In this paper we describe a conversion of the Buckwalter Morphological Analyzer for Arabic, originally written as a Perl-script, into a pure finite-state morphological analyzer. Representing a morphological analyzer as a finite-state transducer (FST) confers many advantages over running a procedural affix-matching algorithm. Apart from application speed, an FST representation immediately offers various possibilities to flexibly modify a grammar. In the case of Arabic, this is illustrated through the addition of the ability to correctly parse partially vocalized forms without overgeneration, something not possible in the original analyzer, as well as to serve both as an analyzer and a generator.
展开▼