Bridging linguistic and clinical perspectives through computational models of speech production
Special session 2
Organised by Ben Parrell, Doris Mücke, Antje Mefferd, Sarah Harper
Our understanding of speech production has been greatly informed by the use of computational models, including DIVA (Directions into Velocities of Articulators), TaDA (Task Dynamics Application, combining Task Dynamics and Articulatory Phonology theories), GEPPETO (GEstures shaped by the Physics and by a PErceptually oriented Targets Optimization), ACT (ACTion based model of speech production), and FACTS (Feedback Aware Control of Tasks in Speech). Computational models formalize theories of speech production and allow for the generation of concrete predictions based on these theories that can be experimentally tested against human speech behavior.
Despite the utility of these models, there is a striking divide between models developed and used for the linguistic study of speech and those focused on speech motor control and its associated disorders. For example, the TaDA model has been used extensively to test linguistic predictions of speech behavior derived from the Articulatory Phonology model, including differences in production based on syllable position, effects of gestural overlap, and coordination patterns between individual speech gestures. However, the abstract control and relatively simple execution component of this model makes it difficult to apply to the study of the neural control of speech or the breakdown in control in motor speech disorders. Conversely, the DIVA model has been developed primarily to replicate and explain neural control of speech and has been used to model speech motor disorders such as apraxia of speech and stuttering, with less application to the effects of linguistic factors including the phonological and prosodic system on speech behavior. Unfortunately, the differences between linguistic and motor-theoretic/clinical perspectives on speech in these models reflects a larger divide in the field. Both in training and in practice, linguistically-oriented phoneticians often gloss over the challenges and difficulties in motor planning and execution while clinically focused researchers often overlook effects of linguistic structure in speech production.
This session will bring together researchers from both linguistic and motor-theoretic/clinical traditions to discuss how to better connect the two fields of speech production. A special emphasis will be given to areas where linguistic theory can help inform our understanding of healthy and disordered speech motor control and, vice versa, where the clinical/motor-theoretic perspective can give new insight into linguistic structure and representation. A critical part of bridging this gap is discussion and, as such, this session will have a round-table format with brief presentations from 2-3 researchers from each tradition and substantial time devoted to a moderated panel discussion of these issues. The panel will highlight the strengths and weaknesses of current models of speech production with a focus on how models can help unite linguistic and clinical/motor-theoretic perspectives.
● Fostering dialogue between different areas of research into speech production (linguistic, clinical, and motor-control) that traditionally have had minimal overlap
● Focus on how both empirical research and computational approaches can help unify different theoretical perspectives (advantages of inter-community dialogue to the advancement of both empirical and computational work)
● Opening existing models on speech production for different communities by discussing interfaces and the integration of higher and lower levels of speech production
● Integrating categoriality and continuity in speech production models relevant for clinical and linguistic/motor-control research