Memory-based morphological analysis generation andpart-of-speech tagging of Arabic

摘要

We explore the application of memorybasedlearning to morphological analysisand part-of-speech tagging of writtenArabic, based on data from the ArabicTreebank. Morphological analysis – theconstruction of all possible analyses ofisolated unvoweled wordforms – is performedas a letter-by-letter operation predictiontask, where the operation encodessegmentation, part-of-speech, characterchanges, and vocalization. Part-of-speechtagging is carried out by a bi-modular taggerthat has a subtagger for known wordsand one for unknown words. We report onthe performance of the morphological analyzerand part-of-speech tagger. We observethat the tagger, which has an accuracyof 91.9% on new data, can be used toselect the appropriate morphological analysisof words in context at a precision of64.0 and a recall of 89.7.