Reverse Engineering Autonomous Language Acquisition

  • Emmanuel Dupoux | Laboratoire de Sciences Cognitives et Psycholinguistique

Speech recognition and understanding technologies rely on supervised learning techniques which typically require tens or hundreds of hours of good quality, human annotated speech in order to train acoustic and language models. In this talk, I argue that it is worthwhile considering an alternative approach, based on unsupervised algorithms, and grounded on the study of human infant language learning. Indeed, during their first years of life, infants spontaneously construct acoustic, language and world models despite large variability in signal quality and amount of parental oversight, across widely different cultures and environments. Reverse engineering this process could therefore enable the development of very robust, flexible and autonomous learning systems as well as enable the modeling and monitoring of normal and impaired language development. I illustrate this approach with results from the recent Zero Ressource Speech Challenge (InterSpeech 2015), and present the Big Baby Data project, a project aimed at constructing a large dataset of parent-infant interactions using kinect sensors.

Speaker Details

Emmanuel Dupoux did his undergraduate in computer science at the Ecole Normale Supérieure in Paris and graduated with a PhD in cognitive science on language processing. After a working as a researcher in France Telecom, he became the head of the CNRS Laboratoire de Sciences Cognitives et Psycholinguistics at the Ecole des Hautes Etudes en Sciences Sociales. His main interest is in the study of language acquisition in the infant and the adult, which he approaches with a variety of experimental, brain imagery and computational modeling tools. He has started a new team devoted to the modeling of early cognitive development using machine learning tools. see www.lscp.net/persons/dupoux and www.syntheticlearning.net

    • Portrait of Jeff Running

      Jeff Running

Series: Microsoft Research Talks