ChatGPT French travel poster

A Second Shot at Perfect Posters – ChatGPT’s Image Tweaker

The big ChatGPT news in recent weeks is about images, rather than words. The AI frontrunner has added a facility to selectively re-prompt for parts of an image, allowing us to tweak sections that don’t live up to prompt expectations.

In essence, this new facility gives us a second shot at saving otherwise perfect output from minor issues. And for language learning content, like posters and flashcards, the biggest ‘minor’ issue – the poor spellings that crop up in AI image generation – makes the difference between useful and useless material.

Rescuing ChatGPT Posters

Take this example. It’s a simple brief – a stylish, 1950s style travel poster for France. Here’s the prompt I used to generate it:

Create a vibrant, stylish 1950s style travel poster featuring Paris and the slogan “La France”.

I wanted the text “La France” at the top, but, as you can see, we’ve got a rogue M in there instead of an N.

ChatGPT generated image of a French travel poster

To target that, I tap the image in the ChatGPT app. It calls up the image in edit mode, where I can highlight the areas that need attention:

ChatGPT image editing window

Then, I press Next, and can re-prompt for that part of the image. I simply restate the slogan instructions:

The slogan should read “La France”.

The result – a correct spelling, this time!

ChatGPT French travel poster

It can take a few goes. Dodgy spelling hasn’t been fixed; we’ve just been given a way to try again without scrapping the entire image. Certain details also won’t be retained between versions, such as the font, in this example. Others may be added, like the highly stylised merging of the L and F in the slogan (a feature, rather than a bug, I think!).

But the overall result is good enough that our lovely 1950s style poster wasn’t a total write-off.

Another case of AI being highly imperfect on its own, but a great tool when enhanced by us human users. It still won’t replace us – just yet!

Image tweaking is currently only available in the ChatGPT app (iOS / Android).

Neon robots racing. Can Claude 3 win the AI race with its brand new set of models?

Claude 3 – the New AI Models Putting Anthropic Back in the Game

You’d be forgiven for not knowing Claude. This chirpily-named AI assistant from Anthropic has been around for a while, like its celebrity cousin ChatGPT. But while ChatGPT hit the big time, Claude hasn’t quite progressed beyond the Other Platforms heading in most AI presentations – until now.

What changed everything this month was Anthropic’s release of all-new Claude 3 models – models that not only caught up with ChatGPT-4 benchmarks, but surpassed them. It’s wise to take benchmarks with a pinch of salt, not least because they’re often internal, proprietary measures. But the buzz around this latest release echoed through the newsletters, podcasts and socials, suggesting that this really was big news.

Tiers of a Claude

Claude 3 comes in three flavours. The most powerful, Opus, is the feistiest ChatGPT-beater by far. It’s also, understandably, the most processor-intensive, so available only as a premium option. That cost is on a level with competitors’ premium offerings, at just under £20 a month.

But just a notch beneath Opus, we have Sonnet. That’s Claude 3’s mid-range model, and the one you’ll chat with for free at Anthropic reports that Sonnet still pips ChatGPT-4 on several reasoning benchmarks, with users praising how naturally conversational it seems.

Finally, we have a third tier, Haiku. This is the most streamlined of the three in terms of computing power. But it still manages to trounce ChatGPT-3.5 while coming impressively close to most of those ChatGPT-4 benchmarks. And the real clincher?

It’s cheap.

Haiku costs a fraction of the price per token of competing models to developers. That means it’s a lot cheaper to build it into language learning apps, opening up a route for many to incorporate AI into their software. That lower power usage too is a huge win against a backdrop of serious concerns around AI energy demands.

Claude and Content Creation

So how does it measure up in terms of language learning content? I set Claude’s Sonnet model loose on the sample prompt from my recent Gemini Advanced vs. ChatGPT-4 battle. And the verdict?

It more than holds its own.

Here’s the prompt (feel free to adapt and use this for your own worksheets – it creates some lovely materials!):

Create an original, self-contained French worksheet for students of the language who are around level A2 on the CEFR scale. The topic of the worksheet is “Reality TV in France“.

The worksheet format is as follows:

– An engaging introductory text (400 words) using clear and idiomatic language
– Glossary of 10 key words / phrases from the text (ignore obvious cognates with English) in table format
– Reading comprehension quiz on the text (5 questions)
– Gap-fill exercise recycling the same vocabulary and phrases in a different order (10 questions)
– ‘Talking about it’ section with useful phrases for expressing opinions on the topic
– A model dialogue (10-12 lines) between two people discussing the topic
– A set of thoughtful questions to spark further dialogue on the topic
– An answer key covering all the questions

Ensure the language is native-speaker quality and error-free.

Sonnet does an admirable job. If I’m nitpicking, the text is perhaps slightly less fun and engaging than Gemini Advanced. But then, that’s the sort of thing you could sort out by tweaking the prompt.

Otherwise, it’s factual and relevant, with some nice authentic cultural links. The questions make sense and the activities are useful. Claude also followed instructions closely, particularly with the inclusion of an answer key (so often missing in lesser models).

There’s little to quibble over here.

A language learning worksheet created with Claude 3 Sonnet.

A Claude 3 French worksheet. Click here to download the PDF!

Another Tool For the Toolbox

The claims around Claude 3 are certainly exciting. And they have substance – even the free Sonnet model available at produces content on a par with the big hitters. Although our focus here is worksheet creation, its conversational slant makes it a great option for experimenting with live AI language games, too.

So if you haven’t had a chance yet, go and get acquainted with Claude. Its all-new model set, including a fabulous free option, makes it one more essential tool in the teacher’s AI toolbox.

An illustration of a cute robot looking at a watch, surrounded by clocks, illustrating AI time-out

Avoiding Time-Out with Longer AI Content

If you’re using AI platforms to create longer language learning content, you’ll have hit the time-out problem at some point.

The issue is that large language models like ChatGPT and Bard use a lot of computing power at scale. To keep things to a sensible minimum, output limits are in place. And although they’re often generous, even on free platforms, they can fall short for many kinds of language learning content.

Multi-part worksheets and graded reader style stories are a case in point. They can stretch to several pages of print, far beyond most platform cut-offs. Some platforms (Microsoft Copilot, for instance) will just stop mid-sentence before a task is complete. Others may display a generation error. Very few will happily continue generating a lengthy text to the end.

You can get round it in many cases by simply stating “continue“. But that’s frustrating at best. And at worst, it doesn’t work at all; it may ignore the last cut-off sentence, or lose its thread entirely. I’ve had times when a quirky Bing insists it’s finished, and refuses, like a surly tot, to pick up where it left off.

Avoiding Time-Out with Sectioning

Fortunately, there’s a pretty easy fix. Simply specify in your prompt that the output should be section by section. For example, take this prompt, reproducing the popular graded reader style of language learning text but without the length limits:

You are a language tutor and content creator, who writes completely original and exciting graded reader stories for learners of all levels. Your stories are expertly crafted to include high-frequency vocabulary and structures that the learner can incorporate into their own repertoire.

As the stories can be quite long, you output them one chapter at a time, prompting me to continue with the next chapter each time. Each 500-word chapter is followed by a short glossary of key vocabulary, and a short comprehension quiz. Each story should have five or six chapters, and have a well-rounded conclusion. The stories should include plenty of dialogue as well as prose, to model spoken language.

With that in mind, write me a story for French beginner learners (A1 on the CEFR scale) set in a dystopian future.

By sectioning, you avoid time-out. Now, you can produce some really substantial learning texts without having to prod and poke your AI to distraction!

There may even be an added benefit. I’ve noticed that the quality of texts output by section may even be slightly higher than with all-at-once content. Perhaps this is connected to recent findings that instructing AI to thing step by step, and break things down, improves results.

If there is a downside, it’s simply that sectioned output with take up more conversational turns. Instead of one reply ‘turn’, you’re getting lots of them. This eats into your per-conversation or per-hour allocation on ChatGPT Plus and Bing, for example. But the quality boost is worth it, I think.

Has the section by section trick improved your language learning content? Let us know your experiences in the comments!

An image of a robot struggling with numbreed blocks. AI has a problem with random ordering.

Totally Random! Getting Round AI Random Blindness in Worksheet Creation

If you’re already using AI for language learning content creation, you’ve probably already cried in horror at one of its biggest limitations. It’s terrible at putting items in a random order.

Random order in language learning exercises is pretty essential. For instance, a ‘missing words’ key below a gap-fill exercise should never list words in the same order as the questions they belong to.

Obvious, right? Well, to AI, it isn’t!

Just take the following prompt, which creates a mini worksheet with an introductory text and a related gap-fill exercise:

I am learning French, and you are a language teacher and content creator, highly skilled in worksheet creation.
Create a French worksheet for me on the topic “Environmentally-Friendly Travel”. The language level should be A2 on the CEFR scale, with clear language and a range of vocabulary and constructions.
The worksheet starts with a short text in the target language (around 250 words) introducing the topic.
Then, there follows a gap-fill exercise; this consists of ten sentences on the topic, related to the introductory text. A key content word is removed from each sentence for the student to fill in. For instance, ‘je —— en train’ (where ‘voyage’ is removed).
Give a list of the removed words in a random order below the exercise.

The output is very hit and miss – and much more miss! Perhaps 90% of the time, ChatGPT lists the answer key in the order of the questions. Either that, or it will produce feeble jumbling attempts, like reversing just the first two items on the list.

AI’s Random Issue

One prompt-tweaking tip you can try in these cases is SHOUTING. Writing this instruction in caps can sometimes increase the bullseyes. Put them IN RANDOM ORDER, darn it! It doesn’t help much here, though. It just doesn’t seem worth relying on Large Language Models like ChatGPT to produce random results.

The reason has something to do with the fundamental way these platforms function. They’re probability machines, guessing what word should come next based on calculations of how likely word X, Y or Z will be next. Their whole rationale is not to be random; you might even call then anti-random machines.

No wonder they’re rubbish at it!

A Road Less Random

So how can we get round this in a reliable way that works every time?

The simplest fix, I’ve found, is to find another, non-random way to list things differently from the question order. And the easiest way to do that is to simply list things alphabetically:

I am learning French, and you are a language teacher and content creator, highly skilled in worksheet creation.
Create a French worksheet for me on the topic “Environmentally-Friendly Travel”. The language level should be A2 on the CEFR scale, with clear language and a range of vocabulary and constructions.
The worksheet starts with a short text in the target language (around 250 words) introducing the topic.
Then, there follows a gap-fill exercise; this consists of ten sentences on the topic, related to the introductory text. A key content word is removed from each sentence for the student to fill in. For instance, ‘je —— en train’ (where ‘voyage’ is removed).
Give a list of the removed words in alphabetical order below the exercise.

The likelihood of this order being the same as the questions is minimal. Hilariously, AI still manages to mess this order up at times, adding the odd one or two out-of-place at the end of the list, as if it forgot what it was doing, realised, and quickly bunged them back in. But the technique works just fine for avoiding the order giving the answers away.

A simple fix that basically ditches randomness completely, yes. But sometimes, the simplest fixes are the best!

Random blindness is a good reminder that AI isn’t a magical fix-all for language learning content creation. But, with an awareness of its limitations, we can still achieve some great results with workarounds.

Does AI have a noun problem? Strategies for avoiding it.

AI Has A Noun Problem : Let’s Fix It!

If you’re using AI for language learning content creation, you might have already spotted AI’s embarrassing secret. It has a noun problem.

Large Language Models like ChatGPT and Bard are generally great for creating systematic learning content. They’re efficient brainstormers, and can churn out lists and texts like there’s no tomorrow. One use case I’ve found particularly helpful is the creation of vocab lists – all the more so since it can spool them off in formats to suit learning tools like Anki.

But the more I’ve used it, the more it’s become apparent. AI has a blind spot that makes these straight-out-the-box vanilla lists much less useful than they could be.

A fixation with nouns.

Test it yourself; ask your platform of choice simply to generate a set of vocab items on a topic. Chances are there’ll be precious few items that aren’t nouns. And in my experience, more often than not, lists are composed entirely of noun items and nothing else.

ChatGPT-4 giving a list of French vocabulary items - all nouns.

ChatGPT-4 giving a list of French vocabulary items – all nouns.

It’s a curious bias, but I think it has something to do with how the LLM conceives key words. The term is somehow conflated with all the things to do with a topic. And nouns, we’re taught at school, are thing words.

Getting Over Your Noun Problem

Fortunately, there’s therapy for your AI to overcome its noun problem. And like most AI refining strategies, it just boils down to clearer prompting.

Here are some tips to ensure more parts-of-speech variety in your AI language learning content:

  1. Explicit Instruction: When requesting vocabulary lists, spell out what you want. Specify a mix of word types – nouns, verbs, adjectives, adverbs, etc. to nudge the AI towards a more balanced selection. When it doesn’t comply, just tell it so! More verbs, please is good start.
  2. Increase the Word Count: Simply widening the net can work, if you’re willing to manually tweak the list afterwards. Increase you vocab lists to 20 or 30 items, and the chances of the odd verb or adjective appearing are greater.
  3. Contextual Requests: Instead of asking for lists, ask the AI to provide sentences or paragraphs where different parts of speech are used in context. This not only gives you a broader range of word types, but also shows them in action.
  4. Ask for Sentence Frames: Instead of single items, ask for sentence frames (or templates) that you can swap words in an out of. For instance, request a model sentence with a missing verb, along with 10 verbs that could fill that spot. “I ____ bread” might be a simple one for the topic food.
  5. Challenge the AI: Regularly challenge the AI with tasks that require a more nuanced understanding of language – like creating stories, dialogues, or descriptive paragraphs. This can push its boundaries and improve its output.

Example Prompts

Bearing those tips in mind, try these prompts for size. They should produce a much less noun-heavy set of vocab for your learning pleasure:

Create a vocabulary list of 20 French words on the topic “Food and Drink”. Make sure to include a good spread of nouns, verbs, adjectives and adverbs. For each one, illustrate the word in use with a useful sentence of about level A2 on the CEFR scale.
Give me a set of 5 French ‘sentence frames’ for learning and practising vocabulary on the topic “Summer Holidays”. Each frame should have a missing gap, along with five examples of French words that could fit in it.
Write me a short French text of around level A2 on the CEFR scale on the topic “Finding a Job in Paris”. Then, list the main content words from the text in a glossary below in table format.

Have you produced some useful lists with this technique? Let us know in the comments!

AI prompt engineering - the toolkit for getting better results from your platform of choice.

Better AI Language Learning Content with C-A-R-E

AI isn’t just for chat – it’s also great at making static language learning content. And as AI gains ground as a content creation assistant, prompt engineering – the art of tailoring your requests – becomes an ever more important skill.

As you’d expect, frameworks and best practice guides abound for constructing the perfect prompt. They’re generally all about defining your request with clarity, in order to minimise AI misfires and misunderstandings. Perhaps the most well-known and effective of these is R-T-F – that’s role, task, format. Tell your assistant who it is, what to do, and how you want the data to look at the end of it.

Recently, however, I’ve been getting even more reliable MFL content with another prompt framework: C-A-R-E. That is:

  • Context
  • Action
  • Result
  • Example(s)

Some of these steps clearly align with R-T-F. Context is a broader take on role, action matches to task and result roughly to format. But the kicker here is the addition of example(s). A wide-ranging academic investigation into effective prompting recently flagged “example-driven prompting” as an important factor in improving output, and for good reason: the whole concept of LLMs is built on constructing responses from training data. It’s built on the concept of parroting examples.

Crafting AI prompts with C-A-R-E

As far as language content is concerned, C-A-R-E prompting is particularly good for ‘fixed format’ activity creation, like gap-fills or quizzes. There’s a lot of room for misinterpretation when describing a word game simply with words; a short example sets AI back on track. For example:

– I am a French learner creating resources for my own learning, and you are an expert language learning content creator.
– Create a gap-fill activity in French for students around level A2 of the CEFR scale on the topic “Environment”.
– It will consist of ten sentences on different aspects of the topic, with a key word removed from each one for me to fill out. Provide the missing words for me in an alphabetically sorted list at the end as a key.
– As an example, a similar question in English would look like this: “It is very important to look after the ———- for future generations.”

This produces excellent results in Microsoft Copilot / Bing (which we love for the freeness, obviously!) and ChatGPT. For example:

Creating AI language learning content with Microsoft Copilot / Bing Chat

Creating AI language learning content with Microsoft Copilot / Bing Chat

Providing short examples seems like an obvious and intuitive step, but it’s surprising how infrequently we tend to do it in our AI prompts. The gains are so apparent, that it’s worth making a note to always add a little C-A-R-E to your automatic content creation.

If you’ve been struggling to get reliable (or just plain sensible!) results with your AI language learning content, give C-A-R-E a try – and let us know how it goes in the comments!

The Polish flag. Photo by Michal Zacharzewski from FreeImages

Język polski i ja / Polish and Me

Not long back, a lively online language learning debate caught my eye. It was around the unassailable prominence of English as a medium for discussion in the polyglot community, and the irony of this within a community of a hundred other choices. Where is the diversity, the German, Japanese, Polish, Spanish articles? After all, we are spoilt for choice.

Of course, it is hard to get round this – not least because we all speak a slightly different set of languages. So, at least for now, English looks to keep its place as the most inclusive choice of language for discussion.

That said, I would personally echo that hope to see more blog and social media content in the languages I learn. Above all, being a blogger myself, it seemed like a good cue to lend a little ballast to the non-English side of things, to be brave, to publish non-English content.

Safe, comfortable English is a difficult spot to get out of, though. As a native English speaker, the reason for my reticence is probably one shared by many of my fellow anglophone enthusiasts: fear of mistakes, of others simply doing it better. That kind of anxiety is self-fulfilling; keep your fledgling skills too tightly caged, and they might just wither away.

Luckily, the chance came along to do a bit of writing along these lines, but with support. That made all the difference.

Good Timing

By complete coincidence, my iTalki Polish tutor Jan set a very appropriate homework task for me recently – a simple blog post, in Polish, about my personal history of learning the language. Writing from experience, like diary-keeping, can be an effective way to engage with, recycle and strengthen your language skills. But in this case, it gave me the opportunity to create something original – and not in English – for Polyglossic.

Now, the natural thing to do would probably have been to do this in one of my stronger languages. German, Norwegian or Spanish. You could say that Polish was simply in the right place at the right time. However, maybe that makes it an even better candidate. My lagging Polish is crying out for a bit of extra writing practice.

Let’s overlook for a moment (pretty please!) the discrepancy of this preface to it in English. Hmm. But for a first non-English post in a site full of them, it only seemed fair – at least for the time being. Baby steps.

Finally, huge thanks to Jan for the prompt and the copious corrections to this during class. Check out his own blog, Polish with John, for some fantastic original resources for learners. Any remaining errors below are completely my own!

Język polski i ja

Na Początku

Interesuję się językiem polskim od wielu lat. W latach dziewięćdziesiątych słuchałem polskiej muzyki w radiu u polskiego sąsiada, Pana Wilsona (jego prawdziwego polskiego nazwiska nie znam) i bardzo chciałem się nauczyć tego pięknego języka.

Ale wtedy nie było łatwo uczyć się polskiego. W bibliotekach nie było wielu materiałów do nauki. Jeśli ktoś chciał się uczyć hiszpańskiego, francuskiego, niemieckiego, dostępna była masa materiałów i książek. Niestety do języka polskiego był tylko jeden, bardzo stary egzemplarz “Teach Yourself Polish”. Było to wydanie z lat czterdziestych oparte na starej metodologii. Zastosowana była metoda gramatyczno-tłumaczeniowa. Pięćdziesiąt lekcji gdzie student musi czytać przykłady, nauczyć się listy słów, a potem zrobić długą listę tłumaczeń. Wtedy uważałem, że to było zupełnie normalne, że tak po prostu uczy się języków. To był błąd.

Brak mówiących

Nie było dostępu do mówiących. Pan Wilson nie lubił mówić po polsku (był starym człowiekiem a miał tragiczną historię i złe doświadczenia z wojskiem), a wszystko, co robiłem, było tłumaczeniem zdań nie mających praktycznego zastosowania. Tak nie da się nauczyć języka obcego.

Nawet słownictwo nie miało sensu dla mnie – słowa z lat czterdziestych, słowa I zwroty takie jak porucznik, pułkownik, polsko-brytyjskie przymierze i tak dalej. Myślę, że książka została napisana dla żołnierzy, którzy pracowali w polakami po wojnie. Po prostu nie mi pasowała. Ciekawe słownictwo, oczywiście, ale nie bardzo przydatne – na początku tylko chciałem rozumieć polskie piosenki! Ale nie było innego wyboru.

Nowy Świat

Wiele lat później, świat się zmienił. Nie tylko jest więcej książek, a też więcej metod, szerszy dostęp do materiałów do mówienia i słuchania w internecie, wszystko, co by mi pomogło jak młodemu studentowi.
Wniosek jest taki: nie da się uczyć się języka obcego bez słuchania i mówienia. Sama książka nie wystarczy.

A classroom ready for teaching

Teaching to learn: boost your studies by helping others

The idea of learning through teaching is nothing new. We find the idea in an old Latin proverb, docendo discimus (by teaching, we learn), possibly handed down to us from Seneca the Younger. The premise is simple: being able to explain what we know turns that knowledge from passive into active smarts.

We might also argue that the skill of teaching is facilitating learning, rather than bound to the actual content of that learning. It’s not necessarily about what you know, but how well you can explain (and re-explain) material – even new material. In this light, a natural next question is: can we teach without being experts in that content already? And are there learning benefits for us in doing so?

Primary Languages

The Primary Languages model rolled out in many UK schools is a great example of learn-while-teaching. Many teachers are not language specialists, but rather using teaching materials that allow them to stay one step ahead of the students.

The very best materials, like Linguascope‘s elementary resources, are packaged like ready-made lesson plans, which can be reviewed before class and form a roadmap for the teacher. Great teaching in this context is the skill of presenting, explaining and reviewing content, even if you’re just a few steps ahead of your class.

Peer teaching

In the classroom setting, learning through teaching can be just as powerful between peers. Students may be tasked with learning material in order to teach it to other students, either contemporaries or those in lower year groups. The resulting ‘altered expectations‘ – the knowledge that you’ll have to teach the material you’re learning to others – transform motivations and sharpen focus on really understanding. Also dubbed the ‘protégé effect‘, educational scientists have noted how preparation to teach results in students spending longer on material. One study provides empirical evidence for this ‘teaching expectancy‘ effect.

The idea has achieved some institutional acceptance already; educationalist Jean-Pol Martin has helped to instill the Learning by Teaching (Lernen durch Lehren) model as a popular method in German schools. The modern ‘flipped classroom‘ also has elements of student-turned-self-teacher, too, reversing traditional roles.

Build teaching into your own learning

So, teaching as learning has a long pedigree, and already has some good traction in the real world. But what lessons can we take from this for our own language learning?

Bug friends and family

Share with friends and family what you’re learning. They don’t ‘do’ languages? Then break it down as simply as possible. Tell them about a quirk of your target language that you find unusual. Think you’ll bore them to tears? Then find some way to make it interesting to them. The more challenging, the harder you’ll have to think – and the more that material will stick.

To get the interest of family and friends, I’ve actively looked for things that will make them laugh in the past. Never underestimate the power of humour in learning! Funny-sounding words (Fahrt in German is always a good one), weird idioms (tomar el pelo – literally ‘pulling the hair’ for ‘pulling someone’s leg’ in Spanish) and other oddities speak to the imaginations of the most reluctant listeners. “You’ll never guess what the word for ‘swimming pool’ is in French…”

Find a learn-and-teach partner

You can go beyond sharing humorous factoids and foibles. Find a fully-fledged language partner – someone who is as motivated as you to learn the language – and devise a schedule where you take turns in teaching vocabulary or grammar points each week. You’ll be activating those ‘teaching expectancy’ effects that worked so well in the classroom studies above.

Create resources for other learners

A revision technique I learnt as a student was to condense important points into simple explanations for others. If you can explain something complicated in a new, simpler way, then it’s a good sign that you really understand it.

Something I’ve been doing recently is to revisit my Castilian by creating Spanish revision videos for beginners. It’s been a form of revision for me, activating old knowledge bases that were starting to fade through lack of use. And because of the interconnected nature of knowledge (the neural networks of our brains), switching on a few buried memories triggers and refreshed many more connected informational blobs.

It’s easy to find a platform to share your homemade revision resources these days. Starting a YouTube channel or a Facebook group could be the perfect platform for your own learning through teaching.

Teaching is connecting

At the heart of it, learning through teaching embodies what languages are all about: making connections, building bridges. Try working some of these ideas into your own learning, and enjoy the social splashback!