Chuvash.Org :: Printable Version :: On Chuvash Stress

    Dobrovolsky, Michael (The University of Calgary) (1990). On Chuvash Stress. Altaica Osloensia. Proceedings from the 32nd Meeting of the Permanent International Altaistic Conference. (p. 113- 124). Oslo: Universitetsforlaget.
On Chuvash Stress
There is something fascinating about science.
    One gets such wholesale returns of conjecture
    out of such a trifling investment of fact.
    Mark Twain

    One of the more fascinating aspects of recent work in linguistics is the proliferation of theoretical models that attempt to deal with stress placement in the world's languages within a (broadly speaking) generative framework. The general rule writing conventions established in early generative models were first applied to both segmental and suprasegmental phenomena. Since then there has arisen a progressively independent treatment of primitives, notation, and rule writing conventions.
    An intresting subhistory in the development of these models languages whose stress patterns (or is claimed to pattern) like that of Chuvash. It is my contention in this paper that three recent models of stress placement have failed to describe Chuvash stress adequately. This is especially surprising in view of the fact that straighforward manner. The stress pattern of literary Chuvash (as it assignment in this paper) is determined by the presence or absence 'reduced' (also described as 'regular length' and 'short'). If a word has only reduced vowels, stress falls on the first vowel.
    Stress placement in the literary dialect is largely agreed on by a range of scholars, including Rezjukov 1959, Andreev 1961, Egorov 1971, Sergeev 1969, and Zahemszky 1982, whose interpretation of the metrics of Ivanov's folk poetry supports the standard analysis. However, there is also a tradition of describing secondary stresses in Chvuash that has been ignored in recent formal models. I will return to this question in the course fo the paper.
(1) 	Reduced vowel words (all stressed on first vowel) 
	kắndăr		'south' 
	šắpčăk		'nightingale' 
	nĕ´rsĕrlĕh	'уродливость' 

	Mixed V words (stress on last full vowel) 
	sírĕm		'20' 
	símĕs		'зеленый' 
	simĕslé		'в зеленый' 
	xĕvél		'sun' 
	vărman		'forest' 
	xăramázăr	'without fearing' 
	timĕrĕsém	'3p's plows' 

	Full V words (stress on last full vowel) 
	yál		'village' 
	yaltá		'village+Loc' 
	xirkelén	'упрямиться' 
	krassɨnlá	'пахать' 

	Ivanov's metrics, after Zahemszky 1982 
	 *   . .  *  .   .  * 
	Yal vĕśĕnči sĕm vărman		'The dark forest near the village'
	 * .  .   *   . . * 
	tĕmĕnšĕn pit yanărat		'Is all astir for some reason'

    Kiparsky 1972 deals with Cheremiss stress which is claimed to be identical to that of literarary Chuvash. His rule employs parenthesis notation (disjunctive ordering), which he claims to be an insightful advance in theoretical notation that permits us to join in one rule two related parts of a single process.
	(2) V → [+acc] /       [CøVCø]ø#

    Several implications emerge from this rule. One is the claim that the first syllable stress (I'll call it 'Left margin stress', LSTS for short) is the default or elsewhere option of the first stress rule. Second is the fact that the formalism requires the rule to work 'backwards', scanning the word from right-to-left for the appropriate vowel to stress. The unstated inplication here, as I read it, is that the basic stress falls on the left margin and has otherwise been attracted to the last full vowel of the word (in formal terms, to the first full vowel that it encounters). But as the model is a purely formal one, I may be reading too much into it.
    Since the model is a purely formal one, we may be constrained from asking questions that pertain to speaker performance and/or psychological reality. Nonetheless, I note that the right-to-left scanning model seems to me here an unnecessarily complex way of accounting for a speaker's knowledge of phonological content of the word. Furthermore, it is also worth noting that the implicit claim of the formalism itself, namely, that the two stress patterns (or stressing processes, if you will) are in an 'elsewhere' relationship, is questinable. There is no inherent reason why one could not write a rule that claims that in reduced vowel words, stress is initial, and elsewhere, the last full vowel of a word is stressed. This issue seems to me particulary significant in light of the fact that Kiparsky has to make a strong assumption to get his formalism to work: the notation 'V' has to do double duty as representing only a full vowel during the application of the first part of the rule, and as representing any use of the 'V' notation, it seems to me, falsifies the claim that the two parts of the rule havɨe been united into a single formal representation.
    Hyman 1985 provides a formal solution to Chuvash stress assignment within a theory of syllable weight, which employs a putative universal theoretical primitive he labels a weight unit (WU). He proposes that full Vs in Chuvash contain 1 WU, but reduced Vs none. Given the appropriate formal statement of RSTS, stress falls 'naturally' on a vowel with a weight unit.
	(3) X → [+stress] /          Q ## (where Q does not contain x)

    Hyman's approach reflects neatly the apparent phonetic fact that in Chuvash, full vowels are ordinary vowels in terms of length while reduced vowels are reduced in length, but not necessarily in quality (Bereczki 1979). Hyman (p. 59): '... I propose that reduced vowels are weightless and that full vowels have one WU, i.e. they are regular, single length vowels.'
    Unlike Kiparsky's formalism, Hyman's does not appear to suggest explicitly that LSTS on reduced vowel words is the default option. The difficulty with Hyman's approach, as pointed out in Odden 1986, is that it does not account for LSTS on reduced V words as naturally as it accounts for full V stress. The stress that falls on the initial reduced V of a reduced V word must be accounted for my other formal means. Hyman claims (p. 60) that Halle and Vergnaud's 1980 'Q-variable notation' will handle what appears to be the elsewhere case of LSTS on all-reduced vowel words. Even if this works (and Odden disputes that it will), the effect is still to make a distinction between the formal processes that generate LSTS and  RSTS. Here, then, we have the implicit suggestion that full V stressing is different from the more generally stateable LSTS on reduced V words. Thist suggests to med that LSTS of reduced vowel words is in some meaningful way distinct from the stressing of full vowel words.
    Halle and Vergnaud 1987 provide an analysis of Chuvash stress within another universal theory of stress assignment. Their approach, growing out of the grid-based analysis of Liberman and Prince 1977, Selkirk 1983, and Prince 1983, accounts for stress placement though (a) the creation of stress-related metrical constituents, and (b) the assignment of stress to (in the overwhelming majority of cases) the margins of these constituents at (c) distinct metrical levels. In the case of Chuvash, (assumed on the basis of secondary sources not to have secondary stress, though in a footnote they allow that it might), they must resort to a device they call line conflation to remove unwanted secondary stresses generated by their model.
	(4) 	Assign line 1 asterisks to full vowels (underline = full V) 
		Line Ø: [+HT, - BND, left]
		Locate heads of line Ø constituents on line 1
		Line 1: [+HT,  -BND, right]
		Locate the heads of line 1 constituents on line 2

		Word with full vowels
		 . .   . . . .   *		line 2
		(* .   * . . .   * )		line 1
		(1 2) (3 4 5 6) (7 8 9)		line Ø

		Word with all reduced vowels
		 * . . . . . . . 		line 2
		(*)				line 1
		(1 2 3 4 5 6 7 8 )		line Ø

		Line conflation (removes unwanted secondary stresses)
		 . .   . . . .   *		line 2
		(. .   . . . .   * . .)		line 1	
		(1 2) (3 4 5 6) (7 8 9)		line Ø

    Particularly appealing about this approach is the explicit manner in which stress i assigned to margins. Since full Vs are projected upwards to a level at which only they are represented, the RSTS on full V words is easily represented as a margin stressing rule. This captures  formally any implicit speaker knowledge more directly than the original Kiparsky formulation with its right-to-left scanning [though H/V do employ right-to-left scanning in other stress systems they examine]. Again, the LSTS of reduced vowel words has a kind of default status, in that there is less formal machinery required to account for it.
    However, there are formal and eventually empirical problems. Formally, both the RSTS on full vowel words and the LSTS on reduced vowel words emerge as rightstressing, since the same constituent structure, [+HT, -BND, right], is invoked to account for both the stress of the rightmost full vowel and the first (or leftmost) vowel of the word. Is this an insightful discovery or an artifact of the formal model? Furthermore, the device of line conflation emerges as a piece of repair machinery. It would be simpler to set a parameter from the outset that excludes secondary stresses.
    At any rate, this question must be reserved for those languages that truly show no secondary stress. We must now being to attempt to account for secondary stresses in Chuvash. The following examples of secondary stress (or, in some cases, second stresses, since there is virtually no difference between the two) are taken from tapes of two native speakers of Chuvash made by Houvdhaugen in 1971.
    I consider the term 'stress' to mean a phonological mark on certain vowels or syllables. In the Chuvash I have heard, these marks ar phonetically manifested primarly by pitch changes. The first of them is typically a high pitch (H) and the second is preceeded by a sharp downward drop in pitch (L). After a H, the pitch often remains high, or begins slowly dropping to a lower level until the sharp drop onto the L. (This sound like a description of Turkish word-pitch, but there are differences. In Turkish, the first H often stays at nearly the same pitch level until the low stress-pitch is reached. Turkish syllables are also more of equal length; the full/reduced vowel contrasts of Chuvash given the language a very different sound than that of Turkish, even though the phonological description of word-pitch is very similar.)
    The forms cited here do not imply that these secondary stresses are obligatory. They appear to be determined, as are secondary stresses in many languages, by speech rate. It is still to be expected that they will be generated by the formal stressing algorhythm. It should also be noted that these secondary stresses are not conveniently describable as occuring only on words in absolute sentence initial or absolute sentence final position, where they could easily be attributed to intonation overlay.
	(5) 	Group 1			Group 2
		a. sápălsá [sic ]	a. xùskatakán
		b. sùxalánːă		b. túnisenè [ túnz'n'ɛ]
		c. yàlkassá		c. súlsinené [ sic ]
		d. káyăksém [ sic ]	d. sùlsín
		e. xàllĕxé		e. pùldarɨmás
		f. sàvăndáx		f. t`ĕlĕnterét
		g. šùpaškár		g. mằndărĕpé
		h. yòpleméš		h. sằnːí
					i. togátmăšằ
					j. vărmánsené [ sic ]

    If we consider the secondary stresses that are generated and alter removed by the H/V model, they occur incorrectly on the full vowels of full V words. The model does nicely place an asterisk on the initial (full) vowel of these words, but line conflation removes it, too.
	(6) H/V as a secondary stressing device
		         *	line 2 (incorrect form predicted with line conflation – 
					no 2ndary stress
	( *    *     *   * )	line 1 (incorrect form predicted without line conflation)
	 (*)  (*)   (*) (*)	line Ø
	x ù s k a  t a g á n	[xùskatagán]

    On words with all reduced vowels, no secondary stress is generated. In words with full vowels, without line conflation, too many stresses are generated; with line conflation, too few. Clearly some repair work is required. Data from Group 1 appears to suggest that left-headed binary constituents apply from L to R (alternating stress), but the data from Group 2 show that this is not the case. Group 2 forms show that two vowels may intervene between stresses, and in some cases two succeding vowels are stressed. (The deletion of a full vowel  in tún[i]senè (where, incidentally, the first stress is primary, which I believe is here due to contextual emphasis) is best understood as occuring after the stress placement and dependent on it.)
    What emerges from the data is that we require the initial stress that is removed through conflation. This leads us to the central claim of this paper - that there is not one but two types of stress placement at work in Chuvash (and possibly other languages with analogous stress placement). As I have suggested, all three of the formal proposals reviewed here have not been able to escape this implification, even though they have attempted, for the best theoretical reasons possible, to unify the solutions.
    These two stress types are of an essentially different character on the functional plane. One, the 'pure' margin stressing (be it LSTS or RSTS in different dialects of Chuvash), has a more direct link with word structure in that i explicity marks word boundaries by making the first (or last) syllable of the word. Not uncoincidentally, as we have seen, this word-margin stressing occurs as basic in words provide no obvious or inherent phonetic reason for stress to fall other than at the word margin. At certain speech rates, this type of stressing applies irrespective of the phonetic content of the rest of the word.
    The alternative type of stressing is affected by the natural phonetics of stress - attraction to a full (= 'unreduced', 'longer', or 'WU'd') vowel. Significantly for a universal theory of stress, this system still marks margins - the margin of the full vowel domain in a word - and thus also serves, although less directly than word-margin stressing, as a word-boundary indicator. The formal expression of this type of system is particularly well served by a model that assigns stress to the margins of constituents at different metrical levels. I propose that these two types of stressing are better expressed as distinct by representing their action as occuring on different planes (cf. other uses of planar representation as a formal device). We then overlay Word and Full-vowel plane, where the term 'overlay' implies coexisting stress (since the planes are assumed to be stress-transparent) and not the deletion of stress, except those stresses from the full vowel plane that have already been superceded by the RSTS rule on L2.
    Biplanar representation of Chuvash stress
    Since reduced vowel words have blank full vowel planes, overlay either need not occur or can be assumed to occur vacuously.
    Forms (5) 2i- 2j., are included to show that there are still problems to be addressed. In (5) 2i., a secondary stress falls on the last vowel of the word. In (5) 2j., a second stress occurs on the last vowel of the stem (vărman 'forest') as well as on the last full vowel of the full word, offering a tantalizing possibility of a cyclic approach to stress. At the same time, it should be noted that both of these apparently odd stresses occur on heavy syllables. More data is needed to deal with such issues.
    Of course, the H/V model can doubtless be made to produce the required results for the majority of forms though constituent analysis. But more important than reworking H/V to this paper is the claim that each component of the stress mechanism presented is functionally distinct, and so should be split into two parts in the formal representation. The aim of this paper has been to suggest that several formalizations of Chuvash stress that have appeared over the last decade have led to repeated reanalysis because they are mistakenly trying to unify two distinct processes: one process assigns prominence to a word margin; the other assigns prominence to a full vowel at a certain place in the segmental string.
    Currently, linguistic formalisms are seen as capturing the inherent capabilities of language users without necessarily making claims about psychological reality (recent defenses of abstraction in LP by Mohanen and Kiparsky, for example). Thus, when a rule assigns stress by L-> R scanning, it is not a claim that individual speakers of Chuvash scan an unfamiliar word from left to right to assign stress.
    I have suggested here that it is preferable to construct representations that capture more isomorphically the natural processes that underly the phenomena; or, perhaps better put, that represent the natural processes that are the phenomena. The closer we come to this ideal, the more truly explanatory our rules and representations are. Phonology has been trying to move this way for a while, but I don't think that the unified stress rules of Kiparsky of Halle/Vergnaud do this for Chuvash, for the reasons presented in this paper. The very difficulties Hyman appears to encounter also suggest that a formally unified approach to this type of stress system is also incorrect. In short, we are lumping where we should be splitting.
    Formally, Kiparsky's approach assigns stress R->L and can thus be iconically represented as in (8).

    Halle/Vergnaud assign stress by successive selection of the rightmost element at metrical levels (plus conflation), an thus can be schematically represented as in (9).

    This paper suggests that stress assignment in Chuvash consists of two distinctly motivated processes that should not be formally collapsed; schematically, this can be represented as in (10).
	(10) ←  →

    However, the arrows in this representation do not have the same implication as those in (8) and (9). There, the arrows represent the direction of scanning. In (10), the arrows represent the distinction between the two independently motivated processes of prominence assignment. Of course, this representation is not mean to capture the formal process(es) of prominence assignment; this is accomplished, as shown in the paper, by assigning *'s on two distinct planes.
    Let us return to the question of two stress systems. We see in the case of the (Anatri) dialect described by Ašmarin no bifurcation of the stress pattern - RSTS prevails as main stress in both full V and reduced V words. Interestingly, in this dialect, even though main stress falls on a full V, RSTS still falls on the word-final vowel of a word; some of the forms cited by Ašmarin are given below.
	(11)	iléml`ĕ		'красивый' 
		áśăvằn		'твоего отца (род.пад.) 
		xúnằm		'мой тесть'

    Of course we could capture the facts of this system in the H/V framework by reorganizing the line Ø constituents as right-headed. But in doing so, we begin to suspect that the constituents are artifactual, since the only change in the dialect is the presence of a right  margin stress on reduced vowel words that in never reflected in the internal constituent makeup.
    What I think is most significant about the stress pattern of the dialect described by Ašmarin is the implication that the pervasive initial stress of the literary dialect is not just a superficial rhythmic phenomenon, for the RSTS requirement of Ašmarin's dialect results in the secondary stress being not a LSTS but a RSTS - the dialect patterns consistently internally. In the Virjal stress that is the basis of literary Chuvash, the systems have become functionally distinct: reduced vowel words are now more sharply separated from full V words by the RSTS-LSTS split. In the dialects described by Baitchura 1963 and 1983, it may be the case that we are encountering another combinatory variant of the systems. According to B's interpretation of this data, main stress may be found on the word initial syllable in all cases, at least for some speakers. We are left with the following combinations.
	(12) Stress combinations in different Chuvash dialects 

					Literary	Ašmarin		Baitchura
	Mixed and full vowel words	RSTS		RSTS		LSTS (?)
	Reduced vowel words		LSTS		RSTS		LSTS (?)
	Main Stress:			RSTS		LSTS		?


Categories: Chuvash Language

Full version :: Link Article