Generating Code-switched Text for Lexical Learning

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 562–571, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Generating Code-switched Text for Lexical Learning Igor Labutov Cornell University iil4@cornell.edu Hod Lipson Cornell University hod.lipson@cornell.edu Abstract A vast majority of L1 vocabulary acqui- sition occurs through incidental learning during reading (Nation, 2001; Schmitt et al., 2001). We propose a probabilistic ap- proach to generating code-mixed text as an L2 technique for increasing retention in adult lexical learning through reading. Our model that takes as input a bilingual dictionary and an English text, and gener- ates a code-switched text that optimizes a defined “learnability” metric by construct- ing a factor graph over lexical mentions. Using an artificial language vocabulary, we evaluate a set of algorithms for gener- ating code-switched text automatically by presenting it to Mechanical Turk subjects and measuring recall in a sentence com- pletion task. 1 Introduction Today, an adult trying to learn a new language is likely to embrace an age-old and widely accepted practice of learning vocabulary through curated word lists and rote memorization. Yet, it is not uncommon to find yourself surrounded by speak- ers of a foreign language and instinctively pick up words and phrases without ever seeing the defini- tion in your native tongue. Hearing “pass le sale please” at the dinner table from your in-laws vis- iting from abroad, is unlikely to make you think twice about passing the salt. Humans are extraor- dinarily good at inferring meaning from context, whether this context is your physical surround- ing, or the surrounding text in the paragraph of the word that you don’t yet understand. Recently, a novel method of L2 language teach- ing had been shown effective in improving adult lexical acquisition rate and retention 1. This tech- 1authors’ unpublished work nique relies on a phenomenon that elicits a nat- ural simulation of L1-like vocabulary learning in adults — significantly closer to L1 learning for L2 learners than any model studied previously. By in- fusing foreign words into text in the learner’s na- tive tongue into low-surprisal contexts, the lexi- cal acquisition process is facilitated naturally and non-obtrusively. Incidentally, this phenomenon occurs “in the wild” and is termed code-switching or code-mixing, and refers to the linguistic pattern of bilingual speakers swapping words and phrases between two languages during speech. While this phenomenon had received significant attention from both a socio-linguistic (Milroy and Muysken, 1995) and theoretical linguistic perspectives (Be- lazi et al., 1994; Bhatt, 1997) (including some computational studies), only recently has it been hypothesizes that “code-switching” is a marking of bilingual proficiency, rather than deficiency (Genesee, 2001). Until recently it was widely believed that inci- dental lexical acquisition through reading can only occur for words that occur at sufficient density in a single text, so as to elicit the “noticing” ef- fect needed for lexical acquisition to occur (Cobb, 2007). Recent neurophysiological findings, how- ever, indicate that even a single incidental expo- sure to a novel word …

Content truncated to 500 words