In this study, we will examine the framework of an advanced online language learning support system utilizing AI that grows with people, specifically an AI learning support system with English conversation as the target domain. We will conduct large-scale data collection, prototype construction, and evaluation. English language learning in Japan is at a major crossroads. Since 2020, there has been a growing awareness in the Japanese government of the importance of English language learning, as indicated by English exams for university entrance expanding to include four skills (reading, writing, listening, and speaking) and the introduction of elementary school English classes that progress yearly. According to the recommendations of the Ministry of Education, Culture, Sports, Science and Technology, the future of English language education involves not just ensuring the acquisition of basic word and grammar knowledge, but also the ability to communicate with others of diverse backgrounds.
In the early 1990s, the developer of the current proposal, Waseda University, developed Tutorial English, a communication-oriented English conversation class centering around Professor Michiko Nakano, an emeritus professor of the Faculty of Education from the late 1990s to the 2000s.
The program is based on the generic language competency criteria of the Common European Framework of Reference for languages (CEFR). The basic language communication skills consist of three perspectives: “Linguistic abilities such as grammatical and linguistic knowledge,” “social linguistic capabilities employed in interpersonal interactions, such as honorifics,” and “linguistic abilities for the composition of communicative content.” Based on these perspectives, comprehensive conversation tasks organized in terms of understanding, generation, interaction, and facilitation aspects are specifically defined as achievement items. the CEFR can be said to be a long-standing purveyor of language education wisdom in the EU, which advocates plurilingualism and pluriculturalism, based on the assumption that people with different backgrounds exist.
However, the critical problem in real-world English language education remains the acquisition of grammar and vocabulary knowledge. Moreover, with respect to the ever-increasing need for the development of English communication skills to interact with other cultures, the English speaking education field (including elementary, junior high, high school, and university, as well as general English conversation classes) lacks the necessary supervisory resources. This resource shortage is expected to become even more serious due to the aging society and increasing sophistication of supervisory content. Furthermore, since the teaching of such English conversation is subject to numerous inherent issues regarding the knowledge and experience of each individual leader, both students and educators are expected to find value in AI-based assistance for advanced English conversation learning.
In the CEFR, English speaking ability is largely explained in terms of elements such as range, accuracy, fluency, phonology, interaction, and coherence. Of these, the interaction aspect is particularly important for interpersonal conversations, and there is a scarcity of automated assessment tests. Items included in interaction competency include the generation and control of topics, turn-taking, listening attitude, and conversation structuring capabilities. In the CEFR, operational capabilities relating to interaction capabilities are defined as phenomena that can only be observed if another party is present.
A typical system for practical automatic assessment of English proficiency is to have students think for a while, then speak for about a minute, extract sound and linguistic features from the speech, assess their abilities, and present the results to the candidate. However, as it was often assessed by recording individual voices, in most cases, it was not an adequately deep assessment of interpersonal communication skills such as interaction capability. While many Japanese students have certain skills in listening and reading, huge individual differences exist with respect to the ability to communicate naturally with others in English. It has been reported that the mismatch between actual abilities and minimum abilities required in classes causes problems both in terms of learning efficacy and class progression. Our preliminary studies have demonstrated that there is a gap between regular listening and reading test results and true speaking ability. In general, it has been found that class allocation on reading and writing test results alone tends to yield allocation to higher-level classes than the actual overall English speaking ability. Thus, there is a need to employ highly interactive conversation agents in English conversation education.
There is also a need to consider the historic, exponential growth of online education globally brought about by the advent of the COVID-19 epidemic in early 2020. The proposer, Waseda University, led the country in announcing that all university lectures would be delivered online. As outlined in the above section on Tutorial English, Tutorial English Online, an individualized English-speaking classroom based on the video conferencing function of an online classroom platform called Moodle, was rapidly prepared and began to be offered in May 2020.
In addition, the GIGA (Global and Innovation Gateway for All) School Concept, adopted by the Cabinet in December 2019 and led by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), stipulates that each child will be an offered an individualized and optimized educational ICT environment, with each student participating in compulsory education provided access to an information device and integrated high-speed, high-capacity communication network. As such, the necessity of online educational services, from compulsory education to everyday adults, will not wait for the debate to catch up.
In early 2019, the proposers, the Institute for Perceptual Information Systems, Waseda University, and Academic Solutions, Waseda University, Inc. (the leading industry-academia partner in this research), began investigating a project to support English conversation through conversational AI technology led by the Tutorial English developer, Professor Emeritus Michiko Nakano. In FY2019, we collected pilot data and organized language learning theory to find clues relating to the realization of AI in language education. Given this background, the gathering and analysis of large-scale data and the investigation of ways to comprehensively support English conversation education through various AI technologies have been occurring since the early stages of the plan to go completely online (due to the 2020 COVID-19 pandemic). The Tutorial English program at Waseda University delivered more than 60,000 hours of onsite English conversation classes to more than 12,000 students in the 2018 financial year alone. It is therefore expected that the realization of advanced AI systems will be greatly advanced and accelerated through the utilization of this environment.