Reproduce “To Tell The Truth: Language of Deception and Language Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Sanchaita Hazra, Bodhisattwa Prasad Majumder

Researchers introduced T4TEXT, a dataset of conversational deception grounded in objective truth, and developed a bottleneck LLM framework to detect lies solely from text. This framework achieved 39.3% accuracy in identifying the real individual, performing comparably to human judges with multimodal cues, and generated high-quality, explainable predictions based on linguistic signals.

Key insights

  • Sufficient linguistic cues indicative of deception are present in conversational dialogues to enable truth discernment, even in the absence of audio and visual information.
  • Large language models can effectively identify complex language-based deceptive cues and compose them into valid chains of reasoning, outperforming traditional supervised methods.
  • A structured bottleneck framework enhances LLM performance and explainability by explicitly modeling the analytical process, demonstrating superior interpretability compared to end-to-end LLM approaches.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.