Lexicons limit their contents to information about words, i.e., concepts that are lexicalized. By contrast, an ontology attempts to reflect conceptual structures, regardless of actual concept-word mappings in a given language. A semantic network like WordNet combines properties of both, in that it organizes the lexemes of a language into the kinds of hierarchical structures familiar from ontologies. Such a network reveals both the extent to which concepts and words are mapped in a systematic fashion, and the location of lexical gaps. It also allows one to examine whether there is a non-arbitrary correlation between a word's position in the hierarchy and the formal properties of its encoding.
Verbs in WordNet are organized by semantic relations to other verbs. The most important relations are (cognitive) synonymy, which yields synonym sets (synsets), and a manner relation, which links synsets fitting the formula "to X is to Y in some manner", and which builds subsumption hierarchies. Almost all English verbs can be assigned to synsets and hierarchies (Fellbaum, 1998a). However, VP idioms tend to resist such integration, with the systematic exception of those, like "kick the bucket," that serve a euphemistic function (Fellbaum, 1998b).
We suggest that many VP idioms express complex concepts that, unlike those encoded by simple verbs, have no straightforward semantic relation (e.g. synonymy or super/subordination) to other event concepts. Idioms appear to break the regular lexicalization patterns based on semantic elaboration, which distinguish super- from subordinate word-concept pairs; idioms frequently encode multiple and distinct types of semantic components. Their semantic complexity--which cannot be attributed to the presence of added constitutents such as objects--precludes their integration into an ontology/network populated by simple words and concepts. It is also paralleled by a correspondingly formal complexity.
We propose a rough typology of VP idioms that appear to denote events differing fundamentally from those encoded by literally referring verbs in terms of their semantic elaboration. One class includes VP idioms referring to complex events that could be broken down into sub-events: "walk the plank" (walk down a plank, jump off, and drown); events implying a prior event: "turn the other cheek" (refrain from retaliating after having suffered something), and idioms incorporating an operator ("fish or cut bait"). Another type of idiom expresses a (covert or overt) negation ("make no bones about", "go begging", "fall on deaf ears"). A third type incorporates an aspectual component, such as the initiation or termination of certain categories of activities ("get on the stick", "call it a day", "jump the gun"). In all cases, the events expressed by these VPs have a special ontological status when compared against the bulk of the English lexicon as projected in a semantic net.
We make some proposals as to why such complex events are not expressed "on the fly" by means of the usual rules of composition, but by idiomatic, pre-formed phrases.
Fellbaum, C. (1998a). A Semantic Network of English. In: Fellbaum, C. (Ed.) "WordNet". Cambridge, MA: MIT Press.
Fellbaum, C. (1998b). Towards a Representation of Idioms in WordNet. In: Harabagiu, S. (Ed.) Proceedings of the Workshop on Usage of WordNet in Natural Language Processing Systems, COLING/ACL 1998, Montreal, Canada.
| Workshop Homepage | Workshop Program | Previous Abstract | Next Abstract |