We propose a language production model that uses dynamic discourse information to account for speakers' choices of referring expressions. Our model extends previous rational speech act models (Frank and Goodman, 2012) to more naturally distributed linguistic data, instead of assuming a controlled experimental setting. Simulations show a close match between speakers' utterances and model predictions, indicating that speakers' behavior can be modeled in a principled way by considering the probabilities of referents in the discourse and the information conveyed by each word.

This study investigates three-year-olds' representations of the attitude verbs think and know, in attempt to assess children's understanding of factivity. Know is factive and is therefore used in contexts where the complement is taken to be true. Think, although nonfactive, is still often used in situations where the complement is taken to be true. Are children able to recognize the difference between know and think and to understand that the truth of the complement is presupposed in one case but not the other? Acquisition studies on know and think find that children do not have an adult-like understanding of these verbs and their (non-)factivity before the age of four, but these tasks are often inappropriate for testing preschoolers' understanding of factivity for independent reasons. We designed an interactive game to implicitly evaluate children's knowledge of these verbs in a task that more directly targets factivity. Our results show that some three-year-olds are able to distinguish think and know, particularly in ways that suggest they understand that know presupposes the truth of its complement, and that think does not. The remaining threeyear-olds, however, seem to treat both as non-factive. This suggests that early representations of know may be non-factive, and raises the question of how children come to distinguish the verbs.

Verb prediction is important in human sentence processing and, practically, in simultaneous machine translation. In verb-final languages, speakers select the final verb before it is uttered, and listeners predict it before it is uttered. Simultaneous interpreters must do the same to translate in real-time. Motivated by the problem of SOV-SVO simultaneous machine translation, we provide a study of incremental verb prediction in verb-final languages. As a basis of comparison, we examine incremental verb prediction with human participants in a multiple choice setting using crowdsourcing to gain insight into incremental human performance in a constrained setting. We then examine a computational approach to incremental verb prediction using discriminative classification with shallow features. Both humans and machines predict verbs more accurately as more of a sentence becomes available, and case markers-when available-help humans and sometimes machines predict final verbs.

The salience of an entity in the discourse is correlated with the type of referring expression that speakers use to refer to that entity. Speakers tend to use pronouns to refer to salient entities, whereas they use lexical noun phrases to refer to less salient entities. We propose a novel approach to formalize the interaction between salience and choices of referring expressions using topic modeling, focusing specifically on the notion of topicality. We show that topic models can capture the observation that topical referents are more likely to be pronominalized. This lends support to theories of discourse salience that appeal to latent topic representations and suggests that topic models can capture aspects of speakers' cognitive representations of entities in the discourse.

Japanese speakers have a choice between canonical SOV and scrambled OSV word order to express the same meaning. Although previous experiments examine the influence of one or two factors for scrambling in a controlled setting, it is not yet known what kinds of multiple effects contribute to scrambling. This study uses naturally distributed data to test the multiple effects on scrambling simultaneously. A regression analysis replicates the NP length effect and suggests the influence of noun types, but it provides no evidence for syntactic priming, given-new ordering, and the animacy effect. These findings only show evidence for sentence-internal factors, but we find no evidence that discourse level factors play a role.