Translation Productivity: The Difference Between an Agency and a Freelance Translator

Translation productivity is the measure most often misread when quotations are compared. When an agency is said to be able to produce thousands of words an hour, this is taken to be comparable with the speed of a single translator. In fact the two figures measure different things.

Contents

Understanding the difference is not merely a matter of academic curiosity. It is directly useful when you are setting a delivery schedule, planning a budget and defining your quality expectations.

In this article we look at how productivity is measured, what the published figures mean, what really determines them and how you, as a buyer, can take a share in the productivity gain.

How is translation productivity measured?

The commonest measure is the number of words produced in a unit of time. Simple as it looks, this measure quickly becomes meaningless when it is not clear what is being counted. Is only the first draft counted, or are terminology research, revision and formatting included as well?

The second problem is the nature of the text. The same translator may produce twice as much a day on a familiar subject, yet drop to half speed in a new field dense with terminology. A single average is therefore a weak instrument for estimating how long one particular job will take.

The third problem is the unit of measurement itself. Word counts are not comparable across languages; the same content is expressed with a different number of words in different languages. Character-based measures solve part of this problem but do not remove it entirely.

The fourth problem is that productivity means little on its own. A text produced quickly but rewritten during revision has in effect been translated twice. Serious measurements therefore count not output but accepted output. A productivity figure should not be interpreted without seeing the acceptance rate after revision.

What do the figures say?

Industry measurements have put translators’ average daily output at 2,684 words, while the widely accepted industry benchmark is 2,500 words (DePalma 2012). This is a realistic average covering a full working day.

The same measurement calculated hourly output at 443 words for freelance translators and 5,728 words for language service providers. The gap is roughly thirteenfold and looks surprising at first sight.

The reason for the surprise is a false assumption. The provider figure is not one person’s speed but the total output an organisation produces in an hour. If ten translators are working at the same time, the total naturally rises tenfold. The two figures can be set side by side, but neither can be read as a performance measure for the other.

Measure	Value	What it shows
Average daily output	2,684 words	Individual capacity
Industry benchmark	2,500 words/day	Planning baseline
Freelance translator, hourly	443 words	Individual speed
Provider, hourly	5,728 words	Total organisational output
New word rate	around 59%	Share of reuse

The real determinant: the repetition rate of the text

The variable with the greatest effect on productivity is not the translator’s speed but the text itself. Measurements put the proportion of entirely new words at 59.38% in companies and 59.11% among freelance translators (DePalma 2012). In other words, roughly 40% of an average job is reusable content.

This proportion varies widely by text type. In a literary text repetition is almost nil. In a product catalogue, a technical manual or a family of contracts it can be far higher. The same agency, with the same translator, shows completely different productivity on different texts.

For the buyer the conclusion is clear: bargaining over productivity is done on the structure of your own text, not on the provider’s speed. In a family of documents with a high repetition rate the potential saving is large, and that potential is realised only if a memory is kept. You can see how tiered pricing works on the pricing page.

There is a source-text side to raising the repetition rate as well. Writing the same warning sentence differently in every document lowers the repetition rate and raises cost. In organisations with template and terminology standards, translation cost falls of its own accord over the years, without any price bargaining at all.

Where does the tool’s contribution end?

Computer-assisted translation tools raise productivity in two ways: they bring up previously translated sections ready-made, and they check terminological consistency automatically. Both are real gains, but they have limits.

The first limit is the quality of the memory. A memory fed with poor translation multiplies the error for years. In the literature this risk has been called “blind faith in the translation memory” (Bowker 2005). When it is not cleaned regularly, a memory can stop being a productivity tool and become a source of error.

The second limit is the file format. Tools speed things up with clean, editable files; with scanned PDFs or badly formatted documents the preparation stage swallows the gain. In one case study, 90% of translators stated that they had experienced compatibility problems and 70% that they experienced them often (Pietrzak & Kornacki 2021).

The third limit is text type. In creative texts, advertising slogans and literary translation the gain from tools is close to nil; in these texts time goes not on producing sentences but on making decisions. When productivity expectations are not adjusted to text type, the delivery date rests on an assumption that was unrealistic from the outset.

Does productivity affect quality?

It would not be right to say that gains in productivity come free. Tools divide the text into segments and the translator mostly works sentence by sentence. This raises speed but makes it harder to see the text as a whole.

Research has shown that in segment-based work translators markedly reduce stylistic interventions such as merging sentences (Ehrensberger-Dow & Massey 2014). The result is a translation that stays too close to the sentence structure of the source text. In the literature this situation is also called “collage translation” (Mossop 2006).

The risk can be managed. The solution is not to abandon the tool but to keep a separate step in which the whole text is read for fluency before delivery. This step gives back a small part of the productivity gain; in return it raises readability markedly.

How does the buyer share in the productivity gain?

A rise in productivity does not pass to the buyer automatically. If the provider keeps the memory to itself, does not share the analysis report and quotes a single flat rate, the gain stays entirely with the provider.

The way to obtain a share is transparency. Asking for the analysis report, seeing the price broken down by match tier and writing into the contract that the memory belongs to you: these three steps make sure the gain is shared.

One caution should be added: overstating the repetition discount is not healthy. Segments with high matches also have to be read and checked for fitness in context. Treating that effort as zero comes back later as inconsistent texts. A fair tier structure does not wipe the effort out; it reduces it.

How do productivity figures shape the delivery schedule?

The most practical use of productivity data is in building a schedule. The benchmark of 2,500 words a day is a reasonable starting baseline for a single-translator job (DePalma 2012). To that baseline you add time for formatting, terminology research and final review.

When the volume exceeds one translator’s capacity, the job is split. Splitting raises speed but creates a risk to stylistic and terminological consistency. To meet that risk a single reviser has to read the whole text; this step should be written into the schedule as a separate day.

The third variable is preparation time. Scanned documents, complex tables and embedded graphics require serious preparation before translation begins. This time is invisible in most quotations but it is the item that most often pushes the schedule out. A realistic plan sets aside a noticeable part of the total time for preparation and final review.

The Türkiye context

In Türkiye pricing is traditionally based on characters rather than words. The common unit is a page of 1,800 characters including spaces, which corresponds to roughly 240 words (Pietrzak & Kornacki 2021, p. 83). Without making this conversion, international quotations cannot be soundly compared with local ones.

The structure of Turkish also affects measurement. Because it is an agglutinative language, a structure expressed with several words in English can come down to a single word in Turkish. A fall in the word count in the target language is not a gain in productivity; it merely shows that the unit of measurement is language-dependent. Quotations should therefore be based on the source text.

A third local factor is that Turkish is a relatively low-resource language in language technology. Ready-made memory and terminology resources are more limited than for pairs such as English-German. That makes building up your own memories all the more valuable in Turkish.

What does this mean for you?

Productivity is not a feature to look for when choosing a provider but a result to be built together. The biggest gain comes from putting your own document structure in order.

1. When asking for a quotation, request the analysis report and the match breakdown; that is how you see your own repetition rate.
2. Send source files in editable form; a scanned PDF creates preparation cost.
3. Standardise your document templates; as the repetition rate rises, unit cost falls.
4. Write into the contract that the translation memory and the term list belong to you.
5. Keep the pre-delivery read-through for coherence as a separate item; do not economise on this step in the name of productivity.

If you do not know the repetition rate of the family of documents you hold, measuring it takes only a few minutes. Share sample files through our quotation page and you can see the analysis report and the tiered price breakdown together.

Frequently asked questions

How many words does a translator translate in a day?

Industry measurements put average daily output at 2,684 words and the widely accepted benchmark at 2,500 words (DePalma 2012). The difficulty of the text and the formatting load change this number markedly.

Why is the hourly output of agencies so much higher?

Because that figure is not individual speed but the total of several translators working in parallel. The measurement found 443 words per hour for freelance translators and 5,728 for language service providers (DePalma 2012).

What raises productivity most?

The repetition rate of the text. In a typical job roughly 40% of the content consists of previously translated or repeating sections; taking those sections into the memory delivers the biggest gain.

Does high productivity lower quality?

Not of itself, but it creates certain risks. Working segment by segment can mean the text as a whole is overlooked; for that reason a pre-delivery read-through for coherence should be kept as a separate step.

Sources

• Pietrzak, P. & Kornacki, M. (2021). Using CAT Tools in Freelance Translation: Insights from a Case Study. Routledge Focus, pp. 83–95.
• DePalma, D. (2012). Translation Productivity Benchmarks. Common Sense Advisory.
• Bowker, L. (2005). “Productivity vs Quality? A pilot study on the impact of translation memory systems”. Localisation Focus, 4(1).
• Ehrensberger-Dow, M. & Massey, G. (2014). “Cognitive ergonomic issues in professional translation”. The Development of Translation Competence.
• Mossop, B. (2006). “Has computerization changed translation?”. Meta, 51(4).