{"id":120781,"date":"2026-09-04T12:52:21","date_gmt":"2026-09-04T03:52:21","guid":{"rendered":"https:\/\/easternwest.net\/post\/?p=120781"},"modified":"2026-09-04T12:52:24","modified_gmt":"2026-09-04T03:52:24","slug":"ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions","status":"publish","type":"post","link":"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/","title":{"rendered":"AI Translation Looks Accurate\u2014Until One Critical Error Costs a Company Millions"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Picture a procurement director reviewing a freshly translated supply agreement. The English reads smoothly, the formatting is clean, and the turnaround took eleven minutes instead of eleven days. Everyone signs off. Eighteen months later, a dispute erupts over a single conditional clause, and the company discovers that a modal verb had been quietly flattened during machine translation, turning an obligation into a suggestion. \ud83d\udd0d That is the uncomfortable reality enterprise buyers are waking up to in 2026: AI translation can look finished while being fundamentally wrong, and the gap between &#8220;fluent&#8221; and &#8220;correct&#8221; is exactly where expensive mistakes live.<\/p>\n\n\n\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_87 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/#Why_Fluency_Is_Not_the_Same_as_Fidelity\" >Why Fluency Is Not the Same as Fidelity<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/#A_Real_Case_How_Two_Letters_Cost_a_Patent_Its_Validity\" >A Real Case: How Two Letters Cost a Patent Its Validity<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/#Where_Korean_Adds_an_Extra_Layer_of_Risk\" >Where Korean Adds an Extra Layer of Risk<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/#Comparing_the_Real_Cost_of_Getting_It_Wrong\" >Comparing the Real Cost of Getting It Wrong<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/#What_a_Defensible_Translation_Process_Actually_Looks_Like\" >What a Defensible Translation Process Actually Looks Like<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/easternwest.net\/post\/ai-translation-looks-accurate-until-one-critical-error-costs-a-company-millions\/#Reading_the_Fine_Print_Before_It_Becomes_a_Filing\" >Reading the Fine Print Before It Becomes a Filing<\/a><\/li><\/ul><\/nav><\/div>\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Why_Fluency_Is_Not_the_Same_as_Fidelity\"><\/span>Why Fluency Is Not the Same as Fidelity<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Independent benchmarking in 2026 puts general-purpose AI translation accuracy somewhere between 82% and 96%, depending on the engine, the language pair, and the content type (source: <a href=\"https:\/\/www.tomedes.com\" target=\"_blank\" rel=\"noopener\">https:\/\/www.tomedes.com<\/a>). That sounds reassuring until the arithmetic is applied to a real document. On a 10,000-word technical specification or licensing agreement, even a 96% accuracy rate leaves roughly 400 words that could be wrong, and there is no way to know which 400 without reading the entire translation line by line. At that point, a company is no longer saving time. It is simply outsourcing the discovery of errors to whoever reads the document next, often a client, a regulator, or opposing counsel.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The problem is not that AI engines are careless. It is that they optimize for statistical plausibility, not legal or technical consequence. A sentence can be grammatically flawless and still invert an obligation, drop a negation, or substitute a near-synonym that changes the scope of a claim entirely. Independent industry analysis of legal-document machine translation has found error rates as high as 15% to 25% for contract-heavy content, compared with the 98% consistency typically achieved by professional human linguists working within a quality assurance process (source: <a href=\"https:\/\/www.bluente.com\" target=\"_blank\" rel=\"noopener\">https:\/\/www.bluente.com<\/a>). For marketing copy, that margin barely matters. For a patent claim or an indemnification clause, it can determine the outcome of litigation.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"A_Real_Case_How_Two_Letters_Cost_a_Patent_Its_Validity\"><\/span>A Real Case: How Two Letters Cost a Patent Its Validity<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">This is not a hypothetical. In IBSA Institut Biochimique, S.A. v. Teva Pharmaceuticals USA, Inc., a company filed a U.S. patent application claiming priority to an Italian filing. Somewhere in the translation, the Italian word &#8220;semiliquido&#8221; became &#8220;half-liquid&#8221; instead of &#8220;semiliquid.&#8221; That single substitution introduced enough ambiguity that the Federal Circuit ultimately found the claim indefinite, and the patent was invalidated (source: <a href=\"https:\/\/ipwatchdog.com\" target=\"_blank\" rel=\"noopener\">https:\/\/ipwatchdog.com<\/a>). Years of research, filing fees, and legal strategy collapsed because of a translation choice that a specification writer would never have made, but a general-purpose translation process let slip through unnoticed.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Patent professionals see smaller versions of this story constantly. Rejections tied to translation inconsistency routinely add tens of thousands of dollars in additional legal fees and prosecution delays, and in opposition proceedings, a single mistranslated technical term \u2014 &#8220;average&#8221; swapped for &#8220;median,&#8221; for instance \u2014 has been enough to invalidate a claim in full, with no path to correction once the error is embedded in the granted document (source: <a href=\"https:\/\/www.gorodissky.com\" target=\"_blank\" rel=\"noopener\">https:\/\/www.gorodissky.com<\/a>). Once a patent estate is compromised in one jurisdiction, competitors often notice before the applicant does. \ud83d\udcc9<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Where_Korean_Adds_an_Extra_Layer_of_Risk\"><\/span>Where Korean Adds an Extra Layer of Risk<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">For companies translating into or out of Korean, the exposure compounds. Korean is a context-heavy, honorific-driven language, and the features that make it expressive are precisely the ones automated systems handle worst. Korean sentences routinely drop the subject and leave it to context, and when a machine has to guess that missing subject, it frequently guesses wrong while still producing something that reads fluently to a non-Korean reviewer (source: <a href=\"https:\/\/www.1stopasia.com\" target=\"_blank\" rel=\"noopener\">https:\/\/www.1stopasia.com<\/a>). A 2026 academic study on automatic translation of Korean honorifics found that translation models inconsistently shift the politeness register of a sentence depending on whether the addressee is stated explicitly, meaning the same source content can come out sounding respectful in one instance and casually dismissive in another, with no warning to the reader that anything changed (source: <a href=\"https:\/\/aclanthology.org\" target=\"_blank\" rel=\"noopener\">https:\/\/aclanthology.org<\/a>).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In a boardroom pitch, an investor update, or a compliance filing bound for a Korean regulator, that kind of register slip is not cosmetic. It can read as a lack of professionalism at best, and as a sign of disrespect toward the counterpart at worst \u2014 a serious liability in a business culture where hierarchy and formality shape how seriously a message is taken. As Korea&#8217;s export economy continues to expand and more enterprise buyers route contracts, technical documentation, and regulatory filings through Korean-English pipelines, the honorific and formality gap stops being a linguistic curiosity and becomes a commercial risk sitting quietly inside every unreviewed document. \ud83c\uddf0\ud83c\uddf7<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Comparing_the_Real_Cost_of_Getting_It_Wrong\"><\/span>Comparing the Real Cost of Getting It Wrong<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Document Type<\/th><th>Common AI-Only Failure<\/th><th>Documented Consequence<\/th><\/tr><\/thead><tbody><tr><td>Patent specification<\/td><td>Substituted or dropped technical term<\/td><td>Claim invalidation, loss of protection (Federal Circuit precedent)<\/td><\/tr><tr><td>Cross-border contract<\/td><td>Flattened modal verbs, missing negation<\/td><td>Voided clauses, unintended obligations, litigation<\/td><\/tr><tr><td>Regulatory filing<\/td><td>Compliance terminology mismatch<\/td><td>Fines, delayed approval, resubmission<\/td><\/tr><tr><td>Korean business communication<\/td><td>Incorrect honorific register<\/td><td>Damaged trust, perceived disrespect toward counterpart<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">The pattern across every row is the same: the error is invisible to anyone who cannot read both languages fluently and does not understand the underlying legal or technical framework. That is precisely the review gap that a purely automated workflow cannot close on its own.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"What_a_Defensible_Translation_Process_Actually_Looks_Like\"><\/span>What a Defensible Translation Process Actually Looks Like<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">None of this is an argument against using AI as part of a translation workflow. Modern engines are genuinely useful for first-pass drafts, terminology consistency checks, and high-volume content where the stakes are lower. The distinction that matters is between AI as an assistive tool inside a supervised process and AI as the entire process. A defensible workflow for high-stakes content generally includes:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Subject-matter review by a linguist who understands the specific domain, not just the language pair<\/li>\n\n\n\n<li>A structured comparison against the source document, sentence by sentence, for legal and technical material<\/li>\n\n\n\n<li>Terminology management that locks down key terms across an entire document set, so &#8220;average&#8221; and &#8220;median&#8221; never get treated as interchangeable<\/li>\n\n\n\n<li>A second reviewer for anything that will be filed with a court, patent office, or regulator<\/li>\n\n\n\n<li>Explicit handling of honorifics and register for Korean-language business communication, rather than leaving formality decisions to a default setting<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Enterprise buyers evaluating vendors in 2026 are increasingly asking to see this process documented before a contract is signed, not after a dispute forces the question. That shift in buyer behavior reflects a broader recognition: the cost of verifying a translation is always smaller than the cost of litigating one. \ud83d\udcbc<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Reading_the_Fine_Print_Before_It_Becomes_a_Filing\"><\/span>Reading the Fine Print Before It Becomes a Filing<span class=\"ez-toc-section-end\"><\/span><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The companies that get burned by translation errors rarely had a translation problem in isolation. They had a review problem \u2014 a document that moved from source language to filed document without anyone positioned to catch the specific kind of error that only shows up when technical, legal, and cultural context all have to line up at once. Speed is not the enemy here. Unsupervised speed is. Whether the document in question is a patent claim heading toward a Federal Circuit courtroom or a Korean-language partnership agreement heading toward a boardroom, the question worth asking before submission is not &#8220;did the translation finish quickly,&#8221; but &#8220;did anyone with the right expertise actually check it.&#8221;<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>References<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>source: <a href=\"https:\/\/www.tomedes.com\" target=\"_blank\" rel=\"noopener\">https:\/\/www.tomedes.com<\/a><\/li>\n\n\n\n<li>source: <a href=\"https:\/\/ipwatchdog.com\" target=\"_blank\" rel=\"noopener\">https:\/\/ipwatchdog.com<\/a><\/li>\n\n\n\n<li>source: <a href=\"https:\/\/aclanthology.org\" target=\"_blank\" rel=\"noopener\">https:\/\/aclanthology.org<\/a><\/li>\n\n\n\n<li>source: <a href=\"https:\/\/www.gorodissky.com\" target=\"_blank\" rel=\"noopener\">https:\/\/www.gorodissky.com<\/a><\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Picture a procurement director reviewing a freshly translated supply agreement. The English reads smoothly, the formatting is clean, and the turnaround took eleven minutes instead of eleven days. Everyone signs off. Eighteen months later, a dispute erupts over a single conditional clause, and the company discovers that a modal verb had been quietly flattened during [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_et_pb_use_builder":"off","_et_pb_old_content":"","_et_gb_content_width":"","footnotes":""},"categories":[1],"tags":[],"class_list":["post-120781","post","type-post","status-publish","format-standard","hentry","category-translation-updates"],"_links":{"self":[{"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/posts\/120781","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/comments?post=120781"}],"version-history":[{"count":3,"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/posts\/120781\/revisions"}],"predecessor-version":[{"id":120784,"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/posts\/120781\/revisions\/120784"}],"wp:attachment":[{"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/media?parent=120781"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/categories?post=120781"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/easternwest.net\/post\/wp-json\/wp\/v2\/tags?post=120781"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}