Question generation in the CODA project

In the ongoing CODA project, we are developing a system for automatically converting monologue into dialogue. The dialogue is generated in a two-step approach. Firstly, snippets of input monologue are mapped to dialogue act sequences. Secondly, these sequences are verbalized. The conversion relies partly on analysing input monologue in terms of its discourse relations. This short paper briefly describes the approach to the first step in CODA. This approach involves the use of a parallel corpus of monologues and dialogues to learn mappings from monologue to dialogue acts. Here, we focus on dialogue acts that involve question asking.