Using ChatGPT to Design Language Material and Exercises

by Frederick Poole, Michigan State University

As language teachers, it is important to always be on the lookout for new and effective ways to teach our students. One tool that can be particularly useful in this task is ChatGPT, a language model developed by OpenAI.

ChatGPT is a language model that has been trained on a large amount of data and can generate coherent and natural text in response to a series of questions or prompts. This makes it a valuable tool for language teachers, as they can use it to create realistic dialogues and scenarios in the language they are teaching.

In addition, ChatGPT is able to respond to questions in real-time, allowing teachers to use it as a way to interact with their students and promote conversation in the language they are learning.

In summary, ChatGPT is a valuable tool for language teachers who are looking for new ways to teach and promote conversation in the language they are teaching.

Hold up!!! Something seems a bit off here right? “ChatGPT is able to respond in real-time, allowing teachers to use it as a way to interact with their students and promote conversation in the language they are learning.” It seems a bit odd that I would suggest using Artificial Intelligence (AI) to communicate with your students! Well, that six-sentence introduction above was generated by ChatGPT when I asked it to write an introduction for how ChatGPT could benefit language teachers. Although the second paragraph is a bit odd, the first part of the introduction is not half bad! So, while I would not recommend using ChatGPT to ‘promote conversation’ with your learners, there are a number of things that it can do for you that can significantly improve the efficiency of creating materials. In this article I will introduce some ways that ChatGPT can be used in your classroom, activities that leverage this tool, and finally some pitfalls and concerns that we should be aware of when using this tool. 

What Materials Can You Create with ChatGPT?

As mentioned above, ChatGPT can automatically generate text based on a prompt. In the last few weeks people have asked ChatGPT to write cover letters, rewrite lyrics to famous songs from a different perspective, analyze essays, and so much more. In fact, many people are using ChatGPT as a coding assistant to help them program apps faster. Some have even suggested that ChatGPT will replace Google as one of the most important tools for future generations. For language teachers, this tool can be used to create unique texts, generate comprehension questions and word lists, analyze student texts, and much more. 

Create Customized Texts Customized to Your Needs

Likely the most intuitive and useful function of ChatGPT is that it can create customized text in your target language. You can ask ChatGPT to write you a narrative in the first-person perspective about someone who bought a new house but does not like it as much as their old house in Spanish (or any other language). Then you can get more specific and ask ChatGPT to compare and contrast each room in the house. Table 1 shows the output for a prompt like this. Clearly this would need some editing, but much of the heavy lifting in creating this material is completed in mere seconds.

Image of Output from ChatGPT
Text Output	Me mudé a una nueva casa hace poco, pero no me gusta tanto como la antigua. La nueva casa es más grande, pero la antigua tenía más encanto y personalidad. La nueva cocina es más moderna y tiene más espacio de almacenamiento, pero la antigua tenía una estufa de gas y una isla en el centro. La nueva sala de estar es más espaciosa, pero la antigua tenía una chimenea y ventanas más grandes. La nueva casa tiene un jardín más grande, pero la antigua tenía un porche y un árbol viejo con un columpio. Aunque la nueva casa tiene algunas ventajas, todavía extraño mi antigua casa.
Translation	I recently moved to a new house, but I don’t like it as much as my old one. The new house is bigger, but the old one had more character and charm. The new kitchen is more modern and has more storage space, but the old one had a gas stove and an island in the middle. The new living room is more spacious, but the old one had a fireplace and larger windows. The new house has a bigger backyard, but the old one had a porch and an old tree with a swing. Although the new house has some advantages, I still miss my old house.

Table 1 – Comparing rooms in a house

Likewise you could ask ChatGPT to write a dialogue with two people arguing over the prices of coffee in Paris in French. Table 2 below shows the output for a dialogue about expensive coffee. 

Text Output

Personne 1: Je suis tellement en colère contre le prix du café à Paris ! C’est tellement cher ! Personne 2: Mais tu sais que Paris est une ville très coûteuse. Le coût de la vie est généralement plus élevé ici. Personne 1: Oui, je sais, mais c’est quand même déraisonnable. Pourquoi ne peuvent-ils pas avoir des prix plus abordables ? Personne 2: Je ne suis pas sûr, mais peut-être que c’est à cause de la qualité et de la popularité des cafés parisiens. Personne 1: Peut-être, mais cela ne signifie pas que nous devrions payer des prix exorbitants pour un simple café. Personne 2: Je suis d’accord avec toi. C’est vraiment dommage que les prix soient si élevés. Mais hélas, c’est la réalité de vivre à Paris.

Translation

Person 1: I am so angry about the price of coffee in Paris! It’s so expensive! Person 2: But you know that Paris is a very expensive city. The cost of living is generally higher here. Person 1: Yes, I know, but it’s still unreasonable. Why can’t they have more affordable prices? Person 2: I’m not sure, but maybe it’s because of the quality and popularity of Parisian cafes. Person 1: Maybe, but that doesn’t mean we should be paying exorbitant prices for a simple cup of coffee. Person 2: I agree with you. It’s really a shame that the prices are so high. But alas, that’s the reality of living in Paris.

Table 2 – French dialogue about the price of coffee

The text generation function becomes a powerful tool when you realize how many ways you can customize your text. You could generate multiple different genres of text around the same topic. You could ask ChatGPT to write an advertisement for a house, an email to a friend about a house, a dialogue about two people reviewing a house, and so on. This would allow you to very quickly provide multiple sources around a common topic. In the next section we will explore some of the ways that we can modify texts once we create them. 

Make More Texts Based on your First Text

Generating texts quickly and being able to customize the texts is a powerful tool for any language teacher, but once you have generated the texts you can also ask ChatGPT to make the text simpler (See Table 3) or more complex (Table 4). 

Simplified Version	Personne 1: Pourquoi les cafés à Paris sont-ils si chers ? Personne 2: C’est parce que la ville est très chère. Personne 1: Oui, mais ça ne veut pas dire que nous devrions payer des prix exorbitants pour un café. Personne 2: Tu as raison. C’est vraiment dommage que les prix soient si élevés. Mais c’est la réalité de vivre à Paris.
Translation	Person 1: Why are the cafes in Paris so expensive? Person 2: It’s because the city is very expensive. Person 1: Yes, but that doesn’t mean we should be paying exorbitant prices for a cup of coffee. Person 2: You’re right. It’s really a shame that the prices are so high. But that’s the reality of living in Paris.

Table 3 – Simplified version of French dialogue on coffee

Doing so would create three dialogues around the same topic at varying levels of difficulty. This would not only allow teachers to create materials for learners at different proficiency levels, but it could also be used in a tiered-reading approach in which learners start by reading the simplest version of the text and then move towards more complex versions, completing unique tasks with each version of the text.

Simplified Version

Personne 1: Je suis extrêmement consterné par l’époustouflant prix des cafés à Paris ! Cela est tout simplement inadmissible ! Personne 2: Mais vous devez être conscient que Paris est une ville très coûteuse. Le coût de la vie est généralement beaucoup plus élevé ici. Personne 1: Oui, je suis tout à fait au courant, mais cela n’excuse en rien leur iniquité. Comment peuvent-ils se permettre d’imposer de telles tarifications exorbitantes pour une simple tasse de café ? Personne 2: Je ne sais pas, peut-être que cela est dû à la qualité supérieure et à la popularité des cafés parisiens. Personne 1: Cela peut être une explication plausible, mais cela ne justifie en rien ces prix déraisonnables. Personne 2: Je suis totalement d’accord avec vous. C’est vraiment dommage que les prix soient si élevés. Malheureusement, c’est la réalité de vivre à Paris.

Translation

Person 1: I am extremely dismayed by the staggering price of coffee in Paris! It is simply unacceptable! Person 2: But you must be aware that Paris is a very costly city. The cost of living is generally much higher here. Person 1: Yes, I am fully aware, but that does not excuse their injustice. How can they allow themselves to impose such exorbitant pricing for a simple cup of coffee? Person 2: I do not know, perhaps it is due to the superior quality and popularity of Parisian cafes. Person 1: That may be a plausible explanation, but it does not justify these unreasonable prices. Person 2: I completely agree with you. It’s really a shame that the prices are so high. Unfortunately, that is the reality of living in Paris.

Table 4 – Complex version of French dialogue on coffee

Not only can we vary the difficulty, but also we could also ask ChatGPT to create three versions of the dialogue that are different but similar. Or we could get more specific and ask ChatGPT to have the two coffee drinkers argue about the taste of the coffee rather than the price. The point is that we can quickly create multiple texts that share similar vocabulary and similar structures but still vary enough to make reading the texts meaningful for learners. Further, we can recreate any of these stories into other target languages fairly easily. In Table 5 below, I asked ChatGPT to rewrite the dialogue in Mandarin Chinese. 

Text Output

第一个人：我真的很生气巴黎咖啡的价格！太贵了！ 第二个人：但是你知道巴黎是一个非常昂贵的城市。生活成本通常更高。 第一个人：是的，我知道，但还是不合理。为什么不能有更多的价格？ 第二个人：我不确定，但也许是因为巴黎咖啡馆的质量和受欢迎程度。 第一个人：也许，但这并不意味着我们应该为一杯咖啡付出高昂的价格。 第二个人：我同意你的看法。价格太高真的很糟糕。但遗憾的是，这是生活在巴黎的现实。

Table 5 – Mandarin Chinese version

It is important to note that the Mandarin version is clearly a direct translation and while it does a pretty good job, there are a few mistakes and odd translations. Most notably, many of the words used in the Mandarin Chinese translation are words that we see more often in written texts rather than in spoken language. I tried to get a more authentic Mandarin version by asking the ChatGPT to generate a dialogue about two Chinese speakers in Beijing arguing over the price of coffee, but regardless of the details that I provided, the stories were similar and the prevalence of written text persisted. This would seem to suggest that many of these stories are coming from a common corpus and are then translated into the requested language.

Generate Comprehension Questions, Expansion Questions, and Vocabulary Lists

Next we will look at some of the things that we can do with a text once we have generated our material. First, we can create a list of comprehension questions. I have generated some comprehension questions in English for this dialogue, but again these could be done in any language we choose. Table 6 shows the result of this prompt and it is important to note that while some of these questions may be too simple, they do give us a starting point.

Text Output

What is Person 1 upset about? Why does Person 2 think the cost of living in Paris is generally higher? What does Person 2 think might be the reason for the high prices of coffee in Paris?

Table 6 – Comprehension questions

We could also ask for open-ended discussion questions or even multiple choice questions (see Table 7) if we prefer. 

Open-ended Discussion Questions	What are your thoughts on the high cost of living in Paris? Do you think the quality and popularity of Parisian coffee justifies the high prices?
Multiple Choice Questions	Who is upset about the price of coffee in Paris? a. Person 1 b. Person 2 Why does Person 2 think the cost of living in Paris is generally higher? a. Because it is a popular city b. Because of the quality and popularity of Parisian coffee c. Because it is a costly city to live in d. All of the above

Table 7 – Additional questions

Finally, we can also generate a set of key words for a vocabulary list on the text that we just created (See Table 8). 

Text Output

colère (anger) prix (price) café (coffee) Paris (Paris) cher (expensive) coûteux (costly) coût de la vie (cost of living) élevé (high) déraisonnable (unreasonable) abordable (affordable) qualité (quality) popularité (popularity) exorbitant (exorbitant) dommage (shame) hélas (unfortunately) réalité (reality)

Table 8 – Vocabulary list

To recap, we can quickly generate any text we want in any language we choose and define a set of characteristics to customize the text we generate. Once we have generated the text we can modify the text by adding details, make the text more or less complex, make similar texts but with unique variations, or simply change the language of the text. Finally we can also create a series of questions and/or vocabulary lists to accompany the text. This all would be enough to make your lesson planning and material creation much more efficient. Before looking at some of the activities that we could design around this tool, I think it is first important to explore some of the concerns that are associated with using ChatGPT in the language classroom.

What Concerns are There about Using AI in the Classroom?

While this is an exciting tool, there are a few concerns to consider. Most notably is the lack of transparency around how the algorithm works in this particular tool. Oftentimes it is easy to assume that a tool like ChatGPT is neutral or devoid of any biases. But algorithms are designed by humans, and many of the Natural Language Processing (NLP) models are derived from a specific set of texts. Understanding which texts are being used to make predictions about language models and how the algorithm works would provide more insight into the implications of the texts being created.

The Algorithm is Not Perfect

It is also important to note as you have probably already noticed, these texts are good overall, but not perfect, and sometimes they are just wrong. This does not mean that we should not use this tool, but it does mean that we should check and edit any text before bringing it into the classroom. Further, much like with Google Translate or any other machine translation tool, we should be teaching students and teachers how to provide better prompts and then how to validate the results once they are provided. 

The Algorithm is Not Culturally Agnostic

While I am not 100% sure, I believe that much of the language produced in this tool comes from an English corpus and then is translated into the language specified by the user. This means that many of the texts will be derived from English pragmatic and cultural assumptions. Again, this does not mean that this tool should not be used, but that we should acknowledge this when creating and using these texts in the foreign language classrooms. I would suggest that these tools, much like other adapted reading resources, should not be used to replace authentic texts, but rather to provide our learners with scaffolding for authentic texts. For example, we could copy and paste an authentic text into ChatGPT and then ask ChatGPT to simplify it for us. Students could first read the simplified version in preparation for the authentic text. Finally, it is important to note that ChatGPT is in early stages of development. As more texts are added to the language model, we may get more and more accurate versions in other languages. 

Copyright Questions

Finally, with AI-generated content, we always have questions around who owns the rights to the materials created. If one uses ChatGPT to create a graded reader series, can this material then be sold? Do you own the material simply because you entered a prompt into the text generator? Who owns these rights? Many of these questions are still being discussed by the AI community and we are unlikely to have any clear answers or decisions anytime soon. These are questions and concerns, however, that should be considered when using any AI tool like this.

What Kind of Activities Can We Design around ChatGPT?

So far we have looked at how the teacher would use this tool to generate texts and some of the pitfalls and concerns associated with this tool. Next I will introduce some activities that could be done to both promote understanding of this tool and to leverage the affordances associated with ChatGPT.

Compare AI-Generated Texts to Authentic Texts

First, as was discussed in the prior section, many of these texts are generated from an algorithm that appears to be pulling from a large corpus of text. It is unclear if text generated in other languages is first generated from an English corpus and then translated, or if it is generated from the target language corpus. This clearly would be important for any teacher to know before using any texts like this in a classroom, but for intermediate and advanced learners, this provides a unique opportunity to explore what makes an authentic text in their target language authentic. Teachers could first identify an authentic text (e.g. a blog post critiquing service at a coffee shop), then the teacher could ask ChatGPT to reproduce the authentic text given as many details as possible. The in-class activity would then involve learners analyzing the two texts, identifying similarities and differences, and then having a discussion about what the ChatGPT algorithm got wrong and why.

Compare AI-Generated Texts to Student Texts

In addition to comparing AI-generated texts to authentic texts, students could also compare their own writing and compositions to AI-generated texts. This activity should be done with the knowledge that AI-generated texts may not be good models of authentic text, but that they likely are good models for accurate grammar use. With this in mind, this activity may be more appropriate for novice and intermediate learners. For instance, in the novice classroom, a student may be asked to write about their favorite drink and snack in a café. After writing their short essay, an AI-generated essay could be created, and the learner could compare and contrast their essay with the AI-generated essay and look for new vocabulary and/or structures that they did not use. Then the student would incorporate new vocabulary and structures into their own essay in a second iteration.

Compare Student-Generated Texts to other Student-Generated Texts

Finally, to help students understand how the ChatGPT algorithm works and to bring digital literacies into the foreign language classroom, teachers could assign students with a series of tasks that asks them to generate a series of texts from ChatGPT. For example, a teacher may ask students to generate the following texts:

1. Generate a Twitter post complaining about medical bills
2. Generate a blog post about when something goes wrong while traveling in Peru
3. Generate a funny story about an animal

After students generate the stories, the teacher may ask the students to pair up and compare the prompts that they used to generate the stories and the texts that were generated. Students could then have a discussion about how their prompts elicited unique responses and how they could adjust their prompts to get better responses, and finally about how ChatGPT interprets their prompts in general. This activity would lead to more exposure to unique texts, but it would also allow students to develop skills around this new technology while also questioning how it functions. 

Conclusion

ChatGPT is undoubtedly an exciting and powerful new tool for educators. The internet is currently exploding with peoples’ thoughts about creative and unique ways of using this tool. At the same time, there are also those suggesting that ChatGPT will only make detecting plagiarism more difficult or that it will put language teachers out of a job. Whether or not this tool will eventually replace Google or usher in a host of new problems remains unknown, but what is clear is that we are entering the age of AI. In this article I provided a list of ideas that illustrated some ways that AI can support your teaching, while at the same time addressing some real concerns that should be considered before adopting this tool.