The thought experiment @Judaka proposes wherein we easily say "he" to refer to a male imaginary character like an anthropomorphized cloud without sex is a good one. The claim is that we cannot be referring to sex if the referent lacks sex. But one problem I see is if we take "it" to refer specifically to non-human referents, "it" wouldn't have been an ideal candidate to refer to the imaginary cloud. English lacks a pronoun referring specifically to animate sexless referents.

There is the possibility however that had English a gender which marked non-human, animate things, and if there existed a pronoun in that gender, that this pronoun would be used to refer to the cloud or whatever. I believe some languages distinguish these features by gendered nouns, so it would be interesting to look at fables or modern children's stories to help answer this question. I'm not totally sure the precise candidate pronoun that would fit best if pronouns necessarily marked sex does indeed exist in some language. If it doesn't, that might itself argue in favor of pronouns referring to the gender concept.

Section 2.1 in this paper for example touches on this a bit, I haven't read the whole paper yet but I think looking at languages that make some of these distinctions English is missing would shed light on this question.

I'm not totally sure that the thought experiment covers the possibility that when we use "he" to refer to the male cloud, we are just picking the best word we have based on some notion of semantic similarity. Does that require a concept of gender role or identity? If people use "she" to refer to a boat, I think they are assigning some female characteristics to the boat, via metaphor, where "female" could just as easily refer to sex as to a notion of gender which is based on but different from sex.

What do you think @Judaka?

@Rosie I think you're right that perception is a necessary component of gendered pronoun usage given that speakers aren't omniscient, but how do we know it is perceived gender (ie not sex) as opposed to perceived sex? What is a good test to distinguish between these two interpretations?