Universal Properties of Natural Language and Georgian as a case study

Most theories of labeling address the question as to why the merger of two syntactic objects should receive some label and then formulate some labelling algorithm; often this amounts to one of the two syntactic objects, the head, to project (some of) its features. In this talk, I reverse this question and wonder why it is not the case that both daughters project their features and I formulate an alternative approach to labelling, based on this question, that integrates minimalist structure building with insights from categorial grammar, much inspired by Neeleman & Van der Koot’s configurational matrix. The proposal will provide a unified labeling mechanism for complements, specifiers and adjunction, incl. coordination structures, and will have repercussions for the nature of roots, c-selection, clauses, and abstract case.

Georgian agreement presents two analytic challenges: competition and discontinuity. In launching Distributed Morphology, Halle and Marantz 1993 presented a purely morphological account of these. Though fit for their purposes (increased descriptive adequacy over predecessor work), the account was not particularly constrained. Competition mechanisms could equally easily yield the opposite outcome and discontinuities could be reversed. Béjar 2003 turned to syntax (specifically, where the search for argumental person was headquartered) to cull some of this excess. Two problems remained, however. Number was not wholly amenable to similar treatment: sometimes it behaved like a dependent of person, sometimes not. And, just as on Halle and Marantz’ approach, discontinuities—in particular, the person-number precedence—remained reversible.

I propose to start at the opposite horn of the Georgian dilemma but to head in the same direction. I will provide a syntax-driven account of discontinuous agreement (a move spearheaded by Trommer 2002) and then derive the results of agreement competition with minimal but nonzero stipulation. The account is an application of my previous work on discontinuous agreement in Afroasiatic, Ngumbin, and Sepik (Harbour 2023, building on Harbour 2008, 2016).

The driving claims are (i) that number dominates person in the syntax; (ii) that such dominance maps to postcedence during exponence, [number [person]] maps to person-…-number; and (iii) that only string edges are available to linearisation (leaving aside morphological special ops, like metathesis). If the phi-structure is linearised after the verb, a string of suffixes results, …-person-number. If it precedes the verb, it yields a discontinuity, person-…-number. Bottom-to-top exponence (Bobaljik 2000) allows object agreement to affix nearer to the verb than subject agreement. This resolves competition save for two hangovers. Without further constraint, my account allows for multiple prefixes in one case and multiple suffixes in quite a few. To avoid multiple suffixes, I appeal to properties of Georgian morphophonology: although Georgian is famous for consonant clusters, these are positionally restricted. To avoid the one case of multiple prefixes, I posit that first person has a zero allomorph. This, again, has motivation: other person and person-like prefixes also have zero allomorphs. Neatly, the factor that drives prefixal placement of phi-structures yields precisely the conditioning environment needed for zero realisation of first person. So, the innovation is minimal.

Time permitting, I will discuss the apparent turf squabble between my account and Béjar’s. Although both vie to explain Georgian, I doubt that they are in competition everywhere. In particular, Migrelian, a western relative of Georgian, lacks what I regard as the key diagnostic for a learner to opt for my mechanisms. At the same time, it presents a phenomenon that looks much more probe-like and is missing from Georgian. In other words, both accounts have explanatory mileage and further work is needed to characterise the proper application of each.

Cross-linguistically, many languages allow arguments to be unexpressed either in whole or under certain conditions, a phenomenon commonly referred to as ‘pro-drop’ in the literature. In this talk, I examine instances of third-person agreement gaps and discrepancies that pose challenges to the agreement-based pro-drop account of Georgian. Drawing on Neeleman and Szendrői’s (2007) theory for the radical pro-drop languages, I argue that pronominal features play a crucial role in reconstruction of unexpressed arguments in environments where agreement fails. Specifically, I propose that the agglutinative morphology for case and number in third-person pronominals comes into play in certain syntactic contexts in Georgian, namely: (a) in cases involving null object agreement; (b) when subject agreement is marked by fusional suffixes; and (c) when zero-marked indirect object agreement occurs, implicating the presence of applicative markers. The core claim advanced here is that the agglutination whether on pronominal or verbal constituents is a key mechanism enabling the omission of arguments.

This talk presents a case-study of alignment in Enggano, an Austronesian language spoken in Sumatra, Indonesia. Enggano has three major constructions in which verbs can occur: bu- form, bare form and ki- form (see Kähler 1940, Hemmings forthcoming). There are also two sets of person markers that combine with bu- verbs and bare verbs respectively. Since these agree with the person/number features of S/A we can argue that main clauses are generally accusatively aligned. However, there is one clause-type in which an ergative alignment pattern is found: subordinate clauses headed by the be ‘because’ and a= ‘if, when’. In these clauses, transitive verbs occur in a bare verb construction together with a person marker that agrees with A. However, intransitive verbs take an allomorph of bu- and no person marker for S. Since A is therefore marked in the agreement and S/O are not we can call this ergative alignment and state that Enggano has a system of split-ergative alignment according to clause-type. The obvious question, then, is how such a system arises. In this talk, I will argue on the basis of joint work with Erik Zobel (Zobel & Hemmings 2024), that the ergative alignment type is the more conservative pattern since it is attested in both main and subordinate clauses in languages like Nias (Brown 2001) and subordinate clauses tend to be more conservative than main clauses cross-linguistically (see Bybee 2002). I will also outline possible historical developments that lead to the pattern of split-alignment attested in Enggano.

Brown, Lea. 2001. A grammar of Nias Selatan. PhD dissertation. University of Sydney.

Bybee, Joan. 2002. ‘Main clauses are innovative, subordinate clauses are conservative: consequences for the nature of constructions’. In J. Bybee, and M. Noonan (eds) Complex sentences in grammar and discourse: essays in honor of Sandra A. Thompson, 1-18. Amsterdam / Philadelphia: John Benjamins.

Kähler, Hans. 1940. Grammatischer Abriß des Enggano. Zeitschrift für Eingeborensprachen 30, 81–117, 182–210, 296–320.

Hemmings, Charlotte. forthcoming. Verbal Morphosyntax in Enggano. In I Wayan Arka, Mary Dalrymple and Charlotte Hemmings (eds.) Enggano: Historical and Contemporary Perspectives. ANU Press.

Zobel, Erik & Charlotte Hemmings. 2024. Morphological conservatism in Enggano subordinate clauses. In Iker Salaberri, Annemarie Verkerk & Anne Wolfsgruber (eds.) Comparative approaches towards the diachronic behavior of subordinate clauses. Special Issue of the Italian Journal of Linguistics.

One of the most crucial Natural Language Processing (NLP) tasks is associated with the universality-driven development of language resources for different languages (e.g., Universal Dependencies (UD), UniMorph, PARSEME, etc.). And as a result, this research outlines the creation of a Syntactic TreeBank for the Georgian language within the UD approach to language modelling and the development of the UDPipe model. The talk describes the importance of syntactic annotation and the need to compile a comprehensive treebank for Georgian. It discusses the tools used for annotating sentences and converting them to the UD format, as well as potential training opportunities. Methodologies for adapting Georgian linguistic resources to UD are explored, with a focus on tokenizers and morphological analyzers, while noting the lack of syntactic analyzers. The research, also, outlines the outcomes, including the development of the UDPipe model for Georgian. Version 1.3.1-dev of UDPipe was trained on a dataset of 3164 sentences and 56239 tokens, with results provided for various components of the pipeline.

(presenting joint work with Caroline Féry, University of Frankfurt)

In the absence of information structural biases, prosodic structure reflects constituent structure, such that phonological phrases (φ) are mapped onto syntactic phrases (XP), intonational phrases (ι) onto clauses (CP) (Selkirk 2011). This predicts that major φ-phrase boundaries separate subjects from VP’s in either SVO or SOV languages; see (1a-b).

For some V-final languages, V’s are reported to be aligned with the right-edge of a higher prosodic constituent, when followed by postverbal material (see Turkish in Özge & Bozsahin 2010). This phenomenon is not surprising if the postverbal material is right-dislocated, in which case the V determines the right edge of a clausal domain, aligned with the right edge of an ι-phrase as in (2).

Georgian prosody displays similar phenomena in prosodic phrasing, precisely a high boundary aligned with the right edge of non-final verbs. However, this pattern is puzzling in a language such as Georgian in which postverbal orders can arise through verb fronting (hence, not necessarily resulting from right dislocation). This presentation will outline the relevant prosodic facts and compare two alternative accounts: an account based on the syntax of postverbal material in Georgian and an account based on the role of prefinal rising contours.

References

Özge, Umut & Cem Bozsahin. 2010. Intonation in the grammar of Turkish. Lingua 120. 132–175.

Selkirk, Elisabeth. 2011. The syntax-phonology interface. In John Goldsmith, Jasson Riggle & Alan C. L. Yu (eds.), The handbook of phonological theory, Second Edition, 435–484. Malden, MA: Wiley. DOI: 10.1002/9781444343069.ch14