The Sino-Tibetan tree

Two views of the homeland compete. The first places Proto-Sino-Tibetan in northern China, among the Neolithic millet farmers of the upper and middle Yellow River (the Yangshao and later Majiayao cultures). Both computational phylogenies of 2019 point that way: Laurent Sagart and colleagues dated the family's first split to around 7,200 years ago, Menghan Zhang and colleagues to around 5,900. On this view one group of farmers stayed in the Yellow River basin and became the ancestors of Chinese, while others moved west onto the Tibetan plateau and then south along the great river valleys (Salween, Mekong, Irrawaddy and Brahmaputra) into the Himalayas and Southeast Asia. Several ancient-DNA studies report ancestry from northern Chinese farmers in present-day populations of the Tibetan plateau, which fits this picture without proving the language link.

The second view looks to where the family is most diverse. Many small and deeply divergent branches are packed into the eastern Himalayas, northeast India and the Myanmar borderlands, and by the usual reasoning of historical linguistics a family's greatest diversity lies near its origin. George van Driem has argued for a homeland in Sichuan, and Roger Blench and Mark Post for one in the eastern Himalaya or northeast India. Blench has also questioned whether the earliest speakers were farmers at all, arguing that the family shares little clearly reconstructible farming vocabulary.

Linguistic palaeontology is weak here because so much of the family is thinly documented, and the numbers from phylogenetic dating depend heavily on how cognates are coded and how the tree is calibrated. What is not in doubt is the later history: Chinese spreading south with the Chinese state over three millennia, Tibetan with the Tibetan empire of the 7th–9th centuries, and Burmese into the Irrawaddy valley from the 9th century.

The first thing a reader notices is the syllable. In most Sino-Tibetan languages a word root is a single syllable, and in many of them syllables carry lexical tone: Mandarin mā, má, mǎ and mà mean 'mother', 'hemp', 'horse' and 'scold'. Tone is not inherited from the ancestor, though. Old Chinese, Written Tibetan and many Himalayan languages had none, and tones arose separately in Chinese, Burmese and Lhasa Tibetan as syllable-final consonants and voiced initials were lost. Amdo Tibetan and many Himalayan languages are still toneless.

Word order divides the family. Chinese and Karen put the verb before the object, like their Tai, Hmong and Mon-Khmer neighbours; almost all the rest put it at the end of the clause and use postpositions, with relative clauses and genitives before the noun. Many languages count things with numeral classifiers, as in Mandarin sān běn shū 'three volume book', 'three books'. Morphology ranges from almost none, in Chinese, to elaborate verb agreement for subject and object in the Kiranti languages and rGyalrong. Old prefixes such as a causative s- leave traces across the family, and many Tibeto-Burman languages mark the agent of a transitive verb with an ergative case ending, often only when it needs emphasis.

The family also has an unusual range of writing systems. Chinese characters are the oldest continuously used script in the world. Tibetan and the Meitei and Lepcha scripts derive from Indian Brahmi, Burmese from the Mon script, and the Tangut of the Western Xia state wrote their language, from 1036, in a logographic script of their own modelled on the look of Chinese characters. The Yi script, Fraser's Lisu alphabet and missionary Latin orthographies add further variety.

The biggest open question is the shape of the tree. The standard model, going back to Benedict, divides the family into Sinitic and Tibeto-Burman. Both 2019 phylogenies support it, with Sinitic splitting first. But no one has identified sound changes or grammatical innovations shared by all Tibeto-Burman languages and absent from Chinese, which is the usual test of a branch. George van Driem therefore rejects Tibeto-Burman and treats the family, which he calls 'Trans-Himalayan', as a 'fallen leaves' collection of some forty well-defined subgroups whose higher-level arrangement is not yet known. Others, including Scott DeLancey and Randy LaPolla, argue over whether the verb agreement of Kiranti and rGyalrong is inherited from the ancestor or developed later, which bears directly on what Proto-Sino-Tibetan was like.

Individual languages also resist placement. Bai of Yunnan has so many layers of Chinese loans that scholars disagree whether it is a sister of Chinese or a Tibeto-Burman language that borrowed heavily. The position of Karen, of Tujia, and of isolated Himalayan languages such as Lepcha is similarly unsettled.

Dating remains loose. The two 2019 studies put the first split around 6,000 and 7,000 years ago respectively, some scholars favour an earlier date, and the results depend heavily on method. Outside links are more speculative still. Laurent Sagart has proposed a Sino-Austronesian family on the basis of shared vocabulary; Sergei Starostin and others grouped Sino-Tibetan with Yeniseian, North Caucasian and Na-Dene as 'Dene-Caucasian' (or Sino-Caucasian). Neither has won general acceptance, and most specialists regard Sino-Tibetan as a family with no demonstrated relatives.