This post is for teachers just finding their way with AI - like me! Not written or checked by AI.
Over the last year I've got into the habit of using AI, mainly Chat GPT, occasionally DeepSeek, for creating resources for frenchteacher.net. I thought I would share with you my thoughts so far. And these are overwhelmingly positive.
Speed and productivity
By far the most useful aspect for me is the speed at which I have been able to create worksheets and lesson plans. I am more productive as a result. In all cases, AI has not done anything I couldn't have done myself, but it has done it all so much more quickly. The best example I can think of is the ability to create multi-choice questions, which I have usually avoided since are very time-consuming to write. It's true that the options provided may not be the most subtle or create 'plausible distraction' - a requirement of assessment materials. But for my purposes, the MC questions produced are at the right level and provide the added comprehensible input I like to see on my worksheets and in lessons.
Lesson design
The last point is significant. A good worksheet, in my view, should reflect sound pedagogical practice, so I try to keep the emphasis on providing input and interaction, the two key ingredients for acquisition to occur. In addition, I like the sequence of exercises, typically starting with an aural or written text, to build in repetition and to proceed from easier to harder. AI can plan for this if you ask for a resource or whole lesson plan. In recent months I have added quite a few whole lesson plans to my site.
AI has allowed my to add extra exercises very quickly to older resources, creating recycling of language. My go-to prompts include: write a text, create comprehension questions, create a matching task, create a true/false/not mentioned task, create a gap-fill, create a 'tick the correct statements' task, create a transformational drill, create a vocabulary list on (topic), create four short texts to recycle the same chunks (narrow reading), create a second text with 10 linguistic or factual differences (for a 'Spot the differences' or 'faulty transcript' listening task, create general discussion questions based on the topic at hand.
Creativity
This is a well-worn debate, I believe. Does AI make you more creative, or less? I have found that the material produced by various prompts can spark off more ideas of my own. For example, if I were to ask for an information-gap activity, the example produced might be run-of-the-mill, but it gets me to think of a better alternative. Most of the time the lesson ideas generated are not very original, since they are sources from sites which often share tried and tested activities, plans and games.
I have used Chat GPT to help me write blog posts. On one occasion, I asked it to write a blog based on a number of points I wanted it to include. I then ended up editing it a lot and adding more ideas. But the initial AI effort got me thinking further. I would baulk at getting AI to just write a blog for me. There is a fine line here - we use review tools in Word, we can check content with AI, and we can get it to help us write material, but I would not share something which is not my own. If I were to, I would explicitly state that AI wrote it. On worksheets I often acknowledge AI use.
If I tend to reproduce activities I am familiar and comfortable with, this has nothing to do with AI. I would have done that anyway.
Accuracy
With its well-known hallucinations, AI is an unreliable encyclopedia, but it very good at generating accurate language. Accuracy is far from the be-all-and-end-all, of course, but I do like my resources to be accurate, especially when teachers are paying for them. I sometimes find errors in older reources I have written, which is pretty annoying, but now I can be almost 100% certain worksheets will be accurate. Where I have to edit, it is not a question of linguistic accuracy, it's more about particular choices of vocabulary or the way a sentence is phrased. I have to use my knowledge and experience to make adjustments.
One aspect of accuracy worth mentioning is this: if you prompt AI to produce a task-based lesson plan, what it comes up with is not necessarily very task-based. It may have strong elements of PPP or just be a loosely communicative task, without any obvious real-life authenticity. (There is debate in the literature, in any case, about what constitutes a 'task'.)
Hitting the right level
Because you can prompt the CEFR level for any resource, you can generate texts and activities at the right level for classes. For resources from Y9-Y11 I usually include the prompt Level A2, while for A-level resources I ask for Level B1. (Worth noting that many UK teachers are not very familiar with the CEFR framework, but you can use the prompt GCSE (Foundation or Higher) to get resources at a suitable level.)
If you prompt Chat GPT to create a text using only words from the 2000 most used in French, it does a good job. This may be useful for teachers teaching GCSE and who wish to avoid low-frequency words. (This would not be a massive priority for me.) Below is a paragraph Chat GPT made about hobbies when I prompted as follows:
"Can you write a text of 150 words for me in French about hobbies only using the 2000 most frequently used words?"
J’aime passer mon temps libre à faire des choses que j’aime. Le week-end, je lis souvent des livres ou je regarde des films. Parfois, je vais au parc avec mes amis pour jouer au ballon ou simplement marcher. J’aime aussi écouter de la musique. Cela me détend et me fait penser à autre chose. Quand il pleut, je reste à la maison et je dessine ou je joue à des jeux sur mon téléphone. Mon frère préfère faire du sport. Il joue au football tous les samedis. Ma sœur aime cuisiner. Elle prépare souvent des gâteaux pour toute la famille. Chacun a ses propres loisirs, mais ce qui est important, c’est de faire quelque chose qui nous rend heureux. Avoir un passe-temps permet de se sentir bien et de ne pas penser au travail ou à l’école. C’est une bonne façon de profiter de la vie.
For multi-choice, this is an example of a prompt I have used:
Give me six multi-choice questions in French based on the following text. Three options in the form of short sentences to complete. Level A2. List the answers in a row at the end.
I have noticed a tendency for Chat GPT to favour answer 'b', so have to reprompt to stop it doing that.
The need to edit
It's worth a reminder that material produced by AI always need checking for potential bias, lack of inclusivity or diversity. It can produce content with gender bias, for example mums doing the washing or dads being engineers, that sort of thing. If you prompt AI to avoid this, it will play ball, so I don't find this a major issue at all.
Comprehension questions, true/false and the like need checking since they may be too easy or too hard, or may use inversion question forms you may wish to avoid. Teachers would do well to use their knowledge of the class/students in editing material, to ensure, for example, that students are hearing or reading previously encountered language.
Be precise with your prompts
If you specifiy exactly what you want and how you want it laid out, for example answers at the end rather than embedded within the questions, then you will get a version which is easier to copy into Word. Chat GPT will often ask you if you want a worksheet format specifically for students, but I haven't taken advantage of this very much. I tend to copy and paste into Word then play with the layout myself. (I have noted my tendency to want to rely on my own judgement more than I need to - AI can certainly do more than I have asked of it so far. It's early days.) In general, I would that it is wise to be very precise with requests. When the result is not want you want, you end up having a dialogue with the AI which feels quite a like a human conversation. I find myself saying 'please' and 'thank you' which is a bit weird, isn't it?
Images
I did make use of AI images when writing our two GCSE French workbooks last year, but since then I have not done so. When you create images, you need to check for accurate detail (famously the number of fingers people have). Human images portray classically 'good-looking' white humans, so I tried to get round this with our GCSE books by prompting for different ethnic types, skin colours and people wearing glasses and veils, for example. Again, getting the prompts right is key. I found it hard to generate a greater variety of physical types.
Clearly, if Pixabay cannot produce an image to fit your needs, AI could do so. For example, scenes of people or situations to generate storytelling opportunities.
Music
I have dabbled with creating language learning songs in French using Suno and Songwriter. Once again, getting the prompts right is crucial. So far I have not published anything created since free versions do not allow for commercial use (which is a little ironic since AI mined published material to create its music generators!). I can report however, that the clarity of vocals is good and there is potential for novice-level songs on basic vocabulary. I would not use AI music for any other purpose, however, because of the lack of cultural authenticy.
Copyright
Here is a typical prompt I have used:
Please summarise in French this article. Length 400 words approximately, level B1.
So this would be an article I have found online, possibly copyright. By reworking the text, I am, I believe, strictly speaking breaking copyright. I justify this to myself, rightly or wrongly, by saying that I could have summarised the information in the article myself and shared it with attribution (commercially). Chat GPT did the job for me. In a resource based on a text of this type I add "Information from...". The whole question of copyright with AI is a mammoth one of course.
So there we are. That's where I am at so far. I am not an AI hater by any means. In discussions about this, I have found myself saying it is good where it fits. It's good at reading medical scans, it's good at writing texts, it's good at generating images and songs, it is (apparently) pretty good at computer coding. It is not bad at sourcing information, but makes mistakes and makes stuff up.
Comments
Post a Comment