zompist wrote: ↑Tue Jul 11, 2023 6:19 am
On the other hand, the mere fact that different levels of interruptability exist should suggest that by itself, ‘interruptability’ alone isn’t useful as a criterion. People have made this point more strongly with phonology: the phonological hierarchy is typically postulated to be something like ‘foot < word < phrase’, but in fact there seem to be more (or possibly less) levels than that, which don’t even necessarily line up with each other, so using them as a criterion is cross-linguistically a bit pointless. (See e.g.
https://doi.org/10.1111/lnc3.12364 — incidentally another great article on wordhood!)
Haspelmath is deeply into cross-linguistic comparisons, so that bothers him a lot more than it bothers me.
Oh, that article isn’t Haspelmath’s. And I’m quite deeply into cross-linguistic comparison, but it doesn’t bother me either that (say) the phonological hierarchy doesn’t have exactly three strictly delineated levels — it just means that that theory is straightforwardly wrong, and that it’s meaningless to talk about a ‘phonological level’ unless you’re talking about a specific language.
Also, I don’t really understand what kinds of ‘taking fuzzy categories into account’ you’d like to see here. He’s looked at ten different criteria, and demonstrated that each is neither sufficient nor necessary to define the kinds of things which people have called ‘words’. If those criteria are themselves fuzzy or ill-defined, that just reinforces his point.
Fuzzy doesn't just mean vague.
I know what it means. (Remember, I’ve read your
Lexipedia!)
The thing is, "make very precise definitions with 100% binary values, then use those for definitions" is precisely not dealing with continua. For the most part, he's showing that binary definitions don't work well. (There's a page where he literally shows these as pluses and minuses.) Again, Chomskyans get hung up on those pluses and minuses, but they're probably wrong about that.
Yes, exactly: he’s taking lots of examples of binary definitions, and showing that they just produce nonsense. The pluses and minuses aren’t his, though — they’re his summary of what binary tests
other people have used.
I don't know what an approach using continua would look like, though if anyone wants to use it as a dissertation topic, I'd read that. I expect you'd get a lot of correlations rather than ironclad laws, and that you'd get a lot of clumping that doesn't turn out to be universal.
I agree with this — and it would be a very interesting research topic indeed! I’m less optimistic than you about the clumping, though (or ‘clustering’ as Haspelmath calls it). This was in part the point of the ‘pluses and minuses’: if there were any clumping, we might expect to see correlations there, but I can’t see much of anything.