CAT Tool Features: Which Ones Do Translators Find Most Valuable?

The sentence “we use a CAT tool” does not describe a single thing. These programs are several separate modules packaged together, and each of those modules solves a different problem.

Contents

For the buyer of translation the real question is not which program the supplier uses but which CAT tool features they actually put to work. A program opened without a memory set up is no different from a word processor.

Below we separate the modules one by one and set out when each of them produces value. We also state plainly what the case study we have to hand does and does not say on the question.

CAT tool features: the core components

A professional translation tool contains five main components that can work independently of one another. Telling them apart is what lets you ask the right question when you are assessing a quotation.

Component	What does it do?	When does it produce value?
Translation memory	Stores translated sentences and suggests them	On repetitive and regularly updated documents
Termbase	Enforces approved terms	In corporate language and brand consistency
Quality control	Scans for formal errors	In texts dense with numbers, dates and tags
Concordance search	Searches past usage in the memory	On long-running projects
File conversion	Turns different formats into translatable ones	On software, web and publishing files

The third column of the table shows where the real argument lies. No component is valuable on every project; the value depends on the structure of the document.

There is therefore no single right answer to the question “which feature is the most valuable”. The right question is this: which component makes a measurable difference on my document type?

A sixth component can be added to the list: the machine translation connection. It is presented as a convenience, but on confidential documents it is a decision in its own right and it has to be possible to switch it off.

There is a set-up threshold as well. Most of the components do not produce value straight out of the box; the memory has to be fed, the term list has to be approved and the quality control rules have to be tuned to the project. That effort usually concentrates in the first week of the project.

Translation memory: the most visible return

The memory is the founding function of these tools and it stands at the centre of the argument. Its logic is simple: every confirmed sentence is stored together with its source and target language, and when a similar sentence comes along it is offered as a suggestion.

Its return depends directly on the repetition rate. In industry measurements the proportion of entirely new words in the texts handled by freelance translators was reported at 59.11%; in other words, roughly 40% of the text was of a reusable nature (DePalma 2012).

That 40% explains why the memory is talked about so much. But the same figure also shows the limit: most of the text is still translated from scratch. The memory is an accelerator, not a magic solution.

The memory has a second benefit, and it is independent of speed: consistency. Having the same sentence translated the same way in every version reduces the risk of error, particularly in documents such as user manuals and contracts. That benefit persists in part even when the repetition rate is low.

There is a risk side too. High-match suggestions can be wrong once the context changes. The tendency to accept a suggestion without questioning it is called “blind faith in the translation memory” in the literature (Bowker 2005). The memory therefore does not do away with the revision process.

To this is added a question of accumulation: whose memory is it? The memory produced on your project is an asset that will serve you in later years when you move to another supplier. When it is not handed over, every change of supplier means starting from scratch.

The termbase: the silent backbone

For the corporate buyer the decisive component is more often than not the termbase. Unlike the memory, it works at the level of words and terms rather than sentences.

Its effect is concrete. Having a product name, a module heading or a legal concept appear with the same equivalent in every document bears directly on many areas, from user experience to compliance auditing. The terminology stays fixed even when different translators work on the material.

One further advantage of the termbase is that it works on projects with no repetition. Even when the document is not repetitive, the terms repeat. This component can therefore produce value even on jobs with a low repetition rate.

The termbase is not free, however: it has to be built, approved and kept up to date. That effort usually concentrates at the start of the project and calls for a contribution from the buyer.

The form of that contribution matters as well. The most useful term lists are the ones the translation side proposes and the buyer approves. Lists drawn up one-sidedly either fail to fit the corporate language or are ignored in practice.

Quality control: what does it catch, what does it miss?

The third component is automated quality control. These modules scan the text and flag particular patterns of error.

What they catch is clear: mismatches in numbers between source and target, missing or broken formatting tags, segments left empty, usages that go against the approved term list, and words typed twice.

What they cannot catch is just as clear: shifts in meaning, misreadings of context, mismatches of register and problems of readability. These modules carry out a formal check, not a grammatical one.

The practical upshot is this: automated quality control does not take the place of human revision, but it does let the reviser spend their time on the work that matters. On documents dense with numbers and dates it is a serious reducer of risk on its own.

These modules have a side effect as well: an excess of false alarms. If the rules are not tuned to the project, the report runs to hundreds of lines and the real errors are lost in the noise. A tuned quality control is far more valuable than an untuned one.

Whether the report is shared is an indicator in itself. A supplier who can hand over the quality control report along with the delivery is genuinely running the process; if no report can be produced, the module has in all likelihood never been switched on.

Does the research give a ranking of value?

Here we have to state a limit plainly. The verified figures we report from the 2019 Polish case study on which this series rests are confined to particular subjects.

They are the distribution of tools (SDL Trados 43.3%, memoQ 29.1%, Wordfast 10.6%), licence ownership (91%), compatibility problems (90%, and 70% of those frequently) and the difference in source of livelihood (75% against 41%). All of them carry the label Poland, 2019.

We do not report a numerical ranking of value between the modules, because presenting a ranking we cannot verify in the form of percentages would mislead the reader. The most common problem in industry writing is that percentages with no source turn into fact through sheer repetition; in an area that has not been measured, speaking qualitatively is more useful than supplying an invented figure. The assessments in this section are qualitative judgements resting on industry practice and on the working logic of the modules.

The sample, moreover, consists of 141 people, rests on voluntary participation and belongs to a period before artificial intelligence. We unpacked all of these constraints in our article on the limits of the research.

The price of the features: complexity and compatibility

Every module brings a burden as well as a benefit. The breadth of function in these tools can work against ease of use.

In the same research roughly 25% of those not using a tool gave complexity as their reason (Poland, 2019). The literature calls this cognitive friction: the tool drawing attention away from the text and towards its own interface can lower net productivity (Ehrensberger-Dow & O’Brien 2015).

The second burden is compatibility. Of the participants, 90% said they had run into file compatibility problems, and 70% of those said they ran into them frequently (Poland, 2019). The benefit of the modules therefore depends on the file opening properly at the other end. The detail is in our article on compatibility.

These two burdens weaken the assumption that “more features are better”. Two well-built modules produce more value than five half-built ones.

The practical test is this: a component is valuable if it repays the effort of setting it up within the first three months. If it does not, then not using that component on that project is a valid decision too. You are not obliged to switch on every part of the tool on every project.

The Türkiye context

There is no publicly available study measuring how intensively these modules are used in Türkiye. We give no comparative figure; instead we set out the technical points specific to Turkish.

The termbase works less easily in Turkish than you might expect. Because the language is agglutinative, the same term appears in dozens of inflected forms: “fatura”, “faturayı”, “faturanın”, “faturalandırma”. For term checking to work, the termbase has to be built so that it recognises the root and its inflected variants; otherwise the module either stays silent or produces false alarms.

The picture is similar on the translation memory side. The free word order of Turkish means that two sentences identical in meaning can differ in surface form, and fuzzy match rates can come out lower than expected. On Turkish projects it is therefore safer to assess the repetition analysis not by looking at the percentage alone but by trying it out on a sample file.

A further point is that Turkish is more limited in digital resources than the major languages. Ready-made termbases and general-purpose dictionaries are fewer, and that means terminology work calls for more human effort on Turkish projects.

What does this mean for you?

When you are assessing a supplier, ask about the infrastructure they have set up rather than the name of the program. The four questions below reveal whether the tool is genuinely being used.

Is a translation memory opened specifically for us, or does the work run in a general pool?
Who draws up the term list, and is it put to us for approval?
Is an automated quality control report produced before delivery, and can it be shown to us?
Are the memory and term files handed over in standard formats on request?

A process that answers yes to all four produces a cumulative benefit for your brand language over the years. We have described how we work on our services page, and you can reach all of the related subjects through the knowledge centre.

Frequently asked questions

What is the most basic function of a CAT tool?

The translation memory. It stores sentences that have already been translated and suggests them when a similar sentence comes along. Its return is directly proportional to the repetition rate of the text.

Are a termbase and a translation memory the same thing?

They are not. A translation memory works at sentence level, a termbase at the level of words and terms. What decides how well a text fits the corporate language is usually the termbase.

Did the research rank the features in order of value?

The verified data we use in this article is confined to the distribution of tools, licence ownership and compatibility problems. We do not report a numerical ranking of value between the modules.

Does the quality control module take the place of human checking?

It does not. These modules catch formal errors such as inconsistencies in numbers, dates, tags and terms. They cannot catch errors of meaning or problems of register; human revision is necessary.

References

Pietrzak, P. & Kornacki, M. (2021). Using CAT Tools in Freelance Translation. Routledge Focus, Table 6.2 and Chapter 6.
Bowker, L. (2005). Productivity vs quality? A pilot study on the impact of translation memory systems. Localisation Focus, 4(1), pp. 13-20.
Ehrensberger-Dow, M. & O’Brien, S. (2015). Ergonomics of the translation workplace. Translation Spaces, 4(1), pp. 98-118.