If the GPT ‘genlangs’ can be translated into English reliably, and we know English definitely follows Zipf’s law (and also all the evidence for ‘neuralinga’/‘interlingua’ from NMT), it seems like the genlangs would have to follow Zipf’s law too. If the genlangs are just thin skins by GPT over existing natural languages it knows, then why wouldn’t they follow Zipf’s law?
Seems like it’d be more relevant to show that for genlangs which don’t have any viable translation into an existing Zipfian language, they follow Zipf anyway.
If the GPT ‘genlangs’ can be translated into English reliably, and we know English definitely follows Zipf’s law (and also all the evidence for ‘neuralinga’/‘interlingua’ from NMT), it seems like the genlangs would have to follow Zipf’s law too. If the genlangs are just thin skins by GPT over existing natural languages it knows, then why wouldn’t they follow Zipf’s law?
Seems like it’d be more relevant to show that for genlangs which don’t have any viable translation into an existing Zipfian language, they follow Zipf anyway.