Analyzing Verbal and Nonverbal Features for Predicting Group Performance

Gabriel Murray; McKenzie Braley; Uliyana Kubasova

arxiv: 1907.01369 · v2 · pith:QIAEW3OBnew · submitted 2019-06-26 · 📡 eess.AS · cs.CL· cs.LG· cs.SD

Analyzing Verbal and Nonverbal Features for Predicting Group Performance

Uliyana Kubasova , Gabriel Murray , McKenzie Braley This is my paper

Pith reviewed 2026-05-25 14:53 UTC · model grok-4.3

classification 📡 eess.AS cs.CLcs.LGcs.SD

keywords group performance predictionverbal featuresnonverbal featuresspeech signal analysissurvival task datasetsocial signal processingautomatic predictiongroup conversation

0 comments

The pith

Nonverbal features from the speech signal alone predict group task performance effectively, though the strongest single features are verbal.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether verbal and nonverbal cues in group conversations can automatically forecast how well a group will perform on a shared task. It releases a new annotated survival-task dataset and merges it with an existing one to run comparisons on a larger collection of recordings than prior studies. Nonverbal measures drawn directly from the audio waveform prove highly useful for the prediction task even when no words are transcribed or analyzed. At the same time, when features are ranked one by one, certain verbal measures outperform all others. If these patterns hold, automatic systems could assess collaborative outcomes from audio alone or by focusing on a small set of language-based indicators.

Core claim

Merging the new survival-task recordings with prior data shows that nonverbal features extracted from the speech signal achieve strong predictive performance for group outcomes on their own. Individual verbal features nevertheless rank highest in effectiveness, and the study identifies which verbal measures contribute most to accurate prediction.

What carries the argument

The direct comparison of verbal feature sets (lexical and turn-based) against nonverbal feature sets (prosodic and acoustic) on merged survival-task corpora for regression or classification of group performance scores.

If this is right

Audio-only systems can forecast group success without requiring speech recognition or text transcription.
A small number of verbal features can be prioritized when computational resources are limited.
Merging multiple survival-task collections produces more reliable feature rankings than single-dataset experiments.
Prediction models benefit from treating verbal and nonverbal streams as complementary rather than competing sources.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Real-time group monitoring could operate with lower privacy cost by relying mainly on acoustic measurements.
The same feature comparison could be tested on other collaborative tasks such as software design or medical team decisions to check generality.
If nonverbal features remain robust, they might serve as a baseline for detecting performance changes in noisy or multilingual settings where verbal analysis is harder.

Load-bearing premise

The survival-task conversations collected and annotated represent typical group performance without systematic bias introduced by recording conditions, transcription, or feature extraction methods.

What would settle it

A replication on fresh group recordings in which nonverbal acoustic features fail to exceed chance-level prediction accuracy while verbal features succeed would falsify the reported effectiveness ordering.

Figures

Figures reproduced from arXiv: 1907.01369 by Gabriel Murray, McKenzie Braley, Uliyana Kubasova.

**Figure 2.** Figure 2: GAP and ELEA Group Scores After Scaling Extra Trees (ET). All models were trained using the ScikitLearn Python package. We used the default training parameters except for the number of estimators, which was set at 50. Due to the small amount of data, we employ 5-fold crossvalidation. The accuracy of our models is evaluated using mean squared error (MSE). We compare the performance of the three tree-based… view at source ↗

**Figure 1.** Figure 1: GAP and ELEA Group Scores Before Scaling For these experiments, we compare three tree-based regression models: Random Forest (RF), Gradient Boosting (GB) and 5https://sites.google.com/view/gap-corpus/ home [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 6.** Figure 6: shows the most important features in the RF models when only training with nonverbal features. The top five are all F0 and MFCC features. Finally, given the small sample size and a large amount of features, we also tried feature selection in combination with RF regression, as well as reducing the number of features using [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗

**Figure 4.** Figure 4: shows the relationship between the number of filled pauses and the group score, and there is a clear correlation, with a higher number of filled pauses being associated with a lower (i.e. better) group score. While the number of filled pauses will tend to be larger for longer meetings, the filled pause feature is much more important than the separate meeting length feature. One hypothesis is that group me… view at source ↗

**Figure 5.** Figure 5: shows the most important features in the RF models when trained only with verbal features. Two features are psycholinguistic: age-of-acquisition (AOA 0 1) and SUBTL score (subtl1 0 1). The other three are again dependency relations and part-of-speech tags [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗

read the original abstract

This work analyzes the efficacy of verbal and nonverbal features of group conversation for the task of automatic prediction of group task performance. We describe a new publicly available survival task dataset that was collected and annotated to facilitate this prediction task. In these experiments, the new dataset is merged with an existing survival task dataset, allowing us to compare feature sets on a much larger amount of data than has been used in recent related work. This work is also distinct from related research on social signal processing (SSP) in that we compare verbal and nonverbal features, whereas SSP is almost exclusively concerned with nonverbal aspects of social interaction. A key finding is that nonverbal features from the speech signal are extremely effective for this task, even on their own. However, the most effective individual features are verbal features, and we highlight the most important ones.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper releases a new public survival-task dataset, merges it with prior data, and directly compares verbal versus nonverbal features for group performance prediction, finding nonverbal speech signals effective on their own while certain verbal features rank highest individually.

read the letter

The main point is that this work adds a new annotated public dataset of group conversations on survival tasks and combines it with an existing one to run a larger-scale comparison of verbal and nonverbal features for predicting task performance. That merged-data setup and the verbal-nonverbal split are the concrete additions over prior SSP work that stayed nonverbal-only. The finding that nonverbal features from the speech signal already work well, yet the top individual features are verbal, is the kind of direct head-to-head result that can guide follow-up feature selection. Releasing the data is the part most likely to see reuse. The experiments stay within survival tasks, so the claims are scoped rather than general, but the paper does not overstate that scope. The abstract gives the headline results without numbers or baselines, which makes the strength of the evidence hard to judge from the summary alone; the full paper presumably supplies the model details, cross-validation, and feature rankings. No load-bearing circularity or hidden fitting issues appear in the description. This is useful for researchers who need multimodal group-interaction data or who want to test similar feature sets on team-performance tasks. It is not reshaping the broader field, but the dataset and the comparison are solid enough to merit referee time so that methods and statistics can be checked in detail.

Referee Report

0 major / 2 minor

Summary. The manuscript analyzes the efficacy of verbal and nonverbal features extracted from group conversations for automatic prediction of group task performance on survival tasks. It introduces a new publicly available dataset, merges it with an existing one to increase data volume, and compares feature sets, finding that nonverbal features from the speech signal are highly effective even alone while the strongest individual features are verbal.

Significance. If the empirical results hold with proper validation, the work contributes a public dataset and a direct verbal-nonverbal comparison that is uncommon in social signal processing literature, which typically focuses on nonverbal cues. This could support development of multimodal group-performance predictors.

minor comments (2)

[Abstract and Results] The abstract asserts effectiveness of nonverbal features and superiority of certain verbal features but supplies no numerical performance metrics, baselines, or statistical tests; the results section should include these with error bars and cross-validation details to substantiate the central claims.
[Discussion] Dataset representativeness and potential annotation/extraction biases in the verbal-nonverbal comparison are not explicitly tested; a limitations subsection should address whether the survival-task data generalize beyond the collected scenarios.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work, including recognition of the public dataset contribution and the value of the direct verbal-nonverbal feature comparison. The recommendation for minor revision is noted. No specific major comments were provided in the report, so we have no individual points requiring point-by-point rebuttal or revision at this stage.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical study performing feature extraction, annotation, and machine-learning prediction on merged survival-task datasets. No equations, derivations, or first-principles claims appear; the central findings rest on measured performance differences between verbal and nonverbal feature sets rather than any reduction of outputs to fitted inputs or self-citations by construction. The work is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivations, free parameters, or invented entities; the work rests on empirical data collection, feature extraction assumptions, and standard ML evaluation practices.

pith-pipeline@v0.9.0 · 5676 in / 1076 out tokens · 18606 ms · 2026-05-25T14:53:19.619313+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Boosting Team Modeling through Tempo-Relational Representation Learning
cs.LG 2025-07 unverdicted novelty 6.0

A tempo-relational neural architecture jointly models temporal and relational aspects of team interactions to outperform prior approaches on team performance prediction and enable efficient multi-task prediction of te...

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages · cited by 1 Pith paper · 1 internal anchor

[1]

For instance, a virtual meeting as- sistant with this capability might provide feedback to the group to enhance their performance and efﬁciency

Introduction Automated prediction of a group’s performance on a task is use- ful for a number of reasons. For instance, a virtual meeting as- sistant with this capability might provide feedback to the group to enhance their performance and efﬁciency. As a second ex- ample, inspection of the trained model and its predictions could help a manager or team co...

work page
[2]

Our main ﬁnding is that nonverbal features are very predictive of group performance on their own, though the most effective individual features are verbal

or used a much smaller dataset to compare verbal and non- verbal features [8]. Our main ﬁnding is that nonverbal features are very predictive of group performance on their own, though the most effective individual features are verbal. This trend of research is important, especially when consid- ering that modern society has become increasingly complex and...

work page
[3]

Analyzing Verbal and Nonverbal Features for Predicting Group Performance

Related Work Automated prediction of group performance has been of in- creasing interest to the Social Signal Processing (SSP) commu- nity in recent years. SSP is concerned primarily with the non- verbal aspects of social interactions, including gesture, gaze, and prosody (see [13] for a review). For example, Avci and Aran [7] train models to predict grou...

work page internal anchor Pith review Pith/arXiv arXiv 1907
[4]

Features In the following two sections, we describe the nonverbal fea- tures and verbal features, respectively, that were used in this set of experiments. 3.1. Nonverbal Features We extracted a large number of acoustic features from the au- dio recordings in the GAP and ELEA corpora using the openS- MILE toolkit1. Speciﬁcally, we used the openSMILE conﬁgu...

work page 2010
[5]

Experimental Setup In this section, we ﬁrst describe the new dataset that was col- lected and annotated, and then give an overview of the machine learning models and evaluation metrics that are used in these experiments. 4.1. Dataset In our experiments, we use both the GAP [5] and ELEA [20, 6] corpora. The GAP corpus consists of 28 meetings where in each ...

work page
[6]

The best- performing model in terms of MSE is using just nonverbal fea- tures with RF

Results Table 1 shows the MSE scores for all models. The best- performing model in terms of MSE is using just nonverbal fea- tures with RF. According to paired t-tests, this best model gives a marginally signiﬁcant improvement (0.05 <p<0.1) over the mean baseline, and a highly signiﬁcant improvement ( p<0.01) over the worst-performing model (all features ...

work page
[7]

Speciﬁcally, nonverbal features from the speech sig- nals are most effective as a class, although the most effective individual features are verbal

Conclusion In this work, we show that nonverbal and verbal features from a group conversation are predictive of the group’s performance on a task. Speciﬁcally, nonverbal features from the speech sig- nals are most effective as a class, although the most effective individual features are verbal. In contract with previous work that focuses primarily on nonv...

work page 2018
[8]

Linguistic correlates of team performance: Toward a tool for monitoring team func- tioning during space missions,

U. Fischer, L. McDonnell, and J. Orasanu, “Linguistic correlates of team performance: Toward a tool for monitoring team func- tioning during space missions,” Aviation, space, and environmen- tal medicine, vol. 78, no. 5, pp. B86–B95, 2007

work page 2007
[9]

Alignment and task success in spoken dialogue,

D. Reitter and J. D. Moore, “Alignment and task success in spoken dialogue,” Journal of Memory and Language, vol. 76, pp. 29–46, 2014

work page 2014
[10]

Language style matching as a predictor of social dynamics in small groups,

A. L. Gonzales, J. T. Hancock, and J. W. Pennebaker, “Language style matching as a predictor of social dynamics in small groups,” Communication Research, vol. 37, no. 1, pp. 3–19, 2010

work page 2010
[11]

Automated team discourse anno- tation and performance prediction using lsa,

M. J. Martin and P. W. Foltz, “Automated team discourse anno- tation and performance prediction using lsa,” in Proceedings of HLT-NAACL 2004: Short Papers. Association for Computational Linguistics, 2004, pp. 97–100

work page 2004
[12]

The Group Affect and Performance (GAP) corpus,

M. Braley and G. Murray, “The Group Affect and Performance (GAP) corpus,” inProceedings of the Group Interaction Frontiers in Technology. ACM, 2018, p. 2

work page 2018
[13]

A nonverbal behavior approach to identify emergent leaders in small groups,

D. Sanchez-Cortes, O. Aran, M. S. Mast, and D. Gatica-Perez, “A nonverbal behavior approach to identify emergent leaders in small groups,” IEEE Transactions on Multimedia, vol. 14, no. 3, pp. 816–832, 2012

work page 2012
[14]

Predicting the performance in decision- making tasks: From individual cues to group interaction,

U. Avci and O. Aran, “Predicting the performance in decision- making tasks: From individual cues to group interaction,” IEEE Transactions on Multimedia, vol. 18, no. 4, pp. 643–658, 2016

work page 2016
[15]

Predicting group performance in task- based interaction,

G. Murray and C. Oertel, “Predicting group performance in task- based interaction,” in Proceedings of ICMI 2018, Boulder, USA , 2018, pp. 14–20

work page 2018
[16]

Improving teamwork using real-time language feedback,

Y . R. Tausczik and J. W. Pennebaker, “Improving teamwork using real-time language feedback,” inProceedings of the SIGCHI Con- ference on Human Factors in Computing Systems. ACM, 2013, pp. 459–468

work page 2013
[17]

Automated language-based feedback for teamwork behaviors,

G. Leshed, “Automated language-based feedback for teamwork behaviors,” Ph.D. dissertation, Cornell University, 2009

work page 2009
[18]

Social visualization and negotiation: effects of feedback conﬁguration and status,

M. Nowak, J. Kim, N. W. Kim, and C. Nass, “Social visualization and negotiation: effects of feedback conﬁguration and status,” in Proceedings of the ACM 2012 conference on Computer Supported Cooperative Work. ACM, 2012, pp. 1081–1090

work page 2012
[19]

The impact of increased awareness while face-to-face,

J. M. DiMicco, K. J. Hollenbach, A. Pandolfo, and W. Bender, “The impact of increased awareness while face-to-face,” Human- Computer Interaction, vol. 22, no. 1, pp. 47–96, 2007

work page 2007
[20]

Bridging the gap between social animal and unsocial machine: A survey of social signal process- ing,

A. Vinciarelli, M. Pantic, D. Heylen, C. Pelachaud, I. Poggi, F. D’Errico, and M. Schroeder, “Bridging the gap between social animal and unsocial machine: A survey of social signal process- ing,” IEEE Transactions on Affective Computing, vol. 3, no. 1, pp. 69–87, 2012

work page 2012
[21]

Model- ing dyadic and group impressions with intermodal and interperson features,

S. Okada, L. S. Nguyen, O. Aran, and D. Gatica-Perez, “Model- ing dyadic and group impressions with intermodal and interperson features,” ACM Transactions on Multimedia Computing, Commu- nications, and Applications (TOMM), vol. 15, no. 1s, p. 13, 2019

work page 2019
[22]

Detecting emergent leader in a meeting environment using nonverbal visual features only,

C. Beyan, N. Carissimi, F. Capozzi, S. Vascon, M. Bustreo, A. Pierro, C. Becchio, and V . Murino, “Detecting emergent leader in a meeting environment using nonverbal visual features only,” in Proceedings of the 18th ACM International Conference on Multi- modal Interaction. ACM, 2016, pp. 317–324

work page 2016
[23]

Unleashing the killer corpus: experiences in creating the multi-everything ami meeting corpus,

J. Carletta, “Unleashing the killer corpus: experiences in creating the multi-everything ami meeting corpus,” Language Resources and Evaluation, vol. 41, no. 2, pp. 181–190, 2007

work page 2007
[24]

Predicting group satisfaction in meeting discussions,

C. Lai and G. Murray, “Predicting group satisfaction in meeting discussions,” in Proceedings of the Workshop on Modeling Cog- nitive Processes from Multimodal Data. ACM, 2018, p. 1

work page 2018
[25]

Moving beyond ku ˇcera and francis: A critical evaluation of current word frequency norms and the in- troduction of a new and improved word frequency measure for american english,

M. Brysbaert and B. New, “Moving beyond ku ˇcera and francis: A critical evaluation of current word frequency norms and the in- troduction of a new and improved word frequency measure for american english,” Behavior research methods, vol. 41, no. 4, pp. 977–990, 2009

work page 2009
[26]

Lexicon-based methods for sentiment analysis,

M. Taboada, J. Brooke, M. Toﬁloski, K. V oll, and M. Stede, “Lexicon-based methods for sentiment analysis,” Computational linguistics, vol. 37, no. 2, pp. 267–307, 2011

work page 2011
[27]

An audio vi- sual corpus for emergent leader analysis,

D. Sanchez-Cortes, O. Aran, and D. Gatica-Perez, “An audio vi- sual corpus for emergent leader analysis,” in Workshop on multi- modal corpora for machine learning: taking stock and road map- ping the future, ICMI-MLMI, 2011

work page 2011

[1] [1]

For instance, a virtual meeting as- sistant with this capability might provide feedback to the group to enhance their performance and efﬁciency

Introduction Automated prediction of a group’s performance on a task is use- ful for a number of reasons. For instance, a virtual meeting as- sistant with this capability might provide feedback to the group to enhance their performance and efﬁciency. As a second ex- ample, inspection of the trained model and its predictions could help a manager or team co...

work page

[2] [2]

Our main ﬁnding is that nonverbal features are very predictive of group performance on their own, though the most effective individual features are verbal

or used a much smaller dataset to compare verbal and non- verbal features [8]. Our main ﬁnding is that nonverbal features are very predictive of group performance on their own, though the most effective individual features are verbal. This trend of research is important, especially when consid- ering that modern society has become increasingly complex and...

work page

[3] [3]

Analyzing Verbal and Nonverbal Features for Predicting Group Performance

Related Work Automated prediction of group performance has been of in- creasing interest to the Social Signal Processing (SSP) commu- nity in recent years. SSP is concerned primarily with the non- verbal aspects of social interactions, including gesture, gaze, and prosody (see [13] for a review). For example, Avci and Aran [7] train models to predict grou...

work page internal anchor Pith review Pith/arXiv arXiv 1907

[4] [4]

Features In the following two sections, we describe the nonverbal fea- tures and verbal features, respectively, that were used in this set of experiments. 3.1. Nonverbal Features We extracted a large number of acoustic features from the au- dio recordings in the GAP and ELEA corpora using the openS- MILE toolkit1. Speciﬁcally, we used the openSMILE conﬁgu...

work page 2010

[5] [5]

Experimental Setup In this section, we ﬁrst describe the new dataset that was col- lected and annotated, and then give an overview of the machine learning models and evaluation metrics that are used in these experiments. 4.1. Dataset In our experiments, we use both the GAP [5] and ELEA [20, 6] corpora. The GAP corpus consists of 28 meetings where in each ...

work page

[6] [6]

The best- performing model in terms of MSE is using just nonverbal fea- tures with RF

Results Table 1 shows the MSE scores for all models. The best- performing model in terms of MSE is using just nonverbal fea- tures with RF. According to paired t-tests, this best model gives a marginally signiﬁcant improvement (0.05 <p<0.1) over the mean baseline, and a highly signiﬁcant improvement ( p<0.01) over the worst-performing model (all features ...

work page

[7] [7]

Speciﬁcally, nonverbal features from the speech sig- nals are most effective as a class, although the most effective individual features are verbal

Conclusion In this work, we show that nonverbal and verbal features from a group conversation are predictive of the group’s performance on a task. Speciﬁcally, nonverbal features from the speech sig- nals are most effective as a class, although the most effective individual features are verbal. In contract with previous work that focuses primarily on nonv...

work page 2018

[8] [8]

Linguistic correlates of team performance: Toward a tool for monitoring team func- tioning during space missions,

U. Fischer, L. McDonnell, and J. Orasanu, “Linguistic correlates of team performance: Toward a tool for monitoring team func- tioning during space missions,” Aviation, space, and environmen- tal medicine, vol. 78, no. 5, pp. B86–B95, 2007

work page 2007

[9] [9]

Alignment and task success in spoken dialogue,

D. Reitter and J. D. Moore, “Alignment and task success in spoken dialogue,” Journal of Memory and Language, vol. 76, pp. 29–46, 2014

work page 2014

[10] [10]

Language style matching as a predictor of social dynamics in small groups,

A. L. Gonzales, J. T. Hancock, and J. W. Pennebaker, “Language style matching as a predictor of social dynamics in small groups,” Communication Research, vol. 37, no. 1, pp. 3–19, 2010

work page 2010

[11] [11]

Automated team discourse anno- tation and performance prediction using lsa,

M. J. Martin and P. W. Foltz, “Automated team discourse anno- tation and performance prediction using lsa,” in Proceedings of HLT-NAACL 2004: Short Papers. Association for Computational Linguistics, 2004, pp. 97–100

work page 2004

[12] [12]

The Group Affect and Performance (GAP) corpus,

M. Braley and G. Murray, “The Group Affect and Performance (GAP) corpus,” inProceedings of the Group Interaction Frontiers in Technology. ACM, 2018, p. 2

work page 2018

[13] [13]

A nonverbal behavior approach to identify emergent leaders in small groups,

D. Sanchez-Cortes, O. Aran, M. S. Mast, and D. Gatica-Perez, “A nonverbal behavior approach to identify emergent leaders in small groups,” IEEE Transactions on Multimedia, vol. 14, no. 3, pp. 816–832, 2012

work page 2012

[14] [14]

Predicting the performance in decision- making tasks: From individual cues to group interaction,

U. Avci and O. Aran, “Predicting the performance in decision- making tasks: From individual cues to group interaction,” IEEE Transactions on Multimedia, vol. 18, no. 4, pp. 643–658, 2016

work page 2016

[15] [15]

Predicting group performance in task- based interaction,

G. Murray and C. Oertel, “Predicting group performance in task- based interaction,” in Proceedings of ICMI 2018, Boulder, USA , 2018, pp. 14–20

work page 2018

[16] [16]

Improving teamwork using real-time language feedback,

Y . R. Tausczik and J. W. Pennebaker, “Improving teamwork using real-time language feedback,” inProceedings of the SIGCHI Con- ference on Human Factors in Computing Systems. ACM, 2013, pp. 459–468

work page 2013

[17] [17]

Automated language-based feedback for teamwork behaviors,

G. Leshed, “Automated language-based feedback for teamwork behaviors,” Ph.D. dissertation, Cornell University, 2009

work page 2009

[18] [18]

Social visualization and negotiation: effects of feedback conﬁguration and status,

M. Nowak, J. Kim, N. W. Kim, and C. Nass, “Social visualization and negotiation: effects of feedback conﬁguration and status,” in Proceedings of the ACM 2012 conference on Computer Supported Cooperative Work. ACM, 2012, pp. 1081–1090

work page 2012

[19] [19]

The impact of increased awareness while face-to-face,

J. M. DiMicco, K. J. Hollenbach, A. Pandolfo, and W. Bender, “The impact of increased awareness while face-to-face,” Human- Computer Interaction, vol. 22, no. 1, pp. 47–96, 2007

work page 2007

[20] [20]

Bridging the gap between social animal and unsocial machine: A survey of social signal process- ing,

A. Vinciarelli, M. Pantic, D. Heylen, C. Pelachaud, I. Poggi, F. D’Errico, and M. Schroeder, “Bridging the gap between social animal and unsocial machine: A survey of social signal process- ing,” IEEE Transactions on Affective Computing, vol. 3, no. 1, pp. 69–87, 2012

work page 2012

[21] [21]

Model- ing dyadic and group impressions with intermodal and interperson features,

S. Okada, L. S. Nguyen, O. Aran, and D. Gatica-Perez, “Model- ing dyadic and group impressions with intermodal and interperson features,” ACM Transactions on Multimedia Computing, Commu- nications, and Applications (TOMM), vol. 15, no. 1s, p. 13, 2019

work page 2019

[22] [22]

Detecting emergent leader in a meeting environment using nonverbal visual features only,

C. Beyan, N. Carissimi, F. Capozzi, S. Vascon, M. Bustreo, A. Pierro, C. Becchio, and V . Murino, “Detecting emergent leader in a meeting environment using nonverbal visual features only,” in Proceedings of the 18th ACM International Conference on Multi- modal Interaction. ACM, 2016, pp. 317–324

work page 2016

[23] [23]

Unleashing the killer corpus: experiences in creating the multi-everything ami meeting corpus,

J. Carletta, “Unleashing the killer corpus: experiences in creating the multi-everything ami meeting corpus,” Language Resources and Evaluation, vol. 41, no. 2, pp. 181–190, 2007

work page 2007

[24] [24]

Predicting group satisfaction in meeting discussions,

C. Lai and G. Murray, “Predicting group satisfaction in meeting discussions,” in Proceedings of the Workshop on Modeling Cog- nitive Processes from Multimodal Data. ACM, 2018, p. 1

work page 2018

[25] [25]

Moving beyond ku ˇcera and francis: A critical evaluation of current word frequency norms and the in- troduction of a new and improved word frequency measure for american english,

M. Brysbaert and B. New, “Moving beyond ku ˇcera and francis: A critical evaluation of current word frequency norms and the in- troduction of a new and improved word frequency measure for american english,” Behavior research methods, vol. 41, no. 4, pp. 977–990, 2009

work page 2009

[26] [26]

Lexicon-based methods for sentiment analysis,

M. Taboada, J. Brooke, M. Toﬁloski, K. V oll, and M. Stede, “Lexicon-based methods for sentiment analysis,” Computational linguistics, vol. 37, no. 2, pp. 267–307, 2011

work page 2011

[27] [27]

An audio vi- sual corpus for emergent leader analysis,

D. Sanchez-Cortes, O. Aran, and D. Gatica-Perez, “An audio vi- sual corpus for emergent leader analysis,” in Workshop on multi- modal corpora for machine learning: taking stock and road map- ping the future, ICMI-MLMI, 2011

work page 2011