Analysis of Discourse Structure and Information Structure using
Questions under Discussion

Arndt Riester
IMS, University of Stuttgart

I present a pilot study demonstrating the use of implicit Questions under Discussion (QUDs; Roberts 1996) for the joint analysis of spoken or written discourse in terms of discourse structure and information structure. Assuming that (rational) speakers / writers follow strategies to break down complex questions / issues into simpler subquestions, the annotator’s task is to recover this strategy by transforming a written text or transcript into a discourse tree in which terminal nodes represent assertions (in linear order) and non-terminal nodes represent (typically implicit) questions. Against Roberts (1996), but in line with theories of discourse structure such as RST (Mann & Thompson 1988) or SDRT (Asher & Lascarides 2003), I assume that a subquestion in the tree need not necessarily stand in an entailment relation with its parent question. However, subquestions must at least be anaphorically dependent on previous material. The benefit of determining discourse structure in terms of QUDs is twofold. On the one hand, it provides us with a tool for identifying parallel structures in text (sequences of partial answers to the same QUD), whose information-structural (and therefore prosodic) relevance has been shown by Büring (2003). On the other hand, implicit questions enable us to determine focus constituents in the ongoing discourse. On the basis of a transcribed section from an English-language TV interview, I give a walkthrough of the current analysis procedure. The prosody (pitch accent placement and phrasing) predicted from the resulting information-structural analysis is evaluated against the actual prosodic realisation.