As at all times, it relies upon.
If the system is already sufficiently internationalized, and different localizations have already been carried out, then future localizations could also be well-defined, documented efforts and their scope could also be simply estimated based mostly on previous efforts.
Alternatively, if the system is not internationalized in any respect, then you could must incur the overhead of that effort inside your first localization job. Therefore, it might take a lot, for much longer.
Moreover, the required accuracy of the interpretation effort and the area over which it operates must be factored in. For instance, it may be easy to check the interpretation a general-interest web page from English to French (notably when accuracy is not important), however it will likely be fairly a much bigger effort if the web page is medical literature the place the accuracy of every phrase in context may be of utmost significance.
I do know of no magic formulation which might specific the “localization testing” as a % of the “total testing”. I can not think about that there’s such a normal formulation that might be helpful throughout totally different functions/initiatives. What if the remainder of the challenge seems to be easy to check (maybe as a result of you’ve examined the English model a whole bunch of instances earlier than) – then your localization share should be greater. However what if you’re utilizing model new expertise as a part of your localization effort – then the localization effort may be an insignificant subset of the general testing challenge.
You may be greatest served to research what’s in entrance of you, take a guess on the first localization, use it as a pilot challenge, then estimate the remaining localization based mostly on what you study to that time.
In case you are pressured to current a guesstimate utilizing a share, strive 10% for the “frequent” language, and 20% for the “unusual” language, and hope for one of the best.
Good luck!