Chatting with Ali Mehdi: AI and Indigenous Language Revitalization

How can AI be a tool for cultural preservation and education? We speak with Ali Mendi, Senior Director at Heritage Lab, a nonprofit based in Nunavik, Quebec, that is creating AI from an Indigenous-led perspective.

Hear how Heritage Lab is leading the way by prioritizing data sovereignty and putting digital control back into the hands of communities. This episode highlights the work they have done, how they’re engaging and training local youth to build and maintain practical educational tools, and their future plans to create new Indigenous grammar tools and educational platforms designed to empower Indigenous communities across the world with their own data.

If you want to learn more about Heritage Lab and their work, visit heritagelab.ca/ or follow AI Inuktitut on Facebook. Look out for their full resource launch this November -December.

Transcript

Welcome to Hello Kwey

Stacy 0:07

Welcome to Hello Quay . Let's Talk Education , a podcast where Indigenous and non-Indigenous voices from across Quebec come together to share knowledge and experiences related to education . In each episode , you'll encounter new perspectives that we hope will enrich your practice and deepen your connection to your communities . In this episode , dafta and I will be meeting with Ali Mendi , the Senior Director from Heritage Lab , to chat about AI systems that have been developed in Nunavik to support Indigenous language revitalization . So hi everyone , and welcome to our second episode of Hello Kwe . Let's Talk Education . Today , daphna and I have the lovely privilege of being joined by Ali . So , ali , could you share a bit about your personal and professional background and the work you're doing at Heritage Lab ?

Ali 1:08

Hi , my name is Ali . I'm the Senior Director at Heritage Lab . I was an educational consultant for the Cadavid School Board for a few years before transitioning into this role at Heritage Lab . Heritage Lab being a non-profit organization that builds Indigenous language tools for communities across Canada .

Daphna 1:31

Ali , we met a few years ago when you were working as a consultant for the Katowice School Board . How has that experience informed the work that you do today ?

AI Translation Errors and Cultural Inaccuracies

Ali 1:42

Yeah , it's been a long journey .

Ali 1:44

So back when I was at the Katowice School Board , it was the big birth of not the birth of AI , but the mainstream adoption of AI . The buzz was all around AI and how we can use it as fast as possible . Our school board up north is that teachers started using it without any guidance and any input , which meant that they were asking cultural questions to ChatGPT and Facebook AI and most of the time receiving very culturally insensitive information about the situation in Nunavik and the danger went beyond that , and that the students were also using it to try and learn about their culture and whatnot and converse with these tools , and that created a complete mess in terms of , well , how do we approach this ? Where do you start ? Because it was somewhat of an arms race between Facebook , microsoft and Google to put out these AI tools as fast as they can . Google's Gemini , kfgpt and I met as AI and each of these tools was performing worse than the next in our school system , and so what do you do with the situation ? That was the birthplace of the nonprofit .

Daphna 3:03

And so how did the resources of Heritage Lab address some of these challenges ?

Birth of Heritage Lab Solutions

Ali 3:09

So , at the time , microsoft and Google released the translator into the communities and with the advertisement behind it of saving ineptitude , and so we were all excited , as non-ineptitude speakers , to start using this tool that could translate documents , we quickly realized that it was complete gibberish . It was translating children into pizza . It was a complete mess . Translators that were using it said that it was under 10% accuracy . The issue is that they were advertised as tools that you can reliably trust to translate indigenous languages , not only in the Pitu , and so me , with a background in AI , and a few community members decided to try and build something that was a translator that was functional , and so these companies have been pouring millions into this , but they didn't build it with communities . They built it in San Francisco , away from communities , and served it to the communities and us in a matter of two weeks , built a translator that was over 70% accurate with three people and no budget , so we realized that it was possible .

Daphna 4:29

Amazing . I know that your hope is to have teachers across Quebec access and make use of Heritage Lab . What are some of the challenges that you face in making that a reality and how can some of us help with that endeavor ?

Ali 4:41

It goes back to how . So this is where we were with . We built a translator that was functional . We didn't know what to do with it , went to my supervisor , went to their supervisor , went to their supervisor , went up the chain of command and realized that we wanted this to be an open , non-for-profit

Indigenous Data Sovereignty

Ali 4:59

initiative . We didn't want it to be for profit . We didn't want to be like Facebook and take indigenous data and then sell it back to them at a profit . So we decided we wanted it to be Indigenous-found , indigenous-led , and so we built a board of directors predominantly Indigenous , an Indigenous majority in terms of all decisions that are taken , and we founded this nonprofit all decisions that are taken . And we founded this nonprofit .

Ali 5:28

And from there , as we were collecting information to build the translator , we realized that we were touching on a lot of archives of indigenous data and in our context was Inuit data and that's historical stories , a lot of cultural material that weren't digitalized , were never digitalized , so these were sitting on the shelves . They were archives , but we needed them to build the system we were building . So over two , three months , we digitalized over 6,000 pages by a scanner that we built that can scan pages and pull ineptitude syllabics from the page , so we had to build the whole pipeline from the ground up full ineptitude , syllabics from the page . So we had to build the whole pipeline from the ground up , and so from there we ended up with a big data set of historical information , fascinating stories , stories of people , stories of places , stories of everything , and from there we had a few students develop and build an avatar that you can interact with and learn about community history from an Inuit point of view , and that's something that hadn't been done before .

Ali 6:33

And so the big challenge with this is that we're faced with giants , multi-billion dollar tech corporations that have these tools that everyone are using , and I think a big part of what's important is to have some sensitivity around what information is being shared , and I think of teachers in the South that want to teach their students or explore Indigenous knowledge or communities . The easy answer is always to ask at GPT , and you'll get an answer that always sounds very , very right , but you don't know if it's actually right because you're not indigenous and you don't have a community member at your fingertips to ask .

Stacy 7:20

I love that you brought that up , ali , because last year we did a few workshops on AI and one of the things I was talking about is where is AI getting its information from ? It's using the web and these huge data sources , but a lot of that data was produced by non-Indigenous peoples , so your accuracy always needs to be questioned . So I think we really get at digital citizenship and evaluating sources and triangulation and all those skills that we need our students to know and that we need to also remind ourselves also to do those checks as well , sometimes as educators .

Ali 7:56

Yeah , no , it's like that's it , because we had to deal with Wikipedia back in the day and don't cite Wikipedia . But this , this is on a whole next level , because not only will it , not only is it , is it , is it information ? But it's also continuously generated information that that produces its own fake sources . So it's you can't even trust the sources . So where do you go with this ? There's such a huge weight that is now on teachers' shoulders to vet everything . It's a lot of work . And where do you find those sources ? Well , you have to look at the school board's website . You have to try and find actual resources from the community . But that's a lot more work than asking a question and you can't really keep up with it .

Daphna 8:51

Okay , ali . So now that you've painted such a dismal portrait of this technological landscape , where's the hope ?

Ali 9:04

Well , the flip side of this is that AI has made it easier to build tools like the tools that we're building . So when you want to build tools for good , it's so much more budget efficient . Our costs have scaled down tenfold from when we started the project , so it's a lot easier to build tools like what we're building . Our five-year plan of wanting to build grammar tools , language learning tools , language voice technology we've already achieved our milestones for a year and a half and we're still in our first eight months . So technology is progressing a lot faster , but that means you can also do good a lot faster . So it's big inspiration for initiatives like this . Grassroot initiatives are as easy as they can get to start . It just takes human motivation to put that together .

Daphna 10:06

Amazing

Heritage Lab's Current Tools

Daphna 10:07

Ali . Your work has received international attention .

Ali 10:14

Tell us why . Yeah , I think there's not that many non-profit technological innovations that come out . When you build technology , the first big balloon is how much money can I make off of this and how can I market this to return a profit ? That's not a thing in Indigenous . It's not as much of a driver . In Indigenous communities there's an atmosphere of sharing and collaboration and whatnot that you don't get in Silicon Valley .

Ali 10:47

So that was a huge plus when we're presenting so we built the technology within two weeks we're invited to go present at UNESCO in Paris in front of 1,500 people . So it really gained traction very , very quickly . You can come and build on the infrastructure that we have free of charge . There's no profit margin . There's no we're selling your data to advertisers . We want more language representation and we want you have a community . You can have your own avatar , your own language on the platform , your own translation , dictionary , whatnot , and all the tools that we're building and have that all in one place .

Ali 11:33

And then when we say , have that all in one place , well , it opens the question of data sovereignty , and that's the other huge can of worms that we discovered after going to UNESCO and then the project getting mentioned at the United Nations General Assembly as a need for having an additional sustainable development goal , being cultural and language preservation , which is not NFCGs , but that's another tangent . But yeah , while we were there , we received a lot of requests from other Indigenous communities , be it Cree , be it communities from all across Canada and communities from South America , asking if they can have these tools in their community . And the thing is is that when you share your data with Google and Microsoft and whatnot , you don't control your data anymore . It's put on servers in California , in Ohio , and you never have access to that data anymore , and so that goes into the next stages of colonization and that's taking away digital data , and that's a huge issue that we're currently facing .

Ali 12:47

And going back to data sovereignty , it's well , where is all this information stored ? And that was the big challenge when we were building the platform is , we don't want this to be stored on servers that are not indigenous-controlled , and so our first idea was okay , let's set this up in Kutuak at the school board and try and build a server there , which is where it currently is hosted , so everything is hosted locally . That adds a lot of costs to our operations , but it's a red line for the organization of making sure that digital data is held to the same standards as physical data . So you don't take a book and not return it . The same way you would not move digital data and not return it . So for now it's sort stored in Kuduak . As we look to expand into different languages , we're looking at a partnership with Kanawagi . They have one of the biggest data centers in all of Canada , the 30-minute drive from Montreal , and so looking at that as a potential central hub to store the data that we work with .

Daphna 13:51

Amazing Ali . Ever since I met you , you're always coming up with innovating and exciting new projects . What motivates you ?

Ali 14:03

That's a big one . I feel like the world is a very sad and scary and awful place at times . I feel like part of the motivation and we're all in education and I feel like it's one of the most satisfying and nourishing fields and that you feel that you can have a big positive impact , and you see it when you're a teacher , you see it when you're in the field of education in general . It's a very humbling and noble field and it's how can we make as much impact as we can in the little time that we have and when you can reach everyone through these tools that are being built , it keeps you up at night .

Daphna 14:50

Amazing . So if teachers want to learn more about , or educators at large want to learn more about , heritage Lab and the work that you do with your team and the resources that they can make use of , how do they access ?

Ali 15:06

So for now it's heritagelabca and our Facebook page , our Inuktitut initiative called AI Inuktitut , so you can follow our updates there . We're looking at a launch around the month of November December , where we're going to have resources on how to build tools for your community guides , art challenges , whatnot and everything is going to be public around there .

Daphna 15:36

Oh my gosh , you're so inspirational , ali Stacey , I don't know how you want to follow up .

Stacy 15:45

So maybe , to end , you mentioned like having a five year plan , maybe about where the business is going . Could you give us an idea about maybe what's happening next , like with the language tools and the systems that you're working with ?

Ali 15:59

so right now , what's what's completed and what's concrete is a translation system between english and english that incorporates all the dialects of new nevada . Keep in mind that Microsoft and Google do not include any dialects in their tools and they mishmash entire 20 , 30 dialect languages into one tool . So we have a dialectical system where you can translate specifically into dialects . We have our character , ayavutak , that you can ask questions to and pulls from local history . We have our dictionary that's the official dictionary of the region . That's built into the tool . But we also have a contribution platform where you as a community member can share a story or share a word or term

Youth Involvement and Future Vision

Ali 16:49

, and that will get vetted by our language committee . That will then get added to the platform .

Ali 16:56

So that's the other big component that was a huge request of well , how do we add to this tool and how do we crowdsource the knowledge base ? What's coming soon is , of course , the grammar tools , so having a version of , let's say , antidote , but for Indigenous languages . So building that with community members and breaking it down into dialects , the different grammatical rules depending on the region , and then , of course , having a voice technology platform , which the big interest with that is around the medical field , so communicating with your healthcare provider in your language without having to go through English . For example , tuberculosis is a big issue in Nunavik . Being able to communicate your symptoms in Inuktitut , where a word for that doesn't necessarily exist in English , so that's the big need there . But then also practicing your Inuktitut and learning it on an educational platform , which are the tools that we're looking at building in the next two years .

Stacy 17:56

Amazing . Thank you , I love it . It's a wonderful tool for language revitalization and the idea that community members can also contribute and have their contributions vetted , because thinking about language evolves and it changes over time , so just having that be part of the program shows that it's going to be continually growing and evolving , along with the AI tools themselves .

Ali 18:28

Yeah , and keeping in mind youth involvement . So this entire initiative is built mostly by youth . So we've hired youth from over six communities . We've created a full-time position in language technology for youth that recently graduated , has put in hundreds of hours into curating data from his community , is learning programming , is now building the AI alongside the development team and going back to college in January to go into programming . So it's also an inspiration for the youth and the community of hey , there are technical technology-based opportunities in the community . So yeah , so yeah . It's very exciting .

Stacy 19:16

It's wonderful to have youth be able to see themselves in technology in their own communities , staying in Quebec and playing a role in language revitalization and cultural revitalization . So I don't know , I'm just thinking like how life has progressed . Like Nunavik had Internet in like 2004 , I think like it was really recent and now we're developing AI systems in Nunavik and it just shows what amazingly talented people are there and the dedication and inspiration that's coming from the youth of that area . Thank you , ali , for speaking with Daphna and I today and we look forward I think I speak for both of us to see what's next and staying up to date with what's coming .

Ali 20:11

Yeah , thank you . It was great talking to you too .