Experiments With an MT-Directed Lexical Knowledge Bank

EXPERIMENTS WITH AN MT-DIRECTED LEXICAL KNOWLEDGE BANK B.C. Papegaaij V. Sadler A.P.M. Witkam BSO/Research Bureau voor Systeemontwikkeling P.O. Box 8348 3503 RH Utrecht The Netherlands Abstract A crucial test for any MT system is its power to solve lexical ambiluities. The size of the lexicon, its structural principles and the availability of extra-lin_~istic knowled~q are the most important aspects in this respect. This paper outlines the experimental development of the SWESIL system: a structured lexicon-based word expert system designed to play a pivotal role in the process of Distributed ~an_~ay ~ Translation (DLT) which is being developed in the Netherlands. It presents SWESIL's organizing principles, gives a short description of the present experimental set-up and shows how SWESIL is being tested at this moment. Introduction The DLT project [WITKAM 1983] utilizes an Intermediate Lan£ua_~ (IL), in which form the translated text will be transported to the various receivers, where it will be translated into any of the avaiIable target-languages. DLT, therefore, is essentially a double translation - first into, then out of - the IL. As a consequence of this strategy, one of DLT's central features is the pivotal role played by its Lexical Knowledge Bank. This knowledge bank, a central part of SWESIL (Semantic Word-Expert System for the InterLingua), has been designed to contain the lexical knowledge the DLT system needs: the entire vocabulary, contextual dependencies and a]l the system's semantic knowledf~: the information necessary to distinguish between word meanings, knowledge about the way those meanings mutually influence each other, and the procedural knowledge that enables the system to calculate the probability of each possible interpretation of a given sentence (or fragment) and carry out the intricate process of le___xxical disambiguation and a~digmatic selection. As both sides of the tranlation process greatly rely on the kind of knowledge SWESIL embodies, and since they both use the IL as medium for disambiguation, it appeared advantageous from the start to store SWESIL's knowledge entirely in IL format. This choice is supported by a number of properties of the IL such as its lack of homonyms, freedom from syntactic ambiguities, power of expression, and easy inspectability [see for details: WITKAM 1983]. The structure of the knowledge bank The SWESIL knowledge bank is a multi-dimensional taxonomy, incorporating ~ , 9ntological, and contextual dependency information. Its basic unit is the dictiqnary entrl (Table i). The starting point for each entry is an IL lexeme: a single IL word in its undeclined form. To discriminate between various intended usages of the one lexeme, and to link the lexeme into the taxonomic tree structure, it is given one k e ~ for every meaning to be defined. This keyterm is a more abstract term [or the concept the lexeme itself denotes, and relates to the lexeme through a kind of ISA link. Note the general structure: lexemo <keyterm> <<supeF keyterlq>> followed the depoucy pairs. [)el,elldel~ey pairs wiLh more that% one wc)rd on ei[her side o{ the relator must be read as abb[eviated forms for all the …

Content truncated to 500 words