Bowern_RWP

8/12/2019 Bowern_RWP

1/17

Dening complexity:Historical reconstruction and Nyulnyulan subordination

Claire Bowern, Yale University

Abstract

I use data from subordination strategies in Nyulnyulan languages (Non-Pama-Nyungan,Northern Australia) in order to investigate various alternative means of dening and quanti-fying complexity. While Edmonds (1999) denes 48 distinct types of complexity (concentratingon social and natural sciences), in this paper I concentrate on three facets of complexity: de-scriptive complexity, ontological complexity, and parsimony in reconstruction. While historicallinguists tend to maximise parsimony, in Nyulnyulan languages the minimization of one aspectof complexity necessarily adds complication elsewhere, and it therefore serves as an appropri-ate case study of the interdependencies between ontology, syntactic modelling, and language

ange.

Keywords : complexity, subordination, Nyulnyulan, dia rony, language ange

1 IntroductionWe nd the notion of complexity in various forms throughout the history of resear on Australian

languages. We nd it more o en in the early history of language documentation in its converse of simplicity, although even today in Australia we nd language a itudes whi simultaneously treatindigenous languages as too simple to survive in the modern world, yet too complex for outsidersto study. Even in 1980, Dixon (1980:1.2) felt the need to disabuse potential readers of the simplic-ity of Australian languages. Su claims perhaps persist in the widespread notion that Australianlanguages do not exhibit embedded clauses (for a review of these ideas see, for example, Nordlinger2006b).

I begin with this point because it shows the importance of considering simplicity and complexityin a wider context. Both complexity and simplicity are relative terms, of course; what one resear ermay regard as simple will be treated by another as complex depending on their level of experience anddegree of familiarity with the concept. Linguistic complexity is also theory-dependent; for example,serial verb constructions are complex (and problematic) structures in a syntactic theory that has astrong version of the lexicality hypothesis (see, for example Di Sciullo and Williams 1987), howeverin a theory where complexity is dened in terms of degree of embedding, they are less complex thansubordinate clauses.

In this paper, I use historical reconstruction of subordination strategies in Nyulnyulan languagesin order to explore theoretical issues in the denition and use of complexity in language ange. I

Lexicality is a good area to illustrate arguments of complexity, since it represents one area where we can clearly seethe trade-offs in different areas (maximizing complexity of the lexicon and parsimony in syntax, versus a less restrictedtheory of syntax).

1


2/17

begin with a discussion of more explicit denitions of complexity, especially as they relate to histor-ical reconstruction. A er all, we cannot evaluate an idea su as complexity without teasing apartthe many different ways in whi a time like complexity could apply to the data. I then give threecase studies of subordination strategies in Bardi and the other Nyulnyulan languages. I argue thatgrammaticalization theory itself relies on notions of complexity in other areas of the eld of linguis-tics and that we cannot consider complexity in grammar without also being explicit about what our

theories lead us to consider as a complex answer to a question.2 Dening complexity in (historical) linguistics

e term complexity itself is ambiguous between at least three senses. O is a measure of the inherent nature of the item under study. Ontological complexity, assuming thatall aspects of a system are knowable, is static. at is, a measure of ontological complexity does not

ange according to the way an item is described. is contrasts with , whirefers to the self-complexity of the models whi were made to represent reality (Csnyi 1989:15).A type of semiotic complexity is Gell-Manns (1995) effective complexity (see also Dahl 2004:25ff),whi is a measure of the complexity of the internal structure of an item. Complexity is a relativeterm, not an absolute one (as discussed in the introduction), so we should note that items can onlybe dened as complex with respect to other items. Finally, we should distinguish local complexityfrom global complexity. A structure may be locally simple but globally complex, as is for examplea single nonbran ing node within a tree, or it may be locally complex but globally simple (a singleterminal node is more simple than a bran ing node).

is is all relevant for the denition of complexity within historical linguistics and languageange. If we are trying to trace the evolution of a structure within a family, and trying to make

claims about its complexity, we need to be explicit about whi type of complexity we are talkingabout, and under whi scenario a given event is more simple or more complex.

erefore, if we are to evaluate a possible increase in complexity over time, su an evaluationneeds to take place along several different parameters, including the following:

(1) a measurement of the constructions effective complexity;

a measurement of relative complexity with respect to the reconstruction;

an evaluation of the role of the particular model used in dening the complexity of thestructure.

is paper is also in part a comment on Givn (2001, 2008) and the feeling that there is more to besaid about increasing complexity than hypotaxis originates in parataxis (e.g. Givn 2001:218219),especially when considering the available coordination and subordination strategies in a languageas a whole. Let us now consider some data.

3 Nyulnyulan subordination3.1 Bardi and the Nyulnyulan languages: ba ground to coordination/subordination

Bardi is a non-Pama-Nyungan, Nyulnyulan language spoken now by about 8 people on the North-ern tip of the Dampier Peninsula. ere is extensive unpublished eld data on Bardi dating ba

Edmonds (1999) found 48 distinct denitions of the term in the natural and social sciences, and even within linguistics,there appears some overlap in the terms used. Here, however, I concentrate on 3 senses in whi the term complex maybe used, depending on what it contrasts with.For example, in a model with no recursion, a structure of the form 123123123123 is equally complex as one of the form132321213123.

Rice Working Papers in Linguistics 2 vol. 1, February 2009


3/17

to the Laves collection of the late 1920s; publications include Metcalfe (1975) and Bowern (2004).e Nyulnyulan family is quite close-knit, although there are differences in phonology, lexicon and

syntax whi make the languages mutually unintelligible. e Nyulnyulan languages are all non-congurational and (as far as I can tell from the sources available to me) make use of similar princi-ples of discourse organisation. ere are, however, differences in verb morphology and agreementmarking, and in the marking of embedded clauses.

3.2 e adjoined relative clause in Australiae term adjoined relative originated in Hales (1976) analysis of a certain construction in Warlpiri.

An example is given in (2):

(2) Ngajulu-rlu=rna 1 =1 .

yankirri emu

pantu-rnu,spear- ,

kuja=lpa =

ngapa water

nga-rnu.drink-

I speared the emu whi was/while it was drinking water. (Hale 1976:78)

Two features of this construction are relevant here. e rst is the phrase structure. e Warlpiriadjoined relative clauses are argued to have the following structure:

(3) S

S REL

S

at is, the adjoined relative is a full clause whi is embedded under an S node, but is an adjunctto the main clause. Adjoined relatives in Warlpiri must appear peripherally to the main clause.

e second feature is the semantic interpretation. Su clauses usually have a coreferential ar-gument, but they need not. ey are also ambiguous in interpretation between relative clauses andtemporal clauses. It seems that the marker of adjunction functions to designate a relation of coref-

erence or cotemporality between the clauses.Similar constructions are found in many parts of Australia, although should be noted that theyare not the only type of relativisation or subordination strategy, and there are languages with switreference marking (Austin 1980), with relative clauses marked by pronouns (e.g. in the Yolu bloc),and nominalisation. e special issue of the Australian Journal of Linguistics (26/1) contains infor-mation about a number of languages from several different families.

As Nordlinger (2006a) has pointed out, the analysis of Warlpiri in Hale (1976) has been presumedto apply to a wide variety (and the majority) of Australian languages. at is, it has been largelyassumed that Australian languages have a single multifunctional type of subordination, whi ismoreover syntactically marginal to the mainclause (Nordlinger 2006a:1); Nordlinger also highlightshow inuential this article has been in interpretations of Australian subordination, and how it hasled to a widespread perception that Australian languages do not exhibit syntactic embedding.3.3 Types of clausal dependency in Nyulnyulan languages

Nyulnyulan languages have clausal dependency structures whi supercially resemble the adjoinedrelatives described by Hale (1976), but they also have other strategies. Some of the structures areovertly marked by sentential clitics or verb morphology, others are unmarked. Some are nite,whereas in others the verb appears with gerund morphology (a circumx m(a)- -(a)n in all lan-guages). Bardi also has indirect questions. e constructions mark an array of functions, ranging

I have no data on the other languages in the family on this point.



4/17

from signaling causes and results, simultaneous a sequential actions, purpose, as well as relativisa-tion.

As a way of seeing the heterogeneity of multiple clause types in Nyulnyulan languages, considerthe following list of clausal dependency strategies in Bardi.

(4) a. the conjunction agal and and arragorror but, or (not further discussed here)

b. =b(a) , a Wa ernagel clitic whi primarily translates relative clauses; see 3.4;c. =min , =gid , =(j)amb , =(g)arra , =gorror if ; Wa ernagel clitics whi mark clausal de-pendencies, but not necessarily subordination (see 3.6);

d. Words whi introduce new clauses whi are dependent in discourse on a previousclause, including ginyinggon, ginyinggarra, ginyinggo (all roughly and then), and, inthe Laves corpus (1920s) only, ranana straightaway (discussed in 3.6);

e. Case markers, including the purposive -ngan for, in order to and the semblative -marr when (with nite or non-nite clauses; see 3.6)

f. Verb morphology; the simultaneous action marker -j while Xing (included in this listfor completeness but not further discussed here);

g. Apposition; null marking (for causes, reasons, simultaneous action, or sequential or con-secutive actions; see 3.7).

Nyulnyulan relative clauses are heterogenous. In Bardi, they are marked by =b(a) , a morphemewhi has no cognates in the rest of Nyulnyulan. In Warrwa, they are marked by a morpheme-jarr , whi is a verbal suffix that appears in the verb before the agreement markers (in Bardi itmarks topic aining; see below). In both cases there are problems in considering su clauses asembedded. However, they are not paratactic either; there is a dependency (for example, evidencefrom intonation and word order interleaving strongly indicates that they are a single clause).

In what follows, I consider four facets of dependent clauses in Bardi and Nyulnyulan. First,I discuss Bardi relative clauses (in 3.4) and review the evidence for embedding. In 3.5 I consider

adjoined relatives in other Nyulnyulan languages and the cognate construction in Bardi, whi is notan embedding construction at all. In 3.6 I review the use of case marking as a subordination strategyin Bardi and other Nyulnyulan languages. Finally, in 3.7 I explicitly address the issue of null-marked embedding and clause aining in Bardi, and its consequences for analyses of complexity.Before doing so, however, I review the evidence for embedding.

3.4 Bardi relative clauses

To illustrate the problem of dening hypotaxis in Bardi, let us consider relative clauses. Relativeclauses in Bardi, as mentioned above, are marked by the morpheme =b(a) . e absence of lenition of /b/ to /w/ or would suggest that this morpheme is a clitic rather than an affix, although the point isnot crucial here and this test is not entirely straightforward, as a few items whi are clearly cliticsalso undergo lenition. =b(a) is affixed to the rst word of the dependent clause whi usually (butnot exclusively) appears immediately following the relativised noun. If the word to whi =b(a) is

Because of the difficulties (discussed in 3.4 for describing a clause as strictly embedded, I here use the more neutralterm dependency to cover both true embedding and clauses whi are semantically dependent on one another.It probably appears fossilised in anggaba who (only Bardi has a distinction between who and what; other Nyul-nyulan languages have a reex of Proto-Nyulnyulan *yangki what in both meanings).McGregor (1994a:35ff) and elsewhere treats this type of clause as subordination, although he notes that parataxis isseldom discussed in descriptions of dependencies in Australian languages.Nordlinger (1998:162166) and Den and Evans (1988) provide discussion of case as complementisers in a variety of Australian languages.



5/17

a a ed ends in a consonant and the following word begins with a vowel, the clitic may be optionallyresyllabied as the onset of the initial syllable of the following word.

(5) Aamba man

[malarr- b wife-

i-na-m-bi-na=jin 3- hit.w.hand- =3 .

garrgoyi] completely

diird run.away

i-n-joo-noo.

3- -do/say- . e man who hit his wife ran away.

(6) Aamba man

[diirdi- b run.away

i-n-joo-noo 3- -do/say- .

barda] off

i-na-m-boo-noo 3- -hit- .

boolooman.bullo

e man who ran away killed a bullo . (AKL.F4)

ere are a couple of things to note about the sentences in (5)(6). e rst is that in all sucases, there is obligatory coreference between an argument in the main clause and an argument inthe subordinate clause. Most examples involve subject relativization, however examples of othergrammatical relations are also found, in both main clause and relativised clause. A few examplesare given below:

(7) Aarli sh

[i-na-marra-na= ba =jirr] 3- -cook- . = =3 .IO

joord=amba J=

n-inga.3-name

e sh whi he cooked for them is called joordoo .(8) Aamba

mannga-n-jalali-n-j 1- -stare.at- .

[yandilybara= b boat=

i-na-moogar-in-j.] 3- -make-

I was wat ing the man who/as he made the boat. (AKL.F4)

e example in (8) illustrates an important point about relative clauses in Australian languagesin general. ey almost never full the sole function of relative clauses. Rather, they are o en used

to translate simultaneous or subsequent actions, sometimes consecutive actions, and sometimes theyfunction more like swit reference markers (as is the case in Diyari (Austin 1981); see Nordlinger(2006b) for a survey of Australia more generally). is heterogeneity of function is part of Hales(1976) argument that su clauses are adjoined to the main clause rather than strictly dependent onthe noun. at is, the grammatically-marked relationship in su clauses is one of relations betweenevents, rather than a strict marking of particular participants (see, e.g. Hale 1976:79).

e sentences in (9)(10) would also appear to point to an adjoined analysis. In (9), for example,the antecedent of the relative clause boogoonb inin is ginyinggi ngaarri that devil; we might wantto analyze this as a case of swit reference, or translate more loosely along the lines of the ngaarri devil saw me, the one whi lives in the mangroves. However, note that ginyinggi ngaarri is notmarked for ergative case. If this were an instance of clause aining we would not expect the ergativeto be omi ed. However, we do regularly nd the ergative dropped from the antecedents of relativeclauses (see further Bowern (2007) for the relevant data).

Hale (1976) notes that in Warlpiri the NP-relative interpretation of su clauses applies when there is a coreferentialargument, and the T-relative interpretation when no arguments are shared between clauses. In Bardi, =b(a) is not usedif there are no shared arguments: there are other dependency markers used in su cases. However, relatives are stillambiguous between NP-relatives and T-relatives.Ergative-marked subjects may be the antecedents of relativeclauses, however su sentences are very rare in my corpusand are strongly dispreferred in elicitation.



6/17

(9) Ginyinggi 3

ngaarri devil

injalij=jarrngayoo 3 ( ) see- . -1 .DO

[boogoon=b inside-

inin].3 sit- .

e spirit, whi lives inside [the mangroves], saw me. (Metcalfe 1975:37)(10) Garrma

later jagoord return

anja 2. -do/say-

mayalgarran a ernoon

[booroo=b time-

anjalajan] 2. see- =1sg.IO

[nyoonoo=mbnganggan here-

boogoon].1- be- inside

When you come ba in the a ernoon, youll see me there inside.

Another problem with the adjoined relative clause analysis is that there some examples of sen-tences whi appear to have intertwined subordinate clauses. at is, constituents within theclauses are not clausebound. Consider (11) from the Laves corpus:

(11) Gooyarra 2

[arra irrmoonggoon] know

ingarr jimbina die

nyoonoo here

ingarramarnirr put

aambanim man-

malgin in secret

nyini here

irr.3

ey didnt know that two [men] had died and a man had been put there hidden. (Lavesn.d.:103/72)

e phrase gooyarra two is the subject of ingarrjimbina , but arra irrmoonggoon is the main clause.erefore either guyarra has raised out of the subordinate clause, or it is the object of irrmunggun ,

and the sentence should more literally be translated they didnt know the two, [that] they had died,although this is rather unlikely, as irrmoonggoon does not usually take a nominal complement. Itis not even a verb: it is a noun meaning something like knowledge. We have the same potentialproblem with (9) above; if the relative clause is part of the NP, either it has been extraposed or thephrase is discontinuous. Discontinuity is found in noun phrases in Nyulnyulan languages, but theconditions under whi it is used are not directly comparable to those in Warlpiri. It is mu morerestricted. (And note, incidentally, that Warlpiri discontinuities are clause-bound.)

In summary, Bardi relative clauses have a number of features of adjunction rather than embeddedsubordination. However, in either case, we have cases where constituents do not appear to be clausebound. Clausal embedding of this type is rare in Australia (although not unknown). e etymologyof this construction is unclear in Nyulnyulan languages. It may be tempting to assume one of Givns(2008)pathways, su as clause aining > embedding. However, we have no evidence for this withinthe language. e argument would be purely one from parsimony (that is, given su a strategy isclaimed for languages elsewhere, it is most parsimonious to assume the same dia ronic pathwayhere rather than multiplying entities).

3.5 jarr -marking

A different type of problem in Nyulnyulan subordination can be found in the analysis of wordswhi contain the morpheme -jarr- . It is found in both Eastern and Western Nyulnyulan languages.It is found in all of the eastern languages, where it is either a general subordinator (as in Yawuru),

Nordlinger (2006b:6) points out that while the majority of Australianists have interpreted Hales (1976) claim aboutadjunction as a claim that Warlpiri does not have syntactic imbedding, Hale consistently refers to su clauses as subor-dinate. Hale (1976:85, (22)) assumes a structure [S REL]_S.Many of Nordlingers (2006b) examples of Wambaya centre-embedding are single non-nite verbs. In this case, it isdifficult to tell whether su items are really embedded clauses, or whether they are really nominals. ( is is an issuefor further study, not a claim that Nordlinger is incorrect.)



7/17

a marker of relative clauses (as in Warrwa, where it functions somewhat like Bardi =b(a) ), or it hasadditional functions in Nyikina whi Stokes (1982:322ff) nds difficult to gloss (she uses the termdiffuseness). It appears to be absent from Nyulnyul and Jabirr-Jabirr. In Bardi, these forms are notused in subordination at all, but rather mark topic aining. Examples from the individual languagesfollow.

In Warrwa (McGregor 1994b:58ff), -jarri -yarri functions as a general marker of subordination,

and is glossed as introducing a temporal adverbial clause whi locates the situation referred to bythe main clause as subsequent to the situation referred to by the dependent clause (as in (12)). ismorpheme is also used in marking conditional clauses.

(12) ngambalany- jarri1 :awoke-

bij open

nganandiny 1 :got

ngajanu my

naarda.eyes

When I woke up I opened my eyes. (McGregor 1994b:58)

e marker is argued by Capell (1952:452) to be a relative pronoun, on the basis of examples suas (13).

(13) Warrwagarndirrin-nganaplatform-

waarratake

gurddie

ngindan jarri.3 / -do-

Take the man who died to the tree platform. (Capell 1952:452)

Further examples from McGregors eldnotes show that =jarri in Warrwa also functions as aclausal connector. All the examples I have found (of whi those in (14) are a sample) involve theconjunction of clauses whi have the same subject.

(14) Warrwaa. nyinggan

here

narndin

he:grabbed

- jarri

-

-yirr

-3 .

narndin

he:grabbed

-yirr/

-3 .

nanggana

he:lo ed

-yirr

-3 .

jimbin/

insideWhen he had grabbed them, he lo ed them up. (WM/FN: fm3;13)

b. yalkarn burp

ngandin I:did

kung drink

ngandin I:did

- jarri-

wila water

I burped from drinking water. (WM/FN: fm;9,166)c. mawu

happyngangariny I:got

liyan feel

nganjalin I:saw

- jarri-

I got happy when I saw him. (WM/FN: fm;9,171)d. ngarndany

I:went

-jarri

-

-yina

-3 .

jina

his

-ngana

-

buru

place

nganyjalany

I:saw

-jirr

-3 .

-wili

-du

wirrin

si ?

-mili

-dardarl si

-kurdany -

yuk camp

jina his

When I got there I found them si in bed. (WM/FN: fm;10,78)

ere is not enough data for Nimanburru to determine how subordinate clauses are formed in that language.Pace Givn (2008:2), there is a fourth method of reconstruction; that is, syntactic reconstruction using the comparativemethod; see Harris and Campbell (1995), for example.



8/17

In Nyikina (Warrwas closest relative), -jarri -yarri has these functions, however in addition itmay also mark multiplicity (all the examples given in Stokes (1982:322) involve the object argument),or circuitous movement. In su cases, the marker is not used in subordination. Examples follow.(15) shows a subordinate use of the morpheme, whereas (16) shows a multiple argument use.

(15) Yim-bula-ny-dyarri

3 -come- -

ng-la-ba-na.

1 - -see-If he had come, I would have seen him. (Stokes 1982:321)(16) Ngam-biga-ny- dyarr -irr

1 -have- - -3 .DOmanydja many

yila.dog

I used to have lots of dogs. (Stokes 1982:322)

In one dialect of Nyikina, -jarri is seldom found; instead, the morpheme is -ja . I do not know if both these morphemes have the same etymology.

In Yawuru, like in Warrwa, the morpheme is used as a subordinator, and this is its sole use inYawuru. ere are no constraints on subjecthood or coreference, although it seems to be the casethat there is a coreferential argument in most of the examples given in Hosokawas grammar.

(17) Wa-ng-ga-bula- dyarri ,3i - - -come-

nyamba this

wal-a--dyina 2 - -give-3 i

milimili.le er

When he comes, give this le er to him. (Hosokawa 1991:4.4.2;(82))(18) Yaga-rr-a-miri- dyarri

12- - -nish-nyanga-dyunu!,this-really

wa-ng-ga-rda-dyayrda 3- - -go-12

birndany-dyi stingray-

warli.meat( )

As soon as we nish all this, he will go and cat some stingray for us to eat.(Hosokawa 1991:10.6.2.1, (170))

Yawuru -dyarri marking is unusual in that there is a strong preference for the dependent clause

to precede the main clause. No other Nyulnyulan language is reported as having this restriction. eexamples given for Warrwa in this section, for example, demonstrate that no su order is requiredin that language.

In Bardi, the cognate morpheme is jarr- , and it a a es to the direct object and oblique agree-ment markers. Direct object and oblique spee participant agreement clitics have two forms. (19)illustrates this with a minimal pair using the verb to give:

I suspect in the light of examples from Bardi that the number marking in Nyikina might be a red herring, however Ido not have enough textual data for this language to look into it and context is not provided for the examples in Stokes(1982).It is possible that -ja is cognate with the Bardi simultaneous marker -j ; however in Bardi the two markers are clearlyunrelated functionally. If -jarri and -ja do not have the same source in Nyikina, we would have to assume that there has

been some morphological conation. ere is certainly no sound ange whi would derive one from the other in thislanguage.Hosokawa (1991:10.6.2) suggests that this is a borrowing from the neighbouring language Karajarri, where -nyarri isa continuous aspect marker. However, given the cognates as a subordinate marker throughout Nyulnyulan, the differentinitial consonant, the different placement of the morpheme in the verb, the different functions of the affix, and the factthatverbalmorphologyis noteasilyborrowed, I do notndtheassumption of borrowingvery plausible, despite Karajarriand Yawuru having a long history of contact.In the glossing for (18) 12- refers to a rst person inclusive plural agreement marker. is section closely follows Bowern (2008).



9/17

(19) a. Ana=ngay 2. -give =1 .DO

oola! water

Give me [some] water!b. Ana= jarrngay !

2. -give =1 .DOGive it to me!

As seen from examples su as (20), in Bardi jarr- forms have no relative function. ey do nothave to occur in a dependent clause, and they do not tra arguments or mark argument coreferenceor dependency in general (in fact, they only occur with rst and second persons).

(20) i- 3-

noo- -moondoo wet

-na -cont

-na -pst

-ng -

= jarrngayoo =1sg.DO

He kept on we ing me with it. (Metcalfe 1975:107)

Jarr-forms (as I will call the set) are transparently related to the unmarked set of object agreementmarkers. Aklif (1993) says that the jarr-forms are used a er stems ending in a consonant. Metcalfe(1975) argues that jarr-forms occur on stems containing an odd number of syllables. Neither of these distributions accounts for the data, as syntactic minimal pairs like (19a) and (19b) show. edistribution cannot be phonological.

ere are two very common frames where the jarr-forms occur. e rst place where jarr-formsoccur is where arguments are contrastive, su as in (21b) below. e second is where there is athird person subject and rst or second person object, and the spee act participant is featured inthe discourse over several clauses (that is, the object is a grammatical topic in the sense it is used inframeworks su as LFG: see, for example, Dalrymple (2001)). is is shown in example (21c).

(21) a. Mangir always

inkalan= jarrngay 3 visit-1 .DO,

iiganim si ness-

alig pain

ngandan.1 do/say-

Shes always visiting me when Im si .b. Niiwandi= jarrngay ,

tall-1 .DO joo 2

ngaada= jirri .short-2 .DO

Im tall, [but] youre short.c. Marbiddynim

M.-inanggalajarrngay 3 -visit=1 .DO

bardi,yesterday

gooyarr 2

aalga day

inggoodali= jarran3 lost=1 .IO.

arra darr come

oolarnajan.3 spear- =1.IO.

Marbiddy came to visit yesterday, for two days I didnt know where she was, she didnt

come to my place.e forms with =jarr- are cognate with verb forms marking relative clauses in the related lan-

guages Warrwa and Nyikina. It is not surprizing that a marker with the function of introducingrelative clauses, that is, one that establishes co-reference relations in syntax, should be co-opted totra and signal coreference across clauses. What is surprizing, however, is that the forms are onlyused for spee act participants, especially since relative marking is not restricted to spee act par-ticipants in Nyikina and Warrwa. Perhaps the jarr-forms also have functions whi are linked todiscourse-based obviation (for whi see, for example, Aissen 1997). Given the strong preference foruse of these forms when a participant lower on the person hierar y is acting on someone higher



10/17

up the hierar y, an obviation-based account is plausible. Some of the examples in Warrwa areambiguous between the type of sentence connective that Warrwa has and the Bardi-type exampleswith topic marking, and could be topic aining. I assume that su examples are the source of thereanalysis in Bardi.

In summary, there are several differences between Bardi-type jarr-marking and that found in theEastern languages. In the eastern languages, jarr-marking links clauses in a more or less denite

way. It works rather similarly to =b(a) marking in Bardi. In Warrwa it may link either particularparticipants or events, whereas in Nyikina there is an additional use in non-subordinate contexts. Innone of the eastern languages is jarr-marking limited to spee act participants, in fact almost all of the examples in the grammars involve third persons. In Bardi, however, jarr-marking is not used inany of these functions. Rather, it tra s spee act participants in grammatically marked discoursefunctions.

eoretically, there are several plausible pathways of ange whi would allow us to derivethese results. We could imagine a pathway of ange where a general subordinate clause markerbecame associated with linking participants between clauses [that is, as an adjoined relative clausemarker], then restricted to aining topics before being further restricted to use with spee actparticipants through the rise of obviation. However, we could also imagine the reverse scenario:that is, a marker whi tra ed obviation and spee act participants through discourse could begrammaticalised as a marker of relative clauses [whi further specify information about particularparticipants], and then extended to a more general function once the basis for obviation was lost.Topic aining in discourse through grammatical agreement marking is quite rare, and an obviation-type system whi is only marked on spee act participants seems to be unique to Bardi. ereforeany historical solution is likely to have few (if any) parallels in other languages.

If we assume universal pathways of discourse > syntax > morphology, that could give us ananswer (cf. Givn 2008). Givns (2008) hierar y is parsimonious, and historical linguistics has longmade use of Occams razor in historical reconstruction, whether through internal reconstruction orthrough the use of the comparative method. However, in this case we have no particular reason to

assume one solution is more parsimonious than the other. Moreover, we have no particular reasonto assume that language ange is itself parsimonious (see also analogous arguments for biologicalphylogenetic work by Sober 1991).

3.6 Nyulnyulan case marking and subordination

My next case study within Nyulnyulan takes up this question of discourse leading to syntax or viceversa. Nyulnyulan languages have structures whi look supercially similar to embedding in otherlanguages, in particular what are called XCOMP structures in LFG (Bresnan 2001). Su construc-tions make use of case marking. ey are the preferred method of forming subordinate clauses inNyulnyul. ey also found in Bardi, although theyre mu less common. In su constructions,there is a nite matrix clause. ere is a further clause, whi is either nite or non-nite (depend-

ing on the language) whi is marked for case. e marker appears either on the verb or on the rstconstituent of the clause.

(22) Bardi

Bijorr-o there-

i-n-alinygarna-n 3- -try-cont

[wirr- nganli -

m-arrmi-n].-rise-

Whi distribution applies in ea language is difficult to determine, since all the examples from Nyulnyul and Warrwahave the verb in initial position in the embedded clause. Either distribution may be possible there.



11/17

From there, he tried to rise up (into the sky). (AKL:eldnotes)

In this sentence, there is a matrix verb inalinygaman he tried, whi is nite, and another verbmarrmin , in a nonnite form (whi I have argued is a gerund). ere is argument coreference (thatis, the subject argument of the nite verb is the same as the notional subject of the nonnite verb).

ere is also overt marking of the dependency between the two clauses, in this case by the purposive

case marker -ngan .Similar constructions are found in all Nyulnyulan languages. e most common cases used arethe semblative, the proprietive, the ablative, and the locative. (23) gives examples from Nyulnyul(McGregor 1994a, 1996). However, in these languages, the verbs are usually nite. Nyulnyul doeshave gerund marking, but they tend not to be used in these constructions.

(23) -uk Locativea. imbulkubulkum

it.swelledindam- uk =ngay he.hit- -me

It swelled where he hit me.b. ingalk

she.criedmajikarr sunset

walk sun

injarrjarr- uk it.stood up-

She cried from sunset to sunrise. lit: (She cried at sunset, to the suns rizing.)

Case marking as a marker of nite subordinate clauses is also found in Yawuru. Here is anexample with the dative. Again, the verb is nite.

(24) Dyubagi tobacco( )

kayukayu+so +

nga-na-ngama 1- - (put( ))

bulkar-gun,ashes-

[wanydyi soon

nga-na-ga-lurra- yi ].1- - -burn- ( )

Ill mix the ewing tobacco leaves with ashes (lit. making tobacco so in ashes) so that Ican later enjoy the hot taste of it. (lit. so that I will burn [it]) (Hosokawa 1991:1067,ex176)

I have not recorded clauses of this typethat is, with case-marked nite verbsin Bardi, althoughit is not certain that they do not exist. However, given how common they are in other Nyulnyulanlanguages, their absence from my Bardi corpus is striking. Instead, Bardi uses either non-niteclauses or nite clauses introduced by a linker su as ginyinggo , ginyinggon , ginyinggarra thenor a Wa ernagel clitic. Etymologically, su items are case-marked third person singular pronouns.(25) and (26) are examples.

(25) Booroo look

nganjalagal=joogarra,1 see- . =2 .IO

boogoon= jamb inside=

goorrinkal.2 sit- .

[When] I looked around for you, I saw you inside. (or, I looked around for you, thats whyI saw you inside.)

(26) Birarr behind

ingirrinin 3 .DO

rawin go.as.group

ingarraman.3 .put-

Anyjimadan ba

booroongan=jirr.camp-all=3 . .

Ginyinggo en

oorany woman

joonk run

innyana 3 -cat -

arnbanjarr sing.out.in.fright

ingilirrmanijirr.3 -call.out-pst=3 .O

Case marking in these languages occurs once per phrase, as a suffix to the rst word of the phrase.



12/17

ey went behind, travelling as a group. ey went ba to camp. en a woman ran off andcalled out to them in fright. (L81.27)

Forms su as =jamb , =min and ginyinggo are unlikely to be markers of strict subordination,since they mark their clause as being related in some way to the discourse before it, but not specif-ically to the preceding clause. ey require a preceding narrative, but not necessarily coreferential

arguments.ese clause ainers have a number of forms, including apparently ablative and locative casemarked forms, as well as ginyinggarra ; (-)garra is a common temporal marker in the other Nyul-nyulan languages but it is not otherwise found in Bardi except in fossilised phrases. In the otherlanguages, -karra or -karr has a subordinating function.

e facts from Bardi lead us to a problem. On the one hand, we could assume that the Bardi struc-tures originate from a paratactic structure, as implied by universal pathways of grammaticalizationas outlined in Givn (2008). On the other hand, we have no evidence for this type of constructionanywhere else in Nyulnyulan languages. Indeed, Australian languages seem seldom to use demon-stratives as subordinate clause markers. (Yolngu is one exception that I know of; it is sporadicallyfound elsewhere too.) Moreover, ginyinggi in Bardi is not a straightforward anaphoric pronoun. It is

specically used for reactivating lapsed topics (Bowern 2008). Finally, -karra is not used in parataxisin any other Nyulnyulan language; in Nyulnyul it marks conditional clauses, while in the EasternNyulnyulan languages it has an aspectual use. erefore, we could either reconstruct a pathwaywhi is widely assumed elsewhere in the world, but would be very rarely a ested in these particularlanguages (and whi would also multiply the paths needed for reconstruction within the family,because we would have to assume multiple grammaticalization events within individual languages);or, we could assume that Bardi has fossilised this marker and turned it into a discourse ainer; inthis case however it would be an example of hypotaxis > discourse dependency, and not the otherway around, and therefore apparently a counterexample to Givn (2008).

3.7 Null marking

In addition to the markers discussed in 3.53.6, all the Nyulnyulan languages also make extensiveuse of juxtaposition/apposition to mark dependencies between clauses.

I have sometimes joked that under Greenbergs SVO word order typology, Bardis basic wordorder is not SVO, OVS or VSO, but V. In a text count of 171 clauses, 47% contained no argument NPsat all. It is common to go for long stret es of text with no overt markers. (27) is a short examplewhere there is no overt subject NP.

(27) Aarlingan sh-

arr go

nganjinj 1 do/say-

bardi.yesterday.

Langar bait

arrajana,1 . ,

arra

ngalinyan 1 cat

aarli.sh.

I went shing yesterday. I didnt have any bait, [so] I didnt cat any sh.

Arnbanjarr is a mysterious form; it looks like it is cognate with the subordinator jarri discussed above; however this isotherwise unknown in Bardi. e sentence is from a text from the 1920s.=jamb is perhaps the most syntax-like of these particles in that it appears to be able to precede or follow a clause thatit has some sort of relation to. However, it is unclear if this is a coercion effect of elicitation.An example is garra garra garra , whi is a type of elliptical for stuff that happens in a narrative. e.g. < X didsomething> , garra garra garra (X kept on doing it, e.g. they kept on walking),< then they did something else> .A morpheme -garra is also found in Ngumpin-Yapa languages (where it has clausal and aspectual functions) and maybe a borrowing into proto-Nyulnyulan, or a wider areal feature (p.c. Ken Hale, 1999).



13/17

In textual data one frequently nds series of clauses whi are clearly closely related but whishow no overt markers for conjunction or subordination. In (28), for example, there are three verbs.

e rst two, nganjarrga I ask (uninected for tense) and nganjoogaljirri I said to you are probablyappositive, i.e., I ask(ed), I said to you . e subordinate clause, if you would give me money,also has no overt marking of subordination and could be appositive.

(28) Nga-n-jangarrga 1 ask nga-n-joo-gal=jirri 1 say- . =2 . .DOgoolboo money nganyji a-n-a=ngay.2 give =1 .DOI was going to ask if you would give me money.

In ained examples of this type, the same subject is usually retained across clauses. In (29), forexample, there are no intonation breaks between the verbs and they form a single large prosodicunit. However, subject retention is not obligatory.

(29) Ginyinggon then

roowil walk

innyana 3 cat .

Ngarrigoonbooroo Ng.

baalingan shade-

darr come

inarnajirri

3 spear- . =3 .IO

niimana

many

aamba

men

agal

and

ambooriny

people

Ngoolbirndi.

Ng. en Ngarrigoonbooroo walked to her camp and came across many people at Ngoolbirndi.(Laves n.d.:129/19)

ese multiple verbs have many of the aracteristics of discourse serialization (see Pawley andLane 1998, for example). ey occur in a single intonation contour (although there are also exampleswith breaks, and examples where intonation units and syntactic units are not isomorphic). eyo en have the same tense/aspect/mood marking (at least in the prexal component of the TAMmarking), but I have not tested this systematically, and exceptions are found in the quoted data here.For example, (28) above would appear to show sequence of tense effects.

is construction is found in all Nyulnyulan languages. Some examples are given below for

Nyulnyul, another Western Nyulnyulan language.(30) Nyulnyul (McGregor 1996)

a. ingirriran=yirr,they.speared=them,

ingirrkan they.brought.it

wanyji ba

bur-ung camp-all

ey speared them and brought them ba to camp.b. nyimal

your.handkad bite

wanaw,you.give

layib good

wanyji,you.do

dumbar y

wanyji.you.do

Cut your wings so that you can y well. (cf. cut your wings, youll y well.)c. mangir

always

ngajarrijarrin

I.get.up

rangar-uk

early-loc

jan

my

malirr

wife

arri

not

ilajarrjarr

she.might.get.upI always get up early, but my wife doesnt.

d. kubimin government

inaw it.gave

bina this

wamb man

malirr wife

murrul li le

baab baby

birray mother

jin his

injimb 3 .died

e government gave this man and his wife a li le baby whose mother was dead.

In the textual counts mentioned above, approximately 10% of the clauses could not be clearly divided and so VSV andOVO orders were also included as is. is example was from a text transcribed in 1929 but was conrmed by current Bardi speakers. ere are many suexamples.



14/17

Yaruwu also has apparently paratactic dependency:

(31) Ngurru more

wal-a-lurra-dyaw,2 - -burn-12

marlu not

wa-ng-ga-miri 3- - -nish

dyungku.re( )

Put more wood on the re for us so that it will not go out. (Hosokawa 1991:1046,ex109)(32) I-ny-dyu-nd-dyanu

3- -say- -1 i[nga-ng-ga-rda 1i - - -go

karda-ngarn].yonder-

He told me to go there. (lit. he told me I will go there) (Hosokawa 1991:1061,ex161)(33) Darra+

bel +i-ny-dyu-nda,3- - (say)-

manydya many

i-na-rli-nda.3- -drink-

He burps as he drank a lot. (Hosokawa 1991:1081,ex227)

erefore, in addition to subordination with an overt marker, we also have what appears to beparataxis. However, it turns out to be rather difficult to show whether the structures are clause

aining, serialization, zero marked discourse dependencies, or subordination proper. In favour of the serialization analysis, at least for Bardi, is the fact that su clauses usually occur under a single

intonation contour. In some Nyulnyulan languages, there are sequence of tense effects, whi alsopoint to serialization of subordination. Moreover, in some cases the presence or absence of overtnominal material appears to be grammatically constrained. In the following Bardi sentence thenoun oorany is not omissible:

(34) Jaarla bea (- )

nganjalagal 1 -see-

*(oorany) woman

wiliwilon shing

inkalgal.3 -visit-

I saw a woman on the bea , she was shing.

However, the sentence without oorany is ne as true parataxis, with a pause between the clauses.

4 Summary and conclusionsWe can reconstruct several subordination strategies for these languages:

jarr-marking, probably as an adjoined relative structure, whi descends as:

adjoined relatives in Warrwa

topic- aining in Bardi

general subordination in Yawuru

case marking:

adjoined or embedded? depends on our view of argument structure in the languagesmore generally;

largely lost from Bardi; retained only in limited nonnite clauses, particular as purpo-sives;

retained in both nite and nonnite structures in other languages;

zero-marked clause aining: probably there all along, multifunctional construction; doesntturn into anything



15/17

In conclusion, let me return to a few points brought up early in this paper regarding complexityin explanation. roughout this paper, I have relied on the idea of parsimony in reconstruction.For example, I argue that Bardi is more likely to have desubordinated karr- marking than that theother Nyulnyulan languages have independently innovated a subordination strategy on the groundsthat a single loss event is more parsimonious than multiple gain events, even if the gain follows awell-known grammaticalization pathway. However, su a view minimizes global complexity at the

possible expense of local complexity. Moreover, as Lass (1997) and others have observed, there is noparticular reason why a language family should adhere to Occams razor (see also Sober 1991).

e case of Nyulnyulan subordination exhibits particularly clearly the problem that minimiz-ing complexity in one area of explanation merely increases it elsewhere. Generalizations su ashypotaxis comes from parataxis belie the ways that su structures arise. e complexity is moreinteresting. In this case, we see no overall trend towards greater complexity, and no overall move-ment towards syntaxis or hypotaxis from parataxis. Rather, as Dahl (2004) has pointed out in othercontexts, we see anges and shi s in form and function, and these anges are governed by discourseconsiderations as mu as emerging from it. In these languages, relative clauses are not an isolatedconstruction but are rather multifunctional, and they remain so over any period we can reconstruct.Hendery (2007) provides further examples of multiple pathways to relative clause formation. In sucases, we might wonder whether polyfunctionality compromises participation in macro-pathwayssu as discourse > syntax > lexis. is requires more investigation.

Overall, there seems to be no general rise in relative complexity over the reconstructible period of the Nyulnyulan family. While we note differences within individual languages, the morphology of subordination appears to be reconstructible. Bardi has undergone the most ange. It has largely lostcase-marked nite subordination, and it has lost the general marker of nominal relative clauses andadverbial temporal clauses. Instead, clauses with shared arguments are ambiguous, clauses withoutshared arguments adverbial, but have a different marker, and the inherited subordinator marks topic

aining in spee act participants. It is hard to tell whether this is strictly more complex or not.On the one hand, there are more morphological markers and more constructions, so from a strictly

effective point of view there has been a rise in complexity. On the other hand, the multiplicity of constructions results in less ambiguity in parsing, so from that point of view complexity is reduced.



16/17

ReferencesAissen, Judith (1997). On the syntax of obviation. Language 73(4): 705751.

Aklif, Gedda (1993). e Baardi language. Unnished PhD thesis, Australian National University, Canberra .

Austin, Peter (1980). Classication of southern Pilbara languages. In Peter Austin (ed.), Papers in Australian linguistics , Canberra: Pacic Linguistics, A-71, vol. 17, pp. 117.

Austin, Peter (1981). A grammar of Diyari, South Australia . Cambridge: Cambridge University Press.Bowern, Claire (2004). Diagnostic similarities and differences between Nyulnyulan and neighbouring lan-

guages. In Claire L. Bowern and Harold J. Ko (eds.), Australian languages: classication and the comparative method , Amsterdam: John Benjamins, Current Issues in Linguistic eory vol. 249,

ap. 11, pp. 295318.

Bowern, Claire (2007). Some uses of ergativity in North-Western Australia. In Proceedings of CLS 39 , 1,Chicago: Chicago Linguistic Society, pp. 622633.

Bowern, Claire (2008). Agreement and referentiality in Bardi discourse. In Bre Baker and Ilana Mushin (eds.),Discourse and grammar in Australian languages , Amsterdam: John Benjamins, pp. 5985.

Bresnan, Joan (2001). Lexical-Functional Syntax . Oxford: Bla well Publishers.

Capell, Arthur (1952). Notes on the Njigina and Warwa tribes, NW Australia. Mankind 4(9, 11): 351369,450469.

Csnyi, Vilmos (1989). Evolutionary systems and society: A general theory of life, mind, and cultures . Durham,North Carolina: Duke University Press.

Dahl, sten (2004). e growth and maintenance of linguistic complexity . Amsterdam: John Benjamins.

Dalrymple, Mary (2001). Lexical Functional Grammar , Syntax and Semantics vol. 34. New York: AcademicPress.

Den , Alan and Ni olas Evans (1988). Multiple case-marking in Australian languages. Australian Journal of Linguistics 8: 147.

Di Sciullo, A.M. and Edwin Williams (1987). On the denition of word . Cambridge, MA: MIT Press.

Dixon, Robert M. W. (1980). e languages of Australia . Cambridge: Cambridge University Press.

Edmonds, B. (1999). Syntactic measures of complexity . Ph.D. thesis, University of Man ester.

Gell-Mann, Murray (1995). e quark and the jaguar: Adventures in the simple and the complex . London:Li le Brown.

Givn, Tom (2001). Syntax . Philadelphia: John Benjamins.

Givn, Tom (2008). Towards a dia ronic typology of relative clauses. Presented at the 12th Biennial RiceUniversity Symposium on Language, Rice University, Houston, TX.

Hale, Kenneth (1976). e adjoined relative clause in Australia. In R.M.W. Dixon (ed.), Grammatical categories in Australian languages: Proceedings of the 1974 AIAS Conference , Canberra: AIAS, pp. 78105.

Harris, Alice C. and Lyle Campbell (1995). Historical syntax in cross-linguistic perspective . Cambridge Studiesin linguistics 74, Cambridge University Press.

Hendery, Ra el (2007). e dia ronic typologyof relative clauses . Ph.D. thesis, Australian National University,Canberra.

Hosokawa, Komei (1991). Meaning in Yawuru: A semantically oriented description of an indigenous language of the Kimberley region, Western Australia . Ph.D. thesis, Australian National University, Canberra, tobe published by Pacic Linguistics.

Lass, Roger (1997). Historical linguistics and language ange . Cambridge Studies in Linguistics 81, New York:Cambridge University Press.



17/17

Laves, Gerhardt (n.d.). Field notebooks, Bardi, Jawi, Karajari and other languages. Held at AIATSIS, Canberra.

McGregor, William (1994a). Complex sentence constructions in Nyulnyul, Western Australia. Functions of Language 1(1): 2566.

McGregor, William (1994b). Warrwa , Languages of the World Materials vol. 89. Muni : LINCOM Europa.

McGregor, William (1996). Nyulnyul . Muni : LINCOM Europa.

Metcalfe, Christopher-Douglas (1975).Bardi verb morphology (northwestern Australia) . Canberra: Pacic Lin-guistics B30: Australia National University.

Nordlinger, Ra el (1998). A grammar of Wambaya , vol. C-140. Resear S ool of Pacic and Asian Studies,ANU.

Nordlinger, Ra el (2006a). Introduction. Australian Journal of Linguistics 26(1): 13.

Nordlinger, Ra el (2006b). Spearing the emu drinking: Subordination and the adjoined relative clause inWambaya. Australian Journal of Linguistics 26(1): 529.

Pawley, Andrew and Jonathan Lane (1998). From event sequence to grammar: serial verb constructions inKalam. In Anna Siewierska and Jae Jung Song (eds.), Case, typology and grammar: In honor of Barry J. Blake , Philadelphia: John Benjamins, Typological studies in language vol. 38, pp. 201228.

Sober, Ellio (1991).Reconstructing the past: Parsimony, evolution, and inferences . Cambridge, MA:MIT Press.Stokes, Bronwyn (1982). A description of Nyigina, a language of the West Kimberley, Australia . Ph.D. thesis,

Australian National University, Canberra.

Bowern_RWP

Documents

Transcript of Bowern_RWP