The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9518

Title:Disambiguation of Compound Nouns by Domain-Specific Semantic Modelling
Date: 1995
Abstract:The ambiguity of compound nouns can be a problem both in syntactic and semantic terms. As a compound noun ,may involve two (or more) nouns in its composition, it makes difficult to decide about the grammaticality of them only in syntactic grounds. The resort to semantics can be a solution to particular cases. However, if we put the aim of systematize semantic knowledge in order to apply it largely in the disambiguation task, it becomes clear that compound nouns are also a problem in semantics. Which are the criteria to allow certain combinations and not allow others? The aim of this research is to develop a system which uses semantic information to disambiguate compound nouns. The first step is to exhaust the syntactic tools to do it. The second step is to choose a semantic domain - to work on semantic domains instead of general semantics allows more precision and clarity in the identification of the semantic nature of the chosen compounds. The third step is to build a system of features to describe the domain (and therefore its terms). This system involves, besides the feature description itself, the building of combination rules to allow certain compounds to be formed (and not others). One important element of the feature system here developed is the use of thematic roles built to fit the chosen domain: the agriculture commodities market. This system is a hybrid between a role system and a feature system, trying to take advantage of the qualities of both.

[Search These Pages] [DAI Home Page] [Comment]