Using a Pareto approach with an LLM to learn Devanagari, Part 2.

Good morning, Gemini.

Let us refresh in our respective minds what we did with the approach to the Devanagari script last time, so that we might continue where we left off, in the same vein.

Good morning! We are officially back in the polyglot strategy room.

Let’s refresh our mental workspace and look at the foundation we built for our huliganov.tv masterclass.

Where We Left Off: The One-Hour Masterclass

We completely abandoned the traditional, overwhelming method of memorizing the alphabet from to . Instead, we deployed a high-frequency spy-code approach, targeting the letters most commonly used to write international names and loanwords in modern Hindi.

Here is what we unlocked in our first hour:

1. The High-Frequency Core (10 Consonants)

We mastered roughly 30% of the consonant lineup, which statistically unlocked over 80% of the sounds needed for global maps and transliterations:

  • The Liquids/Nasals: (), (), (), ()

  • The Hard Stops (Tongue Curled Back): (Hard ), (Hard )

  • The Global Heavyweights: (), (), (), ()

2. The Vowel Dashboard (The Matras)

We learned how Devanagari behaves as an abugida—where every standalone consonant has a built-in “ah” sound (schwa) that drops naturally at the very end of a word. We learned how to override that built-in vowel using graphic modifiers:

  • (Long ), ि (Short – written to the left), (Long ), / (/), / (Pure /).

  • Our Modern Upgrades: The English short vowel hacks—the Crescent Moon (ॉ) for the “O” in Boston/Coffee and the Double Feather (ै) for the “A” in Bank/Texas.

3. The Grand Finale: The 94% Literacy Proof

We put our toolkit to the test against all 50 US States. The result? A 94% functional literacy rate. Our readers could perfectly read or highly accurately guess almost the entire map (Texas = टेक्सस, Boston = बॉस्टन, California = कैलिफ़ोर्निया) using just that one hour of concepts.

Moving Forward: Writing Mechanics & Disguised Letters

Now, as we discussed, if we want to move our readers from a 67% overall capability toward that 100/100 total fluency point, we need to tackle how the letters are actually built.

 

In Devanagari, knowing the equivalent of stroke order isn’t just about good handwriting—it is a vital tool for optical recognition. Many letters look incredibly similar at a glance, and understanding which stroke is drawn first helps the brain instantly tell them apart.

The Golden Rule of Writing Devanagari

Unlike Western script where you write along a baseline, or Chinese characters drawn inside an imaginary box, Devanagari is built on a hanging line system.

  1. You write the distinctive curves and body of the letter first.

  2. You draw the vertical “anchor wall” (if the letter has one).

  3. Crucial Step: You draw the horizontal top bar (the shirorekha) last, effectively hanging the completed word from the line.

Are you ready to select the first cluster of letters from our high-frequency squad to analyze their stroke paths and compare them to the lookalikes that might confuse our readers in the next pass?

Continue reading “Using a Pareto approach with an LLM to learn Devanagari, Part 2.”

Using a Pareto approach with an LLM to learn Devanagari, Part 1

Using a Pareto approach with an LLM to learn Devanagari!

This article is basically an experiment in how linguists or polyglots or language enthusiasts (whatever we want to call ourselves, what does language matter 😉 ) can use LLMs, in this case Gemini on a PRO sub – you can try your own favorite one with the same prompts and see if it comes up with the same sort of thing) to learn something which is perhaps not altogether straightforward, but nevertheless very useful. In this case the writing system used, after Latin, Chinese characters and Arabic-based writing, by most people on the planet. The premise here is not to learn either Hindi or Sanskrit or any of the other languages that use Devanagari as languages. This is an exercise purely in getting to grips with a writing system. For this reason, we are looking at how this writing system transcribes international words, such as geographical or personal names, but not looking at any Hindi or Nepali per se, not every phrasebook phrases. Apart from anything else, the article should be as agnostic as possible with regard to which language form the set is of most interst to readers, whilst being aware that statistically Hindi will account for the greater part.

The prompts (that is, my bits) are in the H4 format, which for those unfamiliar with doing WordPress means the fourth header level format, which is what I am writing in now. What Gemini answers is in the other formats.

So hopefully you find this interesting on two levels: firstly the experiment as to whether, right now, in July 2026, I can simply dispense with an expensive book and ask Gemini to write me an instruction book on a topic like this as we go along, allowing me to dictate the order of play, the style and other matters, and the second point of interest is the content itself, that is, the actual learning of what in this article is actually only 20 of what it takes to know Devanagari inside out, but it’s been guided to follow a Pareto approach. The test at the close proves that Pareto has not entirely held true, because we have a 67:20 relation rather than a 80:20 as Wilf Pareto might have predicted based on his principle, but anyway I think it’s a good deal to give my readers 67% of a very useful skill while asking them to invest only one hour to gain it.

Obviously it will take a few iterations to get it well memorised, but each run at it is likely to be doable within an hour.

Here goes.

Continue reading “Using a Pareto approach with an LLM to learn Devanagari, Part 1”

Thule, if you think it over… (Friday AI-Day series #3)

Two distinct prehistoric Arctic groups—Dorset and Thule Inuit—cooperate around a wide ice hole on a frozen landscape. The Dorset figures wear simpler pale furs and use older tools, while the Thule group wears layered parkas and stands with sleds, dogs, and advanced harpoons. Both groups gesture toward a live beluga whale surfacing in the ice hole, which remains alert and able to dive. Multiple polar bears appear as tiny silhouettes on the far horizon, observing from a safe distance. The scene is set under low winter light with long shadows and distant ridges, evoking a rare moment of peaceful interaction.
Two distinct prehistoric Arctic groups—Dorset and Thule Inuit—cooperate around a wide ice hole on a frozen landscape. The Dorset figures wear simpler pale furs and use older tools, while the Thule group wears layered parkas and stands with sleds, dogs, and advanced harpoons. Both groups gesture toward a live beluga whale surfacing in the ice hole, which remains alert and able to dive. Multiple polar bears appear as tiny silhouettes on the far horizon, observing from a safe distance. The scene is set under low winter light with long shadows and distant ridges, evoking a rare moment of peaceful interaction.

This is the first post of this year 2026, and of the second quarter-century of the 21st Century, as I view it at any rate, although few people seem to be focusing on that, maybe they are not accountants.

I obviously intended more posting this year, but the year did kick off in a predictably busy way.

Thankfully there is always AI.  Thanks, or maybe rather “due” to which, whereas before we were all crying out for content, it now seems that the boot is on the other foot and content is crying out for us, like in the Russian reversal jokes. (“In Post-AI internet, content creates you”, etc.)

Clearly not all my exchanges with AI would necessarily interest my readers, so I do need to be selective but in this “Friday AI day” series, of which this is now the third, we at least have the chance to look together with AI (I mainly use Copilot) at some topics.

The topic for today is indeed topical as we are mainly focussing on Greenland, which dominates the news. The aim here is to try and understand better the country and its people but also a little bit a couple of aspects of its wildlife, we do meander off into that at one point, do keep scrolling if that is not your bag, we come back firmly into the linguistic topic and explore a little bit the mystery of Paleo-Eskimos such as the Dorset peoples and their possible intercations with the Thules who are the ancestors of modern Greenlanders.

The main aspect we are going to be exploring below is the area of language. We won’t be learning any Greenlandic, not today anyway, but we are going to be trying to understand what the linguistic landscape looks like and how it fits with other Northern countries.

I will be adopting the simple convention that my questions are in Italics and the AI’s answers the way it gives them, which has sparse use of Italics thankfully.

If you want to find out more, then you can always ask your own AI.  Sometimes minor variations on a question can produce different answers, or the same one, in defiance of Einstein’s maxim, rather different answers depending on the mood the AI is in on a given day, it would seem.

Please respond and let me know what you think.

Continue reading “Thule, if you think it over… (Friday AI-Day series #3)”

Jak mówić po polsku, żeby brzmieć jak z Doliny Krzemowej (ale nie być rozumianym ani tu, ani tam)

How to Speak Polish Like You’re from Silicon Valley (But Be Misunderstood Everywhere)

By David J. James |

Witajcie w świecie pseudoanglicyzmów—językowej krainie, gdzie „zrobić upload” brzmi jak operacja chirurgiczna, „być na callu” to stan egzystencjalny, a „deadlineować” to nowa forma cierpienia. W tym słowniczku pokazujemy, jak polski biznes i młodzieżowy slang tworzą hybrydy, które brzmią światowo, ale są zrozumiałe tylko dla wtajemniczonych.

📘 Słowniczek Pseudoanglicyzmów | Glossary of Pseudo-English Polishisms

🇵🇱 Wyrażenie Znaczenie w polskim kontekście 🇬🇧 Jak to brzmi po angielsku? 🧠 Komentarz
O co kaman? „O co chodzi?”, „Co tu się dzieje?” What’s going on? (but “come on” doesn’t fit) Brzmi jak angielski, ale nim nie jest
Zrobić research Poszukać informacji, przeanalizować Do research (not “make a research”) „Zrobić” wszystko to polska specjalność
Być na callu Uczestniczyć w rozmowie online Be on a call Korpo-slang w pełnej krasie
Zrobić feedback Dać informację zwrotną Give feedback Feedback jako rzeczownik, „zrobić” jako czasownik
Deadlineować Ustalać lub pilnować terminu Doesn’t exist in English Brzmi jak kara boska
Zrobić upload Wgrać plik Upload a file Technicznie brzmi, ale to kalka
Zrobić meeting Zorganizować spotkanie Set up a meeting / Have a meeting Brzmi jak „zrobić obiad”
Zrobić target Osiągnąć cel sprzedażowy Hit the target / Reach the goal Agresywna metafora łowiecka
Zrobić follow-up Wrócić do tematu, przypomnieć Follow up (verb, not noun) Często używane jako rzeczownik, co jest błędem
Zrobić deal Dogadać się, zawrzeć umowę Make a deal / Close a deal Brzmi jak Wall Street, ale to kalka

Continue reading “Jak mówić po polsku, żeby brzmieć jak z Doliny Krzemowej (ale nie być rozumianym ani tu, ani tam)”

What do the British call a television?

What do the British call a television?

The British native languages – especially Celtic.

Among the British languages we have Teledu in Welsh and in Breton the endearing term Skinwel, although despite Britannic it isn’t spoken in the UK. I couldn’t find the Cornish term, but another commentator has it. Nor Manx, although presumably they have them. Irish Gaelic is Teilifís, Scots Gaelic is telebhisean, and the Lowland Scots article in Wikipedia says “Televeesion” although I am not such if that’s official usage, and as far as the Old English word they use, I think we can be pretty sure that’s an anachronism.  As Abe Lincoln famously said, “don’t believe everything you read on the internet”. In that vein the Pictish term for TV is VOD as this enables you to “pict” what you like, when you like.

An early form of television from Chaucer’s time, needing a huge machine underneath to get it working and even then the screen was only in blue.  Legend has it that in the UK, in order to get that effect in “The Wizard of Oz”, they had to carry out your old black and white minstrel television and bring in a Sony Trinitron actually during the film.

British dialects of English

If you are thinking about British versions of English and the regional or slang terms, I can’t think of any regionalisms. “Telly” is an informal way of talking about television as a service or the actual set, and further slang words for the set exist such as “the box” or “the gogglebox”. The term “tube”, hwever, was not widely used in British English slang and is more of an Americanism which I am not sure many of us would have understood prior to YouTube popularising it.

Community languages

Immigrant communities in the UK exist and the Poles have telewizja, while the words in Indian languages are mainly recognisable as something sounding like the original Standard English word (ie Greek roots put together in such a way that Oedipus could have forgiven his father) but written in their own alphabets and there are quite a few of them.

You can probably get by in the British Isles just using “television”, this weird partly Greek partly Latin word which was put together in the UK by an inventor whose command of physics was clearly many metres per second better than his command of philology.

Continue reading “What do the British call a television?”