A computational theory of human linguistic processing: memory limitations and processing breakdown

Abstract

This thesis gives a theory of sentence comprehension that attempts to explain a number of linguistic performance effects, including garden-path effects, preferred readings for ambiguous input and processing overload effects. It is hypothesized that the human parser heuristically determines its options based upon evaluation of possible representations with respect to lexical, syntactic, semantic and pragmatic properties, each of which is associated with a weight. Processing overload effects are explained by the assumption of the existence of a maximum load corresponding to the limited capacity of short term memory: a structure becomes unacceptable at a particular parse state if the combination of the processing weights associated with its properties at that state is greater than the available capacity. Furthermore, it is assumed that the language processor is an automatic device that maintains only the best of the set of all compatible representations for an input string. This thesis assumes a formulation of representational evaluation within a parallel framework: one structure is preferred over another if the processing load associated with the first structure is markedly lower than the processing load associated with the second. Thus a garden path effect results if the unpreferred structure is necessary for a successful parse of the input. Four properties of linguistic representations are presented within this framework. The first two--the Properties of Thematic Reception and Transmission--derivable from the $\theta$-Criterion from Government-Binding (GB) Theory (Chomsky (1981)); the third--the Property of Lexical Requirement--derivable from the Projection Principle of GB Theory; and the fourth--the Property of Recency Preference--prefers local attachments over more distant attachments (cf. Kimball (1973), Frazier (1979)). This thesis shows how these properties interact to give a partially unified theory of many performance effects.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 252 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.