Reconstructing Native Language Typology from Foreign Language Usage

Proceedings of the Eighteenth Conference on Computational Language Learning, pages 21–29, Baltimore, Maryland USA, June 26-27 2014. c⃝2014 Association for Computational Linguistics Reconstructing Native Language Typology from Foreign Language Usage Yevgeni Berzak CSAIL MIT berzak@mit.edu Roi Reichart Technion IIT roiri@ie.technion.ac.il Boris Katz CSAIL MIT boris@mit.edu Abstract Linguists and psychologists have long been studying cross-linguistic transfer, the influence of native language properties on linguistic performance in a foreign lan- guage. In this work we provide empirical evidence for this process in the form of a strong correlation between language simi- larities derived from structural features in English as Second Language (ESL) texts and equivalent similarities obtained from the typological features of the native lan- guages. We leverage this finding to re- cover native language typological similar- ity structure directly from ESL text, and perform prediction of typological features in an unsupervised fashion with respect to the target languages. Our method achieves 72.2% accuracy on the typology predic- tion task, a result that is highly competi- tive with equivalent methods that rely on typological resources. 1 Introduction Cross-linguistic transfer can be broadly described as the application of linguistic structure of a speaker’s native language in the context of a new, foreign language. Transfer effects may be expressed on various levels of linguistic perfor- mance, including pronunciation, word order, lex- ical borrowing and others (Jarvis and Pavlenko, 2007). English,Such traces are prevalent in non-native and in some cases are even cele- brated in anecdotal hybrid dialect names such as “Frenglish” and “Denglish”. Although cross-linguistic transfer was exten- sively studied in Psychology, Second Language Acquisition (SLA) and Linguistics, the conditions under which it occurs, its linguistic characteristics as well as its scope remain largely under debate (Jarvis and Pavlenko, 2007; Gass and Selinker, 1992; Odlin, 1989). In NLP, the topic of linguistic transfer was mainly addressed in relation to the Native Lan- guage Identification (NLI) task, which requires to predict the native language of an ESL text’s au- thor. The overall high performance on this classi- fication task is considered to be a central piece of evidence for the existence of cross-linguistic trans- fer (Jarvis and Crossley, 2012). While the success on the NLI task confirms the ability to extract na- tive language signal from second language text, it offers little insight into the linguistic mechanisms that play a role in this process. In this work, we examine the hypothesis that cross-linguistic structure transfer is governed by the typological properties of the native language. We provide empirical evidence for this hypothe- sis by showing that language similarities derived from structural patterns of ESL usage are strongly correlated with similarities obtained directly from the typological features of the native languages. This correlation has broad implications on the ability to perform inference from native language structure to second language performance and vice versa. In particular, it paves the way for a novel and powerful framework for comparing native languages through second language performance. This framework overcomes many of the inher- ent difficulties of direct comparison between …

Content truncated to 500 words