Research

Definiteness and anaphora

This cross-linguistic collaborative project with Yağmur Sağ, Jian Cui, and Kathryn Davidson investigates how discourse structure shapes the use of referential expressions, like definite and demonstrative descriptions, and seeks to identify the pragmatic constraints that regulate their acceptability. A central thread of this work concerns the behavior of demonstratives in anaphoric contexts. A central thread of this work concerns the behavior of demonstratives in anaphoric contexts. Contrary to prior views in the literature that anaphora does not contribute any novel insights to our theory of definites and demonstratives, this work illustrates that anaphora serves as a strikingly revealing diagnostic that brings the pragmatic differences between these referential expressions to the fore.  By integrating cross-linguistic experimental data from typologically diverse languages with and without overt definiteness marking (English, Bangla, Turkish, German, Mandarin),  we present evidence for robustly replicable cross-linguistic distinctions between anaphoric definite and demonstrative descriptions—shaped by factors such as situational change and focus-sensitive alternatives. The consistent experimental findings span typologically diverse languages, suggesting a deep conceptual universality underlying the definite-demonstrative contrast, and pave the way for a nuanced formally explicit model of context-sensitive definite reference that capture the role of both grammatical and pragmatic discourse constraints in the use of these expressions.

Bare nouns, kind reference and classifiers

This research project centers on classifier languages within the Indo-Aryan language family, particularly Bangla, and provides a more nuanced typology of kind-oriented languages. While most Indo-Aryan languages do not possess classifier systems, a small subset—Bangla, Odia, and Assamese—forms a striking areal group that does. My research examines the nominal systems of these languages, with special attention to the behavior of bare nouns and the role of classifiers. I investigate how kind reference is encoded in these systems, the availability of definite and indefinite interpretations for bare nouns, and how such interpretations interact with the and how such readings relate to the presence or absence of lexical exponents of (in)definiteness.  Although bare nouns in these languages exhibit several key properties of kind-oriented languages, their full distributional profile does not align straightforwardly with canonical kind-referring languages like Mandarin, nor with property-oriented systems like English or Hindi. Focusing on Bangla, I argue that bare nouns in these contexts are best analyzed as singular kind terms—an approach that accounts for both their restricted distribution and their interpretive flexibility. On a closely connected note, I also examine ra, an animacy-restricted classifier in Bangla, and show how its behavior supports the singular-kind treatment of bare nouns. I argue that ra serves as a dedicated, lexicalized type-shifter that maps singular kinds to their corresponding property interpretations, thereby accounting for the extensive use of singular kind reference in the language. An extended version this work is currently in progress, which re-examines the role of other classifiers in Bangla (ta and gulo) in light of the new analysis and synthesizes additional empirical evidence from my fieldwork on Odia and Assamese in this domain. Through this work, the goal is to enrich the empirical foundation on which formal linguistic models are built and contribute to a more typologically informed understanding of kind reference by foregrounding South Asian linguistic variation, which remains underrepresented in semantic theory.

Pragmatic competence of artificial language systems

Recent advances in language models have prompted renewed inquiry into what it means to understand language, and the extent to which pragmatic competence can arise through exposure to statistical patterns of language use. While several benchmarks exist for evaluating LMs’ pragmatic behaviors for phenomena such as conversational implicature or high-level discourse coherence, it has been relatively under-explored whether language models track fine-grained informational-structural constraints in a way similar to humans. The current work, in collaboration with Jennifer Hu, focuses on pragmatically modulated preference between different definite expressions—specifically, the contrast between definite and demonstrative descriptions in anaphora—as a case study to assess the pragmatic competence of LMs. Understanding LMs’ behavior in this domain can also inform ongoing research on how such preferences might operate in humans.