Neural Network for Assessing English Language Proficiency Developed at HSE University
The AI Lingua Neural Network has been collaboratively developed by the HSE University’s AI Research Centre, School of Foreign Languages, and online campus. The model has been trained on thousands of expert assessments of both oral and written texts. The system evaluates an individual's ability to communicate in English verbally and in writing.
Individuals interested in assessing their speaking and writing proficiency are invited to complete a series of tasks, including writing an essay, answering interview questions, and composing a monologue. Completing the tasks takes no more than 60 minutes, and the test is accessible on any digital device. Results, along with brief comments, are available within 30 minutes.
This format enables a representative assessment of the range of skills most commonly used in oral and written communication. When evaluating responses, the neural network considers over 45 parameters developed by experts at the HSE School of Foreign Languages. The result is determined through a comprehensive assessment of lexical, grammatical, and phonetic/stylistic aspects of speech, as well as characteristics such as coherence, logical presentation of thoughts, and skills in description, reasoning, and narration.
Based on the results of the tasks, a final score is calculated, and a language proficiency level is assigned, ranging from elementary to upper-intermediate. The proficiency level scale developed at the HSE School of Foreign Languages and integrated into the neural network aligns with all international language testing systems.
'An AI model capable of evaluating oral and written tasks with open-ended responses remains a rarity in language education. It can be described as a cutting-edge category of AI-based assessment systems, as the neural network has been trained to identify a variety of complex indicators of speech behaviour to determine candidates' levels of communicative competence,' according to Ekaterina Kolesnikova, Head of the HSE School of Foreign Languages.
The model has been trained on thousands of expert assessments of oral and written texts produced by actual participants in academic communication. The process of ongoing training and refinement of the model continues. Currently, the neural network is in its beta stage, undergoing testing and calibration for further integration into the process of assessing student work as part of independent English proficiency examination. It is planned for the neural network to function as one of the experts.
'The project to develop AI Lingua is an example of adapting large foundational models for a specific applied field—education. We have come all the way from testing various scientific hypotheses to implementing models in the final product, which is available as an online service for self-assessing English proficiency. Our solution is distinguished by its ability to simultaneously evaluate not only text responses but also audio speech, producing a language proficiency level based on a comprehensive set of criteria. To achieve this, we studied a substantial number of open-source solutions and subsequently conducted additional training on works analysed by professional linguists,' explains Alexey Masyutin, Director of the AI Centre.
The HSE University's team also notes that there are plans to develop models for other foreign languages. This will assist foreign language learners in conducting self-assessments of their knowledge, receiving timely feedback with the necessary level of detail, and tracking their progress in language proficiency development.
'The integration of AI solutions into the language skills assessment process fundamentally transforms the approach to language learning, making it more convenient and user-friendly. We are developing a tool that not only records the current level of knowledge but also dynamically tracks progress, offering everyone the opportunity for regular and objective self-assessment,' emphasises Sergey Roshchin, Vice Rector of HSE University.
The beta version of the English proficiency self-assessment tool is available for free with a promo code sent via email upon registration.
See also:
HSE University and Yandex to Host International AI Olympiad for Students
The HSE Faculty of Computer Science and Yandex Education are launching their first joint AI competition, Artificial Intelligence and Data Analysis Olympiad (AIDAO), for students from around the world. Participants will tackle challenging tasks in science and industry and interact with experts from HSE and Yandex. The winners will receive cash prizes.
Winners of the International Olympiad in Artificial Intelligence Admitted to HSE University
In mid-August, Bulgaria hosted the finals of the first International Olympiad in Artificial Intelligence (IOAI) among high school students. The Russian team demonstrated excellent results, winning gold medals in the scientific round, silver medals in the practical round, and coming first in both rounds overall. This year two members of the Russian team were accepted into the programmes of the HSE Faculty of Computer Science.
Artificial and Augmented Intelligence: Connecting Business, Education and Science
The history of AI research in Nizhny Novgorod dates back to the 1960s and 1970s. Today, AI technologies, from voice assistants and smart home systems to digital twin creation and genome sequencing, are revolutionising our life. Natalia Aseeva, Dean of the Faculty of Informatics, Mathematics and Computer Science at HSE Campus in Nizhny Novgorod, discusses how the advancement of AI connects science, business, and education.
HSE University Leads the AI Alliance Ranking
The AI Alliance Russia has released a new ranking of Russian universities based on the quality of education in the field of AI. Similar to last year, HSE University has joined the leaders in A+ group alongside MIPT and ITMO. A total of 207 universities from 69 Russian regions participated in the ranking. In 2024, over 35,000 students were enrolled in AI-related programmes at these universities.
Reinforcement Learning Enhances Performance of Generative Flow Networks
Scientists at the AI Research Centre and the AI and Digital Science Institute of the HSE Faculty of Computer Science applied classical reinforcement learning algorithms to train generative flow networks (GFlowNets). This enabled significant performance improvements in GFlowNets, which have been employed for three years in tackling the most complex scientific challenges at modelling, hypothesis generation, and experimental design stages. The results of their work achieved a top 5% ranking among publications at the International Conference on Artificial Intelligence and Statistics AISTATS, held on May 2-4, 2024, in Valencia, Spain.
‘I Came Up with the Idea to Create an Application Useful for Practicing Physicians’
Dmitry Ryabtsev, a 2024 graduate of the master's programme at the HSE Faculty of Computer Science, created an AI-powered software service for ophthalmology during his two years of study. This product is now entering the market, and its developer plans to participate in establishing a working group on software engineering for medical applications at the HSE Faculty of Computer Science, with the goal of promoting more genuinely useful domestic projects. In an interview with HSE News Service, Dr Ryabtsev shared his story of how a professional doctor turned into a programmer.
HSE University and Sber Conduct Foresight in Artificial Intelligence
HSE University, in collaboration with Sber, have conducted a foresight study on artificial intelligence (AI). Its early results were discussed by the participants of a strategic foresight session on exploratory research in AI, held at the Coordination Centre of the Russian Government, headed by Deputy Prime Minister Dmitry Chernyshenko. The results from the foresight study will inform the Unified Research and Development Programme in the Field of AI.
White Papers of AI Conformity Assessment Published on HSE University Website
The Russian Technical Committee for the Standardization of ‘Artificial intelligence’ (TC164), together with the Chamber for Indo Russo Technology Collaboration and the RUSSOFT Non-profit Partnership of Software Developers, has published new White Papers related to Artificial Intelligence Conformity Assessment. It reflects the approaches to the standardization and ethical regulation of AI technologies in two pilot industries — healthcare and agriculture.
Space for Collaboration in Artificial Intelligence
The HSE Laboratory of Artificial Intelligence for Cognitive Sciences (AICS) has launched regular seminars, offering students and scientists from various universities and research centres the opportunity to share their latest research and discuss the most recent developments in artificial intelligence in a friendly and constructive atmosphere. The first seminar was held on May 15.
Scientists Propose Star-Shaped Diffusion Model
Scientists at the AI Research Centre and the Faculty of Computer Science at HSE University, the Artificial Intelligence Research Institute (AIRI), and Sber AI have come up with novel architecture for diffusion neural networks, making it possible to configure eight distinct types of noise distribution. Instead of the classical Markov chain model with Gaussian distribution, the scientists propose a star-shaped model where the distribution type can be selected and preset. This can aid in solving problems across various geometric modalities. The results were presented at the NeurIPS 2023 conference.