Old Chinese Vocabulary: A Historical Perspective
[译]: 古汉语词汇：历史观研究
Sergei [Sergai] A. Starostin 斯塔罗斯金

Abstract 摘要
In this paper I would like to discuss a rather important methodological problem: does historical linguistics possess an objective procedure of evaluating proposed hypotheses concerning genetic relationship of language?

The procedure that I propose below, is the following:

a) to prove that two (or more) languages or linguistic families are related, we must know the set of regular phonetic correspondences connecting those languages. Otherwise any discussion is futile (all proposed equations may be due to chance). This is the standard demand of comparative linguistics.

b) the languages (or linguistic families) compared should share a significant part of basic vocabulary, and the items compared should match each other according to the set of correspondences demonstrated during the step a). This is also a common demand, but it is usually much less clear than the first one. What is basic vocabulary? What part of it is significant? I dare to propose here a test that appears (at least in my experience) to work in all cases of established genetic relationship.

As a rather quick way to test the results of comparison we may take the list of 35 most stable meanings proposed by S.Y. Yakhontov. They include the following (in English alphabet order): “blood, bone, die, dog, ear, egg, eye, fire, fish, full, give, hand, horn, I , know, louse, moon, name, new, nose, one, salt, stone, sun, tail, this, thou, tongue, tooth, two, water, what, who, wind, year’. Actually, the stability of some items in Yakhontov’s list raises doubts (this concerns, e.g., the items ‘one’ and ‘this’). We could easily choose some other list, but this one has an advantage of being already tested on a great many linguistic families of the world. The compared items should completely match semantically (i.e., correlations like ‘fire’:’hot’ or ‘water’: ‘flow’ are not taken into account – in order to exclude discussion of the semantic plausibility of comparison).

I maintain, that in all known cases of established genetic relationship this test yields following results:

a) closely related languages (like Slavic or Germanic) have about 30 or more related items within the 35-wordlist.

b) more distantly related languages (on the level of Indo-European) have more than 15 related items within the 35 wordlist. To establish the precise nature of relationship (in order to distinguish, e.g., the Balto-Slavic level from the Indo-European level) we have to resort to other, more precise, statistical methods.

c) if the compared languages have from 5 to 15 related items within the 35-wordlist, it means that we can suppose a still more distant relationship between them. The precise nature of relationship is difficult to establish (it may be very archaic, like Nostratic, or somewhat more close, like Uralic or Altaic; other statistical methods should be used to obtain more precise results in cases like that).

d) if the languages compared have less than 5 common items in the 35-wordlist, it means either that they are not related at all (and the existing common items must be explained by pure chance or by borrowing), or that the common words may be in fact the ‘Proto-World’ heritage – if one believes in monogenesis. We will not discuss the latter hypothesis here: obviously, if one proposes a theory of genetically relating two languages, he implies that they are more closely related to each other than to all other languages of mankind.