Syntactic and Semantic Factors in Processing Difficulty: An Integrated Measure

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 196–206, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Syntactic and Semantic Factors in Processing Difficulty: An Integrated Measure Jeff Mitchell, Mirella Lapata, Vera Demberg and Frank Keller University of Edinburgh Edinburgh, United Kingdom jeff.mitchell@ed.ac.uk, mlap@inf.ed.ac.uk, v.demberg@ed.ac.uk, keller@inf.ed.ac.uk Abstract The analysis of reading times can pro- vide insights into the processes that under- lie language comprehension, with longer reading times indicating greater cognitive load. There is evidence that the language processor is highly predictive, such that prior context allows upcoming linguistic material to be anticipated. Previous work has investigated the contributions of se- mantic and syntactic contexts in isolation, essentially treating them as independent factors. In this paper we analyze reading times in terms of a single predictive mea- sure which integrates a model of seman- tic composition with an incremental parser and a language model. 1 Introduction Psycholinguists have long realized that language comprehension is highly incremental, with readers and listeners continuously extracting the meaning of utterances on a word-by-word basis. As soon as they encounter a word in a sentence, they inte- grate it as fully as possible into a representation of the sentence thus far (Marslen-Wilson 1973; Konieczny 2000; Tanenhaus et al. 1995; Sturt and Lombardo 2005). Recent research suggests that language comprehension can also be highly pre- dictive, i.e., comprehenders are able to anticipate upcoming linguistic material. This is beneficial as it gives them more time to keep up with the in- put, and predictions can be used to compensate for problems with noise or ambiguity. Two types of prediction have been observed in the literature. The first type is semantic predic- tion, as evidenced in semantic priming: a word that is preceded by a semantically related prime or a semantically congruous sentence fragment is processed faster (Stanovich and West 1981; van Berkum et al. 1999; Clifton et al. 2007). Another example is argument prediction: listeners are able to launch eye-movements to the predicted argu- ment of a verb before having encountered it, e.g., they will fixate an edible object as soon as they hear the word eat (Altmann and Kamide 1999). The second type of prediction is syntactic predic- tion. Comprehenders are faster at naming words that are syntactically compatible with prior con- text, even when they bear no semantic relationship to the context (Wright and Garrett 1984). Another instance of syntactic prediction has been reported by Staub and Clifton (2006): following the word either, readers predict or and the complement that follows it, and process it faster compared to a con- trol condition without either. Thus, human language processing takes advan- tage of the constraints imposed by the preceding semantic and syntactic context to derive expecta- tions about the upcoming input. Much recent work has focused on developing computational mea- sures of these constraints and expectations. Again, the literature is split into syntactic and semantic models. Probably the best known measure of syn- tactic expectation is surprisal (Hale 2001) …

Content truncated to 500 words