Wittgenstein's notion of the language-game gives a credible alternative more in-line with the observed data. He characterizes informational interactions as a game, the rules of which are negotiated by the players and learned by new arrivals. We can agree the sun is round because we have learned what people mean by round, not because roundness is natural to human beings. Therefore it is not really universal, there may be tiny children somewhere whose notion of roundness is incompletely developed. They just have so little leverage on the game as a whole that we can ignore them until they adapt to the general experience and correct their notion of the label.
There are observations that are universal, and always remain that way, but that is only because the humans not open to them do not get purchase on the game. The definition does not change, because learning it early and precisely pays off too well.
And even the game itself is not necessarily universal, because there are humans born who are too autistic to 'get' how to enter any existing game. But it is widespread enough that we simply declare those people defective and run civilization as if they did not exist.
The other formulation of this same idea is Lacan's observation that 'master signifiers' are empty. Anything important to human beings is folded up into layers of references, and the references seem to point back to a single basis. But if you carefully unfold the references, you find that humans do not in fact really share what seems to be shared in the ultimate definition. We don't all really agree on exactly which ovoid shapes are round. The place where each of us puts the cutoff differs, and in the end, the number of people who exactly agree on a precise and complete definition of roundness is going to be zero.
Both of these viewpoints lead us to a notion of logic that is less realist and more psychological. Classical logic is a lovely model of what we wish language were like. By explicitly learning it, we can better negotiate the game. But it is an incomplete model with idealized elements that simply do not refer to anything. The gain is in the coordination of information exchange, and not in describing reality. Reality is better described in a piecemeal and fluid manner.