A Simple Contrastive Learning Objective for Alleviating Neural Text Degeneration
share
The cross-entropy objective has proved to be an all-purpose training objective for autoregressive language models (LMs). However, without considering the penalization of problematic tokens, LMs trained using cross-entropy exhibit text degeneration. To address this, unlikelihood training has been proposed to force unlikely tokens to be assigned a low probability by a LM. But unlikelihood does not consider the relationship between the label tokens and the unlikely token candidates, thus showing marginal improvements in degeneration. We propose a new contrastive token learning objective that inherits the advantages of cross-entropy and unlikelihood training and avoids their limitations. The key idea is to force a LM to generate high probabilities for label tokens at each step while low probabilities of negative candidates. Comprehensive experiments on language modeling and open-domain dialogue generation tasks show that the proposed contrastive token objective yields less repetitive texts, with a higher generation quality than unlikelihood training, achieving the new state-of-the-art performance.
READ FULL TEXT
