Transparent parsing : Head-driven processing of verb-final structures

Abstract

The conceptual guideline underlying this study is that the goal of processing theory should be to construct a transparent parser. A transparent parser is a parser which employs only properties and relations that are available in the grammar, without resorting to processing-specific notions.

Under a lexicalist conception of grammar, such as the Minimalist Program, this conceptual guideline leads to the conclusion that the structure-building mechanism available to the human parser must be head-driven. As a starting point for developing a transparent parser, this study takes the processing model of Pritchett (1992), which is introduced in chapter 2. In this head-driven model, initial analysis is driven by the need to satisfy a grammatical principle (the Theta Criterion) as soon as possible. The model also features a reanalysis constraint, the On-Line Locality Constraint, which specifies two structurally defined conditions under which the human parser can perform reanalysis without evoking a garden path effect.

Head-driven processing models have encountered resistance in the literature. Frazier (1987, 1989) and Schneider (1999) argue that the processing complexity which can occur in headfinal structures, shows that the parser builds structure before it encounters the head. In chapter 3, this argumentation is critically examined and rejected. The discussion involves a variety of locally ambiguous structures, such as relative clauses in Dutch, and preverbal PP-attachment ambiguities.

Chapter 4 discusses the processing of verb-final structures under Pritchett's theory. One language of interest is Dutch, which is verb-final in embedded clauses, and verb-second in main clauses. Another language of interest is Japanese. Although Pritchett s theory goes a long way in explaining the virtual absence of garden path effects in Japanese, it does predict processing difficulties in Japanese in some contexts that are actually processed without effort, as has been pointed out by Mazuka and Itoh (1995). To accommodate these cases, an extension of Pritchett's reanalysis constraint is proposed. It is argued on the basis of Dutch and Japanese that the reanalysis process that is available to the human parser can reanalyze material from the edge of a phase to a position outside that phase.

Chapter 5 gives an overview of the resulting system, and shows how it accounts for the English, Japanese and Dutch data that were discussed. Furthermore, a unified formulation of the reanalysis constraint is proposed, in terms of domain closure. Chapter 6 concludes that it is possible to construct a transparent parser that accounts for garden path phenomena in both OV- and VO-structures, and that garden path data provide no evidence that there is a need to posit a separate parser for head-final languages such as Japanese, as is sometimes proposed. The resulting system comprises a step forward in the formulation of a universal, transparent parser.