Grammaticalization as Optimization Paul Kiparsky 2.1 Meillet’s formal concept of grammaticalization According to the neogrammarians and de Saussure, all linguistic change is either sound change, analogy, or borrowing.1 Meillet (1912) identified a class of changes that don’t fit into any of these three categories. Like analogical changes, they are endogenous innovations directly affecting morphology and syntax, but unlike analogical changes, they are not based on any pre-existing patterns in the language. Meillet proposed that they represent a fourth type of change, which he called GRAMMATICALIZATION. Its essential property for him was that it gives rise to new grammatical categories — that is, to categories previously unexpressed in the language — and thereby transforms its overall system.
Tandis que l’analogie peut renouveler le détail des formes, mais laisse le plus souvent intact le plan d’ensemble du système grammatical, la “grammaticalisation” de certains mots crée des formes neuves, introduit des categories qui n’avaient pas d’expression linguistique, transforme l’ensemble du système.” The “newness” of a category can be either a matter of content, as when a language acquires a new tense category, or a matter of new form for old content, as when postpositions turn into case 1 endings, or word order replaces morphology as the mark of grammatical relations. For Meillet, analogy and grammaticalization are categorically distinct processes, because anal- ogy requires a model for the innovating structure, and grammaticalization by definition does not have one. Meillet noticed that grammaticalization processes have two other distinguishing prop- erties. They proceed in a fixed direction towards “l’attribution du caractère grammatical à un mot jadis autonome”.
For example, languages readily acquire case systems by grammaticalization of adpositions into affixes, but there are no known examples of languages acquiring systems of ad- positions by “degrammaticalization” of case endings (although individual case affixes can become clitics and even adpositions by ordinary analogical change, see section 2. Secondly, grammaticalization is often accompanied by phonological weakening of the grammaticalized el- ement, and never, it seems, by strengthening. For example, case affixes are typically reduced in phonological form compared to the original adpositions that they are historically derived from. To- gether, Meillet’s two generalizations constitute the first formulation of the famous unidirectionality hypothesis.
Meillet’s generalizations are empirical rather than definitional, so they demand an explanation. He suggested that grammaticalization is due to the loss and renewal of expressiveness of speech forms in the use of language, reasoning that, since this is a constant factor in in the ordinary use of language, the changes it triggers must have an intrinsic direction. As for phonological weakening, he saw it as a consequence of the fact that function words (mots accessoires) ordinarily do not carry focus. Therefore, when a lexical item becomes a function word, speakers can afford to give it a reduced articulation, which then can become established as its normal pronunciation: “.
les mots accessoires groupés avec d’autres tendent de ce chef à s’abréger et à changer de prononciation. De plus, et par le fait d’abrègement, et par le fait que, 2 étant accessoires, ils sont prononcés sans effort et attendus sans attention spéciale, ils sont négligés, dénués d’intensité, ils ne sont plus articulés qu‘a demi. les mots accessoires ont des traitements phonétiques aberrants.” The idea that there is a special type of grammatical change which is unidirectional and as- sociated with phonological reduction has gained widespread support in recent work. Meillet’s examples remains staples of the modern grammaticalization literature: (1) a.
The rise of the periphrastic perfect in Romance and its subsequest development into a simple past in French (V ERB > AUXILIARY, P ERFECT > PAST), b. the strengthening of negation by indefinitely quantified elements, which then become negations themselves: Latin ne > ne ūnum > noenum > nōn, Romance (nōn >) ne > ne pas > pas > pas du tout (the trajectory later famous as “Jespersen’s cycle”), c. the rise of auxiliaries: je suis parti, habeo dictum > j’ai dit, I will make, I shall make, je vais faire, d. the rise of light verbs and “clause union” constructions: laissez venir, il vient me dire cela, e.
the rise of complementizers: Greek elo ina > elo na > e na > a, f. the grammaticalization of fixed word order (“. la façon de grouper les mots peut aussi devenir un procédé d’expression grammaticale. In spite of its sketchy nature, Meillet’s pioneering essay makes some precise proposals and raises issues which remain unresolved even now.
One obvious criticism is that taking the creation of new linguistic categories as the defining property of grammaticalization yields too broad a class 3 of changes. Because morphological categories are compositional, new ones can be created by analogical extension of existing combinatoric patterns. An example of a new morphological category that has originated by analogy is the Sanskrit Past Perfect (“Pluperfect”). Sanskrit has a subclass of Perfect forms which have present time reference, e.
ci-két-a ‘I see, I recognize’, ja-grábh-a ‘I grasp’. These Perfects have Past (“Pluperfect”) counterparts, which are formed by adding Past tense morphology to the Perfect stem, e. á-ci-ket- am ‘I saw’, á-ja-grabh-am) ‘I grasped’. The category of Past Perfect originated by analogy to the Past non-Perfect (the “Imperfect”), by a morphological generalization which can be visualized by the proportional schema in (2): (2) Nonpast Past Nonperfect kr-n.ó-mi ‘I do’ : á-kr-n.av-am ‘I did’ Perfect ja-grábh-a ‘I grasp’ : á-ja-grabh-am ‘I grasped’ The morphology and meaning of the Past Perfect is fully predictable from the meaning of the cor- responding Perfect and the Past.2 Although this innovation creates a new grammatical category, it has none of the hallmarks of grammaticalization; it is a straightforward case of analogical change.3 A related problem with Meillet’s view is that it treats analogy and grammaticalization as rad- ically disjoint classes of change.
Consequently it precludes by definition any interaction between analogy and grammaticalization. I argue below that such interactions are, if anything, the norm, in the sense that most grammaticalization processes are constrained by, and partly motivated by, the grammatical structure of the language. The theory I will propose directly addresses this understud- ied grey area of analogy/grammaticalization interactions. Finally, let us note that the causal mechanisms mentioned by Meillet (expressiveness being the primary factor) do not account for common grammaticalization patterns such as univerbation.
A 4 prime example of univerbation is the change of adpositions to affixes, which will be in focus below.2 New functionalist approaches to grammaticalization Meillet’s idea that grammaticalization is a “conséquence immédiate et naturelle” of ordinary language use has been widely adopted and fleshed out with the help of ideas from pragmatics. Grammaticalization has become a research area in its own right. Modern studies of grammatical- ization often cite Meillet’s article as a precursor, but less often do justice to the rather different view it expresses. The major thrust of the new research has been to explain grammaticalization in functionalist terms (Givón 1979, Lehmann 1982, Traugott & Heine 1991, Bybee/Perkins/Pagliuca 1994).
Meil- let’s formally defined concept of grammaticalization has been replaced in several different ways, usually by building in unidirectionality into the definition. There are two principal competing families of definitions: (3) • D EF 1: a grammaticalization is a change “by which the parts of a constructional schema come to have stronger internal dependencies” (Haspelmath 2004). • D EF 2: a grammaticalization is a change “where a lexical unit or structure assumes a grammatical function, or where a grammatical unit assumes a more grammatical function” (Heine, Claudi, and Hünnemeyer 1991). The two definitions pick out different aspects of unidirectional change.
The first is based on change in morphosyntactic form, which proceeds from lexical words to function words, function words to clitics, and clitics to affixes, in short towards increasingly tightly bonded units. The second definition is based on change in morphosyntactic function, which is also assumed to pro- ceed unidirectionally from “less grammatical” function to “more grammatical” function (however exactly that is to be defined). 5 The two kinds of change do not exactly coincide. For example, the change from a postposition to a clitic or suffix involves a strengthening of internal dependencies, but it does not necessarily involve any change in grammatical function, either of the postposition/clitic or of the grammatical unit to which it belongs.
Such changes are grammaticalizations by Def. 1 but not necessarily by Def. On the other hand, when an epistemic modal acquires a deontic meaning, even granting that it thereby assumes a “more grammatical function” in some sense (which remains to be made more precise), it does not necessarily acquire a stronger morphosyntactic internal dependency. Such changes are grammaticalizations by Def.
2 but not necessarily by Def. The two aspects of gram- maticalization do not have to march in lockstep, and neither seems to be a necessary consequence of the other. Thus, the definitions in (3) arguably pick out separate and more or less loosely parallel trajectories of change. There are moreover putative grammaticalizations to which neither definition is straightfor- wardly applicable.
Meillet’s example of the change of the Romance periphrastic perfect into a simple past in French (the second step in (1a)) certainly does not result in stronger internal de- pendencies, but it doesn’t result in a more grammatical function either, at least in any clear sense of “grammatical”. What does seem to be common to such changes is semantic generalization, or bleaching as it is are sometimes called.4 The perfect’s domain extends to cover the ground of the obsolescent simple past.5 Like the increase of grammatical function, semantic bleaching follows a trajectory of its own which does not strictly coincide with any structural changes. For example, in the development of case, bleaching is not necessarily tied to morphological downgrading from postposition to clitic to suffix. The modern view takes unidirectionality, rather than the formation of new categories, as the 6 essential property of grammaticalization.
This reorientation has important consequences. The concept becomes more restricted in some respects, and more inclusive in others. Some changes no longer qualify as grammaticalizations; on the other hand, the concept now includes much of what is traditionally considered analogical change. In some treatments, in fact, it is effectively equated with grammatical (morphosyntactic) change.
The most drastic consequence of this revision is that the unidirectionality thesis becomes either tautological or false. Under (3), grammaticalization is unidirectional by definition, as Newmeyer 2001, Joseph 2001, 2004, 2006, and others have noted. To make the unidirectionality thesis into an empirical claim again, it can be restated in the obvious way (“there is no degrammaticaliza- tion”, cf. Haspelmath 2004), or grammaticalization can be redefined as a change by which an ele- ment acquires “new grammatical functions” (rather than “more grammatical functions”) (Hopper & Traugott 2003:xv).6 Under either of these reformulations, it becomes false, for there are well- documented instances of DEGRAMMATICALIZATION (“upgrading”), that is, of changes in which “internal dependencies” are loosened, or forms acquire a “less grammatical function” (e.
Harris and Campbell 1995: 336-338). Here is a partial list of them, culled from the recent literature. Seto and Võru (South Estonian) Abessive case suffix -lta > clitic =lta. Vepsian Abessive case suffix -tta (< *-ptaken) > clitic =tta.
English and Mainland Scandinavian genitive suffix -s > clitic =s (Janda 1980, 1981, Plank 1992, 1995, Norde 1997, Allen 1997, Newmeyer 1998:266, Tabor and Traugott 1998). suffix -muid > independent pronoun muid (Bybee et al. suffix -mos > independent pronoun nos. English inseparable infinitive prefix to- > separable to (Fischer 2000, Fitzmaurice 2000).
Estonian question marker -s > clitic =es > free particle es (Nevis 1986a, Campbell 1991:290-2). Estonian affirmative marker -p > clitic =ep > free adverb ep (Campbell 1991:291). Modern Greek prefix ksana- ‘again’ > free adverb ksana ‘again’ (Méndez Dosuna 1997). Haspelmath 1999, 2004 and Hopper & Traugott 2003 argue that such counterexamples are not damaging to the unidirectionality thesis because they are sporadic.7 In their view, robust tendencies is the best we can hope for in the functional realm, indeed they are in some sense more interesting than categorical ones.
