Richard Good
In the world of language teaching there’s one buzz phrase on everyone’s lips at the moment: Artificial Intelligence.
Language teachers can now create course materials – texts and quizzes and so on – with just a few prompts to either ChatGPT or Gemini. So if your teacher suddenly turns up to class with a stack of multiple choice questions, you can be certain they’ve got the bug.
But what about AI for learners? The big challenge, aside from the ethical agility needed to use software built on copyright theft, is unreliability.
You only need to spend a little time playing around with Large Language Models to catch them give blatantly false answers. For grammatical queries, that’s quite an issue.
They are though pretty good with vocabulary. And if you keep your AI assistant on a tight rein, it can be very helpful.
A test case
For my test case, I wanted to learn how to talk about Rugby Union. This is probably not your target, but for the purposes of argument, just replace Rugby Union with the speciality of your choice.
I only have a passing interest in Rugby Union – that’s why work needed to be done. But it’s a subject that comes up in conversation and I wanted to be able to discuss the game when it does. A perfect job for Gemini!
I would create myself a rugby lexicon.
The method
It helps when building vocabulary to start with expressions you’ve actually heard yourself (if you want to speak) or read yourself (if you want to write).
So I began by watching a game of rugby and noting down words or phrases that caught my attention. I use Workflowy as my notebook. By the end of the 80-minute match I had around 50 expressions noted.
From my list, I worked out a typology. It evolved as I went on, but from the start there were two broad categories of words. Nuts and bolts: players’ positions, the areas of the pitch, types of action, the rules. Meta vocabulary : words to analyse strategy, describe actions and emotions.
For the nuts and bolts vocabulary I would simply ask my assistant to give me lists of words. For the meta vocabulary, I would use the snippets of language that I’d noted down and ask the assistant to give me example sentences containing those words. This would be a way of checking that I’d heard correctly and also a way to flesh out the meaning of each expression.
I’d add the results to Workflowy to create my lexicon.
The experience
The job took me two hours in total. But I was experimenting as I went along and it would no doubt take less time in the future.
Creating the nuts and bolts list of words was easy and with just a few prompts I had a complete list for each of my categories. ChatGPT proved much better than Gemini at following instructions and was the tool of choice.
The meta-level of vocabulary took longer. For each key phrase, I asked for five example sentences so that I could pick the best one or two for my examples. For this exercise, the more verbose Gemini proved a better tool than the free version of ChatGPT (3.5) that I was using.
GenAI let me down in two types of cases. There were words that I’d heard that Gemini claimed didn’t exist (le fermé – meaning the blindside, for example). I had to double-check on the internet to be reassured that I’d heard right and Gemini was wrong.
And there were words that I had transcribed wrongly (tout pareil difficile instead of tout paraît difficile for example) that Gemini took as being valid. The result was gibberish examples of my own gibberish.
The Conclusion
At the end of my two hours, I’d built myself a satisfying 200-word lexicon of rugby. In the process of compiling it, I was able to absorb a lot of the words.
To make words stick, you really need to repeatedly use them over time. But in the immediate wake of the exercise, I was able to describe actions in the game that had previously left me stumbling for words. When I listened to the commentary of a subsequent game, I was able to pick out many words which had previously floated straight over my head and also to pick up some new ones that had slipped through the net first time round.
Could I have achieved this result more quickly any other way? There are French rugby lexicons out there on the internet already, but none quite like my new one: not formatted how I wanted, nor targeting words that had caught my attention. Had I been a real rugby fan, I might have read match reports over the years, and no doubt I would have picked up much the same vocabulary. But for a quick hit instant lexicon, this felt like as good an experience as possible.
It requires a certain level of patience (and language knowledge) to correct the mistakes of GenAI, so this isn’t really an exercise for beginners or casual learners with little time to invest.
It also depended on me being able to transcribe words that I heard. If that proves too difficult, there are automatic transcription tools (transkriptor, for example) which do the job very well.
No doubt in a not too distant future we’ll be able to ask AI to “give me a lexicon of my subject” and get a good result instantly. But I’m not sure this would necessarily be an improvement. The time spent constructing one’s own list is a valuable memorisation exercise in itself.
