Does “Low-Resource Language” Mean My People Are Low?

Yesterday, during our Indigen World contributors’ discussion, I spoke about collecting the words, expressions, recordings and stories we need to build better tools for Kasem. Somewhere inside that work sits a phrase I keep returning to: low-resource language.

I understand what it means in technology. Still, put it beside the language of your own people and it starts to sound uncomfortable.

Low according to whom? Low in what? Does that mean my people are low?

No. But the answer deserves more than a reassuring sentence, because the gap that label describes is real. So is the danger of mistaking that gap for a verdict on the people living behind it.

What exactly is supposed to be low?

In language technology, resources include things a computer can use: digital text, recorded speech, translations, dictionaries and examples checked by people who know the language. A language can have relatively few of those resources available for a particular task.

That is the basic issue. There may be too little suitable material to build or evaluate a reliable translation tool, speech recogniser or language assistant.

A 2020 study of linguistic inclusion in natural language processing classified languages according to available digital resources and documented substantial disparities. It also showed why speaker numbers alone do not tell the whole story: languages with similarly sized speaking populations can receive very different research attention.

None of that measures how intelligent the speakers are. It does not count their humour, the sophistication of their relationships or everything they know about the places they inhabit.

A database can be small while the world it describes is enormous.

The computer’s ignorance is easy to misread

Imagine walking into a library, finding one thin shelf about your community and deciding that your community has barely lived.

You would have mistaken the collection for the civilisation.

I think we risk doing something similar with AI. When a system handles English confidently but struggles with Kasem, it is tempting to treat English as naturally suited to the future and Kasem as something that cannot keep up.

But the system’s competence depends on the material, design and testing behind it. Its limitations need scrutiny before we turn them into judgements about a language.

A story told aloud still exists before somebody transcribes it. A song carries meaning before it becomes an audio file. Knowledge does not begin at the moment a developer uploads it.

Consider the Royal Court of Tiébélé in Burkina Faso. UNESCO describes its architecture and symbolic decorations as expressions of Kasena social organisation and cultural knowledge, with women preserving and passing on the decorating tradition.

That is one visible example of knowledge travelling through practice. Plenty of what matters in a community will never arrive neatly formatted as training data.

Still, pride cannot fix a missing tool

This is where I have to challenge my own indignation. Being offended by the wording will not produce a working keyboard, an accurate dictionary or a useful learning experience.

The resource gap has practical consequences. If we want a child growing up outside the community to learn Kasem, or a speaker to use it comfortably on a phone, someone has to do the work that makes those experiences possible.

Through Indigen World and Kawuri, we are trying to contribute to that work. As I wrote in “Indigen World Is Building More Than a Culture App,” starting with Kasem means taking context, dialect differences and community knowledge seriously.

Yesterday’s discussion covered contributions ranging from everyday expressions to oral narrations, pronunciation and cultural videos. Those categories matter because a language is much more than a list of English words with local equivalents beside them.

We also have to be honest about what progress means. Filling our contribution jars would be a project milestone. It would not automatically establish that Kasem has become well resourced for every technology, or prove that an AI system understands it reliably. We still have to check the material and test the tools.

The people doing the work deserve a say

One detail from yesterday’s meeting deserves as much attention as any ambitious target: a contributor explained that batches of fifty assignments were overwhelming and asked for smaller batches.

That is useful feedback. A person can care deeply about a language and still have work, responsibilities and a limited amount of time.

If I want people to help build this, I have to make participation manageable. Ten careful submissions can be more useful than fifty rushed ones. An attractive progress page cannot compensate for an exhausting contribution process.

The wider research supports taking participation seriously. A study of participatory machine translation for African languages argues that being under-resourced involves problems beyond data availability and brings people across the development process into the work.

For me, that raises a responsibility: contributors should understand what their submissions will support, receive recognition and be able to question how their knowledge is represented.

And being proud of our culture does not mean every story or recording must become publicly available. People should be able to set boundaries. Building a resource is also a question of trust.

A future where Kasem belongs in ordinary life

I want Kasem to have room for more than solemn speeches about preservation. I want room for songs, jokes, useful lessons, arguments and the ordinary messages people send when they are living their lives.

A language should be allowed to have fun. It should be allowed to make new things.

That is the future I want our work to help make possible. Better digital resources can support it, alongside people speaking, teaching and creating in the language.

So, does “low-resource language” mean my people are low?

No. It means there is work to do on the resources available to technology. I can acknowledge that gap without accepting shame on anyone’s behalf.

We are building these resources because our language is worth using. Its worth was never waiting for a computer to confirm it.