No Meaning Left Unlearned: Predicting Learners' Knowledge of Atypical Meanings of Words from Vocabulary Tests for Their Typical Meanings [Konferenzbericht] Paper presented at the International Conference on Educational Data Mining (EDM) (15th, Durham, United Kingdom, Jul 24-27, 2022).

Autor/in	Ehara, Yo
Titel	No Meaning Left Unlearned: Predicting Learners' Knowledge of Atypical Meanings of Words from Vocabulary Tests for Their Typical Meanings [Konferenzbericht] Paper presented at the International Conference on Educational Data Mining (EDM) (15th, Durham, United Kingdom, Jul 24-27, 2022).
Quelle	(2022), (8 Seiten) PDF als Volltext kostenfreie Datei Verfügbarkeit
Sprache	englisch
Dokumenttyp	gedruckt; online; Monographie
Schlagwörter	Language Tests; Vocabulary Development; Second Language Learning; Second Language Instruction; Artificial Intelligence; Definitions; Models; Prediction; Task Analysis; Classification; Item Response Theory; Semantics; Test Items; Factor Analysis; English (Second Language); Transfer of Training; Programming Languages; Computational Linguistics; Test of English for International Communication + Suchen Sie Ihr Suchwort? Language test; Sprachtest; Wortschatzarbeit; Zweitsprachenerwerb; Fremdsprachenunterricht; Künstliche Intelligenz; Begriffsbestimmung; Analogiemodell; Vorhersage; Aufgabenanalyse; Classification system; Klassifikation; Klassifikationssystem; Item-Response-Theorie; Semantik; Test content; Testaufgabe; Faktorenanalyse; English as second language; English; Second Language; Englisch als Zweitsprache; Training; Transfer; Ausbildung; Linguistics; Computerlinguistik
Abstract	Language learners are underserved if there are unlearned meanings of a word that they think they have already learned. For example, "circle" as a noun is well known, whereas its use as a verb is not. For artificial-intelligence-based support systems for learning vocabulary, assessing each learner's knowledge of such atypical but common meanings of words is desirable. However, most vocabulary tests only test the typical meanings of words, and the texts used in the test questions are too short to apply readability formulae. We tackle this problem by proposing a novel dataset and a flexible model. First, we constructed a reliable vocabulary test in which learners answered questions regarding typical and atypical meanings of words. Second, we proposed a simple but powerful method for applying flexible and context-aware masked language models (MLMs) to learners' answers in the abovementioned vocabulary test results. This is a personalized prediction task, in which the results vary among learners for the same test question. By introducing special tokens that represent each learner, our method can reduce the personalized prediction task to a simple sequence classification task in which MLMs are applicable. In the evaluation, item response theory (IRT)-based methods, which cannot leverage the semantics of test questions, were used as baselines. The experimental results show that our method consistently and significantly outperformed the IRT-based baselines. Moreover, our method is highly interpretable because one can obtain the learners' language abilities from the first principal component scores of the token embeddings representing each learner. [For the full proceedings, see ED623995.] (As Provided).
Anmerkungen	International Educational Data Mining Society. e-mail: admin@educationaldatamining.org; Web site: https://educationaldatamining.org/conferences/
Erfasst von	ERIC (Education Resources Information Center), Washington, DC
Update	2024/1/01