Although ChatGPT is a significant development in our relationship to Artificial Intelligence, giving an easy to access interface, many other related AI tools have been around in the language teaching world for a long time. A search for relevant articles reveals material that predates the launch of ChatGPT, as well as offering a range of very recent articles that have picked up quickly on different aspects of ways that ChatGPT itself can be useful, often combining existing tools with ChatGPT.
People who are language teachers have been exploring the use of computers in their various guises since the 1950s, and while people have looked for solutions that have offered independent computer-based training, the vast majority of work in this area has looked at how computers can assist the learning of languages with the teacher still playing a significant role in the process. My impression of the articles that I have reviewed leaves me thinking that so far, for the most part, Generative AI had not significantly changed this perspective.
Most of the academic articles that I reviewed fell into three topic areas: writing, speaking, and teacher support in designing and managing learning.
Writing
There are several ways that GenAI can support writing, and this may have implications for how publishers might modify existing tools to keep them competitive, or to start creating new tools that can be taken to market.
Reviewing written texts (marking) is a time-consuming task for all teachers and providing useful feedback that will help learners develop is challenging. This is as true for teachers working in the private language school sector where class numbers are quite low, but in many parts of the world, class sizes can exceed a hundred children and teachers will have more than one class to manage.
In an article that explores the potential role of ChatGPT in Automatic Essay Scoring (AES), Mizumoto and Eguchi (2023) argue that:
Employing AES with ChatGPT has several benefits, including shorter rating times and increased consistency of scoring. As such, this approach is an attractive alternative for teachers and researchers.” (p. 9)
It can also support teachers whose level of language may not be up to the task of providing effective feedback, as, “Language models such as GPT can best be understood as an ever-present, logical assistant.” (p. 11). In many parts of the world where English is taught, teachers’ language is very low level and GenAI tools may help to solve this problem.
Another area presented in articles was that of support to learners in their writing. Su, Lin and Lai (2023) explore the role of ChatGPT in supporting the writing of argumentative essays. This article explores some of the earlier tools that also played a role in second language writing and point out that tools like chatbots were very limited in their responses, these often being “pre-set” (p. 2) and often respond with the same answer to multiple questions. ChatGPT can be responsive and ‘learns’ as it ‘engages’ with people. However, as we are aware, ChatGPT can give biased, or inaccurate responses. I used ChatGPT when I first started exploring this topic area and spent a while pursuing an article that had a very tempting title, by key writers whose names I recognised, but which I eventually realised had been made up. Su, Lin and Lai (2023) argue:
ChatGPT can serve as both a writing evaluator that provides feedback to scaffold the structural and language aspects of argumentative essays, and a virtual peer that engages in conversation around the writing topic, troubleshoots the writing process, and offers tips to strengthen the dialogic aspect of argumentative writing. (p. 2)
This fits well with one of the current ELT methodologies for writing, what is termed ‘process writing’ in which writing is not seen simply as a one-off product, but an iterative process of refining ideas over time and coming to editing at the end before submission.
Su, Lin and Lai (2023) say that in order to avoid students over-relying on GenAI to produce texts, they are asked to create an outline of their essay independently, discussing the ideas before feeding them into ChatGPT to get feedback and further ideas. In order to review the essays ChatGPT can be fed with specific rubrics for evaluation: claim-evidence-reasoning (CER) or Task/ Topic, Audience, Purpose (claim, data, warrant, qualifier, rebuttal and backing). In the study ChatGPT was not consistent on language use and would need to be interrogated a number of times to get different outputs, which the students could then decide what sentences they preferred. However, ChatGPT is good at proofreading, checking grammar, etc.
Interrogating ChatGPT in this way allowed a dialogue to be built up that could be exported and become part of a reflective diary, so that students could see progress in their writing.
Working with ChatGPT in this interactive way and not simply asking it to produce an argumentative essay on a particular topic was seen as being very helpful and developing an important new digital literacy skill.
A more technical approach to writing involved teaching students to create their own text generating tools using the programming language Python (Woo, Guo & Susanto, 2023). The students created these on a tool called Hugging Face. The text generators could produce various output to support the writing of short stories, in the case reported. Students learned about how machines could produce natural language. The study found that the students used the tools to overcome writer’s block and to improve their stories.
Speaking
There appears to currently be less written about spoken language, although dialogue systems have become more common in daily life. A systematic literature by Zhao and Wibowo (2023) and which focuses on higher education suggests that such systems can provide positive improvements, particularly with students who are less confident. It is interesting to note that almost all of the studies cited are not from Europe or North America and don’t focus on activity post ChatGPT.
AI dialogue systems can be used to assist students in finding information, or providing feedback, and this allows teachers to focus on other important areas of development like critical thinking, or problem solving. AI systems can help to provide materials to the students that they are more interested in, thus promoting motivation. This provides a “more effective and engaging learning experience for each student.” (p. 22) At the same time, these systems can help develop a better understanding of learners’ needs and lead to better personalisation of the curriculum, something that mass systems fail to do.
Teacher support
Although in essence a good deal of what we have talked about already is about teacher support, there are quite a few articles that explore this topic specifically. Bonner, Lege and Frazier (2023) do a good job of showing the variety of ways that Large Language Models (LLMs) can support teachers, providing concrete examples. A summary of the topics they cover is:
- “Summarise and level texts for learners
- Automatically correct grammar and sentence mechanics
- Compose narrative writing prompts
- Create presentation notes
- Generate lesson ideas
- Level texts for testing or reading practice” (p.25)
They talk about the way that “LLMs allow for the off-loading of more mundane tasks (Pokrivcakova, 2019)”.
A lot of emphasis is put again on the value that such tools have in providing support for individualised learning.
LLMs […] can support the teacher in processing vast amounts of information about the students and their learning process, then use this information to support the creation of adaptive learning environments that are catered to the needs of the individual.”
For references to the full articles and other resources see here:
