Language Transfer Hypotheses with Linear SVM Weights

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1385–1390, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Language Transfer Hypotheses with Linear SVM Weights Shervin Malmasi Centre for Language Technology Macquarie University Sydney, NSW, Australia shervin.malmasi@mq.edu.au Mark Dras Centre for Language Technology Macquarie University Sydney, NSW, Australia mark.dras@mq.edu.au Abstract Language transfer, the characteristic sec- ond language usage patterns caused by na- tive language interference, is investigated by Second Language Acquisition (SLA) researchers seeking to find overused and underused linguistic features. In this pa- per we develop and present a methodology for deriving ranked lists of such features. Using very large learner data, we show our method’s ability to find relevant can- didates using sophisticated linguistic fea- tures. To illustrate its applicability to SLA research, we formulate plausible language transfer hypotheses supported by current evidence. This is the first work to ex- tend Native Language Identification to a broader linguistic interpretation of learner data and address the automatic extraction of underused features on a per-native lan- guage basis. 1 Introduction It has been noted in the linguistics literature since the 1950s that speakers of particular languages have characteristic production patterns when writ- ing in a second language. This language transfer phenomenon has been investigated independently in a number of fields from different perspectives, including qualitative research in Second Language Acquisition (SLA) and more recently though pre- dictive computational models in NLP. Motivated by the aim of improving foreign lan- guage teaching and learning, such analyses are of- ten done manually in SLA, and are difficult to perform for large corpora. Smaller studies yield poor results due to the sample size, leading to extreme variability (Ellis, 2008). Recently, re- searchers have noted that NLP has the tools to use large amounts of data to automate this analysis, using complex feature types. This has motivated studies in Native Language Identification (NLI), a subtype of text classification where the goal is to determine the native language (L1) of an author using texts they have written in a second language (L2) (Tetreault et al., 2013). Despite the good results in predicting L1s, few attempts have been made to interpret the features that distinguish L1s. This is partly because no methods for an SLA-oriented feature analysis have been proposed; most work focuses on testing fea- ture types using standard machine learning tools. The overarching contribution of this work is to develop a methodology that enables the transfor- mation of the NLI paradigm into SLA applications that can be used to link these features to their un- derlying linguistic causes and explanations. These candidates can then be applied in other areas such as remedial SLA strategies or error detection. 2 Related Work SLA research aims to find distributional differ- ences in language use between L1s, often referred to as overuse, the extensive use of some linguis- tic structures, and underuse, the underutilization of particular structures, also known as avoidance (Gass and Selinker, 2008). While there have been some attempts …

Content truncated to 500 words