In the Arabic wordnet (wn-data-arb.tab), instead of arb:lemma we get either arb:lemma:brokenplural or arb:lemma:root, however there are a small number of arb:lemma:brokenPlural (note the uppercase P):

grep ':lemma:' wns/arb/wn-data-arb.tab | cut -f2 | sort | uniq -c
   2770 arb:lemma:brokenplural
    180 arb:lemma:brokenPlural
  14683 arb:lemma:root

I think we should normalize brokenPlural to brokenplural.

There is a separate issue where the Arabic wordnet file without diacritics (wn-nodia-arb.tab) only has lemma for that column, not even arb:lemma. I'm not sure what to do with this file.