Democratizing Language Learning using Machine Learning

Abstract

Most apps available in the market for learning a new language are severely limited in terms of the number of languages users can learn on the platform (“target language”), as well as the language in which users can receive instruction (“source language”). Users are also limited by the fixed sets of lessons or the curriculum provided by these apps. In this work, we present an app framework which allows users free choice of source and target languages, as well as the flexibility in learning any input word, phrase or sentence of their choice using machine translation, whose performance and coverage of new languages is continuously improving. The app provides real-time feedback on the correctness of user pronunciation for any input using a text-based similarity metric, and helps learners practice their pronunciation until they perfect it. The app also provides a conversation platform where intermediate and advanced learners can engage in simple, real-time conversations with a chatbot on topics they are most likely to engage in while learning a new language. The conversation platform uses machine translation and speech-to-text tools to convert user query in any source language into an English query, gets the chatbot response and converts it back to the source language. This simple and novel approach allows users to freely converse in any language they want to practice, while having the chatbot trained only on English language-based corpus. The chatbot also integrates a novel intent classification module that classifies user query into one of several available topics, thereby enabling the chatbot to continue conversation with the user in the same topic. Finally, the chatbot is also capable of integrating search capability for specific queries (e.g., weather) with the help of available public domain resources so that it can provide users with real-time updates for such queries, thus making language learning fun and interesting.