I'm making a separate issue because the reason for the duplicates seems sufficiently different.

The Arabic wordnet doesn't just have arb:lemma for its entries in the TSV file, but also arb:lemma:root and arb:lemma:brokenplural. Often the ...:root one is identical to the regular lemma. The script to check for duplicates only looks at the offset+pos and the lemma, not the second column, so these look like duplicates:

 05169813-n	arb:lemma	شأن
-05169813-n	arb:lemma:root	شأن
 05169813-n	arb:lemma:brokenplural	شؤون

We are currently not doing anything with roots and brokenplurals when converting to WN-LMF, but maybe we want to keep the redundant lemmas here in case we want to add alternative <Form> elements or something later.