How to build a multilingual inheritance lexicon

Carole Tiberius (Surrey Morphology Group, University of Surrey)

WO4: General Issues On Lexicons

This paper discusses a fairly new approach to multilingual lexical representation which abstracts away from the traditional MT architecture to multilingual lexicons. Rather than linking the monolingual lexicons at the level of semantics only, we aim to construct a multilingual lexicon in which information can be shared at all levels of linguistic description using an inheritance-based formalism. In this paper, we present different architectures that can be used to build such a multilingual inheritance lexicon. Two main approaches are contrasted, a non-parameterised and a parameterised. In a non-parameterised approach, language is not used as a parameter in the multilingual lexicon. The multilingual lexicon consists of a set of monolingual hierarchical lexicons plus a shared hierarchical lexicon containing what the monolingual lexicons have in common.  In a parameterised model, on the other hand, all information is integrated into one single hierarchy, and language is used as a parameter to indicate which parts of the hierarchy are valid for which language. The advantages and disadvantages of the different approaches will be discussed with respect to small sample fragments of Dutch, English, Danish, and Icelandic nouns implemented in DATR.

Inheritance-Based multilingual lexicons parameterised non-parameterised

83.pdf