Research Area - Embodied Language Processing

Embodied Language Processing


In the last decade, new findings and accounts of cognitive models had a major impact and influenced the way we view language technology in human communication. Among the most promising trends, Embodied Language Processing (ELP) embraces a multimodal approach including non-verbal behavioural cues and psychological variables such as attitude and affect. Embodied Language research & technology is, in principle, a multidisciplinary field blending and fertilizing innovations and findings from social psychology, cognitive psychology, neuroscience, computer vision, speech and signal processing, computational linguistics and cognitive linguistics. Pioneering contributions are expected in the coming years, extending the focus on traditional linguistics. Thus, it is clear that new algorithms and technologies are required and computational frameworks need to accommodate the above into more sophisticated models.

Successful analysis and understanding of social interactions would be the key in a number of rich applications, including:

  • future smart environments, Ambient Assistive Living
  • companions, personal assistants and social robots
  • embodied conversational agents (ECAs), avatars
  • applications in Health and Education, Ambient Intelligent applications
  • computational intelligence in games - storytelling
  • human robot interaction
  • multimodal dialogue systems
  • human computer interaction
Research at ILSP focuses on:
  • the integration of cognitive theories, knowledge representation models and algorithms in a computational framework that will enable rich models of human-machine interaction
  • the development of novel machine learning techniques in the analysis of the non-linear and socially situated multimodal streams in human communication

Accomplishments so far:

  • the POETICON data collection: a 12h corpus of 6 everyday scenes capturing language use and sensorimotor experience
  • the Greek Interviews Corpus (an annotated resource of TV interviews)
  • POETICON toolkit: a set of tools that extract information from POETICON annotated corpora facilitating the construction of the conceptual knowledge base PRAXICON
  • POETICON annotated corpora including cognitive, travel and caricature corpora
  • annotation codebooks & multimodal schemes
  • the PRAXICON conceptual system
  • PRAXICON toolkit: a set of tools that try to analyze natural language and establish connections to the conceptual knowledge base, PRAXICON

Current R&D focus:

  • synthesizing sensorimotor representations and semantic representations of natural language in everyday human interaction
  • enrichment of existing multimodal resources
  • analysis of socially situated multimodal streams with machine learning techniques
  • computational multimodal dialogue models
  • multimodal perceptual binding and the temporal perception of correlates within and between different modalities
  • integration of the above in rich cognitive and computational models dealing with human
  • interaction and communication