Research map / Probabilistic and scientific modeling / Molecules and proteins
Probabilistic Modeling of Interpersonal Coordination Processes
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a latent probabilistic (dynamic Bayes net) model that defines interpersonal coordination as the degree to which teammates' latent state distributions are blended over time, and shows on synthetic and real search-and-rescue team data that inferred coordination predicts team score in the second mission.
Problem
Interpersonal coordination lacks a precise computational definition; existing measures (cross-correlation, entropy, CRQA, latent change scores) conflate synchrony from shared causes with causal influence and struggle with multiple modalities and time scales.
Approach
Model coordination as a latent continuous variable C in [0,1] that controls how much each person's latent state at time t is drawn from a partner's previous-state distribution versus their own, within a dynamic Bayes net that handles multiple modalities at different time scales; infer C with NUTS and use posterior predictive error and team-score regression to validate it.
Key results
- On synthetic data, the coordination model gave RMSE 0.11 vs 0.18 for the ablated model on coordinated data, with no advantage on random data (1.04 vs 1.05).
- On real data, M_voc+link reduced vocalic forecast RMSE on ASIST from 1.59 (no coordination) to 1.36.
- Coordination predicted second-mission team score (e.g., ASIST No-Advisor mission B MAE 93 for M_voc vs 113 for M_avg) but not the first mission.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| include a coordination variable (ablation) | M_x baseline on synthetic data generated with coordination (D_c) · Synthetic D_c | RMSE: 0.18 (.004) -> 0.11 (.004) | Helped |
| include a coordination variable (ablation) | M_x baseline on synthetic data without coordination (D_x) · Synthetic D_x | RMSE: 0.05 (.002) -> 0.06 (.003) | Hurt |
| include a coordination variable (ablation) | M_x,real on ASIST vocalic data · ASIST | RMSE: 1.59 (.05) -> 1.45 (.05) | Helped |
| include a coordination variable (ablation) | M_x,real on ToMCAT vocalic data · ToMCAT | RMSE: 1.45 (.04) -> 1.32 (.04) | Helped |
| add semantic links to vocalic modality (ablation) | M_voc · ASIST | RMSE: 1.45 (.05) -> 1.36 (.03) | Helped |
| add semantic links to vocalic modality (ablation) | M_voc · ToMCAT | RMSE: 1.32 (.04) -> 1.34 (.07) | Hurt |
| use coordination-derived features instead of the average-score baseline | M_voc+link · ASIST Combined, mission B | MAE: 143 (17) -> 105 (14) | Helped |
| use coordination-derived features instead of the average-score baseline | M_voc · ToMCAT Advisor, mission B | MAE: 104 (22) -> 74 (17) | Helped |
| use coordination-derived features instead of the average-score baseline | coordination models · Mission A | MAE | Mixed |
| use semantic links without vocalics (ablation) | M_voc+link · ASIST, Advisor, mission B | MAE: 101 (20) -> 192 (27) | Hurt |
| combine vocalic modality and semantic links instead of using vocalics alone (ablation) | M_voc · ASIST, Advisor, mission B | MAE: 125 (26) -> 101 (20) | Helped |
| combine vocalic modality and semantic links instead of using vocalics alone (ablation) | M_voc · ToMCAT, Advisor, mission B | MAE: 74 (17) -> 106 (18) | Hurt |
Surprising findings
- Using only the semantic modality (M_link) did not improve team score prediction over baselines, while combining it with vocalics sometimes did.
- Coordination was predictive of team score in the second mission but not the first, which the authors attribute to teams still learning the game.
Limitations
- Did not study robustness to spurious coordination from common causes.
- Did not explore non-linear dependencies between coordination and team score.
- Semantic link extraction is simplistic and was not validated against human coders.
Experimental setup
Datasets: ASIST Study 3 dataset ToMCAT dataset
Benchmarks: ASIST Study 3 ToMCAT
Metrics: RMSE of posterior predictive vocalic forecasts mean absolute error of final team score (LOOCV)
Baselines: M_avg M_hyp M_x,real (no-coordination ablation) M_link
Builds on: Butner et al. latent change score model Rational Speech Act models Moulder et al. multi-level vector autoregression
Closest papers by meaning
- LLM-Assisted Semantically Diverse Teammate Generation for Efficient Multi-agent Coordination (ICML 2025 · similarity 0.85)
- Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators (NeurIPS 2024 · similarity 0.84)
- Mental-Models for Multi-Agent Systems (NeurIPS 2026 · similarity 0.84)
- Emergent Coordination in Multi-Agent Language Models (ICLR 2026 · similarity 0.84)
- Optimising Spatial Teamwork Under Uncertainty (AAAI 2025 · similarity 0.84)
- Narrowing the Collaboration Gap, Probably (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.