Microsoft Word L 2-03-Greenberg-paper doc

s
and *
z
(the 
latter often becoming 
r
). Yet simply a comparison of certain related English and 
German words will show instances where the expected outcome has been overridden by 
non-phonetic factors: English 
was
: German 
war
; 
hare
: 
Hasen
; 
born
: (
ge-
)
boren
; 
rose
: 
Rose
. In general, across 
n
languages there will be 2
n
sporadic correspondences. 
These are just some of the reasons why, as all sophisticated etymologists know, 
etymology can never be a completely exact science in which all problems can be solved 
by the application of rigorous methods. We are dealing with probabilities, which are, 
however, in many instances very high. 
We now come to the question of the actual methodology of classification in the 
light of the characteristics of genetic resemblances at the trait level, which figured in 
the previous discussion. What we are interested in here is the higher level of languages 
as such and their genetic classification. In doing so, we consider, in regard to each 
principle, both the positive methodological procedure it gives rise to and the 
consequences of its disregard which leads in each case to a specific and often widely 
held fallacy standing in the way of progress in regard to the whole problem of genetic 
classification. 
The first of these principles flows from our consideration of the nature of genetic 
and typological resemblances. There are, as has been seen, resemblances which are 
purely genetic and those which are both typological and genetic; for example, 
agreement in certain features of inherited word order among languages of the same 
genetic stock. However, in going about classification, there are two reasons for 
disregarding the latter in carrying out a genetic classification. One is that, given the 
small number of typological alternatives, the possibilities of accidental convergence are 
high. The other is that the very possibility of distinguishing typological resemblances 
which are also genetic from those which are only typological depends logically on the 
prior establishment of a genetic classification. The use of typological criteria to classify 
languages genetically, at least as soon as one passes beyond the most obvious 

---

Joseph H. Greenberg
124 
groupings, was very common in the nineteenth century. The problem is now much 
better understood, but arguments of this type are still fairly persistent, generally in 
negative argumentation, as when it is asserted that a particular non-tonal language 
cannot be genetically affiliated to a group in which the other languages are tonal. 
Since genetic non-typological resemblances were defined earlier as those 
involving sound and meaning simultaneously, what this means is that, in effect, we 
shall begin with lexical items as well as grammatical morphemes, considering the latter 
with regard to both sound and meaning. We shall call such grammatical resemblances 
concrete, as distinct from those which are typological. For example the agreement of 
English and German in having an adjectival comparison marker 
-er
is both concrete and 
typological, while the agreement of French and Tucanoan, a group of South American 
languages in having masculine and feminine gender is not. Concrete grammatical 
markers are extremely valuable as evidence in carrying out genetic classification and 
they figured in a central way in the earliest work on Indo-European. However, lexical 
comparisons are, so to speak, the bread and butter of genetic classification for two 
reasons. One is that they are always present, at least in so-called basic vocabulary. 
There is always a word for ‘nose’, but relatively few languages have overt markers for 
the comparative of the adjective. The second is purely practical. There is a vast number 
languages in the world, some of them now extinct, for which these are essentially all 
the data that we have. 
In moving from the trait to the language level, we shall necessarily be concerned 
not with single resemblances in sound and meaning, but their clustering in such a way 
as to lead to the grouping of whole sets of languages. This aspect of method, namely 
the relationship between the trait and the languages level, brings into play two 
important considerations: the relative independence of each trait and its relative 
weighting. 
Essentially each item is independent. We may state this in the form of a maxim. 
Just because you call a