Tree Adjoining Grammar at the Interfaces

• Main Author: Longenbaugh, Nicholas Steven
• Language: en_US
• Published: Harvard University 2014
• Online Access: http://nrs.harvard.edu/urn-3:HUL.InstRepos:12717344

This thesis constitutes an exploration of the applications of tree adjoining grammar (TAG) to natural language syntax. Perhaps more than any of its major competitors such as HPSG and LFG, however, TAG has never strayed too far from the guiding principles of generative syntax. Indeed, following the pioneering work of Frank (2004), TAG has been successfully incorporated into Chomsky’s (1995) Minimalist Program (MP). In large part, however, Frank (2004) leaves unexplored the issue of how TAG applies at the PF and LF interfaces. Given the fundamental importance of interfaces within the MP, no minimalist syntactic theory is complete without at least some notion of the means by which syntactic structure relates to pronunciation and interpretation. In this thesis we attempt to provide insight on this very issue: we address how TAG interfaces with the articulatory and interpretive components of the language faculty, and what insights it provides to minimalist conceptions of these interfaces. Ultimately, our aim is both to reaffirm the viability of TAG as a minimalist syntactic theory as well as to demonstrate that TAG makes clear otherwise arcane facts in natural language syntax. The central proposal of this thesis is twofold. First, TAG may be naturally extended to interface with the articulatory and interpretive components of the language faculty by making recourse to synchronous TAG (STAG). Second, once such a framework has been adopted, minimalist ideas regarding the interaction between syntax and linear order can be applied to deal with certain problematic examples in the TAG framework. TAG thus offers confirmation that in at least some cases, certain aspects of linear order are dependent on post-syntactic operations, so that syntax does not always wholly determine linear order. As a corollary of our proposal, we also demonstrate, through a case study in Niuean raising, that the TAG system makes clear predictions on phenomena that are difficult to describe in mainstream minimalist theories. Our argumentation for these proposals proceeds in three major stages. First, we formalize the synchronous TAG system that has to date been applied in a mostly piecemeal way by various researchers (see Shieber & Nesson 2006, Frank & Storoshenko 2012 for some examples). As a part of this formalization, we argue that the derivation of the LF object, but not the PF object, should make recourse to a more expressive version of the TAG system: multicomponent TAG, a variant that relaxes some constraints on the primitive units in the TAG system to yield greater expressive power. Second, we argue that the STAG system lends credence to the view that at least some word order is determined post-syntactically. In the past, researchers have presented ad hoc extensions of the expressive power of TAG to handle various difficult examples such as subject-to-subject raising in English questions and Irish and Welsh main clauses. We demonstrate that these extensions are both theoretically suspect and ultimately unnecessary given minimalist notions of the derivation: for many of the data motivating these extensions, there is independent evidence that their derivation in fact relies on post-syntactic rearrangements of certain verbal heads. Such examples are therefore well within the generative capacity of a framework with a TAG-based syntactic component that allows certain specific and well motivated post-syntactic rearrangements. Third, we demonstrate that not only is our particular system well motivated within the theoretical bounds of the MP, but also that it makes surprising and accurate empirical predications in cases that have otherwise defied analysis. Specifically, the Austronesian language Niuean features a peculiar instance of raising that has defied a satisfactory analysis since its discovery by Seiter (1980, 1983). We show that TAG makes the clear prediction that there is no raising in Niuean, then argue that this prediction is borne out under a careful examination of the facts. Given that the framework was developed almost exclusively based on the Indo-European language family, its ability to capture confounding behavior in a typologically dissimilar Austronesian language is a strong confirmation of its status as a reasonable alternative to mainstream minimalist syntactic theories.