Chinese dialects classified on shared innovations

Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 Classifying Chinese dialects/Sinitic languages on shared innovations L. Sagart 1. Earlier classifications and the need for a new one. 1.1. Li Fang-kuei (1938) Li Fang-kuei's highly influential classification of Chinese dialects took as its unique criterion the fate of the Middle Chinese voiced stops (and affricates; I'll say stops from now on). Note: Middle Chinese (c. 500-600 CE) had three series of stops: 1. voiceless unaspirated p, t, ts, k etc; 2. voiceless unaspirated ph, th, tsh, kh, etc 3. voiced b, d, dz, g etc. In most dialects, series 1 and 2 are unchanged; series 3 has a variety of fates. Li's theory was formed at a time when very few Chinese dialects had been analyzed for corespondences with Middle Chinese. Those that had been were mostly the dialects of large cities. Based on these data, Li proposed the following scheme: Mandarin dialects were those in which the MC voiced stops devoiced as aspirated in tone Ping, (examples: 平田前常群) and as unaspirates in the other tones (examples: 柱跪, 代並, 局白); Gan and Hakka dialects were those were the MC voiced stops had all become aspirates; Xiang dialects like Changsha were those where the MC voiced stops had all become voiceless unaspirates; Cantonese dialects were those where the MC initials had become aspirated under Ping and part of Shang (坐近重), while the rest of Shang, and all of Qu and Ru were unaspirated; Min dialects (like Xiamen) were those in which, although voicing had been lost, you could not predict what MC words with voiced stops became aspirated and what words became unaspirated; 1/7 Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 Wu dialects (like Suzhou) were those where the MC voiced stops remained voiced. Li did not give a tree. A straightforward interpretation of his scheme is to assume that first the different dialect groups that compose Sinitic individualized, and that devoicing then happened separately in each dialect group, before they had time to differentiate into subdialects. There are two kinds of problems with Li's analysis. First, there are many more dialects analyzed now than there were in 1938, and there are many, many exceptions to Li's generalizations: There are Mandarin dialects where MC voiced stops are still voiced: Nantong 南通 suburbs, Hangzhou 杭州. Jishou 吉首 in Hunan has devoiced in Shang, Qu and Ru, but retains voicing in Ping. Some Mandarin dialects in Shanxi have aspirated their old voiced initials in all tones. Moreover, certain early Mandarin dialects as recorded in Yuan and Ming times in Chinese-Mongol or Chinese-Korean bilingual lexica still have voiced stops. Likewise, Cantonese, Gan, Min and Xiang—at least—are not homogeneous regarding devoicing. It is clear that there are just too many exceptions to Li's generalization. Devoicing occurred fairly recently in Chinese dialects, after Mandarin, Cantonese, Gan and Min had begun to diversify, so that devoicing is not a good predictor of dialect affiliation in Chinese. The second problem relates to Wu. To say that the Wu dialects are a branch of Sinitic supposes that they have a common ancestor which is more recent thant the common ancestor of all Chinese dialects. To make that claim one needs to find a feature that they share exclusively; a feature that is not found in other Chinese dialects and that the ancestor of all Chinese dialects did not share either. We could then say that the change that brought this feature about took place in the ancestor of Wu, and that is why all Wu dialects have it, and all non-Wu dialects do not have it. But Li is assuming (rightly) that the ancestor of all Chinese dialects had an independent series of voiced stops. For that reason, he cannot at the same time use the presence of this feature to define Wu. Perhaps the Wu area is just a zone in East China where dialects that are very different in origin have failed to lose voiced stops; and the voicing isogloss around them is just the line beyond which 2/7 Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 devoicing has not spread (yet). In fact it does seem that the southernmost dialects in the area are much closer to northern Min than to northern Wu. The principle we should remember is that one should classify languages on shared innovations (=changes that occurred after the breakup of the most recent ancestral language of the group to be classified). One cannot use shared retentions (=features that were already present in the ancestor). You can use the change of b, d, g into p, t, k as a classification feature, but you cannot use the absence of a change. This principle was first formulated by August Leskien in 1876. Evolutionary biologists also use it. Classifying on shared retentions is a common fallacy. 1.2. Norman (1988) Norman defined three large groups of dialects: a northern group, corresponding to Mandarin, defined by lexical innovations like having 他 for 3SG pronoun; 的 for attributive particle, 不 for general negation, 站 'to stand', 走 'to walk', 儿子 'son', 房子 'house'. This is sound and the move towards using lexical features is welcome. However his southern group (Min + Hakka + Cantonese) is defined by the fact of NOT having these innovations. That is the same fallacy as in Li's classification (Norman also has a central group: Wu+Gan+Xiang, which is transitional between north and south). 1.3. Need for a new classification. Li's Wu group and Norman's southern group are based on shared retentions and lack evidence of internal unity. We need a classification based on shared innovations and nothing else. 2. Other methodological decisions. 2.1. Preference for innovations in the basic vocabulary and morphology. This is because sound changes spread easily (for instance palatalization of velars has spread across Chinese dialects from Manchuria to south Jiangxi and SW Fujian in the last 300 years), while basic vocabulary and morphology spread less. Sound changes will be used only as supporting material. 2.2. Method for inferring a genetic tree from a set of innovations. Because of language contact, not all genuine innovations will stay confined in the language where 3/7 Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 they first appeared. When that happens, an innovation will give a false testimony. Not all the innovations one has assembled will be mutually compatible, in other words, compatible with the same genetic tree. The approach followed here is to look for the largest possible set of innovations (in the basic vocabulary/morphology) that is compatible with the same tree, and regard that tree as an approximation of the true phylogeny. This is a type of compatibility method known as ‘clique analysis’ (Meacham and Estabrook 1985). Clique analysis can be applied by computer-illiterate linguists who possess a good knowledge of linguistic facts pertaining to a group of languages and the ability to discern innovative from non-innovative changes. In Chinese the fact that the Old Chinese vocabulary is known from texts and palaeography makes the detection of lexical innovations in modern dialects in many cases straightforward. 3. Old Chinese features lost in all modern Chinese dialects. OC can be defined as a language ancestral to all attested forms of Chinese save the Jiaguwen: Chinese of the Jin Wen, the Shu Jing, the Yijing, the Shi Jing, the Chunqiu, Zuozhuan, Lunyu, Zhuangzi, Mengzi, the excavated texts of the Zhanguo period, everything written in the Han, Middle Chinese, modern dialects, donors to Vietanmese, Hmong-Mien, Tai-Kadai, Bai etc. A new reconstruction of OC is going to be available on the CRLAO website in a couple of days. Compared with the rest, Modern Chinese dialects do share some non-trivial innovations. Thus, while the Old Chinese words for 'head' and 'to be' were 首 and 惟, no modern Chinese dialect has these forms. They either have 頭 and 是 or forms which replaced 頭 and 是. In phonology, words like 葉 and 用 had a l-type initial in OC: *lap and *m-loŋ: no modern dialect does. In all modern dialects, the initial of these words is explainable as a development out of y. Similarly, OC *t, *th and *d in the same context had changed to palatal affricates. And words like 海 and 黑 must have had a kind of nasal in their initial (Baxter-Sagart reconstruct *m̥), but in all modern dialects, the initial of these words is explainable as a development out of x-. This means that the Most Recent Common Ancestor (MRCA) of modern Chinese dialects was a form of Chinese which already had replaced 首 with 頭 and 惟 with 是; in which *l- had changed to y (though *lˤ-, *lˤr- and lr- were still laterals) and m̥ had changed to x-, and *t, *th and *d to palatal affricates. The change of l to y gives us an element of datation: in 330 and 329 BCE Alexander the Great founded in Central Asia a series of cities, all named Alexandria. One of these was cited in the Hanshu as 烏弋山離, where 弋 =lək. Note the correspondence of OC l- to Greek l. The date at which the transcription was made is not known, but it clearly cannnot be before 330 BCE. Therefore the change of *l to y- is later than 330 4/7 Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 BCE. And since the MRCA of all modern dialects had y, it too has to be later than 330 BE. 4. First split The first language to branch off from the mainstream was likely located in the south or southwest on Chinese. It has two modern descendants: the Wǎxiāng 瓦鄉 dialect of northwest Hunan and the Càijiā 蔡家 language of western Guizhou (1000 speakers, all old!). This last language has only been minimally described (Bo 2004) and it is not even clear that it is a Chinese dialect, but at the very least it includes a Chinese layer in it and that layer is clearly connected to the Waxiang dialect. Both Caijia and Waxiang are very archaic, lexically and phonologically: they preserve the old words for 'dog' 犬, 3sg 伊, 'face' 面, 'love' 字, 'walk' 行. No other Chinese dialect preserves 'love'. Phonologically they have the same preservations as Min; they also preserve OC *lˤ- and lr- as laterals for instance 田 is CJ len31, WX lɛ13; they preserve OC *r as ɣ and z: 來 CJ ɣɯ31, WX zɛ13; etc. But what makes them a subgroup is two innovations that they have in common: • 'two' is 再 (in shangsheng, not qusheng): CJ ta55, WX tso53. • 'milk' is mi 55 in both, perhaps a word of non-chinese origin. 4. Next split. As mentioned before, all Chinese dialects save these two have 愛 for 'love'. This is an innovation that must have happened after Waxiang-Caijia branched off. 5. Then the group that has 愛 for 'love' split into two branches, the main branch and a south-eastern group in which some innovations appeared: 骹 leg human being 箬 leaf Ningbo kʰɔ1 ȵioŋ2 ȵiaʔ8 Wenzhou kʰuo1 naŋ2 ȵia8 Fuzhou kʰau1 nøyŋ2 nuoʔ8 Xiamen kʰa1 laŋ2 hioʔ8 (based on Wang Futang 1999 Hanyu Fangyan yuyin de yanbian he cengci. Beijing: Yuwen. p. 67) This south-eastern group is ancestral to Min and to the earliest layer in the Wu dialects. This layer is very thin in northern Wu, so on the chart the main line of evolution into the Wu dialects is shown as a much later branching-off. Then within this group Min branches off with the unique Min innovations shown on the tree: rice in the husk 穀> 粟; field 層; house 戍; etc. They are due to Norman. 5/7 Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 6. In the other branch, the mainstream branch, new developments occured. The 3rd person pronoun changed from 伊 to 佢; a new word for 'to hit' appeared: 打; there were also phonological developments: *st- changed to sy- (phonetically ɕ-) in words like 室書. From this language a south- central group separated, which developed its own innovations: 禾 rice plant, 佬 person. This group then split into a more northerly branch which now exists only as the earliest layer in dialects classified as Xiang, Gan and even southern Mandarin; and a more southerly group which includes Hakka and Cantonese (and probably some 'tuhua' like Lianzhou in W. Guangdong and Yizhang in S. Hunan). Shared lexical innovations of this southerly branch: 係 ‘to be’, 生 ‘to live’, MC thuwX ‘to rest’, 發夢 ‘to dream’, 屎窟 ‘the arse’. Meanwhile in the main line of development, new lexical innovations were appearing: 吃 ‘to eat’ (no text occurrences before Qing times, but th eword must be old, and vulgar), 太陽 ‘sun’. In phonology all remaining laterals changed to d- or th- and dr or trh-; and r- changed to l-. This last change allows us to date the time of the individualization of this branch between the first century CE (when evidence for a change of laterals to d and dr first appears) and maybe 400 CE (change of *r to *l). 7. At this point the phonology approaches EMC; a southern branch separates, and contributes a very thick layer on the early form of Chinese (Min-related) spoken in the lower Yangzi region to form the "Wu dialects". I have not attempted to isolate uniquely shared innovations of the Wu group. I am not sure it is doable, even if the task is limited to exclusively shared innovations of Wenzhou and Suzhou. Therefore I cannot guarantee that Wu is a valid group. Certainly the Chuqu dialects in south Zhejiang, which are usually classified as Wu because they have voiced initials, are basically Min dialects. 8. Then some more innovations in the main branch: 走 and 臉 become the main words for 'to walk' and 'face'. Nanchang separates (I cannot give shared innovations for Gan); we are now in the Late Middle Chinese regions of phonology. Finally the 3rd person pronoun changes to 他, and Changsha separates. We are now in early Mandarin times. Baxter regards Changsha as part of (pre-devoicing) Mandarin. The end 6/7 Séminaire Sino-Tibétain du CRLAO Nogent, 28 mars 2011 References Bo Wenze (2004) Caijia hua gaikuang. MZYW 2004, 2:68-81. Leskien, August (1876) Die Declination im slavisch-litauischen und germanischen. Leipzig: Hirzel. Li Fang-kuei (1938) Languages and dialects of China. The Chinese Yearbook, 1938. Shanghai: Commercial Press. Norman, Jerry (1988) Chinese. Cambridge language surveys. Cambridge: Cambridge University Press. Wang Fushi (1982) Hunan Luxi Waxiang hua yuyin. YYYJ 1982, 1: 135-147. Wang Fushi (1985) Zai lun Hunan Luxi Waxiang hua shi Hanyu fangyan. ZGYW 1985, 3: 171-177. Wu, Yunji and Shen, Ruiqing. 2010. Xiangxi Guzhang Waxianghua diaaocha baogao. Shanghai: Jiaoyu. Yang Wei 1999: Yuanling Xiang hua yanjiu. Hunan Fangyan Yanjiu Congshu. Hunan Jiayu Chubanshe. 楊蔚 1999 沅陵乡话研究. 湖南方言研究丛书. 湖南教育出版社 7/7