Inconsistent terminology is one of the less visible problems in cross-border legal work. International firms work across jurisdictions and languages every day, but the terminology they use has often developed locally and incrementally. It may sit in old matters, precedent banks, translation memories, internal glossaries, local-office know-how and the preferences of individual lawyers.

A German legal concept may therefore be rendered one way in one transaction and differently in another. Both translations may be linguistically defensible. The more important question is whether they point to the same legal concept, and whether the English terminology being used preserves the distinctions that matter in German law.

For many years this was mainly a human coordination problem. Lawyers and translators reconciled terminology when inconsistencies became visible. The growth of legal AI changes the scale of the issue because the same fragmented material is increasingly made available to systems used for research, drafting, review and knowledge retrieval. An AI system can often infer that different expressions are related, but it may also encounter the same English label attached to concepts that are not fully equivalent, or several English labels for the same local concept.

The underlying problem is therefore broader than translation consistency. It concerns the organisation of legal information. If a firm wants its lawyers and AI systems to work from a common understanding, it needs some way of recording which legal concept is being discussed, the terminology used for it, the jurisdiction in which it operates and, where relevant, how it corresponds to concepts elsewhere.

This doesn’t require every historical document to be rewritten or every terminological variation to be eliminated. Variation is often legitimate. The useful distinction is between variation that reflects style and variation that reflects a different legal meaning. A data strategy should help make that distinction visible.

This becomes particularly important in cross-border workflows. A lawyer reviewing a contract, researching foreign law or drafting for another jurisdiction may encounter familiar terminology and reasonably assume that it carries familiar legal consequences. The AI system assisting that lawyer can make the same approximation. Sometimes it will be right. Sometimes the local concept will overlap only partially with the concept suggested by the English label.

A structured terminology layer can provide a more stable reference point. The local concept can be identified and defined in its own jurisdictional context, preferred and alternative terminology can be recorded, and relationships with concepts elsewhere can be analysed. The information can then be reused across different workflows rather than reconstructed each time.

The practical value is consistency, but not consistency for its own sake. The objective is to ensure that the same underlying legal distinctions remain available when information moves between lawyers, offices, languages and AI applications.

As legal organisations make more of their knowledge available to AI, terminology therefore becomes part of a wider data-governance question. The issue is no longer simply how a term should be translated, but how legal concepts should be represented so that people and machines can work from the same structured legal foundation.