This book takes an empirical approach to language processing, based on applying statistical and other machine-learning algorithms to large corpora.Methodologyboxes are included in each chapter.Each chapter is built around one or more worked examplesto demonstrate the main idea of the chapter. Covers the fundamental algorithms of various fields, whether originally proposed for spoken or written language to demonstrate how the same algorithm can be used for speech recognition and word-sense disambiguation. Emphasis on web and other practical applications. Emphasis on scientific evaluation.Useful as a reference for professionals in any of the areas of speech and language processing.

Daniel Jurafsky joined the faculty of the Department of Linguistics, Department of Computer Science, and the Institute of Cognitive Science at the University of Colorado at Boulder.

Preface This is an exciting time to be working in speech and language processing. Historically distinct fields (natural language processing, speech recognition, computational linguistics, computational psycholinguistics) have begun to merge. The commercial availability of speech recognition and the need for Web-based language techniques have provided an important impetus for development of real systems. The availability of very large on-line corpora has enabled statistical models of language at every level, from phonetics to discourse. We have tried to draw on this emerging state of the art in the design of this pedagogical and reference work: Coverage In attempting to describe a unified vision of speech and language processing, we cover areas that traditionally are taught in different courses in different departments: speech recognition in electrical engineering; parsing, semantic interpretation, and pragmatics in natural language processing courses in computer science departments; and computational morphology and phonology in computational linguistics courses in linguistics departments. The book introduces the fundamental algorithms of each of these fields, whether originally proposed for spoken or written language, whether logical or statistical in origin, and attempts to tie together the descriptions of algorithms from different domains. We have also included coverage of applications like spelling-checking and information retrieval and extraction as well as areas like cognitive modeling. A potential problem with this broad-coverage approach is that it required us to include introductory material for each field; thus linguists may want to skip our description of articulatory phonetics, computer scientists may want to skip such sections as regular expressions, and electrical engineers skip the sections on signal processing. Of course, even in a book this long, we didn't have room for everything. Thus this book should not be considered a substitute for important relevant courses in linguistics, automata and formal language theory, or, especially, statistics and information theory. Emphasis on Practical Applications It is important to show how language-related algorithms and techniques (from HMMs to unification, from the lambda calculus to transformation-based learning) can be applied to important real-world problems: spelling checking, text document search, speech recognition, Web-page processing, part-of-speech tagging, machine translation, and spoken-language dialogue agents. We have attempted to do this by integrating the description of language processing applications into each chapter. The advantage of this approach is that as the relevant linguistic knowledge is introduced, the student has the background to understand and model a particular domain. Emphasis on Scientific Evaluation The recent prevalence of statistical algorithms in language processing and the growth of organized evaluations of speech and language processing systems has led to a new emphasis on evaluation. We have, therefore, tried to accompany most of our problem domains with aMethodology Boxdescribing how systems are evaluated (e.g., including such concepts as training and test sets, cross-validation, and information-theoretic evaluation metrics like perplexity). Description of widely available language processing resources Modern speech and language processing is heavily based on common resources: raw speech and text corpora, annotated corpora and treebanks, standard tagsets for labeling pronunciation, part-of-speech, parses, word-sense, and dialogue-level phenomena. We have tried to introduce many of these important resources throughout the book (e.g., the Brown, Switchboard, callhome, ATIS, TREC, MUC, and BNC corpora) and provide complete listings of many useful tagsets and coding schemes (such as the Penn Treebank, CLAWS C5 and C7, a