A Semi-Automatic and Low Cost Approach to Build Scalable Lemma-based Lexical Resources for Arabic Verbs

—This work presents a method that enables\udArabic NLP community to build scalable lexical\udresources. The proposed method is low cost and efficient\udin time in addition to its scalability and extendibility. The\udlatter is reflected in the ability for the method to be\udincremental in both aspects, processing resources and\udgenerating lexicons. Using a corpus; firstly, tokens are\uddrawn from the corpus and lemmatized. Secondly, finite\udstate transducers (FSTs) are generated semi-automatically.\udFinally, FSTs are used to produce all possible inflected\udverb forms with their full morphological features. Among\udthe algorithm’s strength is its ability to generate\udtransducers having 184 transitions, which is very\udcumbersome, if manually designed. The second strength\udis a new inflection scheme of Arabic verbs; this increases\udthe efficiency of FST generation algorithm. The\udexperimentation uses a representative corpus of Modern\udStandard Arabic. The number of semi-automatically\udgenerated transducers is 171. The resulting open lexical\udresources coverage is high. Our resources cover more\udthan 70% Arabic verbs. The built resources contain\ud16,855 verb lemmas and 11,080,355 fully, partially and\udnot vocalized verbal inflected forms. All these resources\udare being made public and currently used as an open\udpackage in the Unitex framework available under the\udLGPL license