Memory-based learning, keeping full memory of learning material, appears a viable approach to learning tasks, and is often superior in generalization accuracy to eager learning approaches that abstract from learning mate- rial. Here we investigate three Iw'tial memorybased learning approaches which remove from memory specific task instance types estimated to be exceptional. The three approaches each implement one heuristic function for estimating excepttonality of instance types: (i) typicaltry, (ii) class prediction strength, and friendly-neighbourhood size. Experiments are performed with the memory-based learning algorithm ml-Ia trained on English word pro- nunciation. We find that removing instance types with low prediction strength (it) is the only tested method which does not seriously harm generalization accuracy. We conclude that keeping full memory of types rather than tokens, and excluding minority ambiguities pear to be the only performance-preserving optimi -ations of memory-based learning.

This paper presents a decision-tree approach to the problems of part-ofspeech disambiguation and unknown word guessing as they appear in Modem Greek, a highly inflectional language. The learning procedure is tag-set independent and reflects the linguistic reasoning on the specific problems. The decision trees induced are combined with a highcoverage lexicon to form' a tagger that achieves 93,5% overall 'disambiguation accuracy.

This paper presents a machine learning approach to the problems of part-of-speech disambiguation and unknown word guessing, as they appear in Modern Greek. Both problems are cast as classification tasks carried out by decision trees. The data model acquired is capable of capturing the idiosyncratic behavior of underlying linguistic phenomena. Decision trees are induced with three algorithms; the first two produce generalized trees, while the third produces binary trees. To meet the requirements of the linguistic datasets, all three algorithms are able to handle set-valued attributes. Evaluation results reveal a subtle differentiation in the performance of the three algorithms, which achieve an accuracy range of 93-95% in POS disambiguation and 82-88% in guessing the POS of unknown words. INTRODUCTION It has recently become apparent that empirical ML can find in NLP an exciting application area. The increasing use of corpus-based learning in place of manual encoding has led to ...