Posterior Control of Blackbox Generation
Posterior Control of Blackbox Generation
复制标题
DOI：
10.18653/v1/2020.acl-main.243
复制
发表时间：
2020-05
影响因子：
4.2
通讯作者：
Xiang Lisa Li;Alexander M. Rush
中科院分区：
文献类型：
--
作者：
Xiang Lisa Li;Alexander M. Rush
文本生成通常需要遵守特定于任务的规则的高精度输出。这种细粒度的控制很难用现成的深度学习模型来执行。在这项工作中，我们考虑通过结构化的潜变量方法学习离散控制状态来增强神经生成模型。在这种表述下，特定于任务的知识可以通过一系列丰富的后验约束进行编码，这些约束被有效地训练到模型中。这种方法允许用户基于先验知识进行内部模型决策，而不会牺牲神经生成模型的代表性。实验考虑这种方法的文本生成的应用程序。我们发现，这种方法比标准的基准测试有所改进，同时还提供了细粒度的控制。
Text generation often requires high-precision output that obeys task-specific rules. This fine-grained control is difficult to enforce with off-the-shelf deep learning models. In this work, we consider augmenting neural generation models with discrete control states learned through a structured latent-variable approach. Under this formulation, task-specific knowledge can be encoded through a range of rich, posterior constraints that are effectively trained into the model. This approach allows users to ground internal model decisions based on prior knowledge, without sacrificing the representational power of neural generative models. Experiments consider applications of this approach for text generation. We find that this method improves over standard benchmarks, while also providing fine-grained control.