The Stupidly Complicated Linguistics of Toddler Languages

[@Truttle1]

It’s ironic how the g sound is less preferred by toddlers when our onomatopoeia for toddler sounds in English is “goo goo ga ga”

[@4rumani]

true!

[@MichaelDarrow-tr1mn]

yeah it really is-Woah wait I just noticed that's Truttle1! Like. The, Truttle1! Hi Truttle1.

[@fernandomarchi5606]

G was the first consonant sound my baby learned to do. But now in his toddlerhood, I don't hear him using G anymore

[@YEWCHENGYINMoe]

Why is truttle here

[@TechieSewing]

Had the same thought 😂 It's one of the most common sounds for _infants_, and I think 'goo goo ga ga' is actually about infants? Ma, da, ba, goo, ga were the most common in my kids aged 6-9 months.

[@Defektyd]

There have been absolutely crazy experiments with toddlers but still never expected toddler language to be studied. You learn something new every day.

[@ellotheearthling]

I think it has to do with studying how children acquire language

[@helenjohnston3178]

Edinburgh uni has a whole team specialising in language acquisition. They regularly ask for kids to take part on Facebook.

[@fungoose2195]

Its an entire unit of my highschool English language class. The final project was to transcribe and analysise a childs speech according to development stages. I would argue that the most important thing to study in children is their language.(also one of the nicer things to study becouse it helps the research if they're having fun)

[@AloisMahdal]

I can assure that *your* toddler language was studied as well, with loving fondness (and occasional amusement).

[@johnnye87]

Interesting thing about child language acquisition: children learning to talk are late to acquire dorsal sounds but babies babbling make them all the damn time. The classic "goo goo" isn't just a trope, they absolutely say that along with all kinds of gurgling sounds. My five-month-old was straight-up saying /ʁ/ and I was like "do you know how long it took me to get the hang of that in French class?" To me this is evidence that learning speech is a process of learning oromotor control rather than anything else. They can *say* /g/ (when just flapping their mouths at random) and they can *hear* /g/, but learning "how do I make a /g/ on demand?" is the difficult part because unlike coronal and labial sounds it's harder to see and imitate what other people's mouths are doing.

[@ريحانة-و8ك]

your theory would especially make sense if blind children don't always learn coronal/labial sounds first, since they can't see them. i wonder if there's a study on that...

[@peppidesu]

7:45 booba???

[@mordirit8727]

To romanticise a bit since I have no way of even beginning to back this up with anything approaching data, newborn predilection for the easy /g/ might even be the very reason /g/ is set so low on preferences for toddlers. You can imagine that "that one sound I make all the time" grows tiresome to the toddler as their intended meaning is quite literally never understood, I could see this leading to wishing to be away from it when they first start forming words, "get rid of that sound that never got me anywhere, I have these new sounds that actually make people understand me."

[@jonathanodude6660]

@@mordirit8727 its always so wild to hear advanced logical reasoning personified to explain negative feedback loops. i highly doubt the toddler is thinking any of that or has that much intention behind their actions/reaction, but the core argument may be plausible.

[@EmmaMaySeven]

Syntactician: It's not _my_ fault you don't understand generative syntax trees. Also syntactician: * Hugs Chomsky books to chest. * (Source: am syntactician.)

[@Copyright_Infringement]

So I'm a linguist and have never gotten a good explanation from other linguists as to how generative syntax and optimality theory even conflict in any way. Not my area of research, but tbh they seem largely compatible, if not in certain areas identical. As a syntactician yourself, can you please help? I'm so confused

[@PhoenixClank]

As a computer scientist I do love my abstract syntax trees. Have no idea what a generative one is tho.

[@EmmaMaySeven]

@@Copyright_Infringement I'm not a phonologist so I really can't comment intelligently. I know that's a frustrating response, but OT is so on the margin of my studies. My apologies. (Although, in general, I think the problem is more to do with different theoretical assumptions about how language is processed. The question is more "what *makes* a good model?" rather than "what *is* a good model?" So OT might work but be unacceptable for other reasons. But don't quote me.)

[@pennyfarting]

​@@PhoenixClank "Generative" refers to the generative syntax model that powers the tree. Basically, in order to figure out how English syntax works, syntacticians will try to build a theoretical ruleset that will generate all possible grammatical sentences in English, and no ungrammatical sentences. If there is a grammatical sentence that the model fails to generate, or if it generates an ungrammatical sentence, then you make adjustments to your model and try again. The trees are essentially how you execute your model to test which sentences your proposed rules of syntax will and will not generate.

[@derjulius4120]

@@Copyright_Infringement I'm an undergrat student so I don't know anything, really, but doesn't it make a difference that (if I'm not mistaken) GS proposes "rules" that categorically cannot be violated for them to still produce grammatical utterances, while the very idea of OT is that all constraints can be violated in certain cases? So, GS is looking for a set of rules that describe how sentences are generated in a certain language and that are never violated in this language, while OT is looking for rules that may be broken, so they must come to different rules pretty often, right?

[@Protoplanetary]

ngl my intuition has always been that toddlers simply lack the access to muscle memory that adults have developed, despite being able to process the sounds properly. It's physically easier to do a coronal or a bilabial than a velar. It's easier to drop consonants, and it's easier to devoice at the end of a word. I'd equate it to understanding the techniques of a guitar player and the sounds he's producing without being able to produce the sounds yourself because you lack the actual physical technique, you don't necessarily lack understanding of it.

[@nicolasglemot6760]

Yes, and that's exactly why they rank their structure constraints higher than their faithfulness constraints

[@comradewindowsill4253]

@@nicolasglemot6760 except. this is supposed to model the way someone's brain works. saying that, because someone is physically incapable of producing certain sounds, that's actually part of the way their mental model turns words into sounds, is just... the wrong way round. like, look, this all depends on the idea that 'ta' is what the child intends to produce. *it's not.* because the 'bat' that they're hearing from other speakers *isn't an underlying representation.* it *cannot* be, because you can't *hear* an underlying form-- only surface forms are audible. the child is attempting to reproduce the *surface form.* that other speakers produce-- 'bat'. there are speech differences that are essentially processing issues, where the bit between UR and SR creates something that other speakers wouldn't say, but the speaker cannot recognize that it's incorrect-- but with children, they have a lot more physical limitations than mental ones. first of all, Klein seems to be underestimating just how difficult it is to reproduce adult speech with an underdeveloped vocal tract, and bad fine motor control of the articulators, particularly the tongue and lips. children, at the same time as they have to learn how to eat solid food and drink from a cup while possessing a highly variable number of teeth, a short oral cavity and a not fully developed hard palate, have to learn how to speak. is it any great surprise that, as there are difficulties with the first task until the child's mouth develops fully and gains better muscle control, there might also be similar difficulties with the second? it's either disingenuous or ignorant, to try to have a conversation about speech realisation in toddlers without considering their very obvious physical limitations at all, and the fact that this video doesn't have anything to say about this topic convinces me that this is either a self-taught linguist, or a highly unobservant one.

[@ivythepotato2417]

@@comradewindowsill4253 While I agree with most of what you said, and I do think that physical limitations play a large role in how toddlers speak, I don't know if I'd go so far as to insult Klein himself over it. He pointed out many problems with OT and notes in the description that even this is a simplified explanation, so it's completely possible he simply chose to focus on the theory. It certainly isn't disingenuous or ignorant; at worst, it was a small oversight.

[@GoofballPaul]

It could be both, one reinforcing the other (either one-way or bi-directionally)

[@skuzza405]

we can see this in adult learners of foreign languages that don't have an aptitude for certain sounds - for example, in te reo Māori "ng" can appear at the start of a word, but this trips up alot of english learners who can't pronounce "ng" without a sound before it as "ng" can't appear at the start of the word, it it gets reduced to an "n" as that's the closest equivalent their muscle memory for speaking allows. this happens a lot in the name of my iwi - ngāpuhi - where a lot of learners reduce the name to "nāpuhi" or even "nāpu'i" as the destressed H tends to disappear due to how NZ english tends to reduce unstressed Hs (like in the word "behind", it gets pronounced more like "bey'ind")

[@tropezando]

Ah, this makes me miss being around kids. I love watching kids master language. My youngest nephew was a great communicator by virtue of having three older siblings, so other than using F in place of S and W/O in place of L, he spoke well. One quirk he had was he used to refuse to use the words "yesterday," "today," and "tomorrow," instead weaving sentences of "not this day but the day after this day" hehehe

[@treacherous-doctor]

I can't believe I'm procrastinating on my linguistics homework by watching videos about linguistics

[@snowyyyyyyyyyyyyy]

happens to the best of us

[@leokyle6195]

Wow. Just remembered i used to call my "grandma" "bumma" as a toddler. Now I know why. Amazing

[@TheBigGuyBillyBob]

My stepmother called hers "Bunga," which is very similar.

[@mr.gentlezombie8709]

Along almost the exact same lines, I turned "grandpa" into "boppa"

[@lithiumpoisoning8677]

Grandma is Bamma in my language, nice coincidence

[@COArSe_D1RTxxx]

that's a bummer

[@ryanpmcguire]

@@TheBigGuyBillyBobShe was a caveman

[@videakias3000]

my sister's first sentence when she was a toddler was very impressive. she did something naughty, our mom scolded her a little and then she used all her brain power to say : "dot scod Toto never!".(don't scold Myrto never!). our mom was shocked and impressed. my sister tried really hard to think all of these words and my mom could see her pausing in between them. also note that that the conversation happened in Greek, not English. saying "never don't" is Grammatically correct in Greek by the way. The sentence would also be correct if she said "me" instead of her name. edit: changed the "at" to "in" and added the "our" behind the "mom".

[@lococomrade3488]

In English, we refer to the situation as "in Greek," not "at Greek." But as a dopey American, I assume this is also from Greek grammar. ❤😂

[@videakias3000]

@@lococomrade3488 in on and at are nearly identical words most of the time.

[@lococomrade3488]

@@videakias3000 Lol! They are fairly different in English.. "in" and "on" could almost be opposites. 🤣 To put my hand on your shoulder.. a nice comforting act. To put my hand in your shoulder... that's an injury. 😂

[@videakias3000]

@@lococomrade3488 how can you say "in" Greek since the conversation did not take place inside a language? "at" Greek sounds more correct.

[@lococomrade3488]

@@videakias3000 that's a great question that I don't have a valid answer for.... But I think it may be because we consider a language to be a large sum of words.. and the words used in a conversation are found within a certain sum. Inside the boundaries of a language.. so we do consider it inside the language. "At" is more of a location idea, to be somewhere. You can be "at the store," which is the same as "in the store." But we could also say, "I'm at the store, just pulled in the parking lot." To consider the entire property of the building and the parking ~at the store~ but only "in" when inside the building. "A book written in French." But never "A book written at French." Yet, "A book written at a French restaurant." is the same as "A book written in a French restaurant." 😂

[@Samuel-p17]

I don't pronouce Rat and Rad the same way. Maybe it's just maybe it's the swabian Dialect, but i stop the air flow after the t, so it sounds harder.

[@kklein]

german obviously has massive dialectical variation so it's fully possible. i live in the north of germany so Rad and Rat are definitely pronounced the same

[@playtypus4592]

One thing to keep in mind is that devoicing isn't the same as aspirating. So it's certainly possible that Rat and Rad sound different. The thing is that both German and English have a lenis-fortis contrast which is characterized by both the presence of voicedness and lack of aspiration for the lenis (or weak) form and the inverse for the fortis (or strong) form. However, in some instances you enter a sort of "undefined" state where it may be unaspirated but also unvoiced. A typical example for this is t, p or k after s in English (as in stop, spit, skip). And I strongly suspect that something similar is also the case in German when we talk about word-final consonants. For what it's worth I've also felt that there's an audible difference between words like Rad and Rat, but that might similarly be due to where in Germany I've lived so far.

[@n.bastians8633]

@@kklein I live in an area where Auslautverhärtung is definitely a thing, but a lot of people just won't believe you if you tell that that they do it (even if they clearly do). I blame the power that your native language's phonology has on the way you perceive and interpret sounds.

[@timseguine2]

I'm not a native speaker, but I think I aspirate one but not the other. Problem is, when I think too hard about my pronunciation it changes.

[@salticus]

At least according to Wikipedia, the final devoicing (or "Auslautverhärtung") is only really present in Northern Germany. Which, now that I think about it, is a really big part of the stereotypical northern German dialect, even though I never consciously noticed it. Just think of a northern German saying "Der dreht am Rad".

[@Idkpleasejustletmechangeit]

This just makes me think that I was a weird child, because I used to insert random "g" and "k" when I was starting to speak.

[@MaoRatto]

Regionally, from the South East of the USA. We do something similar. Age vs. Letter H, well... When Age is the final syllable, we unintentionally raise the pitch of the vowel before, but at the expense of voiced consonants... It didn't help we didn't have very articulated Dental consonants like T, D, Eth, Thorn. ( AKA (θ) and eth (ð) )... It mostly happens to /dʒ/, /ʒ/, /tʃ/ < /ʃ/ occurs very badly where it divides the pronunciation of " Appalachia ". ʃ or tʃ, the south it is /tʃ/, but weak. Also I noticed regionally our pronunciation of consonants moves towards the back into fricatives. So even we make the mistake of /g/ < /k/ along side /ɣ/ and its voiceless version.

Perhaps you were overcorrecting?

[@louiserocks1]

I have 2 kids and none of them can say t or d sounds for some reason, they replace them with k and g every time

[@mikk.t.7824]

​@@louiserocks1do they see dogs as gods?

[@louiserocks1]

@@mikk.t.7824 lol yeah I guess both those words would be pronounced like "gog" so yeah. There have been countless funny situations where I've misunderstood them because they say cake instead of take etc

[@arctan4547]

this feels kind of like linguistics picking up a physics like model sounds cool af on paper, works decently in practice but has some flaws that are patched by a different model

[@stephenspackman5573]

So much lx is this way.

[@matt92hun]

As a wise man once said, every model is wrong, but some are useful.

[@pennyfarting]

Pretty much all of "serious" linguistic theory has worked on that physics-esque model since Chomsky came on the scene in the 1960's. And it pretty much all started with those generative syntax trees that Klein throws shade at toward the very end of the video, lol.

[@karlhendrikse]

When my daughter was small she couldn't make k or g sounds, and she consistently replaced them with t and d respectively. I thought that was cool because t is to d as k is to g. So it was always "bid tar" instead of "big car". Her favourite color was deen which became daween, dween, dreen and finally, green. Also I found interesting that one day she switched from "yeyyow" to "lellow". It happened very suddenly, literally on a specific day. She never said yellow or leyyow and was unable to do so when she tried. It took about a year after that until she was about to put both the sounds together and say "yellow".

[@raedev]

Mostly unrelated but this reminds me of a story my family keeps repeating of how my first word was "abba" and my parents were fighting over if I was saying "babbo" (dad) or "mamma" (mom) until my aunt came in with a glass of water ("abba" in my home language).

[@ruufs2384]

As a German, I don’t think this model is that bat

[@varshai4704]

What a throwback to my phonology class! Lovely. I even tutored it. I can only hope that my 'students' were half as excited to have learned something new as I am now, having watched this excellent video.

[@Yulenka-]

To me this model is very intuitive. The generator is not an algorithm or a deterministic process, it represents a choice we make when subconsciously deciding how to voice a particular word. And you don't "generate" a bunch of options and then evaluate them, evaluation of constraints happens at the same time as "generation" so that you only "generate" the final output once. It's just useful analytically to separate the two processes conceptually in order to focus on the constraints themselves - the part that's more intereresting to linguists.

[@bb010g]

my intuition is that it's also a probabilistic generator that yeah isn't fully eagerly generating candidate sets but i am not a linguist

[@kettleworks]

that constraints model is super interesting and seems a really interesting explanation for variances in language acquisition. helps explain why i used to say “lellow” instead of “yellow” as i was learning to talk, while my younger brother went with “yeyyow” instead, despite growing up in the same house and around the same people with only timescale being an outlier

[@TheSkyGuy77]

I remember my sister used to say "lellow" too.

[@eyemotif]

same!

[@1manApocalypse_CP]

I'm pretty sure I used to call yellow "nenno" or something similar.

[@Copyright_Infringement]

And so, the endless squabbling between the generative grammar model and the model for generative grammar continues Which will prevail?

[@kklein]

this comment is so real

[@marieobst8850]

As a German, it was mind-blowing to me that final devoicing isn't just a universal truth across every single language simply because its physical impossible for humans to pronounce voice consonants at the end of a syllable. I legitimately learned I was wrong at age 20 or so from my English speaking friends after I proudly proclaimed that "bad" and "bat" are homophones in English. Nowadays I at least know in theory they're not but I'm still unable to perceive any difference or pronounce them differently.

[@gtc239]

Bruh i used to think otherwise, i was raised into a Sundanese family and they speak indonesian with their final voiced plosives voiced (cus their native language has them phonemically), i used to think that ALL Indonesians do it too until i hear someone from a Javanese descent pronounce "jawab" as "jawap", which... I stubbornly corrected him cus it's b not p, but then i realised that whole ppl in eastern Java devoice their final voiced plosive.

[@TurtleMarcus]

A more extreme example might be Japanese learning English. Japanese usually struggle with consonant clusters, saying "taburu" instead of "table". They may even say that it is impossible, or they believe they are saying it correctly when they are in fact not. Meanwhile, North-Germanic languages thrive on consonant clusters, saying "strength" or "Sprache" without any problem. To explain this to my students, I use the analogy of playing an instrument. It's not that Japanese are less capable of making consonant clusters. Biologically, they have the same mouth and voice box as any other human. But when a Japanese learns English, it's like a clarinett player learning to play the guitar. Your fingers can play the instrument well, just not that particular instrument.

[@astridw4737]

I was 21 in a (german) linguistics class when I learned about deutsche Auslautverhärtung. That's weird, I thought. What's so specifically german about that? Voicing final consonants just seems unnatural. I never noticed that native English speakers do it with ease. It took me even longer to realize that English [w] is different from german "w"([v]).

[@blueastronaut4944]

In Russian, final consonants are also devoiced, so I had the same problems. I still pronounce all verbs ending in ed as et, and it's very hard to get used to it.

[@tonydai782]

@@astridw4737Well yea, anything in your native spoken language will seem natural to you, native speakers don’t think about the way they speak with each word, they just say it the way that feels right to them.

[@johnnzboy]

Fascinating. I'm an experienced English-language teacher but I am still in awe of how deep and subtly human consciousness and language are intertwined - I'd never even imagined this constraints-based system, it blows my mind (if it turns out to be correct) ;)

[@Magikarpador]

this explains why as a bab i wasn't able to say commercial and could only say commershimal and i couldn't hear that i was saying a different word

[@teoo3252]

This seems reminicient of the difference between Newtonian and Lagrangian mechanics: the former states laws that the solution follows, while the latter chooses a solution among infinitely many that minimizes the energy and conserves some force. (I am not a physicist btw, so my understanding may be faulty.) The short story "Story of Your Life" by Ted Chiang explores this physics concept, and is also the basis of the movie "arrival".

[@cavalcadeofbobs3559]

Am I crazy or did you also just watch that Jacob Geller video? It's a mind-bending concept. Also, I believe it's "Story of Your Life".

[@teoo3252]

⁠​⁠@@cavalcadeofbobs3559Thanks for the correction (I made the mistake of truncating the name of the short story collection). I have not watched that video, so thanks for that suggestion as well.

[@rubyredlotus]

As a child I had issues with the pronunciation of words, but I also developed a habit of attaching "at" to the end of certain ones. My speech therapist said that I had invented my own grammar system.

[@friendly_sitie]

When I was a toddler I exclusively said "la" in place of yes, so that generator must be doing some funky stuff in the background

[@berlinflight_tv]

I took a couple of linguistics classes ages ago, and your final remarks very much sum up the feeling I got about field back then. It reminded me a bit of pre-Copernican geocentric astronomy: It seems like we’re missing some kind of important fundamental concept, so we have to build complicated models in order to align our ideas of reality with the observable evidence. Maybe that’s just wishful thinking on my part, though. Maybe there isn’t some hidden underlying linguistic principle that would make everything much easier and more elegant if we were to find it - maybe languages and the human mind are just complicated, messy and sometimes self-contradictory. 🤷‍♂️

[@saskiascott8181]

It can't be as complicated as this though. This theory makes no sense, there has to be a simpler explanation than near infinite possible representations going through a checklist of thousands and thousands of constraints one by one before vocalisation. I'm with you on this one

[@constanza1648]

I like that explanation. It seems to explain why when I hear my niece to mispronounce a word and I try to say some word closer enough to the thing she is pronouncing but far from the word that she is trying to say, she look at me with this look as if I am mocking of her. e.g. in the example bad>ta i would say "oh, tag? tap?" she will look at me saying "ta" again as correcting me for not getting it right when her SR is totally linked in her head to the UR.

[@dragonapop]

This is also why historical spelling is important. Currently in German there is final devoicing, however that rule might change one day. If we were to spell the words "rad" and "rat" with a T at the end, then we'd loose the morphemalogical ability to tell the two meaning apart. An example with English is that the words "soften" and "often" are where the T before the "-en" suffix silent. However younger people (like myself) say these words with the T pronounced, reconnecting their pronunciation to their morphemes of "soft" and "oft". I think changes, where 'unrealized' or silent sounds, will become more common as the world globalizes. Because of loan words from other languages requiring unintuitive nonnatural pronunciations, which force native speakers to contradict their languages rules. Resulting in younger generations not developing the same constraints as their parents, allowing the child's pronunciation to have more nuance than their parents. Source (???) never heard of that word. Great video as is your standard 拜拜~

> If we were to spell the words "rad" and "rat" with a T at the end, then we'd loose the morphemalogical ability to tell the two meaning apart. We would cope. We already can't tell eg Gründung (founding) and Gründung (green manure) apart by spelling alone, and the sky hasn't come crashing down.

[@kasane1337]

They are pronounced differently however (at least in Hochdeutsch / Standard Northern German), with short "ü" vs long "ü". Also, I've never heard of "Gründung" as green manure before, I would call that "Grüner Dünger" or something. Anyway, a different example I like is "umfahren" (to drive around smth.) vs "umfahren" (to drive smth. over). Very different meanings with the exact same spelling, but there's a tiny difference in stress (something like "umfáhren" vs "úmfahren"). And that stress might be enough to save your life.

[@dragonapop]

I never said it was earth shattering important. All i said was it's useful for telling the words apart. Historical spelling is useful for making other languages easier. Take the English words "eight" and "daughter". The "gh" is useful if someone is learning German (or another Germanic language), as the German equivalent is "acht" and "tochter". So you know silent English "gh" means German "ch". C is also a useful letter because it can inform you of how these sounds evolved, showing you the pronouciation it another language. Take English "centaur" and Macedonia. In Russian, they are "Kentavr" and "Makedoniya", which makes it easier to learn other languages. Historical spelling makes it easier to learn other languages. And if every language spelled words hisorically, learning them would be much easier while making the system more complex.

[@dragonapop]

@kasane1337 Yea, it would be nice if Germanic languages would have some way of written word stress. They are mostly unimportant except in specific instances like "incíte" vs "ínsight". Or pérfect and perféct, depending on if it's a verb or adjective. It would also be a cool tool for writing dialog in text. Especially dictionaries though. It would be nice if English used any accent makes, like at all. Also, it's funny that me (the American) will write the German nouns in an English text with lower case letters. But you (the German) default to using capital letters when talking about German words in English.

[@matt92hun]

That happens all the time though. For example in English "red" and "read" can be pronounced the same while "read" and "read" are pronounced differently. Or there's the whole cot-caught merger.

[@wisteria3032]

10:19 no! NO! what do you mean "ignoring why the 'jord' disappears for simplicity"??? keep going I want at least a 1 hour lesson! this was SO INTERESTING! explain MORE!

[@AnarchoPinkoEuroBr]

Wish this video was available in Portuguese since it explains to Portuguese people why we developed an epenthetic vowel to break harsh consonant codas, why we developed palatalization of /ti/ and /di/ and why we have a tendency to delete superfluous consonants at the end of words (including our loss of second person singular conjugation, but also the way we may pronounce "as meninas" or "nossas coisas" as "as menina" or "nossas coisa"), things that are pretty common across world languages but that seem odd and "incorrect" in the context of only knowing English and nearby Romance languages.

[@Rickmonas90]

that doesn't happen in European Portuguese

[@AnarchoPinkoEuroBr]

@@Rickmonas90 by we, I obviously mean Brazilians

[@asailijhijr]

The English word 'wolves' satisfies one of the requests you made in the video.

[@adamadamadamadam]

I can think of a bunch of English words with unvoiced finals specifically as singular nouns, mostly v/f. Hoof, knife, wife, life, gift/give, half, calf. Maybe of germanic origin?

[@liamboardman8776]

This is a fascinating topic, thank you for the introduction to it! It seems like OT might be able to help explain why some autistic/ND children start speaking later but with surprising complexity- they're not needing to prioritise the rules related to physical constraints

[@fabibi_ha]

4:08 Reminds me of my teacher encouraging us to write down multiple versions of a word and choosing the best looking one, if we weren't sure about the spelling (double consonants and whatnot)

[@nemo-x]

Simple explanation for your concerns at the end: the evaluator is actually an abstract representation of linguistic ease. This is also how languages evolve! Structural constraints for ease of pronunciation, faithfulness constraints for ease of understanding. Just think about how the germanic word bad HAS evolved into bath in english! Or i believe it's the other way around in make > mache! How about du>thou>you Street strass. Heim home. Leg lay. Good gut. Nacht night. Flieg fly. Sorge sorrow. All words evolve according to the aame constraints. What is easier to pronounce, what is easier to understand and differentiate from others. So it is a BRILLIANT model. And the generator is just random noise in the brain, and in speech according to the constraints themselves. The way people mishear things or accidentally misspeak! "i wanted to eat.... A croissant" -"quaso!" "Gn8" "good night to you too" All the examples of eggcorns. Incorrectly applied foreignisms like in habañero. Sometimes literally just will to make a language more clear. Aluminum aluminium. Or more fancy colour color. All these things generate new possible pronounciations and spellings. And depending on which constraints the overall speakerbase finds more important those alternatives generated with those constraints survive. I'm in frikkin love this is the best linguistic model ever!

[@YourRyeBread]

Its likely a stimming factor cuz when you mentioned each sound in order of preference I was like "oh, yeah cuz those dont feel very good when struggling to speak already" cuz I struggle with occasional speech loss and just end up sounding quite like a toddler (and I’m autistic which is why I even noticed the way it feels cuz I’m sensitive to the vibrations and other physical feelings that are involved with talking)

[@makkurokokkuri]

when i was a wee bairn i called my sister (nearly 9 years my senior) "dudu". every day when she'd get off the school bus and walk toward the door, i'd go "DUDU!!!!!!" and it's nice to know why!

[@liamtahaney713]

Dutch does the same ending consonant tning but they actually spell it correctly. Objectief vs objectieve. Never realized that was a thing phonetically just kinda learned the right thing. Woooow

[@shytendeakatamanoir9740]

Oh yeah French is kinda the same too! Sorta.

[@skreame9064]

Well, not really - Dutch has a lot of final devoicing as well. We don't have a lot of variation in word-final voiced consonants, but every word ending in -d is pronounced with -t: bad, raad, vod; and every word ending in -b is pronounced with -p: lab, club, web. This is also the reason so many people struggle with d/t/dt (e.g. spelling "hij wordt" as "hij word" or "gehaast" as "gehaasd") - it all sounds the same. Sidenote: We even devoice many voiced consonants in the middle of words: even -> efen, gezien -> gesien :) This is only true for Dutch in the Netherlands though, Flemish (and also Brabants and Limburgs) does generally voice its consonants in the middle of words.

[@liamtahaney713]

​@@skreame9064I get the impression with dutch it's a lot less prescriptivist in general. Am living in Antwerp and learning Flemish, and my god the dialects make my head spin...

@@skreame9064 Dutch and Northern German dialects (Plattdeutsch) are relatively closely related, and (mostly) mutually intelligible.

[@warido37]

that whole intro was the last thing i was expecting after reading the title, hilarious loved it

[@TheStickCollector]

I haven't heard of this concept before, sounds like something they would do.

[@rmt3589]

This is amazing and gamechanging!!! Is gonna completely change my approach to my conlangs. Thank you so much!!!

[@thejontao]

For me, personally, the theory helped me understand (that is, gave me an explanation) why I’m often misunderstood when speaking my second language… I know that my pronunciation isn’t correct. I try, but I will always have an accent, and some words are just hard to pronounce. So, when someone is confronted with the sounds that come out of my mouth… if we imagine them generating a list of all possible utterances I might be saying, and then using constraints to eliminate them (possibly adding in grammatical and context constraints), we can see them ultimately arriving at a word I didn’t intend because their generator function and constraints are the ones of their mother tongue… ones that I haven’t come to comprehend, and different from the ones I am using, which are influenced by mother tongue.

[@dhonors999]

I feel like this idea is very similar to principal component analysis (from linear algebra) which on one hand makes me like it, but on the other hand makes me feel like some of it's explanatory power just comes from the effectiveness of PCA and not that this stuff is actually happening in the brain. Like PCA can be used to identify real physical phenomena, but it is fundamentally just a good way to create a model that fits whatever data you fed it.

[@adrianblake8876]

11:23 The rule applied isn't no coda, but one that all syllables must have a head (a consonant must preceed the nucleus). E.g. in French there's a rule that alveolar consonants are dropped at the end of a word, so "les femmes" is /lefam/, but if the next word begins with a vowel or "h", the rule doesn't apply so "les hommes" is /lezom/, because you can't have /leom/...

[@ezdispenser]

a lot of the other comments are making great points about the theory, but i'd imagine that the "generator" works by taking the original word and just running it through the constraints, changing it as need be and stopping when it doesn't feel like the same (or a real) word anymore. toddlers don't have as much experience with language or words, so they have much more flexible limits on what "the same word" sounds like, and as they get older the limits become more refined and strict. (though i guess HOW strict, and what those limits are, depends on your dialect and the people you're around. like regional slang and accents and such.) i'm not a linguist, but this seems like the best explanation based on how my brain attempts to conjugate words i don't know (take the original word, run it through various ways to conjugate it, see what "sounds right" using context clues like spelling/pronunciation and origin)

[@tuluppampam]

I feel like OT could still be explained through rules, just like how you can combine functions in maths.

[@Head0.25s]

An amazing introduction to OT, kinda makes me wish I'd seen this video before starting my OT course.

[@simonwxyz]

It'd be interesting to approach child sign language acquisition with this framework. Given that anecdotally they can pick up sign much easier than speech, so perhaps the range of constraints are also simpler?

[@florisvdvelde5447]

Dutch also has final devoicing, and as I learnt English I had trouble not devoicing final consonants. Really cool to see this might be why!

[@rextanglr4056]

I frequently joke that toddlers have their own language that no one else understands. Turns out, babylangs are real…

[@tj-co9go]

Except parents

[@AloisMahdal]

@@tj-co9go even with parents, it's often mostly an educated guess.

[@SilentGamer._]

@@AloisMahdal all of language is an educated guess

[@SisterSunny]

you've successfully made me fascinated in radio gaga

[@mertaliyigit3288]

Turkish has a very similar case like german. If a word ending with a hard consonant gets a suffix, the consonant gets softened. For example "kitap" (book) becomes "kitabım" (my book). This is because those words were borrowed from Arabic and got applied a hard-consonant in the word end rule. So when the word gets a suffix, it doesnt "get softened", it goes back to its original form instead

[@tregaskin98]

Also many people in Turkey doesn't pronounce the 'r' sound in present continuous suffix '-(i)yor' even in İstanbul due to anatolian influence caused by migration to urban areas and it wasn't the case in İstanbul at least 30 years ago. However in some places such as Adana, people still pronounce the 'r' when they spoke in their accent.

[@UnQuacker]

Yeah, except it doesn't happen with the Arabic loanwords specifically...

[@davidkinnear1905]

Constraits theory seems to be in line with language learning difficulties. Such as when spanish speakers find it difficult to pronounce the /s/ sound without a starting vowel if /s/ is followed by a consonant.

[@thanelinway3042]

In my toddler language I used to derive new words. Ex: "Gabish" was washing machine (no traceback, prob. onomatopeia), and detergant was "gabishia".

[@-beee-]

Heading your description sounds so much like what we see in machine learning. It’s interesting how there are such parallels across fields.

[@jeremylee48]

Silverman (2006) critical introduction to phonology offers a different approach to this, and it's a usage-based one for anyone interested

[@rox4884]

This makes sense when you think about singing. There are sounds at the end of musical phrases that you turn into a softer sound. This makes the off rhymes work in poems and songs as well.

[@lulujuice1]

This sounds like how sound changes happen.

[@danteregianifreitas6461]

In English there are words with the devoicing rule. Words that end in -ed are switched to -t if their root verb ends in certain letters. Like "cracked", "help", "watched", "kissed".

[@Moses_Caesar_Augustus]

The word for 'pigeon' in my language is /kəbuːtər/, and I remember when I was little I pronounced /kəbuːtər/ as /bʊmbʊm/, which is SO weird because these two words don't sound similar at all but still I understood what people were saying and people understood what 2 year-old me was saying.

[@ananas22anne]

as someone who studied a bit of linguistics, that is actually so interesting!

[@strangeWaters]

There's a great paper called "The behavioral approach to open and interconnected systems" that talks about a similar move from rules to constraint-based modeling in control theory

[@iskren4086]

Devoicing happens in some Slavic languages as well. In Bulgarian боб [bob] 'beans' is pronounced [bop] but in Old Bulgarian it was бобъ [bo-bŭ], but the end vowels (ь and ъ) started to be dropped and that caused many two syllable-words to become one-: градъ [gra-dŭ] → град [grat] 'city'; прагъ [pra-gŭ] → праг [prak] 'doorstep'. But in Serbian there's no devoicing.

[@Pablo360able]

As someone who once took an organic chemistry class, this reminds me of resonance structures. These are hypothetical, "rational" arrangements of bonds in a molecule that aren't necessarily how any molecules are shaped but instead "contribute" in some way to the molecule's actual shape, which is often hard or impossible to explain just by talking about conventional bonds. Look at the chemical structure of TNT for a good example. Maybe there is no generator, and there are no candidates, but rather there's something conceptually analogous to a fuzzy phonetic space "surrounding" the UR, and the constraints are simply a representation of the rules that determine how much each element of that space contributes to the SR.

[@citrezene]

as a german speaker I never thought about final devoicing. very interesting, thank you!

[@vigneshanand8490]

How do we know that the reason why toddlers don't pronounce words correctly is because of a mistake in the evaluator instead of a mistake in the underlying representation? Also, it seems like this model could explain any inaccuracies in toddlers' pronunciation of words by saying that the inaccuracy is due to the constraints being in the wrong order. This means that no evidence would falsify this theory, since it is always possible to come up with an order of the constraints that supports why a toddler made a specific pronunciation error.

[@amazingcaio4803]

The video explains it. Take jorgubbar and bubba, for example. The perception is unchanged, despite the different pronunciations. I think OT is very much falsifiable. Ordering doesn't necessarily lead to any possible surface representation. For example, no amount of reordering FinDev and IdentIO[voice] would yield [pat] from /bad/, because either [bat] or [bad] is ranked higher. Besides, the more data you have, the more difficult it is to determine the correct constraints and ordering.

[@willguggn2]

They can clearly understand words they can't pronounce properly.

[@stekeln]

​@@amazingcaio4803 A different ordering could produce 'pat' if there was a (theoretical) InitDev constraint further down the chain. Moving it up the chain could mean eliminating 'bat' before reaching IdentIO[voice], resulting in 'pat' being the only remaining candidate.

[@amazingcaio4803]

@@stekeln You're not wrong, but how does that relate to my original point?

[@stekeln]

@@amazingcaio4803 I don’t quite see how OT is falsifiable based on the point you made about ordering. I was giving an example of how promoting an arbitrary fringe constraint (essentially the same as introducing a brand new constraint) could change the surface form. By extension, nearly any surface form may be achieved when reordering, given how difficult it is to identify *every* constraint necessary to model a particular language, or even an individual speaker. That said, OT does not claim that reordering a given set of constraints can result in *any* surface form. This is easily disproven, and it doesn't invalidate OT. Apologies if I’m misunderstanding your argument. I’m trying to explain where my confusion lies.

[@moritz584]

Such a nuanced perspective! I love it

[@thecoolkittensarecool]

earliest ive ever been to a K klien video!!!!!!!!!!!!!!!!!!!!!!! btw best thumbnail EVER

[@TobiasFangorIsntCis]

🤝

[@hananc]

for some years I have been thinking about the possibility of computers understanding toddlers speaking. this video brought me closer to realizing there will be solutions to this.

[@FleckerMan]

I guess it seems like a useful model, it's okay for part of it to be "and then a list is magically generated" - models are not reality, but ideally it gets superseded with a better version/something better eventually

[@franticranter]

It seems strange to me to say thatr the findev phonological constraint in German is the sames as, say, your constraint against /g/ as a child. the "constraints" in children are not primarily phonological constraints (as we can see that they can understand the phonological outputs of the adults, and thus their phonological constraints), but rather constraints in their ability to pronounce something (dunno if there's a word for this, but I'll call it "production constraints"). Ofc phonological constraints often arise from production constraints, but until they're phonologised, they're still categorically different, and have different effects on how the people understand the sound and the word. So in the case of Rat and Rat, it's not just that if a german said "rad" then you'd be impressed that they were able to do so, you'd think they were pronouncing it wrong. In contrast, when you said "bubba" as a kid, it's not that you thought "jordgubbar" was wrong, you clearly phonologically understood the phonological constraints that the adult was following there, you just found it difficult to pronounce there.

[@barihong5629]

When I was a child, I said (Polish phonemes) /i ɛ ɐ ɔ ʊ ə/ as [i ɜ ʌ̽ o ʊ̽ ə], however, I didn't grew up from them - I stay with them in accent as well in my native tongue. Word “jagoda” was replaced with [hʌ̽jodʌ̽]. I also know from my family, that I wasn't able to pronounce /t͡s d͡z s z n/ - instead, I had only /t͡ʂ d͡ʐ ʂ ʐ t͡ɕ d͡ʑ ɕ ʑ ɲ/ - I just replaced basic set with palatal, only before voiceless plosive with retroflex set Feature common in my family & some others near, is that voicelesss plosives are aspirated & fricatives are geminated, except wordfinally - & I think that it is just lenis-fortis distinction, that is neutralised wordfinally - wordfinally there is no aspiration, no gemination, etc., just voiceless plosive - Patryk is /pʰatrɘk/, /k/ isn't aspirated (/t/ varies, some pronounce it as aspirated some not)

[@saphir7632]

I've learned and I'm learning german and I never noticed that there was a "final devoicing" rule in this language, even tho I was applying it for years. It sounds pretty incredible to me.

[@RyanLynch1]

i actually really like this because it represents well how people start from a somewhat random but mostly similar blank slate then get molded towards what they were taught. they can still change, it's just harder since they've practiced with one constraint ordering for so long cool theory imo

[@tj-co9go]

My mom says I used to say "aija-a-ii-a-aa" instead of "karjalanpiirakkaa" (Karelian Pie, it is a type of bread) in Finnish. Couldn't pronounce consonants or very few of them initially, so k, r, l and p were out of the picture and only a, i and j available.

[@vermillion2023]

bro speaks in morse code

[@YuriTarrdid]

damn dude I'm usually a fan of you obscure takes on language but this is over my head

[@TKDB13]

The bit about avoiding dorsal consonants is something I've noticed in my 3-year-old niece. She replaces "k" and "g" sounds with "t" and "d", respectively. Originally it was always just "t", but more recently she started distinguishing the voiced and unvoiced consonants, which I find incredibly fascinating.

[@A.F.Whitepigeon]

Have we considered the possibility that toddlers are in fact attempting to say "bat," but their tongues and lips do not yet have the coordination to do so, and the closest they can get is "ta"?

[@brunnomenxa]

8:13 🤨

[@kasane1337]

Kinkshaming is so real, even Linguists are doing it, smh smh

[@Tordek]

"Can't we just say the word is spelled different?" Me: ????NO???? you have a perfectly good rule, why would you do that?!?!

[@jucom756]

i feel like cognitive linguists if they want to do any halfway decent job at a model should work together with, or take a class in, neurology. because a lot of these algorithms are like they would be performed by an exact computer rather than a weight based neural network. i think that is why optimality theory is somewhat more reasonable than some other genrative theories, because through it's ranking of restrictions it emulates the weights certain inputs have to the final decision.

[@LOONACORE]

final devoicing reminds me of korean! internet in korean is pronounced... well, internet (인터넷). the ㅅ is typically a s sound, but at the last of a syllable block it is pronounced like t, hence internet. however as soon as you put it next to a ㅇ it turns back into an s sound, so 인터넷을 would be internesul ! amazing how grammar seems to have common threads no matter where in the world you are (idk ipa, sorry)

[@kklein]

this is not final devoicing but it is another phonological process! i'm not sure what is happening here since i don't have enough data and don't know korean, but it might be intervocalic fricativisation, where because a hard plosive (stop) sound like /t/ is appearing between two vowels (internEsUl), it becomes a softer, continuous, fricative sound made in the same place like /s/. this is just a possible constraint idea from the very limited data though. but i agree, the fact that similar rules can be found across languages is fascinating

[@coyo_t]

when i was learning english i had my own pseudolanguage that still sticks mom's mom is "nana", mom's dad is "geekoo" and one i really wanna bring back, "banana" is "nomenae" not sure how this fits in the constraints system but idk. i want more people to call them Nomenae -v-

[@Lucroq]

Oh shit, this explains why I used to say "Bampe" instead of "Lampe" and my brother just used to leave out unnecessary sounds in the front of words alltogether, like "Neemann" instead of "Schneemann", because we were still able to effectively communicate the word.

[@derpauleglot9772]

0:53 You pretty much nailed the pronunciation there, nice^^

[@harjutapa]

idk I'm pretty sure toddlers go through that phase of not fully pronouncing words because they're still figuring out how to make their mouths make the right sounds, and certain sounds are just harder to mechanically replicate with our mouth parts. Some kids are better at moving their lips and tongues around in the required ways, so don't have such a phase (or the phase is much shorter), but I've watched my brother and his wife teach his niece to say words in two very different languages (Japanese and English), and she had problems with the same hard sounds in each. It seemed to me as being much more mechanical than it did any sort of "linguistics engine" problem.

[@comradewindowsill4253]

also their mouths are soft and underdeveloped! when they first start babbling they have like 4 teeth. their tongue muscles aren't strong yet, and their hard palate isn't all that hard yet, and their oral cavity is just too short and small for fancy backing maneuvers.

[@Jerdifier]

The absolute most basic part of the “linguistics engine” is probably true (that there’s an underlying word we’re trying to say), and that adults are mostly limited by learned rules. But for toddlers, I’m almost certain you’re right about it mostly being physical instead of mental. There’s a reason why J/Ch/Th/The are almost always the last sounds acquired when babies are learning English (the first two are combinations of d-zh and t-sh, and the last two require teeth to pronounce).

[@Archimedes.5000]

I mean you just described a physical constraint lol

[@stekeln]

Optimality theory doesn't make any claim that the constraints need to be psychological or learnt. Physical constraints are _the most_ constrictive and are therefore ranked above any constraints in the language itself.

[@anoakenstaff]

This is genuinely one of the most fascinating things I have ever watched, Linguistics is amazing. I was a bit curious about one thing, though it isn't directly _related_ to the video, sorry- why did we ever invent or rank faithfulness constraints as higher than structure constraints? Why did we ever need to make a change at all?

[@kklein]

the idea is because of homonymity - essentially, without creating more distinctions, all our words would sound the same as each other

[@anoakenstaff]

@@kklein That makes sense, thanks a bunch for the reply! I'm honored to receive an explanation lol.

[@keegisuvakas6847]

that is indeed interesting, esp the "no consonant at the end of syllable" thing. From my experience studying finnish and being estonian, finnish has more vowels at the end of syllables, but their words are longer for it. Compare the words for "on the ground": maas(est) vs maassa(fin). It's a bit difficult for me to imagine the longer word being the easier one, but maybe that's just me growing up in estonian. thank you for the video!

[@sarah.s.flanagan]

Adding this to my Watch Later and coming back after I've had my morning coffe 😆

[@leirbag75]

If you don't like Chomsky's syntax trees, what you need is dependency grammar as opposed to Chomsky's constituency grammar. It's so much cleaner, with exactly one node per word, no preferential treatment for subject vs object... Just makes more sense all around. And it has a longer history, too-it goes all the way back to Panini, an Indian grammarian who apparently lived somewhere between the 7th and 4th centuries BC

[@placer7412]

i havent eaten all day and feel tired as shit and i started laughing my ass off at 10:02 because I have no idea what any of this shit means i dont even know how i got here

[@placer7412]

like i get a ranking system but my brain just can not heat up enough to understand all this

[@AloisMahdal]

sounds like you had fun watching, maybe your brain "knew" that's what you need (relax, have fun, get a good sleep...) and navigated you here 😀 (my brain does that well, although it almost always wants to skip the third part)

[@hydroxa4330]

It's always been my intuition that these sounds are like falling short. They reach to say one letter, and say another that's close but not quite there. When I learn languages that have sounds i haven't pronounced before, I'm not coming up with a substitution sound, I'm actually trying and missing the mark, with my mouth falling into place of a different, similar sound. I distinctly remember having this as a kid, too, when learning my mother tongue. The difference between then and now is that I care about sticking to rules now. Back then, I only cared that the listener understood me.

[@vari1535]

omg is the next (or at least a future (please)) video about generative syntax trees

[@chillero3heftig712]

meanwhilst me as an adult: using glutteral stops in my conlang like crazy and no voiced consonant besides d XD

[@brandonharwood9066]

Thank you for this, this somewhat explained why, as a toddler, I was adamant that the "milk" is actually "noop"

[@eueumesmoaquelecara4638]

I found very interesting this theory about the Surface and Underlying Representations, I think it helps me to understand better why it sometimes seems that what we're pronouncing is different from what we think we're saying. Maybe it can also help make sense of some grammatical rules that exist in languages; which natives follow so naturally. Per example the English -ed, which can be realized as -t, -d or -id. I'm not a native English speaker, so I ask you guys: when you say words like "trapped" and "grabbed", do your minds think you're saying "trapt" and "grabd", or the former ones themselves?

[@nngnnadas]

It's intresting that English speakers are doing something similar, final f is becoming voiced in the plural shelf -> shelves. But they do conceptualize it in the second way, as the consonant changing from one to the other.

[@comradewindowsill4253]

you say 'is becoming', but to be clear, it's been that way for a long time. it's evidenced in the spelling, it's not exactly a new phenomenon. actually, this is a whole separate pluralizing morpheme from the standard [z]-- there's a whole class of words like this, ending in /f/ in the singular and /vz/ in the plural, including leaf, self, sheaf, dwarf, etc. they're a different class, because instead of the pluralizing phoneme assimilating its voicing to the previous phoneme, the opposite happens-- the last phoneme (always [f]) of the stem assimilates to the voicing of the always-voiced pluralizer.

[@jantomasanwasenpule1459]

From a Wikipedia article it briefly mentions how this final devoicing only occured in old English with the sound f/v. (I could be misinterpreting how it worded it ) The example it gave was half/halves. But it sounds like in old English there was NOT a spelling difference, so that mustt be relatively recent. Another example I thought of was with dental fricatives. Moth/moths, mouth/mouths, bath/baths, though we don't make a spelling distinction is this sound so not as noticable.

[@LorenzoF06]

​​@@jantomasanwasenpule1459 I'm pretty sure Old English had intervocalic voicing, rather than final devoicing. Some examples from Wikipedia: stæf ('letter') /ˈstæf/: [ˈstæf], stafas ('letters') /ˈstɑfɑs/ > [ˈstɑvɑs]; smiþ ('blacksmith') /smiθ/: [smiθ], smiþas ('blacksmiths') /ˈsmiθɑs/ > [ˈsmiðɑs]; hūs ('house' noun) /ˈhuːs/: [ˈhuːs], hūsian ('to house') /ˈhuːsiɑn/ > [ˈhuːziɑn]; forþ ('forth') /forθ/: [forθ], compare eorðe ('earth') /ˈeo̯rθe/ > [ˈeo̯rðe], fæþm ('fathom') /ˈfæθm/ > [ˈfæðm]

[@FenrirRobu]

The generator is very congruent to parallelization and speed, which can be seen manifested in artificial neural networks. Although it doesn't line up perfectly 1:1. The idea of an initial listeme that gets converted to a matrix, and then every single option in the matrix gets checked, with only one survivor is quite efficient, because in the last step you would only need to "combine" all of the results, which is like the sum of 0+0+0+1+0+0+0 = 1. I can't say if current ML models will spontaneously come up with this structure, but we might be able to "discover" it one day.

[@Mobin92]

German is my native language, and I absolutely would never do the thing you explained in the first section.

What thing specifically?

[@Mobin92]

Saying Rat instead of Rad

[@vitriolicAmaranth]

I'm only a minute in but I already see what you mean, this toddler language called German is very complicated indeed.

[@klank9409]

first 3 minutes of this would have really helped me back in the day

[@klank9409]

wow that was more than i expected excited for the five hour rant about generative syntax trees