EXPRESSION focuses on studying human language data conveyed by different media: gesture, speech and text. Such data exhibit an intrinsic complexity characterized by the intrication of multidimensional and sequential features. Furthermore, these features may not belong to the same representation levels, basically, some features may be symbolic (e.g., words, phonemes, etc.) whereas others are digital (e.g., positions, angles, sound samples) and sequentiality may result from temporality (e.g., signals).
Within this complexity, human language data embed latent structural patterns on which meaning is constructed and from which expressiveness and communication arise. Apprehending this expressiveness, and more generally variability, in multidimensional time series, sequential data and linguistic structures is the main proposed agenda of EXPRESSION. This main purpose comes to study problems for representing and characterizing heterogeneity, variability and expressivity, especially for pattern identification and categorization.
The proposed research project targets the exploration and (re)characterization of data processing models in three contexts:
- Expressive gesture synthesis and recognition
- Expressive speech synthesis
- Expressiveness in text and language