13 - On-line processing, working memory and modularity

Summary

Introduction

In the previous chapter we outlined two opposing theories of the role that syntactic processing plays in sentence comprehension. According to one view – the modular theory, inspired by early psycholinguistic attempts to apply Chomsky's generative grammar – a specialized syntactic parser assigns grammatical structure to an input sentence, yielding an intermediate representation which strongly constrains the assignment of meaning, but which needs to be further operated upon by interpretive (semantic and pragmatic) processes to yield the full meaning of the utterance. According to the opposing view, dubbed the interactive model, sentence meanings are assigned incrementally to word sequences as soon as they are identified, making maximal use of whatever constraints can be applied from the speakers' tacit knowledge of the grammar of their language, pragmatic knowledge and expectations, or even collocational restrictions on word usage (such as habitual phrases or idioms). Sometimes these cues will conflict, in which case constraints may compete to produce local ambiguities which are usually resolved by further input.

In principle, it should be possible to decide between these opposing models (or some intermediate theory between the two) if we had some means of observing changes in state of the language processor as it steps through the input sentence in real time. We may never fully achieve this privileged perspective, but over the past two or three decades a variety of ‘on-line’ techniques, based initially upon behavioural reaction time measurements and latterly upon functional neural imaging techniques, have been devised, which arguably enable us to observe local fluctuations in ‘processing load’, as sentences are judged or comprehended in real time.