Search papers, labs, and topics across Lattice.
This paper introduces EmoStance, a novel approach to empathetic response generation that leverages multi-annotator emoji distributions as weak supervision to control the affective orientation of responses. By constructing the EmojiDialogue dataset and employing a frozen instruction-tuned LLM, EmoStance effectively models both the speaker's affective expression and the listener's response orientation, resulting in improved contextual specificity and perceived responsiveness. In evaluations, EmoStance achieved a 62.2% decisive win rate, highlighting its effectiveness in generating empathetic responses compared to traditional methods.
EmoStance achieves a remarkable 62.2% win rate in generating more contextually aware and responsive empathetic dialogues by harnessing emoji distributions as weak supervision.
Empathetic response generation requires models to decide not only what to say, but also how to respond to the previous speaker's affective situation. We formulate this as response-side affective-orientation control and use multi-annotator emoji distributions as weak affective--attitudinal evidence, rather than as output symbols or gold labels, to induce a latent control space that operationally approximates listener stance. We construct EmojiDialogue, an utterance-level extension of EmpatheticDialogues with emoji votes and confidence scores, and propose EmoStance, which models source-side affective expression, predicts a soft response-side orientation from dialogue context and speaker roles, and steers a frozen instruction-tuned LLM through continuous prefix embeddings. In blind pairwise evaluation with 20 annotators and 800 judgments, EmoStance achieves a 62.2% decisive win rate, with the clearest gains in contextual specificity and perceived responsiveness, while remaining complementary to external-knowledge methods. Code, annotation metadata, and reconstruction scripts are available in our GitHub repository: https://github.com/18277390221/EmoStance.
