Default unification operations combine strict information with information from one or more defeasible feature structures. Many such operations require finding the maximal subsets of a set of atomic constraints that are consistent with each other and with the strict feature structure, where a subset is maximally consistent with respect to the subsumption ordering if no constraint can be added to it without creating an inconsistency. Although this problem is NP-complete, there are a number of heuristic optimizations that can be used to substantially reduce the size of the search space. In this article, we propose a novel optimization, leaf pruning, which in some cases yields an improvement in running time of several orders of magnitude over previously described algorithms. This makes default unification efficient enough to be practical for a wide range of problems and applications.

Not that long ago, all linguistics was corpus linguistics.For much of the twentieth century, though, changes in theoretical fashion put linguists for whom collections of texts are the primary object of study at the margins of the field.Now things are shifting again, and a new generation of linguists with a wakening interest in linguistic data is returning to the field's corpus-based roots.What they are finding is that technological and methodological advances have quietly revolutionized corpus linguistics.In Corpus Linguistics: Readings in a Widening Discipline, editors Geoffrey Sampson and Diana McCarthy have put together a volume aimed at filling in some of what has happened in corpus linguistics while no one was watching.This collection of 43 reprinted papers, with original publication dates ranging from 1965 to 2002, includes a few of the usual suspects and a number of more-often-citedthan-read classics, making it an ideal source for an upper division course on corpus linguistics, or to supply what new corpus linguists (or their teachers) wish they had learned in grad school.Beyond these core contributions, however, the editors have wisely included an idiosyncratic selection of more-obscure papers guaranteed to stir the imagination of anyone with a professional interest in language.Papers on, for example, second language teaching, the discourse properties of Internet Relay Chat, and "nonindigenous minority languages" (such as South Asian languages spoken in the UK) show some of the possibilities that corpus-based methods have to offer beyond their philological foundations.While presented in chronological order, the papers in this volume for the most part fall into three classes.The largest is made up of papers on corpus design and methodology, such as Nelson Francis on the construction of the Brown corpus, Burnage and Dunlop on the British National Corpus, and B öhmová and Hajičová on the Prague Dependency Treebank.The second group are papers on the descriptive analysis of corpora.One, an excerpt from Charles Fries's The Structure of English, predates computational corpus analysis, but most of the rest are quantitative studies of English grammatical features.This group includes papers on cleft and pseudo-cleft constructions (Collins), that vs. null complementizers (Rissanen), and the use of terms of abuse (McEnery et al.).The third class of papers addresses technical issues in computational linguistics.Some of the classics in this group include Gale and Church on smoothing corpus counts, Hindle and Rooth on prepositional phrase attachment, and Briscoe and Carroll on parser evaluation.In this group we also find a number of less-well-known papers, such as contributions on coding dialog moves and on predicting intonation in spoken language corpora.Finally, there are a handful of papers that offer some personal and historical perspective on the field, most notably Sampson's "Reflections of a dendrographer," the text of a lecture originally delivered at a 2001 conference in honor of pioneering corpus linguist Geoffrey Leech.

The order of prenominal adjectival modifiers in English is governed by complex and difficult to describe constraints which straddle the boundary between competence and performance. This paper describes and compares a number of statistical and machine learning techniques for ordering sequences of adjectives in the context of a natural language generation system.

This paper describes new and improved techniques which help a unification-based parser to process input efficiently and robustly. In combination these methods result in a speed-up in parsing time of more than an order of magnitude. The methods are correct in the sense that none of them rule out legal rule applications.