Semantic Parsing for English as a Second Language

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6783–6794 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 6783 Semantic Parsing for English as a Second Language Yuanyuan Zhao1,2, Weiwei Sun1,3, Junjie Cao1∗and Xiaojun Wan1 1Wangxuan Institute of Computer Technology, Peking University 1The MOE Key Laboratory of Computational Linguistics, Peking University 2Academy for Advanced Interdisciplinary Studies, Peking University 3Center for Chinese Linguistics, Peking University {zhao_yy,ws,wanxiaojun}@pku.edu.cn junjie.junjiecao@alibaba-inc.com Abstract This paper is concerned with semantic pars- ing for English as a second language (ESL). Motivated by the theoretical emphasis on the learning challenges that occur at the syntax- semantics interface during second language acquisition, we formulate the task based on the divergence between literal and intended meanings. We combine the complementary strengths of English Resource Grammar, a linguistically-precise hand-crafted deep gram- mar, and TLE, an existing manually annotated ESL UD-TreeBank with a novel reranking model. Experiments demonstrate that in com- parison to human annotations, our method can obtain a very promising SemBanking qual- ity. By means of the newly created corpus, we evaluate state-of-the-art semantic parsing as well as grammatical error correction mod- els. The evaluation profiles the performance of neural NLP techniques for handling ESL data and suggests some research directions. 1 Introduction There are more people around the world learning English as a second language (ESL) than there are native speakers of English with this gap con- tinually and steadily expanding (Crystal, 2012). Accordingly, an extremely large volume of non- native English texts are generated every day. We need an automatic machinery to annotate such large-scale atypical data with in-depth linguistic analysis. High-performance automatic annotation of learner texts, from an engineering point of view, enables it possible to derive high-quality informa- tion by structuring the specific type of data, and from a scientific point of view, facilitates quan- titative studies for Second Language Acquisition (SLA), which is complementary to hands-on ex- periences in interpreting interlanguage phenom- ∗Now works at Alibaba Group. ena (Gass, 2013). This direction has been re- cently explored by the NLP community (Nagata and Sakaguchi, 2016; Berzak et al., 2016a; Lin et al., 2018). Different from standard English, ESL may pre- serve many features of learners’ first languages1. The difference between learner texts and bench- mark training data, e.g. Penn TreeBank (PTB; Marcus et al., 1993), is more related to linguis- tic competence, rather than performance (Chom- sky, 2014). This makes processing ESL different from almost all the existing discussions on domain adaptation in NLP. Despite the ubiquity and importance of interlan- guages at both the scientific and engineering lev- els, it is only partially understood how NLP mod- els perform on them. In this paper, we present, to the best of our knowledge, the first study on Se- mantic Parsing for English as a Second Language. Motivated by the Interface Hypothesis (Sorace, 2011) in SLA, we emphasize on the divergence between literal and intended meanings. To obtain reliable semantic analyses in order to represent the two types of meanings, we …

Content truncated to 500 words