Finding variants of out-of-vocabulary words in Arabic

Abstract

Transliteration of a word into another language often leads to multiple spellings. Unless an information retrieval system recognises different forms of transliterated words, a significant number of documents will be missed when users specify only one spelling variant. Using two different datasets, we evaluate several approaches to finding variants of foreign words in Arabic, and show that the longest common subsequence (LCS) technique is the best overall.