Personalized Text Retrieval for Learners of Chinese as a Foreign Language

Proceedings of the 27th International Conference on Computational Linguistics, pages 3448–3455 Santa Fe, New Mexico, USA, August 20-26, 2018. 3448 Personalized Text Retrieval for Learners of Chinese as a Foreign Language Chak Yan Yeung and John Lee Department of Linguistics and Translation City University of Hong Kong chak.yeung@my.cityu.edu.hk, jsylee@cityu.edu.hk Abstract This paper describes a personalized text retrieval algorithm that helps language learners select the most suitable reading material in terms of vocabulary complexity. The user first rates their knowledge of a small set of words, chosen by a graph-based active learning model. The system trains a complex word identification model on this set, and then applies the model to find texts that contain the desired proportion of new, challenging, and familiar vocabulary. In an evaluation on learners of Chinese as a foreign language, we show that this algorithm is effective in identi- fying simpler texts for low-proficiency learners, and more challenging ones for high-proficiency learners. 1 Introduction Since extensive reading is beneficial for learning a foreign language, students are encouraged to read a wide range of texts, beyond their textbooks and graded readers. Reading materials for language learning need to be carefully selected. On the one hand, if the text is too difficult, it would hinder comprehension and leave the student overwhelmed and discouraged. On the other hand, if the text is too easy, it would not serve to stretch the student’s language skills. The best material should be challenging yet highly readable, containing just the right proportion of new, challenging, and familiar vocabulary. Graded readers created with specific language proficiency levels in mind alleviate this problem to some extent, but they provide only a limited amount of materials, and the grade levels are still coarse-grained. It often falls to language teachers to identify appropriate material, but they cannot do so for every individual student. This paper describes a personalized text retrieval algorithm that helps language learners select the most suitable reading material in terms of vocabulary complexity. The user first rates their knowledge of a small set of words, chosen by a graph-based active learning model. The system trains a complex word identification model (CWI) on this set. Most current CWI approaches assume vocabulary knowledge to be a dichotomy, labeling each word either as “non-complex”, if the user is familiar with the word, or as “complex”, if the user is not familiar with it (Paetzold and Specia, 2016; Ehara et al., 2014). There are, however, many dimensions of vocabulary knowledge (Richards, 1976). As our first contribution, we take one step in this direction by attempting to identify not only known and unknown words, but also challenging words. Further, we apply CWI to the text retrieval task, and measure its effect on the vocabulary complexity of retrieved texts from the perspective of a language learner. Specifically, the algorithm attempts to find texts with a minimum proportion of non-complex words and a maximum number of challenging words for the learner. As our second contribution, we show that the algorithm is effective …

Content truncated to 500 words