Kernelized Bayesian Softmax for Text Generation

Ning Miao, Hao Zhou, Chengqi Zhao, Wenxian Shi, Yitan Li, Lei Li

Abstract: Neural models for text generation require a softmax layer with proper token embeddings during the decoding phase. Most existing approaches adopt single point embedding for each token. However, a word may have multiple senses according to different context, some of which might be distinct. In this paper, we propose KerBS, a novel approach for learning better embeddings for text generation. KerBS embodies two advantages: (a) it employs a Bayesian composition of a word embedding with multiple senses; (b) it is adaptive to word variances and robust to rare sentence context by imposing learned kernels to capture the closeness of words (senses) in the embedding space. Empirical studies show that KerBS significantly boosts the performance of several text generation tasks.

Code Link: https://github.com/NingMiao/KerBS

CMT Num: 6783