Ambiguity and Linguistic Preferences

Gregor Erbach

CLAUS-Report Universität des Saarlandes 23 6/1992.


Attempting to treat ambiguities in typed feature formalisms presents a dilemma. The exploitation of linguistic knowledge for adding additional constraints (e.g. word order, selectional restrictions) to the grammar may indeed help disambiguation, but it also rules out some perfectly grammatical non-ambiguous strings. Ways out of this dilemma are discussed in this paper. They include - reliance on processin g strategies without addition of additional knowledge - processing guided by statistical probability - leaving desambiguation to external knowledge sources - using additional constraints only if they are needed for disambiguation. We propose the addition of preference values to typed ferature structures. Using preference values has the effect that violation of the additional constraints needed for disambiguation only decreases the preference value, but does not make the sentence unacceptable. Disambiguation is achieved by selecting the reading with the highest preference value. We think that it is possible to define a processing strategy (preference-driven linguistic deduction) that finds the preferred reading first.

Weitere Links

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence