Training and Competence 6 August 2026
Cognitive Friction: When Do Tools Get in the Way?
Tools save time, but they spend attention. We look at where cognitive friction comes from, how it erodes quality and how it can be reduced.
In the translation process, cognitive friction describes the way a tool or a working environment consumes the mental resource that ought to be reserved for the work itself. The concept comes from ergonomics research, and it brings a much needed balance to the debate about translation technology.
Contents
The productivity gain from tools is usually measured in word counts. Friction, because it is never measured, stays invisible. Yet the twenty minutes a tool saves can be taken straight back by the attention it scatters.
Below we look at how the concept is defined, where friction comes from, what part segmentation plays, how it affects productivity, how it can be observed and what can be done to reduce it.
What does the concept describe?
Translation is work that runs several operations at once: understanding the source text, generating options in the target language, weighing those options and keeping track of the whole. All of these operations draw on the same limited reserve of attention.
Friction arises when a tool takes a share of that reserve. Instead of translating, the user is working out what the tool is doing. Even when the time spent is short, gathering attention together again carries an extra cost.
Ergonomics research addresses this relationship directly. Slow software, interrupted workflows and badly designed interfaces are counted among the main sources of strain in the translation workplace (Ehrensberger-Dow and O’Brien 2015).
An important feature of the concept is that it moves the blame from the user to the system. A translator may be slowing down not because they handle the tool clumsily, but because the tool demands attention.
Friction never disappears altogether. Every tool asks for some attention. The aim is to bring friction down to an acceptable level and to know where it occurs. A known source of friction can be managed; an unknown one is experienced as a personal failing and damages morale for no reason.
Where friction comes from
The table below sets out the sources of friction most often met in a translation workflow, together with their effects.
| Source | How it appears | Effect on attention |
|---|---|---|
| Response lag | Waiting when moving between segments | The train of thought is broken |
| False quality warnings | Error reports that are not real errors | Warning blindness, real errors missed |
| Tag management | Moving formatting tags by hand | Constant switching between meaning and format |
| Interface clutter | Many panels and much information at once | The cost of visual search |
| Incompatible files | Transfer problems between different tools | A complete break from the work itself |
| Interruptions | Notifications, email, several projects at once | Time needed to refocus |
The last row matters particularly. After an interruption, attention takes time to return to its earlier level, and that time can be longer than the interruption itself. A five-second notification can cost several minutes of recovery, and across a day that cost accumulates.
The incompatibility item should not be underestimated either. In one study, around 90 per cent of translators reported running into file compatibility problems, and 70 per cent of those said the problem recurred often (Pietrzak and Kornacki 2021). A frequently recurring hitch consumes more attention than a single large problem, because it has to be solved again every time.
Segmentation and the integrity of the text
The most insidious form of friction is the way a tool changes the unit of work. Translation tools split the text into sentences, and the user comes to treat that unit as natural.
The result is that the text as a whole drops out of view. The flow of the paragraph, the function of its connectives and the rhythm of its repetitions are invisible in the segment window. Mossop (2006) calls this collage translation.
The effect leaves a measurable trace. Sentence merging has been observed to fall by roughly half once tools are in use (Ehrensberger-Dow and Massey 2014). In other words, translators are not joining the sentences that ought to be joined.
This matters particularly for Turkish. The difference in sentence length between Turkish and English is marked, and decisions about splitting and joining are needed often. Staying faithful to segment boundaries produces an artificial rhythm in the target language.
The answer is not to abandon the tool. Carrying out the final review outside the tool, on the whole text, before delivery undoes most of this friction.
A second measure is to use segment merging and splitting deliberately. Most tools allow both, yet leaving the default segmentation untouched has become a habit. Noticing the habit is the first step towards loosening the boundary the tool imposes.
Where does the productivity gain go?
The productivity promise of tools rests on measurable ground. Industry measurements show translators producing an average of 2,684 words a day, with the benchmark taken to be around 2,500 words (DePalma 2012).
The same measurements report hourly output as 443 words for freelance translators and 5,728 words for translation companies (DePalma 2012). The wide gap comes from work being run in parallel inside companies and from the infrastructure being ready in advance.
This second figure is telling for the friction argument. An individual translator solves technical problems personally; in a company that burden has been handed to a separate role. The most effective way to reduce friction is to share it out.
Part of the gain also comes from reuse. The proportion of entirely new words has been measured at around 59 per cent; the rest is reusable in one form or another (DePalma 2012). That gain is real, but no net gain can be calculated without taking friction into account.
The right question, then, is not whether to use tools, but which tool configuration produces a net gain on which job.
How is friction noticed?
Because friction is not measured, it usually goes unnoticed. Even so, everyday work offers a few reliable signs, and watching for them takes no special measuring equipment.
The first sign is a deviation in time. If two texts of the same difficulty take unexpectedly different amounts of time, the difference usually comes from the environment rather than from the text. When a time estimate does not hold, the reason should be questioned.
The second sign is the kind of tiredness. Fatigue that comes from the difficulty of the text feels different from fatigue that comes from wrestling with a tool. With the second kind, the word count at the end of the day is low while the exhaustion is high.
The third sign is the kind of error. Terminology errors point to gaps in knowledge; inconsistencies and omitted sentences point to lost attention. Counting correction requests by type makes this distinction visible.
The fourth sign is putting things off. If certain file types are always left until the end of the day, that file type is producing high friction. The observation shows directly which step needs improving.
Read together, these four signs make it clear enough where friction is concentrated. No special software is needed to measure it; a simple log kept for a few weeks is usually enough, and it puts the decision to improve something on an evidential footing.
How can friction be reduced?
The first measure is to narrow the rule set. If automated quality checks are tuned to the text type, the number of false warnings falls and warning blindness does not set in.
The second measure is to separate preparation from translation. If file conversion, tag cleaning and memory setup are done in a separate session, the need to switch context during translation drops.
The third measure is uninterrupted blocks. Periods with notifications switched off and given over to a single job remove the cost of refocusing. Moving between several projects in the same day incurs that cost again at every switch; grouping projects together helps wherever it is possible.
The fourth measure is to record recurring problems. Once the solution to an error has been written down, the second encounter produces no friction. Shared across a team, that record turns into organisational knowledge.
The fifth measure is a final review outside the tool. Reading the whole text through fluently is the simplest and most effective step for offsetting the fragmentation that segmentation creates. Doing that reading the day before delivery also gives the distance needed to look at the text afresh.
The Türkiye context
In Türkiye the most widespread source of friction is the quality of the source files. Official documents often circulate as scanned PDFs; because the text exists as an image, it first has to be recognised and then have its formatting rebuilt. Both steps consume attention before the translation has even begun.
The structure of Turkish forms a second source. Because the language is agglutinative, automated checking tools can produce more false warnings when matching terminology. Seeing false warnings constantly leads to the genuine ones being ignored as well.
The third source is the shifting of sentence boundaries. Turkish requires decisions about splitting and joining often; taking those decisions while working in a segment window demands more attention than it does for structurally close pairs such as English and French.
The fourth source is the lack of support. Turkish technical resources and community support are not as broad as those for the major languages, so a translator who runs into a problem spends extra time looking for a solution. Professional associations and peer groups close part of this gap.
What does this mean for you?
Friction reaches your invoice in two ways: longer delivery times and errors of coherence that slip through. The second is the more costly of the two, because it is hard to spot.
You can find out how your supplier manages friction through the following steps:
- Ask whether they want your source file in an editable format; that reduces the preparation burden from the outset.
- Confirm that the whole text is read outside the tool before delivery.
- Make sure the proofreading is done by a different person.
- On urgent jobs, ask openly which step is being shortened.
- Find out how consistency between sections is maintained in long texts.
These five steps make the balance between speed and coherence visible. We have set out the steps in our process on the services page; you will find how delivery times and prices are calculated on the pricing page.
Frequently asked questions
What is cognitive friction?
It is a tool or a working environment consuming the mental resource that ought to be reserved for the work itself. Slow software, warnings that cannot be understood and interrupted workflows are the main sources of this friction.
How does friction affect translation quality?
It does not produce errors directly; it reduces the capacity for attention. Reduced capacity first hits the operation that demands the most attention, namely keeping track of the text as a whole and reviewing decisions.
Is segmentation a source of friction?
Partly. Working segment by segment makes it harder to see the text as a whole. Research has measured sentence merging falling by roughly half once tools are in use.
How can friction be reduced?
Uninterrupted blocks of work, narrowing the rule set to the text type, a final review outside the tool before delivery and a written record of the solutions to recurring technical problems are the most practical measures.
Sources
- Ehrensberger-Dow, M. and O’Brien, S. (2015). Ergonomics of the Translation Workplace. Translation Spaces, 4(1), pp. 98-118.
- O’Brien, S. et al. (2017). Irritating CAT Tool Features that Matter to Translators. Hermes, 56, pp. 145-162.
- Mossop, B. (2006). Has Computerization Changed Translation? Meta, 51(4), pp. 787-793.
- DePalma, D. (2012). Translation Productivity Benchmarks. Common Sense Advisory report.