Main Page
This wiki contains a glossary of terms related to the field of explainable AI, which is relevant to the TRR projects. The glossary consists of three categories of explanation moves: “Explainer moves”, “Explainee moves” and “General moves”. General moves are those, which could happen both from the side of the explainee and the explainer. At the moment, the categories assigned to some of the moves are just rough estimates, and they might be classified differently after further refinement. For example, some explainee move could be classified as a general move at a further refinement step.
Looking at the glossary from a different perspective, the explanation moves in the glossary include: adult–child interaction moves (such as embedded correction, and repetitions to place a common ground), short answer questions (such as verification and disjunctive questions), long answer questions (such as procedural, expectational, and assertion questions), backchannels (such as phrasal, substantive, and non-lexical ones), and Other dialog moves (such as Assert, Reassert, Accept, Accept-part, Reject, Signaling emotions, and Open option).
Moreover, the wiki contains guidelines for annotation of data with explanation moves. In fact, the glossary has been enriched with the extracted examples for each explanation move either from a real-world dataset or from literature. This information is therefore considered as a guideline for the human agents, who will annotate Explainer-Explainee dialogues.