René Alix has basically covered it in his answer (Vote #1 René Alix’s answer to Why can’t we perceive onomatopoeia in other languages as easily as in our native language?)
But there’s something that’s only implicit in René’s answer, that I’ll make explicit:
No actual dogs really sound like that. And so you get the transcription of a facsimile of a sound representing entire microcosms of sounds, and end up with varying, though often vaguely similar words. Oh, and then somebody goes and transliterates them from another language with a different writing system, so we find out that Mandarin Chinese use ”wong wong” for their dog bark, and Arabic speakers use “nabah”.
To make it explicit: onomatopoeias in language are still somewhat arbitrary mappings of sound to meaning. There is a somewhat arbitrary choice of a particular rendering of a bark or a splash into sound, and that choice becomes conventionalised as a word of a particular language.
Because the choice is somewhat arbitrary and conventionalised, it will make sense to you once someone has pointed it out to you, and you are already familiar with it. But without forewarning, you may be surprised. As English speakers are indeed surprised at wong wong or nabah—and as Arabic or Chinese speakers are surprised at woof woof or arf arf or bow wow.
You aren’t surprised at Chinese onomatopoeia because you learned those words, just as you learned any word in Chinese. You can recognise the limited degree of non-arbitrariness of the words, their Iconicity. You are surprised at English onomatopoeia, because you expect onomatopoeia not to be conventionalised, but purely transparent in its iconicity. And that’s what I’m saying: onomatopoeia is not as iconic as you expect, and our native language knowledge of onomatopoeia blinds us to that.