Extracting Linguistic Knowledge from Speech: A Study of Stop Realization in 5 Romance Languages

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3257–3263 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3257 Extracting Linguistic Knowledge from Speech: A Study of Stop Realization in 5 Romance Languages Yaru Wu1,2,3, Mathilde Hutin2, Ioana Vasilescu2, Lori Lamel2, Martine Adda-Decker3 1CRISCO/EA4255, Universit´e de Caen Normandie, 14000 Caen, France, 2LISN, Univ.Paris-Saclay, 91405 Orsay cedex, France, 3Laboratoire de Phon´etique et Phonologie (UMR7018, CNRS-Sorbonne Nouvelle), France yaru.wu@unicaen.fr, {mathilde.hutin, ioana.vasilescu, lori.lamel}@lisn.upsaclay.fr, martine.adda-decker@sorbonne-nouvelle.fr Abstract This paper builds upon recent work in leveraging the corpora and tools originally used to develop speech technologies for corpus-based linguistic studies. We address the non-canonical realization of consonants in connected speech and we focus on voicing alternation phenomena of stops in 5 standard varieties of Romance languages (French, Italian, Spanish, Portuguese, Romanian). For these languages, both large scale corpora and speech recognition systems were available for the study. We use forced alignment with pronunciation variants and machine learning techniques to examine to what extent such frequent phenomena characterize languages and what are the most triggering factors. The results confirm that voicing alternations occur in all Romance languages. Automatic classification underlines that surrounding contexts and segment duration are recurring contributing factors for modeling voicing alternation. The results of this study also demonstrate the new role that machine learning techniques such as classification algorithms can play in helping to extract linguistic knowledge from speech and to suggest interesting research directions. Keywords: corpus-based linguistics, Romance languages, speech transcription and alignment, decision tree 1. Introduction The “big data” revolution has affected many research domains due to both the amount of available data and to many methodological changes needed to efficiently exploit the data. As is the case for many domains, speech technologies and corpus-based linguistics have been impacted by the increasing amounts of data avail- able for training automatic systems and for linguistic studies, and for novel machine learning techniques that can benefit from large corpora. The current scientific and technological state of the art reinforces the collab- oration between the two communities, allowing them to jointly take advantage of the abundance of data and the emergence of new methods. A shared topic concerns the patterns of variation in connected speech: statis- tically supported modeling of phonetic variation is of high interest for linguists that aim to ground their hy- potheses in ”naturalistic” data and for speech technol- ogy specialists that aim to ”master” the variation re- sponsible for processing errors. In fact, modeling variation in connected speech has been a long term research topic in linguistics and a challenge for speech technologists. For instance, pho- neticians and phonologists are particularly interested in sound variation, whether it be synchronic, observed at the present moment, or diachronic, happening over time. It is now accepted by both communities, that studies of such phenomena can be facilitated by speech technologies, e.g. by analyzing large, varied corpora for linguistic exploration. We aim to shed a new light on the study of voicing alternation in connected speech …

Content truncated to 500 words