Processing Judeo-Arabic Texts
Kfir Bar, Nachum Dershowitz, Lior Wolf, Yackov Lubarsky, Yaacov Choueka · 2015
Judeo-Arabic is a set of dialects spoken and written by Jewish communities living in Arab countries. Judeo-Arabic is typically written in Hebrew letters, enriched with diacritic marks that relate to the underlying Arabic. However, some inconsistencies in rendering words in Hebrew letters increase the level of ambiguity of a given word. Furthermore, Judeo-Arabic texts usually contain non-Arabic words and phrases, such as quotations or borrowed words from Hebrew and Aramaic. We focus on two main tasks: (1) automatic transliteration of Judeo-Arabic Hebrew letters into Arabic letters, and (2) automatic identification of language switching points between Judeo-Arabic and Hebrew. For transliteration, we employ a statistical translation system trained on the character level, resulting in 96.9% precision, a significant improvement over the baseline. For the language switching task, we use a word-level supervised classifier, also showing some significant improvements over the baseline.