Hi Sigrid (and everyone on the mailing list),
The problem you mention is inherent in the original SemEval Lexical Substitution dataset which we have borrowed for our task. The 'contexts.xml' file actually is taken straight out of the trial dataset without any modifications whatsoever, and contained this mis-alignment. We have on our part corrected the 'substitutions' and 'substitutions.gold-rankings' files so that they are actually aligned with the 'contexts.xml' file in an absolute way. That is to say that right now instance with id 50 is actually aligned with line 48 in the other two files. So, in effect, if you were to ignore the id numbers in the xml file, the instances are aligned in a one-to-one relation with the lines in the 'substitutions' and 'substitutions.gold-rankings' files.
However, I am attaching herewith a corrected version of the xml file in which I have replaced the instance id tags with the correct numbers. This mistake also happens later in the file around id 135. It is, however, not present in the test set.
We apologize for the inconvenience.
Regards,
Sujay.