The Influence of Discourse on Syntax: A Psycholinguistic Model of Sentence Processing

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1179–1188, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics The Influence of Discourse on Syntax A Psycholinguistic Model of Sentence Processing Amit Dubey Abstract Probabilistic models of sentence com- prehension are increasingly relevant to questions concerning human language processing. However, such models are of- ten limited to syntactic factors. This paper introduces a novel sentence processing model that consists of a parser augmented with a probabilistic logic-based model of coreference resolution, which allows us to simulate how context interacts with syntax in a reading task. Our simulations show that a Weakly Interactive cognitive architecture can explain data which had been provided as evidence for the Strongly Interactive hypothesis. 1 Introduction Probabilistic grammars have been found to be useful for investigating the architecture of the human sentence processing mechanism (Jurafsky, 1996; Crocker and Brants, 2000; Hale, 2003; Boston et al., 2008; Levy, 2008; Demberg and Keller, 2009). For example, probabilistic models shed light on so-called locality effects: contrast the non-probabilistic hypothesis that dependants which are far away from their head always cause processing difficulty for readers due to the cost of storing the intervening material in memory (Gibson, 1998), compared to the probabilistic prediction that there are cases when faraway dependants facilitate processing, because readers have more time to predict the head (Levy, 2008). Using a computational model to address funda- mental questions about sentence comprehension motivates the work in this paper. So far, probabilistic models of sentence pro- cessing have been largely limited to syntactic factors. This is unfortunate because many out- standing questions in psycholinguistics concern interactions between different levels of process- ing. This paper addresses this gap by building a computational model which simulates the influence of discourse on syntax. Going beyond the confines of syntax alone is a sufficiently important problem that it has attracted attention from other authors. In the literature on probabilistic modeling, though, the bulk of this work is focused on lexical semantics (e.g. Pad´o et al., 2006; Narayanan and Jurafsky, 1998) or only considers syntactic decisions in the preceed- ing text (e.g. Dubey et al., 2009; Levy and Jaeger, 2007). This is the first model we know of which introduces a broad-coverage sentence processing model which takes the effect of coreference and discourse into account. A major question concerning discourse-syntax interactions involves the strength of communica- tion between discourse and syntactic information. The Weakly Interactive (Altmann and Steed- man, 1988) hypothesis states that a discourse context can reactively prune syntactic choices that have been proposed by the parser, whereas the Strongly Interactive hypothesis posits that context can proactively suggest choices to the syntactic processor. Support for Weak Interaction comes from experiments in which there are temporary ambi- guities, or garden paths, which cause processing difficulty. The general finding is that supportive contexts can reduce the effect of the garden path. However, Grodner et al. (2005) found that sup- portive contexts even facilitate the processing of unambiguous sentences. As …

Content truncated to 500 words