Are ten findings in a text many or few? The answer depends on the length of the text. Error density makes this question answerable by relating the number of findings to the volume, and makes it possible to compare jobs of different sizes.
Definition
It is the ratio of the number of findings to the volume of text. The measurement is proportional. This is how comparison becomes possible.
The reason for a ratio
The raw number depends on text length. Two jobs cannot be compared. The ratio solves this.
Unit of volume
The unit used for the ratio is defined. Words or characters may be used. The unit is fixed.
Keeping the unit fixed
The unit does not change between periods. If it changes, the trend cannot be read. Consistency is preserved.
Weighted density
Findings are multiplied by severity coefficients and added up. A score is used instead of a raw count. Measurement becomes finer.
The inadequacy of a raw count
An unweighted count equates ten minor flaws with one critical error. The result misleads. A coefficient is needed.
Calculation by sampling
In large jobs density is calculated on a sample. The whole text is not examined. The method is practical.
Representativeness of the sample
The chosen section must represent the whole text. Easy sections are not chosen. The selection is methodical.
Comparison with the threshold
The calculated density is compared with the acceptance threshold. The decision comes from here. The criterion is written in advance.
Threshold by document type
A different threshold may be defined for each document type. One threshold does not fit every job. The criterion is adapted.
Critical findings work separately
A critical finding does not enter the density calculation. A single instance exceeds the threshold. The rule is separate.
Periodic monitoring
Density is monitored from period to period. A single measurement shows no trend. A series is built.
Reading the trend
A rising trend shows a problem in the process. The root cause is sought. Measures are taken.
Monitoring by translator
Density can be monitored per resource. It is used in matching decisions. The data is interpreted with care.
The interpretation trap
Difficult texts naturally produce higher density. The text, not the resource, may be responsible. Comparison is made with care.
The effect of text difficulty
If source quality is low, density rises. This is not translator performance. The distinction is preserved.
Sharing with the client
Density data is shown in the report. A common criterion is set. The discussion becomes objective.
Distribution within the document
How the findings are distributed through the text is also looked at. Clustering in one section shows a different problem. The distribution is examined.
The meaning of clustering
Findings gathered in one section may show that the section was translated under different conditions. The root cause is sought there. An examination is made.
Conditions for comparison
Only jobs measured with the same unit, the same typology and the same severity scheme can be compared; a comparison made without these conditions produces numbers but not information. That is why the measurement scheme is kept fixed. Consistency is preserved.
Limits of measurement
Density alone does not show the usability of the text. A critical finding is always evaluated separately. The limit is known.