Functional Arabic Morphology Formal System and Implementation

types than just strings. Inflected forms need not be merged with roots yet, and can retain the internal structure: . . . ? paraVerbC Feminine Plural "u" FCuL −→ "u" >>| FCuL |<< "na" . . . ? merge "k t b" (Prefix "u" >| FCuL |< Suffix "na") −→ "uktubna" uktubna á . J » @ fem.pl. ‘write!’ The highlight of the Arabic morphology is that the ‘irregular’ inflection actually rests in strictly observing some additional rules, the nature of which is phonological. Therefore, surprisingly, ElixirFM does not even distinguish between verbal and nominal word formation when enforcing these rules. This reduces the number of paradigms to the prototypical 3 verbal and 5 nominal! Yet, the model is efficient. Nominal inflection is also driven by the information from the lexicon and by phonology. Note that the morphophonemic patterns and the Morphs a templates are actually extremely informative. We can use them as determining the inflectional class and the paradigm function, and thus we can almost avoid other unintuitive or excessive indicators of the kind of weak morphology, diptotic inflection, and the like. 5.4 Applications The ElixirFM linguistic model and the data of the lexicon can be integrated into larger applications or used as standalone libraries and resources. The language-independent part of the system could rest in the Functional Morphology library (Forsberg and Ranta, 2004). Among other useful things, it implements the compilation of the inflected word forms and their associated morphosyntactic categories into morphological analyzers and generators. The method used for analysis is deterministic parsing with tries, cf. also (Huet, 2002; Ljunglöf, 2002). Nonetheless, ElixirFM provides its original analysis method exploiting the inflectional invariant defined in Chapter 3. We can, at least in the present version of the implementation, dispense with the compilation into tries, and we use rather minimal computational resources. We define a class of types that can be Resolved, which introduces one rather general method resolveBy and one more specific method resolve, for which there is a default implementation. It says that the form in question should be resolved by equality (==) with the inflected forms in the model. The generic resolveBy method can be esp. used for recognition of partially vocalized or