Grounded Cross-Modal Language Acquisition

by Peter Gorniak (MIT Media Lab)

We build machines which acquire concepts, words and grammar by interacting naturally with people. This involves developing and carefully integrating learning algorithms, speech processors, visual processors, and robots. Furthermore, we develop applications that leverage off these systems to provide multilingual, adaptive and assistive human-computer interfaces. In particular, I will cover:

My group's two central premises are: First, it is not too early to start integrating various modalities into one learning system, because one source of information can be used to structure another, thus making the learning problem easier than it would be in isolation. Second, we strongly believe in providing an intelligent system with semantics derived from its own real experience to provide it with a robust and grounded knowledge representation.

Please see the Cognitive Machines page for more information.

Back to the LCI Forum page