Grant: Discovery Horizons

Learning a second language (L2) entails learning new articulatory patterns of the lips, jaw and tongue, or modifying pre-existing ones to produce speech sounds specific to the target language. Ultrasound imaging is as a useful tool to observe and characterize tongue shape and motion during speech, and to develop a better understanding of how new articulatory patterns are developed by learners as they acquire an L2. However, characterizing patterns in the deformation of tongue shape (as opposed to static tongue shapes) develop over time requires methodological tools that do not yet exist. While shape analysis is a well-developed area of medical image analysis, there are no tools for analyzing non-periodic motion and deformation patterns that occur at short time scales (e.g., during speech) as they evolve over longer time scales (e.g., as a learner acquires an L2). Our objective is to develop the computational tools needed to characterize the development and evolution of new articulatory patterns in different groups of L2 learners, some of whom will have access to ultrasound-based articulatory biofeedback training. We will devise new statistical descriptors of tongue motion patterns in the space of linear dynamic systems (called the Grassmann manifold), and use these descriptors to investigate the evolution of tongue motion patterns and its relationship to production of the target L2 speech. Our experimental design will rely strongly on knowledge and approaches from the disciplines of computer science (i.e., medical image analysis), linguistics (specifically phonetics and phonology), and L2 pedagogy. As such, an interdisciplinary approach will be used in the design of the new computational tools in order to provide motion pattern descriptors that are linguistically interpretable. Our team is comprised of researchers in medical image analysis (C. Laporte), phonetics (L. Ménard), and phonology and L2 pedagogy (W. Cardoso). We will train two postdocs, two PhD students, 2 Master's students and 4 undergraduate students in the disciplines of engineering, linguistics and language education who will be jointly supervised by the co-applicants. The training approach will be collaborative interdisciplinary, with engineering trainees gaining knowledge in articulatory phonetics, and linguistics and language education trainees gaining skills in programming and signal processing in order to provide holistic answers to the proposed research questions.