How Do US Regulated Teams Control Terminology Drift in Translation?

Table of Contents
What is the direct answer?
Terminology drift is the gradual or sudden use of different target-language terms for the same source concept. US regulated teams control it with an approved bilingual termbase, named term owners, termbase access for every translator, terminology checks in review and final quality assurance, and an audit trail of each term change. Without these controls, legal, medical and financial documents diverge across languages.
What is terminology drift, and what is its scope?
Terminology drift is the loss of consistency between the approved term for a concept and the term that appears in translated content. It happens across documents, language versions, vendors, release cycles and reviewers.
The MQM Council’s FULL Typology defines terminology errors as terms that do not conform to normative subject-field or organizational terminology, or that are not the correct normative equivalent of the source term. It lists two relevant subtypes: inconsistency with a terminology resource, and inconsistent use of terminology, meaning multiple terms for the same concept.
A study of terminology translation in the journal Fachsprache describes a constant tension between the need for precision and stability and a pull toward variation. It identifies irregular, incoherent translations as a cause of inconsistencies in the target language.
How does it differ from linguistic or semantic drift?
Linguistic or semantic drift is a change in the meaning of a word across a language community over time. Terminology drift is an operational failure inside a defined content set. A term had an approved equivalent, and the content stopped using it.
The same Fachsprache study also notes that variation can be a valuable contribution when it comes from an author’s deliberate cognitive distinction. Intentional, authoritative variation is a governed change. Drift is variation that nobody approved.
What is in scope for a regulated US team?
Scope covers any content where a term has a defined meaning that a reader, auditor or court relies on:
Legal: contracts, policies, discovery material, regulatory filings.
Life sciences: instructions for use, labeling, clinical documentation, regulatory submissions.
Finance and banking: disclosures, product terms, compliance procedures.
Software and IT: user interface strings, security documentation, support content.
Machinery and manufacturing: manuals, safety instructions, parts lists.
The data category matters too. Documents that contain personal data, protected health information or confidential designs add a security duty to the terminology duty. The vendor must meet both.
Which steps prevent terminology drift, and who owns each control?
The steps below assign an owner and a control to each stage. Each owner is a role, not a person, and the roles can sit in localization, regulatory affairs, legal or the vendor.
Step 1: Define the approved terms
Owner: Terminology owner, usually in localization or regulatory affairs, with a subject-matter expert for each domain.
Control: A bilingual termbase with an approval status for each entry. Each entry records the source term, the approved target term, a definition, the domain and the language variant.
The LISA terminology survey found that adherence to company or organization standards led term-selection criteria, cited by 85% of respondents. Historical use in context and industry standards followed at 75 to 80%. The same survey notes that over-applying historical use can create resistance to changing incorrect terms. The owner should therefore approve a correction even when the old term is established.
Step 2: Put the termbase into the requirements
Owner: Localization manager and procurement.
Control: A written requirement that the termbase applies throughout the project. ASTM F2575-2014, the US standard for translation services, states that terminology management should apply throughout the entire project to ensure consistency. It recommends that unusual or specially translated terms go into a bilingual term base supplied by the requester or developed by the provider. Any provider-developed terminology is made explicit in Additional Tasks.
The requirement should name who supplies the termbase, who approves new terms and how long approval may take.
Step 3: Give translators and reviewers access
Owner: Vendor project manager.
Control: Access control that places the current termbase and translation memory in front of every linguist. The LISA survey found that all respondents considered giving translators access to a bilingual termbase an effective strategy.
Access also means version control. A linguist working from last quarter’s glossary produces drift that looks like compliance.
Step 4: Check terminology in review and final QA
Owner: Reviewer, then the quality assurance lead.
Control: A terminology check in subject-matter review and a second check in final QA. In the LISA survey, 70% of respondents found checking terminology during review effective. Virtually all service providers rated it very effective. Of translation and localization service providers, 80% found reviewing terminology during final testing and quality assessment effective.
Automated QA flags terms that do not match the termbase. A human reviewer decides whether a flag is an error or an approved exception.
Step 5: Manage change with an audit trail
Owner: Terminology owner.
Control: An audit trail that records who changed a term, when, why and who approved it. Each change triggers a decision on whether to update already published content. The record lets an auditor trace why two documents use different terms.
Step 6: Close the feedback loop
Owner: Localization manager and the vendor.
Control: Review of QA logs and reviewer comments to find which terms fail repeatedly. David Calvert’s paper Lost for Words summarizes supplier best practice as consistency, documentation and communication. It says feedback loops improve existing terminology.
What criteria decide how much control a document needs?
Not every document needs the same control. Calvert recommends risk-matrix thinking: grade the likelihood of a consequence against its severity for specific error types. Use these criteria to place each content type.
Criterion | Lower control | Higher control |
Consequence of a wrong term | Internal confusion | Patient safety, legal liability, regulatory rejection |
Reader | Internal staff | Patients, customers, regulators, courts |
Term status | Descriptive wording | Defined term, product name, regulated claim |
Data category | Public content | Personal data, protected health information, trade secrets |
Update frequency | Static | Frequent releases |
Language count | One or two | Many, including regional variants |
Calvert grades terminology consequences from impaired internal communication to danger to life or limb. He notes that wrong job titles cause embarrassment and lost trust. Wrong product names risk expensive late-stage corrections once presses roll.
The decision follows from the table. A defined term in a regulated document read by an outside party gets the full control set: approved termbase, subject-matter review, final QA and an audit trail. Marketing copy gets lighter review, but its brand and product names still come from the termbase. Consistent terms also support brand consistency across markets, which matters for global marketing and SEO teams.
What should a vendor be required to do?
Vendor duties map to the steps above. Write them into the statement of work:
Load and use the client’s termbase and translation memory on every project.
Use linguists qualified in the subject field.
Run terminology checks in review and final QA, and report the results.
Propose new terms for approval and never invent them silently.
Keep an audit trail of term and content changes.
Protect client data under a stated security and privacy regime.
Hold quality certifications relevant to the sector, such as ISO 17100 for translation services.
What do examples of terminology drift look like?
The MQM FULL Typology gives three examples that show the pattern.
Termbase violation. The termbase specifies “USB memory stick”, but the text uses “USB flash drive”. The meaning is the same, but the document no longer conforms to the approved term.
Concept splitting. A component is called “brake release lever”, “brake disengagement lever”, “manual brake release” and “manual disengagement release” in one text. A technician may read four parts where there is one. In machinery documentation, this affects safety instructions and parts orders.

Product name corruption. A product specified as “Acme Turbo2000” appears as “Acme Turbo” or “Turbo200”. In regulated labeling or a contract schedule, a changed product name can mean a different product.
How does this look in a medical device file?
A device manual uses one approved target term for a warning concept in the first language. A later update, handled by a different linguist without the termbase, uses a near synonym. A reviewer reading only the new text sees nothing wrong. A comparison against the termbase shows a mismatch, and the mismatch then needs a decision: correct it, or approve the new term and update all earlier content.
How does this look in legal content?
Legal terms often lack exact equivalents across legal systems. A study of UN human-rights documents by Guzmán and Prieto Ramos analyzed more than 5,000 bi- and trilingual segments from 2000 to 2017. It found that legal asymmetry significantly affects translation decisions. It linked variation within a text to source text length and document series. For a contract or policy set, this means long documents and long series need more controls, because variation grows with length and with the number of documents.
If a defined term in a contract is rendered two ways, the parties may argue over which meaning applies. The controls above exist to prevent that dispute. Where a dispute concerns a specific jurisdiction’s law, the reader should consult counsel.
What are the failure modes, and what are the mitigations?
Failure mode | Cause | Mitigation |
Linguist never saw the termbase | No access control or outdated version | Central termbase with version control and access for all linguists |
Termbase is incomplete | Terms added after problems appear | Extract candidate terms before translation starts and get approval |
Several vendors, several glossaries | Each vendor keeps its own list | One client-owned termbase shared with all vendors |
Source terms change | Product or regulation updates | Change log, impact review and update of existing content |
Outdated term kept by habit | Historical use overrides correctness | Owner approves corrections and records the reason |
Reviewer checks only fluency | Terminology not in the review scope | Add a terminology check to the review checklist |
Final QA skipped | Schedule pressure | Make final QA a release gate |
Machine output ignores terminology | Tool not connected to the termbase | Constrain output with the termbase and add human review |
Client does not drive terminology | No owner on the client side | Name a terminology owner |
The last row has evidence behind it. A 2010 SDL survey of more than 1,500 linguists, reported in a Tradumàtica article, found that 92% rated terminology management very important. Only 15% said clients drove it during projects. A named client-side owner closes that gap.
The cost of weak control is also documented. A study on measuring terminology consistency notes that the European Commission spends millions of euros per year on translation quality control. It adds that language-version discrepancies cause interpretation problems mostly related to terminology. The study defines terminological inconsistency as the use of two or more translation variants for one source term, which makes the problem measurable.
A further failure mode is losing the clear distinction between drift and approved variation. Not every difference is an error. A term list can mark approved synonyms, regional variants and context-specific equivalents, so the QA tool and the reviewer know which differences are intended.
Where does AD VERBUM fit?
AD VERBUM is a language service provider with 25+ years of industry experience. It delivers AI+HUMAN hybrid translation with subject-matter experts, drawn from a network of 3,500+ linguists. It integrates client translation memories and term bases. Its proprietary LangOps System is hosted on EU servers.
AD VERBUM holds ISO 17100, ISO 18587, ISO 13485, ISO 27001 and ISO 42001, among other certifications, audited by Bureau Veritas. It aligns to GDPR and HIPAA and supports 150+ languages, including regional variants. See the AD VERBUM services for translation, localization, interpretation, multilingual SEO, voice over and multilingual documentation.

What is our view on terminology drift?
Terminology drift is a governance problem before it is a language problem. A linguist can translate every sentence correctly and still break a document set, because the set has no single owner for its terms. We treat the client’s termbase and translation memory as controlled assets. Our AI+HUMAN hybrid process places subject-matter experts at the review step, where a term’s regulatory meaning is decided. Quality management aligned to ISO 17100 and ISO 18587 gives that review a defined place in the workflow. Regulated clients also need to know where their terminology data is processed. Our LangOps System runs on private EU-hosted infrastructure, so the termbase and the documents stay inside a controlled environment. Eric Brown
What are the key points?
Terminology drift is inconsistent use of target terms for one source concept, and it is distinct from semantic change in a language.
The core controls are an approved bilingual termbase, termbase access for every linguist, terminology checks in review and final QA, and an audit trail.
ASTM F2575-2014 states that terminology management should apply throughout the entire project.
Each control needs a named owner, and the client needs a terminology owner of its own.
Control level follows consequence, reader, term status and data category.
Variation across long document series is more likely, so large sets need tighter control.
Approved variants belong in the termbase so that QA does not flag intended differences.
Vendor duties for access control, QA reporting and data protection belong in the statement of work.
FAQ
How do I measure terminology drift?
Count the translation variants used for each source term across a corpus. Research on terminology consistency measurement defines inconsistency as two or more variants for one source term. Automated QA against the termbase finds the same mismatches in a live project.
Who is responsible when a translated term changes meaning?
Responsibility follows the contract and the workflow. The client owns the termbase and approves terms. The vendor owns adherence to it and reports on QA. Writing both duties into the statement of work avoids disputes over who should have caught the change.
Does a translation memory prevent terminology drift?
A translation memory stores previously approved sentences and helps with consistency, but it does not govern terms on its own. A termbase defines the approved term for a concept. Both should be loaded in every project and maintained by an owner.
Is a term variation always an error?
No. The Fachsprache study notes that variation can be a deliberate and valuable distinction. Record approved synonyms and regional variants in the termbase so reviewers can tell intended variation from drift.
Sources

