Translation Technology 5 August 2026
The Translation Memory Ownership Problem: Who Owns the Memory?
Who owns the translation memory that builds up on a project? The translator, the agency or you? If the answer is not in the contract, it stays unsettled in practice.
When a translation project ends, the delivered file is not all that is left behind. The translation memory built up on that project is left behind too: source and target sentences in aligned form, term records, corrected wordings. This accumulation is a real asset, and it determines what you will pay in future.
Contents
Even so, the question of translation memory ownership often does not appear in contracts at all. Nobody asks and nobody writes it down; the memory stays, in practice, with whichever party is holding it. The problem surfaces when the relationship ends or when a data request arrives.
This article takes up whose labour the memory is built from, what the contract should say, how the personal data inside it is assessed, the terms of use of free services and a workable solution.
Why is the translation memory ownership question never asked?
The first reason is invisibility. The memory is not delivered to the client as a file; it lives inside the tool, on the translator’s or the agency’s computer. Ownership of an asset that nobody sees naturally never comes up.
The second reason is the vagueness of the legal framework. The memory is neither a work in the classical sense nor an ordinary heap of data. Inside it, fragments coming from the source text, wordings produced by the translator and terms set by the client all sit together. All three parties have contributed.
The third reason is habit. The established practice in the sector is for the memory to stay with the service provider. This need not be a practice in bad faith; more often it comes from nobody having thought otherwise.
The fourth reason is that the dependence suits somebody. The party holding the memory has an advantage on price and turnaround in the next project. For the client, changing supplier means building the accumulation up again from scratch. This is a quiet but effective mechanism of attachment.
Whose labour is the memory built from?
Three parties contribute to the content of a memory. The client supplies the source text, the domain knowledge, the term preferences and the correction feedback. The translator produces the target text. The agency sets the process up, maintains it and checks the records.
This threefold contribution makes a one-sided ownership claim difficult. A practical hierarchy can be set up nonetheless: the source text and the terminology belong to the client; the target text is transferred to the client by the translation contract; and management of the memory as a file rests with the service provider.
In practice the balanced solution is this: the memory is treated as belonging to the client, while the service provider holds the right to use it for the duration of the project and within the limits defined in the contract. That way the accumulation is preserved and the dependence disappears.
The translator’s side should not be overlooked either. According to the case study in the book, 91 per cent of translators hold their own tool licence (2021 data); that is, the memory is most often built up on the translator’s own installation. Memory transfer therefore has to be regulated separately in subcontractor agreements as well.
What should the contract say?
What removes the uncertainty is not long legal text but a few clear sentences. The table below summarises the scenarios that come up most often and the consequences they produce.
| Scenario | Who actually holds the memory? | Risk for the client | Suggested clause |
|---|---|---|---|
| No clause in the contract | The service provider | Loss of the accumulation on changing supplier | An ownership and delivery condition should be added |
| Written that the memory belongs to the client | The client | Low | The delivery format and frequency should be stated |
| Shared use of a memory | Held jointly | Register and confidentiality bleeding together | A memory kept separate for each client |
| Transfer to a subcontractor | The last link in the chain | A break in the confidentiality chain | A ban on transfer and an undertaking to destroy |
The minimum elements that have to go into the contract are clear: ownership of the memory, the format in which it will be delivered on request, how long it will be retained, who it may be shared with, and whether it will be destroyed once the work is finished.
These clauses need not sit in the body of the contract; a short annex or a service description will do just as well. What matters is not length but making sure the parties understand the same thing. A one-sentence ownership clause heads off long arguments when the relationship ends.
The delivery format matters as well. Delivering the memory in a tool-independent interchange format is what makes the accumulation genuinely portable. A memory delivered only in a proprietary format is yours on paper but cannot be used in practice.
Personal data inside the memory
A translation memory is not only language data. Party names in contracts, identity details in powers of attorney, diagnostic wordings in medical reports and contact details in correspondence are all written into the memory as segments. That turns the memory into a recording medium containing personal data.
Tool makers are aware of this. As reported in the book, memoQ shows the user a warning dialogue recalling their data protection obligations (Pietrzak & Kornacki 2021, p. 25). The warning states that responsibility lies with the user; a technical measure on its own, in other words, is not enough.
What is distinctive about the risk is that the data goes on living after the project has ended. Even if the delivery file is deleted, the memory record stays, and years later it can come up on screen as a suggestion in another client’s project. This is the quietest form of data leakage.
The risk is particularly marked in document types such as powers of attorney, medical reports, court judgments and employment contracts. In these documents the text itself is personal; stripping the data out is not confined to a few fields but concerns every segment.
The workable measure is to use a separate memory for documents dense with personal data and to destroy that memory within a defined period at the end of the project. Cleaning proper names and identity numbers out of the memory is, in addition, a maintenance operation that is technically possible.
The terms of use of free services
The ownership debate takes its sharpest form with free online translation services. These services cost nothing, but what that costs you is written into the terms of use.
As Novović reports, Google’s 2016 terms of service granted the company a worldwide licence to use the content a user uploaded (Novović 2017, p. 202). Clauses of this kind are held to be necessary for the service to work; they do not, however, offer an acceptable framework for a confidential document.
Today the data processing terms in enterprise and paid versions are drawn up differently. What is decisive is not the name of the service but which version is being used under which agreement. The difference between the free version and the enterprise version is decisive where confidentiality is concerned.
A serious service provider therefore states in writing which engines are used on confidential documents and whether the data leaves the premises. In a chain that cannot answer, an undertaking of confidentiality is worth nothing in practice.
The other assets alongside the memory
The ownership debate is not confined to the translation memory. Three further assets build up on a project and carry value: the termbase, the style guide and the decision records. All three matter as much as the memory, and all three are usually absent from contracts.
The termbase records which renderings you use for your organisation’s concepts. Product names, job titles, technical concepts and the renderings you do not want all sit here. Unlike the memory, it is a directly readable list, and it changes hands easily.
The style guide defines forms of address, how dates and numbers are written, units of measurement and how brand wordings are to be used. This document is often never written at all; the decisions stay in people’s memories and are lost when the team changes.
Decision records are the asset most easily forgotten. If the reasoning behind a translation choice that was debated and settled on a project is not put on record, the same debate is held again six months later. Even a short decision log heads off that repetition.
Ownership and delivery conditions for all three of these assets should be regulated in the same clause as the memory. A client who asks only for the memory has left behind the readable and transferable part of the accumulation.
The Türkiye context
In Türkiye the legal framework for the subject is set by Personal Data Protection Law no. 6698. In a translation service the client is usually the data controller and the translation provider the data processor. The distinction matters, because it is the data controller who bears the duty to inform and who is the addressee of applications by data subjects.
Under the Law, a written contract with the data processor is expected, and the data is expected to be kept limited to the purpose of processing. Within that framework the translation memory is often an overlooked recording medium; if the contract speaks only of the delivery file, the memory is left unregulated.
Transfer abroad is a separate heading. Cloud-based tools and online translation engines can mean the data travelling to servers outside the country. For documents that call for sworn translation and notarisation, the confidentiality obligation also carries a dimension of professional liability. This article is for general information and is not legal advice.
What does this mean for you?
For the party buying translation, the subject matters for two reasons: not losing what you have accumulated, and knowing where your data sits. Both are settled by a few lines added to the contract.
- Have it written into the contract that the memory and the termbase belong to you.
- State that on request they will be delivered in a tool-independent interchange format.
- Define the retention period and the destruction conditions at the end of the work in days.
- Add whether transfer to a subcontractor is allowed and, if it is, that the same obligations apply.
- Ask for a separate memory to be used on documents dense with personal data.
You can find the details of how your memory is stored and of the delivery conditions on the translation services page, and you can send us your internal confidentiality requirements through our contact page.
Frequently asked questions
Who does a translation memory belong to?
There is no single answer in law; what is decisive is the contract. Where the contract has no clause on the point, the memory stays in practice with whichever party is holding it. Ownership and delivery conditions should therefore be settled in writing at the start of the work.
If I ask for my memory, what format do I get it in?
Common practice is for the memory to be delivered in TMX format. This format is tool-independent, and it lets you carry the content over if you decide to work with another supplier.
Does a translation memory contain personal data?
It can. Party names in contracts, identity and contact details and wordings from medical reports are written into the memory as segments. The memory should therefore be treated as a recording medium containing personal data.
Can free translation services use my document?
It depends on the terms of use. The terms of some free services grant broad rights to use the content uploaded. These services should not be used on confidential documents, and written confirmation that they were not used should be obtained.
Sources
- Pietrzak, P. & Kornacki, M. (2021). Using CAT Tools in Freelance Translation: Insights from a Case Study. London: Routledge Focus, p. 25 and pp. 71–95.
- Novović, M. (2017). Google Translate and data protection. Journal of Intellectual Property Law and Practice, pp. 199–205.
- Personal Data Protection Law no. 6698 (2016). Resmî Gazete, 7 April 2016, No. 29677.
- LISA (2005). TMX 1.4b Specification. Localization Industry Standards Association.