2.1. Evans’ correspondences

Evans (Reference Evans2001) contains the most comprehensive study of Rma diachrony to date, with over 1,000 proposed correspondence sets for Rma. Nonetheless, there are methodological issues that undermine the usefulness of some of Evans’ reconstructions, especially for the vowels. In general, Evans (Reference Evans2001) identifies forms with the same vowel across varieties, reconstructs an identical vowel, and treats exceptions as exhibiting proto-variation. For example, Evans (Reference Evans2001: 176) states that Longxi Rma /o/ :: Mianchi Rma /o/ and that both reflect *o. Evans gives a small number of problematic sets (5) to support this claim. Evans (Reference Evans2001: 177) also states that Longxi /u/ :: Mianchi /u/ and that both reflect *u. Again, there is a rather small number of problematic sets (5) given to support this claim. Nonetheless, Evans goes on to state that:

Table 1 gives a summary of correspondences for these vowels found in the forms reconstructed by Evans.

Table 1. Vowel correspondences found in Evans 2001

There are no fewer than eleven distinct correspondence patterns in Table 1, including three different, unpredictable, outcomes for both *u and *o, two for *u/o and, three for *o/u. Thus, the starred forms do not predict the attested forms. This approach is methodologically unsound, as reconstructions which do not predict the attested forms are not useful. As Hill (Reference Hill2012: 64) makes clear:

Evans (Reference Evans2006a: 113) recognizes that “some revision of PSQ [Proto-Southern-Rma vowels] is needed”, but has not yet revised his reconstructions. The following section gives an alternative analysis that explains the relationship between these Longxi and Mianchi vowels.

2.2. The present analysis

I propose that Longxi /o/ :: Mianchi /ou/ and that Longxi /u/ corresponds with Mianchi /o/ for native forms. Other correspondence patterns are the result of borrowings from other languages or dialect mixing, onomatopoeia, or analogical levelling.

2.2.1. Longxi /o/ :: Mianchi /ou/

Longxi /o/ :: Mianchi /ou/. Evidence is given in Table 2. Note that Longxi H and L tones correspond to Mianchi H and L.Footnote ⁴

Table 2. Longxi /o/ :: Mianchi /ou/

This correspondence is validated by a regular correspondence with Ronghong /u/. Consider Table 3.

Table 3. Longxi /o/ :: Mianchi /ou/ :: Ronghong /u/

2.2.2. Longxi /u/ :: Mianchi /o/

Longxi /u/ :: Mianchi /o/. Consider the evidence in Table 4. Note again the regular tonal correspondences between Longxi and Mianchi.

Table 4. Longxi /u/ :: Mianchi /o/

The correspondence is again confirmed by comparison with Ronghong. See Table 5 for examples.

Table 5. Longxi /u/ :: Mianchi /o/ :: Ronghong /(w)ə/

Table 5 also suggests a change *wə > ə took place after bilabials in Ronghong. The vowel correspondences proposed for native forms are given in Table 6.

Table 6. Vowel correspondences for inherited forms

This raises the question, what about the other nine correspondence patterns in Table 1? The following section attempts to answer this question.

2.3. Irregularities

Loanwords introduce exceptions to regular sound correspondences. Tibetan and Chinese loans have long been known in Rma (Sun Reference Sun1988), but Rma-internal loans have not yet been identified.

2.3.1. Tibetan loans

Tibetan loans are more frequent in the Rma varieties of Heishui county (Liu Reference Liu1981), and also in Songpan County (Huang et al. Reference Huang, Baofeng, Mingjun and Xi2019). See Liu (Reference Liu1981) on Tibetan loans into Mawo Rma.

Table 7 gives three different forms which are probably Tibetan loanwords and shows how these forms have different vowel correspondences from those we find for inherited words.

Table 7. Probable Tibetan loans

Regarding “dragon”, Evans (Reference Evans2001: 235) notes that “It is probable that this is an ancient borrowing from Tibetan”,Footnote ⁵ and does not reconstruct a proto-form.

For “demon”, Evans (Reference Evans2001: 289) reconstructs *du L. LaPolla and Huang (Reference LaPolla and Chenglong2003: 355) give the Ronghong form du-gu̻-mi “ghost, spirit” and note that mi = “eyes”.Footnote ⁶ The vowel correspondences are irregular and suggest that these forms are Tibetan loans.Footnote ⁷ The same can be said for “hole, cave”.Footnote ⁸ Recognizing the aforementioned Tibetan loans explains the otherwise exceptional vowel correspondences.

2.3.2. Chinese loans

Chinese loans have long been discussed in the literature (Sun Reference Sun1988; LaPolla and Huang Reference LaPolla and Chenglong2003: 46–7). Consider Longxi lò pú “radish” and Mianchi lò pú “radish”. These forms, both from Chinese 蘿蔔 luóbo “radish”, show irregular correspondence for both [o] and [u]. Evans (Reference Evans2001) discusses many instances of borrowings from Chinese. Table 8 gives forms which have irregular vowel correspondences within Rma due to being borrowings from Chinese.Footnote ⁹

Table 8. Chinese loans

Nonetheless, certain loanwords from Chinese have gone unnoticed, and, in some cases, Evans reconstructs Proto-Southern Rma forms for the Chinese loans. Table 9 gives forms which have not been recognized as Chinese loans. For these forms, Evans reconstructs a proto-Southern form. However, treating these as native forms ignores the irregular vowel correspondences.

Table 9. Probable Chinese loans

It should be noted that in Southwestern Mandarin, the variety of Mandarin in contact with Rma, the words “gong”, “elder brother”, “loose”, and “clump, heap” share the same vowel nucleus: [o].

Some of the forms in Table 9 are straightforwardly loans from Chinese. For example, “gong” is a loan due to the vowel correspondence, but also because this musical instrument is generally accepted to be Chinese in origin (see Kunst Reference Kunst1947). Evidence that “trough” is a loanword comes from the fact it is found in Ronghong wə-tsʰu “manger” (horse-trough) (LaPolla and Huang Reference LaPolla and Chenglong2003: 347). Note the irregular vowel correspondence for Ronghong.

Some of the forms are possibly Chinese loans, but the relationship is not as certain. These include: “clump, heap”, “row of grain”, “to dip pen”, and “elder brother”.

For example, it seems possible that the forms for “to dip pen” are loans from Chinese 書 shū “book, to write”. The meaning “to write” for the Chinese form is more archaic. This comparison is admittedly speculative, but seems more plausible than reconstructing a literary term for a language without a written tradition.

Nonetheless the explanation of loans from Chinese seems the most likely at this point. Even if these are ultimately not Chinese loans, they should not be reconstructed in this way as they introduce multiple correspondences between proto-Southern Rma and the modern varieties such that the reflexes are not predictable based on the proto-forms.

2.3.3. Cross-dialectal borrowings

Cross-dialectal borrowings are more difficult to detect than borrowings from closely related languages, but are an important part of sorting out the historical phonology of a language. I can find in prior work no identification of cross-dialectal borrowings. In this section, I invoke cross-dialectal borrowings in order to explain vowel correspondences which are exceptional to the patterns summarized in Table 10.

Table 10. Potential cross-dialectal loans

While the Mianchi :: Ronghong correspondences are regular, the Longxi forms have o where we would expect u. There is no indication that these forms are borrowings from another language. One solution is to propose that the Longxi forms are Mianchi loans. There are two reasons for this. The first is phonological: *H.L tones correspond regularly to collapsed monosyllables in Ronghong (cf. “flea”, “listen” in Table 3).Footnote ¹⁰ Thus the etymological tone for “winter” should be H.L, as seen faithfully reflected in Mianchi, and not the aberrant L.H in Longxi. For this reason, we would expect the directionality of borrowing to be from Mianchi into Longxi, and not the inverse. The second reason is sociological. Liu (Reference Liu1998) reports 3,300 Longxi speakers and 15,700 Mianchi speakers. These numbers are outdated, yet it is clear that Mianchi is a much larger variety. This makes a loan from Longxi > Mianchi less plausible. The Mianchi and Longxi varieties are geographically close, and are both spoken in Wenchuan county. Thus, positing interdialectal loans from Mianchi to Longxi is not unreasonable.

2.3.4. Onomatopoeic forms

Evans (Reference Evans2001) reconstructs several onomatopoeic forms, and thereby introduces irregular vowel correspondences. Consider the forms in Table 11.

Table 11. Onomatopoeic forms

Evans (Reference Evans2001: 268) notes that “cat”, “appears to be onomatopoeic”. Evans’ (Reference Evans2001) reconstruction of these onomatopoeic forms is problematic.Footnote ¹¹ The importance of excluding onomatopoeic forms has long been known to the comparative method (Lottner Reference Lottner1862; see discussion by Lehmann Reference Lehmann1976). Authorities on the comparative method caution against comparing onomatopoeic forms, which are frequently renewed by more imitative forms and thus generally unreliable in historical linguistics (Hock Reference Hock1991: 50; Harrison Reference Harrison, Joseph and Janda2003: 216; Campbell Reference Campbell2013: 317; Millar and Trask Reference Millar and Trask2015: 207; Rankin Reference Rankin, Joseph and Janda2017: 184). Most of the forms in Table 11 are pointed out by Campbell (Reference Campbell2013: 321) as cross-linguistically likely to be onomatopoeic (overlapping forms in bold):

A way to reduce the sound-imitative factor is to omit from consideration words which cross-linguistically are often imitative in form, for example, words meaning “blow”, “breathe”, “suck”, “laugh”, “cough”, “sneeze”, “break/cut/chop/split”, “cricket”, “crow” (and many bird names in general), “frog/toad”, “lungs”, “baby/infant”, “beat/hit/pound”, “call/shout”, “breathe”, “choke”, “cry”, “drip/drop”, “hiccough”, “kiss”, “shoot”, “snore”, “spit” and “whistle”, among others.

2.3.5. Unresolved issues

This section presents an unsolved problem regarding vowel correspondences for certain numerals in Rma. Consider the data in Table 12.

Table 12. Rma numerals

Note that these numbers do not pattern like native forms, nor like loans from Tibetan or Chinese. Looking closer, we see alternations in compounds with these three numbers as well. Consider Table 13.

Table 13. Rma numerals (continued)

We see consonantal and vocalic alternations in Ronghong, as well as consonantal and tonal alternations in Longxi and Mianchi.Footnote ¹² Note that lack of tonal correspondences between Longxi and Mianchi. This is suggestive of either early borrowings from another language, inter-dialectal borrowing, analogical levelling, or some combination.Footnote ¹³

3.1. A “far”-fetched etymology

Matisoff (Reference Matisoff2003: 195) gives a problematic etymology of the Mawo Rma word for “far”. Writing about his PTB root *g-wəy-n FAR:

The bracketed text (italicized in the original) indicates that it is the first of the two Mawo Rma syllables that is meant to lend evidence for a velar prefix. This is odd given that the Taoping Rma form χuɑ³³ “far”, which has a uvular onset, is also cited as evidence of a velar onset. Of the languages cited here, only Queyu has velars. Baxter and Sagart (Reference Baxter and Sagart2014) reconstruct Chinese 遠 yuǎn “far” < Middle Chinese *hjwonX < Old Chinese *C.ɢʷanʔ, with a uvular initial.Footnote ¹⁴

Issues of uvularity aside, the above analysis is rooted in a misunderstanding of the Mawo form. Mawo Rma gwə˞ χe is a compound in which the first syllable is “road” and the second syllable is “far”. Consider the Ronghong Rma forms gwa-ha “far”, and gwe-ɲi “near” (LaPolla and Huang Reference LaPolla and Chenglong2003: 381), these forms contain the same morpheme found in Ronghong gwəːʴ “road”. Consider also the parallel situation in Mianchi Rma: ʐí-χwà “far, distant”, ʐí-zɛ̀ “near”. Both contain the Mianchi word ʐì “road”. The Mianchi forms are correctly analysed by Evans (Reference Evans2001: 361) as compounds with “road” as the initial element. In this case, a lack of recognition of the internal structure of the Rma form has led to problematic comparisons with other languages.

3.2. A “cloudy” etymology

Matisoff (Reference Matisoff2003: 271; STEDT #5656) lists Mawo zdɤm “cloud” as a reflex of his PTB *s-dim CLOUD, without discussion of the fact that the -m in Mawo is secondary. As Jacques (Reference Jacques2015b) has pointed out, the -m in this form must be secondary, as Rma lost all codas (see LaPolla and Huang Reference LaPolla and Chenglong2003 and references therein).

The Mawo form zdɤm “cloud” is nearly identical to forms found in Rgyalrong, i.e. Japhug Rgyalrong zdɯm “cloud” and Kyomkyo Situ zdeʔm “cloud”.Footnote ¹⁵ Thus, one possibility is that the Mawo form is borrowed from Rgyalrong. Another possibility is that the -m coda in Mawo Rma “cloud” comes from the second syllable of a compound form. Evans (Reference Evans2001) analyses the apparently cognate forms Longxi dá mù̻ “cloud” and Mianchi dá mò “cloud” as compounds with the second element meaning “dark”. Note that the change *Cd > d is regular in Longxi and Mianchi (see Evans Reference Evans2001: 162–3). In this view, the Mawo coda -m would come from a second element meaning “dark”. Both explanations are possible, but in either case, the -m in Mawo zdɤm “cloud” would not be a reflex of the ancient coda -m posited by Matisoff.

3.3. A “corralled” etymology

On the basis of Longxi tɕʰý “to enclose (sheep)” and Taoping ŋu⁵⁵ tɕi⁵⁵ “enclosure”, Evans (Reference Evans2001: 280) reconstructs a proto-Southern Rma form *tɕ/tɕʰ i/y H. Segmentally, this comparison is problematic. The Longxi verb is related to Mawo tɕʰu “to pen (sheep)”. The unrelated Taoping noun appears to be a compound of zɿ³¹ ŋu³³ “bovine” and tɕi⁵⁵ ko³³  “home”.

3.4. A “thorny” etymology

Evans (Reference Evans2001: 390) gives an etymology of “numbing pepper” (Zanthoxylum bungeanum) as being related to “mutton”. Mianchi tsʰɛ-̀nə́ “numbing pepper” is glossed “mutton?-red” and Evans notes that “(one kind tastes like mutton)”. An issue with this etymology is that the Mianchi form tsʰɛ̀ is not “mutton” but “goat”. It seems more likely that the first syllable is related to Chinese 刺 cì < Middle Chinese *tsʰje H < Old Chinese *[tsʰ]ek-s “thorn” or Tibetan tsher.ma “thorn” and that any similarity with “goat” is coincidental. The second syllable is indeed “red”. Thus, this compound is “red thorn”, which makes more intuitive sense.Footnote ¹⁶ See Jacques and D'Alpoim Guedes (Reference Jacques and Guedes2023) for a discussion of the etymology of Zanthoxylum.

3.5. An “edgy” etymology

STEDT (#0594) links the first syllable of Longxi Rma tɕà qó “knife edge” with PTB *m-dzya EDGE/SIDE. The Longxi form is a compound in which the first syllable comes from tɕà pjá “knife”Footnote ¹⁷ and the second syllable, -qó, means “edge”. Consider the Longxi form tɕà pjá tsò “back of knife blade”. Thus, the comparison does not work well semantically. Even granting a comparison of Longxi “knife” with *m-dzya EDGE/SIDE, the segmental correspondence (PTB *m-dz :: Longxi tɕ-) is not well supported. Consider the data in Table 14.

Table 14. Longxi Rma :: Matisoff's PTB

This etymology should be abandoned due to the weak semantics and problematic segmental correspondences.

3.6. A “bollocks” etymology

STEDT (#1654) draws a connection between the Rma form for “testicles” and Matisoff's Proto-Tibeto-Burman *pu EGG. While many languages have replaced “testicle” with “egg” (i.e. Spanish), there are some issues with this analysis for Rma.

First, neither the onset nor the vowel fit well with the other Rma :: PTB correspondences proposed by Matisoff. Consider the words in Table 15. Longxi forms are from Evans (Reference Evans2001), except for “smell bad” which is from Zheng (Reference Zheng2017).

Table 15. Longxi Rma :: Matisoff's PTB

We see that Longxi /b/ generally corresponds with Matisoff's *b, and that Longxi /u/ generally corresponds with Matisoff's PTB *u. Thus, a comparison of bɚ̀ “testicle” and his PTB *pu EGG does not work well segmentally.

Second, looking within Rma, we see a more likely etymology. The rhotic vowel may be explained as resulting from a collapse of two syllables. Evans (Reference Evans2001: 410) gives the Longxi possessive forms for “testicles” as both bə̀ ə̀˞ and bə̀ ɹə̀. This suggests that the rhotic vowel is from a second syllable which has undergone coalescence.Footnote ¹⁸ Evans and Sun (Reference Evans, Sun and Sybesma2015) give the Hongyan Rma form for “bollock” as ʐə, which seems to be the origin of the second syllable. The first syllable, with the b- initial, is possibly related to Japhug Rgyalrong tɯ-mbɯ “penis” and Tangut ₅₃₆₂ biʶj 2.33 < *mbej “penis” (see Jacques Reference Jacques2014: 168; Gong Reference Gong2020). In brief, the Rma forms seem to come from an old compound and thus comparisons with forms meaning “egg” in other languages are probably incorrect.

4.1. Comparisons with Proto-Tibeto-Burman

The modus operandi of Rma etymologists has been to give top-down comparisons of Rma with reconstructions of proto-languages. For instance, Evans (Reference Evans2001) references Proto-Tibeto-Burman forms from Benedict (Reference Benedict1972), and also references proto-Tibeto-Burman forms from Matisoff (Reference Matisoff2003: 228), Proto-Lolo-Burmese forms (2003: 305, 307), as well as his own reconstructions of proto-Qiangic (2003: 228, 296, 302, 314) and of proto-Northern-Rma (2003: 226). Evans (Reference Evans2001) does make direct comparisons with an impressive number of attested languages.Footnote ¹⁹ Nonetheless, because of the intermittent nature of these comparisons, it is not possible to ascertain the sound correspondences with any one language. In a similar fashion, LaPolla and Huang (Reference LaPolla and Chenglong2003) give a lexicon of the Ronghong variety with intermittent comparisons with Benedict's Proto-Tibeto-Burman, except for the numerals, which follow Matisoff (Reference Matisoff1997). Evans (Reference Evans2006a) gives is a study of the history of certain Rma vowels filtered through Matisoff's (Reference Matisoff2003) starred forms. Sims (Reference Sims2014) gives a lexicon of the Yonghe variety of Rma and posits a smattering of connections between Rma forms and Matisoff's (Reference Matisoff2003) starred forms throughout.

A general problem with this approach is that reconstructed forms may change in light of new data or analyses.Footnote ²⁰ A more specific problem is that neither Benedict's nor Matisoff's starred forms were arrived at through the standard comparative method (Chang Reference Chang1973; Miller Reference Miller1974; Sagart Reference Sagart2006; Reference Sagart2008; Hill Reference Hill2019; Fellner and Hill Reference Fellner and Hill2019). See also an interview with Li Fangkui on the methodological problems (Chan and LaPolla Reference Chan and LaPolla1998).Footnote ²¹

Aside from the matter of how these forms were arrived at, the issue still stands that the starred forms do not predict attested forms. As an example, let us consider the 70 connections posited between Benedict's PTB and Ronghong Rma by LaPolla and Huang (Reference LaPolla and Chenglong2003) in their lexicon of the Ronghong variety. Let us consider only the forms linked with PTB *a. This is widely agreed to be the least controversial vowel in the family (Matisoff Reference Matisoff2003: 162; Sagart Reference Sagart2006; Hill Reference Hill2019: 236) and should be the most straightforward. We will further limit ourselves to cases where LaPolla and Huang posit a link with open syllables. This yields 22 different examples, in which we find no fewer than six different reflexes of Benedict's *a in Ronghong.Footnote ²² Examples are given in Table 16.

Table 16. PTB :: Ronghong

Bringing these examples together reveals a major division between Ronghong /ɑ/ and /ə/ unaccounted for by Benedict's starred forms.Footnote ²³ Because these are all open syllables, it is a problem that cannot be said to result from lost codas.

Vowel raising and fronting has been observed in many Trans-Himalayan languages in the region. This tendency has been called “brightening”, and has been argued to be a common innovation in Qiangic languages (Matisoff Reference Matisoff, Lin, Hsu, Lee, Sun, Yang and Ho2004). One might propose that the apparent vowel split of Benedict's PTB *a into ɑ and ə in Ronghong is the result of brightening. Nonetheless, this is problematic for two reasons. First, recent work has shown that brightening is not a shared innovation but a set of parallel developments (see Chirkova Reference Chirkova2012: 5 fn. 5; Chirkova and Handel Reference Chirkova and Handel2019; Lai Reference Lai2022; Hill Reference Hill2022). Second, there is no obvious phonological condition for “brightening” in Ronghong Rma. Consider the forms for “fat” and “salt” in Table 16. These two forms are segmentally identical in Benedict's system but have different forms in Ronghong.

This belies a more fundamental problem with the way reconstructions are conceptualized. Because reconstructions are merely a shorthand way of indexing the known facts of regular correspondences between cognates, they, by definition, predict attested forms. Reconstructions are not meant to be an amalgam of the phonetic material in attested forms, as we see in Evans’ *q/k i/ou H pu L “cuckoo” or Matisoff's Proto-Tibeto-Burman *m/p/s-(l/d)ap ARM / HAND / WING.Footnote ²⁴

4.2. Regularity of sound change

Some linguists who have worked on Rma have not strongly committed to the principles of regularity of sound change, and this has hindered progress in discovering sound laws. Evans (Reference Evans2001) in particular has expressed doubts about Ausnahmslosigkeit as it relates to Rma vowels:

Evans states that the tones of Rma are “unstable … and often unpredictable” (2001: 231) and subject to “irregular and unpredictable tone sandhi” (2001: 244). Although much has been made of the irregularity or unpredictability of Rma tones, the forms in Tables 2 and 3 attest to the regularity of tonal correspondences for true cognate forms.

Evans (Reference Evans2001: 257) also alludes to grammatically conditioned sound change: “At present I have no explanation for this apparent palatalization, other than the unpredictability of phonological developments in functors”. This type of explanation is not in line with the comparative method (see Hill Reference Hill2014).

Evans (Reference Evans2006b: 113) makes a broader claim that:

This is correct, but it highlights an underlying problem with the correspondence sets themselves, not a fundamental unruliness of the Rma language.