General system for normal and phonetic inflection

Generating all inflected forms for a natural language is a very difficult task not only because of the large number of inflection rules but also because of the large number of exceptions. The operation is more difficult if we consider not only the normal inflected forms (where the inflected forms are represented in the normal alphabet of the involved language) but also the phonetic forms (where the inflected forms are represented in a phonetic alphabet). This paper presents a general system (i.e. a system that can be used for any inflected natural language) that generates all the normal and phonetic forms of words from a given lexicon. A set of results obtained by using this system for Romanian language is also presented. The system is based on GRAALAN metalanguage (grammar abstract language) used for representing the linguistic knowledge concerning a natural language and requested by the inflection process. It also uses a set of tools that allows the handling of this (meta) language. The system starts with a set of linguistic knowledge bases described in GRAALAN containing: normal and phonetic inflection rules, the description of the normal and phonetic alphabets that are used, the base forms (lemmas) that will be inflected, taken from a lexicon (also in normal and phonetic forms) and the description of the morphological structure of the language (morphological categories and their values). All these linguistic knowledge bases are created using special tools. The inflected normal/phonetic forms will be automatically generated by applying the normal/phonetic inflection rules on the lemmas of the lexicon, taking into account the morphological structure and the normal/phonetic alphabets. At the end of the process, the linguist is able to verify and eventually correct the generated forms. When corrections are made, new inflection rules are automatically generated. The presented system is part of larger system for natural language processing.