In absence of reliable repair annotations, or in any certainty that the repair categories we're using at the moment are really the useful ones, can we:
- predict adherence numbers directly
- predict dialogue type directly (where this is based on number of NTRIs)
Svennevig 2008 paper re open-class repair initiators being ostensibly addressing hearing problems. Maybe this one?
- Svennevig, Jan 2008. Trying the easiest solution first in other-initiated repair. Journal of Pragmatics 40 (2), 333-348.
NTRI environments seem to have in common:
- contexts of misalignment
- Dr's prior turn/question is "unexpected"
- (can topic models/shifts help predict NTRIs?)
- inappropriate question
- direct/challenging/topically controlling
- NTRIs often used to "resist" Dr's question (design)
- (can question-question sequences help predict NTRIs?)
- Dr questions often biased (expecting + or - answer)
- (can Dr question form help predict NTRIs?)