Terminology Is Invisible Until It Breaks Something

Somewhere in your localization workflow, the same clinical term is probably written three different ways. One version appears in the patient intake form, another in the discharge instructions, and a third in the translated portal your health system launched last year. Nobody flagged it. Nobody caught it. And the patient reading it in Chuukese — one of the most medically underserved language communities in the United States — had no way to know which version was correct.

That is a terminology problem. And according to Rita Prazeres Gonçalves, terminology expert and founder of The Language Worker, it is the most critical yet invisible discipline in the entire localization industry.

She is right. And in languages like Chuukese and Pohnpeian, the stakes are not abstract.

Why Nobody Knows Terminology Exists

Rita makes a sharp distinction: terminology is not undervalued. It is invisible. People do not dismiss it — they simply do not know it exists as a standalone profession with its own methodology, theory, and career path.

In places like Quebec and Catalonia, language policy created a market for terminologists as recognized, hireable specialists. Everywhere else, terminology duties get folded into someone else's job without the training that work actually demands. The result is accidental terminology management: a translator builds a glossary in Excel, someone else updates it inconsistently, and no single person owns it.

Most localization buyers — health systems, school districts, government agencies — have never heard the word "terminologist." They have heard of translation. They have not heard of the discipline that makes translation consistent.

What an AI Glossary Cannot Do

Here is the specific trap that is catching teams right now: asking an LLM to generate a terminology list and treating the output as a source of truth.

It feels efficient. It is not.

The core problem is domain contamination. A term that means one thing in cardiology means something different in psychiatry. A term that appears in English-language medical literature may have no equivalent in a given language, or it may have a folk equivalent that patients actually use, or it may have been borrowed and slightly corrupted across dialects. An AI model trained on general web text will not know which case applies. It will pick one and move forward.

Rita's recommended sequence flips the process: do the human analysis first, build a grounded understanding of the content domain, and then bring AI in with enough specificity that it can actually help. AI as a drafting assistant under human supervision is useful. AI as a terminology authority is a liability.

For standard European languages with decades of professional glossaries and parallel corpora, a human reviewer can catch AI errors quickly. For Chuukese or Pohnpeian, there is often no published glossary to check against. The error goes undetected.

The Rare-Language Terminology Problem

This is where generalist agencies hit a wall that most of them will not admit exists.

Chuukese and Pohnpeian are spoken by tens of thousands of people living across Micronesia, Hawaii, Guam, and the US mainland — particularly in states like Arkansas, Oregon, and Washington where Compact of Free Association migrants have settled. These communities have among the highest rates of tuberculosis, diabetes, and hepatitis B of any group in the United States. Health literacy materials in their languages are not a nice-to-have. They are a legal and clinical requirement under Title VI and the Affordable Care Act.

The problem is that standardized medical glossaries for these languages do not exist. There is no published Chuukese oncology glossary. There is no peer-reviewed Pohnpeian pharmaceutical terminology database. When a health system needs a patient consent form translated, someone has to make real-time decisions about which term to use — and there is no authoritative reference to verify the choice.

At TXLOC, building usable terminology for these languages means working directly with native-speaking linguists who have domain knowledge, cross-referencing community usage, and documenting decisions in a glossary that grows with each project. That glossary is not generated. It is constructed, entry by entry, with human judgment at every step. When a Chuukese-speaking patient from Weno encounters the word for "informed consent" in a hospital form, that word should be the same word they saw in the prenatal brochure and the same word the interpreter used at check-in. Consistency at that level requires intentional terminology management, not autocomplete.

The table above is not an indictment of anyone. It is a description of a gap that exists, that affects real patients, and that requires a different workflow than standard language pairs.

What a Minimum Viable Terminology Process Looks Like

Rita's recommended starting point is practical and accessible: pick one client, pull a 10,000 to 20,000 word sample of their actual content, and have someone with genuine domain knowledge do a manual pass to identify roughly 200 term candidates. Start with a simple structured file. Build ownership into the process — one person responsible for validating entries, flagging inconsistencies, and keeping the glossary current.

For rare-language work, that same framework applies with one additional requirement: the person doing the human analysis needs direct competence in the language and the subject domain. A general bilingual reviewer cannot substitute for a Chuukese-speaking health educator when the content involves treatment adherence for latent TB infection. The terminology decisions made in that context will appear in materials that patients act on.

Grow the glossary project by project. Every new health system engagement, every school district assessment translation, every government benefits document adds entries. Over time, the glossary becomes an asset — a documented record of how your community uses language in a specific domain.

The Takeaway

Terminology management is not a luxury feature for enterprise localization programs. It is the mechanism that makes translation consistent, trustworthy, and safe — especially when the language has no published reference glossary to fall back on.

If your organization translates clinical or legal materials into Chuukese or Pohnpeian, the question worth asking is not whether you have a translation vendor. It is whether you have a terminology process and who owns it.

If you want to understand what that process looks like in practice for Pacific Island languages, reach out to the TXLOC team.