Discussion
In the course of the past decades, a number of alternative approaches have been proposed to deal with grammaticalization phenomena. Some of these approaches highlight the pragmatic aspects of the process (Bybee and Hopper2001), others focus on semantic aspects (Traugott and Dasher 2002), and still others seek to account for the process in terms of syntactic frameworks (Roberts and Roussou 2003; van Gelderen 2004; Kiparsky 2005). On the basis of the four parameters discussed in the preceding paragraphs, grammaticalization can be portrayed more generally as a process leading
(a) from concrete meanings to more abstract ones,1
(b) from fairly independent, referential meanings to less referential, schematic grammatical functions having to do with relations within the phrase, the clause, or among clauses,
(c) from open-class to closed-class items,
(d) from grammatical forms that may have internal morphological structure to invariable forms, and
(e) from longer grammatical forms to shorter ones.
Part of our reconstruction methodology is well known from what in historical linguistics is known as the method of internal reconstruction. When there is a development from category A to B, certain A-properties are likely to survive, while others will be replaced by B-properties. In specific cases, the presence of A-properties associated with a given B-category can be interpreted meaningfully only if there has been an earlier A. Such surviving A-properties can be used as evidence to reconstruct an earlier A.
But grammaticalization theory is not confined to internal reconstruction. We may illustrate this with the following example from English: The definite article freely combines with both singular nouns (the tree) and plural nouns (the trees) whereas the indefinite article combines with singular nouns (a tree) but not with plural nouns (*a trees). We know that the indefinite article is historically derived from the numeral one, and it has retained some relics of its etymological source. One such relic is its incompatibility with plural nouns,2 another one can be seen in the fact that the indefinite article has retained the nasal consonant when preceding vowels (an apple). Accordingly, surviving properties such as the inability to qualify plural nouns and the presence of a nasal allow us to hypothesize that the indefinite article is historically derived from some element that had these properties. This is as far as internal reconstruction can go. Grammaticalization methodology can go one step further: We have a number of cases of attested developments of numerals for ‘‘one’’ developing into indefinite articles, so we can safely assume a parallel development in those unattested cases. At the same time, since no language has been found where an indefinite article has given rise to a numeral, we can strengthen the hypothesis by arguing that the English indefinite article must have had the numeral as its historical source (Fritz Newmeyer, p.c.).
But even if we had no diachronic knowledge of the fact that there is a universal process from numeral ‘‘one’’ to indefinite article, we would still be able to reconstruct further details of the process on the basis of the parameters described in the preceding topics. Whereas the numeral has a fairly concrete lexical meaning, the indefinite article has only a functional meaning, hence it is semantically ‘‘impoverished’’ vis-a̒-vis the numeral (desemanticization). Furthermore, the indefinite article is syn tactically more restricted: It is a clitic that cannot occur without a following noun (*I have a.), whereas the numeral ‘‘one’’ is a free word that can occur on its own (I have one.). Now, according to one of these parameters, decategorialization, free words tend to lose their independent status and to turn into clitics (sometimes further into affixes). Finally, the more a linguistic item is grammaticalized, the more it tends to lose in phonetic substance (erosion). In accordance with these three parameters, we will hypothesize that the indefinite article can be traced back in a principled way to an element that had a more concrete meaning, was a free form, and had more phonological substance than it now has.
1 Conceptual shift from concrete to abstract, as understood here, is anthropocentric in nature, in that it leads from meanings that are close to human experience and easy to describe, to meanings that are more difficult to understand and describe.
2 As an anonymous referee rightly points out, there are some exceptions, such as Spanish, Old French, and Bulgarian (see Heine 1997b).
