Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Annual Reviews is collaborating with JSTOR to digitize, preserve and extend access to Annual
Review of Anthropology
LINGUISTIC TYPOLOGY
Bernard Comrie
INTRODUCTION
145
0084-6570/88/10 15-0145$02.00
Areal Typology
Linguistic typology, genetic classification, and areal classification are in
principle distinct enterprises. Thus a classification valid in one is no more
likely than any other arbitrary classification to be valid in either of the other
two. While linguists generally recognize this today, the three enterprises have
been confused in the past. Armenian was wrongly classified as an Iranian
language because the overwhelming similarities in vocabulary between Arme-
nian and the Iranian languages were not recognized as the result of Armenian
borrowing from various Iranian languages. The Uralic languages were
assigned to the same genetic family as the Turkic languages on the sole basis
of shared typological features, such as word order patterns.3 The problem is
complicated by the fact that the study of areal similarities among languages is
3There may be valid reasons for relating Uralic and Turkic languages genetically. Similarly,
although Armenian is not an Iranian language, it is distantly related to Iranian languages as a
separate branch of the Indo-European family.
Linguistic typology should be about issues rather than schools. The field is so
diffuse, however, that it is convenient to organize this discussion in terms of
several recent traditions. For recent North American linguistic typology
outside the generative framework, perhaps the major impetus has been Green-
berg's paper on work order universals (4). Although this paper was primarily
about universals, it revealed that, of the numerous logically possible word
order types, four dominate the world: (a) subject-object-verb, possessor-
noun, adjective-noun, noun-postposition; (b) like a but with noun-adjective;
(c) verb-subject-object, noun-possessor, noun-adjective, preposition-noun;
(d) like c but with subject-verb-object. The implications of this classification
are discussed in the following section. More generally, this seminal paper
reinvigorated studies of language universals and linguistic typology in North
America.
4Various terms are used-e.g. adjunct or operator for dependent, operand for head; Venne-
mann's terms are operator and operand. As a rule of thumb, the head of a phrase is the constituent
after which the phrase is named-i.e. the verb is head of the verb phrase, the noun of the noun
phrase, and the adposition of the ad-(pre-/post-)-positional phrase.
at the far left or at the far right of their matrix clause, rather th
suggests that humans find it easier to process clauses one at a time than to
interrupt the processing of one clause in order to process another. As illus-
trated by schemata a and b, below, a verb-last language with prenominal
relative clauses will have at least many instances of strictly left-branching
subordinate clauses (namely, whenever the noun phrase with the relative
clause starts the sentence, as in a), while a verb-last language with postnomin-
al relative clauses would have internal relative clauses in every instance (since
the verb of the main clause must always follow, as in b).
Semantic Typology
One recent promising basis for a typology of rather general validity is the
distinction between head-marking and dependent-marking, developed by
Nichols (14). If a construction has dependent-marking, then the relation
between the head of the construction and its dependent(s) is indicated by a
change in form of the dependent(s). For instance, in the English possessive
construction the man's house, it is the form of the dependent the man that
changes to show the relation. By contrast, in the Hungarian head-marking
construction az ember haz-a (the man house-his), it is the head haz 'house'
that changes. As with many typologies, there are mixed cases-e.g. Turkish
adam-in ev-i (man-POSSESSIVE house-his) marks both head and dependent,
while Haruai (Papua New Guinea) nobo ram (man house) marks neither;
moreover, some languages have head-marking in one part of their grammar
but dependent-marking elsewhere. The recognition of head-marking as a
phenomenon worthy of attention is one of the main insights of Nichols's
work, especially since many current theories of syntax, themselves based on
the predominantly dependent-marking major European languages, are ill
equiped to deal directly with head-marking constructions (usually treating
them, if at all, as deformations of putative underlying dependent-marking
constructions).
More important for linguistic typology is Nichols's claim that the head/
dependent-marking distinction correlates with other, logically independent
typological distinctions. Extended and justified, this basic structural distinc-
tion between construction types would be the basis of a typology of unusually
wide range. For instance, Nichols has suggested (in a lecture at the University
of Southern California in 1987) that the distinction between alienable and
inalienable possession correlates highly with head-marking. She explains this
by noting that the explicit marking of inalienable possession typically in-
volves nouns obligatorily marked for the person-number of their possessor, a
clear instance of head-marking. It should be noted that in this example the
motivation for a link between the logically independent typological contrasts
is quite specific; it does not depend (as arguments too often do in linguistic
typology) on abstract principles of untested validity.
GENERATIVE GRAMMAR
Pro-Drop Parameter
One of the most widely discussed parameters within Government and Binding
Theory, the main current version of generative grammar, is pro-drop. Pro-
drop is a syntactic parameter; its manifestations are said to be correlated (la,
Sect. 4.3, 4.5). The most obvious characteristic of a pro-drop language, and
the one after which the parameter was named, is that unstressed personal
pronouns can be dropped. In Spanish (a pro-drop language) alongside yo
llegue 'I arrived' one can also say simply llegue, whereas in English *arriv
is not a possible sentence in isolation. As had been noted by many traditional
grammarians, this possibility correlates with the richer verb agreement system
of Spanish; Spanish llegue is unequivocally first person singular, contrasting
with llegaste 'you arrived', llego' 'he arrived', etc. English, by contast, is
almost completely lacking in verb-pronoun agreement, the distinction be-
tween I, you, we, they go and he goes lacking the richness of a pro-drop
language. Generative grammar is more innovative in relating these to two
further phenomena. Pro-drop languages allow free inversion of subject and
verb/predicate, so that in Spanish alongside Juan llego one can also say llego
Juan, whereas English allows only John arrived (not *arrived John). Pro-
drop languages also allow so-called long-distance movement-i.e. extraction
of items out of configurations that would be impossible in non-pro-drop
languages. Contrast Spanish ,quien crees que saldrd? and English *who do
you think that will leave?5
The pro-drop hypothesis stimulated investigation of a wide range of phe-
nomena, but the current consensus among generative grammarians seems to
be that this parameter cannot be confirmed. Crosslinguistic evidence suggests
5The English sentence is grammatical if modified by omitting the conjunction that-i.e. who
do you think will leave? The conjunction is crucial. The pro-drop parameter predicts that if a
language allows omission of the conjunction then, other things being equal, the resulting sentence
will be grammatical even if the language is non-pro-drop. In Spanish, omission of the conjunction
que is impossible, but the sentence is nonetheless grammatical when que is retained.
that even the simplest correlation, that between omission of unstressed pro-
nouns and rich verb agreement, does not hold. For instance, Brazilian Portu-
guese has a verb agreement system that, depending on which tense-aspect one
examines, is either just as rich as or less rich than that of German: Brazilian
Portuguese distinguishes at most four person-numbers: eu acho 'I find',
voce^lele acha 'you/he find(s)', nos achamos 'we find', voce'sleles acham 'you
(plural)/they find; often only three: eulvoce^lele achava 'I/ you/ he found', nos
achavamos 'we found', voce^sleles achavam 'you (plural)/they found'; with a
further reduction of one in even educated colloquial speech through the
replacement of nos by a gente, which takes third person singular agreement: a
gente achalachava 'we find/found'. German regularly distinguishes four
person-numbers: ich liebe 'I love', du liebst 'you love', erlihr liebt 'he/you
(plural) love(s)', wirlsie lieben 'we/they love', with a five-way distinction for
a wide range of irregular verbs through different vowels in the third person
singular and the second person plural: er gibt 'he gives', ihr gebt 'you (plural)
give'. Yet Brazilian Portuguese allows omission of unstressed subject pro-
nouns-e.g. acho '(I) find'-whereas German does not: *liebe is not a
possible sentence in isolation. Brazilian Portuguese also allows the extraction
discussed above for Spanish-i.e. quem voce^ acha que vem? 'who do you
think (literally: that) will come?'-while GermIan does not:? (*wer glaubst
du, dass kommen wird?).6 Brazilian Portuguese even has some inversion
parallel to that of Spanish-e.g. o homem chegou or chegou o homem 'the
man arrived', though admittedly the possibilities are much more severely
restricted than in Spanish. In German, subject-object inversion is not free,
being conditioned either semantically-e.g. as a marker of questions (kommst
du? 'do you come?'), or syntactically-e.g. if some constituent other than the
subject starts the clause (e.g. gestern kam sein Sohn, literally 'yesterday came
his son'). Thus the precise typological correlations that are valid in this area
remain to be characterized more adequately.
Configurationality
The second parameter of interest to typologists is configurationality-the
extent to which a language's syntactic structure is hierarchical. The im-
mediate constituents of an English sentence like the big dog bit the kangaroo,
6'he apparently inverse case is presented by a number of East Asian languages, including
Chinese and Japanese, which lack verb agreement but where unstressed pronouns can, and
usually are, omitted. In Japanese, for instance, alongside Taroo ga Ziroo ni hon o ageta
(Taroo-NOMINATIVE Jiro-to book ACCUSATIVE gave) 'Taro gave a book to Jiro', one also
has the perfectly good sentence ageta 'he gave it to him', which is just as complete as the English
translation given. It is now generally agreed in generative grammar, however, that these East
Asian examples are a phenomenon distinct from pro-drop, involving instead an 'empty topic'
position that is coindexed with the missing arguments (9).
for example, are the big dog (a noun phrase) and bit the kangaroo (a verb
phrase). Immediate constituents of the verb phrase are in turn the verb bit and
the noun phrase the kangaroo. Thus this sentence structure is fairly hierar-
chical, in particular with big and dog grouped as one constituent, bit and
kangaroo as another. English is thus a configurational language. In an
extremely nonconfigurational language, hierarchical structure would be ab-
sent. Hale (5), who introduced this parameter, claims extreme nonconfigura-
tionality for the Australian Aboriginal language Warlpiri (Walbiri). The
Warlpiri translation, or at least one possible translation, of the English
sentence just discussed is tjarntu-ngku wawiri yarlku-rnu wiri-ngki (dog-
ERGATIVE kangeroo bite-PAST big-ERGATIVE); in fact, any permutation
of these four words is a grammatical sentence with essentially the same
meaning. In Warlpiri not only can the major constituents of the clause (e.g.
subject, verb, object) be moved around freely, but even the components of
what in a configurational language would clearly be constituents (e.g. an
attributive adjective and its head noun) can be separated. In the example just
given, tjarntu-ngku 'dog' and wiri-ngki 'big' are separated in this way. The
interpretation of the sentence relies heavily on the morphology: tjarntu-ngku
and wiri-ngki are in the same case (ergative), so despite their separation they
must belong together semantically, and the ergative case shows that they must
refer to the subject/agent of a transitive verb. The question of this parameter's
validity is significant for syntactic theory, since many mainstream syntactic
theories are designed specifically to deal with configurational languages.
More recent work suggests that languages are neither configurational nor
nonconfigurational but exhibit this parameter to various degrees. For in-
stance, while Warlpiri seems nonconfigurational within the clause, boundar-
ies between clauses seem well defined, so that in a sentence like ngatjulu-rlu-
rna yankiri pantu-rnu, kutja-lpa ngapa nga-rnu (I-ERGATIVE-I emu spear-
PAST, COMPLEMENTIZER-AUXILIARY water drink-PAST) 'I speared
the emu while it was drinking water', the sentence as a whole has its two
immediate constituents-the clauses ngatjulu-rlu-rna yankiri pantu-rnu and
kutja-lpa ngapa nga-rnu-a bit of hierarchical syntactic structure. Moreover,
when a head noun and its attributive adjective are adjacent in the order
noun-adjective, instead of both noun and adjective taking a case marker, one
needs only a single case marker for the whole expression, as in another
version of 'the big dog bit the kangaroo': tjarntu wiri-ngki wawiri yarlku-rnu
which suggests that in this version tjarntu wiri 'big dog' does form a single
constituent, whence it receives a single case marker.
In his analysis of Malayalam as a nonconfigurational language, Mohanan
(12a) claims verb phrases are absent. In English, the verb phrase node gives
added hierarchical structure by grouping the verb and direct object together to
the exclusion of the subject. In Malayalam manusyan kutti-ye kantu 'man
some but not all of the same operations. Its theoretical background is too
complex to summarize here, but even as an empirical classification of relative
clause types Lehmann's work is an impressive typological achievement.
CONCLUSION
Literature Cited