Microsoft Word L 2-03-Greenberg-paper doc

---

L
ANGUAGE AND
L
INGUISTICS
2.2:111-135, 2001
The Methods and Purposes of Linguistic
Genetic Classification
*
 
Joseph H. Greenberg 
Stanford University 
This paper discusses three questions relating to genetic classification. First, 
regarding the criterion problem, it concerns the nature of the linguistic 
resemblances and distinguishes the different properties and characteristics of 
typological classification, areal classification and genetic classification. Secondly, 
with regard to the methodological problem, it discusses several principles of 
genetic classification and considers both the positive application and the limit of 
the three methods of genetic classification, namely comparative method, 
multilateral comparison and glottochronology. Finally, with regard to the 
justification problem, by comparing the genetic classification with both the other 
two classifications and other fields of knowledge, it provides explanations why 
genetic classification has had a central and unique position in linguistics. 
Key words: linguistic classification, typological classification, areal classification, 
genetic classification, methodology 
Like any other set of objects, individual languages can be classified by many 
different criteria or combinations of criteria. By a classificational criterion will be 
meant a property or set of properties such that all the objects which possess them 
belong to the same set and those which do not belong to different sets. Moreover every 
object belongs to some set and no object belongs to more than one. The sets that result 
are said to be mutually disjoint and exhaustive of the universe of objects being 
classified and to constitute a partition. 
The foregoing is, of course, based on the traditional notion of classification in 
which the ideal is to specify the necessary and sufficient conditions for any group of 
objects to constitute a class. Such a classification is often called categorical. In the last 
two decades, however, the idea that it is justifiable and useful to relax such 
*
Professor Greenberg contributed this paper to a conference at the Center for Advanced Studies 
in Behavioral Sciences at Stanford, organized by L.L. Cavalli-Sforza, M. Feldman, and myself. 
It was to appear in a volume of proceedings, which unfortunately never materialized. The 
paper discusses central issues in theoretical linguistics and contains numerous important 
insights. With Professor Greenberg’s consent and support from the editors of 
LL
, it is 
published here. 
－
W. S-Y. Wang 

---

Joseph H. Greenberg
112 
requirements has been widespread and is generally formulated by means of the concept 
of “prototype”. We have in place of necessary and sufficient conditions a cluster of 
properties which are empirically found generally to coöccur, although not in every 
instance. Objects which most fully partake of this cluster of characteristics are then said 
to be prototypical. 
Such an approach to classification is in many instances sensible and useful, but it 
seems undeniable that categorical classifications do exist and that in a sense the 
approach through prototypes is a form of “sour grapes”. Put perhaps paradoxically, 
categorical classification is the prototypic form of classification. For the moment at 
least, we shall adhere to it, so that by classification 
tout court
we shall mean categorical 
classification. 
There is a further general characteristic often found in classifications which we 
may call hierarchy. There are levels of classification based on the logical property of 
containment. An obvious example in linguistics is genetic classification of languages, 
and it finds expression in the family tree diagram. Thus, English belongs to the class of 
Germanic languages, while the Germanic languages in turn are contained in the higher 
grouping known as Indo-European. As we shall see later, this same property of 
hierarchy is found elsewhere, such as in typological and areal classifications. However, 
where the lowest level is not categorical it may have somewhat different logical 
characteristics with regard to the property of logical containment. 
There is yet another factor to be considered in regard to classifications and which 
arises in more than one of the major kinds of linguistic classifications. When we carry 
out a classification in which languages as wholes figure as units (as when in genetic 
classification we place English and German together as members of the Germanic 
family of languages, or as in the nineteenth century morphological typology we classify 
Turkish and Tamil together as agglutinative), we can distinguish two levels in dealing 
with the evidence. One is a lower level of individual resemblances; e.g., cognates in the 
case of genetic classification and individual typological resemblances or what are 
sometimes called the dimensions of a typology. For instance, in regard to word-order 
typology, in appendix II (Greenberg 1963a) 24 types of languages are distinguished 
based on subject-verb-object order, the relative position of a noun and its dependent 
genitive, the existence of prepositions as against postpositions, and the relative order of 
adjectives and the nouns which they modify. On this basis Hindi, Mordvin (a Finno-
Ugric language), Japanese and many others are classified together under Type 23. This 
common membership is based on agreement in the four factors mentioned above; they 
are all SOV, have the genitive preceding the governing noun, have postpositions and in 
all of them the adjective regularly precedes the noun. Similarly, in areal classification 
the individual traits such as the existence of a suffixed article which help to delineate 

---

The Methods and Purposes of Linguistic Genetic Classification
113 
the Balkan 
Sprachbund
are on a different level than the languages themselves that are 
being grouped together areally. 
The individual comparable properties may be called traits of the classification, as 
against the level of classification proper in which languages as a whole are assigned to 
a single group. Many important questions arise regarding this relationship; for example, 
the number and degree of independence of the various traits, and whether or not (as in 
the typology of word order) these traits are organized on a set of dimensions. In the 
subsequent discussion we shall talk about the trait level and the language level when it 
is necessary to make this distinction, regardless of the type of classification being 
discussed. 
When linguists talk of the classification of languages and do not add 
qualifications, as when they say that English is to be classified as a Germanic language, 
they are employing what is often called genetic, or historical linguistic classification. In 
the orthodox view at least, such a classification is categorical: if English is a Germanic 
language, then it cannot be a Romance language. 
The basic purpose of the present discussion is to answer three distinct but related 
questions relating to genetic classification. Since, when classifying, whether 
categorically or prototypically, we are always concerned with resemblances, the first 
question we ask concerns the nature of the resemblances which are to be considered 
relevant to any particular kind of classification, as distinguished from other modes. We 
may call this the