The Problem: Language Distribution Nobody Mapped
BigLaw litigation support teams run language identification late in the workflow, if at all. By the time a Relativity project lead realizes that 40% of the custodian population communicated in Mandarin and another 15% in Spanish, the review budget has already been allocated against an assumed English-dominant corpus. Certified translation costs in ICC arbitrations run $150K–$400K per matter when left unmanaged — not because the volume is unavoidable, but because nobody produced a language-distribution analysis before committing to a review structure.
How an AI Agent Approaches It
An AI Labor Company agent runs a full language-distribution scan across the document universe in Relativity and Nuix at the outset, producing a breakdown by custodian, date range, and document type. From that map, it prioritizes translation spend on the clusters most likely to contain responsive, high-relevance material — applying machine-translated summaries to the remainder so bilingual reviewers in Relativity can make quick privilege and responsiveness calls with the original document alongside. The result is a defensible, tiered translation workflow where certified human translation is reserved for the documents that actually need it.
What This Is Worth
Certified translation is one of the most controllable cost lines in an international arbitration — but only if you have the data to make the call. Teams using this approach typically see translation cost reductions in the 55–75% range, not by cutting corners on quality but by eliminating translation spend on low-relevance documents. The agent is live and producing the language-distribution analysis in roughly 10 weeks. The capacity freed from manual language triage also lets your review team move faster on the substantive document review, which matters when an ALJ scheduling order is running.
Is machine-translated content defensible in ICC arbitration proceedings?
Machine-translated summaries are used for internal review prioritization, not for production. Documents produced to the tribunal are handled with certified human translation where required. The agent's output helps counsel decide which documents warrant that certified spend.
Can this workflow integrate with our existing Relativity review structure?
Yes. The agent loads machine-translated summaries and tagged exhibits into Relativity alongside originals, so your existing review batching, coding panes, and quality control workflows stay intact. Bilingual reviewers see both the original and the summary in their standard review interface.
What if the document population includes more than two languages?
The language-distribution analysis covers the full custodian corpus regardless of how many languages are present. The prioritization model weights translation spend decisions on relevance clusters, not on language count — so a three-language or five-language population is handled with the same logic.