Sequence-to-Sequence Learning as Beam-Search Optimization
Sam Wiseman Alexander M. Rush School of Engineering and Applied Sciences Harvard University Cambridge, MA, USA
Abstract
Sequence-to-Sequence (seq2seq) modeling has rapidly become an important general-purpose NLP tool that has proven effective for many text-generation and sequence-labeling tasks. Seq2seq builds on deep neural language modeling and inherits its remarkable accuracy in estimating local, next-word distributions. In this work, we introduce a model and beam-search training scheme, based on the work of ?), that extends seq2seq to learn global sequence scores. This structured approach avoids classical biases associated with local training and unifies the training loss with the test-time usage, while preserving the proven model architecture of seq2seq and its efficient training approach. We show that our system outperforms a highly-optimized attention-based seq2seq system and other baselines on three different sequence to sequence tasks: word ordering, parsing, and machine translation.
中文速览
seq2seq（序列到序列）模型虽然在机器翻译、句法分析等任务上表现出色，但其训练方式存在三大隐患：训练时只见"正确历史"导致的曝光偏差（exposure bias）、词级损失与序列级评估指标不匹配，以及局部归一化引发的标签偏差（label bias）。为此，这篇论文提出了"束搜索优化"（Beam Search Optimization，BSO）方法：将seq2seq解码器改为输出非概率的序列得分，并在训练中直接模拟束搜索过程，每当正确序列被挤出候选集时就触发惩罚损失，从而让模型在训练阶段也能接触到自身的错误输出。实验在词序恢复、句法分析和机器翻译三个任务上均超过了经过高度调优的注意力机制seq2seq基线，在搜索难度较大的任务上提升尤为显著。这项工作的意义在于，它在几乎不改变原有模型结构和训练效率的前提下，系统性地弥补了seq2seq训练与推断之间的鸿沟，为结构化预测与神经序列生成的结合提供了一条实用路径。
原文 arXiv:1606.02960;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1606.02960v2