zbMATH — the first resource for mathematics

Natural language understanding. 2nd ed. (English) Zbl 0851.68106
Redwood City, CA: The Benjamin/ Cummings Publ. Co., Inc. xv, 654 p. (1995).
This textbook on methodologies of natural language understanding is the revised second edition of the volume which originally appeared in 1987. The book is divided into three major parts, which deal with syntactic processing, semantic interpretation, and the use of discourse context and world knowledge.
The first part contains an outline of English syntax, grammar formalisms (context-free phrase structure grammar, recursive transition network grammar) and basic parsing algorithms (mainly based on charts), the augmentation of these basic grammar formalisms by feature systems (including descriptions of definite clause grammars and unification grammars) and the extension of the chart parser to handle a feature-based grammar, advanced topics of natural language grammars (e.g., questions, relative clauses and other forms of movement phenomena), their incorporation in a feature formalism and computational mechanisms to handle these more complex phenomena (e.g., hold list mechanism in augmented transition networks or gap threading in logic grammars), methodologies for efficient parsing (e.g., parsing with preferences, shift-reduce parsing, deterministic parsing, efficient encoding of ambiguity, partial parsing) and statistical methods for syntactic ambiguity resolution using part-of-speech tagging, lexical probabilities, probabilistic context-free grammars and a probabilistically driven best-first parser.
The second part on semantic interpretation considers basic issues of semantics and logical form (word senses and semantic primitives, the encoding of semantic ambiguity, semantic representation of verbs, thematic roles, the definition of semantic structure in terms of model theory, etc.), the linkage between syntax and semantics (compositionality, the computation of a logical form for predicate-argument structures, prepositional phrases, verb phrases and simple types of questions, lexicalized semantic interpretation and the use of semantic roles, etc.), semantical issues of ambiguity resolution (selectional restrictions, type hierarchies and their usage for semantic filtering in a constraint propagation algorithm, semantic networks as a generalization of type hierarchies, statistical word sense identification based on word collocations and statistical preference computations) and other strategies for semantic interpretation (e.g., semantic grammars, template matching, semantically driven parsing), together with a discussion of scoping phenomena, co-reference and binding constraints and the interpretation of English noun phrases. The third part discusses the role of knowledge representation (in terms of a first-order predicate calculus and a frame-like language, the mapping of logical form expressions to a knowledge representation language (as illustrated by the representation of complex quantifiers), inference methods in a logical and procedural semantics framework), the role of local discourse context and anaphoric reference (anaphora resolution based on history lists and a more advanced centering mechanism, the treatment of singular definite descriptions, plural descriptions and sets, definite reference, ellipsis and surface anaphora), the use of world knowlege for establishing discourse coherence and reference identification (e.g., knowledge about action and causality to generate expectations, script-based processing, plan-based techniques to generate expectations), and approaches to the description of discourse structure (methods for discourse segmentation and their effects on the interpretation of referential expressions, temporal and causal relations in discourse and inference-based discourse understanding). Pragmatic issues of natural language understanding are discussed in a chapter on conversational agents, i.e., ones that can participate fully in a natural dialogue, in which cognitive states of an agent (beliefs, desires, intentions, plans), speech acts, communicative acts and discourse-level planning are considered. Three appendices are provided which give an introduction to logic and model-theoretic semantics, symbolic computation (including search, matching and unification algorithms), and a brief survey of speech recognition and spoken language methods.

68T99 Artificial intelligence
68-01 Introductory exposition (textbooks, tutorial papers, etc.) pertaining to computer science
68T50 Natural language processing
68Q42 Grammars and rewriting systems