On the Learning of Non-Autoregressive Transformers
This work addresses the problem of improving text generation efficiency for applications requiring low latency, but it is incremental as it provides theoretical insights rather than a novel method.
The paper analyzes the challenges of learning Non-Autoregressive Transformers (NATs) for text generation, showing that standard training drops token dependencies measured by conditional total correlation, and proposes a unified framework to explain existing successes and guide new methods.
Non-autoregressive Transformer (NAT) is a family of text generation models, which aims to reduce the decoding latency by predicting the whole sentences in parallel. However, such latency reduction sacrifices the ability to capture left-to-right dependencies, thereby making NAT learning very challenging. In this paper, we present theoretical and empirical analyses to reveal the challenges of NAT learning and propose a unified perspective to understand existing successes. First, we show that simply training NAT by maximizing the likelihood can lead to an approximation of marginal distributions but drops all dependencies between tokens, where the dropped information can be measured by the dataset's conditional total correlation. Second, we formalize many previous objectives in a unified framework and show that their success can be concluded as maximizing the likelihood on a proxy distribution, leading to a reduced information loss. Empirical studies show that our perspective can explain the phenomena in NAT learning and guide the design of new training methods.