Simpler Context-Dependent Logical Forms via Model Projections
Reginald Long Stanford University、Panupong Pasupat Stanford University、Percy Liang Stanford University
Abstract
We consider the task of learning a context-dependent mapping from utterances to denotations. With only denotations at training time, we must search over a combinatorially large space of logical forms, which is even larger with context-dependent utterances. To cope with this challenge, we perform successive projections of the full model onto simpler models that operate over equivalence classes of logical forms. Though less expressive, we find that these simpler models are much faster and can be surprisingly effective. Moreover, they can be used to bootstrap the full model. Finally, we collected three new context-dependent semantic parsing datasets, and develop a new left-to-right parser.
中文速览
用机器理解"把它混合一下"这样依赖上下文的自然语言指令，难点在于系统必须在没有任何词典的情况下，从海量可能的逻辑形式中找到正确的那个。论文提出将完整语义解析模型（Model A，词语与逻辑形式精确对齐）逐步"投影"成更简单的模型：先去掉对齐约束只保留逻辑形式（Model B），再进一步把组合逻辑形式折叠为扁平的具体参数形式（Model C），搜索空间从百万量级骤降到数百量级。实验表明，在有限计算资源（固定束宽）下，最简单的 Model C 反而效果最好，而在无限束宽的合成数据上完整的 Model A 才占优；三者还可以互相引导——用简单模型初始化复杂模型，最终让完整模型也能高效训练。论文同时发布了三个新的上下文依赖语义解析数据集（Alchemy、Scene、Tangrams），并设计了一种新的从左到右解析器，为在真实对话场景下训练无监督语义解析器提供了切实可行的路径。
原文 arXiv:1606.05378;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1606.05378v1