Semi-Supervised Lexicon Learning for Wide-Coverage Semantic Parsing

Proceedings of the 27th International Conference on Computational Linguistics, pages 892–904 Santa Fe, New Mexico, USA, August 20-26, 2018. 892 Semi-Supervised Lexicon Learning for Wide-Coverage Semantic Parsing Bo Chen†‡, Bo An†‡, Le Sun†, Xianpei Han† †State Key Laboratory of Computer Science Institute of Software, Chinese Academy of Sciences, Beijing, China ‡University of Chinese Academy of Sciences, Beijing, China {chenbo,anbo,sunle,xianpei}@iscas.ac.cn Abstract Semantic parsers critically rely on accurate and high-coverage lexicons. However, traditional semantic parsers usually utilize annotated logical forms to learn the lexicon, which often suffer from the lexicon coverage problem. In this paper, we propose a graph-based semi-supervised learning framework that makes use of large text corpora and lexical resources. This framework first constructs a graph with a phrase similarity model learned by utilizing many text corpora and lexical resources. Next, graph propagation algorithm identifies the label distribution of unlabeled phrases from labeled ones. We evaluate our approach on two benchmarks: WEBQUESTIONS and FREE917. The results show that, in both datasets, our method achieves substantial improve- ment when comparing to the base system that does not utilize the learned lexicon, and gains competitive results when comparing to state-of-the-art systems. Title and Abstract in Chinese 基于半监督词典学习的语义解析技术研究 语义解析器的性能往往依赖于词典的准确度和覆盖度。传统语义解析器利用标注好的句 子-逻辑表达式对来学习词典， 这通常会面临词典覆盖度不足的问题。本文提出了一种 基于图的半监督学习框架，该框架能够充分利用容易获取的 大量文本语料和词典资源来 进行词典扩充学习。该词典扩充学习方法首先利用大量文本语料和词典资源来学习词语 与词语之间的相似度，并构建用于图传播的图；接着使用图传播算法从少量标注的词汇 中学习新的词汇。本文 在两个公开数据集上进行了实验，实验结果表明：本文系统相比 未使用新词汇的基准系统取得了显著提升，相比当前 最好的系统，也取得了具有竞争力 的结果。 1 Introduction Semantic parsing aims to map natural language sentences into formal meaning representations, e.g., Figure 1 shows an example of semantic parsing. Semantic parsing plays an important role in natural language understanding, and has attracted increasing attention in recent years (Zelle and Mooney, 1996; Wong and Mooney, 2007; Lu et al., 2008; Liang et al., 2011; Kwiatkowski et al., 2011; Artzi and Zettlemoyer, 2013; Krishnamurthy and Mitchell, 2014; Li et al., 2015; Chen et al., 2016; Xiao et al., 2016; Jia and Liang, 2016; Reddy et al., 2016; Liang et al., 2017). The performance of semantic parsers critically depends on the quality of lexicon, including accuracy and coverage. Specifically, in order to construct the logical form from a sentence, we first need to learn a lexicon,1 which contains the mappings from natural language phrases (e.g., “born”) to logical predicates (e.g., PlaceOfBirth). From the example in Figure 1, we can see that lexicon is the foundation of parsing, and lexicon learning plays an important role in semantic parsing. Traditional semantic parsers are usually domain-specific, which only contains a limited number of logical predicates (Zettlemoyer and Collins, 2005; Kwiatkowksi et al., 2010; Artzi et al., 2014). In this This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 1We follow the lexicon style defined in Berant et al. (2013). 893 Which city was barack obama born in ? Type.City Barack_Obama Type.city�PlaceOfBirth.Barack_Obama PlaceOfBirth PlaceOfBirth.Barack_Obama Lexicon Lexicon Lexicon Grammar Grammar Figure 1: An example of semantic parsing, which uses lexicons to map phrases to predicates, and applies grammars to construct the logical form. case, the mappings for each predicate can be learned relatively easily from training corpus. Recently, a growing body of research has scaled up semantic parsers to open …

Content truncated to 500 words