Translation Technology 5 September 2026
Translation Memory Match Rates: Exact, Fuzzy and Context
The source of the discount in your quotation is how your document is spread across the match bands. Knowing how to read that distribution is the most practical way to manage a translation budget.
If you see expressions such as “100% match”, “fuzzy match” or “repetition” in a translation quotation, the agency you are dealing with works on a memory basis. Translation memory match rates are technical measures that directly determine the amount you will pay.
Contents
Even so, what these rates mean is often unknown on the buyer’s side. The result is a loss of two kinds: either the discount you are entitled to is never asked for, or you fail to notice which quality risk has arisen in return for the discount granted.
This article explains the match types one by one and shows how an analysis report is to be read. By the end you will be in a position to look at your own quotation and understand what is being charged and at what level.
What is a translation memory match rate and how is it calculated?
Once the translation tool has divided the document into sentences (segments), it searches the memory for each segment. It displays the closest record it finds, expressing the similarity between the two as a percentage. That percentage is called the match rate.
The calculation rests on a distance algorithm that measures the difference between two texts. How many words have changed, how many have been added or deleted, and how far the word order has been disturbed all enter into the reckoning. The figure measures the formal proximity of the sentence, not its meaning; that distinction will matter in the sections that follow.
The calculation method is not exactly the same from one tool to another. When you analyse the same document in two different translation environments, the band distributions can come out slightly differently. When comparing quotations from different agencies, therefore, you need to look not only at the percentages but also at which tool and which memory the analysis was run with.
The economic meaning of the rates, however, is shared: the higher the percentage, the less time the translator will spend, and so the lower the charge for that segment. Reasonable though this logic is, it is not mechanical; whether the time really does fall depends on the language pair and the text type.
The exact match (100%) and why it is not enough on its own
An exact match is the case in which the sentence is found in the memory word for word. The character string in the source text and the record in the memory are the same. At first glance this looks like a segment that needs no work at all.
In reality the situation is a little different. The same sentence may have to be translated differently in a different context. The word “Save” means “Kaydet” in a software menu and “tasarruf” in an insurance brochure. The memory cannot tell the two apart; the one who can is the human being.
This is not a technical detail but a measured risk. Bowker (2005) points out that translators can develop “blind faith” in the translation memory, accepting high-rated suggestions without testing them in the new context. Treating exact matches as entirely free of charge amounts to accepting that this check will never be carried out.
The balanced practice is to charge for exact matches at a very low rate but not to zero them out altogether. That way the checking step stays in the workflow.
Context match: stronger than 100%
A context match, called “context match”, “ICE match” or “101%” depending on the tool, is one step above the exact match. Here it is not only the sentence itself but also the sentence before it and the sentence after it that are identical to what the memory holds.
This distinction largely resolves the “Save” problem described above. If the sentence arrives with the same neighbours, it is all but certain that it comes from the same part of the same document. The likelihood of a shift in meaning falls.
In practice, context matches occur in abundance in updated versions of a document. In the new edition of a user manual, the sections that have not changed typically give context matches. The highest discount can be applied to these segments, because both the text and its surroundings have been verified.
The proportion of context matches is at the same time a sign of how well you manage your documents. If source documents are kept in order from version to version, this band widens. If the document is reformatted with every version, if paragraphs change places, or if the text is rewritten in a different program, the context chain breaks and the segments drop to exact match, or even into the fuzzy band. In other words, part of what determines your discount is not your supplier but your own document discipline.
Fuzzy match bands and the threshold setting
A fuzzy match is the case in which the sentence is similar but not the same. Tools divide this range into bands and report each band separately. The boundaries of the bands and the charging that goes with them are not a fixed industry rule; they vary with the tool setting and with the agreement you have made with your supplier.
| Match type | Meaning | The translator’s work |
|---|---|---|
| Context match | Sentence and its neighbours are the same in the memory | Quick verification |
| Exact match (100%) | Sentence the same word for word, context may differ | Context check |
| High fuzzy (95–99%) | Usually a difference of a single word or number | Small correction |
| Medium fuzzy (85–94%) | Part of the sentence has changed | Partial rewriting |
| Low fuzzy (75–84%) | Structure is similar, content considerably different | Mostly translation from scratch |
| No match | No close record in the memory | Full translation |
Tools also have a lower threshold; suggestions below that threshold are not brought to the screen. Setting the threshold too low does more harm than good: the translator spends time reading irrelevant suggestions. As the literature on cognitive friction shows, the unnecessary information produced by the tool is itself a cost item (Ehrensberger-Dow & O’Brien 2015).
Repetitions: matches within the document itself
There is one further line in the analysis report: repetitions. These come not from the memory but from within the document itself. If the same sentence occurs twice in the document, the second occurrence counts as a repetition.
Repetitions matter particularly on a first project. Even while the memory is still empty, the internal repetition rate in a large catalogue or in a contract with many clauses can come out quite high. In other words, even a new customer with no memory history can see a discount on their first job.
Industry data supports this potential: only around 59% of translation work is entirely new; the remaining portion is in some way reusable (DePalma 2012). That figure explains why repetition analysis ought to be the standard rather than the exception.
Practice divides in two over how the repetition line is charged. Some agencies count repetitions as exact matches; others charge the first occurrence at the full rate and apply a discount to the ones that follow. The second approach is the more realistic, because the first translation of the repeated sentence still has to be produced. Asking which method a quotation applies heads off most of the arguments that would otherwise surface later.
Reading the analysis report: the weighted word
The analysis report is a table showing how the words in your document are distributed across the bands. A coefficient is applied to each band and the total comes out as the “weighted word count”. The invoice is calculated on that figure.
There are three things to look at when reading the report: the distribution across the bands, the coefficients applied, and which memory was used. The third is the one most often skipped. If the analysis was run with an empty memory rather than with yours, the repetition rate naturally comes out low.
One more warning: the size of the gap between the raw word count and the weighted word count is not good news on its own. A very high repetition rate may show that the document really is formulaic, but it may equally show that the memory is full of poor-quality records. On first jobs it is a good habit to have a few high-match segments checked by sampling.
The Türkiye context
When Turkish is the target language, match rates tend to come out lower than expected. The reason lies in the structure of the language: in an agglutinative language a single root produces a great many inflected forms, and word order is relatively free. A one-word change in an English source can require half the sentence to be rebuilt in the Turkish target.
This carries a warning in both directions for the buyer. On the one hand, expecting aggressive discounts on translation into Turkish by looking at the match percentage in the source language may not be realistic. On the other hand, a supplier saying “memory is no use in Turkish” is not right either; in formulaic document families the gain is plainly measurable.
A large share of the documents frequently translated in Türkiye, notarised papers, visa files, tender specifications, ISO documentation, contains a high proportion of set formulas. In these files the share of context and exact matches is markedly higher than in free-form text. This analysis can also be asked for in local practice, where pricing is done by page or by character alongside word count; the convention that 1,800 characters including spaces corresponds to roughly 240 words is used for the conversion (Pietrzak & Kornacki 2021, p. 83).
What does this mean for you?
Understanding match rates is the most direct way of managing a translation budget. You can start with a single question: “May I see the analysis report for this quotation?”
- Ask for the weighted word count alongside the raw word count in every quotation.
- Get the band distribution, and the coefficient applied to each band, in writing.
- Ask which memory the analysis was run with; your own memory should be used.
- Do not insist that exact matches be treated as entirely free of charge; the context check is work.
- On the first project, have a few high-match segments checked by sampling.
These five steps are less a matter of bargaining than of asking for transparency; a supplier who works well is producing this information already. You can find our pricing logic set out in detail on our pricing page, and you can request an analysis and a quotation for your own document.
Frequently asked questions
What is the difference between an exact match and a context match?
In an exact match the sentence itself is present in the memory word for word. In a context match the sentence before it and the sentence after it are identical to those in the memory as well. The context match is more reliable and a higher discount is usually applied to it.
How is the fuzzy match percentage calculated?
The software measures the difference between the new sentence and the sentence in the memory at character and word level and gives the similarity as a percentage. The calculation method varies from tool to tool; the same document can therefore give slightly different rates in two different tools.
Are low-rated matches any use?
Suggestions below the threshold usually call for writing from scratch rather than correction. That is why tools have a lower threshold and matches below it are not displayed. Keeping the threshold too low does not add speed; it breaks concentration.
Is the match discount the same for every document type?
No. In formulaic texts such as user manuals, contracts and catalogues the repetition rate is high and the discount is substantial. In marketing and creative texts repetition is low; expecting a discount on these documents is not realistic.
References
- Pietrzak, P. & Kornacki, M. (2021). Using CAT Tools in Freelance Translation. Routledge Focus, p. 83.
- DePalma, D. A. (2012). Translation Productivity Benchmarks. Common Sense Advisory.
- Bowker, L. (2005). “Productivity vs Quality?”. 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.