major achievement for Australian studies and an
impressive foundation for future work, as well as an important contribution
I turn first to worries about data, particularly in Bardi. Most of McG's
data come from Nicolas (1998), a small, very badly transcribed text
collection, or from Metcalfe's published work, which is of considerably
better quality but is also rather sparse. It seems odd to do textual
research on Nicolas' corpus when Aklif's texts are readily available, are
much better transcribed and translated and represent a much bigger corpus
with more sophisticated language. The examples in the Bardi dictionary are
also a wealth of information there for the taking (Aklif 1999); Aklif 1999
also contains a grammar summary considerably more clear than Metcalfe's
early transformational analysis, which would have been better quoted than
McG's reworking of Metcalfe (1975), where several tense categories have
been collapsed which should be treated distinctly.
Bardi is frequently taken to support conclusions made for Nyulnyul, however
Bardi and Nyulnyul also appear to behave quite differently. The differences
are perhaps quite small when comparing Nyulnyulan to Bunuban, which
represent extremes of the scale. However, there are some very interesting
differences within Nyulnyulan, including whether preverbs can appear
without inflecting verbs (they can in some languages but not others), which
inflecting verbs the preverbs take (there is considerable variation among
cognate preverbs even between Bardi and Nyulnyul, which are almost mutually
intelligible), how many inflecting verbs there are in the language
(Nyulnyul has more than double Yawuru's number), whether preverbs
themselves can be inflected for aspect or other categories (they can in
Yawuru, but not in Bardi), the sources of preverbs and their productivity
with respect to reduplication and whether they can also be used in other
word classes (varies a lot from language to language).
Placement of Bardi on the classification scheme there is considerably more
overlap in Bardi categories than there seems to be in Nyulnyul, and this
raises problems for the classifier analysis (I haven't done counts of what
proportion of the preverbs can take more than one inflecting verb, since
the number goes up every time I gloss a text). The categories are not
nearly as discrete as McGregor claims for Nyulnyul many preverbs appear
with more than a single inflecting verb. The largest collocation I have
recorded is 5:
3. dirray 'turn'
+ -banji- 'turn around'
+ -ar- 'rotate something'
+ -boo- 'turn into something'
+ -jiidi- 'swing about'
+ -0- 'swing about, turn back' (historically < *-w-, now a
phonologically null root.)
A good deal of the book argues for a classifier analysis of various verb
constructions, and I agree with this for all the languages I have any
expertise in. There are several parts of the book (especially pp 261,
266ff), however, where McGregor argues for a classifier analysis to the
exclusion of a complex predicate analysis. That is, McGregor argues that
the constructions in Nyulnyul and Bardi are not complex predicates or
compounds containing bleached inflecting verbs. The argument is much easier
to sustain for Gooniyandi, where all portions of the inflecting verb must
appear with a preverb, and the 'preverb' can be reasonably clearly
identified as the head of the verbal predicate (it assigns theta-roles, etc).
McGregor's arguments against a complex predicate analysis are as follows:
a. head tests fail; on these tests either one or neither of the preverb
and inflecting verb are identified as the head, but never both (cf Alsina,
Bresnan and Sells' 1997:1 criterion that complex predicates are multiheaded);
b. if we take the criterion of inflectional locus, the inflecting verb is
the head; however the uninflecting preverb usually governs theta-role
c. if we follow Mohanan (1997) and say that complex predicates contain two
predicative units which jointly determine clause structure, we again run
into problems with Nyulnyulan transitivity alternations, since the
transitivity of the inflecting verb remains constant even though the
transitivity of the clause as a whole can vary (p 263 ff).
While I accept some of McGregor's reservations about tests for headedness,
I do not really follow argument (a). In Bardi, for example, one can point
to examples of pairs which differ minimally, for example with regard to
theta-role assignment or transitivity, where the locus of shift is the verb
morphology, the preverb or the inflecting verb. Some examples are given
below. These examples show pairs of sentences where a difference in aspect
is caused by changing one part of the predicate either the verb
morphology, the light verb used in the CVC, or the uninflecting preverb.
4. Clausal Aspect can be determined by:
(a). verb morphology
Aaman roowil ngannyan, gala inngoorroobinngay iilanim.
as soon as walk 1sg-'catch'-cont, right then 3sg-chase-1sg dog-erg
'As soon as I go for a walk, the dog chases me.'
Moonboorran roowil innyij.
towards speaker walk 3sg-catch-perf
'He is coming towards me.'
(aspect markers: -n 'continuative' vs -ij 'completed action (in the last
(b). the light verb
'He ran away (quickly).'
'He took off with speed.'
(in this CVC, -(i)nya- is unmarked for any specific aspect, whereas the use
of jarrala- is inceptive, and refers to the act of 'starting to run')
(c). the preverb
Bany inamana boorroo.
shoot 3sg-'put'-past kangaroo
'He shot the kangaroo.'
'He kept on shooting (it).'
(reduplication of the preverb shows that the action was iterative)
For further examples see Bowern (2002). Surely the clearest analysis for
data like these is that both the preverb and the inflecting verb are
contributing information to the meaning of the predicate, information which
is normally associated with a head. Therefore I agree with Mohanan's
characterisation of this type of construction, that both parts determine
I don't agree with McG that this causes problems in analysing the
transitivity of verb constructions (point (c) above). Again, my data are
from Bardi rather than Nyulnyul but I think this analysis could be applied
to Nyulnyul mutatis mutandis. My argument is as follows.
Firstly, I draw a distinction between clausal transitivity and predicate
valency (following Margetts 1999 and others). I need this for Bardi to
account for various case marking and argument structure mismatches which
are not related to the question of CVCs and I will not go into detail here.
In Bardi the situation regarding transitivity alternations is as follows:
5. preverb + monovalent inflecting verb > intransitive predicate
preverb + bi/trivalent infl. verb > intransitive or transitive predicate
Thus CVCs using inflecting verbs such as jiidi- 'go' and -ni- 'sit' always
produce monovalent complex predicates, but when a preverb is combined with
a transitive inflecting verb the resulting complex predicate may be
transitive or intransitive, depending often on the argument structure of
the preverb. For example, roowil (i)nya- 'walk' is intransitive and takes a
single NP in absolutive, even though (i)nya- 'catch' (the inflecting verb
used with this preverb) takes transitive verb morphology and when used
without a preverb takes an ergative subject.
Now, I argue here that the morphology of the inflecting verb is transitive
because it is forced by the conjugation class of the verb. These usually
bivalent verbs take transitive prefixes, because that is the conjugation
that these verbs belong to, even though technically not all argument
positions are filled in CVC constructions. The transitive set of prefixes
have to be used for the verb to be morphologically well-formed. The
argument is similar to arguments for impersonal verbs in languages like
Latin; ningit 'it snows' has 3rd person agreement not because some real
third person argument is triggering agreement, but because all verbs aren't
morphologically well-formed without a tense/subject agreement suffix.
Finally, it is not clear to me that the two arguments (complex
predicate/fusion versus verbal classifier) are mutually exclusive. That is,
surely an inflecting verb can be both part of a complex predicate
construction and fulfil the role of a classifier in that construction?
There is an interesting problem of where Nyulnyul and Bardi fall between a
classifier and category system. McG seems to argue that Nyulnyul and Bardi
have dedicated classifiers; however it is difficult to see how the
classifiers are dedicated, since a) the 'classifier' inflecting verbs can
appear without a preverb, and therefore without a lexical event to
classify, and b) there are numerous simple inflecting verbs in Bardi and
Nyulnyul which do not take part in the classification system.
I am also not entirely convinced that all instances of preverb + inflecting
verb are classifier constructions, or if they are, that they are the same
type of construction. There is a good case to be made, for example, for
both syntactically formed and lexically formed preverb + inflecting verb
constructions in Bardi. They behave differently with regard to
reduplication, the independence of the constituents and the flexibility of
the relative ordering of the components (see further Bowern forthcoming)).
I also have several comments regarding the origin of preverbs in
ideophones. I certainly agree that ideophones are a possible source of
preverbs, and McG provides good arguments that this is the case. I am
somewhat sceptical, however, about applying the idea to (Proto-)Nyulnyulan.
Firstly, many non-ideophonic preverbs can be reconstructed to
Proto-Nyulnyulan with a fair degree of certainty. Secondly, McG argues that
the phonotactics of preverbs cannot be explained by regular sound rules but
there is no demonstration of this. There are frequent synchronic
alternations in Bardi between adjectives and slightly syncopated preverb
forms (eg joorrongg (velar nasal + g) 'straight' vs joorroong ma- (velar
nasal only) 'choose'. This would indicate either that some phonological
reduction has taken place with the cliticisation of the preverb to the
inflecting verb, or that there is extra morphological material on the free
form. There are several possible origins of the Nyulnyulan CVCs in
Pre-Proto-Nyulnyulan, and I am not convinced that alternative scenarios
(such as the grammaticalisation of a loose event classifier construction,
the extension of metaphor, or other paths) were ruled out.
Thus in summary, 'Verb classification in Australian languages' is a very
readable and interesting book, and there is much in it that is
controversial and open to dispute. I hope very much that William McGregor's
book will provide an impetus for further research on this topic in
Australian languages and elsewhere.
Aklif, Gedda (comp) (1999). Ardiyooloon Bardi Ngaanka: One Arm Point Bardi
dictionary. Halls Creek: Kimberley Language Resource Centre.
Alsina, Alex, Joan Bresnan and Peter Sells. Complex Predicates. Stanford,
California: CSLI Publications.
Bowern, Claire (2002). 'How Light are North Australian light verbs?' Paper
presented at the Dudley House/Harvard Linguistics Light Verb Workshop,
November 11th, 2002, to be published in Harvard Working Papers in
Linguistics, Vol 9.
Bowern, Claire (forthcoming). A Bardi comparative grammar. PhD
Dissertation, Harvard University.
Margetts, Anna (1999). Valence and transitivity in Saliba, an Oceanic
language of Papua New Guinea. PhD thesis, Catholic University of Nijmegen.
Metcalfe, C. D. (1975). Bardi verb morphology. Canberra; Pacific Linguistics.
Mohanan, Tara (1997). Multidimensionality of representation: NV complex
predicates in Hindi. in Alsina, Alex, Joan Bresnan and Peter Sells. Complex
Predicates. Stanford, California: CSLI Publications. pp 431-471.
Nicolas, Edith (1998). Etude du système verbal du bardi, langue du
nord-ouest australien, avec une présentation contrastive du système bunuba.
PhD, Université Paris VII: Denis Diderot, Paris.
| ABOUT THE REVIEWER:
ABOUT THE REVIEWER Claire Bowern is the world's other Nyulnyulanist. She is currently writing her dissertation on a historical grammar of Nyulnyulan languages, focusing on Bardi, language contact and historical dialectology.