Utterance Classification

Utterance classification aims to automatically categorize individual statements within a conversation, enabling deeper understanding of dialogue structure and content. Current research focuses on improving accuracy using multimodal data (combining text and speech), sophisticated neural network architectures like transformers and logical neural networks, and multi-task learning approaches that leverage relationships between different classification tasks (e.g., identifying needs vs. assets, or classifying speaker roles). This work has significant implications for various applications, including automated analysis of mental health interviews, community needs assessments from social media, and improved meeting summarization.