Where are the -r- s in Old Chinese?

In the early 1960s, Yakhontov and Pulleyblank first
proposed the reconstruction of medial *-r- for the class of Old Chinese syllables which developed into the so-called "second
division" of Middle Chinese. The reconstruction of this medial
element also conditioned the development of Middle Chinese
retroflex initials out of Old Chinese dental stops. While this medial
*-r- hypothesis neatly accounts for observed historical
developments within Chinese, in the broader context of Sino-
Tibetan comparison it has proven to be problematic.
Three objections may be raised. First, within the Sino-
Tibetan family it is typologically unusual to find medial -rclustering
with dental initials (other than s-). Second, the presence
in Old Chinese of etymologically related variants involving the
presence or absence of medial *-r- has led to hypotheses that *-rwas
a derivational infix. Infixation of this type is rare in Tibeto-
Burman, and no wide-spread equivalent to the hypothesized *-rinfix
is found in reconstructions of Proto-Tibeto-Burman. Third, in
many proposed cognate sets involving Old Chinese words and
Tibeto-Burman forms or reconstructions, a Tibeto-Burman feature
corresponding to Old Chinese medial *-r- cannot be found,
resulting in awkward Proto-Sino-Tibetan reconstructions.
While these objections are not, taken individually, fatal to
the medial *-r- hypothesis, they do raise concerns. Drawing
primarily on the work of Gong Hwang-cherng (1995) and W. South
Coblin (1986), this paper, part of a larger study of Old Chinese
medials, revisits the medial *-r- hypothesis in search of a solution
which is compatible with comparative evidence while retaining the
explanatory power of the original hypothesis. In light of recent
scholarship suggesting that Old Chinese had a rich derivational
system of morphological prefixes I conclude that, at least in the
case of Old Chinese syllables with dental stops and affricates,
medial *-r- should be revised to préfixai *r-. This revision is

compatible with comparative evidence showing that Old Chinese
medial *-r- often corresponds to Tibeto-Burman préfixai *r-, yet in
many other cases has no Tibeto-Burman equivalent.
This is more than a formulaic revision, since it has
implications for our understanding of Old Chinese syllable
structure, typology, and morphology. The reconstruction of the
Chinese syllable (or "sesqui-syllable") can be brought into closer
alignment with Tibeto-Burman and at the same time made more
internally consistent.
The proposal also allows us to reexamine our criteria for the
evaluation of the validity and reliability of individual Chinese/Tibeto-
Burman comparisons, and to posit more regular rules of
correspondence between these two major branches of Sino-Tibetan.

METHODOLOGY

Because this paper makes use of Tibeto-Burman data to aid
in the reconstruction of Old Chinese, a brief explanation and
justification of the methodological approach involved is warranted.
Ideally, under the assumption of the branching model of
linguistic divergence, the comparative method calls first for the
rigorous comparison of closely related languages to achieve
reconstructions of their "meso-level" ancestors. These reconstructed
ancestor languages are then compared to each other, allowing the
reconstruction of an older common ancestor... and so forth, until the
limits of the method are reached and the earliest possible protolanguage
has been reconstructed. According to a strict
interpretation of this method, only languages descended from Old
Chinese - in addition, of course, to contemporary textual
evidence- should be used in the reconstruction of Old Chinese
itself. The characteristics of its sister language Proto-Tibeto-
Burman should not be used to draw conclusions about the nature of
Old Chinese.

In actual practice, however, historical linguists have been
more pragmatic in their work. Hypotheses concerning the nature of
the proto-language are advanced based on preliminary evidence
from daughter languages; these hypotheses are then tested against
additional data from daughter languages (or, ideally, newly
discovered daughter languages), and are then accepted, modified or
discarded. The process is bi-directional; at any given moment, the
historical linguist is working with partial, hypothetical
reconstructions at different time-depths, as well as with incomplete
data for the daughter languages. The linguist develops the
reconstructions incrementally, working forward and backward in
time, bringing each stage of reconstruction into alignment as new
evidence is amassed and processed.
If this practical approach is applied to the comparison of
reconstructed Old Chinese with reconstructed Proto-Tibeto-Burman
in the reconstruction of their parent language, Proto-Sino-Tibetan, it
makes perfect sense to modify a working reconstruction of Old
Chinese to better accommodate a hypothesized Proto-Sino-Tibetan
reconstruction, as long as one takes care that the Old Chinese
reconstruction is at all times consistent with internal Chinese
evidence. The revised Old Chinese reconstruction will in turn have
ramifications for the Proto-Sino-Tibetan reconstruction; the process
of revision continues in both directions until equilibrium is reached.
At that point, with hindsight, the whole structure can be presented
as if it were arrived at according to the strict interpretation of the
comparative method.
There is another sense in which the use of Tibeto-Burman
evidence may be helpful in reconstructing Old Chinese. It is
sometimes the case that the simplest, most elegant solutions to
problems in historical reconstruction can evade discovery for years,
although once proposed they might appear perfectly obvious in
hindsight. The insight leading to the solution is sometimes triggered
by an example from another language or language family. One
never knows which Tibeto-Burman language might possess the

phonological feature which proves to be the key to solving a
significant problem in Old Chinese reconstruction. This key insight
might be independently verifiable with Chinese evidence alone; but
it might never have seen the light of day without the example of
Tibeto-Burman. Axel Schuessler's 1974 proposal, now widely
accepted, regarding the reconstruction of Old Chinese *r and *1 is
an example of the benefits of this approach.

Taking all the evidence presented above into account,
several conclusions can be drawn. In the case of dental stops and
affricates, there is no need to propose both *r-T- and *Tr- clusters
for ОС. Given that PTB seems to have had *r-T- but not *Tr-, it is
simpler to suppose that ОС also had *r-T- rather than *Tr- than it is
to suppose widespread regular metathesis from PST *r-T- to ОС
*Tr-

One may well ask how the development of Middle Chinese
second-division vocalism could be conditioned by a préfixai, rather
than medial, *r, as in examples (lb) and (2b). This is in fact not a
surprising development. Once the dental initials became articulated
as retroflexes through assimilation to préfixai *r-, this feature could
in turn have spread to the vowel: *r-T- > *rT- > *rT- > T- +
second-division vocalism. In other words, the vocalic development
which Baxter (1992) terms *r-color can instead be attributed to
*retroflex-color.

Conclusions about medial, prefix, and initial *r are listed in
Table 7, along with the parallel developments of *I, which are
beyond the scope of this paper. In this table the cover symbol С
represents grave initials (velars and labials). Having established an
ОС prefix *r-, it seems probable that prefix-initial combinations like *r-K- and *r-P- existed in OC, but these combinations would
presumably have developed identically to unprefixed *K- and *P-,difficult to say in which words they should be reconstructed

Coblin follows Li in reconstructing *-rj- in the ОС form for 'ten' to account for the palatal initial in MC. However, as Pulleyblank (1962) has shown, this
palatalization can for the most part be accounted for in other ways, and *-rj should
instead be reconstructed to account for the development of MC chdngniu third-division words. For more details, see Baxter (1992 : 211), who reconstructs
ОС 'ten' as *gjip. As for 'goose', the WB form is an isolate in ТВ, which can perhaps be equally well compared with 鹅 é 'goose' (OC *ngaj in Baxter's system,
*ngar in Gong's), which has no medial *r. The WB form is treated by Benedict as
an -n-suffixed form of a hypothetical basic ТВ root *rja 'goose' for which there is
no direct evidence (Benedict 1972 : 99 note 284). A more likely explanation is that
the two Chinese words for 'goose' belong to that class of words which Baxter
(1995) now follows Starostin in reconstructing with final *-r. These words
normally merge with final *-n but develop like *-j in some ancient dialects,
leading sometimes to doublets.
10
While the WT form for 'corner/angle' indeed shows no evidence for medial *-r-,
the PTB root *krsw 'horn' (Benedict 1972 : #37) does. And, despite the lack of a
final stop, this root is closer in meaning to the basic sense of the Chinese word.
Note the words for 'horn' in these other ТВ languages:
(i) Dulong (Dulonghe) tçi3Ixjui55, Dulong (Nujiang) arj31 Ш3'хлш", Darang
jau", Geman kiâg ", Cuona rw13
Dulong xi- is a regular reflex of PTB *kr-. The Dulong morphemes are
unmistakably descended from *kraw. The Geman form appears to be related, but
here we have the unexpected appearance of a final velar. (Perhaps there is some
connection between this final and the final *-k of the Chinese form.) In Cuona the
initial *k- has dropped, leaving an r-initial word. These same variations are
discernible in various meso-level reconstructions for 'horn':
(ii) Proto-Loloish *kro', Proto-Lolo-Burmese *kruw\ Proto-Tamangic *Ahru,
Proto-Northern-Naga *rurj
In this context, note that the Tibetan word for 'horn' is rwa ~ ru and that there is
another Tibetan word for 'comer, angle' grwa - gru (Benedict 1972 : 113), neither
of which show evidence of a final stop. It is possible that all of these words are
part of a larger word family, and that the Chinese form is to be related to a PTB
root which does contain a medial *-r-.

As for the morphological function of préfixai *r-, if any, it is difficult to
speculate based on the data which appears in this paper. Sagart (1999) proposes a
number of morphological functions for his infix *-r- covering a broad semantic
range, some of which could be attributed to the prefix reconstructed here.
However, I do not feel that there is sufficient evidence at this point to draw firm
conclusions.