1 / 9

Activity-based dialogue analysis as evaluation method

Activity-based dialogue analysis as evaluation method. Bilyana Martinovska Ashish Vaswani Institute Creative Technology University of Southern California. Questions. How can we recognize speech activities? How does the language and media determine and reflect the speech activity?

nuwa
Download Presentation

Activity-based dialogue analysis as evaluation method

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Activity-based dialogue analysis as evaluation method Bilyana Martinovska Ashish Vaswani Institute Creative Technology University of Southern California

  2. Questions • How can we recognize speech activities? • How does the language and media determine and reflect the speech activity? • Which descriptive features are more informative and what story do they tell us about the activity? • Can activity-based dialogue analysis be used to evaluate spoken dialogue systems?

  3. Method • Quantitative activity-based dialogue analysis of radio dialogue during Call for Fire training • Describe the activity: • context: micro and macro • language • pragmatics • Compare to other speech activities • Compare instructive radio talk between humans (HH) and between humans and a machine (HM)

  4. Processing of data • Transcription on Transcriber/Text files • Qualitative analysis of activity • Formulation of a coding scheme • Coding of data • Counting parts of speech with a rules based tagger (Brill 1.14, used in Penn treebank tagging) • Counting moves • Statistics • Measures: liveliness, caution, etc.

  5. Analysis categories • Dialogue Moves • Dialogue Regulators • Grounding Acts • Stress • Pauses • Intonation • Overlap • Own Communication Management • Parts of Speech

  6. Interactive liveliness (2. ToOvPtok) + OvePutt + (2.StressPTok) -(2.PausPTok) – MLU* —————————————————— Total number of tokens (*MLU: mean number of tokens in utterance; ToOvPTok percent of overlapped words as tokens in relation to the total amount of words as tokens in a group; StressPTok percent of stressed words as tokens in relation to the total amount of word- tokens in a group; PauPTok percent of pauses in relation to the total amount of words as tokens in a group)

  7. Interactive caution (2.Pauses) + Stress + (2.OCM) + (2.FB) + Numerals – (2.Pronouns) – (2.Overlap) ————————————————— Total number of tokens

  8. Compare activities: liveliness and caution

  9. Conclusions • CFF radio training is livelier only in comparison to sermons and auctions but more cautious compare to courtroom examinations and dinners. • Question-answer fixed activities appear less lively but more cautious. • The complex measures of liveliness and caution would be improved if information about the duration of pauses is included. • HH interactive training relies on explicit explanations, coordination and meta-grounding acts whereas the HM relies on repetition of pre-recorded actions with not much explicit explanation, less speech management and coordination, but also less modal expression and hesitation sounds. • Training modes are equally efficient, in our sense, but differ in style of instruction and learning.

More Related