How do we deal with segments that have been modified? +1 and -1? Or could it be an additional component with a different fudge factor?
Also, for the aspect of combining the multiple sources, we should be able to refer to the original modification for the actual change diff instead of the translation source. If caching is made and attached to the page version itself, the information shouldn't be to hard to reach.
consider this change scenario:
en, fr, es : in synch : a, a, a : ~20
mod on fr : a, ab, a : +3
mod on es : a, ab, ac : +10
fr -> es : +4 (notice the translation created an additional segment to create the same meaning)
es -> en : 20/(20+3+10) = 61%
Also, we should probably round the percentages to something meaningful, like increments of 5%.
Reply to Question
Seb Mon 25 of Feb, 2008 22:20 GMT
The work of modifying consists pretty much of removing old text and adding in new text, so I figured modifications could be counted as a removal+addition combo. So that would be +1 and -1/N.
Am I correct to say that you point out that every content element, while not explicitly represented in your architecture, can be identified precisely? If so, then yeah, I guess we could derive a quite accurate formula combining the lengths of all the elements where they first appeared with the current length of the target to get our metric.
I was trying to keep it simple, but if this is easy to compute we should do it.
Also, for the aspect of combining the multiple sources, we should be able to refer to the original modification for the actual change diff instead of the translation source. If caching is made and attached to the page version itself, the information shouldn't be to hard to reach.
consider this change scenario:
en, fr, es : in synch : a, a, a : ~20
mod on fr : a, ab, a : +3
mod on es : a, ab, ac : +10
fr -> es : +4 (notice the translation created an additional segment to create the same meaning)
es -> en : 20/(20+3+10) = 61%
Also, we should probably round the percentages to something meaningful, like increments of 5%.
Am I correct to say that you point out that every content element, while not explicitly represented in your architecture, can be identified precisely? If so, then yeah, I guess we could derive a quite accurate formula combining the lengths of all the elements where they first appeared with the current length of the target to get our metric.
I was trying to keep it simple, but if this is easy to compute we should do it.