Is there any language with a vowel system that cannot be analyzed as having no more than three degrees of height with the addition of some other feature (length, tenseness etc)?

  • Your title specifies European languages; the question just asks about “any language” – are you specifically looking for European languages only? Such languages do exist within Europe (e.g., Danish has minimal groups such as mile, mele, mæle, male that show four vowel heights), but including languages from elsewhere would significantly increase their number. May 17 at 7:41
  • Sorry, this was a brain fart - yes, I was asking about European languages specifically; and "European" in a broad geographic and political sense, not linguistic. OTOH note that it's not about vowel height alone, but about how much is necessary to clearly identify phonemes when considering other contrasting features. May 17 at 18:08
  • 1
    Danish contrasts long and short vowels, but not (according to most) tense and lax. Some vowels are of course tenser than others, but tense/lax is not usually said to be phonologically distinctive in Danish. The four vowels I cited in the example above are all long (and tense), and I’m not aware of any analysis that manages to reduce them to only three vowel heights. May 17 at 18:14

1 Answer 1


You describe the standard possibilities given the SPE features [high], [low], [tense]. Bearing in mind that [+hi,+low] describes a physical contradiction so is not expected to exist, that allows three combinations, each of which can be combined with [+tense] or [-tense], admitting as many as 6 formal distinctions. English has 5 of them: [i ɪ e ɛ æ] and some dialects have two front unrounded vowels [a ae]. The crucial distinction in your question (which was the question raised in feature theory) seems to first exclude tense / lax from the domain of "vowel height". So is [tense] (alternatively [ATR]) a vowel height feature, or is it something else? On what basis could we decide?

One (theoretical, but practically unworkable) approach is to try to answer based on physical production. The idea is to identify a physical correlate of "vowel height", either (horizontal) tongue position on some coordinate system, or (first) formant frequency, and ask if any language has more than three consistent physical clusters. The easy answer is "Yes, way more than three". Once you add in "controlling all other factors" (vertical tongue movement, F2 value), then the answer becomes "not even three". The IPA vowel trapezoid embodies this fact at least for front vowels. There has been a long-standing search for a purely physical classification of vowels, which simply has not panned out. The groupings do not leap out from physical measurements, instead, grouping may be surmised from analysis of the phonology of a language, and post hoc one may look for phonetic correlates of those phonological groupings.

The implied separation of "vowel height" from some orthogonal feature like "tense" presumes that certain features "act together" in a way that significantly excludes another feature. This makes no sense in the SPE view of feature organization, but is embodied in hierarchical theories of features via the notion of feature constituency – that there is a node "vowel height" which contains some specific features (hi, low), and that an orthogonal feature ("tense" or "ATR") outside of that constituent. In that framework ("feature geometry") you question then is potentially answerable. The question is: does any language have more than three phonologically contrastive arrangements of features under the Vowel Height node. We can answer the "under the Vowel Height node" using classic constituency arguments.

Constituency arguments are classically used to argue that in some languages like Akan or Kalenjin, the distinction [i e] vs [ɪ ɛ] is governed by a feature ATR which is outside of the Vowel Height constituent. The argument is based on the vowel harmony rule, where /ɪ ɛ ʊ ɔ/ become [i e u o] when an adjacent syllable contains /i e u o/. The situation where /ɪ ʊ/ become [i u] next to /i u/ doesn't tell you anything, because it can either be seen as complete assimilation of vowel height, or as partial assimilation of ATR. The compelling argument for ATR is the fact that /ɪ ʊ/ become [i u] (becoming physically higher) next to /e o/, which are physically lower.

This is the classic argument that ATR is not a vowel height feature. Well, what's sauce for the goose is sauce for the gander. Now we can turn to languages like Matumbi and Gbe which have multi-feature height assimilation, where for example /ɪ/ becomes [i] next to [i,u], [ɪ] next to /ɪ ʊ/ and [ɛ] next to /ɛ ɔ/; or /ɛ/ becomes [e] next to /e o/ and [i] next to [i u]. There are also languages where [i ɪ] or [e ɛ] is phonologically a vowel height distinction.

Parkinson (1996) undertakes an extensive survey of vowel height distinctions in phonology, ultimately arguing (analogous to one proposed by Nick Clements) for a theory using a single recursive feature, thus allowing in principle any number of vowel heights to be represented. The most compelling case for multiple vowel heights comes from Sesotho, which is claimed to have up to six vowel heights (one of which exists only as the result of an allophonic rule, and one of which is limited to the vowel [a] which is the sole vowel where backness and roundness are not correlated). The argument for the height hierarchy [i ɪ e ɛ] is based on the vowel harmony pattern, which can be stated as "a vowel (other than /a/) raises one degree when followed by a higher vowel", thus /ɛ/ → [e] before /i, ɪ/, but /ɪ/ only raises before /i/, and not /e/ (which it does under the ATR pattern of Akan). Since /ɪ/ both triggers raising and undergoes it, an ATR analysis is not viable.

In other words, Sotho is the base case argument for more than 3 phonological heights.

  • I should have mentioned that the question arose as I was reading Clements' "Hierarchical representation of vowel height", where he points out that e.g. French was successfully analyzed using three heights and tenseness; but then lays out the argument that "tense" doesn't clearly correspond to height, while the distinction is equally clearly about height, and so this is not a satisfactory analysis. Which is clear, but I got curious - if we ignore this argument about tenseness, and just accept it as a convenient fiction, is it sufficient (for contrast) at least for European languages? May 17 at 18:14
  • So I guess the question could be rephrased as: are the 6 degrees effectively provided by the model with [±high] [±low] [±tense] sufficient? May 17 at 18:18
  • 1
    In the sense of "being able to represent phonemic distinctions". Sotho is a significant problem in terms of being able to formalize the rule, since no feature corresponds to "one step higher".
    – user6726
    May 17 at 20:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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