*** note unusual day, time and venue ***

Friday 5th December 4pm, DSB 3.10

Rick Janssen (MPI Nijmegen)

Anatomical biasing of speech sounds: an empirically grounded agent model

The Darwinian principles of variation, selection and reproduction have had widespread success in explaining the emergence of biological complexity. Similar principles might be at work in other domains. Computer models on cultural evolution of language have explained features such as recursion, compositionality and vowel distributions that mirror real-world counterparts without any invocation of biologically innate language modules.
We draw on past hypotheses that speakers’ anatomy constitutes a biasing factor in cultural evolution. As such, speakers with different vocal tract morphologies can be associated with subtle, yet different costs in producing particular sounds. Larger ethnic groups that share a similar morphology might hereby converge on vowel distributions that differ from others. Our study employs an agent model to investigate such anatomical biases. Notably, our agents are characterized by realistic 3D geometries of the vocal tract, making empirical grounding and subsequent validation feasible.
We hypothesize that the shape of the hard palate is one prime source of anatomical biasing. As such, our study uses Bezier curves to model the hard palate’s mid-sagittal profile. Using five parameters, we constrain the degrees of freedom of our curve. Curve fitting by application of an evolutionary algorithm shows that the standard deviation of a fit to sample is low in every case. As a result, we are now able to automate the integration of e.g., MRI samples into our 3D model.

• «Previous Post
• Next Post »

---