Native and foreign language orthotactic probability and neighborhood density in word learning

Laboratory studies on word learning in a foreign language (L2) have identified several variables involved in the learning process, key amongst them the orthotactic probability and neighborhood density of new words relative to learners’ native (L1) lexicons. More recently, learners’ sensitivity to orthotactic probability and neighborhood density relative to their developing L2 lexicons has come into focus. Past studies on word learning have largely focused on early stages of learning, in controlled studies spanning hours or days. Few studies have considered large corpora of ‘real-life’ learning data, spanning several weeks. In this study, we validate past findings outside of controlled laboratory conditions, by analyzing a dataset collected from Duolingo (Settles et al., 2018Settles, B., Brust, C., Gustafson, E., Hagiwara, M., & Madnani, N. (2018). Second language acquisition modeling. Proceedings of the NAACL-HLT Workshop on Innovative Use of NLP for Building Educational Applications (BEA). ), a popular language learning app. Effects of orthotactic probability and neighborhood density observed in controlled studies persist under uncontrolled, big-data conditions for learners of Spanish, but not French. As learning progresses, we observe a previously unreported reversal of the effects of L1 orthotactic probability and neighborhood density, challenging theoretical models of word learning. Finally, we confirm the importance of orthotactic probability and neighborhood density relative to learners’ developing L2 Spanish lexicons, lending support to theories which posit that the same processes underly both L1 and L2 acquisition.