>...You can segment the <para> or <p> at text
> extraction time and put each segment in its own <trans-unit>.
I agree with Asgeir: extracting and segmenting should be two distinct operations. While they can be done transparently at the same time for the user, I think it's important to make a distinction between the representation of the extracted unit and the segments.
>...If you use a spanning mechanism inside source, you will
> have multiple segments in source and target and the number
> of source fragments may not match the number of target
> fragments; that's very bad for TM/MT support and not XSLT
> friendly at all.
I agree with Rodolfo: there are some drawbacks with using spans: order, number of segments, etc. But those issues are maybe a product of segmentation-related processes we will always have. For example an automated tool can create a tentative alignment with n-to-m cases and provide the result in XLIFF for a user to finish/correct the aligned set.
Maybe there are other representations we can have other than using <trans-unit> or using <seg-source> that would allow a more seamless tracking of segments. We need to imagine it.
-ys
