The contribution of this project lies in automatically identifying relevant linguistic features for question classification. The goal is to expand the previous work to include prosodic features as well as the larger discourse context in the analysis process. Furthermore, the aim is to develop LingVis systems that will allow linguists to interact and explore linguistic cues to understand how the mapping from combined surface cues to meaning is achieved. An active learning model combined with the feedback cycle will be of primary importance. This technique will enable weighting of the individual features and facilitate the adjustment of these features and thus, in turn, the improvement of the trained model. Moreover, by focusing on the explainability of the machine learning techniques, the intent is to understand how the different features interact and which trade-offs contribute to the overall classification decision.
