Labial Harmony in Written Manchu*

The following HTML text was created from the original print version. Although care has been taken to transcribe the original correctly, errors may remain. Please refer to the PDF as the text of record for citation. If you encounter a mistranscription, please report it to the editor ([email protected].)

0. Introduction

Manchu is a Tungusic language (Vovin 1993), one of a number of related languages spoken in northern China and in the former Soviet Union. Written Manchu (WM), also known as Classical Manchu, refers to the language of the Manchu court from the late sixteenth to the early twentieth century (Ard 1984). Since it is a written language, its phonetic system, and hence its phonology, must be reconstructed from the orthography. The correspondence between orthography and phonetics is particularly problematic in WM, and this has misled some students of its phonology. We will provide evidence for a particular interpretation of certain aspects of the WM vowel system.

WM exhibits two different harmony processes. The first process, with which we will not be concerned here, has been interpreted in different ways by different linguists: some are proponents of some version of coronal harmony, involving a distinction between front and back vowels (Vago 1973, Odden 1978, Walker 1993b); some assume instead that it involves relative height (Hayata 1980, Ard 1984); Zhang (1994) proposes an analysis in which the relevant feature is RTR, distinguishing vowels with retracted vs. advanced tongue root. This harmony process has been much discussed, but we will not say more about it here.

We will be concerned with another harmony process in WM, namely labial harmony. With respect to this type of harmony, most writers are in agreement that it involves the feature LABIAL, or ROUND (depending on one's feature system). Nevertheless, our account of labial harmony in WM is new with respect to explaining the conditions under which it occurs.

It has sometimes been assumed that whether a vowel is transparent or opaque to vowel harmony is a consequence of its featural representation, which is determined partially by the contrasts that it enters into (a number of examples of such an account are presented below). WM appears to counterexemplify this assumption in terms of labial harmony. However, we will argue that vowels in WM appear to be opaque to harmony for an independent reason. We will show that labial harmony in WM is subject to a condition which has not been observed before in vowel harmony, to our knowledge. We will further support our analysis by showing that the same condition holds on labial harmony in the modern Tungusic language Oroqen.

1. Labial harmony in Khalkha and Standard Evenki

Before looking at WM, we will begin by considering two other languages, Khalkha Mongolian and Evenki Tungus. According to van der Hulst and Smith (1988), these two form a 'minimal pair' in terms of labial harmony: both languages have a similar process of labial harmony, but differ with respect to what vowels may intervene.

Labial harmony in Khalkha affects only underlyingly low vowels. The round high vowels /u/ and /U/ are opaque, but /i/ is transparent to labial harmony. The transparency of /i/ to labial harmony in Khalkha is shown in (1); the suffix -AAs rounds to -OOs, where the harmony appears to skip over the intervening high vowel /i/:

(1) /i/ is transparent to labial harmony in Khalkha (van der Hulst & Smith 1988)

Standard Evenki shares the opacity of /u/ to labial harmony. However, unlike Khalkha, /i/ in Evenki is opaque to labial harmony, as shown in (2), where the /i/ of –iglA appears to block the spread of rounding to the suffix vowel /A/:

(2) /i/ is opaque to labial harmony in Evenki (van der Hulst & Smith 1988)

Van der Hulst and Smith explain the different behaviours of /i/ in Khalkha and Evenki with respect to labial harmony in terms of the contrasts of the vowel systems of these two languages. Their vowel systems are shown in (3) and (4), respectively:

(3) Vowel inventory of Khalkha Mongolian (based on van der Hulst & Smith 1988)

(4) Vowel system of Standard Evenki (based on van der Hulst & Smith 1988)

The basic claim of van der Hulst & Smith (1988), translating from their formalism, is that in (3), the positively specified features are Labial, Low, and ATR. /i/ is not Labial or Low, and is not specified for ATR due to its lack of a non-ATR counterpart. /i/ is, therefore, not specified for any relevant feature and so is transparent to labial harmony in Khalkha. In Standard Evenki (4), Front must also be specified, though the ATR value of front and round vowels is predictable. /i/ is Front; since /i/ has a feature, it is opaque to labial harmony in Standard Evenki, according to the theory of van der Hulst & Smith.

However, the contrastive specification of vowel systems alone is not sufficient to explain the opacity of /i/ to labial harmony in WM, which is genetically close and geographically contiguous to Khalkha Mongolian and Evenki Tungus.

2. Labial harmony facts in written Manchu

The vowel inventory of WM is shown in (5):

(5) Vowel inventory of written Manchu (Zhang 1994)

As in Khalkha and Evenki, (6) shows that in WM the trigger and target are both Low in labial harmony:

(6) o-o(-o) in stems (Norman 1978)

As we expect, given (5) and the hypothesis of van der Hulst & Smith, the high vowel /u/ is opaque to labial harmony because it is specified for Labial. This is shown in (7), where rounding does not spread across /u/:

(7) o-u in stems (Norman 1978)[1]

However, we would also expect, given (5), that /i/ ought to be transparent to labial harmony in WM. This is because /i/ is not Labial or Low, and does not need to be specified for ATR, due to the lack of a contrasting non-ATR counterpart. The position of /i/ in the WM inventory (5) is similar to that of /i/ in Khalkha (3) with respect to contrasts; however, it behaves like /i/in Evenki (4) in being opaque to labial harmony, as shown in (8):

(8) o-i in stems (Norman 1978)

The opacity of /i/ to labial harmony in WM is difficult to explain from the point of view of contrastive specifications.

The problem goes further than just the status of /i/. Based on what has been said to now, there is no reason why labial harmony is blocked in (9a-b), where there are no intervening high vowels at all:

(9) a. o-a (oo-a) in stems (Norman 1978)

b. o (oo) in stems (Norman 1978)

Any explanation which attributes the failure of labial harmony in (7) and (8) to the presence of an intervening high vowel will have to account for the failure of harmony in (9) by some other means, perhaps by labeling these forms as exceptions to labial harmony. We shall argue, however, that the forms in (9) are not exceptional, but fall under a simple generalization.

Before proceeding to discover what this generalization is, we need to clarify the phonetic and phonological status of the sound written with double oo. There is a dispute among scholars as to the phonetic value of this symbol and whether it is phonemically a long vowel or a short vowel. Maddison (1984: 283) describes the WM vowel system as having six short vowels and one long vowel, namely, oo. Against this view, Seong (1989), among others, argues that oo is an allographeme of o. Thus, it is very often the case that words with oo can be found to be represented also with o in WM, as seen in (10a):

(10) a. variation of oo and o (Norman 1978; Seong 1989)

b. 'Minimal pairs' (Seong 1989)

Seong (1989) notices that oo occurs mainly in the initial syllable of a word; words with oo in the noninitial syllable are rare, and almost all of them are compounds (e.g. taiboo 'Grand Guardian'). He points out that, just as the words in (10a) are due to orthographic fluctuation, so similarly the apparent minimal pairs in (10b) are not true phonemic minimal pairs. Thus, the word for 'river perch' can be found written with oo and also with o, though the word for 'obedient' is found only with o.; presumably, the contrast found in the last two words in (10b) is accidental, and not phonologically significant.

There is a class of words written only with oo which never varies with o. As pointed out by Seong (1989), these are usually Manchu loan words borrowed from Chinese, in which oo represents a sound pronounced [au] in Chinese, as shown in (11):

(11) oo in words borrowed from Chinese (Norman 1978; Seong 1989)

That oo represents Chinese /au/ is also confirmed by the fact that there is no diphthong /au/ in the diphthong system of WM.[2] It appears, then, that oo is the graphemic representation of the Chinese diphthong /au/ in these borrowed words. Thus, we follow Seong (1989) in regarding oo as an orthographical variant of o, and not as a counterpart of o in length. As we shall see, oo- patterns with a single o and not with a sequence of two short os in labial harmony.

3. Mechanism of labial harmony in written Manchu

Since we have claimed that o and oo are neither phonetically nor phonologically distinctive in terms of vowel length, all the forms in (6)-(9) can be schematically represented in (12), where C stands for any consonant:

(12) schematic representation of labial harmony in WM

As we have seen, labial harmony in WM occurs only in (12a), but never in (12b-e). Next, we will discuss how labial harmony in WM has been explained in previous work, looking in particular at accounts by Odden (1978), Hayata (1980), and Walker (1993b).

Odden (1978) formulates the rule of rounding assimilation shown in (13):

(13) rounding assimilation (Odden 1978)

(a front mid vowel becomes back when preceded immediately by o)

The harmony rule in (13) is problematic in several ways. We note first that it is based on the assumption that the underlying suffix vowel in WM is /e/, an assumption we do not accept. However, we will not enter into this issue here. For even if we accept this assumption, the rule in (13), while accounting for the forms in (12a-c), fails to explain the forms in (12d-e), where no high vowels intervene between the mid vowels.

Walker (1993b) claims that [Labial] is a morpheme-domain feature for low vowels, but a segment-domain feature for high vowels. In other words, low vowels spread the feature [Labial], but high vowels do not.

For example, the underlying representations and the result of labial association of the forms in (12b-c) can be shown using the words bofu-la 'wrap' and gosi-ngga 'affectionate' in (14a-b) and (15a-b), respectively:

(14) a. underlying representation for bofu-la 'wrap'

b. association of Labial

(15) a. underlying representation for gosi-ngga 'affectionate'

b. association of Labial

(14) shows that the morpheme feature [Labial] cannot link to the high round vowel /U/ because the segment already has [Labial] underlyingly. The association of the morpheme feature [Labial] cannot cross /U/ without violating locality. This analysis is similar to that of Van der Hulst & Smith which we discussed earlier, in that the presence of a feature on the high round vowel causes it to be opaque to labial harmony.

What about the high front vowel /i/? Recall that in terms of contrastive specifications, /i/ does not need to be specified for V-place, and this causes a problem for Van der Hulst & Smith's theory, since we would expect /i/ to be transparent to labial harmony. Walker (1993a,b) also adopts a theory of contrastive specification, and thus assumes that /i/ has no V-place node, as shown in (15a). Since it has no V-place node, [Labial] cannot link to /i/. But why then can the harmony not cross over the /i/ to the following low vowel? Walker proposes that in labial harmony, the anchor and the target for Labial association are different: although the target is V-place, the anchor is Vocalic. According to Walker, this is the marked option; the default is that the target and anchor are the same. If the anchor for labial harmony is Vocalic, then [Labial] cannot pass over the /i/ to associate to /E/ because doing so would violate locality, for the high vowel h as a Vocalic anchor.

Walker's analysis offers an account for the forms in (12b-c). However, we note that the decision to make the anchor Vocalic does not follow from any principle, but is a stipulation designed to allow /i/ to be opaque. Also, like Odden's, Walker's analysis does not give any explanation for the forms in (12d-e).

Finally, the labial harmony rule for WM given by Hayata (1980) is shown in (16):

(16) rounding assimilation rule (Hayata 1980)

V—> [+round]/ V C1 V + C1______

[+round]

[+low]

(a suffix vowel is converted into o when preceded by a polysyllabic stem ending in o)

The rule in (16) must be amended in a number of ways. First, the vowel which is the target of the harmony must be specified to be a low vowel. Second, the target vowel is not necessarily a suffix vowel: since WM has stems with three low round vowels, such as in (6) dorolon, foholon, etc., but no stems with two os followed by a, it follows that harmony is obligatory also within a stem in the context of (16). With these revisions, the rule accounts for all the harmony facts represented in (12).

However, the rule lacks explanatory adequacy in a number of ways. First, it simply stipulates that labial harmony can affect only immediately adjacent vowels; thus, Hayata does not deal with the question of why the high vowels are opaque to labial harmony. The second problem is more subtle. As the rule is written, the first V in the environment may be any vowel, and the second vowel must be /o/. In fact, there is no o in the second syllable of a stem in WM unless there is an o in the first syllable. Therefore, the generality of rule (16) is somewhat misleading, and, in our view, actually conceals an important generalization about WM labial harmony.

Given the harmonic and disharmonic forms in (12), it is obvious that labial harmony in WM has something to do with the number of low vowels in a stem. We, therefore, propose the syllable condition on labial harmony in WM given in (17):

(17) syllable condition on labial harmony in written Manchu:

[Labial] has to be linked to two adjacent syllables of low vowels in a morpheme in order to spread.

With (17), all the forms in (12) can be well accounted for. (12a) is expected to exhibit labial harmony in the suffix because [Labial] is linked to two syllables in the stem underlyingly. Similarly, labial harmony should spread to a third stem low vowel. In these cases, labial harmony is obligatory, for no forms of the type (12a') are found. When [Labial] is linked to only one syllable in a stem, it cannot spread to any subsequent syllables, no matter whether high vowels intervene, as in (12b-c), or not, as in (12d-e).

The labial harmony mechanism for some Manchu words is shown in (18)-(20):

(18) a. underlying representation for monggo-ro 'speak Mongolian’[3]

b. Labial spreading

In (18), [Labial] spreads to a low vowel because it is linked to two syllables underlyingly; in (19) and (20), [Labial] cannot spread because it is linked to a single syllable underlyingly:

(19) a. underlying representation for golmi-kan 'rather long'

b. Labial is linked to one syllable and cannot spread; default realization

(20) a. underlying representation for coba-la 'to lift with a lever'

b. Labial is linked to one syllable and cannot spread; default realization

It follows, then, that the failure of harmony in cases (12b) and (12c) is not due to the presence of the high vowel at all. We have thus almost solved the problem of why the vowel /i/ is opaque to labial harmony in WM, even though its position in the vowel system would lead us to expect it to be transparent. The answer is that labial harmony fails before /i/ for an independent reason. This, however, is only a partial answer, because of forms like those in (21):

(21) o-o-i in stems (Norman 1978)

In (21) we list stems which have two round low vowels followed by /i/. When a suffix with a low vowel is added to these stems, the low vowel does not undergo labial harmony, even though [Labial] is associated with two syllables in the stem. These cases show that the high vowel /i/ blocks the spread of labial harmony in WM. Therefore, like some of the other analyses we have discussed, we also have to limit labial harmony so that it does not skip any syllables. Recall that Walker achieved this result by stipulating that the anchor for labial harmony is the feature [Vocalic]. Although we cannot dispense with such a condition, we may now be able to see why it exists. Notice that the condition in (17) requires that [Labial] be associated to two adjacent syllables. That is, the rule of labial harmony must in any case check the syllable tier in order to permit the harmony to occur. We propose that this adjacency condition must remain in force and apply to the spreading of [Labial] as well.

The syllable condition on labial harmony in WM is not an isolated phenomenon.

The same condition on labial harmony is also found in other Manchu-Tungusic languages in China. We will demonstrate this by looking briefly at labial harmony in the Oroqen language.

4. Labial harmony in Oroqen

Oroqen [ɔrɔ’tɕɛn] is a Tungusic language spoken in the northeast of China with about 2,000 speakers at present. The Oroqen data reported in this paper is from Zhang's fieldwork (Zhang 1989; 1995b).

Oroqen has the vowel inventory shown in (22):

(22) vowel inventory of Oroqen (Zhang 1989; 1995b)

Unlike WM, Orogen has both short and long vowels. In Oroqen, the feature

[Labial] may occur on a low vowel only if it is associated with the first two moras of a stem as its domain. The two moras may be associated with either two short vowels or one long vowel, or a short vowel plus a long vowel, as shown in (23):

(23) [Labial] has to be linked to two moras within a morpheme (Zhang 1995a)

a. [Labial] linked to two short vowels

b. [Labial] linked to one short vowel and one long vowel

c. [Labial] linked to one long vowel

d. [Labial] is not allowed to be linked to just one short low rounded vowel

[Labial] has to be linked to two syllables (not two moras) in order to spread. A long vowel by itself does not trigger labial harmony:

(24) [Labial] has to be linked to two syllables in order to spread (Zhang 1995b)

a. [Labial] is linked to two short vowels; [Labial] spreads (*CɔCɔ-Ca, *CoCo-Cə)

b. [Labial] is linked to a short vowel plus a long vowel; [Labial] spreads

c. [Labial] is linked to a long vowel plus a short vowel; [Labial] spreads

d. [Labial] is linked to one long vowel; [Labial] does not spread (*Cɔɔ-Cɔ, *Coo-Co)

A brief comparison of labial harmony in WM and in Oroqen is given in (25). We find that the conditions on labial harmony in these two languages are the same: both languages require that [Labial] be linked to two syllables in order to spread. In Oroqen, [labial] must have two moras as its domain in a morpheme; in WM, [Labial] can have one syllable as its domain in a morpheme. This difference is evidently due to the existence of long vowels in Oroqen but not in WM.

(25) comparison of labial harmony in WM and in Orogen[4]

Short vowels

Long vowels

Short & long vowel co-occurrence

Domain

Spreading

5. Conclusion

The syllable condition on labial harmony in written Manchu has never been described in the Chinese and western literatures available to us. That this is a genuine condition governing labial harmony and not just an accident of the data is supported by the existence of the same condition in Oroqen (Hu 1986; Zhang et al 1989).[5] It appears that the syllable condition on labial harmony is a characteristic of the Manchu-Tungus languages in China.

We would like to conclude with two observations. The first concerns the syllable condition itself. Such conditions are rare in harmony systems, but they are more common in stress systems. For example, it is common to find a condition whereby a metrical foot must have at least two moras, or two syllables. It may be that the syllable condition on labial harmony has a basis in the prosody of the Manchu-Tungus languages; however, we have not yet found independent evidence for this hypothesis.

The second observation concerns the status of neutral vowels in harmony systems. We have seen that a number of approaches, such as those of Van der Hulst & Smith and of Walker, have attempted to predict whether neutral vowels will be transparent or opaque to vowel harmony on the basis of the system of contrasts found in the language. Written Manchu appears at first to be a counterexample to such analyses; but we have shown that the opacity of the high vowel /i/ to labial harmony in this language may indeed be predictable, once we take into account the syllable condition. Some recent analyses of vowel harmony, such as that of Cole & Kisseberth (1994) in the framework of Optimality Theory, have given up on the idea that transparency or opacity of neutral vowels can be predicted. Rather, the required result is derived by language-particular rankings of constraints, or by other stipulations. Our investigation suggests, however, that it may be too early to give up on the search for general principles governing the behaviour of neutral vowels in vowel harmony.

*We would like to thank members of the phonology group of the Department of Linguistics at the University of Toronto and the audience at the 1995 CLA annual meeting for their comments. We are grateful for the support of SSHRC research grant 410-92-0885.

References

• Ard, J. 1984. Vowel Harmony in Manchu: A Critical Overview. Journal of Linguistics 20: 57-80.
• Cole, J. & C. W. Kisseberth. 1994. An Optimal Domains Theory of Harmony. Cognitive Science Technical Report UIUC-BI-CS-94-02 (Language Series). Urbana IL: The Beckman Institute, University of Illinois.
• Hayata, T. 1980. Non-abstract Vowel Harmony in Manchu. Gengo Kenkyu 77: 59-79.
• Hu, Z. & Zork. 1986. Ewenkeyu Jianzhi [A Brief Introduction to the Ewenki Language]. Beijing: Minzu Chubanshe.
• Hu, Z. 1986. Erlunchunyu Jianzhi [A Brief Introduction to the Orogen Language]. Beijing: Minzu Chubanshe.
• Hulst, H. van der & N. Smith. 1988. Tungusic and Mongolian Vowel Harmony: A Minimal Pair. In P. Coopmans & A. Hulk (eds.), Linguistics in the Netherlands. Dordrecht: Foris. 79-88.
• Maddieson, I. 1984. Patterns of Sounds. Cambridge University Press.
• Norman, J. 1978. A Concise Manchu-English Lexicon. Seattle and London: University of Washington Press.
• Odden, D. 1978. Abstract Vowel Harmony in Manchu. Linguistic Analysis 4: 149-165.
• Seong, B. 1989. Vowel Length in Manchu. Proceedings of the XXVIII Permanent International Altaistic Conference in Venice, 8-14 July, 1985. Wiesbaden: Otto Harrassowitz. pp. 195-216.
• Vago, R. M. 1973. Abstract Vowel Harmony Systems in Uralic and Altaic Languages. Language 49: 579-605.
• Vovin, A. 1993. Towards a New Classification of Tungusic Languages. Ural-Altaische Jahrbücher 65: 99-113.
• Walker, R. 1993a. Contrastive Specification and Vowel Harmony Mechanisms in the Altaic Languages. Forum paper. University of Toronto.
• Walker, R. 1993b. Manchu Vowel Harmony. Ms. University of Toronto.
• Zhang, X. 1994. Vowel Harmony in the Manchu-Tungus Languages in China. Ms. University of Toronto.
• Zhang, X. 1995a. Vowel Harmony in Oroqen (Tungus). Toronto Working Papers in Linguistics 14: 161-174.
• Zhang, X. 1995b. Oroqen Fieldwork Notes. University of Toronto.
• Zhang, Y., B. Li & X. Zhang. 1989. The Oroqen Language. Changchun: Jilin University
• Press.

1. Words with the structure in (2b) are rare in WM.

2. Written Manchu has the following diphthongs (Zhang 1994):

   Roman letter: ai ei eo iya iye iyo (io) oi uwa uwe ui

   phoneme: /ai/ /ai/ /au/ /ia/ /ia/ /io/ /oi/ /ua/ /uə/ /ui/

3. The velar-uvular alternation in WM is ignored here.

4. Only examples with non-RTR vowels are given here for Oroqen.

5. Ewenki, another Tungusic language in China, has been noted to have a similar condition on labial harmony to Oroqen (Hu & Zork 1986).