Cross-Lingual Consistency of Phonological Features: An Empirical Study

Phonological Features for 0-Shot Multilingual Speech Synthesis

“…PFs offer a shared model topology across languages, similar to the "byte-like" representation used in [10], and maintain the connection to abstract, phonological categories, while also providing explanatory power on a level closer to the acoustics of an utterance [12]. While the applicability of a specific PF set to all languages is questionable, certain phonological contrasts, such as "front-back", have been shown to generalise across various language families [13]. At a lower bound, where phonological categories such as "fricatives", "rounded vowels", etc.…”

Towards Induction of Structured Phoneme Inventories

“…Overall, our findings are mixed. A specially designed experiment for predicting contrasts in unvoiced labial consonants between Bengali and Spanish produced consistent and cross-lingually robust predictions (Johny et al, 2019), also for a variety of auditory representations (Gutkin, 2020), despite the conflicting status of some of the allophones of the phonemes in the experiment. Similarly robust were contrasts between front and back vowels, as well as the vowel height and continuant manner of articulation distinctions (Skidmore and Gutkin, 2020).…”

“…We followed a simple method: to consider a phonemic contrast to be consistent or robust across languages, it needs to be easily predictable on heldout languages in a binary classification task (Johny et al, 2019). An instance of this problem consists of a span of a speech signal (e.g., a vowel in surrounding context) and a positive or negative label (e.g., front vowel vs. back vowel).…”

Managing Phonological Inventory Data in the Development of PHOIBLE

“…There has also been a broad range of published research that uses the PHOIBLE data, including studies on linguistic and genetic diversity (Creanza et al 2015), phonetics , phonology (Cohen Priva 2017), typology , historical linguistics (Barrack, McCloy, & Wright 2014), and computational linguistics ( Johny, Gutkin, & Jansche 2019). 20 Some of my own research has involved testing whether there is a correlation between the population size of speech communities and the number of sounds in their languages (Moran, McCloy, & Wright 2012), investigating whether there are compensations Interoperable, and Reusable) principles (Wilkinson 2016) and provides a new digital object identifier (DOI) for each new release along with a bibliographic citation that users can use when citing a particular release of the data in their research papers.…”