A Phrase-Structured Grammatical Framework for Transportable Natural Language Processing

We present methods of dealing with the syntactic problems that arise in the construction of natural language processors that seek to allow users, as opposed to computational linguists, to customize an interface to operate with a new domain of data. In particular, we describe a grammatical formalism, based on augmented phrase-structure rules, which allows a parser to perform many important domain-specific disambiguations by reference to a pre-defined grammar and a collection of auxiliary files produced during an initial knowledge acquisition session with the user. We illustrate the workings of this formalism with examples from the grammar developed for our Layered Domain Class (LDC) system, though similarly motivated systems ought also to benefit from our formalisms. In addition to showing the theoretical advantage of providing many of the fine-tuning capabilities of so-called semantic grammars within the context of a domain-independent grammar, we demonstrate several practical benefits to our approach. The results of three experiments with our grammar and parser are also given.