INDUCTION OF INFLECTION RULES WITH CLASSIFICATION AND ASSOCIATIVE MEMORY FOR HUNGARIAN LANGUAGE.

Inflection is a vital element to express semantic in synthetic languages. Proper induction is crucial for text generation and reporting systems. The induction of inflection rules is an open question in computational linguistics. The existing solutions use dictionary, transformation rules or statistical observations to inflect a stem. These methods have drawbacks either in precision and cost efficiency. This paper present a hybrid method which is based on classification and associative memory. The words which belong to non- frequent categories are stored in the associative memory thus the classification process can be performed faster. The transformations for the regular words are determined by the classifier. Precision, size and time cost of the algorithm are measured with different sized associative memory. The tests were performed on a training set of (stem, inflected form) pairs for the accusative case in Hungarian. The precision of the hybrid algorithm can exceed the 90 per cent based on the experimental results. [ABSTRACT FROM AUTHOR]