SRL-Based Verb Selection for ESL

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 1068–1076, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Abstract In this paper we develop an approach to tackle the problem of verb selection for learners of English as a second language (ESL) by using features from the output of Semantic Role La- beling (SRL). Unlike existing approaches to verb selection that use local features such as n-grams, our approach exploits semantic fea- tures which explicitly model the usage context of the verb. The verb choice highly depends on its usage context which is not consistently captured by local features. We then combine these semantic features with other local fea- tures under the generalized perceptron learn- ing framework. Experiments on both in- domain and out-of-domain corpora show that our approach outperforms the baseline and achieves state-of-the-art performance.1 1 Introduction Verbs in English convey actions or states of being. In addition, they also communicate sentiments and imply circumstances, e.g., in “He got [gained] the scholarship after three interviews.”, the verb “gained” may indicate that the “scholarship” was competitive and required the agent’s efforts; in contrast, “got” sounds neutral and less descriptive. * This work has been done while the author was visiting Mi- crosoft Research Asia. Since verbs carry multiple important functions, misusing them can be misleading, e.g., the native speaker could be confused when reading “I like looking [reading] books”. Unfortunately, accord- ing to (Gui and Yang, 2002; Yi et al., 2008), more than 30% of the errors in the Chinese Learner Eng- lish Corpus (CLEC) are verb choice errors. Hence, it is useful to develop an approach to automatically detect and correct verb selection errors made by ESL learners. However, verb selection is a challenging task because verbs often exhibit a variety of usages and each usage depends on a particular context, which can hardly be adequately described by convention- al n-gram features. For instance, both “made” and “received” can complete “I have __ a telephone call.”, where the usage context can be represented as “made/received a telephone call”; however, in “I have __ a telephone call from my boss”, the prepositional phrase “from my boss” becomes a critical part of the context, which now cannot be described by n-gram features, resulting in only “received” being suitable. Some researchers (Tetreault and Chodorow, 2008) exploited syntactic information and n-gram features to represent verb usage context. Yi et al. (2008) introduced an unsupervised web-based proofing method for correcting verb-noun colloca- tion errors. Brockett et al. (2006) employed phrasal Statistical Machine Translation (SMT) techniques to correct countability errors. None of their meth- ods incorporated semantic information. SRL-based Verb Selection for ESL 1,2Xiaohua Liu, 3Bo Han*, 4Kuan Li*, 5Stephan Hyeonjun Stiller and 2Ming Zhou 1School of Computer Science and Technology Harbin Institute of Technology 2Microsoft Research Asia 3Department of Computer Science and Software Engineering The University of Melbourne 4College of Computer Science Chongqing University 5Computer Science Department Stanford University {xiaoliu, mingzhou, v-kuli}@microsoft.com b.han@pgrad.unimelb.edu.au sstiller@stanford.edu 1068 Unlike the …

Content truncated to 500 words