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).

Lots of owls - not quite the Duolingo one, but they're looking strict nonetheless!

Going Cold Owl (Once More, With Feeling)

Well, friends – I finally did it. I beat my Duolingo Leagues addiction.

Yes, it’s been months in the making. It was back in May last year that I wrote about that tipping point where Duolingo’s gamification switches from incentivising learning, to subverting it. After more than two years in Diamond, my main Duo activity had become all about chasing points, rather than finishing courses.

I’d even sussed out a ‘good’ trick – an easy lesson on Hindi alphabet characters – that I could mindlessly run through to amass points in my idle moments. Then I’d justify it with the two or three daily lessons I’d do that were relevant to my language goals.

Only it wasn’t good at all, in the grand scheme of things. What is was, was a great way to de-direct my learning time on what is otherwise a nice little supplementary app for my language plan.

Surely my idle moments were worthy of something more.

Letting Yourself Fall (and Trusting in a Soft Landing)

Gamified apps like Duolingo are adept at creating a sense of worth out of otherwise valueless tokens. A league title here, an achievement there – it feels good to notch these accolades up. But the real value of a language learning app is never the dressing, it’s the language-y filling in the centre. When it stops being about that, it’s time to stop.

So, in a final moment of take-the-plunge bravery, I stopped chasing the demotion cut-off last week. I let myself fall. And I won’t pretend I didn’t have several fleeting moments of panic when I thought – it’s not too late, I can still catch up!

But nope – I was resolute. And when, on Monday morning, I woke up to – gasp – Obsidian and not Diamond, do you know what? It felt liberating.

I’d broken the cycle.

didn’t feel like I’d left an exclusive club. I didn’t feel like I’d lost anything valuable. I just felt relieved.

Duolingo – Done Right (aka all things in moderation)

Since then, I’ve kept up my lengthy Duolingo streak. But not through mindless tap-tap-tap lessons – instead, through a couple of Swedish and Gaelic lessons to progress through the courses. Just as it should be.

If you’re in this autopilot pattern yourself, ask yourself: what do I gain? Reframing any kind of daily screen-addictive behaviour in this way is the first step in changing it.

Keen on more tips for breaking digital additions? Check out:

An illustration of a robot taking a picture of a book page, to illustrate AI image analysis in the context of language learning course books.

AI Image Analysis for Language Learners : Your Course Book Assistant!

Image upload and analysis is one of the most game-changing recent additions to AI platforms. Combined with a knack for text recognition, it’s possibly one of the most revolutionary for language (and other!) learners, too.

In short, if it’s on a page, you can now get it into AI and do things with it. Because of this. image analysis has huge potential for extending, and breathing new life into printed materials, producing the very best synthesis of old and new tech.

A screenshot of AI chat in the Bing App, with an arrow showing the 'upload an image' function.

The image upload icon in the Bing app.

At its very simplest, it’s a handy summary and explanation tool. Just upload your page image, and prompt:

Analyse this page from my [LANGUAGE] course book. Summarise it in a few short bullet points I can use for revision.

Useful in its own right. But with some extra prompt magic, you can produce individually tailored support material on the fly – material that will help you to delve really deeply into those language learning texts, making it work for you.

Let’s see what it can do for starters!

Working with Vocab Lists

Vocab in Context

Take the most conventional form of book-based, language learning data. Most course books have vocabulary lists and glossaries of words in the current chapter top. But beyond the dialogues or passages they are attached to, there’s rarely any other in context use of them.

Personally, I find it really helpful to see individual items embedded in sentence examples as an aide memoire. I usually seek them out in mass sentence banks and other manual-search online resources.

Even easier with AI:

Analyse this entire page from my [LANGUAGE] course book, noting all of the vocabulary items internally. Then, create a useful, practical sentence using each and every item. The sentences should relate to real-world contexts where possible. Make sure you include every single entry – don’t leave any out. Constantly double-check that the language is natural-sounding and grammatically correct. Output them in table format listing the word, your sentence and an English translation of that sentence.

ChatGPT Plus analysing a page of Swedish vocabulary.

ChatGPT Plus analysing a page of Swedish vocabulary.

ChatGPT Plus analysing a page of Swedish vocabulary.

The trick here is the analyse the entire page instruction. LLM / AI platforms tend to take shortcuts when working with lists, sometimes skipping list items. Adding this stipulation is great at keeping it on track!

Rationalising Vocab Lists

You can also sort such material in an order that works better for you. For instance, I work best with vocab when I classify it first, be that by parts of speech, topic or otherwise. AI makes light work of it:

This is the material I’m currently studying in [LANGUAGE]. First, analyse the entire page, noting all of the vocabulary items listed. Then, rewrite that list, grouping the items by their grammatical part of speech and in alphabetical order. Where the word isn’t in its simple dictionary form, provide that too. Include any entries you couldn’t categorise at the end. Double-check throughout the process that a) you haven’t left out any items, and b) that your categorisation of each item is correct. If you detect errors, start again.

ChatGPT Plus analysing a page of Swahili vocabulary to create model sentences for context.

Microsoft Bing analysing a page of Swahili vocabulary to create model sentences for context.

Microsoft Bing analysing a page of Swahili vocabulary to create model sentences for context.

You can also combine this with the AI Anki decks trick to really digitise those paper lists.

AI Translation Exercises

Now, how about some methods for actively working with vocab? Personally, I’m a  big fan of the translation method. Now I know this isn’t everybody’s cup of tea (it’s one thing that turns some off Duo) but if it works for you, you can produce a raft of exercises in seconds:

Here’s a page from my [LANGUAGE] course book. Analyse the entire page, noting all vocabulary items internally. Then, create a set of 20 practical, useful sentences using this vocabulary in context. Make them relevant to real-world, current affairs contexts where possible. Present half of these sentences in English and half in the target language for me to translate for practice. Add a key for any extra words you used that aren’t included in the list, as support. Add the translations of all sentences at the end as an answer key.

ChatGPT Plus analysing a page of Hebrew vocabulary to create translation exercises.

Bing analysing a page of Hebrew vocabulary to create translation exercises.

AI Exercises

You can also extend course book pages with worksheet-style practice exercises. Here’s a prompt that should produce a diverse set of activities in an output perfect for copy-pasting into a note, or PDF, to pore through on the move:

Here’s a page from my [LANGUAGE] course book. First, analyse the entire page, noting all vocabulary items, sentence frames and grammatical structures internally. Then, create a set of worksheet-style activities for me to practise using that material. Vary the activity types, including exercises like gap-fill / cloze, matching and translation. Add an answer key to all exercises at the end.

You might even like to try a more dynamic approach with this paper-to-exercise technique. The following prompt should set up a turn-based game (my favourite kind!) that recycles chapter vocab in live conversation:

Here’s a page from my [LANGUAGE] course book. First, analyse the entire page, noting all vocabulary items, sentence frames and grammatical structures internally. Then, let’s have a conversational, turn-based activity using the material. Present me, turn by turn, with a sentence in the target language using the vocabulary. I have to provide the missing word. Don’t give me any clues or model answers until I’ve made my response each turn!

Admittedly, turn-based language gaming worked better in Bing before recent updates forced it to focus solely on being a fancy search engine. If it does stray, just remind it that you’re playing a vocabulary game!

Choose Your Platform!

All these prompts have one thing in common: they play to the power of AI to take information and display it in different ways. That’s gold for learners, as the human brain learns best when presented with material in multiple, not monotonous formats. For one thing, this helps beat the context trap of repetitious learning. Recycling vocab in as many ways as possible is key to remembering it in unlimited, unpredictable future situations.

Tech-wise, you’ll see that I’ve used Bing in most of these examples. It’s an excellent place to start if you’re new to AI yourself, not leave because it’s accessible, user-friendly and completely free! Additionally, the Bing app allows you to snap a book page easily with your phone camera. And Bing’s internet connectivity out-of-the-box gives it more breadth and up-to-the-minute relevance when creating your materials.

That said, you can use these prompts with any platform that allows image uploads. ChatGPT, for instance, has the added bonus of multiple uploads – ie., pages – so you can process a larger chunk of chapter in one go.

Whichever platform you choose, the most important piece of advice remains the same: don’t just stick with these potted prompts. Instead, experiment constantly to find what works for you, building up your own prompt library to copy-paste. AI can, and should, be an incredibly personalised experience. Good luck making it your own!

Have you used AI image analysis in your learning? Let us know your own tips and tricks in the comments!

A neon lock with a glowing owl motif, reminiscent of Duolingo

Green Handcuffs – Duolingo and the Walled Garden of Welsh

What happens when you pour your heart into learning a language on service X, then service X mothballs the resource? It’s a situation many Welsh learners found themselves in this week, as Duolingo announced an indefinite pause to further development of its Welsh course

The immediate question is where do these learners go? There are other Welsh courses, of course, like the excellent materials at But the Duolingo announcement begs a further question: how do these learners take their progress with them? Progress data, the result of weeks, months, years of hard learning work, is locked into Duolingo’s proprietary system.

Now, you can already request your personal data from Duolingo. The Duolingo Data Vault is definitely a welcome addition in data transparency, allowing you to access personal data records the site holds on you. But crucially, language-based progress is missing. There’s nothing to say what you’ve studied, item for item. That means there’s no way to pick up where you left off elsewhere, with a true record of where you are.

No wonder users feel a bit stuck inside a course consigned to gather dust.

Duolingo Data Vault files, unzipped.

Duolingo Data Vault files, unzipped.

Interoperability and Language Learning

It all sounded very familiar after tech activist Cory Doctorow’s recent discussion of open internet practices in The Internet Con. This quick read (well worth a look for anyone invested in apps and services – ie., all of us) bemoans the walled gardens that Big Tech firms have become. They’re great places to be, when they work for/with us. But when they suddenly change at the whim of execs, the lack of interoperability – standards or conventions that allow you to use data from one service on another – leaves us stranded and at their mercy.

Don’t like a recent update? Tough, you’ll just have to stay, or start from scratch on another service.

It’s not for a lack of standards. Language learning platforms have long used industry-wide formats to allow interoperability. Take the plain old CSV (comma-separated value) spec. You’ve long been able export your Anki deck in this plain text format, and import it into another service like Quizlet or Educandy.

Not to be too hard on Duolingo (we love it really), there’s a clear counterargument to allowing full export of full vocabulary and phrase lists, as with Anki and Quizlet decks. The full complement of learning text is the result of lots of hard work on company time; it’s a copyrighted resource just as a course book is.

Opening the Duolingo Garden Wall

But when it’s tied to user progress, it becomes something else; a personal record of items we’ve committed to memory. Other programs export this as a matter of course. Anki, for example, will export frequency and accuracy data alongside vocabulary item entries. It shouldn’t be too hard to export this subset of Duolingo material in a universal format that could be loaded without fuss into an app like Anki.

Duolingo might well fret about losing users if the effort costs of leaving were reduced like this. No big tech corp is under obligation to organise its data in a way that helps users migrate. But you can imagine a world of interoperable ‘take your data with you’ standards to have a double-edged benefit.

First off, it could incentivise Duolingo to strive for constant betterment, to be additive rather than reductive in its updates. The race would be to the top, rather than the bottom, to maintain a winning app for all. There’d be an open door, but nobody would feel the need to defect (or the resentment that they can’t).

Likewise, there’s a general benefit even if the resources simply aren’t there for a Welsh continuation on Duo. Course migration standards would allow smaller companies to step in and fill in the gaps. Duo could focus on its core projects and nobody would feel linguistically homeless. And, of course, if Duolingo offered the missing service again in future, it would be easy to move right back.

Perhaps it’s time to make a request of our beloved owl in the name of an open web for linguists.  As the trailblazer that you are, could you be a leader in open standards, prising ajar the door to these walled gardens?

The Verb Blitz Adage : Keep It Simple

They say it’s best to keep things simple. And so it is with the Verb Blitz apps.

Verb Blitz, if you missed it, is a solid, old-school reference and drill tool to practise verb conjugations. I created the first over a decade a go as a nerdy hobby project in machine morphology, and it’s now available in 23 languages. Originally intended as a support for my own learning, it’s now helping lots of other learners grapple with endings, stem changes, and all other manner of verb fiendishness.

It was definitely high time for updates. The original apps were developed in XCode with Objective-C and storyboards, which are now very much ‘the old’. Since then, Swift and SwiftUI have become the smart new kids on the block for all things Apple. The longer you leave things, the harder it is to catch up, so a conversion project was as much about up-skilling myself as keeping the apps functional and easily updateable.

A screenshot of Verb Blitz for Scottish Gaelic.

Reining It In

The thing to guard against is that overzealous rush you get when you start a new project. It has a lot in common with the euphoric optimism polyglots get when they start a new language. After a handful of words, we’re promising ourselves that we’ll reach C1 within a year, that’s we’ll commit large swathes of each day to linguistic endeavours. Time and other commitments get in the way, and overpromising can sometimes dent our motivation a little.

For that reason, I found myself having to rein it in a little with the new, fresh Verb Blitz apps. I have a lot of exciting ideas for further developments, but to let them take over would be to jeopardise updating the existing functionality in good time. The fact is that by focusing on getting the foundations right – the existing activities – I take care of the urgent needs first, and have lots of time later to do the more fun stuff.

Isn’t that just like learning a language? It can be so tempting to skip the boring introductory units, and head straight to the meaty chapters of a new course book. I feel that urge with every new language project I start. But it’s definitely worth reining it in. Deal with the urgent needs first – basic communication – and then all the fancy bells and whistles can come later, when you’re up and running.

It’s a nice reminder of the importance of sobriety and moderation in project management. Once again, good learning strategy seems to have a lot of touching points with well-planned tech development. Not least the oft-forgotten advice when setting out: first and foremost, keep it simple!

A screenshot of Lingvist in use, demonstrating its lovely, clean interface.

I Befriended a Lingvist (and It Was About Time) [Review]

I gave Lingvist a whirl this week, a sentence-based language learning app from Estonia that had mysteriously passed way under my radar until now. The verdict: Lingvist, I’m glad I finally found you!

It’s a bit of a match made in Heaven, to be honest, given my love of mass sentence techniques. This app uses in-context, useful sentences to illustrate all of its vocabulary items, drawing on a massive library of items for each language. The sheer size of its libraries should keep even the most avid learners busy for a while, and it’s available in an impressive number of languages:, including Estonian (as you’d expect!), Japanese, Korean, Norwegian, Polish and Swedish, as well as the other ‘biggies’.

It boasts a very smart, clean app design, using an eminently readable font (always an easy thing to overlook in a language app). It has a sensible, just-forgiving-enough approach to mistakes, particularly with accents. And – most impressively – it has the most user-friendly automatic voice input mode I’ve come across in such an app. Even more impressively, it allows Japanese input in all three writing systems.

A screenshot of Lingvist in use, demonstrating its lovely, clean interface.

In use, the app has an excellent approach to exposition and testing. Items, new and old, appear as gap-fill challenges as you perform sentence repetitions. That makes for an engaging routine, even when words you already know pop up – it’s not just learning, it’s practice. As such, it’s the perfect tool if you already know some of the language, but want to start filling in the gaps.

Not a Newbie? Not a Problem!

Talking of non-beginners, Lingvist also features a great placement test mode. For a start, it’s not overlong. Isn’t it always a bit soul-sapping when a new app makes you churn through a 10-minute test off the bat? Lingvist’s snappy check pretty accurately chooses a spot to skip you to very quickly.

To check it out, I performed the test in my strongest foreign language, German. It airlifted me about 85% of the way to the end of its mammoth list. And, proving there’s always something more to learn, the sentences were actually complex and interesting enough to challenge me. That bodes well for forging ahead it with it in my more nascent languages – you can reach a very decent level of language with it.

It tracks that gap-filling with what seems like a quite sophisticated spaced repetition algorithm. Despite that sophisticiation, you have at-a-glance access to all of those stats in a clear, unjargonised format, which  makes the spaced rep process understandable even if you’re new to it. Again, no wonder I love it, given my constant proselytising of the spaced repetition original, Anki.

Generous Trial Period

The best things, of course, are usually worth paying for. And true enough, Lingvist is a premium app, although its pricing is very competitive compared to the likes of Babbel and others, at just over £5 a month on an annual plan. But what’s striking about Lingvist is how generous the team have been with the free trial (at least on iOS). You get a whopping 14 days to try the software out. That’s significantly more than the usual seven days with most – if you’re lucky enough to get a trial at all.

I’m just puzzled over why it took the algorithms so long to push Lingvist in my path, especially since the App Store says it’s been around since 2016. I hope it’s not been such a hidden, under-the-radar app for others, as it really deserves to be up there amongst the best.

If it’s your first time hearing about it too, check it out!