Restricted archive / case files
The project is organised around three case studies. Each asks the same question in a different system: when language is forced through an institution or technology, what meaning survives, what is standardised, and who pays for the loss?
Asylum interpretation is the anchor case. A person gives testimony through an interpreter, the account is turned into an official record, and that record becomes the basis for credibility findings. Procurement, accreditation, domain knowledge, and transcript practice all shape whether the applicant appears coherent, inconsistent, credible, or impossible.
Singlish particles such as lah, leh, and meh carry stance, relation, emphasis, and social meaning. Correction systems often treat them as errors because they are not legible to standard English. The result is not neutral correction. It is the quiet removal of meaning.
AI systems inherit the uneven distribution of language data. Languages with less online presence become harder to model, less accurately processed, more expensive to use, and easier for automated systems to ignore. The hierarchy is not flattened by automation. It is encoded into it.