INTRODUCTION TO GENERATIVE LEXICONJAMES PUSTEJOVSKYGenerative Lexicon is a theory of linguistic semantics which focuses on the distributednature of compositionality in natural language. Unlike purely verb-based approaches tocompositionality, Generative Lexicon (henceforth, GL) attempts to spread the semanticload across all constituents of the utterance. Central to the philosophical perspective ofGL are two major lines of inquiry: (1) How is it that we are able to deploy a finite numberof words in our language in an unbounded number of contexts? (2) Is lexical informationand the representations used in composing meanings separable from our commonsenseknowledge?IntroductionGenerative Lexicon introduces a knowledge representation framework which offers a richand expressive vocabulary for lexical information. The motivations for this are twofold.Overall, GL is concerned with explaining the creative use of language; we consider thelexicon to be the key repository holding much of the information underlying this phe-nomenon. More specifically, however, it is the notion of a constantly evolving lexiconthat GL attempts to emulate; this is in contrast to currently prevalent views of static lexi-con design, where the set of contexts licensing the use of words is determined in advance,and there are no formal mechanisms offered for expanding this set.One of the most difficult problems facing theoretical and computational semantics isdefining the representational interface between linguistic and non-linguistic knowledge.GL was initially developed as a theoretical framework for encoding selectional knowl-edge in natural language. This in turn required making some changes in the formal rulesof representation and composition. Perhaps the most controversial aspect of GL has beenthe manner in which lexically encoded knowledge is exploited in the construction of in-terpretations for linguistic utterances. Following standard assumptions in GL, the com-putational resources available to a lexical item consist of the following four levels:(1) a. LEXICAL TYPING STRUCTURE: giving an explicit type for a word positionedwithin a type system for the language;b. ARGUMENT STRUCTURE: specifying the number and nature of the arguments toa predicate;c. EVENT STRUCTURE: defining the event type of the expression and any subeven-tual structure it may have; with subevents;d. QUALIA STRUCTURE: a structural differentiation of the predicative force for alexical item.1The qualia structure, inspired by Moravcsik’s (1975) interpretation of the aitia of Aristotle,are defined as the modes of explanation associated with a word or phrase in the language,and are defined as follows (Pustejovsky, 1991):(2) a. FORMAL: the basic category of which distinguishes the meaning of a word withina larger domain;b. CONSTITUTIVE: the relation between an object and its constituent parts;c. TELIC: the purpose or function of the object, if there is one;d. AGENTIVE: the factors involved in the object’s origins or “coming into being”.Conventional interpretations of the GL semantic representation have been as feature struc-tures (cf. Bouillon, 1993, Pustejovsky, 1995). The feature representation shown belowgives the basic template of argument and event variables, and the specification of thequalia structure.αARGSTR =ARG1 = x. . .EVENTSTR =E1 = e1. . .QUALIA =CONST = what x is made ofFORMAL = what x isTELIC = function of xAGENTIVE = how x came into beingTraditional Lexical RepresentationsThe traditional organization of lexicons in both theoretical linguistics and natural lan-guage processing systems assumes that word meaning can be exhaustively defined by anenumerable set of senses per word. Lexicons, to date, generally tend to follow this orga-nization. As a result, whenever natural language interpretation tasks face the problem oflexical ambiguity, a particular approach to disambiguation is warranted. The system at-tempts to select the most appropriate ‘definition’ available under the lexical entry for anygiven word; the selection process is driven by matching sense characterizations againstcontextual factors. One disadvantage of such a design follows from the need to specify,ahead of time, the contexts in which a word might appear; failure to do so results in in-complete coverage. Furthermore, dictionaries and lexicons currently are of a distinctlystatic nature: the division into separate word senses not only precludes permeability; italso fails to account for the creative use of words in novel contexts.GL attempts to overcome these problems, both in terms of the expressiveness of nota-tion and the kinds of interpretive operations the theory is capable of supporting. Ratherthan taking a ‘snapshot’ of language at any moment of time and freezing it into lists of2word sense specifications, the model of the lexicon proposed here does not preclude ex-tensibility: it is open-ended in nature and accounts for the novel, creative, uses of wordsin a variety of contexts by positing procedures for generating semantic expressions forwords on the basis of particular contexts.Adopting such a model presents a number of benefits. From the point of view of alanguage user, a rich and expressive lexicon can explain aspects of learnability. From thepoint of view of linguistic theory, it can offer improvements in robustness of coverage.Such benefits stem from the fact that the model offers a scheme for explicitly encodinglexical knowledge at several levels of generalization. In particular, by making lexical am-biguity resolution an integral part of a uniform semantic analysis procedure, the problemis rephrased in terms of dynamic interpretation of a word in context; this is in contrast tocurrent frameworks which select among a static, pre-determined set of word senses, anddo so separately from constructing semantic representations for larger text units.There are several methodological motivations for importing tools developed for thecomputational representation and manipulation of knowledge into the study of wordmeaning, or lexical semantics. Generic knowledge representation (KR) mechanisms, suchas inheritance structures or rule bases, can—and have been—used for encoding of linguis-tic information. However, not much attention has been paid to the notion of what exactlyconstitutes such linguistic information. Traditionally, the