Using Contextual Speller Techniques and Language Modeling for ESL Error Correction

Using Contextual Speller Techniques and Language Modeling for ESL Error Correction Michael Gamon*, Jianfeng Gao*, Chris Brockett*, Alexandre Klementiev+, William B. Dolan*, Dmitriy Belenko*, Lucy Vanderwende* *Microsoft Research One Microsoft Way Redmond, WA 98052 {mgamon,jfgao,chrisbkt,billdol, dmitryb,lucyv}@microsoft.com +Dept. of Computer Science University of Illinois Urbana, IL 61801 klementi@uiuc.edu Abstract We present a modular system for detection and correction of errors made by non- native (English as a Second Language = ESL) writers. We focus on two error types: the incorrect use of determiners and the choice of prepositions. We use a decision- tree approach inspired by contextual spelling systems for detection and correction suggestions, and a large language model trained on the Gigaword corpus to provide additional information to filter out spurious suggestions. We show how this system performs on a corpus of non-native English text and discuss strategies for future enhancements. 1 Introduction English is today the de facto lingua franca for commerce around the globe. It has been estimated that about 750M people use English as a second language, as opposed to 375M native English speakers (Crystal 1997), while as much as 74% of writing in English is done by non-native speakers. However, the errors typically targeted by commercial proofing tools represent only a subset of errors that a non-native speaker might make. For example, while many non-native speakers may encounter difficulty choosing among prepositions, this is typically not a significant problem for native speakers and hence remains unaddressed in proofing tools such as the grammar checker in Microsoft Word (Heidorn 2000). Plainly there is an opening here for automated proofing tools that are better geared to the non-native users. One challenge that automated proofing tools face is that writing errors often present a semantic dimension that renders it difficult if not impossible to provide a single correct suggestion. The choice of definite versus indefinite determiner—a common error type among writers with a Japanese, Chinese or Korean language background owing to the lack of overt markers for definiteness and indefiniteness—is highly dependent on larger textual context and world knowledge. It seems desirable, then, that proofing tools targeting such errors be able to offer a range of plausible suggestions, enhanced by presenting real-world examples that are intended to inform a user’s selection of the most appropriate wording in the context1. 2 Targeted Error Types Our system currently targets eight different error types: 1. Preposition presence and choice: In the other hand, ... (On the other hand ...) 2. Definite and indefinite determiner presence and choice: I am teacher... (am a teacher) 3. Gerund/infinitive confusion: I am interesting in this book. (interested in) 4. Auxiliary verb presence and choice: My teacher does is a good teacher (my teacher is...) 1 Liu et al. 2000 take a similar approach, retrieving example sentences from a large corpus. 449 5. Over-regularized verb inflection: I writed a letter (wrote) 6. Adjective/noun confusion: This is a China book (Chinese book) 7. Word order (adjective sequences and nominal compounds): I am a student of university (university student) …

Content truncated to 500 words