Weng Z, Sokal R R
Department of Ecology and Evolution, State University of New York at Stony Brook 11794-5245, USA.
Hum Biol. 1995 Aug;67(4):577-94.
A series of tests was undertaken to relate lexicostatistical dissimilarities (LAN) among 48 Indo-European languages to distances representing various causal hypotheses. The comparison is limited to languages currently spoken in Europe. The putative causal distance matrices include (1) geographic (GEO) distances between the languages, (2) distances representing the origin of agriculture (OOA), (3) distances representing a model postulated by C. Renfrew (REN) concerning transformations that gave rise to the major Indo-European language families in Europe, and (4) distances representing a competing hypothesis by M. Gimbutas (GIM) concerning the origin and spread of Indo-European languages in Europe. Pairwise Mantel tests of the matrices show that OOA correlates better with LAN than does REN, supporting Renfrew's basic hypothesis of the dispersal of the Indo-European languages with the spread of agriculture but showing less effect for his postulated transformations. Partial correlation of LAN with OOA when GEO is held constant is significant at p = 0.004, whereas REN is no longer correlated with LAN when GEO is held constant. When repeated for only seven languages chosen to represent the seven major families of Indo-European languages currently spoken in Europe, the results differed appreciably, yielding a negative, albeit nonsignificant, partial correlation between OOA and LAN when GEO is held constant. This apparent contradiction led us to develop some new statistical approaches to examine, confirm, and explain the patterns. Decomposing the Mantel correlation coefficients for the 48 Indo-European languages into several additive correlation components showed that much of the positive component of the correlation coefficient was contributed by LAN, OOA correlation within language families, particularly within the Germanic family, covering up the negative contributions between language families. The differentiation of the seven major Indo-European language branches in Europe seems unrelated to the times of the origin of agriculture. This finding fails to support the fundamental assumption of Renfrew's hypothesis. There are also no significant correlations between LAN and REN or GIM. A series of Monte Carlo experiments confirmed these findings. Consideration of the accumulated evidence from genetics supports the model of demic diffusion during the origin of agriculture. However, published genetic studies and the present study lend no support to the notion that the early farmers were indeed the Indo-Europeans.
我们进行了一系列测试，以将48种印欧语系语言之间的词汇统计差异（LAN）与代表各种因果假设的距离联系起来。比较仅限于欧洲目前使用的语言。假定的因果距离矩阵包括：（1）语言之间的地理（GEO）距离；（2）代表农业起源（OOA）的距离；（3）代表C. 伦弗鲁（REN）提出的一个模型的距离，该模型涉及导致欧洲主要印欧语系产生的转变；（4）代表M. 金布塔斯（GIM）关于印欧语系在欧洲的起源和传播的一个竞争性假设的距离。对这些矩阵进行成对曼特尔检验表明，与REN相比，OOA与LAN的相关性更好，这支持了伦弗鲁关于印欧语系随着农业传播而扩散的基本假设，但他所假定的转变的影响较小。当GEO保持不变时，LAN与OOA的偏相关性在p = 0.004时显著，而当GEO保持不变时，REN与LAN不再相关。当仅对选择用来代表欧洲目前所说的印欧语系七个主要语族的七种语言重复进行该测试时，结果有明显差异，当GEO保持不变时，OOA与LAN之间产生了一个负的（尽管不显著）偏相关性。这种明显的矛盾促使我们开发一些新的统计方法来检验、确认和解释这些模式。将48种印欧语系语言的曼特尔相关系数分解为几个相加的相关成分表明，相关系数的大部分正成分是由语族内部，特别是日耳曼语族内部的LAN与OOA相关性贡献的，掩盖了语族之间的负贡献。欧洲印欧语系七个主要分支的分化似乎与农业起源的时间无关。这一发现不支持伦弗鲁假设的基本假设。LAN与REN或GIM之间也没有显著相关性。一系列蒙特卡洛实验证实了这些发现。对遗传学积累证据的考量支持了农业起源时期人口扩散的模型。然而，已发表的基因研究和本研究均不支持早期农民确实是印欧人的观点。
