Topological Parsing

This chapter presents a new formalism used in ALE for parsing with freer word-order languages. This new formalism, which is based on a CFG-like backbone, does not presuppose contiguity of constituents nor does it presuppose any linear order. Where necessary, however, these can be explicitly expressed in order to provide the grammar with sufficient constraints to allow for more efficient parsing.

This is achieved by departing from the traditional notion of constituency where a single region (namely conventional grammatical categories such as NP, VP, etc.) is used to define aspects of interpretation (e.g. thematic role assignment), linear order, and contiguity. In the topological-fields (TF) approach, two types of constituency are involved. One type consists of the regions used to capture aspects of interpretation only, and the other involves the regions that capture aspects of linear order and contiguity. This idea was first introduced by Curry (1961) but was not completely formalised. In Curry's terms, the regions that capture interpretation belong to the domain of tectogrammar and those responsible for linear order and contiguity belong to the domain of phenogrammar. Traditional grammatical categories stripped of their linear-order and contiguity assumptions correspond to tectogrammatical categories. As for phenogrammatical categories, we extend the notion of topological fields used by traditional German grammarians to define contiguous regions with a fixed order of constituents. The TF approach has three properties:

• Topological Linearity: All word order constraints are defined on some linear region.
• Topological Locality: All discontiguities are characterised in some local region.
• Qualified Isomorphy: Regions that captures linear order and those that capture contiguity are the same (i.e., topological fields).