Fundamental frequency as an acoustic cue to phonological phrase boundary in Spanish

Prosodic structure is encoded by speakers through a range of different articulatory and acoustic cues, i.e., gestural stretching, time lengthening (Cho, 2011; Cho, 2016) and pitch movements (Baek, 2019) among other possibilities. In Spanish, Lahoz-Bengoechea (2015) confirmed the presence of prosodic cues to phonological word boundaries from a production perspective. Polo-Cano & Elordieta (2016) approached the influence of phonological phrase boundary in phonological operations that involve segmental changes. However, the actual acoustic cues to phonological phrases in Spanish are yet to be explored. This work is focused on fundamental frequency as one of these cues and so it aims to study how phonological phrase boundaries model F0 contours. 30 Spanish speakers from Madrid were recorded reading the experiment corpus aloud in a recording booth. These texts gathered 60 sentences grouped in couples. Each couple of sentences had the same two syllables before and after a phonological word boundary (PW) or a phonological phrase boundary (PP). This is an example of a possible couple in the corpus: (a) El algodón decente no causa esos problemas, (b) Las débiles fibras de ese algodón dejarán bolas al lavarlo (Quality cotton doesn’t cause those issues, That cotton’s weak fibers will bobble after washing it). The two sentences contain syllables dón and de. On the one hand, (a) presents these syllables in a context of a PW; that is, between a noun and its adjective (Polo-Cano, 2015; Prieto, 2006). On the other hand, (b) presents those syllables in a context of PP; that is, between a long subject and its verb (Prieto, 2006). Those recordings were then segmented using Praat TextGrids to extract the interval that spanned from the beginning of the word before the prosodic boundary to the end of the word afterwards. Python’s Parselmouth (Jadoul et al., 2018) was used to extract F0 data from each sample and to interpolate. Pitch floors and ceilings were adjusted manually for every single participant. To normalize duration differences between samples, a fixed number of points were extracted from the words before and after the boundary. The whole set of F0 contours was analyzed using Functional Principal Components Analysis (Gubian et al., 2015), which allows to account for factors of variation in a dataset as a function of normalized time. PC1 and PC2 were the most informative components for this analysis. PC1 captured interspeaker variance including male/female distinction in F0. PC2 captured prosodic boundary differences. Both PW and PP constituents cause a rise followed by a drop across the boundary. Findings related to PP influence on F0 contours match what previous prosodic studies have set. PPs tend to align with syntactic boundaries (Selkirk, 2011), which was the starting premise for this study’s corpus design. Moreover, in Spanish there seems to be a tendency to mark PP with a boundary tone (level 2 phrase break in ToBI) (Prieto, 2006), which was the main result for PC2. Contours obtained using the reconstructing function for PC2 show a generalized trend to have a H- tone before the PP boundary.