Natural Language Processing: The PLNLP Approach

Natural language is easy for people and hard for machines. For two generations, the tantalizing goal has been to get computers to handle human languages in ways that will be compelling and useful to people. Obstacles are many and legendary.
Natural Language Processing: The PLNLP Approach describes one group's decade of research in pursuit of that goal. A very broad coverage NLP system, including a programming language (PLNLP) development tools, and analysis and synthesis components, was developed and incorporated into a variety of well-known practical applications, ranging from text critiquing (CRITIQUE) to machine translation (e.g. SHALT). This books represents the first published collection of papers describing the system and how it has been used. Twenty-six authors from nine countries contributed to this volume.
Natural language analysis, in the PLNLP approach, is done is six stages that move smoothly from syntax through semantics into discourse. The initial syntactic sketch is provided by an Augmented Phrase Structure Grammar (APSG) that uses exclusively binary rules and aims to produce some reasonable analysis for any input string. Its `approximate' analysis passes to the reassignment component, which takes the default syntactic attachments and adjusts them, using semantic information obtained by parsing definitions and example sentences from machine-readable dictionaries. This technique is an example of one facet of the PLNLP approach: the use of natural language itself as a knowledge representation language -- an innovation that permits a wide variety of online text materials to be exploited as sources of semantic information.
The next stage computes the intrasential argument structure and resolves all references, both NP- and VP-anaphora, that can be treated at this point in the processing. Subsequently, additional components, currently not so well developed as the earlier ones, handle the further disambiguation of word senses, the normalization of