ICT Lead: Kallirroi Georgila
Dialogue involves coherent interaction among multiple participants, and multiple utterances and turns. Dialogue context includes information required to understand and coherently participate in a dialogue beyond the immediate input (a new sentence or communicative intention). Dialogue embeddings are vector representations for encoding the dialogue context. We propose to develop new methods for generating dialogue embeddings. These new dialogue embeddings will be designed and trained for specific dialogue-related tasks and dialogue types, but we will also investigate to what extent transfer learning can be used to leverage similarities between multiple dialogue-related tasks and dialogue types.