PaperlayerView exact revision on arXiv

Source-linked AI summary

Time-Dynamic Semantic-Activity Entanglement: Quantifying Communication-Based Synchrony in Collaborative VR

Pooja Pol, Jianing Zhang, Qing Huang

arXiv:2610.37427v1cs.HC

TL;DR

Time-varying team communication in collaborative VR is difficult to characterize with aggregate summaries alone. The paper introduces TSAE, combining speaking activity and contextual semantic alignment, and reports broadly distributed entanglement with a descriptive contrast to a core-periphery pattern.

  • Problem

    Aggregate end-state summaries can miss temporal dynamics in collaborative problem solving, motivating time-resolved measures of individual and interpersonal behavior.

    Source: §1 · p. 1

  • Method

    TSAE combines relative speaking activity and contextual semantic alignment for participant pairs, aggregates dyadic scores into participant trajectories, and summarizes their distribution with Gini.

    Source: §1 · p. 1, §2.1 · p. 2

  • Results

    Across sessions, entanglement shares are broadly distributed, providing a descriptive contrast to the core-periphery pattern associated with Gloor et al.’s findings.

    Source: §4 · pp. 4–5

  • Takeaways & Limitations

    TSAE provides a time-dynamic speech-based analytical layer for examining team communication, but does not establish subjective team flow or collaboration quality.

    Source: §4 · p. 5

    Robustness to uncorrected automatic speech-recognition errors remains untested, and generalization may be limited by the German-language dataset and limited demographic information.

    Source: §4 · p. 5

Abstract

from arXiv · show

Collaborative virtual reality (VR) produces rich multimodal traces of team behavior, yet practical methods for characterizing time-varying team communication remain limited. We present time-dynamic semantic-activity entanglement (TSAE), inspired by the entanglement concept introduced by Gloor et al. (2022), to describe communication-based team synchronization within discrete time windows. TSAE is modeled by combining relative speaking activity with contextual semantic alignment for participant pairs, which is aggregated into participant-level trajectories and summarized through the Gini coefficient to understand the distribution of team-level communication synchrony. This approach is employed with German-language transcripts from fifteen collaborative VR gameplay sessions. Transcript utterances are represented using a long-context embedding model with a late-chunking strategy and analyzed within fixed game-time windows. A representative session illustrates shifting dyadic alignment and low-to-moderate participant contributions, with sporadic periods of high entanglement. Across fifteen game sessions, the metrics indicate complementary insights, offering a qualitative contrast to the core-periphery pattern reported in the foundational work. Hence, these descriptive findings support the study of communication as a dynamic process and provide a mathematically grounded approach to analyze its evolving distribution in shared interactive VR spaces.

1 Introduction and Background

Collaborative problem-solving in VR involves changing individual and interpersonal behavior, but aggregate end-state summaries can miss temporal dynamics. The paper adapts communication entanglement to undirected VR dyads and introduces TSAE to examine evolving communication structure.

1 Introduction and Background

  • Aggregate end-state summaries can miss relevant temporal dynamics in collaborative problem solving, where individual and interpersonal behavior changes across phases.Speech offers an accessible starting point for identifying episodes of group alignment among modalities such as gaze and movement.Source: §1 · p. 1
  • Gloor et al.’s communication-volume-based metric studied directional email networks, whereas this work adapts entanglement to undirected dyads because conversational addressees can be ambiguous in free-flowing VR [5].The adaptation also draws on team flow and the notion of being in sync.Source: §1 · p. 1
  • TSAE combines relative speaking activity and the semantic context of team members’ speech as a time series, with dyadic scores aggregated into participant trajectories and summarized by Gini.The authors present this as a descriptive, mathematically grounded method for examining communication in shared interactive spaces.Source: §1 · p. 1

2 Method

The method combines speaking activity and contextual semantic alignment for participant pairs, aggregates pairwise scores into participant trajectories, and uses Gini to summarize their distribution. It is applied to timestamped German-language VR gameplay sessions, with utterances transcribed, verified, and represented using late-chunked embeddings.

2.1 Time-Dynamic Semantic-Activity Entanglement: Formulation

  • Window-level semantic representations use late-chunked utterance embeddings, which the method describes as stable while avoiding biases induced by extrema-based pooling.The method defines pairwise components for activity and semantic comparison between participants within each time window.Source: §2.1 · p. 2
  • Pairwise TSAE requires substantial joint speaking activity and contextual semantic alignment, while greater separation between participant centroids attenuates the score.Adding one to the distance denominator prevents division by zero and makes the distance term an attenuation factor.Source: §2.1 · p. 2
  • Participant-level entanglement averages pairwise values, and the Gini coefficient summarizes the distribution of normalized participant shares.Using the fixed denominator of n−1 possible teammates keeps the participant-level scale consistent across window sizes; higher Gini indicates greater concentration among fewer participants.Source: §2.1 · p. 2

2.2 VR Task and Dataset

  • The study analyzes fifteen German-language sessions, each with one team of five to eight adults completing a collaborative problem-solving task in co-located VR.Sessions average about 25 minutes; player-level 48 kHz mono audio and complete game timestamps were recorded, with demographic and prior-relationship metadata unavailable for privacy reasons.Source: §2.2 · pp. 2–3

2.3 Data Processing and Pipeline

  • Utterances are transcribed with Whisper, manually checked for completeness, aligned to a common game clock, and encoded with Jina v2 using late chunking [13] [6] [11].The approach preserves conversational context for short local utterances; silence or inactivity yields zero activity and zero entanglement.Source: §2.3 · p. 3

3 Results

The representative session shows entanglement shifting across dyads and time, with mostly even participant contributions but occasional concentration. Across fifteen sessions, longer windows smooth transient differences, while Lorenz curves show relatively distributed shares with some variation in concentration.

3.1 Communication-based Entanglement and Gini Coefficients in a Representative Session

  • In one six-participant session, dyad (1, 3) reached E_ij = 0.58 at 225 s and dyad (3, 4) reached E_ij = 0.55 at 540 s, while most Gini values were 0.07–0.27.Gini rose to 0.39 at 540 s and 0.42 in the final window, amid intermittent shifts in entanglement.Source: §3.1 · pp. 3–4

3.2 Cross-Session Analysis

  • Across fifteen sessions, corrected Gini medians decline from approximately 0.25 at 45 s to 0.17 at 90 s and 0.14 at 200 s as longer windows smooth transient differences.Most sessions follow this trend.Source: §3.2 · p. 4, Figure 3 · p. 4
  • The cross-session median Lorenz curve assigns approximately 31%, 54%, and 74% of total entanglement to the top-ranked 20%, 40%, and 60% of participants, respectively.Session-level Gini coefficients range from 0.139 to 0.355, indicating relatively distributed entanglement with variation in concentration.Source: §3.2 · p. 4, Figure 3 · p. 4, Table 1 · p. 4

4 Implications and Discussion

The discussion frames team communication as a temporally varying distribution of activity and semantic alignment, contrasting the observed broadly distributed shares with a previously reported core–periphery pattern. The authors note limits to generalization and emphasize that TSAE does not establish subjective team flow or collaboration quality.

4 Implications and Discussion

  • The observed broadly distributed entanglement offers a descriptive contrast to Gloor et al.’s account of higher performance associated with a core–periphery pattern.Concentration may reflect task-specific contributions and warrants examination alongside ongoing activity.Source: §4 · pp. 4–5
  • Generalization may be limited by the German-language dataset and limited demographic information, while robustness to uncorrected ASR errors remains untested.The transcripts were manually quality-checked.Source: §4 · p. 5
  • TSAE offers a time-dynamic speech-based layer for examining communication, but does not establish subjective team flow or collaboration quality.Comparisons with other communication or behavioral measures, participant reports, and task outcomes could help test hypotheses across CPS tasks.Source: §4 · p. 5

5 Conclusion

TSAE combines speaking activity and semantic alignment to describe communication-based synchronization over time. Across sessions, entanglement shares are relatively evenly distributed with intermittent redistribution, providing a descriptive contrast to a core–periphery pattern.

5 Conclusion

  • TSAE combines speaking activity with semantic similarity to quantify time-varying communication synchrony at dyadic, participant, and team-distribution levels.The Gini coefficient characterizes the team-level distribution; validation against explicit hypotheses is needed to establish what the descriptive patterns signify in CPS.Source: §5 · p. 5
Loading 2610.37427v1…