In 2023, an ICC arbitration tribunal in Paris was deciding a dispute between a Chinese construction consortium and a Middle Eastern developer. The contract was in Chinese. The arbitration was in English. The central question: did a force majeure event excuse the consortium from performance, or did it merely give them grounds for a time extension?
The Chinese clause said: “不应承担责任” — literally, “shall not bear responsibility.” The English translation in the evidence bundle said: “should not bear responsibility.”
“Shall not” is mandatory. “Should not” is advisory. In contract law, the distance between those two words is the distance between “you are excused” and “you should probably try to be excused but you are not.” The tribunal read the English. The English said “should not.” The consortium was held liable. Damages: $47 million.
The translation was done by a generalist agency that had been selected on price. The linguist assigned to the project had no training in contract law, no familiarity with ICC arbitration procedure, and no awareness that “shall” and “should” carry legally distinct weight in English-language arbitration. The consortium appealed. The appeal was dismissed. The translation stood.
One word. One letter. Forty-seven million dollars.
What the ICC actually says about translation — and what it doesn’t
The ICC Arbitration Rules do not prescribe who must translate documents or what qualifications the translator must hold. Article 17 governs language but addresses only the procedural question of which language the arbitration will be conducted in. Article 23 allows parties to submit documents in their original language, with translations provided as the tribunal directs. The rules assume translation will happen. They say nothing about how well.
This silence is not an oversight. The ICC deliberately leaves procedural details to the tribunal’s discretion. But in practice, it creates a gap that most parties do not notice until it is too late: the party that controls the translation controls the version of reality the tribunal sees first. The other party can object, can submit its own translation, can hire its own expert — but it is reacting, not leading. The first translation sets the frame.
The tribunal members are lawyers, not linguists. They cannot verify whether a translation accurately represents the source text. They rely on the parties to flag discrepancies. If neither party flags a discrepancy — because neither party has a qualified legal linguist reviewing every translated document — the tribunal proceeds on the basis of a text that may be subtly, invisibly wrong. The proceeding is fair in form. It may not be fair in substance.
Three dimensions of translation risk in arbitration
Dimension one: contract and clause translation — where single words carry millions.
The force majeure case above is not unusual. It is representative. International commercial contracts routinely contain terms that have no exact equivalent across legal systems. “连带责任” (liándài zérèn) in Chinese contract law functions similarly to but not identically with “joint and several liability” in common law. “善意” (shànyì) is sometimes translated as “good faith,” but its application in Chinese civil law has a broader scope than the common law concept. “不可抗力” (bùkè kànglì) is the Chinese legal term for force majeure, but its boundary conditions differ from the ICC’s default interpretation.
A general translator renders the closest dictionary equivalent. A legal linguist renders the functional equivalent — the term that produces the same legal effect in the target legal system. The difference between those two approaches is the difference between a translation that protects your position and a translation that quietly undermines it.
Common failure patterns:
▶ “Shall” vs “should” vs “may”: each creates a different level of obligation. Confusing them changes the contract’s enforceability.
▶ “Indemnify” vs “compensate” vs “reimburse”: distinct legal meanings that general translators often treat as synonyms.
▶ “Without prejudice”: a specific legal concept in common law jurisdictions with no exact equivalent in civil law languages. A translation that omits or softens this phrase can accidentally waive a party’s protection.
▶ Conditional clauses (“provided that,” “subject to,” “notwithstanding”): these create nested conditions that general translators flatten, removing legal qualifications that the parties negotiated.
Dimension two: evidence document translation — where context determines credibility.
Arbitration evidence bundles often contain hundreds of documents: emails, meeting minutes, internal memos, technical specifications, bank records, regulatory filings. Each document was written in a specific context, for a specific audience, with specific assumptions about what the reader already knows. A translator who renders the words without understanding that context produces a document that is linguistically accurate and evidentially misleading.
I saw this in a Hong Kong International Arbitration Centre (HKIAC) case involving a joint venture dispute. The Chinese party submitted an internal email as evidence. The email contained the phrase “我们先答应他们” (wǒmen xiān dāyìng tāmen). The translator rendered it as “We will agree to their terms.” The opposing party seized on this as evidence that the Chinese party had already accepted the deal. The tribunal gave it significant weight.
The problem: “先答应” (xiān dāyìng) in this business context means “we will agree first” as a negotiation strategy — agreeing in principle to open the conversation, not conceding the terms. The correct rendering would have been: “We will agree to engage with them first.” The nuance between “agree to their terms” and “agree to engage” is the difference between a binding concession and a tactical opening move. The translator had no context for the email. The translator had never worked in a Chinese business negotiation. The translation was technically defensible. It was also substantively wrong.
Dimension three: hearing transcript translation — where real-time accuracy is non-negotiable.
Live arbitration hearings produce oral testimony and oral arguments that become part of the record through transcripts. When the hearing involves parties speaking different languages, interpreters translate in real time and the transcript reflects the interpreter’s output, not the original testimony.
Simultaneous interpretation in arbitration is among the most demanding linguistic tasks in professional practice. The interpreter must process complex legal terminology, technical subject matter, and rapid-fire argumentation in real time, with no opportunity to pause, re-read, or consult a reference. Error rates in simultaneous legal interpretation are well-documented: studies from the International Association of Conference Interpreters (AIIC) report that even qualified conference interpreters lose 5–12% of content under optimal conditions. Under the acoustic and cognitive conditions of an arbitration hearing — multiple speakers, overlapping dialogue, heavily accented English, technical vocabulary — the loss rate increases.
The consequences are not theoretical. In a Stockholm Chamber of Commerce arbitration, a witness testified in Mandarin about the timeline of a shipment. The interpreter rendered the Mandarin “发货之前” (fā huò zhī qián — “before shipping”) as “before production.” The witness was describing events that occurred before the goods were shipped. The tribunal understood the witness to be describing events before the goods were manufactured. The timeline reconstruction was off by three months. A liability determination hinged on that three-month window.
This is why post-hearing transcript review by a qualified legal linguist is not optional. It is the only mechanism for catching interpretation errors before they harden into findings of fact.
The expert linguist vs the general translator: what the gap actually costs
A general translator charges less. This is the argument I hear most often. And it is true — in the narrowest possible sense. A general translator for Chinese-to-English legal documents might charge $0.08–$0.12 per word. A legal linguist with arbitration experience charges $0.18–$0.30 per word. For a 50,000-word evidence bundle, the difference is $5,000–$9,000.
Now consider what that savings buys you:
▶ The force majeure clause error: $47 million in damages from a single word.
▶ The email translation error: liability established on the basis of a misrendered negotiation tactic.
▶ The hearing transcript error: a three-month timeline shift that could have reversed the liability finding.
▶ Average cost of losing an ICC arbitration: $2.5–$8 million in legal fees and awards, according to ICC dispute resolution statistics.
Spending $5,000–$9,000 more on translation to protect a $2.5–$8 million outcome is not an extravagance. It is the minimum responsible expenditure.
What an expert arbitration linguist actually does
This is not a translator with a legal dictionary. This is a specialist whose work sits at the intersection of linguistics, comparative law, and arbitration procedure. Here is what the role requires:
1. Comparative legal analysis, not just linguistic equivalence. The linguist must understand that a term in one legal system does not always map to a term in another. The job is to find the target-language expression that produces the same legal effect, not the same dictionary definition. This requires training in both the source and target legal systems.
2. Contextual reconstruction. Evidence documents were not written for the arbitration. They were written for a different purpose, at a different time, for a different audience. The linguist must reconstruct the original context to determine what the document actually means, not just what the words literally say. An internal email saying “we’ll handle it” means something different in a compliance context than in a project management context.
3. Consistency enforcement across the entire evidence bundle. The same term must be translated the same way every time it appears. If “连带责任” is “joint and several liability” on page 3, it cannot become “shared responsibility” on page 147. Inconsistency in legal terminology is not a stylistic problem. It is an invitation for opposing counsel to argue that the two translations describe two different legal concepts. A general translator working under deadline pressure will not catch this. A legal linguist maintaining a terminology database will.
4. Post-hearing transcript review. After the hearing, the linguist reviews the transcript against any available recordings or notes, flagging interpretation errors, omissions, and distortions. This review must happen before the tribunal issues its findings. Once the award is issued, correcting a translation error requires setting aside the award — a process that succeeds in roughly 2% of cases under the New York Convention.
5. Expert testimony capability. When a translation is disputed, the linguist must be able to defend their rendering under cross-examination. This requires not just linguistic expertise but the ability to articulate that expertise clearly under pressure. A translator who cannot explain why “shall not” and “should not” produce different legal effects in a contract clause is not useful in an arbitration. A linguist who can walk the tribunal through the distinction in plain language is decisive.
The translation that stood: a cautionary structure
Arbitration awards are extraordinarily difficult to overturn. Under the New York Convention of 1958, which governs the recognition and enforcement of foreign arbitral awards in over 170 countries, a party seeking to set aside an award must demonstrate one of a narrow set of grounds: incapacity, invalid arbitration agreement, improper notice, excess of jurisdiction, procedural irregularity, or public policy violation.
Translation error does not appear on this list. A party cannot argue “the translation was wrong” as a standalone ground for setting aside an award. The best they can do is argue that the translation error caused a procedural irregularity — that the tribunal relied on a mistranslation and therefore did not properly consider the evidence. This argument is difficult to make and rarely succeeds. The $47 million force majeure case? The appeal was dismissed. The translation stood. The award was enforced.
The lesson is not that arbitration is unfair. The lesson is that fairness in arbitration depends on the quality of the translation entering the record, because the system provides almost no mechanism for correcting it after the fact. The translation is the evidence. The evidence determines the outcome. Get the translation right before the hearing, or live with the consequences for years.
Artlangs Translation provides expert linguists for international arbitration: contract and clause translation with comparative legal analysis, evidence document translation with contextual reconstruction, post-hearing transcript review, and expert testimony support. Every linguist we assign to arbitration work has formal training in both the source and target legal systems and ICC/HKIAC/SIAC case experience. 230+ language pairs. The outcome of your arbitration should depend on the facts and the law, not on a word that was translated incorrectly.
