How Native Is Eiken's Official Website English?

The Paradox: Literal Translations Score Higher

On this blog, I translate Japanese articles into English with AI and then auto-evaluate the quality through back-translation. The English gets translated back into Japanese, and I measure how close it lands to the original using cosine similarity over embeddings.

There's a catch with this approach: the more literal the English, the higher the score tends to be.

A word-for-word "literal" translation stays close to the source when you flip it back to Japanese. The semantic preservation is high. But to a native English reader, that same translation often feels stiff and unnatural — and yet this scoring system happily labels it "high quality."

So what does "grammatically correct but slightly off to a native speaker" English actually look like? I stumbled onto a perfect example: EIKEN's official website.

I Read the English on EIKEN's Site

I went through the English version of the website for EIKEN (the Test in Practical English Proficiency).

Grammatically, it's solid. No typos. The content covers everything it needs to. But reading it from end to end, while it's what you'd expect from an educational institution's official documentation, it feels a touch stiff for a modern website.

I'm not saying this is bad English. What's interesting is the structural reason this kind of English gets produced in the first place. And it happens to be exactly the type of English that would probably score well on back-translation metrics.

What Does It Actually Look Like?

Here are some passages pulled straight from the site.

Passive Voice Everywhere

EIKEN can be used as a language skills certificate for study abroad as it is recognized in approximately 400 universities and educational institutions in North America, Australia, and throughout the world.

Two passive constructions in one sentence ("can be used", "is recognized"). The subject is "EIKEN," but we never find out who's doing the recognizing, or who gets to use it. Something like "Over 400 universities worldwide accept EIKEN scores" — written in the active voice — would land much more directly with the reader.

That said, this isn't a quirk unique to EIKEN. It's pretty much standard fare on Japanese public institution websites.

Cramming Everything into One Sentence

EIKEN is an abbreviation of Jitsuyo Eigo Gino Kentei (Test in Practical English Proficiency), one of the most widely used English-language testing programs in Japan.

EIKEN is produced and administered by the Eiken Foundation of Japan (formerly the Society for Testing English Proficiency, Inc.), a public-interest incorporated foundation based in Tokyo.

Every sentence opens with "EIKEN is...", slots in some parenthetical context, then keeps piling on modifying clauses. It's the cram-everything-into-one-sentence style. Accuracy and completeness over readability.

This isn't a problem with the translation — it's that the Japanese version is written that way to begin with. When you take a Japanese passage that carefully layers information piece by piece and convert it straight into English, this is what you get.

The Institutional Tone

There are eight tests within the EIKEN framework, each representing a different ability level. The levels are called grades and are given on a pass-or-fail basis.

Group test sites (junkaijo in Japanese) are the approximately 18,000 junior high schools, high schools, colleges, and other institutions that have been approved to administer the EIKEN tests on an individual basis.

A dry, matter-of-fact walkthrough of how the system works. Quite a contrast with TOEFL or IELTS sites, which lead with benefits like "Prove your English skills to the world." But I think EIKEN's site is leaning into its role as a user guide rather than marketing copy. Different goals, different tone.

Why Does It Come Out This Way?

It's easy to write off EIKEN's English as "bad," but it's more productive to think about the structural reasons it ends up this way.

It Started in Japanese

EIKEN's site has a Japanese version, and the English version is its translation. When you're producing the English version as a work project, you typically have to convert the Japanese content completely and faithfully into English.

Japanese organizations tend to treat "an expression that exists in Japanese but not in English" as a missing piece of information, and translators often get caught in the middle.

• Accuracy first: Even if you paraphrase for readability, a reviewer will flag "this part of the original is missing from the English" and send it back.
• Risk avoidance: Cutting words gets read as "you changed the content on your own authority." Leaving everything in — however verbose — is the safer play.
• Public institution formality: It's tied into the MEXT framework, and it's hard to deviate from a near-regulatory "official document" register.

This isn't a knock on EIKEN's translators — it's about what the organization expects from translation. The criteria for choosing between accuracy and readability are heavily weighted toward the Japanese side.

The Opposite Extreme: Movie Subtitles

For another flavor of Japanese-to-English translation, look at the English subtitles for movies and anime — they take the exact opposite approach.

Subtitles operate under tight constraints on character count and on-screen time, so the priority isn't "leaving nothing out" but getting the meaning across in as few words as possible. A small loss of nuance is fine — what matters is that the viewer gets it in the moment.

Japanese honorifics and roundabout phrasings, for instance, get ruthlessly trimmed in English subtitles. Information density drops, but the English flows naturally.

Set EIKEN's site next to subtitle translation and you can see clearly: "good" translation depends entirely on the goal. And if you ran back-translation scoring on both, subtitles would almost certainly score lower. They're discarding information, so the result drifts further from the Japanese when you reverse it. But to a native speaker, the subtitles are far more natural.

A higher score doesn't always mean a better translation — the recurring theme on this blog shows up again here.

Not a Style Sample, but a Brochure

EIKEN's site isn't a place to showcase model English — it's a brochure explaining how a test works. Yes, it reads stiffer than the Cambridge English or IELTS sites, but those were almost certainly written in English from the start, while EIKEN's site is converting a Japanese-language institution into English. There's a fundamental difference.

It might not even be fair to compare them on the same terms.

Tying It Back to Back-Translation Scores

Coming back to where we started.

The English on EIKEN's site faithfully preserves the structure of the original Japanese. The heavy passive voice, the long sentences, the refusal to drop information — all of these come from prioritizing "accurate transmission of the Japanese content" above all else.

In other words, the semantic preservation rate is high. Back-translate it into Japanese and you'd get something quite close to the original.

Meanwhile, a site like IELTS — cutting information, leading with benefits — reads naturally to natives but drifts away from a literal back-translation. The score drops.

"High score" doesn't equal "good English." This is a structural limitation of back-translation scoring, and it's a challenge this blog's translation pipeline needs to confront going forward.

"Good Writing" Has Different Shapes in Japanese and English

The stiffness of EIKEN's English isn't unrelated to how English is taught in Japan.

Japanese tends to layer information carefully and arrive at the conclusion at the end. English does the reverse — lead with the conclusion, then add supporting information. Map Japanese structure straight onto English and the English ends up roundabout.

Since 1963, EIKEN has had an outsized presence in English education in Japan. The fact that its official site reads as "English that traces the contours of Japanese structure" is, in a way, a snapshot of where Japanese English education stands today.

The test itself has been steadily modernizing — stronger writing components, new grade levels — so it's a shame that the English on the front door hasn't kept pace. But as long as you're "translating" the institution into English, the structure naturally tends this way. Changing it would require more than better translation: it would mean rewriting the English version with English logic from scratch.

What This Means for My Pipeline

Looking at EIKEN's English drove home the limits of measuring translation quality by back-translation score alone — again.

• High score = meaning is preserved. But it tells you nothing about whether the English feels natural to a native.
• Low score = meaning has drifted. But it might also mean the translator paraphrased into more natural English.

This blog's translation pipeline needs a way to evaluate "native feel" as a dimension orthogonal to the score. Back-translation works as a sanity check, but it's incomplete as a grading system. EIKEN's site turned out to be a great concrete example of just how that incompleteness shows up.