MTPE Works Great — Until the MT Doesn't Exist

The MTPE Pitch Assumes MT Exists

About 50% of companies now post-edit their machine translations, according to Nimdzi's industry research. The workflow is straightforward: run your content through an MT engine, hand the output to a human editor, ship a finished translation faster and cheaper than a fully human workflow would allow.

For Spanish, French, Japanese, or Arabic, that math works. Productivity for full post-editing runs roughly 600-800 words per hour versus 200-300 for pure human translation. Light post-editing pushes even higher — around 1,000 words per hour. The Crowdin MTPE guide lays this out clearly, and the efficiency gains are real.

But the entire model collapses the moment MT output doesn't exist. And for millions of people served by US healthcare systems, school districts, and government agencies, that moment arrives regularly.

Two Languages That Break the Standard Playbook

Chuukese and Pohnpeian are the primary languages of the Federated States of Micronesia. Together they represent the dominant languages spoken by Micronesian communities across Hawaii, Guam, the CNMI, and growing mainland US populations in states like Arkansas and Oregon.

Neither language has a commercially viable, production-ready MT engine. No major MT provider — not DeepL, not Google Translate at any useful quality level, not any enterprise CAT-integrated engine — offers Chuukese or Pohnpeian as a supported language pair with output reliable enough to post-edit.

That's not a gap waiting to be filled next quarter. It reflects the structural reality of low-resource languages: no large parallel corpora, limited standardized orthography across dialects, and no commercial incentive for the major players to invest in training. The situation isn't changing soon.

So when a hospital in Hawaii needs a discharge summary in Chuukese, or a school district in Guam needs an IEP notice in Pohnpeian, the MTPE cost model is irrelevant. There is no MT output to edit.

What Title VI Actually Requires

Title VI of the Civil Rights Act requires recipients of federal funding to provide meaningful access to people with limited English proficiency. For healthcare systems and school districts, that obligation extends to Chuukese and Pohnpeian speakers — two of the most frequently cited languages in LEP assessments across Pacific-facing jurisdictions.

The law doesn't care whether an MT engine exists. It cares whether the patient understood their diagnosis, whether the parent understood their child's education plan, whether the community member could access government services without a family member doing ad hoc interpretation in a hallway.

For these language pairs, compliance means 100% human translation, every time. There is no shortcut, and building your language access plan around MTPE assumptions for Chuukese or Pohnpeian is a compliance risk, not a cost strategy.

Where MTPE Logic Still Applies — and Where It Doesn't

This isn't an argument against MTPE. For the language pairs where it works, the productivity gains are legitimate and the quality ceiling for full post-editing is high. Here's an honest comparison of how the two workflows actually differ when you're operating across both common and rare language pairs:

For organizations managing multilingual content across both common and rare language pairs, the mistake is applying a single workflow assumption across the board. A content operations team that runs Spanish through MTPE and assumes the same process applies to Chuukese will eventually produce a compliance failure — not because anyone made a bad decision about MT in principle, but because they didn't account for where MT infrastructure simply doesn't reach.

The Specific Problem With Healthcare Content

MTPE guides correctly identify medical and pharmaceutical content as requiring full post-editing — the highest-effort, highest-accuracy tier — even for language pairs where MT exists. The reasoning is obvious: a mistranslation in a medication instruction or a consent form isn't a brand problem, it's a patient safety problem.

For Chuukese and Pohnpeian in healthcare contexts, that standard applies with no MT shortcut available. A qualified Chuukese translator working on a discharge summary, a medication guide, or an interpreter services intake form is doing work that requires both linguistic competence and subject-matter familiarity. Finding that person is harder than finding a Spanish MTPE specialist by an order of magnitude. The talent pool is small, geographically dispersed, and heavily competed for by healthcare systems, school districts, and federal agencies simultaneously.

Planning around that scarcity — building relationships with qualified Chuukese and Pohnpeian linguists before a compliance deadline, not after — is the practical lesson that the MTPE productivity discussion tends to skip.

The Takeaway

MTPE is a genuinely useful workflow for the languages it covers. Adopt it where MT output is reliable enough to edit. Use full post-editing for anything with legal, medical, or reputational stakes. And build a separate, explicitly human workflow for Chuukese, Pohnpeian, and any other language where the MT infrastructure your vendor assumes simply does not exist.

If you're a healthcare system or school district assessing your language access coverage for Pacific Islander communities, we're glad to talk through what a compliant Chuukese or Pohnpeian translation program actually requires.