Formalizing Word Sampling for Vocabulary Prediction as Graph-based Active Learning

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1374–1384, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Formalizing Word Sampling for Vocabulary Prediction as Graph-based Active Learning Yo Ehara∗ National Institute of Information and Communications Technology ehara@nict.go.jp Yusuke Miyao National Institute of Informatics yusuke@nii.ac.jp Hidekazu Oiwa Issei Sato Hiroshi Nakagawa The University of Tokyo {oiwa,sato}@r.dl.itc.u-tokyo.ac.jp nakagawa@dl.itc.u-tokyo.ac.jp Abstract Predicting vocabulary of second language learners is essential to support their lan- guage learning; however, because of the large size of language vocabularies, we cannot collect information on the entire vocabulary. For practical measurements, we need to sample a small portion of words from the entire vocabulary and pre- dict the rest of the words. In this study, we propose a novel framework for this sam- pling method. Current methods rely on simple heuristic techniques involving in- flexible manual tuning by educational ex- perts. We formalize these heuristic tech- niques as a graph-based non-interactive active learning method as applied to a spe- cial graph. We show that by extending the graph, we can support additional function- ality such as incorporating domain speci- ficity and sampling from multiple corpora. In our experiments, we show that our ex- tended methods outperform other methods in terms of vocabulary prediction accuracy when the number of samples is small. 1 Introduction Predicting the vocabulary of second language learners is essential to support them when they are reading. Educational experts have been continu- ously studying methods for measuring the size of a learner’s vocabulary, i.e., the number of words ∗The main body of this work was done when the first author was a Ph.D. candidate in the University of Tokyo and the paper was later greatly revised when the first author was a JSPS (Japan Society for the Promotion of Science) research fellow (PD) at National Institute of Informatics. See http: //yoehara.com/ for details. the learner knows, over the decades (Meara and Buxton, 1987; Laufer and Nation, 1999). Ehara et al. (2012) formalized a more fine-grained mea- surement task called vocabulary prediction. The goal of this task is to predict whether a learner knows a given word based on only a relatively small portion of his/her vocabulary. This vocabu- lary prediction task can be further used for predict- ing the readability of texts. By predicting vocab- ulary unknown to readers and showing the mean- ing of those specific words to readers, Ehara et al. (2013) showed that the number of documents that learners can read increases. Word sampling is essential for vocabulary pre- diction. Because of the large size of language vo- cabularies, we usually cannot collect information on the entire vocabulary. For practical measure- ments, we inevitably need to sample a small por- tion of words from the entire vocabulary and then predict the rest. We refer to this sampling tech- nique as word sampling. Word sampling can greatly affect the perfor- mance of vocabulary prediction. For example, if we consider only short everyday general domain words such as “cat” …

Content truncated to 500 words