Word network (computational linguistics)

from Wikipedia, the free encyclopedia

A word network is a type of semantic network . In a word network, lexical signs and concepts are connected to one another by semantic relations . The difference to semantic networks is that

  • that the relationship of instances to classes is not modeled in word networks;
  • that the set of lexical-semantic and conceptual relations is limited to the psycholinguistically interesting relations.

A word network is technically an object-relational database .

To develop word networks

In the mid-1980s, a team of psycholinguists at Princeton University, led by George A. Miller, began building a word network for the English language . The main motive for the development of an extensive lexical resource was the expectation that with the help of this data theories about the mental lexicon could be tested more comprehensively and thus more precisely. The word network should include the basic English vocabulary, but should be limited to the meaningful word classes noun , verb , adjective and adverb . The Princeton WordNet has thus developed into a lexical database, the structure of which is based on psycholinguistic knowledge about the structure of the mental lexicon.

Princeton WordNet was quickly discovered as a valuable lexical resource for applications in computational linguistics and language technology. As a result, word networks were built for numerous other languages. A list of the languages ​​for which word networks exist can be found on the website of the Global Word Association , which also coordinates the work on word networks. The German word network GermaNet is being developed and maintained at the Eberhard-Karls-Universität Tübingen .

Structure of word networks

The elementary unit of description in a word network is the concept . Concepts are represented by one or more lexical characters. The lexical signs with the same meaning or synonymous are summarized in synsets . Example:

{Apfelsine, Orange}
{öffnen, aufmachen}

The meaning of a lexical sign can thus be inferred from its affiliation to a synset and from the place of this synset in the network. Some word networks, e.g. B. GermaNet, additionally specify the meaning of a concept by means of a gloss and examples of use.

The synsets are connected by the following conceptual relations:

  • Hyperonymy - orange → fruit or hyponymy - open → unlock
  • Meronymy - book → text , tomato saladtomato or holonymy - barracksmilitary ; Cake basecake
  • Entailment - 'x conditionally y' - snoresleep
  • Causation - 'x causes / causes y', openopen

Individual lexical units are linked by the following lexical-semantic relations:

  • Synonymy - lexical units with the same meaning are summarized in synsets. This indirectly models the synonymy relationship.
  • Antonymy - open - close ; open - close
  • Pertonymy - this relationship is primarily one of word formation and can be paraphrased with 'Expression x is related to Expression y' - financialfinance

The relations of hyperonymy / hyponymy and holonymy / meronymy create hierarchical relationships between the networked synsets, since both relationship types are transitive. The hierarchical organization of hyperonym / hyponym corresponds to the inheritance mechanism used in computer science : a hyponym inherits the semantic characteristics of all higher-level concepts and adds at least one characteristic to them. Most word networks also model multiple inheritance - a concept can be the hyponym of several concepts, e.g. B. 'Banana' → 'Tropical Fruit', → 'Plantation Plant'.

Synonymy, antonymy and pertonymy group the lexical units.

Cross-language structures

Many monolingual word networks are linked to one another via a common conceptual structure, which is represented by an interlingual index . For example, through the common concept 'DRIVE', the German expression drive is linked with the Italian expression guidare and the Spanish expression conducir .

See also


  • Claudia Kunze: Lexical-semantic word networks. In: K.-U. Carstensen et al. (Ed.): Computational Linguistics and Language Technology: An Introduction. Heidelberg; Berlin: Spektrum, Akademischer Verlag, pp. 386–393.
  • WordNet. An Electronic Lexical Database. Edited by Christiane Fellbaum. MIT Press 1998

Web links