Reproduce “Utterance-level Dialogue Understanding: An Empirical Study” 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

Deepanway Ghosal, Navonil Majumder, Rada Mihalcea, Soujanya Poria

An empirical study by Ghosal et al. systematically quantified the impact of conversational context on utterance-level dialogue understanding tasks such as emotion, intent, and dialogue act classification. The research demonstrated that contextual models generally offer superior performance and robustness, although the optimal contextual factors vary by task.

Key insights

  • Contextual models generally demonstrate superior performance and substantially increased robustness against adversarial perturbations compared to non-contextual models.
  • The importance of context length, directionality (past vs. future), and speaker-specific dependencies is highly task- and dataset-dependent, suggesting no one-size-fits-all context modeling approach.
  • Dataset biases, such as strong intra-speaker label-copying tendencies, can significantly influence how models utilize and even benefit from contextual information.

Summary by alphaXiv.

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