Source-linked AI summary
PolyDebate: A Game-Orchestrated Multimodal System for Debate Skills Practice and Evaluation
Jianing Yin, Weng Pan Kuan, Xiaoyun Liu, Zhiyuan Wen, Yuxuan Li, Milos Stojmenovic, Jiannong Cao
TL;DR
AI debate systems often center on argument generation or transcript scoring, leaving complete multimodal learner practice under-supported. PolyDebate combines staged game-like 1v1 debate, a stage-aware AI opponent, multimodal rubric-based evaluation, and structured feedback, with studies reporting skill-aware responses, broader assessment coverage, and a usable, motivating workflow.
Problem
Existing AI debate systems are largely text-centered and rarely provide staged guidance, strategy scaffolding, immediate feedback, or evidence of oral and visual delivery, although debate trains communication skills.
Method
PolyDebate orchestrates staged 1v1 debates with skill cards, props, coins, a stage-aware AI opponent, and rubric-aligned evaluation of transcript, speech, and visual evidence.
Results
PolyDebate produced skill-aware opponent responses, broader-than-text assessment, and a usable, motivating practice loop across its evaluations.
Takeaways & Limitations
PolyDebate provides a practical workflow that integrates debate interaction, gamified scaffolding, multimodal assessment, and structured feedback for debate skills practice.
Takeaways & Limitations
Learner-profile-driven practice, additional debate formats, and classroom deployment remain future work, leaving learning gains to be examined.
Abstract
from arXiv · showhide
Debate is a structured form of persuasive communication that trains argument construction, rebuttal, oral delivery, and audience awareness. These skills are valued in education, language learning, and professional communication. Recent AI debate systems and LLM-based judges have advanced argument generation and debate evaluation, but most remain text-centered and rarely support learners through a complete multimodal practice experience. We introduce PolyDebate, a game-orchestrated multimodal system for English debate practice and evaluation. PolyDebate guides learners through staged one-on-one (1v1) debates with an AI opponent, while skill cards, props, and coins make persuasive strategies explicit and turn practice into a game-like interaction. During each session, the system captures learner speech and visual delivery evidence, generates context-aware opponent responses, and produces rubric-informed stage-level and overall feedback. PolyDebate is available as both an immersive Unity 3D game version and a web platform version that share the same workflow and evaluation services. Four studies covering AI opponent quality, evaluation coverage, AI judge feedback, and user perception show that PolyDebate brings debate interaction, gamified scaffolding, multimodal assessment, and structured feedback together in a practical workflow for debate skills practice. The demonstration video is available at https://youtu.be/mHwBG1_8Ebk.
1 Introduction
PolyDebate addresses the difficulty of scalable, complete debate practice by integrating staged 1v1 interaction, game-like scaffolding, multimodal evidence capture, and rubric-aligned feedback in an English debate system.
- Motivation: Debate practice develops argument construction, evidence use, rebuttal, spoken communication, critical thinking, and audience awareness, but is difficult to provide at scale.Effective practice requires a responsive opponent, a clear procedure, and formative feedback on argument quality and oral delivery.
- Problem: Existing LLM debate systems generate arguments or score transcripts, but generally lack a complete learner-facing practice process with staged guidance and multimodal assessment.The introduction identifies these as two remaining limitations of current AI-supported debate practice.
- PolyDebate: PolyDebate combines a game-orchestrated practice module, a stage-aware AI debater, and a multimodal evaluation workflow for English debate practice and evaluation.The practice module supports complete 1v1 rounds with side assignment, stage control, speaking turns, skill cards, props, and coins; AI responses use motion, side, stage, history, and assigned skill card.
- Contributions: The system integrates staged oral debate, an AI opponent, two interface versions, learner-facing feedback, game support, and rubric-aligned multimodal evidence into one workflow.Transcript, speech/audio, and visual evidence are linked to actionable learning feedback.
2 Related Work
Prior AI debate systems emphasize autonomous argumentation and performance, while debate-learning systems provide narrower forms of staged guidance or scoring. PolyDebate combines a learner-facing 1v1 workflow with skill-card guidance, multimodal feedback, and AI-supported debate practice.
- AI debaters and LLM-based debate systems: AI debate systems use evidence retrieval, multi-agent collaboration, genetic algorithms, adversarial search, and retrieval-augmented memory to improve argumentation and rebuttal.Examples include early autonomous debaters, Agent4Debate, DebateBrawl, and R-Debater.
- AI debaters and LLM-based debate systems: These systems target AI debating performance rather than staged practice guidance, skill-card guidance, or multimodal learner feedback.PolyDebate instead places the AI opponent within a learner-facing 1v1 practice workflow.
- Automated debate evaluation: Automated debate evaluation spans chronological stages and dimensions including argument strength, relevance, organization, style, tone, emotional appeal, factual authenticity, and logical validity.The cited systems illustrate increasingly multidimensional LLM-based and mixed subjective-objective judging approaches.
- AI-supported debate learning: AI-supported debate learning includes prompt-based topic selection, debate flow and difficulty control, simple scoring, role-driven preparation, and guidance through evidence, warrant, and rebuttal stages.These systems usually cover only part of the practice experience.
3 PolyDebate
PolyDebate is a multimodal, game-oriented system that guides learners through staged 1v1 debates with a context-aware AI opponent in Unity 3D and web versions. It combines explicit strategy scaffolding, speech and visual evidence capture, and rubric-aligned feedback in a complete practice loop.
- System workflow: PolyDebate combines staged oral debate, AI opponent interaction, rubric-aligned evaluation, and game mechanics in a fixed four-stage 1v1 practice round.The system supports both immersive Unity 3D practice and browser access through a web platform version.
- Gamified scaffolding: Skill cards assign concrete techniques to both learner and AI debater, while stage guidance and props make abstract debate strategies explicit.Examples include Data-Driven, Chain of Reasoning, Address Opponent, and Emotional Appeal.
- Gamified scaffolding: Coins convert evaluated-turn scores into visible progress, and learners spend cumulative coins on limited props or strategic effects before subsequent stages.These mechanics preserve rubric evaluation while adding learner agency and engagement to repeated practice.
- AI opponent: The AI debater generates stage- and side-consistent responses from the motion, debate sides, current stage, latest learner speech, history, and assigned skill card.This supports distinct functions such as establishing a position, probing the opponent, attacking weak points, and summarizing.
- Multimodal evaluation: The evaluation rubric derives from ELC2012 and covers Analysis, Persuasiveness, Clarity, and Appropriacy using transcript, video, and audio evidence.The categories are weighted 30%, 30%, 25%, and 15%, respectively.
4 Demonstrations
The demonstrations present PolyDebate’s complete 1v1 practice-evaluation loop in an immersive Unity 3D arena and show representative screens from a browser-based platform sharing the same workflow. An accompanying video presents both versions.
- Immersive Unity 3D Demonstration: Figure 2 shows a complete 1v1 session in the immersive Unity 3D arena, from session start and side assignment through final feedback.The loop includes constructive, cross-examination, rebuttal, and closing stages, followed by the final coin outcome and overall feedback.
- Web Platform Demonstration: Figure 3 presents representative screens from the browser-based web platform version, which shares the Unity arena’s workflow.The passage identifies the web platform as a second version of the same practice-evaluation process.
- Demonstration Video: The accompanying demonstration video presents both the immersive Unity 3D and browser-based web platform versions.The video is available at https://youtu.be/mHwBG1_.
5 Experiments and Evaluation
PolyDebate was evaluated across four analyses covering AI opponent quality, multimodal evaluation coverage, AI judge feedback, and user perception. Results indicate that its staged, multimodal, and gameful design supports pedagogically targeted responses, comprehensive diagnosis, and positive experiences across web and Unity versions.
- Evaluation design: The evaluation covered AI opponent response quality, framework-level evaluation coverage, AI judge feedback with ablations, and user perception.These analyses examined both system components and learner-facing experiences.
- AI opponent quality: PolyDebate achieved the highest overall opponent score at 4.0, versus 3.1 for the generic LLM opponent and 3.6 for the stage-only opponent.Scores used a 1–5 scale across anonymized next-turn responses.
- AI opponent quality: Skill-usage scores increased from 2.1 to 3.9, indicating that PolyDebate’s skill cards shaped stage-aware pedagogical responses rather than serving only as interface cues.The comparison was between the generic and PolyDebate opponents.
- Evaluation coverage: Compared with representative frameworks, PolyDebate jointly covered argument content, oral delivery, visual behavior, and learner-facing feedback across text, audio, and video.Its framework adapted the ELC2012 rubric and organized assessment around analysis, persuasiveness, clarity, and appropriacy.
- AI judge feedback: The full AI judge performed best on all reported metrics, while removing the rubric reduced weighted rubric coverage and removing the skill card or multimodal evidence reduced weakness F1.The comparison included a generic baseline and four component ablations using 100 multimodal samples.
- User perception: Both platform versions received positive ratings across six dimensions; the web version led on usability, feedback usefulness, skill-card support, and overall value, while Unity led on gameful motivation.AI opponent usefulness was similar across versions, which the study characterized as complementary user experiences.
6 Conclusion and Future Work
PolyDebate combines staged 1v1 debate interaction, game-like scaffolding, multimodal evidence, rubric-aligned evaluation, and structured feedback across Unity 3D and web versions. Future work will investigate learner-profile-driven practice, additional debate formats, and classroom deployment to examine learning gains.
- Conclusion: PolyDebate connects staged 1v1 interaction, skill cards, props, coins, an AI opponent, multimodal evidence, rubric-aligned evaluation, and structured feedback in Unity 3D and web versions.Both versions share one workflow.
- Future Work: Future work will use prior-round feedback to support learner-profile-driven practice.
- Future Work: Future work will explore team-based and multi-role debate formats and classroom deployment to examine learning gains.