Training and Competence 5 August 2026
Automation Anxiety: Will the Translation Profession Lose Its Work?
Will machines take the translator’s work away? We approach the question through historical evidence and the automation literature rather than through exaggerated claims.
The debate about translation, automation and unemployment is the oldest and most frequently repeated item on the profession’s agenda. Every time a new technology is announced, the question returns in the same form: will this profession disappear?
Contents
The question deserves to be taken seriously, but the answer has to rest on evidence rather than on exaggerated claims. Both the promises of technology companies and the pessimistic forecasts within the industry have been proved wrong many times over.
Below we look at what automation anxiety is, at the great misjudgements in its history, at what the economics of automation says, at which tasks in translation are becoming automated, and at what can be done in this picture.
What is automation anxiety?
Automation anxiety is the widespread fear that technology will replace human labour and do away with a profession. The idea is not specific to translation; it has surfaced in every major technological wave in industrial history.
In translation this anxiety differs from technology anxiety. Technology anxiety is about using the tool; automation anxiety is about the tool making the person redundant. The first has to do with learning, the second with the future of the profession.
Vieira (2018) stresses that automation anxiety in translation is not only anxiety about income but also concerns professional identity. Even someone who does not lose their work may feel that their role has been devalued.
Another dimension of the anxiety is uncertainty. When it is not known how much will change and over what period, planning becomes difficult. Uncertainty can be more wearing than a concrete threat.
Finally, the anxiety is not entirely groundless. The distribution of tasks is changing, and that change really does make some ways of working harder. The right response is not to dismiss the anxiety but to put it in proportion.
What does history teach us?
The history of machine translation is full of examples in which both excessive optimism and excessive pessimism turned out to be wrong. Weaver’s memorandum of 1949 is regarded as one of the texts that launched the field, and it raised great expectations.
The Georgetown experiment of 7 January 1954 carried that expectation to its peak. A limited demonstration of roughly 250 words created the impression that the translation problem would be solved within a few years. That is not how things turned out.
The ALPAC Report of 1966 then swung the other way. The report concluded that machine translation would be of no use in the near term, and funding in the field stopped for many years. Seen from today, that assessment also looks too categorical.
The Météo system, which went into operation in 1976, showed that automation really does work in a narrow domain. Weather bulletins have a limited vocabulary and a repetitive structure, and automation succeeds in those conditions.
The lesson history offers is this: automation depends on the structure of the task. Narrow, repetitive tasks with low uncertainty become automated, while broad, context-sensitive tasks that carry consequences resist.
The second lesson concerns timing. Expected changes generally arrive later than predicted but across a wider area than predicted. Short-term panic and lasting complacency are therefore both misleading.
What does the economics of automation say?
The economics literature offers an important correction to the automation debate. Autor (2015) argues that automation generally targets tasks rather than occupations. An occupation is made up of many tasks, and only some of them become automated.
The second point is complementarity. When some tasks are automated, the value of the remaining human tasks can rise. If the machine speeds up, checking and decision-making become more critical.
The third point is the demand effect. When costs fall, demand can rise. This effect has been observed in translation: content that was never translated before is now being translated. The global market being measured at 49.6 billion dollars on 2019 data (CSA) is one of the signs of that expansion.
The fourth point is the difference in pace. Automation is gradual rather than instantaneous. It takes time for organisations to change their workflows, for contracts to be renewed and for quality criteria to settle.
The fifth point is distribution. The effect of the change is not felt equally by everyone. Those working on repetitive, low value-added jobs are affected first; specialised fields are affected later and less. This inequality also explains why forecasts that look at averages are misleading.
Which tasks in translation are changing?
The table below breaks translation work down into tasks and assesses how open each task is to automation.
| Task | Openness to automation | Why |
|---|---|---|
| Producing raw translation | High | In formulaic texts machine output can reach a usable level |
| Repetitive content | High | Translation memory already covers most of it |
| Terminological consistency | Medium | Automated checking helps, but the decision stays with the person |
| Post-editing | Low | Telling fluent but wrong output apart requires subject knowledge |
| Defining scope and purpose | Very low | It calls for negotiation with the client and for accountability |
| Data security decisions | Very low | They have legal consequences and a person carries the responsibility |
The lower rows of the table show where the profession’s centre of gravity is moving. Value is shifting from producing text towards making decisions.
This shift is not new. A similar one occurred when translation memory became widespread; the way repetitive content was priced changed, but the profession did not disappear. At that time, too, the debate started out as one about lost work and gradually turned into a debate about pricing and scope.
Lost work or shifted tasks?
The data available do not support a wholesale loss of work. The signs that the composition of the work is changing are, on the other hand, clear. The proportion of those who have tried machine translation was measured at 55 per cent among freelance translators and 44.02 per cent among companies (DePalma 2012), which shows that the tool has entered the workflow.
There is a structural shift on the institutional side as well. The number of in-house translators at the European Commission has been reported to have fallen by roughly 50 per cent over twenty years (Strandvik 2017). That decline, however, has more to do with work being outsourced than with automation.
The distinction matters. A reduction in staff does not always mean a reduction in work; more often it means the work has moved. The high share of freelance working is consistent with this picture too (78.4 per cent, Pym et al. 2012).
A shift in tasks can, on the other hand, put pressure on income. Post-editing is priced differently from raw translation, and if the level of editing is not defined at the outset the translator may put in the same effort for a lower fee.
The real risk is therefore not unemployment but a lack of definition. When it is not clearly defined what the work is, its value cannot be priced correctly either.
The same lack of definition produces risk on the client side. When it is not known which step has been left to the machine and which to a person, the nature of the service being bought remains unclear. A transparent statement of work protects both sides and moves the discussion from price haggling to clarity of scope.
What can be done to adapt?
The first strategy is to take ownership of the tasks that do not become automated. Subject expertise, defining scope, evaluating output and decisions about data security make up this set.
The second strategy is to put the statement of work in writing. The level at which post-editing will be carried out, which errors fall outside the scope and how the fee is calculated should all be settled in advance.
The third strategy is to develop the skill of evaluation. Noticing output that is fluent but wrong is harder than correcting output that is obviously broken, and this skill develops together with subject knowledge.
The fourth strategy is to go deep in a narrow field. Automation advances faster in broad, general work. In texts that have regulatory consequences or carry high risk, human responsibility cannot easily be handed over.
The fifth strategy is to turn anxiety into knowledge. Following industry data regularly reduces uncertainty. Less uncertainty means that decisions are taken with a plan rather than out of anxiety.
The sixth strategy is to diversify sources of income. Working for a single client or on a single type of job magnifies the effect of a shift in tasks. Someone working across different fields can offset a contraction in one with another.
The Türkiye context
In Türkiye the automation debate is shaped by the characteristics of the language pair. Turkish has less parallel data than the major language pairs. That makes evaluating machine translation output more critical and increases the post-editing effort.
Sworn translation practice and notarised document translation also form a local buffer. In this work the responsibility is personal; the translator who signs is answerable for the accuracy of the text, and that responsibility cannot be handed over to a piece of software.
On the education side, adaptation depends on the pace of the curriculum. When translation and interpreting programmes and translation studies programmes include courses on machine translation evaluation and post-editing, graduates start out ready for the shift in tasks.
Finally, KVKK is a local factor that sets the limits of automation. Sending documents containing personal data to external engines depends on the contract and on the legislation. This article is for general information only and is not legal advice.
What does this mean for you?
If you are on the buying side, this debate reaches you as price and as risk. A supplier that uses automation may be faster and more affordable; but if you do not know which step has been automated, you do not know what you are buying either.
To secure that transparency, you can follow these steps:
- Ask openly whether machine translation is used, and get the answer in writing.
- If it is, define the level of post-editing: comprehensibility or publishability?
- For your confidential documents, set out in the contract which engine may be used.
- For high-risk texts, ask for an independent proofreading step.
- Ask to see, item by item, which steps the price covers.
These five steps turn automation into a benefit for you and remove the hidden risks. We have described which process we apply to which job on our services page, and you can find the price items on our pricing page.
Frequently asked questions
What is automation anxiety?
It is the widespread fear that technology will replace human labour and do away with the profession. In translation this anxiety strengthens at regular intervals as machine translation develops.
Has machine translation put translators out of work?
Industry data do not show a wholesale loss of work of that kind. The change observed runs towards a shift in the distribution of tasks: raw translation production is falling while the share of post-editing, terminology management and quality evaluation is rising.
Why does the ALPAC Report matter in this debate?
The ALPAC Report of 1966 lowered expectations of machine translation sharply and caused funding in the field to stop for years. It is an example showing that excessive expectation and excessive pessimism can both be wrong.
What should a translator do in this period?
It helps to take ownership not of the step that becomes automated but of the decision that does not: subject expertise, evaluating output, terminology management, data security and defining scope with the client.
Sources
- Autor, D. (2015). Why Are There Still So Many Jobs? Journal of Economic Perspectives, 29(3), pp. 3-30.
- Vieira, L. N. (2018). Automation Anxiety and Translators. Translation Studies, 13(1), pp. 1-21.
- Strandvik, I. (2017). Evaluation of Outsourced Translations. Berlin: Language Science Press.
- Pietrzak, P. and Kornacki, M. (2021). Using CAT Tools in Freelance Translation. London: Routledge Focus, Chapter 3.