PhD Proposal: Understanding Implicit Content in NLP

Human communication hinges on drawing implicit meaning from text, going beyond what is explicitly stated. In this proposal, we investigate the role of implicit content in NLP, consisting of the propositions, beliefs, attitudes, and assumptions that remain unstated but are important for understanding communicative intent. Across three interconnected chapters, we demonstrate how computational methods for recognizing and representing implicit content significantly improve NLP systems in diverse settings.
First, we examine implicit content in individual utterances by extending pragmatic inferences to include plausible inferences obtained from domain knowledge along with abductive and deductive inferences, going beyond the classical concepts of presupposition and implicature to capture the communicative intent of users. We also introduce PairScale, a framework combining implicit content extraction with pairwise comparisons to measure subjective constructs from public attitudes to psychological stress.
Next, we shift to the conversational level, where implicit content involves a participant's beliefs about their partner's beliefs and intentions. Through analysis of collaborative problem-solving dialogues, we demonstrate that misalignments in these shared assumptions, or the conversational common ground, create measurable friction that can lead to task failure, while revealing current models' limitations in tracking implicit belief states during dialogue both as an observer and a participant.
Third, we show how explicit representation of implicit content enhances real-world applications. In conversational systems, inferring users' unstated information needs improves query rewriting and response quality. In sensitive domains like maternal health, identifying implicit false assumptions in questions enables safer, more complete answers that address both surface queries and underlying misconceptions. Finally, we also explore how understanding implicit content in a conversational setting can help prevent harmful responses from AI chatbots in mental health conversations. Together, these contributions show that modeling implicit content is essential for building NLP systems that are not only more accurate, but also more aligned with human communication and needs.