An Empirical Evaluation of LFGDOP.pdfVIP

An Empirical Evaluation of LFG-DOP Rens Bod Informatics Research Institute, University of Leeds, Leeds LS2 9JT, Institute for Logic, Language and Computation, University of Amsterdam rens@scs.leeds.ac.uk Abstract studied from a mathematical perspective by Cormons This paper presents an empirical assessment of the LFG- (1999) who also accomplished a first simple experiment DOP model introduced by Bod Kaplan (1998). The with LFG-DOP. Next, Way (1999) studied LFG-DOP as parser we describe uses fragments from LFG-annotated an architecture for machine translation. The current sentences to parse new sentences and Monte Carlo paper contains the first extensive empirical evaluation techniques to compute the most probable parse. While of LFG-DOP on the currently available LFG-annotated our main goal is to test Bod Kaplans model, we will corpora: the Verbmobil corpus and the Homecentre also test a version of LFG-DOP which treats generalized corpus. Both corpora were annotated at Xerox PARC. fragments as previously unseen events. Experiments with Our parser uses fragments from LFG-annotated the Verbmobil and Homecentre corpora show that our sentences to parse new sentences, and Monte Carlo version of LFG-DOP outperforms Bod Kaplans techniques to compute the most probable parse. model, and that LFGs fun