Fast parsing using pruning and grammar specialization

Abstract

We show how a general grammar may be automatically adapted for fast parsing of utterances from a specific domain by means of constituent pruning and grammar specialization based on explanation-based learning. These methods together give an order of magnitude increase in speed, and the coverage loss entailed by grammar specialization is reduced to approximately half that reported in previous work. Experiments described here suggest that the loss of coverage has been reduced to the point where it no longer causes significant performance degradation in the context of a real application.

Learning efficient parsing

“…In contrast to our approach in which no manual annotation is required, Rayner and Carter (1996) report that for each sentence in the training data, the best parse was selected manually from the set of parses generated by the parser. For the experiments described in the paper, this constituted an effort of two and a half person-months.…”

“…An overview of grammar specialization techniques is given in (Sima'an, 1999). For instance, Rayner and Carter (1996) use explanation-based learning to specialize a given general grammar to a specific domain. They report important efficiency gains (the parser is about three times faster), coupled with a mild reduction of coverage (5% loss).…”

Applying explanation-based learning to control and speeding-up natural language generation

“…By using these filters it is possible to restrict the range of structural properties of candidate phrasal templates (e.g., extract only saturated NPs, or subtrees having at least two daughters, or subtrees which have no immediate recursive structures). These filters serve the same means as the "chunking criteria" described in (Rayner and Carter, 1996).…”

Hybrid language processing in the Spoken Language Translator

“…The practical utility of the specialized grammar is largely determined by the loss of coverage incurred by the specialization process. We show in [19] that suitable "chunking" criteria and a training corpus of a few thousand utterances in practice reduce the coverage loss to a level which does not affect the performance of the system to a significant degree.…”

“…To achieve efficient parsing, we need to be able to focus our search on only a small portion of the space of theoretically valid grammatical analyses. Our work on parsing (see [19] for a fuller description, albeit one based on a slightly earlier version of the system) is a logical continuation of two specific strands of research aimed in this general direction.…”