Out of vocabulary guesser : Swahili

Abstract

Free texts include also such words, which are not listed in the analysis system. Yet they need to be treated as part of the vocabulary, so that the unknown elements in text do not unnecessarily disturb the translation process. They cannot be fully treated as the known lexical items, but if we know some basic propertied of the words, we can figure out the structure of the sentence kore precisely. Traditionally, the heuristic guessing of such unknown words was done on the basis of the morphological form of the word only. In this report it is suggested that the unknown words should be treated in two phases. First, we give a tentative assignment of the word in the word-level guesser. In the second phase we test the assignment in context. The first assignment may have two or more assignment candidates, and in the second phase we test which one is the correct one in the context.