Australian Linguistic Society
The *Dative Split in Martuthunira
Alan Dench Centre for Linguistics, UWA, Perth email@example.com
Martuthunira has a consistent nominative/accusative case-marking pattern -- the result of a relatively recent innovation from the predominantly (split-) ergative Pama-Nyungan Australian type. The case frame for transitive verbs is an old, marked nominative-dative frame which has extended to become the unmarked pattern (see Dench 1982). Thus, the modern accusative suffix is a reflex of an old dative, confidently reconstructed as *ku (O'Grady 1966, Hale 1970).
In the modern accusative languages of the Pilbara, the suffix has a wide range of functions. The accusative marks the objects of (di)transitive verbs, and accusative marked beneficiaries can appear with almost any verb. In addition, accusative arguments can appear in a range of non-verbal clauses. In many related languages, these additional functions are assumed by the dative (see Blake 1987, Dixon 1980).
In most of the non-accusative languages of the Pilbara, reflexes of the *ku dative also double as the genitive. However with Martuthunira as the single exception the accusative languages do not show evidence of the suffix in this function. Martuthunira alone retains reflexes of *ku as both a genitive and as an accusative -- but the set of forms has split to become two separate morphemes. The allomorphs of the two suffixes overlap only on (the relatively less frequent) consonant-final stems, as shown in Table 1.
L is any lateral, N is any nasal, rr is the apical tap/trill.
How are we to account for this split? We should first consider the possibility that the split is the result of regular phonological changes which have affected the suffix forms in different ways. In order to support such an account it would be necessary to show that the phonological environment of the *ku suffix in its relational (dative or accusative) function was distinct from the environment of its adnominal (genitive) function. However, it is clear that we cannot make such a case on the basis of the modern distributions of the suffixes; for example, the accusative form of a nominal presents exactly the same phonological environment as the genitive form in a nominative NP.
Whatever the solution, it must involve a generalisation of one or other (or both) suffixes beyond an original conditioning environment. The question can be approached as follows. First, the general phonological changes effecting lenition and loss in Martuthunira can be considered. Second, the history of other Martuthunira morphemes with a similar phonological shape can be investigated in an internal reconstruction. Third, the Martuthunira reflexes of *ku can be compared with those in other languages, with a view to a comparative reconstruction of the history of this particular morpheme. Finally, we can consider the morpho-syntactic context in which the Martuthunira reflexes are to be found and relate these to the earlier findings.
Patterns of Lenition and Loss affecting *k
Martuthunira shares a history of intervocalic consonant loss and lenition with a number of other Pilbara languages. O'Grady (1966) describes the patterns of change in Yindjibarndi, in particular, and these are revisited in the light of more data from Martuthunira and Kurrama, in Dench (1987). Austin (1981) describes similar patterns in the history of Tharrkari and Purduna. The changes affecting the bilabial and velar stops are summarised in the Table 2.
The Martuthunira changes are quite clear: *k is lost between like vowels and is lenited to w between dissimilar vowels. Changes in the other leniting languages are slightly different -- in particular, the two peripheral stops pattern together, and the lenition/loss is blocked if the preceding syllable contains w.
Morphophonemic alternations involving *k
There are a number of Martuthunira morphemes, in addition to *ku, which show evidence of the lenition of *k between vowels. These fall into two groups depending on whether the morpheme-initial *k is realised consistently as w (as with the genitive), or has been lost (as with the accusative).
The forms to consider here are the 'Belonging', 'Side' and 'Allative' suffixes. For the other morphemes, the data is incomplete or, in the case of the two verbalisers, is limited and speculative. Clearly, neither of the two patterns of allomorphy is entirely consistent with the sound changes presented in the preceding section. Given that the loss or lenition of *k is more generally dependent on the identity or dissimilarity of the flanking vowels, we might have expected allomorphs exhibiting lenition and loss for each morpheme, as shown in Table 4.
Instead, it appears that the pattern found on a-stems has been generalised to all stems. That this should have occurred is not particularly surprising given that a-stems are much more common than those involving either of the other two vowels. In a list of 857 nominals with a final vowel, 422 (49%) have a final /a/, 222 (26%) a final /u/, and 213 (25%) a final /i/. In text, the proportions are higher. Table 5 lists the number of tokens of particular allomorphs on the different stems for the complete Martuthunira corpus. Figures for the 1sgOBL pronoun, nganaju, are given separately as particular case forms of this item are disproportionately represented in the elicited data.
A number of observations can be made on the basis of these figures. First, we can assume that the relative proportions of stems reflected in the figures for the accusative are representative for nominal stems in general. Second, although there are few tokens, the 'Belonging', 'Side' and 'Allative' suffixes also most often occur on a-stems. Given these frequencies, it is easy to imagine that the patterns found on a-stems might serve as the basis for a levelling of the allomorphy.
However, while this line of argument might be taken to explain the patterns of allomorphy for 'Belonging', 'Side' and 'Allative' suffixes, and can also explain the consistent wu genitive on vowel-final stems, it does not account for the forms of the accusative. On the contrary, the accusative forms now stand out as exceptions to what might be seen as a tendency to generalise the a-stem allomorphs. The problem then is to explain the exceptional forms of the accusative.
Reflexes of *ku across the Pilbara languages
We can approach the question from a different direction by considering the range of allomorphs of the original *ku suffix across a selection of Pilbara languages (see Table 6).
With just Thalanyji as an exception, all of the languages show some intervocalic lenition of the *ku suffix, even where the language is otherwise phonologically quite conservative. There are essentially two different patterns of allomorphy revealed in Table 6. Purduna and Yingkarta show the simplest alternation: *ku is lenited to wu on vowel final stems. Warriyangka and Tharrkari show further elaboration of this pattern: on i-stems, the suffix-initial glide is y rather than w, while on u-stems the suffix is realised as lengthening of the stem-final vowel. In Jiwarli, this latter pattern extends to i-stems.
The second pattern is slightly more complicated. In Nyamal, Nyiyaparli, Panyjima and Yinhawangka (all quite conservative languages), the ku form is retained on stems of three or more morae but is lenited to yu on dimoraic stems. In Ngarluma, the dimoraic allomorph assimilates to the final vowel of the stem. Yindjibarndi and Kurrama show a further lenition of the trimoraic allomorph to wu, realised as lengthening of a stem-final u. This last change is consistent with general phonological changes in the two languages.
Since Martuthunira has a phonological history similar to that of Yindjibarndi and Kurrama, it might be expected that its reflexes of *ku would show similar degrees of lenition. However, while the accusative forms fulfil this expectation, the genitive is more conservative than might be expected from the comparative evidence. Thus, in contrast to the findings of the last section, it now appears that it is the Martuthunira genitive forms which need to be explained as exceptions. The dilemma is this: the genitive forms are consistent with regular phonological changes, (allowing for some analogical adjustment), while the accusative forms are consistent with general patterns of lenition for this particular suffix across the Pilbara languages. I suggest that the answer to the problem lies in first accounting for why the *ku morpheme should show patterns of lenition which extend beyond the more general phonological changes, and then in explaining why the Martuthunira forms in genitive function are apparently protected from the fullest extent of these changes.
I suspect that the lenition of *ku is related to its usual occurrence in a weakly stressed environment. By the typical stress patterns of Pilbara languages, words bear a primary stress on the initial syllable and may, on longer forms, bear a secondary stress on either the penultimate or antepenultimate syllable. The metrical patterns of a phonological word depend on its morphological structure: disyllabic morphemes initiate a stress domain, monosyllabic morphemes typically do not. Thus, the monosyllabic suffix *ku does not initiate a stress domain, but may itself carry secondary stress if it is followed by a single unstressed monosyllabic morpheme.
Where the *ku suffix functions as a dative or accusative relational suffix, it typically occurs in final position and thus will not bear stress. The Panyjima examples in Table 7 illustrate.
These patterns suggest that the yu allomorph of the Panyjima accusative (and by extension the other languages which share this pattern) may be an artefact of what is a maximally unstressed environment -- the last syllable of a word bearing only an initial stress.
It must be pointed out, however, that while the dative/accusative is the last inflectional morpheme in a word, it may be followed by post-inflectional clitics. There are thus some contexts in which the suffix will bear a secondary stress. In Martuthunira, for example, 15% of accusative suffixes in naturally occurring text host a post-inflectional clitic, in Jiwarli 13% of dative suffixes (in relational function) host a clitic. Nevertheless, the generally 'weak' position of the suffix may account for its greater tendency towards lenition.
Towards an explanation
If the relational *ku, as a final inflection, occurs in a phonologically weak position we may be tempted to argue that the adnominal (genitive) *ku occurs in a stronger position by virtue of the fact that it may be followed by some relational case suffix and is thus less subject to lenition. However, the data does not clearly support such a position. First, Southern Pilbara languages show similar lenition of *ku in both dative and genitive functions and there is no immediately obvious reason why Martuthunira might not have followed the same path.
Second, text counts do not strongly support a contention that the genitive is generally followed by other inflections. In fact, there are relatively few examples of the genitive functions of *ku in the Martuthunira and Jiwarli texts (as opposed to elicited data), and so it is difficult to make comparisons. If anything, the figures (Table 8) suggest opposing tendencies.
The figures suggest that while the genitive is often followed by a nominal suffix in Martuthunira, this is rare in Jiwarli. But a number of additional observations must be made in relation to these patterns. First, the relative lack of inflected *ku genitives in Jiwarli is perhaps in part due to the fact that the language has a separate pronominal and demonstrative dative/genitive suffix, -mpa. In Martuthunira, the *ku genitive occurs in these contexts also. Second, Jiwarli permits a sequence of dative suffixes (adnominal 'genitive' followed by relational dative) only where the allomorphs are distinct (Austin 1995), otherwise just the single suffix appears. One such example occurs in the Jiwarli texts (Austin 1997, p121, line 22):
(1) Kuwarti-thu kumpa-inha kurlkanyu-rri-ngu now-top sit-pres thinking-inchoat-imperfSS walypala-wu(*-wu)-rru kujinu-wu. white man-dat(-dat)-now medicine-dat. 'Nowadays (people) think about white man's medicine.'
While Martuthunira shares a general ban on sequences of identical forms, it freely permits a sequence of original *ku suffixes, as in (2). The modern genitive and accusative suffixes are sufficiently distinct that there is no possibility of a sequence of identical forms occurring.
(2) Ngayu kampa-lalha mimi-wu-u thanuwa-a. 1sgNOM cook-PAST uncle-GEN-ACC food-ACC 'I cooked uncle's food.'
Given the shift in alignment type, the relational functions of the *ku suffix in Martuthunira are much more extensive than they are in non-accusative languages. At the same time, the lack of a distinct pronominal dative/genitive suffix means that the opportunity for GEN +*DAT sequences is further magnified.
The genitive/accusative split circumvents the ban on sequences of identical suffixes. 'Weaker' post-vocalic allomorphs have generalised as the accusative, the stronger have generalised as the genitive. If the regular phonological rules had been allowed to apply, sequences of identical suffixes would have occurred and important grammatical distinctions would have been lost. If the Martuthunira genitive were to have followed the pattern of lenition shown by the accusative allomorphs, the resulting overlong vowel would have shortened leading to the collapse of the GEN+ACC/ACC contrast. The maintenance of different forms thus avoids the loss of the nominal genitive.
The history of the Martuthunira *dative thus appears to present a case of 'preventative analogy', or 'morphologically conditioned sound change' (see Anttila 1972:99: Hoch, 1988:46), but one which through preserving a contrast has led to the development of separate morphemes.
 The accusative languages include Jurruru, Kurrama, Martuthunira, Ngarluma, Panyjima, Yindjibarndi and Yinhawangka.
 There are other restricting conditions, but these are not important to the argument here. Further, it should be noted that there are exceptions to these patterns in Martuthunira (Dench 1995:48).
 Austin (1994) suggests that for the i-stem allomorph, the vowel of the suffix has assimilated to the glide (and/or to the stem-final vowel). He represents the Jiwarli i-stem allomorph as -yi, though the glide is consistently lost by phonetic rule (Peter Austin, pers. comm. April, 16, 1998).
 Jiwarli figures, based on a total of 210 tokens, are calculated from the texts published in Austin (1997).
 For example, in Jiwarli the genitive is realised as lengthening of a stem-final high vowel in forms such as [thártuupàrnti] (thartu-wu-parnti, 'dish-GEN-ABL') (Peter Austin, pers. comm. April, 16, 1998).