Context Effects in Language Production: Models of Syntactic Priming in Dialogue Corpora
This thesis addresses the cognitive basis of syntactic adaptation, which biases speakers to repeat their own syntactic constructions and those of their conversational partners. I address two types of syntactic adaptation: short-term priming and longterm adaptation. I develop two metrics for syntactic adaptation within a speaker and between speakers in dialogue: one for short-term priming effects that decay quickly, and one for long-term adaptation over the course of a dialogue. Both methods estimate adaptation in large datasets consisting of transcribed human-human dialogue annotated with syntactic information. Two such corpora in English are used: Switchboard, a collection of spontaneous phone conversation, and HCRC Map Task, a set of task-oriented dialogues in which participants describe routes on a map to one another. I find both priming and long-term adaptation in both corpora, confirming well-known experimental results (e.g., Bock, 1986b). I extend prior work by showing that syntactic priming effects not only apply to selected syntactic constructions that are alternative realizations of the same semantics, but still hold when a broad variety of syntactic phrase structure rules are considered. Each rule represents a cognitive decision during syntactic processing. I show that the priming effect for a rule is inversely proportional to its frequency. With this methodology, I test predictions of the Interactive Alignment Model (IAM, Pickering and Garrod, 2004). The IAM claims that linguistic and situation model agreement between interlocutors in dialogue is the result of a cascade of resource-free, mechanistic priming effects on various linguistic levels. I examine task-oriented dialogue in Map Task, which provides a measure of task success through the deviance of the communicated routes on the maps. I find that long term syntactic adaptation predicts communicative success, and it does so earlier than lexical adaptation. The result is applied in a machine-learning based model that estimates task success based on the dialogue, capturing 14 percent of the variance in Map Task. Short-term syntactic priming differs qualitatively from long term adaptation, as it does not predict task success, providing evidence against learning as a single cognitive basis of adaptation effects. I obtain further evidence for the correlation between semantic activity and syntactic priming through a comparison of the Map Task and Switchboard corpora, showing that short-term priming is stronger in task-oriented dialogue than in spontaneous conversation. This difference is evident for priming between and within speakers, which suggests that priming is a mechanistic rather than strategic effect. I turn to an investigation of the level at which syntactic priming influences language production. I establish that the effect applies to structural syntactic decisions as opposed to all surface sequences of lexical categories. To do so, I identify pairs of part-of-speech categories which consistently cross constituent boundaries defined by the phrase structure analyses of the sentences. I show that such distituents are less sensitive to priming than pairs occurring within constituents. Thus, syntactic priming is sensitive to syntactic structure. The notion of constituent structure differs among syntactic models. Combinatory Categorial Grammar (CCG, Steedman, 2000) formalizes flexible constituent structure, accounting a varying degree of incrementality in syntactic sentence planning. I examine whether priming effects can support the predictions of CCG using the Switchboard corpus, which has been annotated with CCG syntax. I confirm the syntactic priming effect for lexical and non-lexical CCG categories, which encode partially satisfied subcategorization frames. I then show that both incremental and normal-form constituent structures exhibit priming, arguing for language production accounts that support flexible incrementality. The empirical results are reflected in a cognitive model of syntactic realization in language production. The model assumes that language production is subject to the same principles and constraints as any other form of cognition and follows the ACT-R framework (Anderson et al., 2004). Its syntactic process implements my empirical results on priming and is based on CCG. Syntactic planning can take place incrementally and non-incrementally. The model is able to generate simple sentences that vary syntactically, similar to the materials used in the experimental priming literature. Syntactic adaptation emerges due to a preferential and sped-up memory retrieval of syntactic categories describing linearization and subcategorization requirements. Long-term adaptation is explained as a form of learning, while shortterm priming is the result of a combination of learning and spreading activation from semantic and lexical material. Simulations show that the model produces the adaptation effects and their inverse frequency interaction, as well as cumulativity of long-term adaptation.