In cross-border litigation and arbitration proceedings, translation is often treated as a purely administrative step: documents are rendered into the procedural language so that the court or tribunal can understand them. Yet translation is often not only instrumental to evidence but rather part of the evidentiary infrastructure in itself, as it shapes how the facts of the case are perceived and whether they are admitted or not. Every translated clause, email, and witness statement helps establish what the tribunal will accept as “the facts.”

Within the EU, this dynamic is heightened by the coexistence of national languages, EU procedural languages, and private contractual languages. For a Spanish supplier’s correspondence with a Portuguese client, a German buyer’s submissions in English, or an arbitration tribunal seated in Paris, each link in the procedural chain is mediated through translation, and so the infrastructure that enables multilingual justice also silently conditions the evidentiary environment. The credibility of a piece of evidence, or its semantic “coherence,” is thus inseparable from decisions made by legal linguists who are, so to speak, behind the scenes.

There is a widespread assumption that a translation – a good one, at least – is transparent: that translation, by definition, is a frictionless rendering of the meanings of words in one language into words in a different language, in a one-to-one relation. Yet every translation involves terminological, syntactic, and conceptual choices that reflect a specific perspective – in the case of legal translation, a legal viewpoint. When translated documents enter the legal record, those translational decisions become part of what the tribunal recognizes as the facts of the case. Thus a witness statement phrased differently in Italian and in English, or a contract clause with multiple renderings of the Spanish legal term cláusula resolutoria, can subtly determine whether intention, obligation, or good faith are established, and how. (We discuss these choices extensively in our Untranslatables series). 

The practical impact of these choices emerges clearly in EU cross-border procedures. Ontanu and Pannebakker describe the example of a French consumer suing a German trader using the European Small Claims Procedure (ESCP) with help from the European Consumer Centre (ECC) Kehl/Strasbourg, which provides legal advice and dispute resolution for cross-border consumer issues within the EU, Iceland, Norway, and the UK. The claim is drafted in French on standardised ESCP forms, but, to be processed by the German court and understood by the defendant, it must be translated into German. The ECC often has to explain legal terms to the consumer, help them formulate the narrative, and arrange translation so that the claim is both linguistically and legally intelligible for all actors. In practice, the way the facts and legal characterisation are “packaged” across French and German versions determines whether the story of the dispute is clear enough to be admitted and acted upon at all.

Another example in the same study concerns the EU Judicial Atlas and e-Justice portal, which provide guidance on languages accepted for cross-border proceedings. The authors show that, while the Judicial Atlas correctly lists which languages France accepts for enforcement (French, English, German, Spanish, Italian), the e-Justice portal wrongly suggests that France accepts all of these languages for every ESCP form, including the claim- and answer forms used during proceedings. A claimant relying on the portal might file forms in English or German and only later receive a rectification request, incurring delays and extra translation costs before the matter can even enter the evidentiary phase. Here, inaccurate “translation” of language requirements into user-facing guidance shapes whether evidence can be validly filed in the first place and affects equality of arms between parties.

In arbitration proceedings, where the procedural framework is multilingual by design, translation makes it possible for the tribunal to understand the documents, but also plays a key role in establishing admissibility timelines, procedural fairness, and even jurisdictional scope. For instance, an inaccurately translated clause might alter the perceived scope of an arbitration agreement or the characterization of a breach. In that sense, the translator becomes an unacknowledged participant in the construction of legal reality.

This constitutive role of translation also appears clearly in EU regulatory practice. A striking example is the mistranslation of the General Block Exemption Regulation (GBER) into Romanian, which we discussed in another post, where a translation error in the authentic Romanian text broadened which companies could receive State aid compared with other language versions. On the basis of that erroneous wording, Romanian authorities granted aid to beneficiaries who would not have been eligible under the correctly translated provision. When the Commission later adopted a correcting regulation to fix the Romanian version, national authorities began reassessing and seeking recovery of aid already granted, raising litigation and forcing courts to decide whether the “correct” multilingual meaning could be applied retroactively to individuals who had relied in good faith on the mistranslation. Here, a single erroneous term in one language version effectively redefined the legal criteria for eligibility and then, through its correction, reshaped the evidentiary and legal framework around who was entitled to aid and who might have to repay it.

A similar dynamic appears in EU VAT fraud case law following the Court of Justice’s Kittel line of judgments. In some language versions of the relevant provisions and judgments, the taxable person must be “implicated” or “involved” in fraud, which suggests a relatively high threshold before they can be denied their right to deduct input VAT. In others, the wording is broader and speaks of being “connected” with fraud, which can be read as a lower threshold. When national courts and litigants argue over which language version and which phrasing best reflects the authentic EU meaning, they are not merely clarifying an existing rule: they are effectively constructing the operative legal test, that is, how close a trader must be to fraud to lose their deduction. That multilingual interpretation exercise directly shapes factual findings (what the trader “knew or should have known”) and ultimately who is treated as complicit and liable.

This constitutive role of translation becomes even more visible in high‑stakes investor-State and sovereign‑related disputes, where the language of the arbitration and of the underlying investment treaty can determine who has access to arbitration and under what conditions. Kılıç İnşaat v. Turkmenistan and Içkale İnşaat v. Turkmenistan are illustrative: these are companion ICSID cases in which Turkish construction companies sued Turkmenistan over public works contracts, where the key issue was Article VII(2) of the  1992 Agreement between the Republic of Turkey and Turkmenistan concerning the Reciprocal Promotion and Protection of Investments (BIT), which regulates recourse to local courts and international arbitration. The authenticated English version of Article VII(2) contained a convoluted “provided that, if” clause that could be read as making recourse to domestic courts optional before arbitration, while the Russian version, also authenticated, used wording that in expert evidence was translated as “on the condition that” or “provided that,” suggesting a mandatory domestic‑litigation requirement. Both tribunals had to interpret divergent language versions of the same clause using the Vienna Convention on the Law of Treaties, especially Article 33 on multilingual treaties, and both accepted that the Russian and English texts were authentic and that the Russian text, properly translated, clearly imposed a domestic litigation step.

The tribunals, however, drew different legal consequences: in Kılıç, the tribunal treated compliance with the domestic‑court step as a jurisdictional condition precedent and declined jurisdiction because the investor had not litigated in Turkmen courts for up to one year as the Russian version required. In İçkale, by contrast, the tribunal, while acknowledging Kılıç, characterised the same requirement as one of admissibility, and after analysing the factual context held that the claims were admissible and that a futility argument could not simply override the clause, before ultimately dismissing the claims on the merits.

Taken together, the two awards show how the translation and hierarchization of authenticated language versions of a regulation effectively redefines access to arbitration by deciding whether prior domestic litigation goes to jurisdiction, admissibility, or neither, and thus which disputes and factual narratives ever reach the merits phase. In this case, multilingual treaty texts, together with their translations and legal readings that make them operable in proceedings, do not simply transmit pre‑existing rules but actively construct the evidentiary and jurisdictional framework: indeed, they shape which claims may be heard at all, which facts are considered relevant, and how closely a State’s conduct must align with one linguistic version for a breach to be found.

Acknowledging translation as evidentiary infrastructure requires a shift in advocacy and procedural design. Counsel should approach translated evidence with awareness of its constitutive power: preserving source documents, disclosing translation processes, and engaging expert legal translators early. In cross-border dispute resolution, translation makes evidence intelligible, but also shapes evidence. Recognizing that transforms the way we pursue procedural fairness in a multilingual legal order, particularly in an EU setting where every step, from how forms are drafted and translated to how language options are communicated, helps decide which facts reach the court at all.