What Is Error Severity? Weighting Errors in Localisation

In a localisation audit report, the sentence “forty-two errors found” tells you nothing on its own. Forty-two small spelling problems and a single error that renders the product unusable appear on the same page in exactly the same way. Yet the commercial consequences of these two situations are entirely different.

Contents

The concept of error severity exists precisely to fill that gap. Miguel A. Jiménez-Crespo puts the point plainly: knowing how many errors were found may sometimes be enough, but it is also necessary to know how serious the errors are, how important they are and what effect they have on the content in question (2024, p. 71).

This article looks at the concept of error severity, at severity levels, at how weighting scores work, at the priority scale used in games testing, and at how you can set up a simple scale on your own project.

Why does error severity matter more than the error count?

The error count is the most easily measured output of an audit. Because it is easy, it tends to settle at the centre of reports. But a count does not show risk; it shows only volume. Decisions call for information about risk, not volume. Confusing the two pushes the wrong work to the top of the list.

The comparison drawn in the book captures the difference well: some errors are critical, for instance functional errors that cause a game to crash. Others are not even noticed by the average user, such as a missing accent mark on a Spanish or French word (Jiménez-Crespo 2024, p. 72).

Severity information also sets the order of correction. When only a limited number of fixes can be made in the time available, someone has to decide which findings to close first. That decision can be taken quickly and without argument only if severity labels exist.

Reading the count and the severity together has a further benefit: the real state of the content becomes visible. A large number of small findings usually points to a missing style guide, while a small number of large findings points to missing context information. The two pictures call for different corrective actions.

The difference between importance and severity

The MQM metric contains two separate concepts for measuring the overall weight of errors: “importance” and “severity types”. The two are frequently confused, yet they operate at different levels (Jiménez-Crespo 2024, p. 71).

Importance concerns the relative value assigned to each macro category of error. In some document types, for example, fluency or style may not matter much, while accuracy may matter a great deal. Importance is a decision about the category itself, and it is taken at the start of the project.

Severity types, by contrast, concern the importance or level of meaning of each individual error instance detected. Importance therefore looks at the category, severity at the single finding. When the two are used together, the overall weight reflects both the priorities of the content type and the concrete impact.

Four severity levels

The metric defines four severity levels: critical, major, minor and zero. The definitions of these levels are quite concrete and rest directly on the effect on the user (Jiménez-Crespo 2024, p. 71).

Level	Definition	Example
Critical	Renders the product unfit for its intended use	Wrong dosage information, a screen that crashes
Major	Prevents the reader from grasping the meaning, but causes no harm	A conditional clause turned back to front
Minor	Does not affect usability; most users do not notice it	A missing punctuation mark
Zero	A change that cannot be counted as an error; no penalty is applied	Replacing a term with a synonym

Critical errors occupy a place of their own. Errors of this kind can lead to legal consequences and other damaging effects; they can have an adverse effect on the brand name of a company, institution or organisation. For that reason, when a critical finding is detected, delivery is usually halted without regard to the total score.

How do weighting scores work?

Scores are assigned to the four severity levels by default: 100 points for critical, 10 for major, 1 for minor and 0 for zero (Jiménez-Crespo 2024, p. 71). Keeping the gap this wide is a deliberate choice; the intention is that a single critical error should outweigh a hundred minor ones.

The precedent for this approach was already present in the LISA QA model, the industry’s first standardised model. There the penalties were applied as 10 points for critical, 5 for major and 1 for minor, and if the total score exceeded a threshold the text was deemed unpublishable (2024, p. 68).

The total obtained by combining the scores is compared against a predetermined quality threshold. If the threshold is exceeded, the job goes back; if it is not, the job is accepted. What the threshold should be for a given content type is itself a decision taken at the start of the project.

Scoring can also mislead when it is left on its own. If the volume of the audited text is not stated, the same total score means very different things. Weighted scores are therefore usually normalised against a fixed unit such as a thousand words, and only in that form can they be compared across projects.

Priority levels in games testing

Severity levels are used in other evaluation settings as well. In games testing, the severity of linguistic errors is grouped into four priority levels (Bernal-Merino 2020, cited in Jiménez-Crespo 2024, p. 72).

Level A: Cannot be sold in this state.
Level B: Playable, but playability is affected.
Level C: Noticeable, but playability is not affected.
Level D: Barely visible.

What stands out about this scale is that it is tied directly to a commercial outcome. The top level is defined by the words “cannot be sold”; the criterion, in other words, is not linguistic but commercial. Language of this kind makes it easier to move the quality discussion out of the technical team and towards decision-makers, and it speeds up approval of the correction budget. A similar scale can be set up for software and web projects too.

The user impact scale: unnoticeable or unusable?

The user’s point of view is frequently brought into the severity assessment. The aim is to measure the effect of errors on the user’s perception of quality and on the degree to which the text reaches its purpose (Jiménez-Crespo 2024, p. 72).

In some metrics this impact is handled through a separate scale. The scale starts from “not noticeable by the average user”, passes through the level of “irritates the user” and rises as far as the level of “renders the text unusable”.

The value of this scale lies in moving the discussion away from grammatical correctness and towards impact. A reviewer may dislike a particular choice; but the question “will the user notice this, and if so will it bother them” offers common ground.

The limit of the scale is equally clear: user impact is estimated, not measured. Seeing the real effect calls for data such as user testing, support records or store reviews. The reviewer’s estimate is a good starting point, but it is no substitute for evidence. It is therefore recommended that impact estimates be compared against real data over time.

Why are weights not widely applied?

The literature offers an honest observation on this point. Lommel states that weights are not widely applied in LQA processes; even so, he stresses that they provide a mechanism reflecting the relevant priorities within an evaluation (cited in Jiménez-Crespo 2024, p. 72).

The reason is a practical one. Weighting requires two separate labels to be assigned to every finding: category and severity. This lengthens the audit. On projects with short delivery times, teams often settle for the category label alone.

A middle path is possible. Rather than assigning a severity to every finding, marking only those at critical and major level is enough to reach a decision on most projects. The rest are treated as minor by default and the process does not become any heavier.

For the scale to work consistently, the definitions have to be shared. Two reviewers giving different severities to the same finding renders the scale itself useless. A severity definition document backed by short examples is therefore seen as the most critical precondition for weighting.

The Türkiye context

In Türkiye, localisation audits mostly run without severity labels. Findings are gathered in a single list, they all appear to carry equal weight, and the order of correction is often set by the order of that list. This can mean that the wrong items get fixed in the limited time available.

There is also a risk specific to Turkish: a suffix error that looks small can reverse the meaning. A dropped negative suffix, or a sentence built with the wrong case ending, may look like a small intervention in spelling terms, yet for the user it can produce an error at major or critical level.

In health, finance and technical safety content, carrying this distinction into the contract matters particularly. Setting down in writing what counts as a critical error, and what happens when one is found, protects the provider and the buyer alike. This article is for general information only and does not constitute legal advice.

What does this mean for you?

When you read audit reports, look at the severity distribution rather than the total error count. A report with zero critical errors and thirty minor ones is a far better picture than one with two critical errors and five minor ones.

Ask your provider to deliver findings with a severity label attached. This is a request that adds no cost but noticeably improves the quality of your decisions. Write the definition of a critical error yourself, to suit your own content; the best definition comes from the party that knows your product.

You can see how audit line items feed into pricing on our pricing page, and you can discuss which severity scale suits your own content through our contact page.

Frequently asked questions

What is the difference between error severity and the error count?

The error count tells you only how many problems were found. Severity measures the effect of each problem on the product and on the user. A text carrying thirty small spelling errors may pose less risk than a single critical error that renders the product unusable.

What counts as a critical error?

An error that renders the product or the project unfit for its intended use. Errors of this kind can lead to legal consequences and can damage the brand value of a company or institution. For that reason, when a critical error is found, delivery is usually halted whatever the total score.

What does the zero level mean?

The zero category is reserved for changes made to the text that cannot be counted as errors. The change is recorded in the quality system, but no penalty score is applied. A reviewer replacing a term with a synonym is a typical example.

Should weighting scores be used on every project?

Not necessarily. The literature notes that weights have not been widely adopted in practice. Even so, weights offer a mechanism showing which priorities apply within an evaluation. On small projects a simple three-level scale is often enough.

References

Jiménez-Crespo, M. A. (2024). Localization in Translation. Routledge, pp. 68, 71-72.
Lommel, A. (2019). Weighting mechanisms; cited in Jiménez-Crespo 2024, p. 72.
Bernal-Merino, M. Á. (2020). Linguistic error priorities in games testing; cited in Jiménez-Crespo 2024, p. 72.
LISA QA model penalty scores; cited in Jiménez-Crespo 2024, p. 68.