Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Prediction of who will be the next speaker and when using gaze behavior in multiparty meetings
Ishii R., Otsuka K., Kumano S., Yamato J. ACM Transactions on Interactive Intelligent Systems6 (1):1-31,2016.Type:Article
Date Reviewed: Sep 21 2016

The authors study an issue that, as far as I know, has not been studied in the past: predicting the next speaker and the timing of the start of his or her utterance in multiparty meetings.

Several gaze-transition patterns like X or X-L1M-X were analyzed. Here X denotes a situation where a speaker or listener looks at a nonperson (like the floor or ceiling), and L1M denotes when a speaker or listener looks at another listener with mutual gaze.

I was impressed by the detail of this type of sequence analysis. Even if in some analyzed situations the supporting amount of data was significantly less (33 cases) than the average amount of corpus data (of the order of 700 cases), both in quantity and in variety (here 12 meetings were used, of four women seated in a semicircle, each meeting recorded for ten minutes by four cameras, and the gaze sequences manually annotated by three annotators), the overall analysis was detailed and convincing. For the record, the X mentioned above happened 597 times while in turn-keeping and 74 times with turn-changing, and the sequence X-L1M-X happened 72 times while in turn-keeping and only nine times with turn-changing. Facts like these were used in a next-speaker predictor based on support vector machines (SVM) with a sequential optimization algorithm (SOA) that is implemented in the Weka [1] data-mining software.

There are cultural dimensions that affect the results that are not analyzed or quantified. The authors mention but do not analyze some limitations of that type. I believe that an ethnographer/ethnologist, if added to their team, would help explain the nature and importance of specific gestures and timings, and would help explain some unexpected results and the absence of expected ones.

There is some corroboration of known and established sociolinguistic results, a fact that shows that the authors’ research program is on the right track.

There are some timing-related decisions the authors made in their study, like the 200 ms of silence to define the interpausal unit (IPU), which is used without adequate justification. As the authors pointed out, this value cut about 23.4 percent of their data, and they plan to diminish it in their future work. Also, the 1000 ms prior to an IPU that was set in the analysis of gaze behavior and timing structure was later found by the authors to affect the relation between the timing structures of making eye contact and when turn-keeping and turn-changing happen. Some deeper timing analysis is needed here on how to select these values, and the authors are aware of that.

Overall, this work is a very nice first step that is on the right track. I recommend it to professionals and researchers in the field, and am eagerly waiting the authors’ next steps in their research program.

Reviewer:  Constantin S. Chassapis Review #: CR144782 (1612-0901)
1) Bouckaert, R. R.; Frank, E.; Hall, M. A.; Holmes, G.; Pfahringer, B.; Reutemann, P.; Witten, I. A. Weka–experiences with a Java open-source project. Journal of Machine Learning Research 11, (2010), 2533–2541.
Bookmark and Share
 
User/ Machine Systems (H.1.2 )
 
 
Natural Language Processing (I.2.7 )
 
Would you recommend this review?
yes
no
Other reviews under "User/Machine Systems": Date
The future of interactive systems and the emergence of direct manipulation
Shneiderman B.  Human factors and interactive computer systems (, New York,281984. Type: Proceedings
May 1 1985
Reading continuous text from a one-line visual display
Monk A. International Journal of Man-Machine Studies 21(3): 269-277, 1984. Type: Article
Oct 1 1985
Readings on cognitive ergonomics - mind and computers
van der Veer G., Tauber M. (ed), Green T. (ed), Gorny P. (ed)  Readings on cognitive ergonomics - mind and computers,Gmunden, Austria,1984. Type: Whole Proceedings
May 1 1986
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy