Enhancements to a Morphological Generator to Capture Arabic Morphology

We describe an enhanced version of the MORPHE tool, a morphological analyzer/generator designed to interface with a knowledge-based machine translation system. MORPHE uses a hierarchy (tree structure) to relate various morphological forms to each other based on common and distinctive features. Transformational rules are attached to the leaf nodes of the hierarchy. In generation, MORPHE takes as input a feature structure and pushes it through the hierarchy, which acts as discrimination net. When a leaf node is reached, MORPHE applies the attached rule. Each rule may contain several mutually exclusive clauses, each of which attempts to match a pattern against the base string contained in the feature structure and, if the match is successful, applies operators to the string to produce a transformed string. Our enhancements to MORPHE were motivated by attempting to use the tool to generate Arabic morphology. The non-concatenative morphology typical of Semitic languages has spurred the development of sophisticated formalisms and computational engines, as well as produced brute force approaches. In this paper we show how the relatively straightforward formalism used in MORPHE can be extended in simple ways to produce an elegant treatment of Arabic morphology that captures the inflectional regularities of the language. The result is the ability to describe Arabic morphology, as well as the morphology of any language whose word forms undergo stem changes, using a set of rules that contains minimal duplication, is easy to understand and maintain, and is useful for language learning as well as for machine translation applications.