Nous accueillons à partir du 22 mai 2018 le professeur Louis Goldstein (University of Southern California – USA) pour une série de quatre séminaires sur le thème « Recent Developments in Articulatory Phonology »:

Ces séminaires se dérouleront les mardis 22 et 29 mai, 05 et 12 juin de 17h00 à 19h00 à l’ILPGA – 19 rue des Bernardins – 75005 Paris – Salle Rousselot.

Résumé des séminaires :

Articulatory Phonology (AP) is a theoretical approach to phonological and phonetic representation first developed in the mid 80s and early 90s by Browman and Goldstein [1]. Its fundamental units are gestures, simple actions of the vocal tract articulators that are modeled as dynamical systems that achieve some goal state of the vocal tract. Over the last 10 years, the original (“classical”) theory and the task dynamics model that instantiates it (TaDA) have been substantially developed in new directions that have not yet been widely disseminated. The seminar will cover these developments, with attention to relevant phonological data.

Week 1: Introduction.
A tutorial introduction to classic AP will be presented, covering dynamical systems, synergies, gestures, and TaDA, the task dynamic simulation system that can be used to model AP. Key points include (1) how dynamical systems and task-oriented synergies can resolve the apparent incompatibilities between phonological and phonetic descriptions of speech, including discreteness vs. continuity and context-independence vs. context- dependence, (2) evidence for vocal tract constrictions as goals of dynamical phonological primitives (or gestures), including very recent evidence from electrocorticography in the sensorimotor cortex [2], and (3) how gestural scores constitute a description of how the discrete phonological primitives of an utterance are deployed over time in an overlapping fashion, and how they, much like phonetic transcriptions, can be force-aligned to an audio signal.

Week 2: Sensorimotor Representation and Speech Production.
Classic AP and task dynamics hypothesized that the goals of gestures (phonological units) were all defined with respect to constriction formation within the vocal tract, rather than its auditory (and visual) consequences, and the production of gestures was modeled without use of auditory feedback. Over the last 15 years, much has been learned about the role of sensory feedback (both actual and generated by an internal “forward model”) in speech production, though behavioral and neural experiments using altered feedback. These findings, which will be reviewed in this lecture, serve to sharpen the concept of gesture as a coordinated unit of articulatory action that achieves a goal, where the goal can be defined in a variety of different possible reference frames (e.g., auditory, acoustic, areodynamic, geometric). Variation of goal as a function of gesture type and possibly as a function of language will be presented, as will a new version of TaDA that can adapt to altered auditory feedback, as speakers do. More generally, recent studies using electrocorticography have revealed cortical maps of phonetic units in motor and sensory areas during both speech perception and production that show substantial topological differences. While this appears to leave auditory and motor representations untethered to one another, a new representation will be introduced, the Spatiotemporal Modulation Signature, which shows closely correlated temporal patterns of articulatory and auditory/acoustic changes over the course of an utterance, and can serve as a basis for sensorimotor integration.

Week 3: Dynamics of Intergestural Timing.
Careful reflection suggests that the relative timing among the gestures of an utterance is intrinsically dependent on the utterance’s prosodic organization (including both syllable structure and length contrasts as well as higher levels of prosody), but is to a first approximation independent of the specific consonant and vowel gestures that are deployed and their dynamics. Recent experimental evidence for this dissociation will be presented, along with parallel evidence from birdsong. Within AP, a model of intergestural timing has been developed over the last 15 years [5] in which the relative time at which gestures are triggered is predicted by a network of coupled planning clocks (oscillators) each of which is responsible for an individual gesture. The graph represents syllable structure by means of modes of coupling: onset consonants are coupled in-phase to the nucleus gesture, while coda consonants are coupled anti-phase to the nucleus gesture. Coupling between gestures of adjacent syllables is also represented. Two issues that have not been previously probed will be discussed: (1) Predicting language variation in relative timing (and overlap) of heterosyllabic consonant sequences, as well variability in timing, as a function of the topology of the language-specific coupling graph (the graph not only predicts timing, but also variability). (2) Integrating sonority with the coupled oscillator model. The coupled oscillator model is blind to the properties of the individual consonant and vowel gestures (such as sonority), but sonority lies at the heart of most models of syllable structure, and evidence (to be reviewed) shows that sonority does modulate the patterns of relative gesture timing.

Week 4: Dynamics of Qualitative and Quantitative Phonological Processes.
It is often found in languages that some phonological process is exhibited as a categorical alternation in some limited set of contexts, and in the contexts in which it does not occur, gradient, quantitive variability is found that mimics the categorical alternation (to a reduced extent). For example in contemporary Seoul Korean [4], tense vs. lax consonants are associated with categorically distinct values of f0 (high vs. low tone) in the initial syllable of an Accentual Phrase. Internal to the Accentual Phrase, however, this tonal contrast (bimodal distribution of f0) is not found, but nonetheless, vowels following tense consonants are produced with a slightly higher f0 than those following lax consonants. The analytical challenge is to model these effects as arising from the same source in the grammar, and one solution is to model the alternations using a dynamical law to govern the selection of values of the relevant phonetic variable (i.e., f0 in the above example). The dynamical law must have two discrete attractors, corresponding to exhibited categorically distinct values. The tilted anharmonic oscillator was introduced by [3] as having the relevant properties, and they employ it to model Hungarian vowel harmony. The dynamical model will be presented as will application to three phonological alternations: tonal association and alignment in Tashlhiyt Berber, stress in Tagalog, and tense-lax tonal alternations in the Seoul Korean Accentual Phrase.


[1] Browman, C.P. & Goldstein, L. (1992). Articulatory phonology: an overview. Phonetica 49 : 155–180.
[2] Chartier, J., Anumanchipalli, G.K., Johnson, K. & Chang, E. (in press). Encoding of articulatory kinematic trajectories in human speech sensorimotor cortex. Neuron.
[3] Gafos, A. & Benus, S. (2006). Dynamics of phonological cognition. Cognitive Science, 30:905-943.
[4] Lee, Y. & Goldstein, L (2016). Global and local interaction of consonant type and tone in the Korean Accentual Phrase. Poster presented at the 24th Japanese/Korean Linguistics Conference in Tokyo, Japan.
[5] Nam, H., Goldstein, L., & Saltzman, E. (2009). Self-organization of syllable structure: a coupled oscillator model. In F. Pellegrino, E. Marisco, & I. Chitoran, (Eds). Approaches to phonological complexity. Berlin/New York: Mouton de Gruyter. pp. 299-328.