Microsoft Word L 2-03-Greenberg-paper doc

hand
in English, but also concrete grammatical markers, e.g., the 
-er
of the 
adjectival comparative. In other words, our unit is the 
morpheme
as the term was used 
in American structural linguistics. Generativists employ the term 
formative
for roughly 
the same concept. 
Given the existence in all languages of numerous morphemes associating specific 
sounds with specific meanings, we can abstract from one or the other. If we consider 
the sound in abstraction from the meaning we have a phonological typology. For 
example, we could classify the languages of the world into those which have voiced 
stops and those which do not. In such a classification we are abstracting from the 
meanings of the forms containing voiced stops. 
The obvious counterpart of this is to consider meaning in abstraction from sound. 
The most interesting typologies here are those which involve grammatical morphemes. 
For example, we could classify languages which possess a morpheme for the dual 
number in the noun into one class, and those which do not into another. 
Classifications involving lexical items as such seem to be in many instances 
uninteresting. Thus we might ask whether languages had a word for ‘nose’, abstracting 
once more from the particular sounds involved in expressing this concept. However, 
this sort of typology is not 
always
uninteresting. We might, for example, want to find 
out just what concepts are expressed in all languages and which are not. Further, there 
are areas of vocabulary that are quite structured, though these are few. Among them 
would be numeral systems and systems of kinship terminology. Here complex 
typologies are possible and interesting. Much of the work done by ethnosemanticists 
falls into this area. To ask then whether a language has distinct terms for ‘mother’s 
brother’ and ‘father’s brother’ is interesting both because languages differ in this regard 
and because correlations with social structure can be established, as well as other 
semantic facts about the languages. Thus if a language has distinct terms for ‘mother’s 
brother’ and ‘father’s brother’, it almost always has separate terms for ‘mother’s sister’ 
and ‘father’s sister’. 
In the framework for typologies discussed thus far, there has been no provision for 
what is probably at present the most popular of all typologies, namely that which has to 
do with the order of morphemes or words. It would seem to involve a combination of 
form and content quite analogous to that
 
involved in the association of sound and 

---

Joseph H. Greenberg
116 
meaning. There is a formal aspect, namely whether something precedes or follows 
something else (which would correspond to sounds) and the grammatical categories 
involved (which seems to correspond to a kind of meaning). Thus if we state that a 
certain language belongs to the SOV type, the ordering of the three elements is a formal 
criterion akin to that of the sounds in lexical items, while the grammatical categories of 
subject, verb, and object are by contrast meaningful. In fact, we find across languages 
that order and grammatical morphemes are alternatives for expressing particular 
grammatical relations. For example, possession is expressed in some language purely 
by order, while some have a grammatical morpheme for the genitive, while still others 
use some combination of both. 
All this suggests that our attempt to define typological classification in terms of 
the arbitrariness of the sign, so that we basically had two kinds of typology, 
phonological and grammatical-lexical, to which we then added order typologies in an 
ad hoc
fashion, is not adequate, however useful, as an initial approach. 
If we consider for a moment the order typologies themselves which involved us in 
the theoretical problem with which we are now concerned, we may approach more 
closely to the essential features which distinguish typological classification. This has to 
do with the number of theoretical possibilities involved. Consider for a moment the 
typology which utilizes the order of S, O, and V. Logically there are only six possible 
orders and, of these, two are extremely rare. This strong limitation in possibilities 
applies also to lexical typologies of the kind exemplified above by the existence of a 
word for ‘nose’. There are only two possibilities. Either a language has a word for 
‘nose’ or it does not. 
Both the limited number of possibilities and the fact that these possibilities tend to 
be distributed very unevenly among languages (e.g., that SOV languages are very 
common and OSV languages exceedingly rare or perhaps even non-existent), bring it 
about that languages can quite easily belong to the same type “accidentally”, that is, 
from a historical point of view. Even where the number of logically possible types is 
quite large, as with systems of kinship terminology, the constraints both of cognitive 
and social origin are so powerful that the actually occurring systems are a very small 
proportion of the logically possible ones.
1
As a result languages may easily be similar 
typologically without a historical connection as the basis for the coincidence. An 
example of a phonological phenomenon for which this holds is tonality in Africa, East 
Asia and Mexico; with regard to word order, SOV in Somali and Turkish. It is, of 
1
An example is Nerlove and Romney (1967) dealing with sibling kinship terminology. Out of 
245 systems investigated, 240 fell into 18 of the 4,140 logically possible types. With a handful 
of exceptions most of these were in 12 types predicted in advance by a combination of marking 
theory and a cognitive principle of the avoidance of disjunct categories.

---

The Methods and Purposes of Linguistic Genetic Classification
117 
course, possible for a typological resemblance between two or more languages also to 
be genetic when the agreement results from common inheritance from an ancestral 
language. However, as we shall see more fully later, such resemblances which are both 
typological and genetic simultaneously play a very different role in the actual 
methodology of classification while they furnish certain kinds of insights regarding 
linguistic change which are not derivable from other sources. 
The problem of categorical versus prototypical definitions arises in reference to 
the delimitation of typological criteria. With regard to word order, the tendency of some 
analysts has been to classify languages in terms of two basic types: VSO and SOV. 
Since each of these is more or less associated with other criteria in a polar manner (so 
that, for example, almost all SOV languages are GN and virtually all VSO languages 
are NG), we may say that an SOV language which has GN order is more prototypical 
than one which does not. Similar problems arise at the logically lower level of the 
definition of the typological traits themselves. For example, a language like French in 
which adjectives normally follow the noun but a few may precede or follow, is less 
prototypically NA than Tagalog in which the adjective invariably follows the noun. 
A parallel problem arises regarding the meaning of grammatical categories in 
typologizing. For example, when we seek to identify genitive constructions on a 
universal basis in order to typologize them, what we find is a cluster of characteristics 
on the semantic side. In most languages, a construction which is used to express 
possession of a house or of domestic animals is likewise used to indicate a person’s 
relation to his own head, doubtless because one seems to have an analogous sort of 
control over it. But a person’s head is also part of his body and from this the extension 
to part-whole relations is not difficult. Hence we find a cluster of characteristics usually 
found to coöccur; our enumeration, of course, is by no means complete. However, we 
do find languages like Finnish in which there is a case form which expresses, among 
other things, possession, but also a
 
separate case called the partitive, which we would 
probably not want to identify with the prototypical possessive. Our purpose here is not 
a full discussion, which would obviously be complex and the subject of a separate 
study. We merely wish to point out that the problem does arise in the case of 
typological criteria, particularly in regard to grammatical categories such as “subject”, 
the crosslinguistic identification of which raises difficulties and concerning which the 
notion of prototypicality has, in fact, been utilized by many linguists. 
A further characteristic of typological classification is relevant in the context of 
the present discussion, namely that the number of possible typologies is infinite. There 
is, further, no contradiction if, in classifying languages along typological lines, two 
languages belong together in one typology and do not in another. Given the infinite 
variety of possible typological classifications it will of course result that very many of 

---

Joseph H. Greenberg
118 
them are quite pointless. A fruitful typological classification is one that shows strong 
correlations with one or more others suggesting some causal connection of a universal 
nature among the properties involved. It is, of course, for this reason that most recent 
work in typology has been in connection with the search for linguistic universals. When 
this occurs the common practice of typologists is not to talk of connections among 
typologies but to combine them in multidimensional typologies in which the separate 
dimensions are logically independent but empirically related. This once more shows the 
typical “arbitrariness” of typological procedures which
 
permits great freedom of 
manipulation in regard to the definition of types in the search for universal linguistic 
principles. 
Typological classifications may be hierarchical, but the hierarchies display the same 
characteristic of arbitrariness as the classifications themselves in the sense that has just 
been explained. For example, we might in a typology of phonological tone classify 
languages as being tonal or non-tonal. The tonal languages might in turn be divided into 
those which have level tones only, those which have contour tones only, and those which 
have both. We might also divide non-tonal languages into those which have phonological 
stress and those which do not. Clearly we have here a hierarchy within a typological 
classification. Moreover there is the same type of arbitrariness that we found to be 
generally true of typological classifications. We might for example have divided tonal 
languages into those in which there are significant limitations in their sequence based on 
the word as a unit and those which do not; that is into word-accentual and those which are 
not word-accentual. This would cross-cut the classification first described, but there would 
be no logical contradiction in this. It would simply be a question of fruitfulness in regard 
to further results as noted in the earlier discussion. 
There remains one important type of classification to discuss before we consider 
genetic classification in detail in relation to the questions raised at the beginning of this 
paper, namely areal classification. The problem with which areal classification deals 
arises in the following manner. If we plot on a map the geographical distribution of 
linguistic traits, we often find that this distribution is not a random one. This is 
equivalent to saying that they cluster in such a way that if languages which are 
continuous or not distant to each other share one trait they often share a whole series of 
others. However these traits must first be analyzed in order to determine the reasons for 
this non-randomness. A linguistic area is defined by a set of traits whose common 
occurrence in the languages has arisen by a process of linguistic contact over time. Just 
as we found that in particular instances a trait might be both genetic and typological, so 
we may find that a particular trait may be both typological and areal without there being 
any contradiction. However it cannot be both areal and genetic at the same time since 
this would involve two different and mutually exclusive historical explanations. 

---

The Methods and Purposes of Linguistic Genetic Classification
119 
Initially we shall only consider typological traits and, in fact, these are the ones 
most commonly employed in defining linguistic areas. However, the actual distribution 
of typological traits found on a map is, as it were, a surface phenomenon. This is 
because resemblances can result in three different ways and only one of them is 
relevant for areal classification. The first of these is sheer accident. For example, given 
the large number of SOV languages in the world a whole group of contiguous 
languages could share this characteristic for accidental reasons. The term “accidental” 
in this context means historically independent. As is evident already from the statement 
that areal resemblances are those arising from language contact, we see that areal 
classification shares one important property with genetic classification: namely that it 
is, as opposed to typological classification, historical, whereas typology is ahistorical. 
By this we mean that a typological resemblance remains a typological resemblance 
whether it results from historical processes or not. 
The second type of resemblance in a set of geographically contiguous languages 
are those which result from unchanged genetic inheritance from an ancestral language. 
These also are not relevant for areal classification since they do not result from 
language contact. On the other hand, it does count as evidence in defining a linguistic 
area if a set of contiguous languages all develop a dual number not inherited from a 
common ancestral language and as the result of a historical process by which bilingual 
speakers innovate this category in one of the languages they speak because of the 
structural influence of the other. Since what we are interested in here is the influence of 
one language on the other, we need not confine ourselves to the typological traits which 
we have just been considering. Hence we can include loan words, which, of course, 
involve resemblance in form and meaning simultaneously and are thus not typological. 
Thus far we have been considering the types of traits which are significant for 
areal classification, but we have not shown how they result in an areal classification of 
languages. The possibility of classifications of this kind depends on the existence of 
situations in which particular languages and sets of languages have more similarities 
resulting from contact in one geographical direction than another. A classic case is that 
of the languages of the Balkans. Romanian, Albanian, Bulgarian, and Greek share, in 
addition to many loan words which have diffused from one of the languages to one or 
more others, a series of typological characteristics. These include the absence of an 
infinitive, a suffixed definite article, and the formation of a future tense by means of a 
particle which derives from a verb meaning ‘to wish’, or in the case of Romanian is the 
conjugated verb itself. Serbo-Croatian shares these characteristics to a lesser degree, 
and in certain respects, e.g., the possession of vowel length, resembles Hungarian to the 
north, which is not geographically a Balkan language. 

---

Joseph H. Greenberg
120 
The Balkan languages thus form what is sometimes called a