Hi Fredrik, Thanks for the example. I understand your concern better now. However, I wonder if we are not going too far with those processing expectations. For instance: An XLIFF document may be the input for an aligner tool. It would break sentences and align the resulting segments. So its input may be like this: <unit id='1'> <segment>
<source>Error detected: File not found.</source>
<target>Erreur détectée : Fichier non trouvé.</target> </segment> </unit> And its output like this: <unit id='1'> <segment>
<source>Error detected: </source>
<target>Erreur détectée : </target> </segment> <segment>
<source>File not found.</source>
<target>Fichier non trouvé.</target> </segment> </unit> The input would have an existing target and it would be perfectly reasonable to re-segments and align the content. As long as the operations are controlled the results should be valid. In other words, doing anything to the target can make that target a mismatch for the source. Re-segmenting an existing target is no more dangerous that other operations. Also: <target> has an order attribute that can be used to modify the order in which the target segments need to be re-constructed. So even the example you gave could probably be handled properly. Cheers, -ys
