Discourse Segment Annotation and Analysis: A Tagalog Telephone Corpus

Each discourse segment represents a conversational activity identified by the annotators. The conversations were presented to annotators as audio recordings accompanied by transcripts.

Every utterance in the transcripts is labeled by five to nine annotators. Two probabilistic models to infer a single ground truth label from the multiple labels were compared. The dataset contains the conversational data and annotation task description, the annotation results, code to apply each of the probabilistic models to the annotations, the output of the code, and a paper that describes the dataset and modeling.