A Model of Discourse Predictions in Human Sentence Processing

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 304–312, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics A Model of Discourse Predictions in Human Sentence Processing Amit Dubey and Frank Keller and Patrick Sturt Human Communication Research Centre, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB, UK {amit.dubey,frank.keller,patrick.sturt}@ed.ac.uk Abstract This paper introduces a psycholinguistic model of sentence processing which combines a Hidden Markov Model noun phrase chun- ker with a co-reference classifier. Both mod- els are fully incremental and generative, giv- ing probabilities of lexical elements condi- tional upon linguistic structure. This allows us to compute the information theoretic mea- sure of surprisal, which is known to correlate with human processing effort. We evaluate our surprisal predictions on the Dundee corpus of eye-movement data show that our model achieve a better fit with human reading times than a syntax-only model which does not have access to co-reference information. 1 Introduction Recent research in psycholinguistics has seen a growing interest in the role of prediction in sentence processing. Prediction refers to the fact that the hu- man sentence processor is able to anticipate upcom- ing material, and that processing is facilitated when predictions turn out to be correct (evidenced, e.g., by shorter reading times on the predicted word or phrase). Prediction is presumably one of the factors that contribute to the efficiency of human language understanding. Sentence processing is incremental (i.e., it proceeds on a word-by-word basis); there- fore, it is beneficial if unseen input can be antici- pated and relevant syntactic and semantic structure constructed in advance. This allows the processor to save time and makes it easier to cope with the con- stant stream of new input. Evidence for prediction has been found in a range of psycholinguistic processing domains. Semantic prediction has been demonstrated by studies that show anticipation based on selectional restrictions: listeners are able to launch eye-movements to the predicted argument of a verb before having encoun- tered it, e.g., they will fixate an edible object as soon as they hear the word eat (Altmann and Kamide, 1999). Semantic prediction has also been shown in the context of semantic priming: a word that is pre- ceded by a semantically related prime or by a seman- tically congruous sentence fragment is processed faster (Stanovich and West, 1981; Clifton et al., 2007). An example for syntactic prediction can be found in coordinate structures: readers predict that the second conjunct in a coordination will have the same syntactic structure as the first conjunct (Fra- zier et al., 2000). In a similar vein, having encoun- tered the word either, readers predict that or and a conjunct will follow it (Staub and Clifton, 2006). Again, priming studies corroborate this: Compre- henders are faster at naming words that are syntacti- cally compatible with prior context, even when they bear no semantic relationship to it (Wright and Gar- rett, 1984). Predictive processing is not confined to the sen- tence level. Recent experimental results also …

Content truncated to 500 words