An Intelligent Interactive Support System for Word Usage Learning in Second Languages

Second-language learners typically encounter difficulty in learning how to use words properly. One reason for this is that words that express multiple meanings - that is, polysemous words - vary from language to language. For example, the word "figure" can mean either a number, an image, or a person. However, this is not the case in all languages. Although the word "figure" is often used with both meanings, some words have only rare usage cases. Thus, ideally, a second language vocabulary learner would benefit from the following learning aids. For words all meanings of which are frequently used, all of the senses of the word should be presented with a high learning priority, whereas senses of a word that are rarely used should be given a low priority. Furthermore, learners should be able to visually understand the semantic proximity of word senses. In this study, we propose an intelligent, interactive user interface to support learning such word usages in second languages. The proposed interface estimates the difficulty of the usage example of the word by estimating its exceptionality. Our method measures semantic closeness with contextualized word embeddings. The model also incorporates a deep anomaly detection model to measure the exceptionality of each usage example. Using our interface, learners of English as a second language (ESL) learners can learn about the semantic closeness between word usage examples and the exceptionality of the examples.