Currently we are dealing with a particularly crafty vandal on Chinese Wikipedia. The person in question is likely running automated scripts through open proxies. Their behavior pattern consists of inserting random phrases after punctuation marks and is fairly easy for a human to identify from the diff page (example).

However, as AbuseFilter currently does not support word-level diff, I have resorted to a hacked toghether regex in hope that it can be filtered out. Would it be possible to support word-level diff in AbuseFilter like /source/mediawiki/browse/master/includes/diff/WordAccumulator.php?