In this paper a statistical study of linguistically structure-based Portuguese syllables (henceforth, PS) is described for the Portuguese language. This study is particularly attractive for assigning statistical information to PS. Our belief is that a formalised statistical approach of PS prototypes will help to improve Portuguese language teaching/learning systems.
Location of syllable boundaries is a well-known problem of non-consensual resolution in Portuguese language, particularly when we focus on acoustic-phonetic decoding models of language processing.
A complete set of all syllable structures for the Portuguese language is presented, which complements the traditional approaches. Based on segment sequences of C (consonants) and V (vowels), syllable structure rules are implemented and verified using about 400K Portuguese words. All sequential CV Portuguese patterns are automatically detected and related to syllabic structures. When the map of all grapheme components of PS structures is outlined in a statistical form, some confirm our expectations, some apparently run counter to expectation and some appear to be new.
PS splitting, together with the stress determination and the grapheme-to-phone transcription, is an essential tool in the architecture of our text-to-speech systems. This PS is part of the pre-processing module of that system and its function is to split words into syllables. We argue that the syllabification mechanism will ultimately be useful for improving pronunciation modelling.
Finally, the present study suggests that the phones-in-PS and PS-in-words context might also be an interesting factor for acoustic modelling We will explore this in future work.