Design and Implementation of the Slovenian Phonetic and Morphology Lexicons for the Use in Spoken Language Applications

Phonetic and Morphology Lexicons that can be used in Spoken Language Applications are costly and time-consuming to build. This paper reports on a project aiming at the semi-automatic development of large phonetic (SIflex) and morphology (SImlex) lexicons for Slovenian language. The main goal of the project is to build the phonetic and morphology lexicon for Slovenian language that will be used within the framework of various applications in speech processing (e.g. speech synthesis and recognition), natural language processing (e.g. spell checking) and for studying and assessing automatic grapheme-to-phoneme transcription. In automatic speech recognition one of the major problem is extremely high variability of pronunciations. One part of this variability can be taken into account through a training of the acoustic-phonetic units from a large amount of data. Another part of variability must be modeled in the lexicon as pronunciation variants. In the case of text-to-speech systems it is also very usable to be able to detect homographs and choose the correct pronunciation according to the context information. All this was our motivation for developing both lexicons for Slovenian language. Currently the created phonetic lexicon (SIflex) contains more than 130.000 items, whereas the morphology lexicon (SImlex) consists of approximately 600.000 inflected forms, including information on the orthography, pronunciation, stress and morphosyntactic features, as defined in the framework of the Multext project.