Dynamics of Phonological Cognition

Finished · Speech Production

Summary

Gafos and Benus argue that cases of incomplete neutralization (e.g. German word-final devoicing) and vowel harmony patterns (e.g., in Hungarian) provide a need for a model of the phonetics-phonology interface in which there is no translation of symbolic representations into phonetic space. Instead, articulatory phonology is used together with a nonlinear dynamical model to account for both phenomena. There are attractors (i.e. faithfulness constraints) that are based on the symbolic representation (gestural representation) of a segment, but other attractors (i.e. markedness constraints) can result in the actual production of a gesture along a continuous dimension. Depending on the strength of different attractors (i.e., global linguistic knowledge in my dissertation), neutralization of a contrast like word-final obstruent voicing can fall along a continuous dimension of incomplete neutralization, up to and including practical complete neutralization.

Most importantly, the vowel harmony pattern in Hungarian is very interesting for how transparent front vowels like [i] can affect the harmony pattern. In Hungarian, suffix vowels harmonize in backness to the preceding stem vowel: [haz-nak] `house-Dat' vs. [hez-nek] `hand-Dat'. however, a vowel like [i] is transparent in that the preceding stem vowel defines suffix vowel harmony: [zafir-ban] `sapphire-Inessive' vs. [zefir-ben] `zephyr-Inessive'. Gafos and Benus explain this transparency by arguing that there are gestural differences in these two [i]s brought about by the preceding vowel, but this difference is not perceptual on the [i] itself. Nevertheless, they show using nonlinear dynamical models, that the different articulations of [i] can create a tipping point in which the vowel in a suffix is dependent on that [i], either being realized as a front or back vowel depending on what side of the critical value of tongue backness the [i] takes. At values less than the critical value, the shape of the function demonstrating the stable state shifts to result in back productions, but at values above the critical value, the shape of the function is distinct, resulting in a stable state of tongue frontness for the production. When the value is on the critical value, there are two stable states, resulting in both back and front realizations. Interestingly, in Hungarian, for some words variation is exactly what is observed following certain stems with an [i] vowel, and this [i] shows an intermediate articulation in tongue position compared to [i] following a front or back vowel in the stem.

Ultimately, they show that a perceptually identical [i] can have fine-grained articulatory differences, resulting in the stable state being different. This is interesting for cases of opacity, but only if there is some type of incomplete neutralization. If there really is no measurable difference between two segments, this analysis won't go through.