Over the past several years, there as been a growing interest in extracting and summarising information from meetings. One type of information of particular importance to meeting summaries is information about when someone agrees (or disagrees) with someone else.
In our research in the field of Automatic Agreement Detection, we were especially interested if it is possible to label segments of the transcript as being an agreement or disagreement to what someone else said beforehand. Using supervised machine learning techniques, we explored a variety of multimodal features (e.g., lexical, prosodic, dynamic or structural).
One novel task in the research of automatic agreement detection is the detection of the speaker who is the target of the (dis-)agreement. For this task, we explored different sources of knowledge and found that the adjacency pair annotations overlap in a significant amount with the target annotation.