Ahmad Jabbar

Using the Hindi-Urdu discourse particle 'lo' as a case study, the present paper models mirativity within the Table model framework (Farkas & Roelofsen, 2017). This comes with an enrichment of the Table model; we propose the discourse structure to encode, in addition to discourse commitments, the time these commitments are publicized. We also incorporate a component that contains the public record of private beliefs. With these two additions, we seek to capture mirativity in the Table model, emphasizing recency as a condition for surprise and underlining the need and the conceptual difficulty of incorporating private beliefs in a public scoreboard of conversation. We show that the difficulty can be overcome to fulfill the modeling need.

Compositional generalization benchmarks seek to assess whether learning agents can successfully combine familiar concepts in novel ways. COGS (Kim & Linzen 2020, COGS, EMNLP) provides a suite of such tasks in the area of interpretive semantics (mapping sentences to logical forms). A noteworthy finding for COGS is that model performance varies widely across tasks. In this paper, we argue that these performance differences reflect deep properties of these tasks. We focus on two COGS tasks: an easy task (models are generally successful) and a hard task (no present-day models get any traction). Using both experiments and conceptual analysis, we argue that the easy task requires only a single distributional generalization that is well-supported by the training data, whereas the hard task involves a learning target that is ambiguous or even contradicted by the training data. We additionally argue that pretraining can disambiguate the hard task without compromising the goal of testing compositional generalization. Overall, our findings offer practical guidance to designers of compositional generalization benchmarks and also yield new insights into the nature of compositionality itself.

This paper presents a study into the Hindi-Urdu 'na' as a sentence-final particle. Although also used as a topic marker and negation, 'na' occurs sentence-finally across clause-types. In light of the data, we think the following hypothesis offers the best fit: 'na' signals the speaker’s belief that the content of na’s containing clause is a reasonable inference, given what’s common ground. Notably, in addition to other clause-types, we explore na's distribution in exclamations and exclamatives. We link our work to recent research on the polar question particle 'kya' in Hindi-Urdu.

Recent scoreboard models of conversation, in addition to modeling update effects of assertions and questions, also make sense of how one may react to such discourse moves. An account of acceptance and rejection is captured by Farkas & Bruce (2010), while Bledin & Rawlins (2020) have recently made sense of how one may resist an assertion too. For rejection and resistance of assertions, truth comes out to be a crucial notion. Between X and Y, if Y doesn't believe p to be true, then Y might resist or reject X's assertion of p. For questions however, truth cannot serve this purpose. A complete model of discourse would seek to offer explanation of resisting questions. Much like Bledin & Rawlins, one can non-exhaustively enumerate a class of resistance moves. In many ways, one can resist a request to make a particular question QUD. In this paper, we raise following question: which notion can serve to explain question uptake in discourse? In our paper, we show how the expected utility value (EUV) of questions can serve this role for questions. EUV is used by van Rooij (2003) to give an account of interpretation of questions; we think it can serve our intended purposes too. To demonstrate plausibility, we build a workable formalism.

We propose a framework that makes space for both non-indexical contextualism and assessment-sensitivity. Such pluralism is motivated by considering possible variance in judgments about retraction. We conclude that the proposed pluralism, instead of problematizing, vindicates defining truth of a proposition w.r.t. a context of utterance and a context of assessment. To implement this formally, we formalize initialization of parameters by contexts. Then, a given parameter, depending on a speaker's judgment, can get initialized by either the context of utterance or the context of assessment. An upshot of the proposal is that one need not exclusively espouse one of non-indexical contextualism or assessment-sensitivity about a particular class of expressions, for instance predicates of personal taste, in language; one can be pluralist in this sense. (This manuscript is part of an ongoing research project).

We consider a puzzle in the question semantics literature. The puzzle concerns data when 'know' embeds interrogative complements. For the exhaustive strength in the literature known as intermediately exhaustive, first person ascriptions don't seem to exist, but third person do. By arguing against the only solution in the literature, we suggest that the puzzle is more interesting than previously thought. We provide a compositional semantics for 'know' where the interpretation of 'know' is relativized to an information state. The proposed semantics, we think, is compatible with relativism and nonindexical contextualism, but we opt for relativism for its relative ease in capturing disagreement data we provide.