Grammaticalization, viewed in a diachronic perspective, is a movement from left to right through a grammaticalization scale. Thus, any grammatical phenomenon in a language can originate by grammaticalization of the phenomenon positioned at its left on a grammaticalization scale. This is, of course, not the only way that new grammatical features may emerge in a language. Another very important origin is by analogy. These general rules are also valid for agreement. In what follows, I will concentrate on the origin and development of agreement through grammaticalization, following the precedence of Greenberg 1978 on internal class agreement and Hale 1973 and Givón 1976 on verbal agreement. It must, however, not be forgotten that agreement may come about in different ways. Third person verb agreement in Iroquoian, for instance, has developed, according to Chafe 1977, largely through analogy.
7.1. External agreement
To the left of external agreement on the grammaticalization scale IX, we find anaphora. When I said at the end of § 4.2. that agreement markers in person-domain, i.e. external, agreement, are of a basically pronominal nature, this may now be interpreted diachronically: If agreement develops by grammaticalization, it will first appear as anaphoric, then as syntactic agreement. Thereby, all grammatical categories which play a role in pronominal reference, viz. person, number and nominal class,75 enter into the agreement mechanism. The way is predetermined by the stages of scale IX: demonstrative pronouns become free personal pronouns, the latter becoming in turn clitic and affixal pronouns, finally agreement affixes and markers of finite verb inflection. Beyond that, subject agreement affixes may become markers of the verb (as a part of speech), object agreement affixes may become transitivity markers, and possessive agreement affixes, relationality markers.
There is much historical evidence for this diachronic development. The demonstrative pronoun ille of classical Latin becomes a free personal pronoun in Late and Vulgar Latin and develops into the clitic pronouns le, lo, la etc. of the Romance languages, which already begin to enter into syntactic agreement (cf. examples (50) and (51)). The verbal object affix of classical Arabic, of exclusively anaphoric use, becomes an optional agreement marker in modern Lebanese and Syrian Arabic (cf. Koutsodas 1967). Originally free personal pronouns become, in Walbiri, clitic to the auxiliary and are used in syntactic agreement (Hale 1973, § 8.). The English free object pronoun him develops, in Tok Pisin, first into an agreement suffix of the verb and then into an invariable marker of all transitive verbs (ex. (55)). The subject pronoun he suffers a similar fate (see Givón 1976, § 8.2. and Sankoff 1977).76
Special attention has been devoted in § 6.2. to the transition between anaphoric and syntactic agreement. Left-dislocation, as illustrated by (50), and right-dislocation ('afterthought' construction), as illustrated by (56), are the diachronic basis in anaphoric agreement on which syntactic agreement develops (cf. especially Givón 1976). (75) is a further example in the less often illustrated realm of adpositional agreement.
(75) š-šərkemā fi-habarake
SYR DEF-partnership NEG in-F.SG blessing
"There's no advantage in partnership."(Cowell 1964:542)
As mentioned above, Syrian Arabic does have object-verb agreement, but it lacks (as yet) prepositional agreement. In (75), the referent of the pronominal complement of the preposition is clearly left-dislocated. If the clause boundary between the preposition and the referent disappears, the referent becomes integrated into the clause, and a certain appositive relationship between it and the agreeing pronoun comes into being. This stage is, for adpositional agreement, exemplified by (56). For object-verb agreement, cf. again (51), which represents the stage subsequent to right-dislocation, and (52), which results from left-dislocation.
The final step in the genesis of syntactic agreement is the interpretation of the referent of the agreeing pronoun as the direct complement of the relational expression which supports the agreeing pronoun. At the same moment, the latter loses its anaphoric and pronominal function, becoming a mere agreement marker. This stage is, for prepositional agreement, illustrated by (33) and (34), and for object-verb agreement, by (53). Similar considerations apply to the other forms of external agreement.77
Once the level of syntactic agreement has been reached, agreement then tends to become increasingly obligatory, i.e. to occur in the presence of kinds of nominal complements where it was formerly unnecessary or impossible. In the case of object-verb agreement, this expansion moves down the animacy hierarchy; see 42. This has to do with the specific semantosyntactic role of the direct object and is not likely to be replicated in the development of the other kinds of external agreement.
The expansion of verbal agreement supposedly follows the hierarchy of syntactic functions (first set up in Keenan/Comrie 1977 for different purposes): subject > direct object > indirect object > other oblique complements. This may be assumed on the basis of existing implicative generalizations of the form: if a language has verbal agreement on a given position of this hierarchy, it also has agreement on the higher positions (cf. Ingram 1971:23, Moravcsik 1978:364f and 1974, § 1.1., furthermore, with a slight change in the hierarchy, Givón 1976, §§2. and 6.). Moreover, known historical facts such as the acquisition of object agreement signs by modern Romance languages and Arabic vernaculars long after subject agreement existed, attest to the accuracy of this assumption.
The hierarchy must be modified somewhat in order to accommodate verb-argument relations not only in accusative languages, where the subject is indeed the primary argument, but also in ergative languages, where the absolutive is the primary argument. There are languages such as Avar (Caucasian),78 where the verb agrees only with the absolutive, not the ergative argument. Therefore, the absolutive function must be put in the hierarchy on the same level as the subject, and the ergative, on the same level as the direct object (cf. already Lehmann 1979, ch. IV. 3.1.1.).
If a language has agreement beyond that of the personal pronoun, it has verbal agreement (cf. already Bazell 1979(S):15; Moravcsik 1978:340). This means that the agreement of the verb with its complements is more fundamental than the agreement of the possessum and the adposition with their complements. Keenan (1974:303) proposes the following "implicative universal: If heads of possessive constructions agree with their possessors in a given language then verbs agree with subjects in that language". Here we see again the hierarchy of syntactic functions at work. The central role of the verb in the clause brings it about that adverbal syntactic functions rank higher in the hierarchy than adnominal functions. Specifically, the subject function ranks higher than the possessor function. It seems that the latter is likened, in the languages of the world, either to the subject or to the object function. Thus we may explain the fact that possessive agreement affixes tend to be formally similar or even identical to either the subject agreement affixes (Turkish, Cahuilla, Mohave, Yucatec Maya, Quechua, Piro)79 or the object agreement affixes (Hungarian, Arabic, Hixkaryana). All this suggests that in the diachronic expansion of external agreement, possessive agreement will follow subject-verb and possibly also object-verb agreement.
The last step in the expansion of external agreement is the acquisiton of adpositional agreement. It is framed after the model of either object-verb or possessive agreement, more often the latter.80 Examples of both types may be found in Arosi (Capell 1971:74-79). One may with confidence propose the implicative generalization: if a language has agreement of the adposition with its complement, it also has agreement of either the verb with its object or the possessum with its possessor.
The combination of these implicative relationship between the steps of evolution of external agreement yields the following hierarchy:
Although this is not a unilinear hierarchy, its interpretation is straightforward: If a language has a given form of agreement, it also has the forms of agreement connected with the latter by at least one upward path in the hierarchy. And conversely in diachrony, a language may acquire forms of external agreement only along downward paths in the hierarchy, starting from the top, and may lose them only in the opposite direction.