Ask HN: What are you working on (August 2024)?

I've been building a web extension for an Inuit language (ᐃᓄᒃᑎᑐᑦ ᖃᓂᐅᔮᖅᐸᐃᑦ).

Most Inuit in Canada speak Inuktitut, which is a language with long words and two different writing systems: the latin alphabet, and syllabics. Syllabics are specially adapted for Inuktitut and well-loved by the Inuit, but unfortunately can be a pain to input on a computer, so often times the more cumbersome latin alphabet is used in casual writing.

Tom Scott has an awesome video about how syllabics work[1], but breifly, the shape of the character determined its inital consonant spund, and the rotation determines its vowel sound. So ᐱ = pi, ᐳ = pu, ᐸ = pa, ᑎ = ti, ᑐ = tu, ᑕ = ta, etc. The word "Inuktitut" becomes ᐃᓄᒃᑎᑐᑦ in syllabics.

Transliterating between the two is fairly simple, but there are edge-cases around dialects and whatnot. The more interesting problem from a technical perspective is having a web extension that can detect Inuktitut on a web page (in wither writing system), and transliterate that into whatever writing system the user desires, whilst never accidentally transliterating the other text on the page ("inhabitants" could be picked up and transliterated as "ᐃᓐᕼᐊᖯᐃᑕᓐᑦᔅ", for example, even though that makes no sense).

The project is mostly using Rust via WebAssembly, which has been a lot of fun to work with and has let me do some awesome things, like avoiding heap allocation and using compile-time hashmaps to do conversions on the text. The build system has to do a lot and I eventually settled on python. Right now I'm trying to wrangle JS and the DOM (there's a lot of edge cases to deal with), and that's been difficult as it's not my wheelhouse.

1. https://www.youtube.com/watch?v=xW4hI_METac&t=15s

It looks like a great language to work with.

Inuktitut, Rust, Python, JS, or WASM? :D

Inuktitut is very cool. I'm not a speaker so I'm relying heavily on research, but I've always loved making tools for syllabics and working with text generally.

Then congrats all the more on this work, I had assumed a motivation from attachment also. As for the tech stacks and langs yeah they're fun, but I meant in how abstract Inuktitut looks, I guess the human language in the loop is the real meat of the problem and of the "working with".

If you're curious, I don't have any direct connection with the Inuit but I got into syllabics because I have some Ojibwe speakers in my family, and Ojibwe also uses syllabics. None of the current speakers I know can write in Ojibwe (in any writing system) due to their residential schooling, but we have letters from the generations that have passed on.

I made a few syllabics converters for Ojibwe when I was younger, but Ojibwe and Cree are harder to build converters from than Inuktitut because they have a lenis-fortis distinction in the latin script but not syllabics, so the word "anishinaabe" can be converted unambiguously to ᐊᓂᔑᓈᐯᒃ, but trying to convert the other way could render any one of anishinaabe, ani*zh*inaabe, anishinaa*pe*, or ani*zh*inaa*pe*. I don't think it could be done accurately without some sort of machine learning. There's also a lot more complexities surrounding how different dialects are encoded. In the end, Inuktitut was simpler to work with and has a lot more everyday speakers and text to work with, especially because the Canadian territory of Nunavut publishes its government documents and websites in Inuktitut too.

There's a lot to cover but I think the technical challenges might be interesting to the Hacker News crowd, I may make a series of videos about it someday!

Yes it is right up my hacking alley. I'm working too with two-way asymmetric conversions between languages/charsets as noted in this thread. It's interesting to have such a niche tech challenge associated with your ethnic group of languages. In Manitoba are any of those spoken?

Ojibwe and Cree are both spoken in Manitoba, but it's too far south for Inuktitut.

I read through your other comment on this post and read about the work you're doing. I didn't quite understand –I think it's hard to explain a lot of these linguistics tools briefly without uploading a lot of context– but it seemed up my alley as well.

> It's interesting to have such a niche tech challenge associated with your ethnic group of languages.

That's true, I think it can be a lot of fun and rewarding for technologists. Kevin King is a font designer who was tasked with making a new font for syllabics, then found that some syllabics had never been adopted into Unicode and set about getting them approved[1]. He worked with the communities and the Unicode committee and got them into Unicode, and that must have been really rewarding.

There's difficulty in approaching from the outside, though. I'm trying to be careful to provide value and not overstep. The Inuit especially can be defensive, because you see a lot of people online who sort of fetishize the culture, or people (certainly linguists) whose work can seem extractive to the tight-knit Northern communities.

I've always wanted to build tools working with text that people use on a day-to-day, but I also worry that this will be just another project I spend months on that will not see any use. I'm trying to make connections so that I can build value without seeming extractive. It's hard to know where to start, however.

1. https://www.unicode.org/L2/L2020/20255-ucas-adds.pdf