Speech-language acquisition involves learning the speech sounds of the language at hand as well as which acoustic cues are relevant to differentiate them. For example, the Dutch vowels /ɑ/ and /a/ in the words 'man' (man) and 'maan' (moon) differ both in their spectral properties (F1 and F2 are both higher for /a/) and in duration (longer for /a/). Perception experiments indicate that the way in which different perceptual cues are being combined and weighted is language specific, however, there are large individual differences in cue weighting within language groups (Escudero, Benders, & Lipski, 2009). Furthermore, the different perceptual cues are not entirely independent from a speech acoustics point of view, meaning that one of the cues could play a more prominent role underneath. The goal of the present project is to unravel (1) what exactly influences how perceptual cues are being weighed, and (2) which cue is actively being manipulated by speakers.

In the pilots, we combined measurements of the perceptual weighting of acoustical cues of phonological contrasts with acoustical, articulatory (EMA), and muscle activation (EMG) measurements of the production of those contrasts.

First results of a pilot study with 45 young adult speakers of Dutch (age 19-29) investigating the Dutch /ɑ/ - /a/ vowel contrast showed (1) a correlation between perceptual acuity for spectral properties and the perceptual weighing of spectral properties and duration and (2) a negative correlation between how participants weigh perceptual cues and how the participants use these cues in production (Terband et al., 2017). Future directions include expanding the phonological contrasts and combined auditory feedback perturbation and EEG-mismatch negativity measurements.

Suprasegmental cues: articulatory correlates of stress and prominence

Stressed or pragmatically prominent syllables are typically produced with longer durations and a more precise articulation. For instance, stressed vowels in English are not reduced to schwas, while unstressed vowels often are. Articulatory strategies to mark pragmatic prominence have been proposed in several studies. One way is to move the tongue to a more extreme position, the other is to increase the sonority contrast in more prominent syllables, by making consonants less and vowel more sonorous (Beckman, Edwards, & Fletcher, 1992; de Jong, Beckman, & Edwards, 1993; Harrington, Fletcher, & Beckman, 2000). However, these studies suffer from very small sample sizes, the studies of Beckman et al. (1992) and de Jong et al. (1993) featured the same two participants while Harrington et al. (2000) recorded 5 participants, and strategies seemed to vary between speakers. The goal of this project is to investigate (1) the production strategies employed to mark stress and prominence in Dutch, and (2) the effect of stress or pragmatic prominence on assimilation of place of articulation in Dutch in a sample of 40 speakers of Dutch.

A pilot study comprised acoustic and articulatory (EMA) recordings of a speaking task in which lexical stress and focus (new and/or contrastive information; see e.g., Vallduv, 1993) were made orthogonal. The target syllable were Dutch morphemes ‘in’ (/ɪn/) and ‘uit’ (/œyt/) in the context of a following labial or alveolar consonant. Stress was manipulated by placing the morpheme in a compound verb (e.g., /'ɪnpɑkə/, to pack) or as preposition in a phonemically identical phrase (e.g., /ɪn 'pɑkə/, in packages). The morpheme was made pragmatically prominent to three degrees by embedding it in a given sentence that served as an answer to a given question. Pragmatic prominence levels varied between low (after a different phrase with narrow focus), neutral (the whole sentence or phrase contains new information), and high (the morpheme alone corrects an assumption).

The results of an in-depth exploratory analysis of a subset of four participants did not reveal any effects of prominence on articulatory or acoustic duration of segments or syllables (Lentz & Terband, 2018). The results did show that consonants were produced at more extreme positions when pragmatic prominence was higher, while at the same time, the speakers did not consistently make larger articulatory movements to achieve this. This suggests that when a prominent consonant is realised, the tongue might be positioned for it during the preceding vowel. This exploration allows for a focussed use of the larger dataset for hypothesis testing. In addition, future directions include the analysis of individual differences to identify possible participant-specific strategies.