Reality Check: Is There An Actual Market To Sell “offline-sourced” Datasets For Zero-resource / Low-ROI Languages (like Shawiya/Tacawit)?

Hi everyone,

I’m a native speaker of Shawiya (Tacawit – ISO shy), an Amazigh language spoken in Algeria. Online, this language is practically non-existent—there is almost zero digital text or parallel corpora available to scrape.

Because the data simply doesn’t exist on the web, I have direct access to offline sources: physical books, un-digitized manuscripts, historical folktales, and real-world native speech that have never touched the internet. I’m considering digitizing, transcribing, and translating these into a clean parallel dataset (Shawiya <-> English/French/Arabic), as well as creating original native text from scratch.

We always hear that AI companies have “hit the data wall” and are desperate for non-web, un-scrapable human data. However, realistically, Shawiya is a niche language with very low commercial ROI. Before I spend hundreds of hours digitizing manuscripts and building this corpus from scratch, I want the harsh market reality from people in NLP and data acquisition:

1. Does “un-scrapable offline data” actually sell if the language has low commercial ROI? Do AI labs or data brokers (Scale, Appen, Defined.ai, etc.) actually pay independent creators for rare offline-sourced datasets like this, or is the “data wall” hype only applicable to major commercial languages?
2. If there are buyers, who are they? Are there specific brokers, labs, or funded research grants that actively buy raw, digitized offline corpora and historical/cultural text for underrepresented languages?
3. Copyright & Sourcing: When digitizing local historical tales, oral folklore, and physical prints, what kind of provenance or licensing proof do buyers demand before they even look at the data?

Please don’t sugarcoat it—I want to know if there’s real money in this niche before I start the heavy lifting. Thanks!