How can one compare grammatical structures from many different languages?

A fundamental characteristic of the typological approach to language is that one

begins by comparing a wide range of languages for the grammatical construction in

question. The most important observation that comes from that process is the

extraordinary diversity of grammatical structures that are found. For example, we noted above that English expresses the predication of possession with a transitive verb have as in That man has a canoe, but the Mokilese expression is literally ‘That man’s canoe exists’, with an intransitive verb ‘exist’. Other constructions for the predication of possession would translate literally as ‘A canoe is located at that man’, ‘That man is with a canoe’, ‘For the canoe exists that man’, and ‘As for that man, a canoe exists’ (Heine 1997: 92). Each of these constructions has its own distinctive grammatical structure, and in addition to these major types there are a number of other less common types.

What is the basis for crosslinguistic comparison? The tremendous diversity of

grammatical structures point to an answer that was offered in the seminal paper in

typology, Greenberg (1966), and has been reiterated since: meaning or function. For

example, in the study of possession, Heine compares whatever grammatical constructions are used in a language to express the function of predication of possession. Possession is what is traditionally described as a semantic relation: an ownership relation between a possessor (usually a person) and a possessum (usually an artifact). Predication is generally described as a pragmatic function, having to do with the packaging of information in a sentence: the possession relation is asserted, in contrast to attributive possession (as in that man’s canoe), in which the possessive relation is used to modify the description of the referent canoe.

The use of function as a basis for crosslinguistic comparison allows one to avoid

problems arising from the usage of traditional grammatical terminology. For example, in English a contrast is made between relative clauses with finite (fully inflected) verb

forms, as in (15), and participles, a nonfinite verb form, as in (16):

(15) the papaya that the ants ate

(16) the papaya eaten by the ants

A crosslinguistic study of “relative clauses” immediately encounters the situation that

some languages such as Turkish (Altaic, Turkey) use participial or nominalized verb

forms heavily, as in example (17) (Comrie 1989: 142; see Keenan and Comrie 1977):

(17) Hasan-ı n Sinan-a ver-diğ -i patates-i yedim

Hanan-GEN Sinan-DAT give-NR-his potato-ACC I.ate

‘I ate the potato that Hasan gave to Sinan.’

However, this is an artifact of the traditional grammatical terminology, which treats

finite and nonfinite constructions completely differently. Keenan and Comrie define

relative clauses functionally, in terms of a proposition modifying a referent (the noun);

using this definition, Turkish has constructions comparable to English relative clauses

(and participles).

One widely held but invalid assumption about the crosslinguistic comparison of

grammatical structures is that one can only compare grammatical structures by using the same grammatical categories across languages. Languages can vary to a remarkable degree in basic categories of grammar. For example, a plausible candidate for a pair of universal grammatical categories are the categories of subject and object of a verb:

(18) The woman didn’t run.

(19) The snake bit the man.

The sentence in (18) has only a single phrase (the woman) referring to a participant in

the event denoted by the verb (run). Such a sentence is intransitive, and the woman is

labeled the ‘subject’; we will refer to intransitive ‘subject’ with the label S. The transitive sentence in (19) on the other hand has two phrases referring to the two participants in the event (bit). It seems completely natural, indeed even necessary, that the first phrase, the snake (labeled A, mnemonic for ‘agent’) should belong to the same grammatical category as the woman in (18). Both the woman and the snake occur before the verb. Substitution of a pronoun for the woman would require the subject form she, not her. The grammatical category grouping S and A would be called ‘subject’. The second phrase in (19), the man (labeled P, mnemonic for ‘patient’) is grammatically different. It occurs after the verb, and substitution of a pronoun for the man in (19) would require the object form him, not he. The grammatical category consisting of P is generally called the ‘object’.

But many languages do not categorize the phrases referring to the participants in

events in the same way. Compare the translations of (18) and (19) in Yuwaalaraay, an

Aboriginal language of Australia:

(20) wa: l n1ama yinar-Ø banaga-n1i

not that woman-ABS run-NFUT

‘The woman [S] didn’t run.’

(21) d1uyu-gu n1ama d1ayn-Ø yi: -y

snake-ERG that man-ABS bit-NFUT

‘The snake [A] bit the man [P].’

Yuwaalaraay does not have subject and object in the English sense. The grammar of

participants is expressed by case suffixes on the noun. In an intransitive sentence like

(20), the phrase labeled S has no suffix (notated here with the zero symbol -Ø). In a

transitive sentence like (21) however, what an English speaker would call the “subject”,

A, has a case suffix -gu, which is called the ergative case (abbreviated ERG), and the

“object” phrase P has no suffix, like the “subject” S in (20). In other words, whereas

English categorizes both A and S together (as subject) and distinguishes P (as object),

Yuwaalaraay categorizes P and S together (called the absolutive, abbreviated ABS) and distinguishes A (as the ergative).

This difference between English (and many other languages) on the one hand, and

Yuwaalaraay (and many other languages) on the other, is very striking. It seems very

unnatural to us to group together S and P against A—subject and object in the English

sense seem to be such basic categories of grammar. Since the terms ‘subject’ and

‘object’ are so loaded with theoretical and traditional connotations, most typologists

discard them in describing the contrast between languages like English and languages

like Yuwaaalaraay, and instead use the terms nominative and accusative, from the Latin case terms, for the English categories. The crosslinguistic difference between English and Yuwaalaraay is called a difference in alignment.

The difference between English and Yuwaalaraay grammatical patterns here seems to

make the two languages entirely different, so that no comparison, let alone universals of language, can be derived from the study of alignment systems. But this is not true.

Crosslinguistic comparability of grammatical relations is based on the semantic roles, or more precisely prototype clusters of semantic roles, represented by S, A and P: one can compare how the agent of ‘bite’ and the single participant of ‘run’ are encoded

grammatically across languages. More strikingly, there are language universals that can be stated no matter what the alignment system of the language is. For example, the absolutive is zero-coded in Yuwaalaraay and many other languages, in contrast to the overt coding of the ergative case. In languages with a nominative-accusative alignment and case inflections, the nominative is usually zero-coded in contrast to overt coding for the accusative. These are not the only types found: the universals for each alignment system can be more precisely formulated to cover all types, as in (22a-b):

(22) a. The accusative is encoded with at least as many morphemes as the nominative.

b. The ergative is encoded with at least as many morphemes as the absolutive.

The universals for the two alignment systems in (22a-b) have a single explanation.

Noun phrases with nominative referents are more frequent in discourse than noun phrases with accusative referents, and noun phrases with absolutive referents are more frequent in discourse than noun phrases with ergative referents. A far-reaching universal encompassing (22a-b) and many other grammatical categories is: categories of meanings that are more frequent in discourse are likely to be encoded with fewer morphemes (Croft 2003; see §6). Hence it does not matter what the grammatical categories of a language are in order to find language universals. One must simply identify how concepts are encoded in a language. The language universals are based on how concepts are encoded, not a set of universal grammatical categories.

The methodological necessity of comparing constructions encoding the same

functions across languages matches the functionalist orientation of most typologists. In order to discover language universals, a typologist compares how function is encoded in linguistic form. A crosslinguistic perspective does not allow a typologist to analyze form autonomously from its function. The very fact of crosslinguistic variation in the formal encoding of linguistic function, however, implies that linguistic form is separate from linguistic function, and that linguistic form is at least partly arbitrary—otherwise, all languages would have the same grammatical structure.