The Spontal corpus contains 120 half-hour spontaneous dialogues in Swedish, balanced for gender combination and previous acquaintance. From this corpus, over 600 questions have been extracted with a view to investigate and describe prosodic variation and to test the hypothesis that a standard type of question intonation such as a final pitch rise which contrasts to a final low of declarative intonation is not consistent with the pragmatic use of intonation in dialogue . The extracted questions have been labeled with respect to four relatively simple queries, Q1-Q4. Q1 is related to question type, whether the question is best described as a y/n question (Y/N), a wh-question (WH), an alternative question which includes a restricted set of alternative answers (ALT), or other (OTHER). Q2 concerned whether a response was required (REQUIRED), possible (OPTIONAL), or prohibited (PROHIBITED). Q3 was labeled with respect to the question "Does the person asking the question ask for something that has not already been said (FORWARD) or is it more a question of verifying or showing attitude towards what has already been stated (BACKWARD)?". Q4 was to be answered in the positive if the question was a case of reported speech (REPORTED), and in the negative if not (DIRECT). Clustering across the different labels enables the extraction of specific question categories – categories that presumably correspond to different prosodic characteristics. One category of questions that has been relatively underexplored is reported questions. In , for example, reported questions were not coded as questions. Here, we present typological, lexical and prosodic characteristics of reported questions in the Spontal corpus. This set of questions is relatively small (N = 48), and statistic analyses were not very revealing; analyses comparing prosodic variation (duration, average pitch, pitch variation and intonation slope) within speakers did not show any differences dependent on whether questions were reported or not. Instead, our prosodic analysis is based on qualitative descriptions of the reported questions – descriptions that apart from prosodic characteristics (pitch level and variation, speed, intensity) involve aspects of voice quality and articulation. We were surprised to find that prosodic marking was not very frequent in the data; around half of the reported questions were not perceived as being prosodically marked. The reported questions were subcategorized by a) whether the question had actually been produced or not, and b) whether the reported question had been produced by the speaker herself or by someone else. Prosodic marking was found to be more common in reported questions that had actually been produced than in hypothetical/rhetorical questions, and also more common in reported questions that had been produced by someone else than by the speaker herself. Regarding the characteristics of the prosodically marked questions, we found that seemingly opposite features can be used to signal reportedness: e.g. either slower or faster speech, higher or lower pitch, stronger or softer speech. Prosodic difference to the preceding context seems to be the key. Apart from the prosodic signaling of reportedness, we also noted lexical signals (where "bara" [ba], Eng. "like:" is the most common) and semantic/pragmatic cues, where the question does not make any sense unless it is a case of reported speech. Regarding the distribution of question types within the group of reported questions, FORWARD-looking YN and WH questions that REQUIRE an answer are the most common type, just as for the non-reported questions in the Spontal corpus.
Tartu, Estonia, 2012.