Lexeme-based morphology: a computationally expensive approach intended for a server-architecture

• Authors: Marc Domenig
• Venue: COLING '90 Proceedings of the 13th conference on Computational linguistics - Volume 2
• Year: 1990

This paper presents an approach to computational morphology which can be considered as being derived from the two-level model but differs from this substantially. Lexemes rather than formatives are the most important entities distinguished in this approach. The consequence is that a new formalism for the specification of morphological knowledge is required. A short description of a system called Word Manager will outline the characteristics of such a formalism, the most prominent of which is that different subformalisms for inflectional rules and word-formation rules are distinguished. These rules are applied separately though not independently and support the concept of lexicalization. The primary advantage of this is that the system can build up a network of knowledge on how formatives, lexemes, and rules depend on each other while individual lexemes are lexicalized. Thus, the system will know the inflectional forms of a lexeme, the destructuring of these forms into formatives, how the lexeme has been derived or composed if it is a word-formation, etc. This requires much memory, yet, the philosophy behind the approach is that the system runs as a server on a local area network, so that an entire machine can be dedicated to the task, if necessary.