Modulating Language Models with Emotions

TL;DR

This paper introduces a modulated layer normalization technique for large-scale language models to generate emotionally aware responses, achieving superior performance and efficiency with limited data.

Contribution

It presents a novel modulation method inspired by computer vision to enhance emotional response generation in language models.

Findings

Outperforms baseline methods in automatic and human evaluations.

Maintains diversity, fluency, and coherence in generated responses.

Effective even with only 10% of training data.

Abstract

Generating context-aware language that embodies diverse emotions is an important step towards building empathetic NLP systems. In this paper, we propose a formulation of modulated layer normalization -- a technique inspired by computer vision -- that allows us to use large-scale language models for emotional response generation. In automatic and human evaluation on the MojiTalk dataset, our proposed modulated layer normalization method outperforms prior baseline methods while maintaining diversity, fluency, and coherence. Our method also obtains competitive performance even when using only 10% of the available training data.