Structural ambiguity

The parsing model to be described in detail in Chapter 4 addresses particular questions that have arisen from two related, but distinct, research programs investigating natural language processing. The first seeks to investigate hypotheses concerning human sentence comprehension within the context of experimental psycholinguistics; the second is concerned with the computational implications of such hypotheses. In recent years there has been a considerable amount of convergence concerning basic principles of language comprehension. For example, the important role of phrase structure in language processing has been convincingly demonstrated, and it is a rare processing model which does not make some reference to syntactic constituents and structural relations. There is a growing consensus that, given ambiguous input, the perceptual mechanism has an initial bias toward the reading consistent with the minimal structural representation.

Further, owing to the important work of Marslen-Wilson (1973, 1975) there is general agreement that models of syntactic processing must be responsive to the experimental demonstrations of the speed and efficiency of human language processing.

As this chapter will make clear, important, and interesting, differences remain, but real progress has been made since the mid-1970s. The purpose of this chapter is to outline some of the main avenues of research responsible for this progress. This will give a context for the parsing model described in Chapter 4. This model, as will become clear, incorporates significant aspects of the parsers described in the following sections.