The Vowel Space [video] | Hacker News Reader

The Vowel Space [video]

youtube.com

93 pointsgus_massa33 comments

I ran into some videos on this channel (Dr Geoff Lindsey) a few weeks ago, and they are really good IMO. If you glance at the video thumbnails they appear like low-quality clickbait, but the content inside is solid linguistics and very informative.

Examples:

• https://www.youtube.com/watch?v=wt66Je3o0Qg (‘"Schwa is never stressed" – FALSE’)

• https://www.youtube.com/watch?v=EaXYas58_kc (Weak forms)

• https://www.youtube.com/watch?v=qfCsiF80TX0 (Iambic pentameter with the Pink Panther)

The video has a link to the following page where you can listen to the sounds of the vowels on the chart at the bottom: https://www.englishspeechservices.com/blog/the-vowel-space/

And this is how Henry Higgins listened to vowel sounds a century ago: https://www.youtube.com/watch?v=EN0wXFN0GO0

(Well, he would have, had he been real.)

Related, a matrix with playable audio of all the vowels:

https://en.wikipedia.org/wiki/IPA_vowel_chart_with_audio

His video was mainly about how that chart is inaccurate and outdated. For example, 12:50 in the video. From what I understood, one of the main downsides of the chart is it focuses on tongue position, but not lip position. There are different vowel sounds that have identical tongue positions but different lip positions.

The IPA vowel chart does show lip rounding - the positions in the chart have pairs of vowels, the first is unrounded (e.g. /i/) and the second rounded (e.g., /y/). And, tbh, it's a binary feature unlike tongue positioning.

it is not binary. Did you watch the video?

No, but I took a few phonetics classes in grad school. Does that put me ahead or behind watching the video?

I don't know, I wasn't in your classes. You would probably find the video interesting nonetheless. It's from a UCLA professor and is in part specifically about how he at least partially disagrees with the way that phonetics are traditionally taught.

It’s inaccurate, but honestly not too inaccurate. The revised chart he ends up with is pretty similar to the traditional one. The important thing is to think of it as a simplified graph in formant space, rather than as directly showing tongue/lip positions.

> From what I understood, one of the main downsides of the chart is it focuses on tongue position, but not lip position.

The main downside is that it claims to focus on tongue position, but really it uses a weird, poorly-defined hybrid of acoustic and articulatory criteria. Lindsey’s point is that focussing on acoustics alone lets us produce a diagram which is generally more sensible and more principled, but still very similar to the traditional one.

The "sbeech" video that Geoff Lindsey did is really interesting:

https://www.youtube.com/watch?v=U37hX8NPgjQ

Appreciated this video too.

It helps one understand that the lenis/fortis opposition is broader and arguably more useful than the voiced/unvoiced one even in English.

I have sometimes wondered if a poet or lyricist could use the maps of how consonants and vowels are laid out to compose under some constraints, so that the sounds danced around the mouth in some deliberate way, and what that might sound like.

GPT-4 could probably do a pretty decent job of this.

Seems completely feasible.

Somewhat related: In the 30s Gilbert Adair wrote a novel in French without using the letter e. It was translated under the same constraint into English, in which it's called A Void.

He was part of a group of artists who reveled in putting strange constraints on their work. They have a name but I can't remember what.

https://en.wikipedia.org/wiki/Oulipo

Sort of related, here are a couple rap songs that I like which self-impose similar constraints:

Blackalicious - Alphabet Aerobics https://www.youtube.com/watch?v=xxmmGToZlns

Nach - Efectos Vocales https://www.youtube.com/watch?v=eOTbm-NvLII

This tool is a fun way to play with vowels sounds and mouth position. https://dood.al/pinktrombone/

And here is a mechanical vocal tract robot. https://m.youtube.com/watch?v=HmSYnOvEueo

The lack of unique letters to describe the wide array of vowels seems like a big flaw in the Latin alphabet. Diacritics or inverted characters just aren't as good as clear and unique letters. Maybe there just isnt enough design space left to create more unique glyphs but it sure would be nice.

He talks about making a set of references values for vowels in this video. This is something I had thought of too in the past, but limited in scope of teaching English to Japanese native speakers. It struck me that instead of teaching to target prononciations of words using the whole vowel space of american or British English, which has some differences that are hard to understand for native Japanese speakers, a vowel space could be made that was optimized listener understandability, but having enough differentiation of sounds, and ease of articulation.

The revelation that I could generate a vowel that sounded awful close to /y/ with unrounded lips by centralizing /u/ slightly is blowing my mind…

In French class I was taught that /y/ can be made simply by doing /i/ while the lips are fully rounded.

Right, that's the canonical way to produce the vowel.

But what the video explains is the fact that the rounding is actually less fundamental than the length of the resonating tract. It means that you can keep your lips unrounded and produce a vowel very, very similar to the canonical /y/

Edit: I made a typo in my original comment, I meant to say "by centralizing /i/ slightly" (not "/u/")

This is why it's pretty hard to distinguish [y] from [ʉ] and [ɨ] from [ɯ]. Indeed, as I understand, in many cases the distinction is strictly phonemic, not phonetic.

> Indeed, as I understand, in many cases the distinction is strictly phonemic, not phonetic.

What is this supposed to mean? It isn't possible to maintain a phonemic distinction between sounds that aren't phonetically distinct.

I’m guessing they got it backwards accidentally: I’d say the distinction between [y ʉ] is strictly phonetic, not phonemic.

What I meant is that whether a given language is analyzed as having /y/ or /ʉ/ is often down to phonemics - i.e. how that vowel groups with others - and not to the minute distinctions in how it is actually pronounced (especially once you take allophones and regional pronunciations into account).

FWIW I'm even not sure there are any languages that actually have a phonemic distinction between /y/ and /ʉ/ or between /ɨ/ and /ɯ/. I'm sure there are some, but I would be very surprised if the actual distinction cannot be more sensibly analyzed in other terms (length etc).

Norwegian has /ʏ, yː, ʉ, ʉː, ʊ, uː/. https://en.wikipedia.org/wiki/Norwegian_phonology

Very interesting, thank you. I knew that vowel inventory in Germanic languages can get very impressive, but that formant chart is quite something:

https://en.wikipedia.org/wiki/Norwegian_phonology#/media/Fil...

In the phonetic realization section, though, there's this:

"The close /ʏ, yː, ʉ, ʉː/ have been variously described as protruded [ʏʷ, yʷː] and compressed [ʉ͍, ʉ͍ː] as well as compressed [ʏ͍, y͍ː] and protruded [ʉʷ, ʉʷː]. The backness of /ʉ, ʉː/ has also been variously described as central [ʉ, ʉː] and near-front [ʉ̟, ʉ̟ː]. Therefore, /ʏ, yː/ may be differentiated from /ʉ, ʉː/ by backness and the type of rounding or even only by the type of rounding"

This last bit is actually even more surprising: two front-ish rounded vowels that are distinguished solely by how the lips are rounded? I can't help but wonder how stable such an arrangement could be long term.

I have also found this about Swedish

> Rounded vowels have two types of rounding: /ɵ/, /ʉː/, /ʊ/ and /uː/ are compressed [...] > /ʏ/, /yː/, /œ/ and its pre-/r/ allophone [œ̞], /øː/ and its pre-/r/ allophone [œ̞ː], /ɔ/ and /oː/ are protruded [...] > Type of rounding is the primary way of distinguishing /ʉː, ɵ/ from /yː, œ/, especially in Central Standard Swedish.

https://en.m.wikipedia.org/wiki/Swedish_phonology

More about different kinds of rounding at https://en.m.wikipedia.org/wiki/Roundedness

/y/ is the rounded version /i/, so that makes sense

From the video: https://www.englishspeechservices.com/ipa-vowels/

I've built real time mobile app to provide visual feedback for vowel articulation based on the formants[0](2d/3d) graph and live recording to teach accent, still need more work, and data for different languages (as the "accepted" vowel region, i.e the region where a native speaker expects the vowel to exist is somewhat different between languages and dialects)

The problem is that when someone tries to learn a new language after years of exposure to his own language, he never really practices pronunciation usually because there are fewer vowels in his language so they map to bigger regions.

I'm sure many people had the experience that said something to native english speaker which they were sure they said correct but the listener just had a blank face and didn't understand.

As this is very delicate thing (see vowel shift) I'd rather get it right the first time and not teach something wrong but one could make his own vowel chart approximation[1] (Although it's not that consistent, and there are multiple algorithm which would produce different results). I would also like to expand it to consonants

There is a good physical explanation for why vowels are charachterized by formants. With vowels you don't restrict the airflow so the pitch wavecreated in the vocal cords travels in 3d in the vocal tract some would go straight to the exit, and some would bounce so there is destructive interference patterns in most places but for some frequencies(overtones) there is a constructive interference so the amplitude('energy') for that frequency is higher. One could model it with quadrilateral pipes of different length and shape as described in the video. I think the frequency change from you pitch is due to the 'bouncing' on the vocal tract some energy would be Refracted and some would reflect

Another cool thing about the vowel space is that the most common vowels are the one that lies on the boundary of the chat as far from each other(highest diff between the formants), this is of course increases understanding as there is a clear cut betweeen the different sounds, but if you could teach clear distinction you could "pack" more data at the same time.

Teachings kids to distinguish more precisly between vowels is just like teaching them "Color terms"[2]

[0] https://en.wikipedia.org/wiki/Formant [1] https://www.youtube.com/watch?v=BGW8J4cG0qY [2] https://en.wikipedia.org/wiki/Color_term

Reply on news.ycombinator.com