Computational Linguistics Colloquium

Monday, November 05, 16:15, Building 17, Conference Room (2.11)
NOTE UNUSUAL DAY AND PLACE

Topological Parsing

Parsing with freer word-order languages is difficult. Part of the reason for this difficulty is that we must potentially consider all possible orderings of the daughters on the right-hand side of a phrase structure rule if we are not allowed to assume that this has been implicitly specified. This kind of ambiguity needs to be resolved either by statistics that indicate the most likely orderings or by the use of semantic, functional or other knowledge to remove the ambiguity.

Another part of the reason, however, is that even the notion of what a constituent is and what its predictions are with respect to order or contiguity may be entirely misrepresented by classical phrase structure. Various alternative representations have been proposed that have found their way into parsing specifications, including Zwicky's (1986) use of "liberation" and LFG's separation of constituent structure and functional structure. Within the HPSG tradition, work in this direction began with several influential publications by Reape on word-order domains, culminating in Kathol's (1995) dissertation on incorporating the traditional topological analysis of German clausal word-order into a domain-based HPSG analysis.

This talk presents an overview of our on-going work on extending topological fields from a means of describing clause-level German syntax to a flexible means of describing the lack of isomorphy between the traditional notion of constituency and word-order variation in a much wider range of constructions in natural language. We will pay particular attention to how topological fields can also be used to control the combinatorial consequences of freer word-order in parsing with such grammar specifications.

If you would like to meet with the speaker, please contact Valia Kordoni.