The structure of grammaticalization chains
Grammaticalization chains are linguistic categories involving a linguistic form or construction that is associated with a range of different, contextually defined uses. The structure of chains can be described with reference to the formal properties listed in (1) (Heine 1992:348–9):
(1) Some properties of grammaticalization chains
a. Other than the fact that all uses share one and the same (or at least a similar) form, no attribute is shared by all uses of the chain.
b. No use combines all the attributes in question.
c. Each use has at least one attribute in common with some other use.
d. The endpoints of the chain have no attribute in common.
e. Immediately adjacent members have more attributes in common than non-adjacent members.
F. Uses located at opposite ends of the chain have the lowest number of attributes in common, while uses located at the center of the chain share the highest number of attributes.
The structure of a grammaticalization chain can be illustrated with the following English example. The sentences of (2), taken from Hopper and Traugott (1993:2–3, 61,88), can be said to include the attributes distinguished in (3).
The classification of structural attributes presented in (3) is based on the information provided by Hopper and Traugott (1993:61). One might think of alternative analyses; for our purposes, however, this would be largely immaterial. The four sentences of (2) can be interpreted as reflecting a minimal grammatical ization chain having all the attributes listed in
(1): a. None of the attributes distinguished in (3) is shared by all four uses, i.e. by all four sentences in (2). For example, the attributes ‘future tense’ and ‘full form’ are shared by three sentences, other attributes again are shared by two (‘any verb’), or by one sentence only (‘progressive’, ‘directional’, ‘verb of activity’, ‘reduced form’).
b. None of the sentences combines all the attributes.
c. Each sentence has at least one attribute in common with some other sentence.
d. The endpoints (i.e. (i) and (iv)) have no attribute in common.
e. Immediately adjacent sentences share more attributes than non-adjacent members. Thus, (i) and (ii) share one attribute, (ii) and (iii) two, (iii) and (iv) two, but (i) and (iv) have no attribute in common.
f. As just noted (e), the peripheral sentences (i) and (iv) share no attributes.
The synchronic properties just sketched have a systematic diachronic correlate: In accordance with general findings on the evolution of grammatical categories we will expect that the initial member of a grammaticalization chain represents the earliest form and the final member the latest form. With reference to the present example this means that (i) can be expected to represent the earliest and (iv) the latest stage in the development of the English be going to-construction.
Grammaticalization chains share some characteristics with prototypes. At the same time, however, they are also incompatible with some main properties of prototypes; they are most appropriately described as linear family resemblance categories; the reader is referred to Heine (1992:349ff.) for more details.
To conclude, the structure of a grammaticalization chain can be defined with reference to the following variables: (a) the endpoints of the chain, (b) relative distance along the chain, and (c) location on the chain.
That such chains cannot only be identified within a given language has been demonstrated by Spike Gildea (1997). Cross-linguistic chains, like the one described by him, differ from the language-internal chain looked at above especially in the following:
Rather than being confined to one language, they include uses of a form that is etymologically the same in genetically closely related languages.
The languages concerned can be classified with reference to the range of uses they exhibit along a given grammaticalization chain.
The grammatical behavior of a form or construction can be explained meaningfully only if all its uses involved are taken into account, even if these uses extend beyond the limits of a single language.
While language-internal chains can be accounted for with reference to established descriptive techniques, cross-linguistic chains require an explanatory framework that is comparative in nature.
With the example looked at in the following section I wish to demonstrate that the reconstruction of cross-linguistic chains can contribute to a better understanding of the nature and structure of some linguistic categories. The example chosen differs in several ways from those discussed in Heine (1992) and Gildea (1998). The languages looked at belong to the Khoe (or Central Khoisan) family of southern Africa.
