クリスティーナ・ヨキネン,岩本秀明,田中英輝
Manual for tagging the SLDB English
dialogues
with speech acts and topic
Abstract:We describe a dialogue tagging project on the bilingual ATR spoken dialogue corpus. This
project is a part of a larger research effort, the goal of which is to promote research and development involving speech translation systems based on multi-level information. The discourse level
information contains dialogue act and topic types. The novel feature in our tagging research is
the use of topic tags, which represent the information content of utterances, and thus complement
dialogue act tags which represent the speakers' intentions. In this report we describe the design
and use of the tag sets for these two features. We also report on a tag browser which we have
developed to check the consistency of the tag sets, and on the preliminary results concerning tag
prediction and bilingual surveys on our corpus.