I stumbled upon the claim that vowel quality model based on tongue height and frontness has been known to be incorrect – and the so called openess and frontness are actually two formant frequencies.

On one hand, I understand why that might be the case: an [a] played by an audio recorder, spoken by a parrot or even generated by a speech synthesizer has no tongue position, but is an [a] nonetheless. So I guess if one had to define vowels on the lowest abstraction level possible, they'd have to define them in terms of spectrogram properties.

On the other hand, consonants shouldn't be defined by place and manner of articulation then – and yet they are. So what's so inaccurate about the tongue position vowel quality model: doesn't tongue position directly result in different formant frequencies? Are there some cases where humans can articulate different vowels without changing tongue position, or what?

"Incorrect" is a pretty strong word after all.

P.S. I'm aware that the raising/fronting movement trajectories of the tongue aren't perfectly vertical/horizontal nor perpendicular to each other, and that the real oral cavity isn't shaped like a perfect right-angled trapezoid, but that IMO doesn't make the model wrong – more like a crude approximation, which it certainly is.

  • 1
    I’m not primarily a phoneticist, but I’ve never seen it claimed that vowel quality being based on tongue position is wrong. Obviously, it’s the formants (not the position of the tongue) that our ears pick up and use to distinguish the vowels acoustically, but as you say, those formants are formed by how the tongue’s position affects the oral cavity. Also, why would an [a] spoken (mimicked) by a parrot not have tongue position? Parrots have tongues and use their oral cavity for sound production (though syrinx-based instead of larynx-based). Nov 26, 2023 at 14:17
  • @JanusBahsJacquet, syrinx of some birds is kinda similar to a loudspeaker and allows them to generate a great diversity of complex sounds. This is how parrots mimic human speech, and how lyrebirds reproduce all kinds of sounds from other birds' calls to the sounds of a chainsaw, car alarm and camera shutter. Similar to a loudspeaker, they do it without having to reproduce human articulatory gestures (otherwise how would they say bilabial sounds, without lips at all?)
    – Slavus
    Nov 26, 2023 at 15:36
  • @JanusBahsJacquet, I imagine they might use their oral cavity and tongue to further modulate the sound produced in the syrinx, but the tongue and oral cavity shapes are so drastically different that you wouldn't expect the "same" articulatory gestures in humans and parrots to produce even remotely similar sounds. Parrots might have some tongue frontness and closeness when they speak (like you say, they have tongues), but it's not what they rely on to say different human vowels.
    – Slavus
    Nov 26, 2023 at 15:45

2 Answers 2


There is a traditional phonological view that high vowel have in common a specific degree of tongue raising, mid vowels have a lesser degree of tongue raising, and low vowels have a degree of tongue lowering (there is an assumes reference position). Therefore theoretically you could quantify vowel height in terms of physical differences in the X-ray y coordinates and backness in terms of x coordinates. However, physiological studies have shown that this is not correct, in that front vowels don't have in common an x coordinate property and high vowels don't have in common a y coordinate. However, vowel distinctions can be reasonably modeled in terms of relations of formants to each other, hence various acoustic graphs roughly resemble the traditional vowel box.

Changes in tongue position have an effect on where the constriction lies that divides the vocal tract, where that division (strongly) influences the resonance frequencies of the vocal tract. Another factor that influences resonance is the length of the vocal tract, which can be extended with the help of the lips, and also whether the tube is closed (by a labial constriction) or open. The larynx also affects resonance frequency because it can be raised or lowered. Finally, the position of the tongue is strongly influenced by the position of the jaw, which because of how it is hinged plays a lesser role in back vowels than it does in front vowels.

Humans can articulate different vowels without changing tongue position, as observed in many languages that have vowel length contrast, tone contrasts, phonation contrasts, and nasalization contrasts on vowels. If we toss out the ubiquitous rounding distinction and focus only on IPA [i ɪ e ɛ æ a ɑ ʌ ɤ] etc those are primarily made with tongue distinctions though often there is a subtle distinction in lip spreading / protrusion between front and back unrounded vowels (even in the presence of back rounded vowels), since rounding enhances the formant-lowering that typifies backness.

I should note that one does have to factor in F3 in order to get a reasonable account of vowel distinctiveness.

The IPA labels do retain the traditional physiological idea ("close", "near-close" etc), just as consonant terminology reflects articulatory state. I think the problem that you are alluding to comes from the notion of "defining" symbols in terms of physical properties. This comes from the Jakobsonian-Halle view of what sounds are made of, that "high" vowels have in common a physical property as do "back" vowel and "low" vowel, and that these are precise articulatory instructions. So it is a significant problem when a front mid vowel has a higher tongue position than a back high vowel – that kind of fact is essentially a refutation of the underlying theory of the theory of vowel "stuff". Every theory can be said to be a crude approximation of reality, and when observation doesn't match theoretical prediction we can always say "we just need some more epicycles to refine the model". The model is pretty good as a model of phonological distinctions, it's just that the assumed simple mapping to the physical domain is quite complicated. It is an open and interesting question what the trade-off is in characterizing vowels in terms of physical targets vs. acoustic targets. However, as far as articulation is concerned, the most plausible coordinate system (which has yet to be discovered) would probably be heavily leveraged by the specific effect of contracting each muscle such as the Superior and Inferior Longitudinal, Genioglossus, Transverse, Hyoglossus, Styloglossus and Palatoglossus. With all of these muscles pulling in various directions, you would not expect the coordinate system of articulatory targets to be as simple and rectangular as predicted by the classic rectangle-and-grid system.

The main current objection to that model as a model of vowel articulation is simply that is lacks explanatory value – it doesn't solve any problems of vowel production.


I'm afraid you got your messages a bit crossed here, by which I mean to say that in a lot of phonetics and phonology it's less between articulatory vs acoustic phonetics but about articulatory and acoustic phonetics. Some phenomena are better dealt with in one subfield than in the other, some are convincingly described by both fields. For example, there's a surprising and surprisingly frequent 'jump' in articulatory phonetics when it comes to the common [kj] > [tʃ] development: [k] is velar but [t] is apical, and there's seemingly no intermediate stage along the hard palate. However, acoustically the development is less of a big step. The vowel triangle (or trapezoid if you prefer, or inverted pyramid) has originally been justified by tongue positions, but it has turned out that the formants also serve well as (rough) coordinate axes for the geometric figure to become apparent.

It's all about models, their usefulness and their predictive power; some prefer to use long vowels vs short vowels to describe a given system, others prefer to deliver a description in terms of tense vs lax articulations. Neither is wrong or right (except for the models that are really wrong of course :-), they exists in different but related frameworks and may lead to different predictions, some of which may evade the other models.

Edit how can I conclude but with the old adage, "all models are wrong; some models are useful (for the task at hand)"

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.