HSE: AI Lingvo Neural Network to Assess English Proficiency

The main articles are:

• Neural networks (neural networks)
• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding

2024: Presentation of "Lingvo AI"

Lingvo AI Neuronet was developed jointly by the Center, artificial intelligence the School of Foreign Languages ​ ​ and the online campus. HSE The model is trained on the basis of thousands of expert assessments of oral and written texts. The system allows you to assess the ability to communicate orally and in writing on. English The HSE announced this on October 1, 2024.

Those wishing to learn their level of knowledge of the skills of speaking and writing are invited to complete a block of tasks: write an essay, answer interview questions and build a monologue. The time for completing tasks is no more than 60 minutes, passing the test is available on any device, receiving a result and concise comments on it takes 30 minutes.

This format allows you to representatively assess the range of skills most often used in oral and written communication. When assessing responses by the neural network, more than 45 parameters developed by experts from the School of Foreign Languages ​ ​ of the Higher School of Economics are taken into account. The result is determined on the basis of a comprehensive assessment of the lexical, grammatical, phonetic/stylistic aspects of speech, as well as characteristics such as connectivity and logic of presentation of thought, ability to describe, reason and narrative.

Based on the results of the tasks, the final score is determined and the language proficiency level from the initial (Elementary) to the free (Upper-Intermediate) is assigned. The level scale developed at the School of Foreign Languages ​ ​ of the Higher School of Economics, incorporated into the neural network, correlates with all international examination systems.

The project to create Lingvo AI is an example of adapting large fundamental models for a specific applied sphere - for education. We went from testing a number of scientific hypotheses to introducing models into the final product, available as an online service for self-testing English language proficiency. Our development is distinguished by the ability to evaluate at the same time not only text responses, but also audio responses, and based on a set of criteria to assess the final level of language proficiency. To do this, we needed to study a significant number of open solutions and, as a result, conduct further training on works tested by professional linguists, "said Alexey Masyutin, Operations Director of the Center for Artificial Intelligence.

The HSE also noted that it is planned to develop models for other foreign languages. This will help students of foreign languages ​ ​ to conduct self-assessment of knowledge, receive prompt feedback in the required amount and with due frequency, monitoring the development of their level of language proficiency.

The introduction of AI solutions into the process of assessing language skills radically changes the approach to learning languages, making it more comfortable. We are creating a tool that allows not only to record the current level of knowledge, but also to dynamically track progress, providing every person with the opportunity for regular and objective self-assessment, - said Sergei Roshchin, Vice-Rector of the Higher School of Economics.