Transdutor de estados finitos para conversão de grafema para a pronúncia da variedade linguística potiguar

Abstract

The Potigrafone is a finite-state transducer that automatically transcripts the graphic forms from the Portuguese language to the phonetic forms belonging to the“potiguar” linguistic variety. It was through the development of this system, by using of the Atlas Linguístico do Centro-Oeste Potiguar and to the Atlas Geolinguístico do Litoral Potiguar, that it has been possible to identify and separate the phonetic patterns in the linguistic variety belonging to the studied region, presenting a phonetic transcription tool while using the Speech Assessment Methods Phonetic Alphabet - SAMPA; and, in addition, a performance assessment in the corpus of written texts in Portuguese has been made through the use of two multi-level measures: Exact Match Ratio and Labelling Fscore. To do so, the identification and separation of phonological patterns present in the phonetic letters of Atlas, the definition of some relevant criteria to inputs in the system, the implementation of rules relating to underlying form with the surface form and, also, two system evaluation measures have been needed. The results have showed that the Potigrafone performance was 81% for the Exact Match Ratio measurement and 89% for the Labelling FScore, presenting thus, increased accuracy in total or partial correct attempts, when the transcript is made without stress markup and without the phonetic phenomena of nasality, the palatalization, the diphthongization, the monophthongization, the deletion, the epenthesis; another important result is that the consonant modules and the syllabic core also interfere with this performance, since the values with all modules and without these two modules are similar. The Potigrafone performs well during the phonetic transcription, but there are some occasional errors that have made necessary the creation of new rules to correct them and, consequently, help to improve the system performance, both in accuracy and in full and partial transcription match.