Neural networks have already established a firm foothold in language learning. They are used by learners to develop their skills independently and by teachers to make their work easier, whilst major EdTech companies and universities are integrating neural network-based tools into teaching processes to create new lesson formats.
And whilst some teachers use AI tools themselves, others view this technology with apprehension, fearing that ‘neural networks are taking away our livelihood’. After all, whereas previously only a few managed to master a foreign language entirely on their own, artificial intelligence now seems to have opened up such opportunities to anyone who wishes to do so.
I will try to summarise the specific tasks in language teaching where neural networks are already being used, and whether there is anything left for human teachers to do.
Conversation practice with AI
Previously, conversation practice required a teacher or a native speaker — now large language models (LLMs) can take on this role. For example, Duolingo has been integrating artificial intelligence into the learning process for over five years and has been actively developing speaking trainers featuring GPT-4 in recent years.
Fancy a break from lessons and work? Then le bandit slot can help you with that
The app now offers a subscription that includes three AI features:
‘Role-play’. The user selects a scenario and starts a dialogue with a character in the app. After the conversation, the neural network provides feedback — it evaluates the answers and shares recommendations for the future.
‘Explain my answer’. Users can receive personalised feedback during lessons — the AI explains why an answer is correct or incorrect, tailoring the explanation of the rule to the specific user.
‘Video call with Lily’. Lily is an animated character from Duolingo; you can now call her in the app to practise speaking. The neural network supports conversation on any topic and adapts to the user’s language level.
Knowledge assessment with AI
Assessment in foreign language learning is important both for the initial evaluation of language proficiency (for example, to determine which study group a new student should join and which level of the curriculum they should follow) and to check whether a student has acquired new knowledge during the learning process or whether they meet a specific language proficiency grade (when it comes to issuing a certificate to that effect).
In such assessments, the most costly stage for the organisers of the proficiency test is the speaking test.
It is conducted in the form of oral conversations, which allows one to understand, in real-life communication, how effectively a person has learnt to use vocabulary and grammar.
There are many students to assess, and paid teaching time must be allocated to communicate with each one — consequently, this proves resource-intensive for the educational institution. The implementation of AI helps to automate the process and make it less expensive.
Personalised learning through contextual lessons
Just a few years ago, language learning was a separate activity that had to be fitted into one’s schedule — now AI helps to do this without taking time away from other tasks. To this end, the Google Labs team has launched an experimental project called Little Language Lessons, based on the Gemini neural network. The main idea is to integrate practice into everyday life and make it personalised. The experiment features three formats that help users work through the material via mini-lessons:
Tiny Lesson. The idea is that the user selects a situation from those offered, and the neural network picks out relevant words and expressions, explains the grammar and voices the phrases (for example, advising how to politely ask for the bill).
Slang Hang. The AI generates natural dialogues using colloquial expressions and slang. The user hears how the language is used in real life and learns to speak less formally.
Word Cam. The user points their smartphone camera at an object — Gemini recognises it, says the word in English and suggests phrases containing it.
