Bridging the Gap: oorNews Writing Formal Scots to Train AI

In a recent study titled “The Sociolinguistic Foundations of Language Modeling,” researchers Jack Grieve and his colleagues delve into the crucial role of linguistic diversity in artificial intelligence. Published in Frontiers in Artificial Intelligence, the article underscores how language models reflect the varieties of language they are trained on. The authors argue that the societal value and performance of these models hang on the quality and diversity of their training data.

One of the pressing issues highlighted in the study is the representation of the Scots language within AI systems. Scots, often incorrectly viewed as a dialect of English, requires careful handling to prevent AI outputs from sounding informal or out of place. Such misrepresentation can be both discriminatory and offensive to Scots speakers, reinforcing harmful stereotypes of low status and low education.

In response to these challenges, oorNews is making significant strides by providing a large corpus of formal register Scots. This initiative aims to ensure that AI outputs in Scots are not only accurate but also respectful of the language’s status and use in formal prose. By curating content that reflects formal Scots, oorNews helps prevent AI from locking into damaging stereotypes associated with informal language use.

The integration of formal Scots in AI training data is essential for creating language models that honor and accurately represent Scotland’s linguistic diversity. This approach aligns with the recommendations from Grieve et al., advocating for training corpora that genuinely reflect the specific varieties of language being modeled.  All Scots speakers are welcome to contribute to oorNews by suggesting any changes they want to see in an article via the comments section beneath.