Generalized Monotonicity for Reanalysis Models

Abstract

A common assumption in psycholinguistic theory is that reanalysis is constrained by a preference to preserve certain aspects of the representation built in response to previous input. In this chapter, we discuss this notion of representation-preservation in the wider context of models of reanalysis as a whole, and point out that in order to define a representation-preserving constraint on reanalysis, we must specify not only which aspects of representation should be preserved, but what is meant by the notion of preservation. We propose that the appropriate notion of preservation is that which is assumed in monotonic models of parsing, where structural relations between linguistic elements are updated totally non-destructively from state to state. Previous monotonic theories of parsing have limited themselves to consideration of phrase structure representations. In contrast, we propose a general framework within which one may formulate models which apply the same notion of preservation to other representation types. The framework is discussed with reference to a model which preserves thematic structure.

Much of the material in this chapter was presented at seminars given at the University of Freiburg, Germany, and the University of Torino, Italy, during July and September 1996. We would like to thank the seminar participants for their instructive comments, especially Barbara Hemforth, Lars Konieczny, Leonardo Lesmo, Vincenzo Lombardo, and Gerhard Strube. We would also like to thank Suzanne Stevenson and an anonymous reviewer for their comments and suggestions which have greatly improved the quality of this chapter. Any remaining errors are our own. The research reported here was supported by the award of ESRC studentship No. R00429334338 to the first author.