LINGUIST List 9.873

Sat Jun 13 1998

Sum: Rising and Falling Diphthongs

Editor for this issue: Martin Jacobsen <>


  • Chris Golston, Sum: Rising and Falling Diphthongs

    Message 1: Sum: Rising and Falling Diphthongs

    Date: Fri, 12 Jun 1998 10:15:39 -0800
    From: Chris Golston <>
    Subject: Sum: Rising and Falling Diphthongs


    Summary: Rising and Falling Diphthongs

    I posted a query to the Linguist List last fall as follows: 'Does anyone know of a language that contrasts rising [ia, ua] and falling [ai, au] diphthongs?' The following, to whom huge sums of thanks are due, replied:

    Jerry McMenamin (Spanish) Patricia Donegan (Khmer, Frisian) Markus Hiller (Swabian, MHG) Asya Pereltsvaig (Russian) Rob Pensalfini (Jingulu) Vivian Franzen (Rumanian) Shobhana Chelliah (Meithei) Paul Boersma (Geleen Limburgian) David Parkinson (Inuktitut) Sean Jensen (Mandarin) Antony Dubach Green (Irish) Philip Franz Seitz (Vietnamese) Z.S. Bond (Latvian) Thomas Becker (Middle High German) Dom Watt (German, Tyneside English) Horst Simon (Bavarian German) Vivian Franzen (Romanian) Vincent Jenkins (Maori) Peter T. Daniels (English) Mark Mandel (Spanish) James L. Fidelholtz (Spanish) Stan Whitley (Spanish)

    I hope I haven't left anyone out. Here's a general summary of what's going on:

    Diphthongs are much less common than monophthongs and most languages lack them entirely (e.g. Tuscarora, Williams 1974); they occur in only a third or so of the languages of the world (Lindau, Norlin & Svantesson 1985). With very few exceptions, no language has all the diphthongs one could get by arbitrarily pairing all the vowels of the language. Consider the exponential problem of complex nuclei: a language with five vowels and branching nuclei should have thirty distinct nuclei, all things being equal: 5 short vowels, 5 long vowels and 20 diphthongs. There are of course languages with large diphthong inventories, but it is very rare to have a language in which the number of branching nuclei is the square of the number of simple nuclei. Part of this is due to sonority dispersion, of course: if low vowels are more sonorous than mid vowels which are more sonorous than high vowels languages should prefer [ai] to [ia] and [ou] to [uo]. And one sometimes reads in the literature that languages do not contrast rising and falling diphthongs: 'languages have on or the other type of diphthong, but not both at the same time' (Rubach 1998). But this is not the case, as we will now see.

    A number of languages contrast rising and falling 'diphthongs' across what appear to be syllable boundaries: [ka.i] vs. [ki.a] and the like. This probably includes Maori (Jenkins, p.c.) and Jingulu (Pensalfini, p.c.). I leave such cases aside and concentrate on true (tautosyllabic) diphthongs. A number of languages have what might be called orthographic diphthongs: languages in which the first (or last) member of a diphthong is always a high vowel and patterns like a glide. Spanish provides a clear case, as the orthography uses the symbols <i> and <u> interchangeably for [i, j] and [u, w]:

    (1) Spanish cuando [] 'when' causa [] 'cause'

    We can see that this is merely orthographic by considering rhyme. If the <ua> in cuando is all in the nucleus it should rhyme only with other words that contain <ua>; if the sequence is actually [wa], with [w] in the onset and [a] in the nucleus, it should rhyme with words that contain just [a] in the nucleus. The facts here are quite clear: cuando rhymes with Armando, showing that orthographic [u] is actually a glide [w] in the onset. The stress-facts of Spanish are somewhat problematic but it is probably safe to assume that post-vocalic hi voicoids like the [w] in causa are gllides in the coda (Harris 1995). If causa has a simple nucleus [a] and cuando does too, there is no contrast between [au] and [ua] in the rhyme in Spanish: [wa] spans the onset and rhyme, [aw] is contained within the rhyme.

    Similar arguments obtain for Romanian, which has orthographic diphthongs that are best treated as Glide + Vowel (Mallinson 1986, 339); for Russian, where [ia, ai, iu, ju, io, oi, ie, ei] are best treated as [ja, aj, ju, uj, jo, oj, je, ej] (Asya Pereltsvaig, p.c.); for Chinese languages such as Mandarin and Cantonese, where prenuclear high vowels are best treated as part of the onset (Duanmu 1990); and for Vietnamese, where prevocalic [u] is best analyzed as a property of the preceding onset (Philip Franz Seitz, p.c.).

    Pike & Pike's (1947) analysis of Huautla Mazatec is another good instance of orthographic diphthongs, although here the orthography is IPA. They posit the following diphthongs, noting that (i) they are of the same phonetic length as monophthongs and (ii) there are no diphthongs that begin with [e]:

    (2) Mazatec diphthongs (omitting contrastive nasalization and breathiness) ae ao ai oe oi ie oa ia io

    Three problem pairs arise: [oa, ao; ai, ia; oi, io], all of which are featurally identical and yet contrast. But Golston & Kehrein (1998) have argued that all putative complex nuclei in Mazatec are glide-vowel sequences, with labial (w), palatal (j) and velar () on-glides in the preceding onset. The reanalysis accounts for the peculiar facts about Mazatec 'diphthongs' already noted by Pike & Pike: (i) they are no phonetically no longer than monophthongs (because they are monophthongs) and (ii) they never begin with [e] (because there is no corresponding glide for [e] distinct from [j]). On this analysis Huautla nuclei contain a simple short vowel (a, e, i, o) with or without tone, nasality, breathy voice or creaky voice.

    Languages like Kurdish admit of plausible reanalysis for a different reason. In these languae all diphthongs end in high vowels [i] and [u]:

    (3) Kurdish (Abdulla & McCarus 1967) ai au ei eu oi ui i u

    Analyzing [i] as [j] and [u] as [w] is harmless and accounts for the otherwise puzzling fact that diphthongs never end in mid or low vowels.

    I hasten to add that little hinges on such reanalysis in most cases: no language to my knowledge contrasts [ai] with [aj] or [au] with [aw], so interpreting VV sequences as VG has little consequence.

    Some languages admit of reanalysis because of peculiar gaps that appear given published descriptions. Consider Acoma (Miller 1966), the diphthongs of which include the following:

    (4) Acoma (Miller 1966)

    ai au ei eo iu Ii (I= barred i) ui

    Curiously absent from Acoma is [oi], a very common diphthong in other languages. If the diphthong [ui] is a misphonemicization of [oi], Acoma can neatly be reanalyzed as follows:

    (5) Acoma reanalyzed ai au ei eo oi iu Ii

    A similar case comes from !Xu (Maddieson 1984) which seems to contrast [ao] and [oa]:

    (6) !Xu diphthongs (omitting contrastive nasalization and pharyngealization) ae ao ei eu oe oi ia ui oa

    Conspicuously absent is the diphthong [ua], alongside extant [ia]. If we analyze [oa] as [ua] mirror-image diphthongs disappear from the language altogether.

    There are in fact a number of langugages that have mirror-image diphthongs that are NOT susceptible to the type of reanalysis that Acoma and !Xu) are susceptible to. These include Greleen Limburgian, Inuktitut, Irish, Khmer, Latvian, Meithei, Swabian, Vietnamese and White Hmong. Consider the diphthongs of White Hmong:

    (7) White Hmong (Smalley et al. 1990) ai aI au (I= barred i) ia ua

    The phonetic description that Smalley et al provide makes it clear that the bottom row of diphthongs in (7) is somewhat centralizing: 'ia roughly similar to see a (cat)... ua roughly similar to sue a (man).' This is confirmed in an acoustic study by Jones (1998), who finds the F2-F1 endpoint for [Ai] is about 170 Hrz. further back than monophthongal [i] and that the endpoint for [iA] is about 120 Hz. further front than monophthongal [A]. Let us call the centralization of the second part of a diphthong a curtailed trajectory. It is tempting to reanalyze diphthongs with rising sonority as eg, [i] and [u], but the contrasts available to Hmong monophthongs make this an unlikely solution: Hmong has only six monophthongs, [a, e, O, i, I, u], lacking a schwa entirely.

    Curtailed trajectories are very common in languages with mirror-image diphthongs. Consider Swabian. Standard sources (Frey 1975, Russ 1990) include the [ai, ia, au, ua], but according to Markus Hiller (p.c., Geumann and Hiller 1996), the pairs [au, ua] and [ai, ia] are not phonetic mirror images. Instead, the end-point of the rising diphthongs [ua, ia] is central, not low like the beginning of the falling diphthongs [au, ai]. The same seems to be the case for related languages like Frisian (Cohen et al. 1961) and Middle High German.

    A number of other languages with mirror-image diphthongs are surprisingly similar: Vietnamese has /ai, ia/ but they are realized as [ai, i] (Philip Franz Seitz, p.c.); Latvian (Z.S. Bond, p.c.) and Irish (Antony Dubach Green, p.c.) both realize /ai, ia, au, ua/ as [ai, i; au, u]; and Greleen Limburgian curtails the initial vowel of the diphthong realizing /iQ, Qi, ua, au/ as [IQ, Qi; Ua, au] (Paul Boersma, p.c.).

    As we have seen, a number of languages seem to have mirror-image diphthongs. Many of these cases are better analyzed as containing GV or VG sequences with the glide in the onset or coda. But a number of other languages have mirror-image diphthongs that are harder to reanalyze. Many of these have a clear phonetic difference between diphthongs with relatively canonical members and mirror-image diphthongs with one centralized member, i.e. with curtailed trajectories. This might be taken as evidence that what appear to be mirror-image diphthongs are in fact featurally distinct since the centering present in one diphthong is absent in its mirror-image. Future research will have to decide the issue.


    Chelliah, Shobhana. 1997. A Grammar of Meithei. Mouton de Gruyter.

    Cohen, A. C. L. Ebeling, K. Fokkema, M.W.S. DeSilva. 1961. Fonologie van het Nederlands en het Fries. The Hague: Nijhoff.

    Duanmu, San. 1990. A formal study of syllable, tone, stress and domain in Chinese languages. PhD dissertation, MIT.

    Frey, Eberhard. 1975. Stuttgarter Schwaebisch. Marburg: Elwert.

    Geumann, Anja, and Markus Hiller. 1996. Diphthong dynamics in Swabian. Paper presented at the 3rd ASA/ASJ joint meeting.

    Golston, Chris and Wolfgang Kehrein. 1998. Mazatec Onsets and Nuclei. CLS 33.

    Harris, James. 1995. Projection and ewdge marking in the computation of stress in Spanish. In John Goldsmith (ed.), The Handbook of Phonological Theory. 867-887.

    Huffman, Frank. Introduction to Modern Cambodian. 197x.

    Jones, Rosemary. 1998. Distinction between canonical and noncanonical vowels in White Hmong. Ms, California State University, Fresno.

    Lindau, Mona, Kjell Norlin and Jan-Olaf Svantesson. 1985. Cross-linguistic differences in diphthongs. UCLA Working Papers in Phonetics 61, 40-44.

    Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press.

    Mallinson, Graham. 1986. Rumanian. London: Croom Helm.

    Miller, Wick R. 1966. Acoma grammar and texts. U.C. Publications in Linguistics 40. University of California Press, Berkeley and Los Angeles.

    Pensalfini, Rob. 199x. Jingulu grammar, dictionary, and texts. Doctoral dissertation, MIT. MIT Working Papers in Linguistics.

    Pike, Kenneth L. and Eunice Pike. 1947. Immediate constituents of Mazatec syllables. IJAL 13:78-91.

    Rischell, Jrgen. 1974. Topics in West Greenlandic Phonology. Copenhagen: Akademisk Forlag.

    Rubach, Jerzy. 1998. A Slovak argument for the onset rhyme distinction. Recent Linguistic Inquiry, 168ff.

    Russ, Charles V. J. 1990. Swabian. In Charles V. J. Russ (ed.), The dialects of Modern German. London: Routledge. 337-363.

    Seitz, Philip Franz. 1986. Relationships between tones and segments in Vietnamese. PhD dissertation, University of Pennsylvania.

    Smalley, William A et xxxx. 1990. The mother of writing. xxxx

    Wiesinger, Peter. 1989. The central and southern Bavarian dialects in Bavaria and Austria. In Charles V.J. Russ (ed.), The Dialects of Modern German. A Linguistic Survey. Stanford University Press. 438-519.

    Williams, Marianne Mithun. 1974. A grammar of Tuscarora. Doctoral dissertation. Yale University.

    Selected gems from various of the respondents follow in no particular order:

    Boersma, Paul (p.c., 11.17.97) "While [i] and [Au] have equal F1 contours, their rising counterparts [i] and [UA] do not start at the same height."

    Bond, Z.S. 'Latvian might qualify. We have eg, uiela 'street', laiva 'boat', ala 'egg', lauva 'lion'. The second component of the ie ua diphthongs is phonetically rather obscure, not a clear /a/.'

    Green, Antony Dubach (p.c. 11.17.97) "Irish contrasts [au] with [u] and [ai] with [i]."

    Hiller, Markus (p.c., 11.21.97) "The Swabian opening and closing diphthongs are NOT mirror images of each other (even though Frey's and Russ's transcriptions look like that). The same goes, by the way, for Middle High German, although orthography at first glance seems to suggest otherwise..."

    Jenkins, Vincent (p.c. 11.18.97) "Maori has the following pairs, but you may want to call them separate syllables rather than diphthongs..."

    Seitz, Philip Franz (p.c. 11.18.97) "Roughly, Vietnamese contrasts upgliding and ingliding diphthongs. However,the ingliding diphthongs can be analyzed as underlying long vowels (and they are realized with little or no inglide in some contexts) in some cases and as a combination of a rounded initial consonant plus plain vowel in the rest of the cases. Here are some minimal-pair examples in VIQR ("Vietnamese Quoted Readable," the representation of VN orthography used on the Internet and in telegrams). In the phonetic transcriptions, "x" stands for schwa (mid-central unrounded short vowel) and "S" stands for the alveopalatal voiceless sibilant. Tones are not indicated in the transcriptions; in VIQR, "'" is the high rising tone, "`" the low falling tone, and no extra mark is the high level tones (all examples have one of these three tones out of the five or six VN tones). 1. [a:u], [au], and [xu] contrasting with [wa], where the rounded element can be analyzed as a property of the initial consonant.

    1a. tao [ta:u] 'I (arrogant)' toa [twa] 'a prescription for medicine' or 'a car of a train'

    1b. ta`u [tau] 'ship' or 'Chinese' toa` [twa] 'classifier for large building or court'

    3. [a:i], [ai], and [i] contrasting with [i], where [i] can be analyzed as an allophone of long /i/ (/i:/).

    3a. chai [tSa:i] 'bottle' chia [tSi] 'to separate'

    - Note here, e.g., chie^'n [tSin] ~[tSi:n] 'struggle': syllables with final consonants, especially with the high rising tone, show variation between an ingliding diphthong and a long vowel (cf. 2a).

    3b. cay [kai] 'piquant' kia [ki] 'over there'

    3c. cay [ki] 'classifier for stick-like objects' kia [ki] 'over there'

    Department of Linguistics, M/S 92 California State University Fresno Fresno CA 93740

    Office: (209) 278-4895 Home: (209) 292-7049