In this paper, we use a feature model of the semantics of plural determinersto present an approach to grammar checking for definiteness. Using neuralnetwork techniques, a semantics -- morphological category mapping was learned.We then applied a textual encoding technique to the 125 occurences of therelevant category in a 10 000 word narrative text and learned a surface --semantics mapping. By applying the learned generation function to the newlygenerated representations, we achieved a correct category assignment in manycases (87 %). These results are considerably better than a direct surfacecategorization approach (54 %), with a baseline (always guessing the dominantcategory) of 60 %. It is discussed, how these results could be used inmultilingual NLP applications.
展开▼