Analyzing Learner Understanding of Novel L2 Vocabulary

Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning (CoNLL), pages 126–135, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Analyzing Learner Understanding of Novel L2 Vocabulary Rebecca Knowles and Adithya Renduchintala and Philipp Koehn and Jason Eisner Department of Computer Science Johns Hopkins University {rknowles,adi.r,phi,eisner}@jhu.edu Abstract In this work, we explore how learners can infer second-language noun meanings in the context of their native language. Mo- tivated by an interest in building interac- tive tools for language learning, we collect data on three word-guessing tasks, analyze their difficulty, and explore the types of er- rors that novice learners make. We train a log-linear model for predicting our sub- jects’ guesses of word meanings in varying kinds of contexts. The model’s predictions correlate well with subject performance, and we provide quantitative and qualita- tive analyses of both human and model performance. 1 Introduction Second language (L2) instruction includes an em- phasis on vocabulary, as reflected in curricular ma- terials and educational technology. Learners ac- quire new vocabulary in several ways, including direct instruction, memorization, and incidental acquisition. In this work, we seek a predictive model of the circumstances in which incidental ac- quisition is possible. That is, when can a learner guess the meaning of a novel word? We present novice learners with new L2 words inserted in sentences otherwise written in their na- tive language (L1). This experimental design al- lows us to assume that all subjects understand the full context, rather than needing to assess how much of an L2 context each subject understood. We also present novice learners with the same novel words out of context. This allows us to study how cognateness and context interact, in a well- controlled setting. Cognates and very common words may be easy to translate without context, while contextual clues may be needed to make other words guessable. In the initial experiments we present here, we focus on the language pair of English L1 and Ger- man L2, selecting subjects who self-identify as fluent English speakers with minimal exposure to German. We confine ourselves to novel nouns, as we expect that their relative lack of morphologi- cal inflection in both languages1 will produce less noisy results than verbs, for example. (For verbs, naive learners would be required to attend to tense and mood in addition to the lemma.) The goal of this work is to develop intuitions that may transfer to less artificial learning set- tings. Even experienced L2 readers will encounter novel words when reading L2 text. Their ability to decipher a novel word is known to depend on both their understanding of the surrounding con- text words (to understand a text, a reader needs to understand at least 95% of its words (Huckin and Coady, 1999)) and the cognateness of the novel word. We seek to evaluate this quantitatively and qualitatively in “extreme” cases where the con- text is either completely comprehensible or absent, and where the cognateness information is either present or absent. In …

Content truncated to 500 words