Acquiring Plausible Uni cation-Based Grammars using Model-Based and Data-Driven Learning

Undergeneration is a problem that undermines successful parsing of unrestricted texts. A popular solution to this problem is automatic grammar correction (or machine learning of grammar). Broadly speaking, grammar correction approaches can be classiied as being either data-driven, or model-based. Data-driven learners use data-intensive methods to acquire grammar. They typically use grammar formalisms unsuited to the needs of practical text processing. That is, data-driven learners acquire grammars that overgen-erate and fail to assign linguistically plausible parses. Model-based learners are knowledge-intensive and are reliant for success upon the completeness of a model of grammaticality. But, in practice, the model will be incomplete and so since we deal with undergeneration by learning, we hypothesise that the combined use of data-driven and model-based learning would allow data-driven learning to compensate for model-based learning's incompleteness, whilst model-based learning would compensate for data-driven learning's inadequacy. We describe a system that we have used to test this hypothesis empirically. The system combines data-driven and model-based learning to acquire uniication-based grammars that are more suitable for practical text parsing. The system is used to illustrate the strengths and weaknesses of both learning styles. Finally, using the Spoken English Corpus as data, and by quantitatively measuring undergeneration, overgeneration and parse plaus-ibility, we show that the hypothesis is promising.