An implemented constraint-based model of DO/SC ambiguity resolution

McRae, K.
University of Western Ontario

Hare, M. L.
Bowling Green State University

Elman, J. L.
University of California, San Diego

The resolution of temporary syntactic ambiguities has been a fruitful domain for understanding what information is used, and when, in sentence processing. One important such ambiguity arises in sentences involving verbs that can take either a Direct Object or Sentential Complement, e.g., admit, as in "The teacher admitted the students had little chance of
succeeding." Previous experiments have suggested that comprehenders are sensitive to the frequency with which such verbs are typically used in one or the other frame (Garnsey et al., 1997). More recently, it has been demonstrated that in many cases, such variation is correlated with alternate senses of a verb (Roland & Jurafsky, in press). Thus, with the
meaning "let in/allow", admit is overwhelmingly used with a DO, whereas with the meaning "acknowledge", it usually takes an SC. Hare, McRae, and Elman (submitted) have shown that, depending on prior sentence contexts that bias toward one or the other sense of a verb, comprehenders will initially parse the same ambiguous sentence in different ways, according to which syntactic frame is most appropriate for the inferred sense.

In the present work, we describe an implemented computational model that helps understand the time course of these disambiguation effects. In this model, various types of information are represented numerically and are integrated immediately in a word-by-word manner. The relevant onstraints were estimated using off-line sources of information (primarily corpus analyses). The constraint weights were estimated by fitting the model to off-line completion norms. When using the model to simulate word-by-reading times, we find that it exhibits similar patterns of disambiguation as were observed with the human data in Hare et al., including a reversal of expectations when the temporary ambiguity is
resolved. We discuss the assumptions that underlie the model and their implications.

References

Garnsey, S. M., Pearlmutter, N. J., Meyers, E., & Lotocky, M. A. (1997). The contribution of verb-bias and plausibility to the comprehension of temporarily ambiguous sentences. Journal of Memory and Language, 37, 58-93.

Hare, M., McRae, K., & Elman, J. L. Sense and Structure: Meaning as a determinant of verb subcategorization preferences. Manuscript submitted to Journal of Memory and Language.

Roland, D., & Jurafsky, D. (in press). Verb Sense and Verb Subcategorization Probabilities. In P. M. a. S. Stevenson (Ed.), 1998 CUNY Sentence Processing Conference . Philadelphia: Benjamins.