In Classical Arabic, there is a fairly rare verb šani'a شنئ, meaning "to hate", probably best-known from the third verse of Surat al-Kawthar: 'inna šāni'aka huwa l-'abtar "For he who hateth thee, he will be cut off (from Future Hope)". (Cognate words are found elsewhere in Semitic, for example Hebrew śānē', Syriac snā "hate".) This has barely survived in spoken Arabic, but (according to de Prémare) the causative šənnā is still used in Tangier (Morocco), meaning "to taunt someone by showing him something he wants that you won't give him."
Phonetically, šani'a is a perfect match for šnu (the glottal stop/hamza becomes y in colloquials, and Arabic final-y verbs normally end up in Kabyle as final-u, for reasons I won't go into) - but semantically, surely this is absurd?
So I would have thought, until, idly browsing through a glossary of the rather conservative Bedouin Arabic dialect of the Nefzaoua area in southern Tunisia (Boris 1951), I found the following entry:
شنى šnệ... inacc. yẹ́šni...; noms d'act. šänyân et šạ́ni: 1) "critiquer en vers, faire la satire"... 2) "détester"."Hate" to "criticise in verse" is a credible change, and so is "criticise in verse" to "sing". Suddenly, a connection that looked impossible becomes almost obvious.
شنى šnē... impf. yašnī...; verbal nouns šanyān and šany: 1) to criticise in verse, to satirise... 2) to hate
In this case, as in many others, Kabyle has preserved an Arabic word that almost every Arabic dialect in North Africa has lost - but to make sense of the connection you have to look at a wide range of Arabic dialects, not just checking Classical Arabic and stopping there. The converse also applies: when looking into Berber loans into an Arabic dialect, it's not enough to look just at the Berber spoken next door. People move around, and words that were familiar in one generation may be forgotten in the next one.
Of course, if the Nefzaoua data weren't available, there's no way you could accept a comparison like this - and, if several thousand years had passed since the word was borrowed, instead of less than 1500, that intermediate step probably would not have survived. In other words, semantic change can rather easily erase connections beyond any reasonable hope of retrieval. This is one of the main difficulties in long-range historical linguistics - the further back you go, the more cases like this.