FlowSeq: Non-Autoregressive Conditional Sequence Generation with Generative Flow

Abstract

Most sequence-to-sequence (seq2seq) models are autoregressive; they generate each token by conditioning on previously generated tokens. In contrast, non-autoregressive seq2seq models generate all tokens in one pass, which leads to increased efficiency through parallel processing on hardware such as GPUs. However, directly modeling the joint distribution of all tokens simultaneously is challenging, and even with increasingly complex model structures accuracy lags significantly behind autoregressive models. In this paper, we propose a simple, efficient, and effective model for non-autoregressive sequence generation using latent variable models. Specifically, we turn to generative flow, an elegant technique to model complex distributions using neural networks, and design several layers of flow tailored for modeling the conditional density of sequential latent variables. We evaluate this model on three neural machine translation (NMT) benchmark datasets, achieving comparable performance with state-of-the-art nonautoregressive NMT models and almost constant decoding time w.r.t the sequence length. 1

Non-Autoregressive Machine Translation with Latent Alignments

“…Non-autoregressive neural machine translation (Gu et al, 2018) aims to enable the parallel generation of output tokens without sacrificing translation quality. There has been a surge of recent interest in this family of efficient decoding models, resulting in the development of iterative refinement (Lee et al, 2018), CTC models (Libovicky and Helcl, 2018), insertion-based methods Chan et al, 2019b), editbased methods (Gu et al, 2019;Ruis et al, 2019), masked language models (Ghazvininejad et al, 2019(Ghazvininejad et al, , 2020b, and normalizing flow models (Ma et al, 2019). Some of these methods generate the output tokens in a constant number of steps (Gu et al, 2018;Libovicky and Helcl, 2018;Lee et al, 2018;Ghazvininejad et al, 2019Ghazvininejad et al, , 2020b, while others require a logarithmic number of generation steps Chan et al, 2019b,a;Li and Chan, 2019).…”

A Bilingual Generative Transformer for Semantic Sentence Embedding

“…Decoder Architecture. Many latent variable models for text use LSTMs (Hochreiter and Schmidhuber, 1997) as their decoders (Yang et al, 2017;Ziegler and Rush, 2019;Ma et al, 2019). However, state-of-the-art models in neural machine translation have seen increased performance and speed using deep Transformer architectures.…”

Iterative Refinement in the Continuous Space for Non-Autoregressive Neural Machine Translation

“…Learning On the other hand, Ma et al (2019); Shu et al (2020) proposed to use continuous latent variables for non-autoregressive translation. By letting the latent variables z (of dimensionality T × D) capture the dependencies between the target tokens, the decoder p θ (y|z, x) can be factorized over time.…”

“…Meanwhile, another line of work proposed to use continuous latent variables for non-autoregressive translation, such that the distribution of the target sentences can be factorized over time given the latent variables (Ma et al, 2019;Shu et al, 2020). Unlike the models discussed above, finding the most likely target sentence under these models requires searching over continuous latent variables.…”