The best way to show how difficult it is to assess quality is to use a recognized solution.
The Localization Industry Standards Association, or LISA for short, has developed a program, based on the American standard SAE J2450, that evaluates the quality of localization. This is the LISA QA Model (currently Version 3.1). This program is designed to simplify quality assessment and it does this by predefining error categories and assigning weights (levels of severity) to each category (a missing comma is obviously not as serious an error as a mistranslation - for example, "open valve" instead of "close valve").
You therefore only have to enter the number of errors that were determined for each of the various categories and you receive the result in the form of "PASS" or "FAIL".
The problem lies elsewhere: LISA defines 46 (!) categories that are very subjective and this means that, in practice, there are always intense debates about how a specific error should be classified and whether it can even be considered to be an error.
One reason for this is the specifications that are made by companies in order to preserve the company identity. As these are continually enhanced, this often leads to contradictions and ambiguities. In various quality discussions with our customers, we were able to prove that an "error" was due to such contradictions and that it should therefore not be considered to be an error.
Another problem is the quality of the source text. The poorer the quality of the source text, the more difficult it is to deliver a good translation. (LISA does allow for this in the QA model).
In Germany, there is an organization that deals with quality issues and that can provide information and support:
©2009 Eule Lokalisierung GmbH