Evidence-Indexed Slide Narration
Paper at the GroundLM 2026 workshop, co-located with EMNLP 2026, on keeping generated slide narration tied to what the slide actually contains.
Published in the proceedings of the first workshop on Grounding Language Models: Learning Faithfully and Efficiently (GroundLM 2026), co-located with EMNLP 2026.
Slide narration generated by a model can add visual or decorative claims the slide does not support. This pipeline indexes slide content with deterministic identifiers, generates an internally cited template, and applies heuristic checks before and after an optional language-model rewrite. The study covers 30 publicly sourced presentations (279 slides). The paper reports its results as system-level proxy rates, and is careful to say they are not evidence that hallucination is eliminated.
