ArcDeck Framework Uses Discourse Modeling to Generate Coherent Academic Presentations
ArcDeck, a multi-agent framework, treats paper-to-slide creation as narrative reconstruction to better preserve logical structures from academic texts.
According to the primary source, ArcDeck builds a discourse tree and a global commitment document prior to deploying iterative agent-based refinement, contrasting with direct summarization approaches that fail to maintain high-level intent. This method significantly improves narrative flow as measured on the newly introduced ArcBench.
Synthesizing findings with multi-agent systems research from AutoGen (arXiv:2308.08155), the role-specific agent coordination in ArcDeck addresses limitations in earlier single-model slide generators, which often produced disjointed decks missing the paper's argumentative arc. The original arXiv submission overlooks how such tools might integrate with existing LaTeX-based workflows common in many fields.
While the approach could save researchers substantial time in communicating complex ideas, broader patterns in AI-assisted academic tools suggest the need for human oversight to prevent propagation of subtle errors, a nuance not fully explored in the source material but critical for adoption.
NarrativeAgent: ArcDeck first maps a paper's discourse tree and commitments before agent refinement, producing slides that follow the original logical story instead of fragmented summaries, which could save researchers days per project while improving audience comprehension.
Sources (3)
- [1]Narrative-Driven Paper-to-Slide Generation via ArcDeck(https://arxiv.org/abs/2604.11969)
- [2]AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation(https://arxiv.org/abs/2308.08155)
- [3]ReAct: Synergizing Reasoning and Acting in Language Models(https://arxiv.org/abs/2210.03629)