Testing the stability of a phonological category in naturalistic speech: The case of hiatus in European Portuguese

The sequence /ia/ is phonologically classified as hiatus in European Portuguese, i.e., it consists of two full vowels with a syllable boundary in between. This stands in contrast to most Romance languages which either classify the vowel sequence as a diphthong /ja/, or follow rule-based alternations between hiatus and diphthong. While this typology is supported by small sets of controlled data in laboratory settings, we tested the stability of /ia/ in Portuguese using a large corpus of naturalistic, spontaneous speech. After a substantial cleaning and annotation process, 6806 tokens of /ia/ were selected for analysis. For each token, the first two formants and their duration were analysed using functional PCA, a technique that identifies systematic variations in multidimensional time-varying signals. The main finding was that the acoustic realization of /ia/ in Portuguese ranges from hiatus-like formant configurations when the sequence is stressed to almost monophthongal realizations in post-tonic position. This study highlights the benefitsof testing effects found in laboratory conditions also in naturalistic speech, where reduction processes can expose the phonetic gradience of phonological categories. In addition, the findings show that analyzing large corpora can help uncover unexpected patterns of variation and thus reveal the range of typological diversity.