Research Topics

~.~.~.~. UnDeR cOnStRuCtIoN ~.~.~.~.

Affect x Language

Perspective Sensitivity

Collaboration with Cal Howland at Rutgers Philosophy/Center for Cognitive Science.

Semantics and pragmatics of Wh-Questions

Dissertation

My dissertation research on the semantics and pragmatics of \emph{wh}-questions takes a step in this direction, by using advances in the psycholinguistics and pragmatics to reanalyze (non-)exhaustivity in questions. I argue that a pragmatic explanation, which takes in to account hearer expectations, is better able to capture the contribution of both contextual and linguistic information in judgements of question meaning. Questions like \emph{Where can I get coffee?} may be answered exhaustively by providing a complete list of places to get coffee in the relevant domain (a Mention-All answer), or non-exhaustively by providing a single place, perhaps the most popular or accessible one (a Mention-Some answer). However, it has long been observed that not all questions allow both equally. My dissertation investigates the contextual, lexical and structural factors licensing these answers to questions. I pursue the hypothesis that (non-)exhaustivity is semantically underspecified, thus is resolved via the interaction between the question’s form (which features certain surface-level cues selected by the speaker posing the question), and the context-dependent goals of the questioner (knowledge of which allows the listener to craft a response that resolves the question). Thus, questions can give rise to the range of interpretations based on both linguistic and contextual elements. Dominant accounts argue that question-meaning is exhaustive (Mention-All); however some questions, like \emph{Where can we find coffee?} and its embedded clausal counterpart \emph{Dana knows where we can find coffee}, naturally allow for a Mention-Some interpretation. Semantic solutions have either reduced non-exhaustivity semantically to modal questions (\emph{can}), or have dismissed it as an exceptional ``pragmatic'' phenomenon. I argue on empirical and conceptual grounds that (non-)exhaustivity falls out of joint inference about the speaker's meaning and goal given the linguistic signal, and discourse situation, rather than an exhaustive underlying semantics. Given that questions underspecify a linguistic meaning, the hearer must recruit information from multiple sources to resolve meaning. While the linguistic signal can provide a defeasible cue to the speaker's goal, hearer expectations about intended meaning drive the resolution of (non-)exhaustivity. In one set of experiments, I show empirically that non-exhaustive readings are neither exceptional, nor restricted to modal questions. In another series of experiments, I show that contextual goals operationalized as `high' and `low' stakes determine the level of (non-)exhaustivity that is judged acceptable in the situation. Further, when a discourse context is highly informative with respect to a speaker goal, hearers rely less on the linguistic signal to resolve question meaning than on the context. Further, even when contexts are less explicit about a discourse goal, hearers still look to context by imputing additional information not explitictly provided, but deriving from their world knowledge and expectations about the likely goals behind the linguistic utterance. By identifying how interlocutors effortlessly coalesce cues from the surface form of the question (for example, a modal) with cues from the context and inferences about the questioner's goals, I engage with foundational inquiries into a question's semantic representation and the role of cognition. In this work, I utilize experimental methods to assess semantic theories by testing the predictions they make regarding possible interpretations of questions across linguistic and non-linguistic contexts.

Pragmatics of Discourse Makers

Prosody and Speech Acts

Collaboration with Anouk Dieuleveut,

Induction of underlying forms in phonological learning

Computational models can test the cognitive viability of our linguistic theories. In this line of research, I address two foundational issues in computational learnability: given a finite set of observed data (1) how does the learner infer the target grammar from a large space of possible grammars, and (2) how does the learner generalize beyond the observed data when necessary? I explore an output-driven typological system that models the interaction of stress and vowel length. The logical structure imposed by the condition of output-drivenness mitigates the computational intractability of a large search space, addressing the first question. The output-driven learner exploits this structure successfully to deduce necessary facts about the target grammar from the observed data. However, certain patterns of data prove barriers to learning, namely those giving rise to a novel variation on the subset problem. I show that where deductive reasoning fails, inductive reasoning picks up the slack: if the learner can induce generalizations about ungrammatical forms through ranking induction, then the learner can successfully overcome the subset problems encountered. Details of this work are outlined in Moyer (2016) and Moyer \& Tesar (in prep). Though this work is in computational learnability of phonology, it carries implications beyond phonological learning. Inductive inferences are rampant in language, and indeed are the basis of many pragmatic inferences that we make. Showing that such induction over indirect evidence is viable at early stages in phonological learning suggests its possibility in learning other aspects of grammar. Finally, this Optimality Theoretic model provides tools which could be successfully co-opted for a better theory of pragmatic reasoning more generally.

Acquisition of indexical pronouns: Other minds in discourse

My undergrad thesis, with Kate Harrigan, Valentine Hacquard and Jeff Lidz. From early on, my interests were compelled by the relationship between language and theory of mind in the acquisition of context-sensitive meaning. Indexical pronouns present an interesting case of context sensitivity because only across utterances is their discourse-rich meaning revealed. If a learner is insensitive to context, the acquired meaning would be highly impoverished. In virtue of calling attention to shifting conversational perspectives, learning indexicals requires close attention to discourse and a speaker's referential intentions. Indeed, it is unclear how exactly children would acquire pronouns were they unable to recruit their intention and goal tracking abilities. In collaboration, I developed a novel experimental design to test children's comprehension of indexical pronouns in naturalistic discourse, in which they showcased their ability to do this (Moyer, Harrigan, Hacquard \& Lidz, 2014). Though I have not actively pursued this research line, there are many further directions in which it could be extended. For example, the experimental design lends itself quite naturally to testing the pragmatics of pro-drop in languages such as Spanish or Portuguese, where explicit pronouns trigger additional inferences about the intended referent.