Linguistic palaeontology, a term introduced by Adolphe Pictet in 1859, is an attempt to understand the ancient material culture of a language family on the basis of the lexical items which can be reliably reconstructed for the common ancestral language. The reflexes of the reconstructed roots should be reflected across the various branches of the family, and their phonological development should be historically regular. The two language families which reflect rice agriculture terminology robustly are Hmong-Mien and Austroasiatic.

Strecker's (1987) Hmong-Mien phylogeny recognises three branches, i.e. Hmongic (Miáo), comprising 27 linguistic subgroups arranged in an articulate tree structure, Mienic (Yáo), comprising six subgroups arranged in a tree structure, Ho Nte (Shē) comprising two subgroups, and a fourth entity, the language Na-e, the precise classification of which is problematic. The antiquity of the linguistic phylum as a whole, and indeed the antiquity of most major linguistic phyla, remain largely a matter of conjecture. Indirect arguments like the relative chronology of branching and the possible dating of loans figure prominently in historical linguistics in the absence of absolute dating techniques comparable to, say, a radiocarbon date. However, the perennial inability of linguists to establish a genetic relationship between Hmong-Mien and any other major linguistic Asian phylum is compatible with the assumption that the common Hmong-Mien proto-language could represent a stage roughly coetaneous with Proto-Austroasiatic or Proto-Tibeto-Burman.

Martha Ratliff (2004) ² 2

I thank Martha Ratliff for sharing with me the index of her forthcoming handbook on Hmong-Mien historical phonology, which is slated for publication by Pacific Linguistics in Canberra. I thank Bill Baxter and Wolfgang Behr for discussion of the Old Chinese forms. The reconstructed forms attempt to conform to the latest version of Bill Baxter and Laurent Sagart's Old Chinese reconstruction of 20 February 2011, soon to be published. Many thanks to Scott DeLancey and Wolfgang Behr for helpful editorial suggestions, and I cordially thank Anne-Marie Bacon for the Holocene distribution maps.

Six of the ten forms reconstructible Proto-Hmong-Mien etyma are also found in Old Chinese, where, however, they are they more likely to represent ancient loans into Sinitic from Hmong-Mien rather than the other way around (pace Ratliff, 2004: 158–159). First of all, the Old Chinese forms (shú) “glutinous millet” (i.e. not rice), (tián) “field”, (lián) “sickle”, (jù) “cakes”, (dǎo) “pound, thresh” and (zèng) “steamer” are not reconstructible to Tibeto-Burman, of which Sinitic is but a subgroup (van Driem, 2005 and van Driem, 2007). Similarly, Ratliff relates Proto-Hmongic “seedling” and Proto-Mienic “seedling” to Middle Chinese (yāng), evidently due to a discrepancy in vocalism between the Hmongic and Mienic forms, and relates Proto-Hmongic “rice measure” to Old Chinese (liàng, liáng), but again neither etymon is well reflected in Tibeto-Burman outside of Sinitic. Moreover, not only are the earliest textual attestations of the Chinese forms “field”, “cakes”, “steamer” and the measure word absolutely ambiguous as to what kind of grain they refer to, though “field” may reflect a Hmong-Mien loan into Sino-Bodic rather than just Sinitic. Furthermore the form “cakes” is actually not an Old Chinese form, for its first known attestation occurs in the poetry anthology of the feudal state of Chǔ, entitled Chǔcí, dating from the Hàn period, whereas “sickle” likewise first occurs in the Hàn period as a western dialect word (Wolfgang Behr, personal communication).

The Proto-Mienic *hmei^B “husked rice” appears to correspond to Old Chinese (mǐ), and rice terms containing a bilabial nasal initial appear in other languages in the East of the Tibeto-Burman area, e.g. Bái me ³³ “husked rice”, Jinuo “rice”, Black Lahu mi ³³ “paddy”, Nusu me ³³ me ³¹ “rice”, Garo mi, Dimasa mai “rice”, Tangkhul ma “paddy”, Sgaw Karen me “boiled rice”. However, the meanings of these forms are disparate, viz. paddy, hulled rice, boiled rice, and the forms may represent mere look-alikes, since no regular phonological correspondence is yet known to obtain between them. Benedict (1972: 149) set up a Bodo-Koch proto-form *mey or *may “rice, paddy”, which Matisoff later inflated to Tibeto-Burman “*ma ⇔ *may or *mey” (2003: 216, 231) by adding a “monophthongal allofam” and stressing the uncertainty of the rhyme. In fact, no rice agricultural terminology can be confidently reconstructed for the Tibeto-Burman phylum as a whole, an issue addressed by Blench (2009).

The second language family aside from Hmong-Mien with an impressive reconstructible repertoire of rice agriculture terms is Austroasiatic. Gérard Diffloth (2005) has adduced the following eleven reconstructible Austroasiatic roots: “rice plant”, “rice grain”, “rice outer husk”, “rice inner husk”, “rice bran”, “mortar”, “pestle”, “winnowing tray”, *gu:m “to winnow”, “dibbling stick” and “rice complement”, i.e. accompanying cooked food other than rice. The comparative historical study of Austroasiatic is not as advanced or, for that matter, as tractable as that of Hmong-Mien, and perhaps for these reasons the current repertoire of reconstructible Austroasiatic rice agricultural terms is slightly more modest.

Rice cultivation terminology could have been borrowed into Sinitic from ancient Hmong-Mien rice cultivators at a time when Proto-Sinitic millet growers came into increasing contact with their southern neighbours. The main split in the Hmong-Mien family is between Hmongic and Mienic, and the scattered distributions of modern communities of either branch have approximately the same geographical range, which is roughly bisected by the Pearl River. On the basis of the historical sources, it has long been mooted that the geographical centre of gravity of the family would originally have lain further north along the middle Yangtze (Cushman, 1970). There is currently no palaeobotanical evidence for the co-cultivation of rice and foxtail millet along the middle Yangtze until around 3800 BC (Nasu et al., 2006). This brings us to a problem with rice agriculture archaeology.

The archaeology of rice agriculture is plagued by an empirical quandary. Archaeologists have looked for the remains of early rice agriculture and indeed found them at some sites and not at some others. The recovered remains of early cultivated rice are of differing antiquity and reflect distinct stages of domestication. Unsurprisingly, archaeologists have not found the remains of early rice agriculture in those places where they have not yet looked. Vast swathes of Asia covering the areas identified by rice geneticists (Londo et al., 2006, Molina et al., 2011) as harbouring likely sites for the domestication of Asian rice have not been subjected to systematic archaeological and palaeobotanical investigation.

Palaeobotanical evidence for an early domestication of Asian rice might as yet turn up in Northeastern India and the Indo-Burmese borderlands, regions which have maintained highly diverse rice cultures to the present day. At least five species of wild rice are native to Northeastern India, viz. O. nivara, O. officianalis (O. latifolia), O. perennis (O. longistaminata), O. meyeriana (O. granulata) and O. rufipogon, and reportedly over a thousand varieties of domesticated rice are currently in use in the region (Hazarika, 2005 and Hazarika, 2006). Despite the archaeological work conducted in the Ganges and Yangtze basins, much of the archaeology of ancient rice agriculture simply remains unknown because little or no substantive work has been done on the Neolithic in the most relevant areas, e.g. Northeastern India, Bangladesh, the Indo-Burmese borderlands and Burma. One archaeologist of cereal cultivation in China has already stressed the need to expand the scope of research beyond the Yangtze river basin into these areas, e.g. Lu, 2006 and Lu, 2009.

A related but distinct empirical issue is the archaeological recoverability of rice agriculture sites. The traces of ancient farming communities would tend to have been better preserved in the hill tracts surrounding the Brahmaputran flood plains than on the fertile fields themselves. Even in East Asia, most salvageable rice agriculture sites are in the foothills or at the base of the foothills (Nakamura, 2010). Yet the earliest rice-based cultures may first have developed on those very flood plains. Perhaps the remains of the first rice cultivating cultural assemblages lie buried forever in the silty sediments of the sinuous lower Brahmaputran basin. Maybe the palaeobotanical evidence for the earliest domestication of rice was washed out by the Brahmaputra long ago and now lies submerged in the depths of the Bay of Bengal.

By comparison, the absence of evidence for early rice agriculture of great antiquity in mainland Southeast Asia, despite the relatively more well researched archaeology of the region, presently embarrasses those who have lately taken to espousing Robert von Heine-Geldern's 1917 homeland theory for Austroasiatic around the lower course of the Mekong. However, the fact that the archaeology of Northeastern India, the Indo-Burmese borderlands and the northern Bay of Bengal littoral is virtually unresearched does not compromise homeland proposals in this region. As the old saw has it, the absence of evidence is not the evidence of absence. Political, cultural, geographical and logistic factors appear to have conspired to impede intensive archaeological research in a vast area extending from the lower Brahmaputran basin to the Tenasserim.

Against this background, the archaeology of rice agriculture has nonetheless produced important results. The domestication of japonica rice through genetic modification by selective breeding was possibly effectuated along the Yangtze around the beginning the fifth millennium BC by people, who previously relied far more heavily on the collecting of acorns, water chestnuts and foxnuts before becoming reliant on rice cultivation (Fuller and Qin, 2009, Fuller et al., 2007, Fuller et al., 2009, Nakamura, 2010, Ruddiman et al., 2008 and Zhao, 2010). Rice cultivation reached the Yellow River basin during the third millennium BC (Crawford and Shen, 1998) and Formosa and Vietnam between 2500 and 2000 BC (Higham and Lu, 1998), but only spread throughout the Indo-Chinese peninsula between 1500 and 500 BC (Weber et al., 2010). It has been claimed that rice may have been cultivated in the Gangetic basin as early as 7000 BC (Agrawal, 2002, Pal, 1990 and Sharma et al., 1980), but the current evidence for the actual domestication of rice in the middle Ganges dates from no earlier than the second half of the third millennium BC.

On the basis of evidence presented at the 4th International Conference on Austroasiatic Linguistics in 2009 (van Driem, in press), the early Hmong-Mien were identified as the original domesticators of japonica rice and the ancient Austroasiatics as the original domesticators of indica rice from the wild precursors O. rufipogon stricto sensu and rufipogon, var. nivara respectively. The ancient Austroasiatics were presumed to have acquired knowledge of rice agriculture from the ancient Hmong-Mien, and introgressive hybridisation with the already domesticated japonica were presumed to have introduced the traits sh4, rc and Prog1 into the nivara gene pool during or before this early contact phase.

The recent evidence presented by Molina et al. (2011) in support of a single domestication of Asian rice between 13,500 and 8200 years ago could simplify the 2009 hypothetical model. In the revised scenario, the ancient Hmong-Mien would still be the initial domesticators of rice, and the ancient Austroasiatics may still have acquired the technology secondarily, but this process would then have involved deriving indica from japonica rather than deriving indica from nivara rice. Moreover, an early contact phase lengthy enough to have left some linguistic residue no longer has to be assumed. However, at present, Molina et al. (2011) are unable to exclude the possibility that indica and japonica were independently domesticated, since the gene pools of the original wild precursors may have been lost or may still be inadequately represented by the samples hitherto studied by rice geneticists.

Interdisciplinary reconstructions of prehistory sometimes resemble a “château de cartes”, and, as scientific theories, there would be nothing particularly disturbing about them falling over in the face of new evidence. Carefully reasoned interdisciplinary models of prehistory endowed with a transparent and clearly articulated argument structure consisting of discrete testable subsidiary hypotheses based on facts from the contributing disciplines can be precisely readjusted so as to accommodate new inputs readily. Indeed, models must be malleable enough to accommodate and incorporate new evidence. Models which cannot be thus adjusted must be discarded. In this context, it can apparently not be reiterated too often that a proto-language can only be reconstructed on the basis of linguistic evidence, and that the linguistic ancestors of any modern language community were not necessarily all the same people as the community's biological forebears.

The linguistic evidence has singled out the linguistic ancestors of the Hmong-Mien and the Austroasiatic language communities as the likeliest candidates for the original domesticators of Asian rice. Yet this leaves us with the matter of localising the putative Hmong-Mien and Austroasiatic homelands in time and space. The modern distribution of Hmong-Mien language communities is shown in Fig. 1. The communities in Southeast Asia are generally more recent settlements, and mainland East Asia south of the Yangtze is the Hmong-Mien heartland. The ancient Hmong-Mien probably lived along much of the Yangtze river basin, where the currently available archaeological evidence would suggest that japonica rice was domesticated.

Rice genetics is, in fact, in partial disagreement with the inherently fragmentary archaeological record. Londo et al. (2006) identified three probable geographical ranges for the domestication of Asian rice on the basis of the geographical distribution of genetic markers in the wild precursor O. rufipogon (Fig. 2). Only one of the three regions has been subject to systematic palaeobotanical investigation. Moreover, the phylogenetic evidence adduced in support of a single domestication of Asian rice indicates that the “Indian/Indo-Chinese” population of rufipogon rice was directly ancestral to the first domesticated rice (Molina et al., 2011). The plot of the story will no doubt change as new data pour in from the different disciplines, and in this context it is relevant now to reassess some of the other linguistic palaeontological data for Austroasiatic.

Scholars have sought to situate the Austroasiatic Urheimat as far west as the Indus valley and as far east as the Yangtze delta or insular Southeast Asia. Today, the principal contenders for the Austroasiatic homeland are the Indian subcontinent, the Bay of Bengal littoral, mainland Southeast Asia and the middle Yangtze. The gaping lacunae in palaeobotanical research are convenient to the argument in favour of the middle Yangtze. Linguistically, the old hypothesis that proposed Old Chinese (jiāng) “Yangtze” to be a loan from Austroasiatic emboldened Pulleyblank (1983) to envision a major Austroasiatic presence all along the eastern seaboard from Việtnam to Shāndōng, and to impute an Austroasiatic ethnolinguistic identity to the Lóngshān horizon. This interpretation of the linguistic data has notably been challenged by Zhāng (1998).

As opposed to this gossamer evidence, nine solidly reconstructible etyma in the ancient Austroasiatic lexicon adduced by Diffloth paint the picture of the fauna and ecology of a tropical humid homeland environment: *mraːk “peacock Pavo muticus”, “tree monitor Varanus nebulosus or bengalensis”, “binturong Arctitis binturong”, “pangolin Manis javanica”, “bamboo rat Rhizomys sumatrensis”, “the Asian elephant Elephas maximus”, “mountain goat Capricornis sumatrensis”, “rhinoceros Dicerorhinus sumatrensis” and “buffalo Bubalus bubalus” (Diffloth, 2005: 78). The Austroasiatic root “bamboo rat” has found its way into Malay as a loan, but this does not affect the picture. Diffloth has argued that all of these species were never native to what today is China. This evidence therefore rendered the middle Yangtze homeland hypothesis less likely.

However, Diffloth offers both the Bengal tree monitor Varanus bengalensis and the clouded tree monitor Varanus nebulosus as possible referents for the Austroasiatic etymon . By the same token, it could be argued that the ancient Austroasiatics may have used the root just as aptly to refer to the Indian peafowl Pavo cristatus as to the Javan peafowl Pavo muticus, and that they may have used the root just as readily to refer to the Indian rhinoceros Rhinoceros unicornis or Javan rhinoceros Rhinoceros sondaicus as to the Sumatran rhinoceros Dicerorhinus sumatrensis. The Holocene distribution maps of the tree monitor (Figs. 3 and 4), the peafowl (Figs. 5 and 6) and the rhinoceros (Figs. 7–9) cover Southeast Asia, including the Chinese province of Yúnnán and Northeastern India, though the rhinoceros would exclude the Deccan plateau and indeed much of the subcontinent.

Even if we presume that semantic shifts were limited to species that to jungle dwelling humans resembled each other closely, then for the Austroasiatic root , the Late Holocene geographical distribution of the Chinese pangolin Manis pentadactyla would be just as relevant as the geographical range of the Sunda pangolin Manis javanica. The geographical distribution of the pangolin (Figs. 10 and 11) comprises all of Southeast Asia as far as the Wallace line as well as the northeastern Subcontinent at least as far east as the Nepalese district of Dolakhā, where I recently saw a pangolin, and East Asia south of the Yangtze, including the region around the mouth of the Yangtze.

Likewise, for the root , the range of the large bamboo rat Rhizomys sumatrensis may be augmented by the geographical ranges of the Chinese bamboo rat Rhizomys sinensis and the hoary bamboo rat Rhizomys pruinosus. These enlargements increase the range of possible referents for the bamboo rat etymon (Figs. 12–14), to comprise most of Southeast Asia and parts of mainland East Asia. As linguistic palaeontological evidence, however, this less salient species is inherently more problematic, since arguably this etymon could be transferred to any large rodent or even other small mammal.

The Holocene geographical distribution of the buffalo (Fig. 15) and the Asian elephant (Fig. 18) cover all of South and Southeast Asia as well as East Asia south of the Yangtze. The Holocene geographical range of the serow (Fig. 16) covers Southeast Asia, East Asia south of the Yangtze as well as the Himalayas and the Indo-Burmese borderlands. The binturong (Fig. 17) covers mainland and insular Southeast Asia all the way to the Wallace line as well as Northeastern India and the Indo-Burmese borderlands. In summary, the Holocene ranges of the species for which Austroasiatic has reconstructible etyma covers a vast area. The area of overlap delineated by the reconstructible fauna, however, more specifically comprises Northeastern India, the Indo-Burmese borderlands and Burma.

Linguistic palaeontology is, of course, not the only linguistic tool for localising the original homeland of a language phylum. Both the centre of gravity of the phylum on the basis on the geographical distribution of modern Austroasiatic language communities as well as the deepest phylogenetic divisions in the family tree point to the northern Bay of Bengal littoral (Fig. 19). The deepest historical division in the family's phylogeny lies between Munda in the west and Khasi-Aslian in the east, which would put the homeland on either side of the Ganges and Brahmaputra delta. Even the deepest division within the Khasi-Aslian trunk, i.e. the split into Khasi-Pakanic and Mon-Khmer, would suggest a point of dispersal for Khasi-Aslian between South Asia proper and mainland Southeast Asia proper.

Recent phylogenetic evidence has been adduced in support of a single domestication of Asian rice, but the findings do not, at present, exclude the possibility that indica and japonica rice originated directly from different O. rufipogon gene pools. A single domestication of Asian rice would simplify the 2009 model by reducing the necessity of assuming an early contact phase between ancient Hmong-Mien and ancient Austroasiatics lengthy enough to have left some linguistic residue. Phylogenetic evidence appears to indicate that the Indian/Indo-Chinese population of rufipogon rice was directly ancestral to the first domesticated rice. By contrast, the currently available archaeological evidence at present favours the Yangtze river basin as the locus of the domestication of Asian rice. However, the gaping lacunae in the palaeobotanical research on rice domestication span vast stretches of unresearched territory, including key areas identified by rice geneticists as likely sites for the domestication of Asian rice. In some relevant areas, the recoverability of possible rice agriculture sites might be problematic. The linguistic palaeontological evidence supports Hmong-Mien for the linguistic identity of early domesticators of Asian rice, and the Yangtze river basin is a likely location for the Hmong-Mien homeland. The linguistic evidence indicates that the ancient Austroasiatics also cultivated rice and supports the view that the early Austroasiatics may have been the people behind an independent domestication of Asian rice. Additional linguistic evidence from the reconstructible Austroasiatic lexicon indicates an Austroasiatic homeland in the northern Bay of Bengal littoral, more specifically the northeastern portion of the Indian subcontinent, the Indo-Burmese borderlands and Burma.