DeepFon
We investigate the production, comprehension, acquisition and historical development of language (especially its phonetics, phonology and morphology) by combining evidence from language observations and laboratory experiments with computational simulations using artificial neural networks. We are currently inspired by recent methodological developments in artificial intelligence.
Key objectives
- Creating a computational model of phonetic, phonological and morphological production, comprehension, acquisition and historical development.
- The model should entertain multiple levels of representation, probably closer to 20 than to 5.
- The model should be bidirectional, i.e. the same connections should handle production as well as comprehension.
- In the model, phonological and morphological entities should be emergent rather than innately given.
- Once the model works for phonology and morphology, it can generalize to syntax and semantics, thereby providing a new perspective on language as a whole.
Impact
Practically (short-term goal), we are working on improving the inclusivity of automatic speech recognition. Theoretically (long-term goal), we are working toward finding “the” cortical algorithm, thereby potentially solving how the mind works.
Events
- The DeepFon people can join the BiPhon meeting, usually every fourth Friday at 15:00. This meeting is open to other phoneticians and phonologists, as well as to other ACLC and ILLC members interested in computational language emergence.
Members
Paul Boersma (coordinator)
Golshid Shekoufandeh
Titia Benders
Dirk Jan Vet (engineer)
Marianne de Heer Kloots
Charlotte Pouw
Martijn Kingma
David Weenink
and various thesis and tutorial students, both BA and MA.