The IPA's current tone system can show five different tone levels, and any contours formed from them.

Is there any language for which this is insufficient? In other words, is there any (known, natural, attested) language which uses six or more phonemically-distinct tone levels?

  • The Lanna language traditionally spoken in northern Thailand has six tones but is not widely spoken today.
    – user23078
    Aug 13, 2019 at 5:06
  • 1
    It also depends what counts as "sufficient" - a description like high rising is actually pretty vague. If you only knew the IPA system, you would still be guessing, whereas if you know the language, the IPA system is no more help than a label like H or R, or a spelling which encodes the tone.
    – user23078
    Aug 13, 2019 at 5:13
  • There is audio of the six tones here: en.wikipedia.org/wiki/Northern_Thai_language While the diagrams show two high falling tones, I don't think the third tone actually does fall, in which case the IPA notation could approximate all six tones.
    – user23078
    Aug 13, 2019 at 5:31

3 Answers 3


I don't think there are any attested languages that require more than five (5) phonemic levels of pitch to describe. However, there is one language Cori with six (6) surface pitch realizations, although it can be analyzed with a tone inventory of three level tones. I'm looking for some more non-Wikipedia sources and a better explanation of the situation in Cori, but here's what I have so far.

Bench has five level tones and one rising tone 23. Here's a source that Wikipedia cites.

Cori can be described as having six phonemic levels of pitch, apparently. However, 2 is an allotone of 4 and 3 and 5 are surface realization of sequences of tones. So, a minimal analysis only needs a three tone inventory.

  • Perfect; this is exactly what I was looking for! Bench, if I understand right, can be represented unambiguously by the IPA (with the tone letters for 1, 2, 3, 4, 5, 23), but Cori seems to fit my criteria perfectly. If you can find any other sources for that claim, I'll happily accept.
    – Draconis
    Aug 14, 2019 at 0:44

The situation in Chori exemplifies a widespread problem with discussions of allophony, that "allophone" has different meanings. The classic definition of allophone is that two phonemes (or, tonemes) are allophones iff the variants appear in surface-complementary environments. This is not the case in Chori: all 6 tones are surface-contrastive. The alternative definition is "you could remove some of the tones from underlying forms by applying some set of rules". The data-source, Dihoff, pursues a toneme-minimizing analysis where 1→3/_6, 4→2/_1 and 2→5/_6. The particular rules are well-enough motivated since they apply in phrasal contexts, but the underlying representations are not always motivated. Dihoff notes that these rules may be limited to certain grammatical constructions, for example in the present tense, /1/ becomes [6] after a 3s subject and /4,6/ become [3] after a 1s subject. This analysis may be technically tenable if you decompose any [3] into an abstract sequence /1+6/, but this is entirely contrary to the spirit of the traditional concept "allophone" which is based on distribution in surface forms, not underlying forms. Dihoff's analysis only assumes that you can remove tones /2,3,5/ from lexical entries, it does not assume that you can remove those tones from derived representations, including the surface form. In fact, his discussion starts with a tone-minimizing analysis with just /1,4,6/ but morphs into a full 6-level analysis once he gets to talking about tense-inflection and the like.

IPA allows concatenation of level markers, using either accents or bar-markers, so that contours are irrelevant – you can have rising, falling, high-rising and so on as contours. The only thing that would pose a problem for IPA is a 6th level. As it happens, IPA sort of failed in its remit to specify standard letter-shapes for contours. Given 5 levels, there are very many contour tones than can be created (and that are attested) such as extra-high-to-extra-low vs. extra-high-to-mid vs high-to-extra low etc. But they only provide 5 letters for tone contours, and they leave it to the author to come up with a clever way to indicate more subtle contours. This means that authors dealing with language that have many levels and many contours formed from those levels have to operate outside the IPA system, and use one of the two (mutually incompatible) non-IPA numeric-superscript notational systems.

  • How did you access Dihoff's dissertation? I found this document which cites it. Aug 13, 2019 at 18:19
  • It's available from Proquest (paywall, I'm afraid).
    – user6726
    Aug 13, 2019 at 19:20
  • "But they only provide 5 letters for tone contours, and they leave it to the author to come up with a clever way to indicate more subtle contours." I'm not sure I understand you correctly: surely the IPA system can indicate any contour that exists within the five-level system (such as, if necessary, distinctions between 54, 53, 52, and 51)?
    – Draconis
    Aug 14, 2019 at 0:25
  • There is no pre-designed accent or tone-bar for 54 (for example). For accents, there is no H-to-Extra-Hi accent combination and it is not at all obvious what it would look like. Since the would-be characters are not part of the official IPA list, if you make up a letter (stick it in private space), a reader cannot rely on the unicode value to tell them what the value is supposed to be.
    – user6726
    Aug 14, 2019 at 0:35
  • 1
    @user6726 Surely there is, though? I'd write 54 as /a˥˦/ using the tone-bar system. (If you have a font that supports it, that should display as a single bar, not as two.)
    – Draconis
    Aug 14, 2019 at 4:09

I don't see Vietnamese nặng (low glottalised) or ngã (rising broken) among the IPA symbols. Vietnamese tone pitches and contours cannot be separated from phonation type or manner of articulation of final consonants, so classification based on pitch is not adequate. See Vietnamese Tone: A New Analysis (Pham, 2004).

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.