Microsoft Word L 2-03-Greenberg-paper doc

hound
is 
cognate to German 
Hund
, but 
hound
will have been replaced by 
dog
on the English list. 
We see then that glottochronology both excludes relevant evidence and weighs all 
items equally regardless of their wider distribution. 
It will perhaps have been noticed that the occurrence of borrowings between 
languages as a possible source of error in genetic classification has not been discussed. 
I do not consider this a serious problem. This is true not only because, in most 
instances, it only tends to occur exclusively or mainly in non-basic vocabulary. Even 
when it occurs in a large part of basic vocabulary, there is a more fundamental reason 
why it can be detected. This has once more to do with multilateral comparison. 
Consider, for example, a language like Turkish with numerous Arabic loanwords. 
Outside of the rarity of these words in basic vocabulary, there is the fact that Turkish 
cannot be a dialect of Arabic because the two are mutually unintelligible. But Arabic is 
clearly Semitic. If Turkish is then related to Arabic, the words generally acknowledged 
to be loans will have to be reassessed as cognates and Turkish will be a Semitic 
language. But we may then ask why it shows no independence within Semitic. 
Whenever it resembles Semitic, the resemblance is to Arabic, which is thereby 
identified as the loan source. 
A somewhat different sort of problem is presented by Quechua and Aymara which 
share numerous vocabulary similarities, many of them involving virtual identity of 
form. The question debated is whether all of the resemblances between the two 
languages are the result of borrowing, probably by Aymara from Quechua. If we 
consider the languages in isolation, it is difficult to reach a decision. However, they 
both belong to the Andean subgroup of Amerind, within which they do not form a 
special subsubgroup. The reason for believing that Aymara is related to Quechua is 
simply that they are both Andean languages. As such they show independent 
resemblances to other Andean languages. For example, an Aymara form not found in 
Quechua will occur as a cognate in Araucanian (another Andean language), while in 
other instances it will be a Quechua form not found in Aymara which has a cognate in 
Araucanian or some other Andean language. Note that it is not necessary to decide in 
every case whether a word common to Quechua and Aymara is a borrowing. Common 
membership in Andean is sufficient to show that they are related languages. 
We now come to the last of the three questions raised initially, what was there 
called the justification problem. Nothing stated here in regard to this is intended to 
suggest that other kinds of classification are not legitimate and important; e.g., the 
significance of typological classification for the study of language universals. 
Nevertheless 
qua
classification, genetic classification has a central position as indicated 
by the fact that it is the unmarked meaning of the term when linguists use the term 

---

The Methods and Purposes of Linguistic Genetic Classification
131 
“classification” without further qualification. The basic reasons appear to be the 
following. 
First, as compared to typological classification, it is unique in the sense that there 
can only be one correct one, whereas in regard to typology to ask which is the correct 
one is a meaningless question. The uniqueness of genetic classification is based, of 
course, on the fact that it reflects history, and history could have happened only one 
way. A by-product of this is
the application of its results to culture history. 
Areal classification is also important for history, but it assumes genetic 
classification as a basis and, as we have seen, the boundaries of linguistic areas are 
vague. We may sum up by saying that genetic classification is the only internal way of 
classifying language which is both unique and categorical. 
It is, however, the importance of genetic classification as the point of departure for 
historical-comparative linguistics that linguists think of first if they are asked to 
describe its significance, and this is the reason it dominated the study of language in the 
nineteenth century. Most of what we know about the processes of linguistic change 
derives from the methodology associated with genetic classification, especially for 
areas without written records. This is, in fact, a further reason within the history of 
linguistics itself for the dominant position of the genetic model in language 
classification. During the nineteenth century there was only one form of typological 
classification practiced to any significant extent: that into isolating, agglutinative and 
synthetic languages. And this classification was further associated in a vague way with 
one into analytic, synthetic and polysynthetic. This form of classification, as compared 
with the genetic, did not prove to be fruitful, and, particularly with the advent of the 
Neogrammarians of the latter part of the century, was relegated to a very marginal 
position within linguistics as a whole. 
There are several important relationships between these two modes of 
classification. One is in regard to typological sampling (Bell 1978). As far as possible 
in establishing implicational universals on the basis of typology, we wish to base the 
connection on historically independent cases, and, hence, considerations of both genetic 
and areal factors are important. There is a significant reciprocal value, however, for 
comparative linguistics deriving from typology in its diachronic aspect. The 
comparison of parallel typological developments in historically independent cases adds 
to our knowledge of diachronic processes, and thereby increases the scope of historical 
explanation and reconstruction. 
Finally, we may note that the family tree model, by means of which genetic 
linguistic classifications are frequently represented, has analogues in a number of other 
fields, in some of which it receives a historical processual interpretation, and in some of 
which it does not. The logical structure of such trees is as follows. The individual 

---

Joseph H. Greenberg
132 
members form a set generated by a one-to-many relation, hereafter symbolized as