Computational prosodic typology and its application to understudied languages

Uwe Reichel & Katalin Mády, RIL HAS

Prosodic categories such as rhythm (syllable- vs. stress-timed), headedness (left vs. right) or the presence or absence of certain phrase levels (e.g. accentual phrase) have often been subject to discussion, especially in languages that do not represent typical cases of these categories. In this paper, we attempt to describe the above-mentioned categories by a set of prosodic features, based on well-studied languages such as Hungarian, English, French, and German. For this, numerous pitch, energy and rhythm features based on computational prosody stylization were tested for their appropriateness to distinguish between languages. We created prosodic typology profiles from these features, for which we will give phonetic interpretations. We further trained and tested random forest classifiers on the prosodically categorized data in order to obtain a proof of concept to apply this classification on prosodically understudied languages, namely Estonian and Slovak.