Search papers, labs, and topics across Lattice.
This paper introduces EmoVec, a novel framework for enhancing emotional expressiveness in Large Language Models (LLMs) by steering latent vectors during inference. By extracting emotion-specific directions from paired responses and refining them through debiasing techniques, EmoVec allows for continuous control over emotional intensity without the need to retrain the model. Experimental results demonstrate that EmoVec significantly enhances emotional salience across multiple LLMs and emotions while maintaining semantic integrity and coherence.
EmoVec enables real-time emotional control in LLMs, enhancing their affective responses without retraining.
Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further refines them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectiveness of vector purification and adaptive scaling, establishing EmoVec as a practical inference-time method for affective control in deployed LLMs.
