A Purely Surface-Oriented Approach to Handling Arabic Morphology

摘要

In this paper, we introduce a completely lexicalist approach to deal with Arabic morphology. This purely surface-oriented treatment is part of a comprehensive mathematical approach to integrate Arabic syntax and semantics using overt morphological features in the string-to-meaning translation. The basic motivation of our approach is to combine semantic representations with formal descriptions of morphological units. That is, the lexicon is a collection of signs; each sign 5 is a triple 5 — ＜E,C, M＞, such that E is the exponent, C is the combinatorics and M is the meaning of the sign. Here, we are only concerned with the exponents, i.e. the components of a morphosemantic lexicon (for a fragment of Arabic). To remain surface-oriented, we allow for discontinuity in the constituents; constituents are sequences of strings, which can only be concatenated or duplicated, but no rule can delete, add or modify any string. Arabic morphology is very well known for its complexity and richness. The word formation in Arabic poses real challenges because words are derived from roots, which bear the core meaning of their derivatives, formed by inserting vowels and maybe other consonants. The units in the sequences are so-called glued strings rather than only strings. A glued string is a string that has left and right context conditions. Optimally morphs are combined in a definite and non-exceptional linear way, as in many cases in different languages (e.g. plural in English). The process of Arabic word formation is rather complex; it is not just a sequential concatenation of morphs by placing them next to each other. But the constituents are discontinuous. Vowels and more consonants are inserted between, before and after the root consonants resulting in what we call 'fractured glued string', i.e. as a sequence of glued strings combined in diverse ways; forward concatenation, backward concatenation, forward wrapping, reduction, forward transfixation and, going beyond the multi-context free grammars (MCFGs), also reduplication.