Zompist Bboard Again

Linguoboy wrote: ↑Wed Apr 26, 2023 11:39 am

Raphael wrote: ↑Wed Apr 26, 2023 10:35 amThis reminds me, how did German and apparently some celtic languages end up with the same word - "ach" - meaning very similar things?

Onomatopoeia. Cf. English ugh, yech, yuck, etc. which all have their origins in a similar sound.

Thank you!

this comic made me think... is UG unfalsifiable? like... the cat's spiel could be countered by "okay, so what if instead your name is the entire lexicon of english, and just like pikachu has the freedom of saying different bits of its name in whatever order (pika pika, pikachu, pika chu chuu, whatever)", so you can string together the lexemes embedded in your name, so you are proof of UG.

and like... how could you tell the difference between two theories?

Torco wrote: ↑Wed Apr 26, 2023 2:02 pm this comic made me think... is UG unfalsifiable? like... the cat's spiel could be countered by "okay, so what if instead your name is the entire lexicon of english, and just like pikachu has the freedom of saying different bits of its name in whatever order (pika pika, pikachu, pika chu chuu, whatever)", so you can string together the lexemes embedded in your name, so you are proof of UG.

The cat's spiel is wrong for two reasons:
1. The universe is not deterministic. People have got to get over their Leibniz fixation.
2. A finite genome cannot contain "all utterances I will ever make until I die". Genomes are really really small.

Now, the trainer's idea of "generative principles that allow us to produce an infinite set of utterances using a finite set of resources" isn't exactly UG, but it's also so general that I think it's pretty obviously correct. Not proven, just right. Everybody loves to rag on Chomsky, but asserting that the brain is finite is not one of his many errors.

Chomsky does insist on the infinite nature of language, because that rules out some dumb-but-possible ways to produce language. The hoary old one in comp.ai.philosophy was the Humongous Lookup Table: a list of every possible sentence. (This is pretty close to what the cat claims to have.) This works very well for those dolls that say something if you pull their string... but they can only say a dozen or so things. If language is infinite, you can't explain it with a HLT. If it's not infinite... well, you can't prove that there is no HLT.

But really, things are proven only in mathematics. In the real world we settle for making a really good case. And there's no good case to be made for the HLT: it sounds reasonable only until you start really gawking at the combinatorial explosion of possible sentences. (To restrict ourselves to Spanish-speaking researchers, cf. Borges, J.L., Ficciones, §7.)

Can you detect the difference in practice? Yes, sequence the cat's genome. If it is the size of a solar system, it has an HLT.

How is this comic even talking about Chomsky? It's like saying saturated fats are good for you, therefore Andrea Dworkin was right.

I mean, sure Chomsky would agree that language is a dynamic system that is different from reciting a script, but... is there any linguist who would say that language is rote, deterministic recitation?

My guess is that the author of this comic just threw in the name Chomsky because that's a linguist most people can be trusted to have heard of, and it's a name that comes up in discussions about linguistics and the brain.

Torco wrote: ↑Wed Apr 26, 2023 2:02 pm this comic made me think... is UG unfalsifiable? like... the cat's spiel could be countered by "okay, so what if instead your name is the entire lexicon of english, and just like pikachu has the freedom of saying different bits of its name in whatever order (pika pika, pikachu, pika chu chuu, whatever)", so you can string together the lexemes embedded in your name, so you are proof of UG.

and like... how could you tell the difference between two theories?

This comic made me think like i lost part of my brain it wwas so mindnumbingly bad

I don't know, it gave me a good laugh: not exactly 0% at it.

Can you detect the difference in practice? Yes, sequence the cat's genome. If it is the size of a solar system, it has an HLT.

yes but like... we're talking about a pokemon here, they clearly are from some different, weird universe: assume you find that its genome is just fantastically big: it *could* be that it has an HLT, but it also could be that it has just a dictionary and the rules to combine those tokens in the form of a fantastically inefficient LLM. or it could all just be the information needed to fit inside a pokeball: could an HLT be proven?

I mean... possibly you could move one gene around and listen to the cat mispronouncing "that's roight!" every subsequent time or something.

Torco wrote: ↑Thu Apr 27, 2023 11:19 am

Can you detect the difference in practice? Yes, sequence the cat's genome. If it is the size of a solar system, it has an HLT.

yes but like... we're talking about a pokemon here, they clearly are from some different, weird universe: assume you find that its genome is just fantastically big: it *could* be that it has an HLT, but it also could be that it has just a dictionary and the rules to combine those tokens in the form of a fantastically inefficient LLM. or it could all just be the information needed to fit inside a pokeball: could an HLT be proven?

I think you have to look at scale here, at least at orders of magnitude. The comic actually talks about multiple universes, so it really is an HLT. For my syntax kit I did a rudimentary calculation— just taking in the variants of one ten-word sentence with a 10,000 word vocabulary. It came to 10²⁴ sentences. By contrast the number of base pairs in the human genome is 10¹². That is, of course, not half the number we need, it's a billionth of it. Far less than that, really, because a base pair isn't even a gene, it's like one genetic bit.

Also, the whole idea is a bonkers misunderstanding of genetics. A genome is not a .txt file. It's a set of instructions for building proteins, and how those work is mind-numbingly complex. It's the worst programming language ever; you can't stuff arbitrary data into it as some actual sf writers do, who should know better.

Ironically, it's Chomsky who thinks that Chomkyan grammar, and the lexicon, are encoded in the genome. So you might amuse yourself thinking out what that stupid idea would predict. E.g. there should be mutants who are utterly unable to learn the word "climb", but can learn everything else.

mʲ can sound like n in cases.

foxcatdog wrote: ↑Thu Apr 27, 2023 7:33 pm mʲ can sound like n in cases.

All nasals sound much like all other nasals depending on context.

bradrn wrote: ↑Thu Apr 27, 2023 8:20 pm

foxcatdog wrote: ↑Thu Apr 27, 2023 7:33 pm mʲ can sound like n in cases.

All nasals sound much like all other nasals depending on context.

m and n sound quite distinct to me usually. I think what's going on here is palatisation since its in russian and the name they are saying is spelt *miki (which apparently is the same name as "mickey")

zompist wrote: ↑Thu Apr 27, 2023 4:54 pm [quote=Torco post_id=69955 time=1682612341 user_id=100
Ironically, it's Chomsky who thinks that Chomkyan grammar, and the lexicon, are encoded in the genome. So you might amuse yourself thinking out what that stupid idea would predict. E.g. there should be mutants who are utterly unable to learn the word "climb", but can learn everything else.

Wait... the lexicon as well? really? cause I can imagine ways in which you could encode mentalese grammar (mentalese is such a pointless idea, but still, for argument's sake) like into the brain: make it more 'natural' to think in terms of, say, SVO than VSO or whatever you might want the way it's more natural to drink water and less natural to drink alcohol, something which children generally dislike and need to come to develop the taste for, because presumably taste receptors and sensation processing and whatever... but like, the lexicon? man, grampa Chom really pissed outside the pot in that one.

I recently heard a nanosyntax talk where someone cited generative work that assumes a universal set of abstract roots (which amounts to a universal lexicon, IINM) in a minimalist syntax plus distributed morphology framework. Close to zero people in the audience, which mostly consisted of generative grammar people, agreed and several people voiced their disagreement in the question period. I don't know what Chomsky is saying on the topic though because I don't usually understand his writing, so make of it what you will.

Torco wrote: ↑Fri Apr 28, 2023 9:32 am

zompist wrote: ↑Thu Apr 27, 2023 4:54 pm Ironically, it's Chomsky who thinks that Chomkyan grammar, and the lexicon, are encoded in the genome. So you might amuse yourself thinking out what that stupid idea would predict. E.g. there should be mutants who are utterly unable to learn the word "climb", but can learn everything else.

Wait... the lexicon as well? really? [...]... but like, the lexicon? man, grampa Chom really pissed outside the pot in that one.

Haha yes, the lexicon. "Climb" is his example, in fact— he mentions a journal article that says that the meaning is quite complex, but then himself asserts that "every child learns it perfectly right away", which means it's in the genome.

Of course he has zero interest in actual biology and very little in actual child language acquisition, so he doesn't develop this idea and his theories don't depend on it. But like all of his ideas, he acts like it's obvious and only a fool could reject it.

I'm reading the grammar of Saramaccan Creole by John McWhorter and Jeff Good, and the language basically has two tonal systems generated by the mixture of English, Portuguese and Niger-Congo vocabulary.

The majority of words are from English and Portuguese and follow a pitch accent system. Morae are either H tone or toneless, and the main tone patterns are final H, penultimate H, or ante-penultimate + penultimate H. Because the distinction is H vs unaccented, H tone spreading occurs in certain circumstances when unaccented syllables occur between two H tones in adjacent words.

A smaller group of words, though, has tone on every mora / TBU. These words distinguish H/L and don't have toneless morae. In isolation these words may look like pitch accented words, but in context they behave differently because the syllables with L tone block H tone spreading. Fully tonally specified H/L words make up 10% or less of the lexicon.

It's fascinating that the lexicon is divided in this way. There don't seem to be any words which contain all three of H/L/toneless, although there are some words which are classified differently by different speakers/dialects.

A simple and maybe dumb question: Do languages with a retroflex series of consonants (such as /ʈ ɖ ɳ ʂ ʐ ɽ ɭ/) always also have a palatal series (such as /c ɟ ɲ ɕ ʑ ʎ/)?

WeepingElf wrote: ↑Sun Apr 30, 2023 7:13 am A simple and maybe dumb question: Do languages with a retroflex series of consonants (such as /ʈ ɖ ɳ ʂ ʐ ɽ ɭ/) always also have a palatal series (such as /c ɟ ɲ ɕ ʑ ʎ/)?

I see no reason for this necessarily to be so.

I have a question about mean, as in "gohan means rice". I've noticed that some immigrants, when speaking Swedish, will sometimes use the reverse order and say things like "rice means gohan". This is ungrammical in Swedish; you have to express it like "rice is called gohan". So I wonder if this happens because of interference from their L1. How common is it in natlangs that either order is fine together with the word meaning mean? Which specific natlangs do this?

WeepingElf wrote: ↑Sun Apr 30, 2023 7:13 am A simple and maybe dumb question: Do languages with a retroflex series of consonants (such as /ʈ ɖ ɳ ʂ ʐ ɽ ɭ/) always also have a palatal series (such as /c ɟ ɲ ɕ ʑ ʎ/)?

I don't think so. According to Dixon's survey of Australian languages, while most contrast apical and laminal coronal series, there are some single laminal place languages, and the realisation of the laminal series varies between languages. E.g. Dyirbal mostly has a palatal pronunciation for its laminal series, while two of its neighbours show variation between dental and palatal (Warrgamay) and primarily dental allophones (Nyawaygi). So you don't need to contrast dental and palatal laminals, and you can have a single series which is either just dental or just palatal or you can have a series which varies between the two (not sure what conditions the allophones without some Googling, but maybe neighbouring vowels?).

One caveat here is that in some of these languages, apical doesn't mean retroflex. E.g. I've just looked at Dixon's grammar of Yidiny, which has an apical laminal contrast with the laminal series as palatals, and the description makes the apical series sound alveolar, not retroflex. I would hypothesize (without any evidence) that if you have a two way laminal - apical contrast and the laminals are palatal, then the apicals will be pushed forward and be plain alveolars (because retroflexes are relatively marked and otherwise you're lacking a "plain" coronal series), whereas if you have a dental laminal series, the apicals are more likely to be retroflex.

Linguoboy wrote: ↑Wed Apr 26, 2023 11:39 am

Raphael wrote: ↑Wed Apr 26, 2023 10:35 amThis reminds me, how did German and apparently some celtic languages end up with the same word - "ach" - meaning very similar things?

Onomatopoeia. Cf. English ugh, yech, yuck, etc. which all have their origins in a similar sound.

This seems unlikely to me because, having looked at etymology in Réunionese Creole, many words purported to be of onomatopoeic origin turned out to be of Malagasy origin e.g. tèktèk, a bird species. A very similar word exists in Malagasy but Chaudenson, otherwise great, managed to dismiss the possible link between the two and propose an onomatopeic origin. But the onomatopeic representation of sounds differs significantly from one language to another e.g. Cockadoodledoo and cocorico and koukoukou.

I'm struggling to find specific Australian examples though. Most of the two series laminal - apical languages seem to go for either the apical alveolar option, or alveolar~retroflex allomorphy. It seems like to have retroflexes you maybe need plain apicals too, which means that in Australian languages we're looking for a laminal - apical alveolar - retroflex language. But the most common option for laminal in those languages seem to be either dental-palatal allomorphy (mostly related to the vowel i if it's conditioned) or a plain palatal series with no dental.