MATERIALS AND METHODS
The 12 living language families in Europe fall into five language phyla as follows (1):
- Indo-European (Albanian, Baltic, Celtic, Germanic, Greek, Romance, and Slavic);
- Finno-Ugric [Finnic and Ugric (Hungarian)];
- Altaic (Turkic);
- Afro-Asiatic [Semitic (Maltese)]; and
- Language Isolates (Basque).
A linguistic distance matrix of language-family relationships was constructed by setting the Baltic-Slavic distance to 1 (these are the only two Indo-European families for which close genetic affinities are generally accepted, see ref. 1), all other distances between language families within a phylum to 2, and distances between language families belonging to different phyla to 4. Thus, language distances mostly contrast intraphylum and interphylum distances.
A geographic distance matrix between all pairs oflanguage families was computed from great-circle distances between subjectively chosen centers of language-family regions. Genetic distances were calculated by using frequency data for human blood antigens, enzymes, and proteins of 26 genetic systems, each for a different subset of 3369 localities across Europe.
Because of the different spatial sampling for each genetic system, we computed genetic distances sepa- rately for each system. All localities were also assigned a language-family affiliation. The systems and the sources of the data are described elsewhere (5-7).
Sample sizes range from 50 to many thousands ofpersons. Previous work (5) has shown that the simplest of these distances, that due to Prevosti et al. (8), provided essentially the same information as more elaborate formulations. It was, therefore, adopted here.
To allowforpossible bias due to different ranges ofgene frequencies, we also standardized the distances. For each system we first calculated genetic distances for all pairs oflocalities and subsequently averaged over all locality pairs representing a particular pairwise combination of language families.
This yielded, for a given system, an average genetic distance for each pair oflanguage families. However, for some systems, we lacked localities to represent one or more language families and could not compute distances for certain pairwise combinations of language families.
This resulted in genetic distance matrices for some systems with missing values for some pairwise comparisons. Since different genetic systems are based on different sets of localities, the particular pairwise combinations missing in the genetic distance matrices vary among systems.
The final genetic distance matrix (Table 1) was obtained by averaging over all systems. Nine of the language families are well represented by genetic systems, but genetic distances for Semitic, Baltic, and Albanian are based on only seven, three, and two systems, respectively.
These few systems may furnish unreliable estimates of distances between language families. For this reason, we analyze both the reduced set of 9 language families and the total set of 12. Our conclusions are based largely on the 9 families, with added consideration of the larger data set when appropriate.
The average genetic distance matrix between pairs of language families was subjected to standard numerical tax- onomic clustering and ordination procedures (9). Hierarchic classifications of the language families were achieved by UPGMA (unweighted pair-group) clustering of the average distance matrices (9). Ordinations ofthese distance matrices were obtained by nonmetric multidimensional scaling in three dimensions. All computations were carried out by the NTSYS program (10).
Congruence between genetic, linguistic, and geographic distances was tested for significance as described (5).
Using methods ofquadratic assignment (11), we calculated pairwise Mantel matrix correlations (12, 13) and investigated three- way relations between the distance matrices by computing partial correlations (14). These correlations were tested for significance by Monte Carlo permutation methods. These computations were carried out by using the R package for multivariate data analysis (15).
RESULTS
Fig. 1 shows the hierarchic clustering ofgenetic distances for 9 language families.
A Germanic-Celtic cluster isjoined later by Romance, and a Slavic-Ugric cluster isjoined by Turkic.
Finnic, Basque, and Greek are outliers to these clusters. Including Albanian, Baltic, and Semitic in the analysis changes the phenogram by affiliating Greek with Albanian, Baltic with Turkic, and clustering Germanic-Celtic with Slavic-Ugric before adding Romance. Semitic, Basque, and Finnic are outliers to the clusters of 12 language families.
An ordination ofthe genetic distance matrix (Fig. 2) depicts the relative genetic distances between the nine language families.
Finnic and Basque are outliers at opposite ends of the ordinated space. The Celtic-Germanic and Slavic-Ugric language-family pairs are evident along the first axis which runs roughly East-West. The second axis approximates a North-South gradient. In the minimum spanning tree, Turkic links Ugric and Greek with Romance. The position of Romance is central on the first and second axes, but isolated by the third, explaining its variable affiliation during clustering.
Genetic distance (Gen) correlates significantly with geog- raphy (Geo) but not with language (Lan). The pairwise correlations of distance matrices based on nine language families are as follows: Gen x Geo = 0.468 (P < 0.01), Gen x Lan = 0.182 (P>0.05), and Geo x Lan = 0.177 (P>0.05).
The partial correlations are (Gen x Geo)-Lan = 0.451 (P <0.01) and (Gen x Lan)-Geo = 0.114 (P > 0.05). Geography
determines 20.3% of the variance of the genetic distances, language determines only 1.0%o, and factors common to geography and language determine 1.6%. One might have expected a high and significant Gen x Lan correlation, because speakers ofa particular language (family) tend to be found settled near each other.
In other words, when geographic distances between samples are small we expect linguistic distances to be small, and vice versa.
Ifgeography is likewise correlated with genetics, then genetic and linguistic distances should also be positively correlated.
But the Geo x Lan correlation in this study is low because centers of language phyla are positioned in Europe both relatively close and far apart spatially causing the relationship between geography and language to break down. Therefore, the common effect of geography does not produce a high correlation between genetics and language. This finding contrasts with the significant correlation between genetic and language distances reported by Sokal (5). In that study correlations among genetics, linguistics, and geography were calculated for pairwise locality distances. Since a finer scale ofdistances was used, both the correlation ofgeography and of language and the common effect of geography were greater.
DISCUSSION
This study shows that genetic distances between the European language families do not reflect their accepted linguistic relationships.
If we group the language families by their linguistic origins, there should be a cluster of the Indo- European language families, Baltic and Slavic being most closely related, a separate branch for the Finnic and Ugric speakers, and separate coordinate branches for the Turkic, Semitic, and Basque language families.
The genetic distances of some interphylum language-family pairs, such as those between Slavic and Ugric speakers, or between Turkic and Ugric speakers, however, are closer than some distances within a phylum, as between Greek and Celtic speakers or between Finnic and Ugric speakers.
The low matrix correlation between genetics and language confirms the lack of agreement between presumed language phylogeny and the observed genetic distances.
If genetic distances reflect geographic proximity, we should be able to predict genetic affinity from a clustering of the great-circle distances between language families.
This yields Germanic-Celtic and Romance-Basque as mutually closest pairs, with the two pairs together forming a major cluster.
Likewise, Slavic-Ugric and Greek-Turkic both cluster as pairs. Finally, Finnic is an outlier.
Clustering all 12 language families by great-circle distance enlarges the Greek Turkic cluster into one that also includes Albanian and Semitic and places Baltic-Finnic as an outlying pair. Considerable concordance between geographic proximity and observed genetic relations is evident and confirmed by the significant correlation between geographic and genetic dis- tances.
The role of geography can also be seen in the ordinations.
Disregarding the outliers, Fig. 2 demonstrates an East-West separation of geographically adjacent language family pairs, Germanic-Celtic from Turkic-Greek with Slavic-Ugric intermediate.
The small Greek-Albanian genetic distance is also consistent with geographic proximity, although we attach less confidence to this value.
The observations made here are supported by a largely geographic clustering of European map quadrats characterized by gene frequencies (16).
A geographic gene flow model does not, however, explain why the Basque and Finnic language families are outliers both in the ordinations and phenograms, nor why Semitic is an outlier in the extended dataset. These results reflect the distant origins of speakers of these language groups. The Finnic language family is given its unique genetic profile by inclusion of the Lapps. These populations, ethnically different from other Finnic speakers, apparently migrated to northern Scandinavia from northern Eurasia (17).
The Basques have long been an isolated enclave, presumably descended from the pre-Indo-European inhabitants of Europe (18, 19).
The Semitic speakers have North African origins. These results suggest that some modification of the strictly geographic gene flow model by language origin may provide greater concordance with the genetic relationships between language families.
We conclude that affinities between modern European gene pools have been formed primarily by relatively short- range gene flow between geographically adjacent populations. Yet, between the speakers of some language families and their geographic neighbors, there are genetic differences that apparently reflect their remote historical and linguistic origins.
We thank Barbara Thomson for computational assistance. Prof. L. L. Cavalli-Sforza and Dr. Neal Oden provided helpful comments.
This research was supported by Grant GM28262 from the National Institutes of Health.
This article is contribution 681 in Ecology and Evolution from the State University of New York at Stony Brook.
ЗОШТО МЕНУВАЛЕ НАСЕЛЕНИЕ?..
по јазик и гени се слични.
A керим верата...
