43
Deconstructing grammaticalization Frederick J. Newmeyer Department of Linguistics, University of Washington, Seattle, WA 98195-4340, USA Abstract Grammaticalization is often regarded in the literature as a distinct process requiring explanatory machinery unique to its own domain. I argue, on the contrary, that ‘grammaticalization’ is simply a cover term for certain syntactic, semantic, and phonetic changes, all of which can apply independently of each other. 7 2000 Elsevier Science Ltd. All rights reserved. Keywords: Grammaticalization; Language change; Functionalism 1. Overview To a certain degree, the dierent research programs of dierent schools of linguistics have led their practitioners to investigate — and attempt to explain — a dierent set of natural language phenomena 1 . Generative grammarians, for example, have been consumed with such things as the intricacies of parasitic gap constructions, the precise conditions for the extraction of question elements, and the nuances of quantifier scope judgments. These problems have to a large extent been ignored in the literature of functional linguistics. 0388-0001/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved. PII: S0388-0001(00)00021-8 Language Sciences 23 (2001) 187–229 www.elsevier.com/locate/langsci E-mail address: [email protected] (F.J. Newmeyer). 1 This paper is excerpted from Chapter 5 of Newmeyer (1998) and appears here with the permission of MIT Press. It owes a great deal to the work of, and electronic conversations with, Lyle Campbell, Alice Harris, Richard Janda, and Brian Joseph. However, I must stress that none of these individuals bears any responsibility for ideas expressed here that are not directly attributed to them, nor do I wish to imply that they are necessarily in agreement with the rest.

Deconstructing grammaticalisation

Embed Size (px)

DESCRIPTION

Grammaticalisation

Citation preview

  • Deconstructing grammaticalization

    Frederick J. Newmeyer

    Department of Linguistics, University of Washington, Seattle, WA 98195-4340, USA

    Abstract

    Grammaticalization is often regarded in the literature as a distinct process requiringexplanatory machinery unique to its own domain. I argue, on the contrary, that

    grammaticalization is simply a cover term for certain syntactic, semantic, and phoneticchanges, all of which can apply independently of each other. 7 2000 Elsevier Science Ltd.All rights reserved.

    Keywords: Grammaticalization; Language change; Functionalism

    1. Overview

    To a certain degree, the dierent research programs of dierent schools oflinguistics have led their practitioners to investigate and attempt to explain a dierent set of natural language phenomena1. Generative grammarians, forexample, have been consumed with such things as the intricacies of parasitic gapconstructions, the precise conditions for the extraction of question elements, andthe nuances of quantifier scope judgments. These problems have to a large extentbeen ignored in the literature of functional linguistics.

    0388-0001/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved.PII: S0388 -0001 (00)00021 -8

    Language Sciences 23 (2001) 187229

    www.elsevier.com/locate/langsci

    E-mail address: [email protected] (F.J. Newmeyer).1 This paper is excerpted from Chapter 5 of Newmeyer (1998) and appears here with the permission

    of MIT Press. It owes a great deal to the work of, and electronic conversations with, Lyle Campbell,

    Alice Harris, Richard Janda, and Brian Joseph. However, I must stress that none of these individuals

    bears any responsibility for ideas expressed here that are not directly attributed to them, nor do I wish

    to imply that they are necessarily in agreement with the rest.

  • On the other side of the fence, there are phenomena which functionalists haveshown far more interest in than have generativists. Without any doubt, theforemost of these is grammaticalization. And just as many generativists wouldclaim that, say, parasitic gaps defy functional explanation, functionalists point togrammaticalization as presenting an equal challenge to generative grammar.Traugott and Konig (1991), for example, feel that the study ofgrammaticalization challenges the concept of a sharp divide between langue andparole and . . .also challenges the concept of categoriality (p. 189). In the view ofHeine et al. (1991, p. 1), grammaticalization theory challenges what they see asthe predominant conceptions of theoretical linguists since Saussure. Indeed, theyfeel that it calls for a new theoretical paradigm, counterposed to:

    [m]ost post-Saussurean models of grammar [which] rely explicitly orimplicitly on the following tenets:(a) Linguistic description must be strictly synchronic.(b) The relationship between form and meaning is arbitrary.(c) A linguistic form has only one function or meaning.

    And Paul Hopper (1987) goes so far as to claim that There is, in other words,no grammar but only grammaticization movements toward structure (p.148).In this paper I will put grammaticalization under the microscope and conclude

    that such claims are unwarranted. Indeed, I will conclude that there is no suchthing as grammaticalization, at least in so far as it might be regarded as a distinctgrammatical phenomenon requiring a distinct set of principles for its explanation(see Campbell, and Janda, this issue). Instead, I will attempt to demonstrate thatthe set of phenomena that fall under its label are a simple consequence ofprinciples that any theory whether formal or functional would need to positanyway.The paper is organized as follows. Section 2 describes the two principal ways

    that grammaticalization has been characterized, one as a distinct historical processthat requires an independent theory for its explanation and the other as anepiphenomenal result of independent processes. Section 3 defends at some lengththe latter option, arguing that the component parts of grammaticalization alloccur and must be explained independently of each other. The followingsection (Section 4) takes on the question of the unidirectionality ofgrammaticalization, demonstrating that it is falsified by the many attested andreconstructed upgradings from higher to lower degrees of grammatical content. Iattempt to explain why downgradings do greatly eclipse upgradings in frequency.Section 5 raises two problems endemic to many studies of grammaticalization, onemethodological and one theoretical. The former involves the frequent use ofreconstructed forms as theoretical evidence. The latter is the postulation ofpanchronic grammars, that is, grammars combining synchronic and diachronicstatements. Section 6 addresses the question of whether grammaticalization refutesthe generative view of language, while Section 7 is a brief conclusion.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229188

  • 2. What is the true nature of grammaticalization?

    There are a variety of opinions in the literature on whether grammaticalizationrequires an inherent set of explanatory devices or is an epiphenomenal result ofother principles. The following two sections outline these positions in turn (seeCampbell et al., this issue).

    2.1. Grammaticalization as a distinct process

    Both the major theoretical work devoted to grammaticalization, Heine, Claudi,and Hunnemeyers Grammaticalization: A Conceptual Framework and the majorintroductory overview, Hopper and Traugotts Grammaticalization, begin bydescribing it as an historical process (see also Joseph, this issue).

    What is common to most definitions of grammaticalization is, first, that it isconceived of as a process (Heine et al., 1991, p. 4).

    We define grammaticalization as the process whereby lexical items andconstructions come in certain linguistic contexts to serve grammatical functions,and, once grammaticalized, continue to develop new grammatical functions(Hopper and Traugott, 1993, p. xv).

    Now, the term process is often used informally to mean nothing more thanphenomenon to be explained. Some of the references to grammaticalization as aprocess seem simply to have this use of the term in mind. However, the term hasanother, much stronger, sense. In this use, a process is a phenomenon of aparticular type, namely, one driven by a distinct set of principles governing thephenomenon alone. Let us call such a type of phenomenon a distinct process. Themany references in the grammaticalization literature to unidirectional anddeterministic pathways of development convey strongly the idea that it is anencapsulated phenomenon, governed by its own set of laws. For example:

    [G]rammaticalization . . .may be defined as the evolution of grammatical formand meaning from lexical and phrasal antecedents and the continued formaland semantic developments such material subsequently undergoes.The . . . lexical sources of particular grammatical forms . . . [undergo] formal andsemantic changes which characterize their developmental histories. . . .As alexical construction enters and continues along a grammaticalization pathway,it undergoes successive changes in meaning, broadly interpretable asrepresenting a unidirectional movement away from its original specific andconcrete reference and toward increasingly abstract reference. . . . [T]he mostadvanced grammatical forms, in their travel along developmental pathways,may have undergone continuous reduction from originally free, unbound items,to axes entirely dependent on their hosts (Pagliuca, 1994, p. ix).

    In the view of Hopper and Traugott (1993), two components of

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 189

  • grammaticalization, reanalysis and analogy, arise from independently neededmechanisms, but unidirectionality is said to be unique to that process:

    The subset of processes that are particular to grammaticalization are those thatover time render more independent elements less independent (Hopper andTraugott, 1993, p. 62; emphasis added).

    Since a distinct process requires a distinct theory, it is not surprising to seereferences in the literature to a grammaticalization theory, which makespredictions about its subject matter. The implication that there is a predictivetheory of grammaticalization per se recurs repeatedly, for example, in thefunctionalist literature (see for example Heine, 1990, p. 129, 220; Heine, 1993, p.106; Bybee et al., 1994, pp. 922).

    2.2. Grammaticalization as the result of other processes

    The implication that grammaticalization is a distinct process whose workingsare to be attributed to a distinct theory contrasts with many other references thatsuggest that it is essentially an epiphenomenal result of independent historicaldevelopments, each of which falls out of some independent theory. For exampleBybee and Pagliucas capsule scenario of the causes of grammaticalizationsuggests no process unique to it, much less the need for a distinct theory toexplain the process:

    [W]e suggest that human language users have a natural propensity for makingmetaphorical extensions that lead to the increased use of certain items.. . .Thus the paths of development leading to grammatical meaning arepredictable, given certain lexical meaning as the starting point. As the meaninggeneralizes and the range of use widens, the frequency increases and this leadsautomatically to phonological reduction and perhaps fusion. (Pagliuca, 1982;Bybee and Pagliuca, 1985, p. 76).

    As far as unidirectionality is concerned, there have been a number of attemptsto derive it from the (putative) unidirectionality of independent mechanisms andprocesses. For example, Bybee and Pagliuca (1985), Heine et al. (1991), andHaspelmath (1998) agree that part of the explanation lies in the fact that certaincognitive processes, in particular the human tendency to conceptualize abstractnotions in terms of concrete notions by means of metaphor, are both universaland unidirectional. Haspelmath adds the tendency for humans to associatepragmatic force with novelty (Lehmann, 1985) (Haspelmath, 1998, p. 55), whichaccounts for semantic changes in which no metaphor is involved, say, thereduction of full pronouns to agreement markers. He goes on to note that thephonetic reduction and merger aspect to grammaticalization is also the result ofan independent process: the speakers desire for ease of production. This desire isoften counterbalanced by the hearers need for perceptual clarity. But with

    F.J. Newmeyer / Language Sciences 23 (2001) 187229190

  • familiar frequent items, clarity is less of an issue, so reduction wins out.Haspelmath concludes: The unidirectionality of grammaticalization is [thus] anindirect eect of general principles of human cognition and behavior(Haspelmath, 1998, p. 58).It should be mentioned that nonfunctionalists who have addressed the issue of

    grammaticalization have been virtually unanimous in agreeing that its eects areepiphenomenal (see, for example, Roberts, 1993; Harris and Campbell, 1995;Joseph, 1996; Janda, 1995; see also other papers of this issue). Brian Joseph likensgrammaticalization to lexical diusion,

    in that there is clearly a diusionary eect in the way that sound change isrealized in lexical material, but one need not privilege Lexical Diusion withthe status of a mechanism of change instead, the well-known mechanismsof analogy and dialect borrowing together can give the diusionary eect thathas been referred to as Lexical Diusion (Joseph, 1996, p. 20).

    In the next section, I will argue that it is correct to view grammaticalization asan epiphenomenon.

    3. The epiphenomenal nature of grammaticalization

    This section is devoted to arguing for the correctness of the view that thehistorical changes observed in grammaticalization are the product of wellunderstood forces (see also Campbell, and Janda, this issue). Grammaticalization,as I will argue, is nothing more than a label for the conjunction of certain types ofindependently occurring linguistic changes. I begin in Section 3.1 by showing thatgrammaticalization cannot sensibly be conceived of as a distinct process. Indeed,the very idea that there can be such a thing as a diachronic process will bechallenged. Section 3.2 addresses the relationship between grammaticalization andreanalysis. It argues that there is no fixed order between the reanalysis, thesemantic changes, and the phonetic reductions that comprise grammaticalization.And Section 3.3 stresses that each can occur without the other; it is only when thesemantic and phonetic changes happen to converge with a certain type ofreanalysis that we speak of grammaticalization having taken place.

    3.1. On the notion diachronic process

    As noted in Section 2.1, the term process is often used informally tomean nothing more than phenomenon to be explained. If such is all that isimplied in calling grammaticalization a process, then no harm is done. But Ifeel that the term process is dangerous when applied to set of diachronicdevelopments. The reason for the danger is that it invites one to conceive ofsuch developments as being subject to a distinct set of laws that are

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 191

  • independent of the minds and behaviors of individual language users.However, nothing transgenerational can be situated in any human faculty.Children are not born with racial memories that fill them in with the detailsof the history of the forms that they hear. They have no way of knowingwhether some particular clitic in their grammar, for example, was once a fullnoun phrase or whether it is on its way to axhood. If it truly is the casethat speakers are more likely to reanalyze an item as less lexical, rather thanas more lexical, then we need to look at synchronic mechanisms, that is, atmechanisms that are rooted in the cognition and behavior of living speakers.Several decades ago Paul Kiparsky (1968) warned against the practice, all too

    common in historical linguistic studies, of disembodying language change fromlanguage speakers:

    The point is simply that a language is not some gradually and imperceptiblychanging object which smoothly floats through time and space, as historicallinguistics based on philological material all too easily suggests. Rather, thetransmission of language is discontinuous, and a language is recreated by eachchild on the basis of the speech data it hears (p. 175).

    But to read the functionalist-oriented grammaticalization literature, one has theimpression that words, morphemes, axes, and so on are literally driven to evolvein a particular way. Consider, for example, the quote from William Pagliuca inSection 2.1, which appears to conceptualize language change in such terms. Thereis some irony to the fact that Bernd Heine (1993) would open his book on thegrammaticalization of auxiliaries with the following quotation from WilliamCroft: Languages dont change; people change language (Croft, 1990, p. 257).The very definition provided by Heine of what auxiliaries are, namely, linguisticitems located along the grammaticalization chain extending from full verb togrammatical inflection of tense, aspect, and modality . . . (p. 131), invites one tothink of grammars apart from the minds and activities of speakers. No actualspeaker can be expected to know where some item might fall along a particularchain. The focus of the book on these unidirectional chains of auxiliarydevelopment and on cognitively incoherent panchronic statements (see below,Section 5.2) eectively negates the force of Crofts aphorism, painting a picture inwhich languages are impelled to change regardless of what the people who speakthem are disposed to do or think.Since grammaticalization is not a distinct process, there can be no such thing as

    grammaticalization theory, unless one intends that expression merely as aconvenient way of referring to the set of independent theories needed to explainthe phenomenon (see Joseph, this issue).

    The following sections, as they unfold, will provide further evidence thatgrammaticalization lacks the distinguishing characteristics of what one mightreasonably call a distinct process.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229192

  • 3.2. Reanalysis, semantic change, and phonetic reduction their temporal orderingin grammaticalization

    The idea that grammaticalization is a distinct process is based in large part onits always passing through the same stages. However, there is no consensus on thequestion of what those stages are. Let us begin with the ordering of the semanticchanges and the reanalysis. In the view of Heine (1993, p. 48), [I]n the process ofgrammaticalization like the one considered here, conceptual shift precedesmorphosyntactic shift . . . . Indeed, conceptual is the first obligatory step ingrammaticalization . . . (p. 51; see also Givon, 1991b, p. 123). But for Hopper andTraugott (1993, p. 207), the components occur side-by-side: In general it can beshown that meaning change accompanies rather than follows syntactic change(see also Bybee et al., 1991). Others (e.g. Harris and Campbell, 1995, p. 92),suggest that semantic changes can follow from reanalysis.As far as I have been able to determine, there is some degree of truth to all

    three of these positions. Sometimes the semantic changes precede themorphosyntactic changes, sometimes they accompany them, and sometimes theyfollow them. Since the position that semantic change is the result of reanalysisseems to be the most controversial, let me give a concrete example supporting it.Consider the development of English periphrastic do from its origins as an earlyMiddle English causative verb. Kroch et al. (1982) argue that the rise of do inquestions was a direct consequence of the shift of English word order to SVO the use of do as a dummy allowed that order to be preserved even in questions.Evidence is provided by the fact that do was first used in this capacity where theinversion of the main verb with the subject produced the most extreme violationsof SVO order. It then spread to other environments. But there is no evidence thatthe bleaching of the meaning of do played any role in the causation of thissequence of events. Quite the contrary, it was only as do was co-opted as aquestion marker that it lost its causative properties2.Other reanalyses seem inextricably linked to their accompanying semantic

    changes. In Finnish, for example, a noun-postposition unit meaning on the chestwas (optionally) reanalyzed as a simple postposition meaning next to (Harris andCampbell, 1995). Let us say that the original and still possible structure ofthis phrase was (1):

    2 Traugott and Konig (1991, p. 190) agree that the bleaching of the meaning of do took place very

    late in its historical development. See Hopper and Traugott (1993, pp. 8990) for an account of the

    development of do that is somewhat dierent from that provided by Kroch et al. They argue that the

    semantic changes observed in grammaticalization occur in two stages the pragmatic enrichment

    (metonymy) occurs very early and starts the other developments on their way. Bleaching, however, is

    very late.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 193

  • What could it possibly mean to claim that the meaning of the PP on the chestchanged to next to without an accompanying reanalysis of that PP as apostposition? Such would imply that at some stage the embedded PP consisted ofa full NP followed by a postposition which somehow compositionally yielded themeaning next to. That seems quite implausible. Semantic factors may haveprovided the ultimate motivation for the change, but the semantic change itselfcould have hardly taken place before the reanalysis.If we look at a series of historically successive grammaticalizations involving the

    same element, then of course a semantic change at time T can precede asubsequent reanalysis at time T+1. The semantic change involving the use of havein the sense of obligation preceded the grammaticalization of have and a followingto as hafta. Marchese (1986) gives an example from Tepo, a Western Krulanguage of West Africa, of the same reanalysis being both a response to and acause of semantic change. Mu go can be used in its literal sense with anominalized complement:

    (2) & mu na w&he go drink NOMhe went to drink

    Mu also occurs in semantically bleached form as a future marker, which(apparently) triggered the reduction and loss of the nominalizer:

    (3) n mn n cre they FUT me shave they will shave (my head)

    In other words, the following structural changes took place:

    (4) SV1OV2nom4SV1OV2This led in turn to the reanalysis of V1 as AUX and of V2 as a main verb:

    4S AUXOV

    F.J. Newmeyer / Language Sciences 23 (2001) 187229194

  • But now, according to Marchese, the restructuring itself is setting into motion anew set of semantic changes.As far as the relative ordering of phonetic erosion is concerned, the mainstream

    functionalist position is, I think, that it is a response to frequency of use, whichitself is a response to the semantic changes (for an explicit statement to this eect,see Heine et al., 1991). I certainly have no desire to question the idea that ingeneral erosion comes late. But it is easy to see how the precise opposite chain ofevents might unfold. Phonetic erosion can be the result of natural phonologicalprocesses, say, the loss of unstressed final syllables. Such losses can triggerprofound morphosyntactic changes. So for example, Vennemann (1974) arguesthat shifts from SOV to SVO order have often been triggered by the loss of casemarkers due to phonological change. The word order change was a necessaryfunctional response to allow the grammatical relations among the major sententialconstituents to be unambiguously expressed. But this in turn led to thegrammaticalization of distinct pre-verbal auxiliaries with the attendant semanticshifts. In other words, phonetic erosion was the (ultimate) cause of semanticchange.Finally, it should be pointed out that once the first steps of grammaticalization

    have taken place, there is no inevitability that the other steps will follow. AsHopper and Traugott (1993, p. 95) have noted:

    [T]here is nothing deterministic about grammaticalization and unidirectionality.Changes do not have to occur. They do not have to go to completion, in otherwords, they do not have to move all the way along a cline. A particulargrammaticalization process may be, and often is, arrested before it is fullyimplemented . . .

    Indeed, despite the frequent references encountered to universal paths taken ingrammaticalization (e.g. Bybee, 1985, p. 201), there is little evidence ofdeterminism. For example, Heine et al. (1991, p. 38) point to the fact that onesource concept can give rise to more than one grammatical category and that,conversely, a given grammatical category may be historically derived from morethan one source concept or structure. In later work, Heine (1993, p. 91) gives noless than eight distinct possible grammaticalization outcomes for verbs whosemeaning translates as come. And as far as a particular category arising from amultitude of sources is concerned, Heine shows that tense-aspect markers canoriginate from full verbs, from adpositions, and from adverbs.To summarize, grammaticalization fails to evince the most important

    distinguishing feature of a distinct process the unfolding of its component partsin a determinate sequence in which one step of the sequence inevitably engendersthe following one.

    3.3. The independence of the component parts of grammaticalization

    I will now provide evidence that the component parts of grammaticalization can

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 195

  • occur independently of each other. We find the relevant semantic changes withoutreanalysis or phonetic reduction (Section 3.3.1), phonetic reduction withoutsemantic change (Section 3.3.2), and reanalysis without semantic change orphonetic reduction (Section 3.3.3) (see Campbell, this issue).

    3.3.1. The independence of the semantic changesGrammaticalization is frequently characterized in terms of an increase in

    metaphorical abstractness: Grammaticalization may also be viewed as a subtypeof metaphor . . .Grammaticalization is a metaphorical shift to the abstract(Matiso, 1991, p. 384). But such a definition fails in both directions. First, thereare semantic shifts observed in grammaticalization that are not properlycharacterized as metaphoric. For example, many languages have subordinatingconjunctions with a causal sense (English since, German infolgedessen, Frenchpuisque, and so on). Typically they derive from temporals, whose conversationallyimplicated causal sense has undergone conventionalization. As Traugott andKonig (1991) note, such changes seem more metonymic than metaphoric. Andsecond, metaphorical shifts to the abstract which are not accompanied by theother components of grammaticalization are commonplace. I will illustrate byreference to (5), which has been proposed by Heine et al. (1991, p. 48) torepresent a scale of metaphorical directionality: Any one of them may serve toconceptualize any other category to its right (p. 49). Thus, for example, an objectmay be conceptualized as a person, but not vice-versa. Note that each position onthe sequence (arguably) represents a greater degree of abstractness than theposition to its left:

    (5) PERSON>OBJECT>ACTIVITY>SPACE>TIME>QUALITY

    Heine et al. illustrate with numerous examples. Consider the PERSON >OBJECT>QUALITY progression of metaphors:

    [M]any languages have a comitative adposition with that is used to refer toinstruments with inanimate nouns and to manner with certain abstract nouns.Metaphor, in this instance, has the eect of conceptualizing an instrument as acompanion and a quality as an instrument . . .(Heine et al., 1991, p. 52).

    The OBJECT>SPACE metaphor might be exemplified by the use of body-partterms to denote spatial orientation (cf. the extension of the meaning of chest tonext to in Finnish); the SPACE> TIME metaphor by the grammaticalization ofadpositions meaning behind into subordinating conjunctions meaning after; andso on.The problem is that these metaphors are rampant in all languages in a variety

    of circumstances and seem only tangentially connected with grammaticalization.Saying Ive read a lot of Barbara Vine instead of Ive read a lot of BarbaraVines novels illustrates the PERSON > OBJECT metaphor. Calling anexamination room a torture chamber extends an object to a quality. The

    F.J. Newmeyer / Language Sciences 23 (2001) 187229196

  • OBJECT > SPACE metaphor was presumably involved in the creation ofexpressions like the mouth of the river, the eye of the needle, and so on. Wefind the OBJECT>TIME metaphor in the derivation of month from moon, andpossibly the SPACE > TIME metaphor is exemplified by Paul TherouxsAustralian in Riding the Iron Rooster who measured time in terms ofKalgoorlies. It appears to be the case, then, that there are no metaphoricalextensions of meaning that are unique to grammaticalization.The same point could be made with respect to the conversational implicatures

    identified with grammaticalization. As noted above, it is not uncommon to find anelement with a temporal meaning taking on a conversationally implicated causalsense. But there is nothing specific to grammaticalization in such a development.As Traugott and Konig (1991, p. 194) point out, the classic fallacy of post hocergo propter hoc pervades our everyday use of language. Note that in thefollowing examples, temporality has been strengthened to causation (for detailedlinguistic and philosophical discussions of these and related cases, see Geis andZwicky, 1971; Atlas and Levinson, 1981; Horn, 1984):

    (6) a. After we heard the lecture, we felt greatly inspired (implicates because ofthe lecture, we felt greatly inspired).

    b. The minute John joined our team, things started to go wrong (implicatesbecause John joined our team, things started to go wrong).

    I must agree with Sweetser (1988, p. 389) that one should attempt to treat thesemantic changes attendant on grammaticalization as describable and explicable interms of the same theoretical constructs necessary to describe and explain lexicalsemantic change in general. Surely, that is the most parsimonious route to takeand the one we should adopt in absence of any disconfirmatory evidence. Such aroute leads inevitably to the rejection of the idea that grammaticalization is adistinct process.

    3.3.2. The independence of phonetic reductionI do not disagree with the observation of Heine (1994, p. 267) that When a

    given linguistic unit is grammaticalized, its phonetic shape tends to undergoerosion. But the question is why this should be the case. In this section I willargue that the phonetic reductions identified with grammaticalization areexplicable by reference to forces that must be posited and explained independently of grammaticalization itself. These are essentially least-eort forcesthat lead to more frequently used items being, in general, shorter than lessfrequently used ones.Heine (1993, pp. 109111) provides four factors said to account for the erosion

    observed in grammaticalization (see also Heine et al., 1991, p. 214). The first is theQuantity Principle (Givon, 1991a), according to which there is an iconicrelationship between the amount of information conveyed and the amount ofcoding. The second is a least-eort principle in the sense of Zipf (1935): Peoplewill shorten the linguistic expressions that are used most commonly for economy,

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 197

  • that is to simplify their linguistic utterances . . . The third is based on the idea thatthere is a positive correlation between the information value of a linguisticsymbol and the amount of coding material employed to express it. And the fourthis a physical motivation, along the lines suggested by Gabelentz (1891). Since themore a physical entity is used, the more rapidly it is likely to wear out, it followsthat the frequency of use of a grammatical element leads to a reduction of itsphonetic substance.Whatever the status of the Quantity Principle may be, I fail to see its general

    relevance to grammaticalization3. It is by no means obvious, for example, that anax conveys less information than a full form or that an element with the statusof a functional category conveys less information than one with the status of alexical category. In some conversational situations, for example, the lexicalmaterial coding tense and aspect is wholly extractable from the context of thediscourse; in others conveying clearly that the proposition expressed is in the pasttense rather than in the present tense can make all the dierence in the world.Have in the sense of perfect aspect does not convey less information than have inthe sense of possess; it simply conveys dierent information. Sweetser (1988, p.389) makes the same point with respect to the development of future morphemesfrom words meaning go:

    [In this development] we lose the sense of physical motion (together with all itslikely background inferences). We gain, however, a new meaning of futureprediction or intention together with its likely background inferences. Wethus cannot be said to have merely lost meaning; we have, rather, exchangedthe embedding of this image-schema in a concrete, spatial domain of meaningfor its embedding in a more abstract and possibly more subjective domain.

    Furthermore, in a simple negation like Mary doesnt study linguistics any more,does surely conveys less information than not (does does convey any information?),yet it is the latter that is reduced, not the former. Perhaps there is some way thatthe notion amount of information conveyed can be characterized precisely. Untilthen, I feel that we can safely dismiss the Quantity Principle as a factorcontributing to phonetic reduction.Heines third and fourth motivations seem equally suspect. As for the third, I

    fail to distinguish it from the first it seems no more than a restatement of theQuantity Principle. The fourth, if interpreted literally, is absurd whateverwords are, they are manifestly not physical entities. And if taken loosely, it seemssimply to be a restatement of the least-eort principle.We are thus left with the least-eort principle as the primary cause of the

    phonetic reductions that accompany grammaticalization. Well over a half-century

    3 The quantity principle is presumably an instantiation of Grices maxim of quantity: Make your

    contribution as informative as required (for the current purposes of the exchange) (Grice, 1975, p. 45).

    For other approaches to the grammatical correlates of predictability or familiarity of information con-

    tent, see Prince (1981, 1985) and Gundel et al. (1990).

    F.J. Newmeyer / Language Sciences 23 (2001) 187229198

  • ago George Zipf observed that frequently used expressions and grammaticalelements are, on the whole, shorter than rarely used ones:

    High frequency is the cause of small magnitude . . .A longer word may betruncated if it enjoys a relatively high frequency [either] throughout the entirespeech community [or] if its use is frequent within any special group (Zipf,1935, pp. 3132).

    Zipf cited studies showing that the average length of both words andmorphemes declines in proportion to the frequency of their use (for more recentfindings along these lines, see Nettle, 1995). Extending these results, I wouldpredict that the following propositions are also true (though I know of noconfirmatory studies):

    (7) a. The average member of a lexical category tends to be both longer andless frequently used than the average member of a functional category;

    b. The average member of a functional category tends to be both longerand less frequently used than the average clitic.4

    If (7ab) are true, then the erosion associated with reanalysis may simply be aleast-eort response having nothing to do with grammaticalization per se. That is,in order for the observed fact of phonetic erosion to bear on whethergrammaticalization is a distinct process, one would have to demonstrate somespecial linkage between erosion and the other component parts ofgrammaticalization not attributable to frequency. That might involve, for example,showing that given two equally frequently used homophonous senses of a singlemorpheme, one grammaticalized and one not, the grammaticalized sense is themore reduced of the two. I know of no examples illustrating this. Indeed, asHeine himself notes, the English copula be reduces in its main verb usage as wellas in its grammaticalized auxiliary sense (Heine, 1993, p. 111):

    (8) a. Hes going.b. Hes criticized every day.

    (9) a. Hes a farmer.b. Hes sick.

    Fidelholtz (1975) has confirmed the idea that erosion is at least a part afunction of frequency, independent of grammaticalization eects. Hedemonstrated that the degree of stress reduction on a lax vowel in a pretonic

    4 What about clitics vis-a`-vis axes? Lyle Campbell (personal communication, 1966) points out that

    in Romance and Finno-Ugric, at least, clitics do not tend to be longer than axes. One would predict

    then that the grammaticalization of clitics to axes would be less common than that of lexical to func-

    tional categories or of functional categories to clitics.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 199

  • syllable correlates with its frequency (I repeat his results as summarized in Horn,1993, p. 35):

    (10) a. astronomy ga`stronomyb. mstake m`stookc. abstain a`bstentiond. mosquito Mu`skegon [city in Michigan]

    And an empirical study by Johnson (1983) supports the idea that there is acorrelation between the frequency of a word and the extent to which fast speechrules apply to it. Johnson analyzed several hours of recorded natural speech andfound the only systematic correlation between the application of optional fastspeech processes and any other factor involved the frequency of the wordoperated on by the rule.One good indication that the phonetic reductions observed in

    grammaticalization are simply a subset of those at work in general in fast speechcomes from the fact that they are just as likely to lead to marked syllablestructures as those not implicated in grammaticalization. Consider, for example,the reductions that have taken place in cant and wont we (Forner et al., 1992, p.87):

    (11) a. cant /knt/ 4 [knt] 4 [kt] 4 [kc]b. wont we /wont wi/ 4 [wontwi] 4 [wocwi]

    Presumably if grammaticalization were a distinct process, guided by principlesindependent of those operative elsewhere, one would expect it to avoid thecreation of such marked structures.The frequent use of a form, it might also be pointed out, is no guarantee that it

    will set into motion other grammaticalization eects. Heine et al. (1991, p. 39),citing a study by Bertoncini (1973), note that of the fifteen most frequently usedwords in Swahili, none has been a source for grammaticalization.If, following Slobin (1977), the maximization of processibility has been an

    important factor in language change, it is clear that as a general trend frequencyeects will manifest themselves as reductions. In short, one does not need a theoryof grammaticalization to explain why unbound forms are more likely to becomebound than vice-versa any more than one needs a theory of ice cream to explainwhy it is easier to melt together a scoop of chocolate and a scoop of vanilla in thesame bowl than to reconstitute the original scoops from the melted mass. In theformer case, elementary facts about speech production and perception will sucevery well.Finally, it should be pointed out that many phonetic reductions observed in

    grammaticalization are simply the eects of regular phonological change occurringelsewhere in the language and have no bearing on grammaticalization per se (infact, this point is made in Hopper and Traugott, 1993, p. 147). For example, the

    F.J. Newmeyer / Language Sciences 23 (2001) 187229200

  • phonological reductions in the development of the Modern Greek future markertha from the Classical and early Post-Classical Greek verb thelei want are oftencited as exemplifying the grammaticalization process (Meillet, 1912). But Joseph(1996) shows that some of these reductions were simply manifestations of regularsound change and hence lend no support to the idea that grammaticalization hasan independent reality.

    3.3.3. The independence of reanalysisThe structural reanalysis that is characteristic of grammaticalization also occurs

    without the other features that are typically associated with this phenomenon.Haspelmath (1998) gives a several examples illustrating reanalysis withoutsemantic or phonetic changes, two of which I will now present.In standard German, the complement of the verb anfangen begin is a sentential

    infinitival clause, obligatorily postposed after the finite verb:

    (12) Sie fangt an, zu singen.She begins to sing

    However in Haspelmaths own dialect of German, the infinitival phrase precedesthe finite verb and the verbal particle an precedes the infinitive, as in (13a). Thisorder is impossible with other verbs (13bc):

    (13) a. Wenn sie an zu singen fangt, . . .If she begins to sing, . . .

    b. Wenn sie auf zu singen hort, . . .If she stops singing, . . .

    c. Wenn sie ein zu heiraten willigt, . . .If she agrees to get married, . . .

    Haspelmath makes the reasonable conjecture that (13a) originated as a reanalysisof (12), as in (14b). In this reanalysis, an apparently became part of the complexinfinitival complementizer:

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 201

  • But this reanalysis does not qualify as a grammaticalization none of the othereects associated with grammaticalization accompanied it.Another example discussed by Haspelmath is from French. In colloquial speech,

    the postverbal 3rd person subject pronoun has been reanalyzed as an interrogativeparticle ti (see Harris and Campbell, 1995, for discussion):

    (15) a. Votre pe`re [V[part-PRO[il]]?Does your father leave?b. Votre pe`re V[par]PTCL[ti]?Does your father leave?

    Here again, the other diagnostics for grammaticalization are missing.

    3.4. Summary

    We have examined the component parts of grammaticalization and found thatthey all are manifested independently. By a (not particularly useful) convention ofusage, it has become customary to use the term grammaticalization only when adowngrading reanalysis happens to coincide with certain independently occurringsemantic and phonetic changes. Fig. 1 gives a schematic illustration of howgrammaticalization is simply the set union of the three historical changes.Definitions are free for the making; one obviously has the right to use the term

    grammaticalization to describe the conjunction of certain types of historicalchanges that are manifested independently. No harm is done as long as the use of

    Fig. 1. Grammaticalization as an epiphenomenon.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229202

  • the term in such a way does not invite the conclusion that some dynamic is atwork in grammaticalization that cannot be understood as a product of thesehistorical changes.

    4. On unidirectionality

    This section probes further the question of the purported unidirectionality ofthe conjunction of eects associated with grammaticalization. In Section 4.1, Iphrase the issue in such a way as to give unidirectionality the status of anempirical hypothesis. Section 4.2 makes the point that if grammaticalization werea distinct process, then unidirectionality would be expected as a naturalconsequence. The lengthy Section 4.3 provides numerous counterexamples tounidirectionality, while the concluding Section 4.4 is addressed to why categorialdowngradings (and their associated eects) are much more common thanupgradings.

    4.1. Phrasing unidirectionality as an empirical hypothesis

    Any attempt to evaluate the correctness of the unidirectionality principle mustbegin by making sure that it is phrased as an empirical hypothesis. Such is notalways the case in functionalist accounts of grammaticalization often we findunidirectionality built into the very definition of grammaticalization, as thefollowing quotations illustrate:

    Where a lexical unit or structure assumes a grammatical function, or where agrammatical unit assumes a more grammatical function, we are dealing withgrammaticalization (Heine et al., 1991, p. 2).

    We define grammaticalization as the process whereby lexical items andconstructions come in certain linguistic contexts to serve grammatical functions,and, once grammaticalized, continue to develop new grammatical functions(Hopper and Traugott, 1993, p. xv).

    In other words, grammaticalization is defined as a unidirectional process. Supposethat in some particular case, directionality appeared to be reversed, that is,suppose we observed a lexical item or construction that had developed a lessgrammatical function. Such a circumstance would not bear in the slightest onwhether grammaticalization is unidirectional or not, because nothing could bearon that question. The point is that definitions make no empirical claims. Settingaside for the moment the question of whether grammaticalization is a distinctprocess or not, one might feel that the above definitions are either useful (i.e. theypick out some segment of reality) or not useful (there are not enough instances ofwhat they pick out to warrant such a definition), but, as is the case withdefinitions in general, they cannot be either right or wrong.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 203

  • But elsewhere we find discussions of unidirectionality that can only beinterpreted so as to suggest that it is taken to be an empirical hypothesis:

    The strong hypothesis of unidirectionality claims that all grammaticalizationinvolves shifts in specific linguistic contexts from lexical to grammatical item, orfrom less to more grammatical item, and that grammaticalization clines areirreversible. Change proceeds from higher to lower, never from lower to higheron the cline. Extensive though the evidence of unidirectionality is, it cannot beregarded as an absolute principle. Some counterexamples do exist (Hopper andTraugott, 1993, p. 126; emphasis added).

    In the remainder of this section, I will treat unidirectionality only as anempirical hypothesis. That is, I will take it as a claim about language change. Iwill assume that if unidirectionality is true, then grammatical elements maydowngrade (i.e. increase in degree of grammatical content morphosyntactically),but never upgrade (i.e. decrease in degree of grammatical contentmorphosyntactically). As we will see, the claim is false.

    4.2. If grammaticalization is a distinct process, then unidirectionality (if true) isuninteresting

    The purported (near) unidirectionality of grammaticalization is typicallypresented in the functionalist literature as a remarkable fact, and one which wouldnever have been uncovered by the standard methodology of generative grammar.But suppose that grammaticalization were, in fact, a distinct process. If so, thenunidirectionality would be the most unremarkable fact imaginable5. The reason isthat unidirectionality is a property of natural processes in general. Such processesare either entirely irreversible or they can be reversed only given huge amounts ofeort, time, unusual circumstances, and so on. Consider the aging process as aprototypical example! Or consider any chemical or physical process. Mountainsare eroded and washed down to the sea; mountain-creating mechanisms in no wayinvolve sand grains flowing upstream followed by de-erosion. Sodium hydroxideand hydrochloric acid react to create table salt and water. Extreme amounts ofenergy amounts never encountered under natural conditions on Earth arerequired to reverse the process. Biological evolution has never been known torecreate an ancestral form. And so on.In short, given that unidirectionality is a general characteristic property of

    natural processes, we would expect it to hold of distinct grammatical processes aswell. If grammaticalization were a distinct process, its unidirectionality should notmerit more than a footnote. Indeed, it would be the nonoccurrences ofunidirectionality that should occupy the bulk of the attention of those who subject

    5 I owe this observation to Richard Janda (personal communication, 1966).

    F.J. Newmeyer / Language Sciences 23 (2001) 187229204

  • it to study. But since, as we have seen, grammaticalization is not a distinctprocess, we have no a priori reason to expect rigid unidirectionality.

    4.3. Unidirectionality is not true

    Less than twenty years ago, Christian Lehmann (1982/1995, p. 19) could assertthat no cogent examples of degrammaticalization [i.e. reversals of directionality FJN] have been found. More recently, Heine et al. (1991) have remarked that[a]lthough both degrammaticization and regrammaticization have been observedto occur, they are statistically insignificant . . . (pp. 45). Unfortunately, however,they fail to provide any statistics to back up their claim. My sense is that suchphenomena are rampant, though I would not hazard a speculation about theirstatistical breakdown. In this section, I will document some of the best knowncases of upgrading and conclude with some remarks on their theoretical import.None of these cases will be presented in full detail; the reader might wish toconsult the cited literature for a complete account (see also Campbell et al., thisissue).First, however, I must stress an important methodological point. I take any

    example of upgrading as sucient to refute unidirectionality. Occasionally, onereads of certain upgradings not really counting as counterexamples tounidirectionality because they are not genuine cases of grammaticalizationreversing itself. Rather, they are said to manifest some other process, such as, saylexicalization. My feeling is that attributing upgradings to some process distinctfrom the inverse of grammaticalization is tantamount to covertly buildingunidirectionality into the definition of grammaticalization. Certainly it would havethe eect of ruling out the great majority of potential counterexamples tounidirectionality.In any event, it is certainly the case that complete reversals of

    grammaticalization are extremely rare, perhaps nonexistent. This should hardly bea cause of surprise; as noted by Janda (1995, 1998), given the predominantarbitrariness of the sound/meaning association in linguistic signs and thephonological deformations that accompany downgrading, it would be nothing lessthan a miracle if some aspect of the precise earlier stage of a language wererecreated in degrammaticalization.

    4.3.1. The problem posed by the lexicalization of axesThe claim of unidirectionality in its strongest form is refuted by the

    phenomenon of lexicalization, which is widepread in the languages of the world.One manifestation of this phenomenon is the fusion of axes to the root to createa new, morphologically opaque, lexical item. Many such examples are discussed inRamat (1992). English drench derives from bimorphemic drank-jan cause to bewet; the ancestors of English wash and German forschen to search, investigatecontained an inchoative morpheme (cf. Latin rubesco, pallesco to go red, to turnpale); English elder and French maire once contained a causative morpheme; andso on.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 205

  • What these examples show is that there is an alternative developmentalpossibility for an ax than to weaken to zero. Now, none of these examplesillustrate the upgrading of the ax in and of itself (for such examples, see belowSections 4.3.24.3.4) and in that sense do not refute the claim that a particularelement may move only from the less grammatical to the more grammatical.However, they do indicate a recycling process from morphology to lexicon andhence a more complex picture of linguistic evolution than is typically presented inthe grammaticalization literature. The obvious consequence of this more complexstate of aairs is for historical reconstruction. Contrary to what has becomestandard practice among grammaticalization theorists, internal reconstructionsbased on the assumption that full ungrammaticalized lexical items are startingpoints in historical development must be viewed as methodologically illicit. Thepossibility that such items are themselves the result of a feeding process from themorphology must always be entertained.

    4.3.2. From inflectional ax to derivational axAn example of an inflectional ax upgrading to a derivational ax is provided

    in Lehmann (1989; discussed also in Ramat, 1998). The Latin inflectional presentparticiple sux -nt evolved into the French derivational adjectival sux -ant (cf.Latin currens running and French courant current).Greenberg (1991) gives many examples of what he calls regrammaticalization,

    an important component of which involves an inflectional ax being upgraded tothe status of derivational ax. Citing both attested and reconstructed forms, heoutlines a common path of development from demonstrative to definite article tonoun marker to a derivational ax of nominalization (as in some Bantulanguages) or to one signaling transitivization or causativization (as in theChibchan languages). A somewhat similar development is mentioned in Lehmann(1982/1995). Tok Pisin grammaticalized the English pronoun him as an inflectionalobject marker -im. This element subsequently upgraded to a derivational markerof verbal transitivity. Hopper and Traugott (1993, p. 166) regard suchdevelopments (whose historicity they do not dispute), not as examples of a radicalshift in directionality but rather as natural examples of the sort of generalization,spread, and splitting into dierent functions that accompanies ongoinggrammaticalization. Natural or not, however, they are in fact incompatible withthe strongest hypothesis of unidirectionality, which sees a relentless shift to themore grammatical.

    4.3.3. From inflectional ax to cliticOur first example of an ax upgrading to clitic status was called to my

    attention by Martin Haspelmath and Anders Holmberg (personal communication)and is discussed in Fiva (1984) and Norde (1997a). This involves the developmentof the Old Norse genitive ax -s to the modern Mainland Scandinavian -s, which,as in English, is a clitic marker on the full NP. Consider the following examplesfrom Norwegian:

    F.J. Newmeyer / Language Sciences 23 (2001) 187229206

  • (16) a. NP[NP[den gamle mannen]s hus]

    the old man-thes house

    the old mans house

    b. NP[NP[den gamle mannen med skjegget]s hus]

    the old man-the with beard-thes house

    the old man with the beards house

    Unlike what seems to have happened in English, however (see Janda 1980, 1981),

    there is no possibility that the clitic could be a reduction of his. We have a simple

    case of pure upgrading from ax to clitic. Further evidence is provided by wh-

    movement. Norwegian has two possessive constructions, (17a) and (17b), the

    former with cliticized -s and the latter with sin, the usual possessive reflexive:

    (17) a. Pers bil

    Peters car

    b. Per sin bil

    Peter his (refl) car

    Peters car

    The -s form is older in the sense that it goes back to the Old Norse genitive. The

    form with sin is said to have come into the language in the Middle Ages from

    Low German merchants in Bergen. Normally, under Wh-Movement, both sin and

    s- are extracted along with the wh-pronoun:

    (18) a. Hvem sin (bil) er det

    who REFL (car) is that

    Whose (car) is that?

    b. Hvems (bil) er det

    who+s (car) is that

    Whose (car) is that?

    Sin can regularly be stranded under Wh-Movement:

    (19) Hvem er det sin (bil)

    who is that REFL (car)

    Whose (car) is that?

    Remarkably, for many speakers, -s can be stranded as well, cliticizing to the

    element preceding it. Here, of course, an analysis as an ax is out of the question:

    (20) Hvem er dets (bil)

    who is it+s (car)

    Whose (car) is that?

    Holmberg has even observed forms like (20) in Swedish child language, where no

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 207

  • equivalent of the sin form exists. It would appear, then, to be a spontaneousreanalysis independent of direct Low German influence.For other examples of the upgrading of an inflectional ax to a clitic, see Janda

    (1995) on developments in New Mexican and other nonstandard dialects ofSpanish; Nevis (1986a) and the Campbell paper in this issue on upgrading inLappish; and Garrett (1990) and Joseph (1996) on the rise of clitics frominflectional axes in the transition from Proto-Indo-European to Hittite andGreek.

    4.3.4. From inflectional ax to wordBybee et al. (1994) give an example from the history of Irish of an inflectional

    ax becoming a pronoun. Earlier stages of the language had a rich set of person/number agreement suxes on the verb. All but one of these have been lost andreplaced by obligatory subject pronouns. The exception is the first person pluralsux -mid/-muid, which is now an independent pronoun (replacing the earlier firstperson pronoun sinn ):

    (21) Is muide a rinne ebe 1P.EMP who do.PAST itIts we who did it.

    Bybee et al. remark that there was strong paradigmatic pressure for the reanalysisof [this] person/number sux as a free pronoun (p. 13), though they do notelaborate. But paradigm pressure is hardly a rare motivating factor in historicalchange. If paradigm pressure is able to reverse unidirectionality, it suggests thatpressure towards unidirectional change is perhaps not so overpowering as issometimes claimed. In other words, we might expect to find many more caseswhere unidirectionality has lost out to paradigm pressure.Another example is taken from Rubino (1994). An inflectional sux -Vt is

    reconstructed for Proto-Semitic to mark accusative case on nominals. It shows upin Akkadian only on pronouns (22) and in Kemant (Central Cushitic) onpronouns and definite nouns (23):

    (22) yati 1sACCk(u)ati 2sACC

    (23) -y3t 1sACC-k3t 2sACCN+-t definite accusative nouns

    In Modern Hebrew this sux has developed into a preposition, cet, whichprecedes nouns for which it specifies for definite accusativity:

    (24) a. racti cet ha-isI:saw DEF:ACC the:manI saw the man

    F.J. Newmeyer / Language Sciences 23 (2001) 187229208

  • b. sgor cet hadeletclose DEF:ACC the:doorClose the door

    4.3.5. From derivational ax to wordOne area where we find rampant upgrading and hence clear counterexamples

    to unidirectionality is the relatively common development whereby aderivational ax is detached from its stem and lexicalized as an independentword. This has happened to the sux -ism, for example, not just in English, butalso in Finnish and Estonian (Nevis, 1986b) and in German and Italian (Ramat,1992), in each case acquiring the meaning doctrine, theory. In Italian, -anta (asin quaranta 40) has become used as a noun meaning older than 40: ha passatogli anta he is over 40 (Norde, 1997b, citing Anna Giacalone Ramat). Ade, teen,ex, pro, and many other words of English were at one time restricted to use asderivational axes. And this type of diachronic change is productive witnessthe title of the movie Boyz N the Hood, where hood is AfricanAmerican slang forneighborhood.A particularly interesting example provided by Norde (1997b) is that of the

    derivational sux -tig in Dutch (cognate with English -ty ) which has developedinto an independent word with the meaning umpteen, zillion (see also Norde,this volume).Trask (1997) reports on a series of upgradings of derivational axes to words

    in the history of Basque. Due to lack of adequate records, we cannot beabsolutely sure about cases like ume child and -kume ospring, the latter ofwhich appears in numerous formations like katakume kitten; and ohi habit,custom and -koi fond of. But the Bizkaian dialect has recently converted thesux into a noun kume (or, really, kuma, with dialectal final-vowel lowering)young animal. Clearer cases are talde group and toki place. These derive fromthe synomymous word-forming suxes -alde and -oki. We may be sure of thisorder of development, since, first, the suxes are archaic the latter is confinedto place names and, second, no native Basque word of any antiquity beginseither with a voiceless plosive or with a coronal plosive (apart from three or fourpuzzling cases with initial /k/). More recent still is the noun tasun quality,derived from the common noun-forming sux -(t)asun -ness. The noun is noteven attested before the late nineteenth century, but today it is moderatelyfrequent and has given rise to derivatives like tasunezko qualitative.

    4.3.6. From clitic to wordThe upgrading of clitic pronouns to full lexical pronouns seems to have

    happened repeatedly. Kroch et al. (1982) give an example from the history ofEnglish. In the sixteenth century, inverted subject pronouns such as those in (25)were encliticized to the verb:

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 209

  • (25) a. Where dwellyth she?b. Why bewaylest thou thus soore O Pelargus?

    Their clitic status is indicated by the spellings of the period, in which the verb andpronoun are represented as orthographic units: hastow, wiltow, and wille for hastthou, wilt thou and will he respectively, as well as the spelling of ye as y. Butthere is considerable evidence that the pronoun underwent decliticization after1550. For example, not appears for the first time between the inverted main verband the pronoun, instead of occurring after the subject. Also at this time youbegan to displace ye as the second person subject pronoun. This is important,since before that time the strong form you was not used in this construction,except for emphasis. In other words, as decliticization took place, a preexistingstrong pronoun was coopted to replace the clitic.Consider free-standing relative/indefinite/interrogative words in Indo-European

    languages, such as Latin quis, quod, Greek hos, Sanskrit yas, and Hittite ku-is, ku-it. It is, I believe, uncontroversial among Indo-Europeanists that they deriveetymologically from PIE enclitic particles. These particles followed the firstaccented word in a clause in PIE and were themselves followed by one or moreclitics (for details, see Watkins, 1963, 1964; Gonda, 1971; Jeers and Pepicello,1979). In the course of a discussion of these developments, Jeers and Zwicky(1980, p. 224) conclude that the tacit assumption that clisis is invariably one stagein an inexorable development toward the status of an ax, or toward ultimateoblivion, is simply false6.Rubino (1994) gives an Ilocano example of the development of a clitic to a free

    word. This Austronesian language spoken in the Philippines has a future enclitic -to, with two allomorphic variants -to, after consonants, and nto, after vowels:

    (26) a. Mapan-ak-toINTR:go-1sABS-FUTIll go

    b. Mapan-ka-ntoINTR:go-2sABS-FUTYoull go

    This clitic is attested even in pre-Hispanic times, with no traces of it innoncliticized form. In the modern language, however, it occurs unbound incolloquial speech, usually in an armative response to a question or request foran action to be done in the future:

    (27) Um-ay-ka no bigat, aINTR-come-2sABS FUT morning PARTCome tomorrow. Okay?

    6 For a somewhat similar development in the development of Estonian, see Nevis (1986c) and Camp-

    bell (1991).

    F.J. Newmeyer / Language Sciences 23 (2001) 187229210

  • ToIll do that

    Rubino provides a parallel case from Hungarian (he cites Robert Hetzron, p. c.for the data). This language has a versatile enclitic particle -is, which can beglossed as also and as an emphatic marker, among other things:

    (28) a. Jancsi-is tudja eztJohnny also knows:it this:ACCJohnny also knows this

    b. Tudja-is a valasztknow:PAST the answer:ACCHe did indeed know the answer

    In Modern Hungarian is can be used as a separate word with the sense of indeed(as in (29), where evidence from stress indicates that it is not a clitic) and, inreduplicated form, as an armation of both conjuncts of or questions (30):

    (29) Jancsi meg is erkezettJohnny did indeed arrive

    (30) Kulfoldrol hozzak vagy itt gyartjakabroad:from they:bring or here they:manufacture:itDo they import it or manufacture it here?Is-sREDUP PARTboth

    4.3.7. From functional category to lexical category (and pronoun to noun)The history of English for (Van Gelderen, 1996) shows that even what appears

    at first glance to be a unidirectional change can be more complex than meets theeye. Superficially, the development of this item fulfills the standardgrammaticalization scenario. In Old English, for was a preposition indicatinglocation, translatable roughly as in front of (cf. German vor ). By the late OldEnglish period it had grammaticalized to causative and benefactive senses. Its firstuse as a complementizer (indicating further grammaticalization) is attested around1200. In this use it is never separated from the infinitive marker to. It is only morethan a century later that for and to could be separated by a pronoun in theobjective case ( for him to reade ). In other words, for actually regainedprepositional case-assigning properties. While still functioning as a clausesubordinator, it became less of a grammatical item than it had been previously.A rather dierent case is discussed in Van Gelderen (1997). The origin of

    Modern English expletive there lies not in the locative adverb, as is sometimesclaimed, but in the Old English demonstrative /re. The progression taken by

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 211

  • /re (and by demonstratives /ara and /t as well) was from demonstrative torelative pronoun to expletive pronoun. If there is a lexical noun, as is generallyassumed, then we have an example of an upgrading from determiner to noun, thatis, from a functional category to a lexical category.As Van Gelderen (1997) shows, the history of the word man also presents

    challenges for any sweeping claims about unidirectionality. In Old English itspredominant use was as an indefinite pronoun (cf. German man ). Subsequently itseems to have swung back and forth from pronoun to full lexical noun and backagain. In any event, it is the less grammaticalized use that has survived intoModern English.The upgrading of prepositions to full verbs is widely attested. For example, the

    Spanish verb sobrar to be extra, left over derives from the preposition sobreover, on, above. Harris and Campbell (1995) note that in Central AmericanSpanish the preposition dentro inside has undergone the same development,becoming the root of dentrar to insert. We find similar developments in English7:up the ante, down three drinks, o the pigs (1960s student protester slogan).Pronouns as well can upgrade to verbs as the German duzen, French tutoyer,Spanish tutear, and Finnish sinutella to use the familiar form attest (cf. Norde,this issue).Prepositions and conjunctions may be lexicalized as adjectives, as an iy

    situation (Ramat, 1987). We also find English adjectives for and against (Howmany of you are against?), through in the sense of finished and on and o (Imjust not on today ). Also note that German zu developed an adjectival meaningclosed (eine zuene Tur a closed door), ultimately from zugemacht closed(Janda, 1998).Prepositions and conjunctions can upgrade to nouns, as is illustrated by inn

    (from preposition in ), bye (sporting term, from preposition by ), and out (as in Ihave an out ). Note also the ups and downs and the ifs and buts, and the Frenchderrie`re behind and devant in front of, which have developed the meaningbuttocks and front respectively. There also seem to be cases of prepositions thatderive historically from nouns, and have later reconverted to nounhood: inside,outside, and front (in certain uses).

    4.3.8. From hypotaxis to parataxisAt the level of the clause, not all changes have proceeded in unidirectional

    fashion toward increased hypotaxis. As Matsumoto (1988) has shown, in Japanesethe precise opposite has happened. In the modern language, there are two ways ofexpressing the proposition Although Taro is young, he does a good job, one by

    7 Martin Haspelmath has suggested to me (personal communication, 1966) that many of the preposi-

    tion-to-lexical category examples that I cite from English are actually examples of adverbs changing cat-

    egory. The etymological sources that I consulted have not, in general, been explicit on this point. One

    must bear in mind, in any event, that a well accepted analysis of many adverbs that are homophonous

    with prepositions takes them in fact to be (intransitive) prepositions (Emonds, 1972).

    F.J. Newmeyer / Language Sciences 23 (2001) 187229212

  • the simple conjunction of the two main propositions (31a), the other by use of theadversative subordinating sux -ga (31b):

    (31) a. Taro-wa wakai(-yo). Ga, yoku yar-u(-yo)Taro-TOP young. but well do-PRESTaro is young. But he does a good job.

    b. Taro-wa wakai-ga, yoku yar-u(-yo)Taro-TOP young, well do-PRESAlthough Taro is young, he does a good job.

    The principle of unidirectionality would predict that sentences of the form of (31a)should be historically antecedent to those of the form of (31b). But this is not thecase. According to Matsumoto, paratactically-formed sentences such as (31a) havebeen recorded only since the seventeenth century, while the hypotaxis manifestedin (31b) is observed much earlier8.For an extended argument against the idea that hypotactic constructions in

    general develop out of paratactic ones, see Harris and Campbell (1995, chap. 10).

    4.4. Why unidirectionality is almost true

    While unidirectionality is false every conceivable type of upgrading that onecan imagine has actually occurred it is not all that false. Again, nobody is in aposition oer statistics on the matter, but a rough impression is thatdowngradings have occurred at least ten times as often as upgradings. So we havetwo questions to answer. The first, posed by Paolo Ramat, is why upgradingsoccur at all:

    It may be that degrammaticalization is statistically insignificant when comparedwith the large number of grammaticalization processes . . . , but its examples areby no means uninteresting, and not as scanty as one would prima facie inclineto admit. The question we have to deal with is therefore, why is it thatgrammaticalization and degrammaticalization coexist in natural languages?(Ramat, 1992, p. 553).

    And the second question is why the latter are so much rare than the former.These questions have, essentially, least-eort based answers. Less eort is

    required on the part of the speaker to produce an ax than a full form. Add theelement of frequency-caused predictability to the extreme amount of redundancyin grammatical codings, and it is not dicult to see why the quick-and-easy

    8 For more examples of this sort from the history of Japanese, see Onodera (1995) and for general

    discussion of their theoretical importance, see Traugott (1997). Traugott points out in this paper and in

    Tabor and Traugott (2000) and Traugott (2000) that the transformation of construction-internal el-

    ements into discourse particles is very common diachronically, thereby contradicting the scenario that

    decrease in syntactic freedom is an essential ingredient of grammaticalization.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 213

  • option of axation is frequently chosen. Other downgradings can readily beinterpreted as least-eort eects as well. Functional categories require less codingmaterial and hence less production eort than lexical categories. As a result,the change from the latter to the former is far more common than from theformer to the latter. All other things being equal, a child confronted with theoption of reanalyzing a verb as an auxiliary or reanalyzing an auxiliary as a verbwill choose the former.But all things are not always equal. There are other, conflicting, pressures that

    might lead the child to do the precise opposite. Analogical pressure might result inthe upgrading of an ax to a clitic or in the creation of a freestanding verb froman ax. Paradigm regularization, as we have seen, can lead to the reanalysis of anax as a free-standing pronoun. Now it would appear to be the case that thesavings in production eort generally outweigh the savings that would lead toan upgrading reanalysis. For that reason, we find downgradings greatly exceedingupgradings in frequency. Hence, just as with grammaticalization itself,unidirectionality is an epiphenomenal result of the interaction of other factors.Joseph and Janda (1988) provide a dierent sort of explanation (i.e. one not

    based on least-eort considerations) for why diachronic changes involving themorphologization of syntactic elements are far more widespread than those ofdemorphologization into the syntax. As they note, the morphologization ofphonological processes also tends to be far more widespread than the reverse. Oneof the most common attested changes is for a rule that at one point was subjectonly to phonological conditions to become subject to the properties of morphemesand their boundaries. They point out that some of the best-studied phenomena inthe history of our field attest to this fact: consider German umlaut, GrassmansLaw in Sanskrit, consonant mutations in Celtic, accent shifts in Modern Greek,among other developments. The preference for morphological over phonologicalsolutions is so strong that speakers will choose a fragmented morphologicalanalysis involving considerable allomorphy over a simple purely phonologicalsolution. Joseph and Janda cite the Maori example discussed in Hale (1973) andKiparsky (1971) in support for this idea. This language replaced an across-the-board phonological rule deleting word-final consonants with a large set of suxalallomorphs and a requirement that verb stems be lexically marked for their choiceof allomorph.We also cannot exclude the possibility that purely sociological factors are partly

    responsible for the general trend toward an increased degree of grammaticalcontent. Janda (1998) puts forward a scenario under which such might beexpected to take place. We have known since the work of Labov in the late 1960sthat a central motivation for younger speakers to generalize a sound change to anew context is that this allows them to show their solidarity with older members(by sharing participation in the change via use of common innovative forms) andyet to set themselves apart (by extending a variant to unique new contexts whereit is not in fact phonetically motivated) (Janda, 1998, p. 25). As Labov put it:

    Succeeding generations of speakers within the same subgroup, responding to

    F.J. Newmeyer / Language Sciences 23 (2001) 187229214

  • the same social pressure, carried the linguistic variable further along the process

    of change, beyond the model set by their parents (Labov, 1972, p. 178).

    Janda suggests an analogous eect in grammaticalization:

    There is no inherent reason, after all, why an analog of such socially

    functioning phonological extensions could not also be at work in

    grammaticization: i.e. once a form has been initially grammaticized, its furtherdowngrading could well be at least partly conditioned by the social function of

    age-group marking. (ibid)

    I consider the possibility that Janda raises to be a promising direction for

    grammaticalization research to take. Its primary obstacle as a general explanation

    of unidirectionality is the often painfully slow pace of the changes involved.

    Clearly if a form has remained stable over many generations before undergoingfurther downgrading, the younger members of the speech community would not

    be showing their solidarity with older members by downgrading it still further.

    Nevertheless, I would hope that empirical studies would be undertaken to see if

    Jandas scenario can be documented in the progress of some attested historical

    change.

    As we have seen, many reconstructions of historical antecedents to

    grammaticalized forms are based on assumptions about the unidirectionality of

    semantic change, as well as on the assumption of the unidirectionality of the otherparts of grammaticalization. Recall that it is assumed that change is always from

    the concrete to the abstract, that conversationally implicated meanings are

    conventionalized, but not the reverse, and so on. I suspect that, for whatever

    reason, there is a general directionality to the semantic changes observed in

    grammaticalization. But strict unidirectionality appears to be incorrect. One of the

    longest and most intricately detailed studies of grammaticalization is Frajzyngier

    (1996), which focuses its attention on the Chadic languages. On the basis of

    comparative evidence, Frajzyngier reconstructs for some of these languages lessabstract sequential markers as deriving from more abstract complementizers and

    temporal markers as deriving from conditional markers. He draws explicitly

    methodological conclusions from his findings:

    Thus the unidirectionality hypothesis with respect to grammaticalization from

    one grammatical morpheme to another is shown to be factually incorrect and

    certainly should not be used as a tool in grammaticalization research.In fact,bidirectionality appears to be the most likely possibility for those

    grammaticalization processes that involve metaphor and metonymic extensions

    (Frajzyngier, 1996, p. 467).

    Let us conclude this section with a general observation about the

    methodological consequences of the fact that unidirectionality is not true across

    the board. Many reconstructions are based, implicitly or explicitly, on the reliable

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 215

  • unidirectionality of every aspect of grammaticalization. For example, given theobservation that in some (normally unwritten) language we find homophonous Vand P with related meanings, it is assumed that the directionality of historicalchange was V>P. Assumptions of unidirectionality have led without exception tothe P being taken to represent the later development. Perhaps such claims are notas secure as one might wish them to be9.

    5. Two issues in grammaticalization research

    In this section, I will address two prominent themes in functionalist-orientedresearch on grammaticalization, one methodological and one theoretical. The firstinvolves the disturbing trend to use reconstructed forms as evidence, that is, totake reconstructions as results that can be accorded the same methodologicalstatus as attested historical changes. The second is the hypothesis of a panchronicgrammar, in which synchronic and diachronic statements find equal place.

    5.1. Using reconstructions as evidence

    Uncontroversially, one of the principal goals of historical linguistics is tohypothesize what unattested antecedent forms of a particular language or groupof related languages might have looked like. Certainly one of the major goals ofthe field of linguistics over the past two centuries has been to attempt to sharpenthe picture of the phonological system of Proto-Indo-European. But since thislanguage was spoken perhaps 6000 years ago and more than two thousand yearsbefore its oldest attested descendent, we have no right to call our latest version ofthis picture an established result. Rather, it is a simple hypothesis, awaitingimprovement, perhaps based on new data that have to date eluded our attention,or perhaps on improved theoretical conceptions pertinent to the technique ofreconstruction.A large share of grammaticalization research is devoted to historical

    reconstruction. Routinely in this literature hypotheses are put forward about theorigins of some grammatical marker or construction or, in more theoreticallyoriented work, about the general origins of particular types of elements. There isnothing pernicious about such reconstructions to the extent that they are based onthe sorts of internal and comparative evidence commonly appealed to in historicallinguistics. That does not guarantee that they are correct, of course, but, inprinciple, the reconstruction of a verb of possession as the historical antecedent ofa perfective marker is no more or less dubious than that of a sequence of a voweland nasal consonant as the antecedent of a nasalized vowel.

    9 For more reasons to doubt the assumption that there is an isomorphism between diachronic devel-

    opment of clines of grammaticalization and the synchronic relations among polysemous items, see

    Tabor and Traugott (2000).

    F.J. Newmeyer / Language Sciences 23 (2001) 187229216

  • Methodological problems do arise, however, with certain uses to which thesereconstructions might be applied. It would be unthinkable in mainstreamhistorical linguistics, for example, to take some reconstructed vowel nasalconsonant sequence and add it to a data base which included attested changes, sayfor the purpose of arguing for a particular theory of language change. However,in the functionalist-oriented research on grammaticalization, we find exactly thatsort of methodology. That is, we find reconstructions based on the hypothesis ofunidirectionality in grammaticalization being used as evidence to support theprinciple of unidirectionality.The following subsections illustrate my claim with reference to two of the best

    regarded lines of research in grammaticalization. The first (Section 5.1.1) is drawnfrom the work of Bernd Heine on the origins of grammatical markers in Africanlanguages. The second (Section 5.1.2) is the much-cited work of Joan Bybee andher colleagues on the origin of grammatical markers denoting futurity.

    5.1.1. Heine on progressives and locativesHeine (1994) notes that in the West African language Ewe there are formal

    parallels between progressives and locatives. One means of forming progressivesinvolves the free morpheme le and an -m sux on the verb:

    (32) Kofi le x& tu-mKofi PROG house build-PROGKofi is building a house

    As it turns out, le is synchronically also a locative auxiliary verb (be at) and me(the presumptive source of -m) exists as a locative preposition. On the basis of thehypothesis that spatial expressions are metaphorically employed to conceptualizetemporal notions (p. 269) and that the direction taken by the metaphor isunidirectional, Heine reconstructs progressives such as (32) as deriving historicallyfrom locative expressions.I have no problem with this reconstruction per se it may very well be

    correct. I do have a problem, however, with the next step that Heine takes,namely that of using this very reconstruction as evidence for grammaticalizationtheory. In his concluding remarks (pp. 277281), Heine asserts that this theory isable to explain the sequence of developments that led to the development of thelocative into the progressive. But a (unidirectional) sequence of developments wasassumed in the very reconstruction. There is no known sense of explanation inwhich the assumption of X to demonstrate Y can legitimately allow one toconclude that X has been confirmed.Such a use of a reconstructed form as evidence goes well beyond acceptable

    practice in historical linguistics. Its a little bit like reconstructing the Indo-European consonantal system on the basis of notions about the naturalness ofsound systems in general and then including that reconstruction in a sample ofsound systems in order to argue that some particular systems are more naturalthan others!

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 217

  • At least in the Ewe reconstruction, Heine points to a variety of factors in thecurrent language that can be taken as independent evidence for the correctness ofhis reconstruction. But after finishing his discussion of Ewe, he goes on to discussprogressive constructions in a number of other languages, most of which have nosignificant written history (these include Diola Fogny, Usarufa, Umbundu,Tyurama, and Maninka). Here he reconstructs diachronic ancestors for theprogressive in the absence of any independent evidence. The reconstructions arebased solely on the assumption of unidirectionality of semantic change. Yet thesereconstructions as well are appealed to as confirmation of the idea that there arenatural paths of development for progressive morphemes.

    5.1.2. Bybee et al. on the origins of future morphemesWe find similar methodological problems in the extensive writings of Joan

    Bybee and her colleagues on the origins of morphemes denoting future tense inthe worlds languages (Bybee and Pagliuca, 1987; Bybee et al., 1991; Bybee et al.,1994). A case in point is Bybee et al. (1991), whose goal is to test the followinghypotheses:

    1. That futures in all languages develop from a small set of lexical sources andthat all future morphemes from a given source go through similar stages ofdevelopment;

    2. That the semantic change in grammaticalization is accompanied by formalreduction, whereby the morpheme loses its independence and may fuse tocontiguous material (p. 18).

    After a lengthy discussion, the paper claims that these hypotheses have beenconfirmed:

    [T]here is, as predicted, a highly significant correlation between the semanticclassification we have imposed on the future grams [=grammatical morphemes]and their formal properties. Moreover, these correlations are in the predicteddirection the semantic properties hypothesized to belong to older gramscorrelate with the formal properties that are accrued in the process ofgrammaticalization. We conclude, then, that there is good reason to believethat future grams develop in very similar ways across unrelated languages, andthat semantic and formal changes move in parallel in grammaticalization (p.41).

    Unfortunately, however the authors motivate Hypothesis (1) partly on the basisof an untested assumption and partly by appealing to the correctness ofHypothesis (2). They motivate Hypothesis (2), in turn, primarily by assuming thecorrectness of Hypothesis (1). Let me illustrate.Since Bybee et al. (1991) draws heavily on Bybee and Pagliuca (1987), I will

    begin with a description of the earlier paper. The authors examined 50 languagesand found that the future morphemes in most of those languages bear anunmistakable resemblance to one or more morphemes with a fairly circumscribed

    F.J. Newmeyer / Language Sciences 23 (2001) 187229218

  • set of meanings, including desire, movement, possession, obligation,existence, and becoming. On the basis of virtually no discussion, they assumethat the historical sources (p. 111) for the future are verbs bearing thesemeanings, rather than vice-versa or considering the possibility that the sharedmeanings might have arisen from a common source. The only examples they giveof attested stages in the development of any future morphemes in any languageare English will, shall, and be going to, from desire, obligation, and movement,respectively, and as far as will is concerned, they acknowledge that a futuremeaning was present even in Beowulf.

    They give one argument that these modal flavors do not develop from thefuture meaning, but rather, when present, must be interpreted as retentions . . . (p.118), which I repeat in its entirety:

    The inflectional future paradigm in Quechua is heterogeneous the firstperson singular and plural exclusive inflections are based on an oldmovement morpheme (i.e. going to); the second person is identical to thepresent, and the third person sux appears to have evolved from anobligation marker. The evidence for this is that in most dialects its readingis prediction, but in the dialect of Cajamarca, it is used for obligation(must), probability, and future (Felix Quesada, personal communication).The crucial point is that the movement-derived first singular and pluralexclusive can have readings with prediction and intention but not readingswith obligation and necessity. Only the third person that is, the formderived from an obligation source can obligation or necessity be presentas modal flavors (Bybee and Pagliuca, 1987, p. 118).

    The nature of the argumentation in the above passage is not clear to me. Giventhe (reasonable) assumption that we are not dealing with accidental homophony,we find a first person future that shares a common source with movement and athird person future that shares a common source with obligation. In the absenceof further relevant information bearing on the directionality of developments, wehave no more reason to think that the future senses derived from the movementand obligation senses than vice-versa, or that (quite reasonably, in my opinion)in one person the future is historically basic and in the other historicallyderivative. No principles of language change of which I am aware dictate that allforms in a suppletive paradigm are necessarily semantically innovative.

    Armed with the assumption of common origins of future morphemes in a smallset of lexical sources, Bybee et al. in their 1991 paper propose the following set ofdevelopments for those that develop from a modality sense (p. 29):

    (33) a. STAGE 1. The youngest future morphemes (i.e. the most recent todevelop) have a sense of obligation, desire, or ability.

    b. STAGE 2. The next oldest future morphemes have a sense of intentionor root possibility.

    c. STAGE 3. The next oldest future morphemes have a future sense only.

    F.J. Newmeyer / Language Sciences 23 (2001) 187229 219

  • d. STAGE 4. The very oldest future morphemes have a probability senseor are used as imperatives10.

    In other words, they propose a model instantiating Hypothesis (1). But what isthe evidence for this model? Based on a survey of 75 languages, they show thatthe stages correlate with the future morphemes degree of reduction of form, asmeasured in terms of degree of fusion with the verb, morphophonologicalautonomy, and length. Given the assumption that semantic change isaccompanied by formal reduction, they conclude