This paper presents a model of affect in conversations by leveraging Appraisal Theory as a generalizable framework. We propose that the multidimensional cognitive model of Appraisal Theory offers significant advantages for analyzing emotions in conversational contexts, addressing the current challenges of inconsistent annotation methodologies across corpora. To demonstrate this, we present AppraisePLM, a regression and classification model trained on the crowd-EnVent corpus that outperforms existing models in predicting 21 appraisal dimensions including pleasantness, self-control, and alignment with social norms. We apply AppraisePLM to diverse conversation datasets spanning task-oriented dialogues, general-domain chit-chat, affect-specific conversations, and domain-specific affect analysis. Our analysis reveals that AppraisePLM successfully extrapolates emotion labels across datasets, while capturing domain-specific patterns in affect flow – change in conversational emotion over the conversation. This work highlights the entangled nature of affective phenomena in conversation and positions affect flow as a promising model for holistic emotion analysis, offering a standardized approach to evaluate and benchmark affective capabilities in conversational agents.

This paper is the model description for the Emo-Gen BART dialogue generation architecture, as submitted to the SCI-CHAT 2024 Shared Task. The Emotion-Informed Dialogue Generation model is a multi-task BARTbased model which performs dimensional and categorical emotion detection and uses that information to augment the input to the generation models. Our implementation is trained and validated against the IEMOCAP dataset, and compared against contemporary architectures in both dialogue emotion classification and dialogue generation. We show that certain loss function ablations are competitive against the state-of-the-art single-task models.