Jump to ContentJump to Main Navigation
Meaning Change in GrammaticalizationAn Enquiry into Semantic Reanalysis$

Regine Eckardt

Print publication date: 2006

Print ISBN-13: 9780199262601

Published to Oxford Scholarship Online: September 2007

DOI: 10.1093/acprof:oso/9780199262601.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 17 September 2019

Meaning Change under Reanalysis: Previous Views

Meaning Change under Reanalysis: Previous Views

(p.22) 2 Meaning Change under Reanalysis: Previous Views
Meaning Change in Grammaticalization

Regine Eckardt (Contributor Webpage)

Oxford University Press

Abstract and Keywords

This chapter delineates the issues and questions in diachronic linguistics that are addressed in the book. It's purpose is twofold: it introduces the current debate and criticizes earlier theories of meaning change in grammaticalization, specifically the metaphor approach, the metonymy approach, and bleaching. Finally, this chapter points out that semantic reanalysis is a process visible, but not limited, to grammaticalization phenomena. The emergence of discourse particles offers another domain where semantic reanalysis can be observed.

Keywords:   metaphor, metonymy, bleaching, discourse particles, language change

2.1 Reanalysis versus Grammaticalization

Meaning change under reanalysis is closely tied to grammaticalization, another mode of language change that has received much attention in recent years. After a meticulous survey of definitions of ‘grammaticalization’ that scholars in the field advanced over the last century, Lyle Campbell and Richard Janda (2001: 114) summarize ‘the prototypic (or core) definition most familiar today: some linguistic element > more grammatical’. While even a brief survey of sample cases shows that reanalysis and grammaticalization are not identical concepts, the notions show large overlap. Hopper and Traugott (1993: 32) state that ‘unquestionably, reanalysis is the most important mechanism for grammaticalization, as for all change’, and generally, reanalysis is viewed as a major process in grammaticalization.

The reason why any investigation of meaning change under reanalysis will have to take the literature on grammaticalization as one of its starting points is of a practical nature: reanalysis is virtually always, in practice, investigated as it operates in grammaticalization. While the process was early on perceived as a mode of change in its own right (Langacker 1977), a position that remained unchallenged over the last decades,1 it inspired scholarly interest mostly in the context of grammaticalization. Consequently, many proposals as to the possible modes of meaning change under reanalysis were made with respect to the more limited field of grammaticalization data.

I will therefore take the time to summarize some of the main claims, assumptions, and keywords in grammaticalization and point the reader towards more comprehensive overviews of the field in its current state. Not (p.23) only do the relevant keywords occur over and over again in the referenced literature. The perspective advanced in this book can also shed new light on some issues that are hotly debated in grammaticalization literature. The chapter is organized as follows: After a brief overview of grammaticalization in the present section, sections 2.2 to 2.4 will be devoted to three major approaches to ‘meaning change in grammaticalization’ (which, remember, overlaps but is not necessarily coextensional with ‘meaning change under reanalysis’ as it is investigated in the present monograph). These are the bleaching approach, the metaphor based approach, and the metonymy based approach. Of course, this categorical division of the field into three separate branches is an oversimplification, and areas of overlap and contact will be pointed out in passing. Section 2.4 in particular will address the interrelations between the concepts metaphor, metonymy, and historical pragmatics. Section 2.5 will take a closer look into the development of textual and discourse markers and spell out in what respects these pattern well with grammaticalization and where they do not. In section 2.6 I will review some recent proposals about the nature of the contexts of change. Section 2.7 finally will spell out in some more detail how far meaning change under reanalysis and meaning change in grammaticalization are orthogonal to each other, and I will point out some possible consequences.

The term ‘grammaticalization’ goes back to Antoine Meillet (1912) while the phenomenon was studied long before by Wilhelm von Humboldt, Georg von der Gabelentz, and even earlier by the French scholar Étienne Bonuot de Condillac, or the English John Horne Tooke.2 Detailed accounts of the development of the field over the last three centuries can be found in Lehmann (1982: ch. 1) as well as Hopper and Traugott (1993: ch. 2), leading the reader past a line of authors who laid the grounds for current research.

Campbell and Janda (2001: s. 2) offer the most detailed summary of definitions for ‘grammaticalization’ since Meillet's famous starting point. They characterize the topic of investigation in grammaticalization research as developments that lead from some linguistic element to something more grammatical. The nature of this trend becomes immediately clear once we are presented with typical examples like the following: English will ‘want’ developed an auxiliary use will ‘future’ in which it has a grammatically more specific, more bounded function and has also suffered from phonetic erosion (I'll, you'll, etc.) Germanic thaz/that as a deictic element developed a complementizer use (daβ, that) in which it serves to mark a subordinate clause. In Latin, the collocation verb + habere turned into an analytic future construction. (p.24) Phonetic reduction in French led to a synthetic future tense paradigm that integrates the descendant of a lexical verb as a suffix into the inflected verb, thereby turning syntax into morphology. And so on.

What are the factors that tie all these cases together? In his classic Thoughts on Grammaticalization, Christian Lehmann offered a detailed proposal to define the field of grammaticalization and posited some prominent generalizations (Lehmann 1982 [1995]). Later research has mostly taken Lehmann's picture more or less as a starting point, and I will therefore give a summary of his six ‘parameters of grammaticalization’. Lehmann's book was followed by the rich and differentiated monograph by Paul Hopper and Elizabeth C. Traugott (Hopper and Traugott 1993). The overview of Fischer, Rosenbach, and Stein (2000) offers a more recent update which, basically taking over the large picture drawn in the two earlier books, adds some refinements and reports the state of discussion with respect to more controversial issues that have come up in the last years. Heine and Kuteva's (2002) world lexicon of grammaticalization comprises an impressive range of case studies and grammaticalization trends. The papers in Campbell and Janda (2001) address the soundness of the terminological underpinnings of the field in a particularly well-balanced and illuminating way, carefully teasing apart theoretical assumptions and empirical findings and disentangling some seemingly controversial issues. Traugott and Dasher (2002) finally also endorse an overview of the field while the authors, being concerned with ‘regularity in semantic change’, in fact address a broader range of phenomena.

Grammaticalization, to put it intuitively, takes place if an item starts to get used in a way that is more bound, more functional, less independent, less contentful. From a certain point on, one will want to postulate a new item (with new grammatical properties, new meaning) that has developed from the older one. We have then reached a stage of polysemy (or layering in terms of Hopper 1991: 22). The older item can briefly be dubbed as ‘more lexical’, the newer item as ‘more grammatical’.

Thinking in terms of grammatical categories, it was found that the various stages of items frequently pass through the same sequences of grammatical categories. For instance, the following pathway of subsequent stages was posited for grammaticalizations of verbs:


(Hopper and Traugott 1993: 108). Such pathways are also called clines. While the development of some actual item may not pass the full cline, or leave out certain stages, these clines do reflect universal tendencies: if an item grammaticalizes into a new item, the new item is commonly located to the right of the (p.25) older item on the respective cline. (See Hopper and Traugott 1993: chs. 1.1.2 and 5.)

Can we characterize the tendency from ‘less grammatical’ to ‘more grammatical’ in general, beyond a mere listing of attested clines? Hopper and Traugott (1993) propose the following general scheme of grammaticalization clines, pointing out that the spellout of any actual cline has to remain a case-by-case task.



(Hopper and Traugott 1993: 8)

Lehmann (1982 [1995]: ch. 4) however explicates the trend from ‘less grammatical’ to ‘more grammatical’ in terms of more specific, more graspable properties of linguistic items. He offers a system of three parameters of grammaticalization, each being realized in two dimensions, a syntagmatic and a paradigmatic dimension. The following table of criteria shown in Table 1 emerge (Lehmann 1982 [1995]: T4):

Grammaticalization, according to Lehmann, is characterized by an increase in cohesion along with a decrease in weight and variability from older item to newer item. The system is to be read as a cluster of correlated features rather than a list of necessary and sufficient criteria for grammaticalization. This means that we may encounter cases of grammaticalization that show most, but not all of the listed tendencies. Let me briefly spell out the keywords given in the table, while I refer the reader to the original work for a fuller discussion of the six criteria.

The paradigmatic weight of a sign, or its integrity, measures its distinctness and independence of other signs in terms of both phonology and semantics. Hence both phonological reductions and semantic losses constitute a loss in integrity. The paradigmaticity of a sign reflects the degree to which it functions as part of a paradigm of signs of complementary distribution in certain contexts. Grammaticalization frequently involves a trend for an item to turn

Table 1





structural scope





paradigmatic variability

syntagmatic variability

(p.26) into part of a paradigm of fixed semantic and structural function. Paradigmatic variability, finally, concerns the question whether an item can be freely replaced by other signs of the same paradigm, or be left out altogether. A loss in paradigmatic variability means that the item becomes obligatory in certain contexts.

The syntagmatic weight of a sign, according to Lehmann, is its structural scope. This notion is spelled out as either semantic scope or syntactic scope in different examples; Tabor and Traugott (1998) attempt to develop a formal version of this criterion. Syntagmatic bondedness measures the degree to which an item is dependent on the presence of other signs, or attaches to them in a morphophonologically significant manner. Syntagmatic variability, finally, reflects the degree to which an item has to hold a fixed position or shows freer word order.

Lehmann illustrates each of these criteria with traditional instances of grammaticalization and demonstrates that, even though many cases lack some of these criteria, every one of them reflects an important intuition that has led scholars to class one or the other case of language change as grammaticalization. More importantly, the system was developed against a background of a large number of both case studies and less comprehensive typologies of grammaticalization and to date is viewed as the most elaborated catalogue of criteria available. While Lehmann's two-dimensional parameter system captures all positive instances of grammaticalization, later sections on discourse markers (see section 2.6) allow us to get a feeling for its limits.

Let me come back to the above observation that grammaticalization clines reflect universal trends of language change, even if they do not reflect compulsory ‘road maps’ of change. A vast number of case studies have confirmed over and again that diachronic developments very frequently lead from the more lexical to the more grammatical, from the less bounded to the more bounded, from the less functional to the more functional, or on the cline from left to right. This observation naturally leads to the claim that whenever an item develops a new reading that can be located on any such axis at all, it will necessarily lead from the lexical pole towards the grammatical pole (the unidirectionality hypothesis). The status of this finding is an issue of heated debate. While all scholars agree that it holds true as a statistical tendency, opinions differ as to whether this tendency should be viewed as a law of grammaticalization. Search for counterexamples has yielded an impressive range of changes that contradict the unidirectionality hypothesis (see the surveys in Campbell 2001: 127 f. and Janda 2001: 289 ff. who lists more than seventy critical examples) and which have elicited mixed responses in the literature. In particular, authors tend to acknowledge the existence of counterexamples but go (p.27) on to use ‘grammaticalizaion’, in word or in spirit, as if unidirectionality were nevertheless a 100 percent reliable feature. Campbell (2001) shows carefully how the net result of this strategy—which he diagnoses for several authors in the field, for instance Lehmann, Hopper and Traugott, Heine, Haspelmath, Tabor and Traugott—turns unidirectionality into a defining criterion of ‘grammaticalization’ rather than an empirical generalization.

At what point does reanalysis enter into the grammaticalization literature? Generally, it is acknowledged as one of the major processes in grammaticalization (see Hopper and Traugott 1993: ch. 3, Harris and Campbell 1995, Fischer, Rosenbach, and Stein 2000, Campbell 2001, among others3). Traugott and Dasher write, ‘for most of this century, reanalysis has been considered the major factor in morphosyntactic change’ (2002: 27), thus echoing the earlier statement in Hopper and Traugott (1993: 50): ‘It is best, then, to regard grammaticalization as a subset of changes involved in reanalysis, rather than to identify the two.’ This more specific view on the relation between the two processes leads to three caveats that should be kept in mind when evaluating the literature on grammaticalization with respect to statements about ‘meaning change under reanalysis’:

  1. (a) Not all reanalysis in grammaticalization is necessarily of a kind that is of interest for us. Consider, for instance, the morphosyntactic reanalysis in Early French future constructions that turned the auxiliary habere into an inflectional suffix of the verb -ai, -ais, -a, etc. In spite of the structural reanalysis of the sequence VERB + AUX into VERB + SUFFIX, the meaning of the construction remained unchanged. We will not be concerned here with reanalysis that only affects morphosyntactic form.

  2. (b) Not all changes in grammaticalization need to involve reanalysis. In spite of the general agreement that reanalysis is a major force in grammaticalization, other mechanisms have been proposed as well, for instance analogy/extension and borrowing as a further source of new grammatical items (Campbell 2001: 141).

  3. (c) Not all cases where language changes occur on the basis of reanalysis need be cases of grammaticalization. This point echoes Hopper and Traugott's position quoted above. Apart from isolated examples, discourse markers and their emergence offer rich material. Their problematic status with respect to grammaticalization will be taken up in section 2.5.

(p.28) Three modes of meaning change in grammaticalization were pointed out as particularly relevant: bleaching, metaphor, and metonymy. They will be discussed in more detail in the following three subsections.

Research in grammaticalization at its revival around 1970 was not committed to any particular theoretical paradigm in linguistics. As more and more individual case studies emerged and general tendencies and mechanisms suggested themselves, two major lines of combining diachronic and synchronic linguistics emerged. On one side, language change in general and grammaticalization in particular was adopted as one central argument in advancing conceptual approaches to grammar and meaning against the formal paradigms in syntax and semantics that are connected with the names of Noam Chomsky and Richard Montague. I will however largely ignore this challenging field in favour of a second line along which a synthesis has been proposed: In the recent decade, we have also seen an increasing interest in reconciling the results of grammaticalization theory with formal theories of linguistic research. Specifically, work by Gelderen (1993, 2004), Roberts (1993a, 1993b), Roberts and Roussou (2003), or Faarlund (1990) shows large-scale attempts to elucidate the syntactic side of grammaticalization in terms of formal syntactic theory in the Chomskian tradition. This work has, however, so far not been complemented by truth value based investigations in to meaning change.4

Attempts to capture the dynamics of language change with formal theories have received sceptical comments in the past. Fischer, Rosenbach, and Stein (2000: 13) lay out two competing positions under the headings of ‘formal’ versus ‘functional’ approaches and point out aspects that seem to prove the unsuitedness of formal approaches for gradual processes of change (see Fischer, Rosenbach, and Stein 2000: 13, table 4). Functional approaches are characterized as resting on a holistic conception of language and grammar, considering conceptual, semantic-pragmatic, and language external factors, perceiving diachrony in synchrony, as localizing language change mainly in language use, perceiving language change as gradual, as seeing grammaticalization as the full process from lexical items to grammatical words, including actuation (see below), implementation, and motivation, as offering a description of the whole process, and as looking for explanations inside and outside grammar. Formal approaches, according to Fischer et al., show the following characteristics: they have a modular conception of language and grammar, consider only language internal factors, oppose synchrony to diachrony, (p.29) which is viewed as a comparison of synchronic stages. Their subject matter of investigation is linguistic competence (as opposed to language use), and language change is localized in language acquisition. Grammaticalization is reduced to reanalysis, and seen as the evolution of functional heads out of lexical heads. Only the situation before and after reanalysis is described, and finally, only language internal causes are taken into consideration in explaining language change.

The present view, even though it might at first sight be classed under Fischer et al.'s header ‘formal approaches’, demonstrates that approaches to grammaticalization do not always follow the neat two-way classification suggested by these authors.5 In particular, the present study shares several important features with their ‘functional’ approaches: it aims to tie the onset contexts of reanalysis to pragmatic and language external factors and to explain how these can instigate reanalysis. Consequently, the role of language acquisition in language change is downtuned in comparison to communicative experiences of the adult speaker. Explicating the very process of reanalysis, we will describe the process of change as well as pre-state and post-state, and explanations for change are sought in the communicative situation rather than ‘inside grammar’. Fischer et al. attribute all these features to the so-called ‘functional approaches’. Nevertheless, my approach likewise shares several important features with the ‘formal approaches’. Most importantly, I will make the assumption that a synchronic language stage can be characterized in a reliable and precise way and that the precise description of language stages before and after change is an indispensable precondition for any analysis of language change. It is in line with this assumption that I will concentrate on reanalysis as a mode of language change, another aspect that Fischer et al. would attribute to ‘formal’ approaches.

This defence of a formal approach to meaning change against the atmospheric criticisms that are implicit in two-way classifications like the one by Fischer et al. must necessarily remain vague until brought to life by the contents of later chapters. Eventually I will have to leave it to the reader to decide where the present work should most appropriately be located on the landscape of investigation into language change.

Theoretical characterizations of a field of research are one thing—the range of actual research conducted in the field may sometimes draw a slightly different picture. The following collections offer a good overview of the developments in the field over the last years: Heine, Claudi, and Hünnemeyer (1991), Traugott (p.30) and Heine (1991), Pagliuca (1994), Harris and Campbell (1995), Giacalone Ramat and Hopper (1998), Fischer, Rosenbach, and Stein (2000), Campbell and Janda (2001), Wischer and Diewald (2002), Heine and Kuteva (2002), Lang and Neumann-Holzschuh (1999), and Batllori et al. (2005). The state of the field in the early 1990s is moreover mirrored in the Dictionary of Grammaticalization by Donald A. Lessau (Lessau 1994).

2.2 Meaning Change by ‘Bleaching’

The earliest attempt to characterize meaning change in grammaticalization is based on the metaphor of meaning ‘fading away’, ‘weakening’, or ‘bleaching’ in grammaticalization. This reflects the intuition that the meanings of content words are usually more concrete, more graspable, more precise, more often linked to real things, properties, or activities than the meanings of the respective derived function words. The following passage of Georg von der Gabelentz is frequently quoted as an early manifestation of the bleaching theory (Gabelentz 1901: 241):

Was erst neu und selten war, wird dann alltäglich und damit verliert es an Kraft, verblasst, rückt schliesslich wohl gar in die Reihe jener abstracten Bestandtheile der Rede, die es hatte verbessernd und verstärkend ergänzen sollen … Was von den Formwörtern gilt, das gilt … auch von den Wortformen. Wo deren neue geschaffen wurden, da waren sie periphrastisch …, frischere neue Farben deckten die verblichenen alten.6

About twenty years later, Antoine Meillet (1912) wrote, ‘Laffaiblissement du sens et l'affaiblissement de la forme des mots accessoires vont de par; quand l'un et l'autre sont assez avancés, le mot accessoire peut finir par ne plus être qu'un élément privé de sens propre.'7 Note that the bleaching metaphor is particularly convincing (but also slippery) because it can be read in two different ways, and seems to reflect a fact about language change under both interpretations. Under one interpretation, it concerns the stylistic value of expressions. A newly coined expression will still carry the flavour of originality and strikingness. As soon as it is used repeatedly and imitated by (p.31) more speakers in more contexts, this originality and strikingness will of course get lost, or ‘fade away’, and leave the expression as a neutral part of language. I suspect that at least von der Gabelentz's remark was primarily meant in this sense. According to this view, bleaching would take place when a formerly new grammatical construction gets used more and more and becomes a regular part of language. The order of events is hence: first change (= innovation), followed by later fading which in turn invites further innovation.

According to another interpretation of the bleaching metaphor, however, bleaching takes place at the very moment when a new expression or grammatical construction is coined. Assume that an older content word undergoes grammaticalization. In terms of syntax, it becomes more bound, more functional, more obligatory—in terms of semantics, the older, more concrete meaning bleaches to yield the newer, more abstract meaning. The order of events here is different: change is fading. Taking up current practice, I will only address this second notion of bleaching. At the end of this section we will, however, remain with the result that the notion of ‘bleaching’ is surprisingly tenacious even though it was often criticized as empirically inadequate and theoretically ungraspable. The two-facedness of the bleaching metaphor may be one reason for its lasting appeal.

Can bleaching (henceforth in the second sense) be spelled out more explicitly? The first and simplest idea might be to view bleaching as meaning generalization. The less restrictive the contents of a word, the larger will become its range of application. Different uses of the German heben (full verb)/haben (cognate auxiliary) that range from grasping over concrete to abstract possession can serve as an illustration.



Er hebt einen Stock

‘he has/holds a stick’

(South Germ.)


Er hat ein Auto

‘he has/owns a car’


Er hat einen Schnupfen

‘he has a cold’


Er hat Recht

‘he is right’


Er hat gelacht

‘he has laughed’

This view can at least be traced back to Hermann Paul (1880) who described cases of grammaticalization in this spirit, characterizing the concurrent meaning change as generalization. Lessau (1994) attributes the introduction of the modern term bleaching to Heine, Claudi, and Hünnemeyer (1991: 40) while the notion itself was already discussed in Lehmann (1982 [1995]: 127).

Many instances of grammaticalization, however, do not exhibit meaning generalization. Remember the say-do future tenses that were discussed in Chapter 1: we would hesitate to claim that the concept of SAYING generalizes (p.32) to DO-IN-FUTURE. Clearly, ‘saying’ is not just a more specific variant of ‘intending to do in the future’ because a person can say a lot of things without any intentions for future action.

A similar case was already made in Traugott (1988) and Sweetser (1988) who both point out that an item under grammaticalization may gain meaning (= become more specific in certain respects) as well as lose meaning (= become less specific in other respects). Gains in meaning have since been acknowledged as an integral part of meaning change in grammaticalization under the label of pragmatic strengthening (e.g. Traugott and König 1991: 194, corresponding to the I-principle, Levinson 2000, and also section 2.4 below). Among the cases discussed in the present volume, the history of selbst offers another striking example for grammaticalization/reanalysis without semantic generalization.

A slightly different view of bleaching is endorsed in the work of E. Sapir (1921: ch. V) and was taken up later by Zirmunskij (1966: 83, from Lehmann 1982[1995]: 127 ff., 156). They propose that grammaticalization leads from lexically meaningful, presentational absolute meanings to more abstract, relational meanings. Lehmann critically observes that the relationality of a word usually remains unchanged during grammaticalization. Words that denote concrete relations develop into words for abstract relations without change in arity.

Thinking about changes in relationality during grammaticalization, we may note that neither of two possible extreme positions seems to reflect the full truth. Neither does a word at different stages of the grammaticalization cline generally amass more and more argument places that need to be filled (i.e. get more and more relational), nor does the argumental structure of an item generally remain unaffected during grammaticalization, as Lehmann seems to suggest. Consider once again the emergence of the have + participle construction. Evidently, have must undergo reorganization in argument structure in order to develop from the simple binary possesion verb to part of the complex tense construction. If we follow Musan (2001), have AUX is analysed as a functor that takes a verbal predicate and turns it into a more complex predicate that includes aspectual information. Hence, the word turns from a binary relation into a functor, which means a considerable reorganization.

It is undeniable that the bleaching metaphor, the tendency from the more concrete to more abstract meaning, offers an adequate descriptive characterization of meaning change under reanalysis. Yet it seems highly problematic to turn the descriptive term into a label for ‘meaning change under grammaticalization’ (or ‘meaning change under reanalysis’). The hypothetical postulate that (p.33) ‘meaning change in grammaticalization always and only occurs by bleaching’ will leave several questions unanswered.

First, it remains open whether this mysterious process of semantic bleaching can ever occur elsewhere in meaning change. We have already argued that bleaching is not the same as semantic generalization. In the next section, we will review metaphorization as another mode of meaning change that commonly leads from more concrete to more abstract meanings. Even though metaphorization would fit nicely with the descriptive term ‘bleaching’ in that respect, it will become clear that not all meaning change under reanalysis occurs by metaphoric shift. Under the rigid view of bleaching endorsed in the above postulate, hence, bleaching cannot be the same as metaphorization. This will lead to the consequence that the mysterious bleaching processes can always and only be observed as a semantic change in grammaticalization.

This consequence, however, conflicts severely with current views on grammaticalization: there is strong evidence in favour of the assumption that grammaticalization is an epiphenomenon of interactions of phonological, morphological, and syntactic changes that are generally operant in language change. Arguments in favour of this view—as opposed to regarding grammaticalization as an independent mode of language change that cannot be decomposed—have been advanced by Campbell (2001), Joseph (2001), and Newmeyer (2001). According to this ‘conspiracy’ picture, languages come with an inventory of modes of sound change, morphological change, syntactic change, and—presumably—semantic change. An instance of language change is called ‘grammaticalization’ just if all the interacting changes on all levels produce a case of the appropriate phenotype. Against this background it would be highly surprising to claim that at the level of meaning alone, grammaticalization required a special process of ‘bleaching’ that was operant nowhere else.

Second, even if we grant that meanings get ‘paler’ in grammaticalization (or reanalysis), the bleaching approach still remains tacit as to what actually drives the outcome of the fading process. Given that we know competing approaches that can successfully address this question and offer meaningful answers (to which we will turn in section 2.4), this must be counted as a drawback of the bleaching approach.

Finally, the mystery remains why meanings always seem to get paler in grammaticalization. There is certainly no general trend for meanings to get less specific or more abstract. We find meaning specializations, as in E steorfan ‘die’ > starve (‘die from lack of food’), in G fasz (‘container’) > Faβ (‘barrel’), or in G Kleid (‘clothing’) > Kleid (‘woman's dress’). We also find concretizations, as in G Bestellung (‘order’, as an activity) > Bestellung (‘the thing ordered’), or in Germ. Thing (‘lawsuit’) > E thing, G Ding. To put it (p.34) drastically: ‘Bleaching’ in meaning change is not as inevitable as ‘bleaching’ if you think about stockings in your washing machine. This fact is usually obscured by the force of the bleaching metaphor which seems to mirror some kind of intuition about grammaticalization. At the end of the book, I will come back to this puzzle. After evaluating the case studies in the core part, I will offer a new attempt at explaining in what sense the denotations that emerge in grammaticalization are substantially different from those that are attributed to words in other modes of establishing semantic conventions.

In spite of the dubious theoretical status of the term, bleaching is still used as a pervasive descriptive label for meaning change in grammaticalization. This is witnessed by recent publications inside and in the periphery of the field, e.g. Haspelmath (1998: 318): ‘Grammaticalization comprises … the development of function words from content words, the development of affixes from function words, as well as a large number of concomitant changes … (e.g. desemanticization, …)’, Musan (2001: 370): ‘The past participle hypothesis, however, only has to assume that haben and sein lost at least part of their original meaning by semantic bleaching, which is a general characteristic of grammaticalization in any case,’ or the article ‘How far does semantic bleaching go?’ by Werner Abraham in Faarlund (2001) (emphasis mine).

2.3 Meaning Change by Metaphor

Meaning changes in grammaticalization share core properties with meaning shift by metaphor. A prosperous branch of research on grammaticalization, initiated by B. Heine and his colleagues, argue that meaning change in grammaticalization is metaphorical meaning shift (Heine 1993, 1997a, 1997b, Heine, Claudi, and Hünnemeyer 1991, see also Bybee, Perkins, and Pagliuca 1994, Sweetser 1988, 1990, Stolz 1994 for other proponents of this view). Heine observes that the same small range of content words tend to give rise to the same kinds of functional words in unrelated languages all over the world. Similar grammaticalization patterns arise independently at different times and places, and conversely, the grammatical constructions of foreign languages, even though different from those in our mother language, usually have a motivated flavour about them once one has understood their etymological origin.8 Heine explains this by the fact that metaphors are based on universal human cognitive schemes and therefore universally accessible.

(p.35) The metaphor account looks plausible in many cases. Take the French venir de faire (‘coming from doing’) immediate past tense as an example:

  1. (2.2) Je viens de finir l'essai sur grammaticalisation.

    I come from finishinf the paper on grammaticalization

    ‘I have just finished the paper on grammaticalization.’

Stolz (1994) proposes that this tense form involves a metaphoric shift from movement in space to movement in time. It might be interesting to note that Sweetser (1988) exploits the same metaphor to account for the English going to future. The mapping between ‘movement in space’ and ‘movement in time’ can, according to these proposals, be explored in various different ways, a finding that coheres well with the general observation that metaphors are commonly based on structural similarities between different ontological domains. These similarities can be exploited by more than one instance of metaphoric language use (Lakoff and Johnson 1980).

In spite of its plausibility, the approach does not automatically answer all questions. A first important observation concerns the productivity of metaphor. Traditionally, metaphor has been viewed as a typical instance of creative language use. Metaphors are typically found in literary texts, particularly poetic texts, and are taken as an indication of the creativity of the author. Metaphors are a rhetorical device to present a given content in a striking, impressive, colourful way. Metaphors have also been found useful in scientific research, suggesting new conceptualizations of a given domain that may lead to new kinds of theories. Metaphors can finally be exploited in problem solving and engineering and lead to new solutions to a given technical problem (for a broad overview see Indurkhya 1992, Gentner 1983). Importantly, all these domains require the volitional, intentional use of metaphor as a mode of non-literal speaking.

New grammatical constructions, in contrast, are never coined by volitional use of metaphor. Let me come back to the French venir de faire past in order to illustrate what this would amount to. Given that an English or German linguist can understand the construction as metaphor based, we would expect that they could in principle exploit the same metaphor in their own mother tongue. Yet, the following utterance is highly marked:

  1. (2.3) (Jo has been sitting at her desk since 8 a.m. At noon, her colleague enters the office. Jo looks up and says:)

    I am coming from finishing the paper on grammaticalization.

    I come from finishing the paper on grammaticalization.

(p.36) Such utterances would clearly not be understood (let alone praised) as creative use of language. The hearer might understand what Jo intends to say, but not on the basis of knowledge about movement in space and movement in time but at best on the basis of knowledge about some motivated immediate past tense in French that Jo evidently tries to calque into English. Likewise, such a scenario does not ‘feel’ like the initial point of grammaticalization under way.9

In answer to this observation, it was suggested that we need to distinguish between volitional and subconscious metaphorical shift (creative and emerging metaphor in terms of Heine, Claudi, and Hünnemeyer 1991: 60–2). Metaphor in grammaticalization would then be subconscious metaphor, while in literary or scientific discourse, we meet volitional metaphors. This distinction, however, leads to an account of meaning change in reanalysis that is again descriptive rather than explanatory. Somewhat oversimplifying, we end with the following picture.

Meaning change in grammaticalization is (sometimes, always) meaning shift by subconscious metaphor. While we know a lot about the licensing conditions for volitional metaphors, we do not know equally much about subconscious metaphor. In particular, we do not know when speakers can use subconscious metaphors. In constrast to classical metaphors, a mere similarity between some source and target domain is not sufficient to license emergent metaphor (see (2.3)).

Another drawback of the metaphor based approach to grammaticalization is that it will certainly require additional mechanisms of meaning change, in particular something like pragmatic strengthening, in order to account for the data (as acknowledged in Heine 1997a, or much earlier Sweetser 1988). Again, we can take the venir de faire past to illustrate the point. If we consider the purely spatial use of the verb venir (‘come’), it can easily be seen that it is not restricted (and never was) to short distance movements. We can use venir equally well to report that we ‘are coming from the bakery’ right now as to report that we ‘are coming from Texas’ when arriving at New York airport or even to inform the hearer about our place of birth (je viens de Bordeaux in the sense of ‘I was born in Bordeaux’). Hence, the use of venir de … in the source domain lacks the sense of immediacy that it carries in the target domain.

(p.37) One might object that the meanings of metaphoric uses of words are always coloured by facts about the target domain and in that sense can be more specific than the source domain would suggest. However, the venir de faire past tense again tells us that this cannot be the full truth. ‘Immediacy’ is not a given fact about events in the past, as little as it would be a fact about distances between locations. The metaphor ‘movement in space = movement in time’ alone should perfectly well license a use of venir de faire like that in (2.4):

  1. (2.4) Je viens de faire mon Ph.D. à Stuttgart (en 1996).

    intended sense: ‘I did my Ph.D. at Stuttgart (in 1996)’, or more poetically ‘In the journey of my life, I have also passed a station that was doing my Ph.D. at Stuttgart seven years ago’ (see Lakoff and Johnson 1980: Life is a journey).

Yet, the meaning of the grammaticalized venir de faire simply does not allow this use (i.e. sentence (2.4) is not acceptable French if I utter it today, in 2006). This illustrates the effects of pragmatic strengthening.

A final objection against the metaphor based approach to meaning change under reanalysis is that we do find a considerable number of instances of grammaticalization that are not plausibly metaphoric.10 Once more, the case of the Eastern Bantu say do future can serve to illustrate this claim: even though there is a clear practical link between announcing something and doing something in the future, this link is not a metaphorical one. Another striking example is posed by the development of the French complex ne pas negation which goes back to the transparent Latin non passum (= ‘no step’). This case will be taken up in Chapter 5, but even now we will be inclined to say that whatever the link from ‘step’ to negation may be, it is presumably not a metaphoric link.

Surveys on grammaticalization commonly concentrate on motivated grammatical constructions, or even restrict attention exclusively to these (see e.g. Goossens 1989, Stolz 1994 on grammatical constructions that are metaphorically related to words for body parts). However, even paradigms with a majority of elements that might be explainable by subconscious metaphor can contain other elements for which the origin, whatever it may (p.38) be, does not look metaphoric. In a series of papers, Butt and Geuder address the light verb paradigm in Urdu (Butt and Geuder 2001, 2003). They report the surprising Urdu light verbs dee (‘give’) to contribute ‘forcefulness’ and lee (‘take’) to contribute ‘ability, habituality’ alongside the more metaphor-like maar (‘hit’) for ‘doing something energetically’, gaa (‘fall’) for ‘suddenly, involitionally’, or jaa (‘go’) for inceptive aspect.

In conclusion, an account of meaning change in grammaticalization by metaphor does not cover the field in a satisfactory manner, even though it cannot be denied that many grammaticalizations carry a flavour of metaphor. Most importantly, the metaphor based approach has little to say about the additional enhancing factors that are required to license subconscious metaphor. It is moreover not easy to reconcile the role of pragmatic strengthening with metaphor in an integral fashion.11 Finally, we do know of a considerable number of cases of grammaticalization which cannot be described as metaphor, however widely conceived. Importantly, many of these constructions are still motivated, which proves that motivated grammatical constructions can arise by cognitive processes other than metaphor.

In the next section, I will discuss the proposal that meaning change under reanalysis occurs by metonymy. I will treat it as a separate mode of meaning change, although I am aware of proposals that metaphor and metonymy are not incommensurably distinct but should rather be viewed as opposing poles on a continuum (see e.g. Barcelona 2000, Dirven and Pörings 2002, and earlier Goossens 1989). Following their view, the question about the nature of meaning change in reanalysis might have a quite simple answer: we could observe that all wittnessed changes are located somewhere on this prospective continuum, sometimes leaning more towards the metaphoric pole, sometimes more towards the metonymic. I think that this might be a possible answer but one which would not tell us very much. The proposal will be taken up at the end of the following subsection where I point out some of its merits and drawbacks in more detail.

2.4 Metonymy and Historical Pragmatics

The approaches to meaning change in grammaticalization (and reanalysis) that will be reviewed in the present section are all similar in that they seek the causes of language change in actual communication. In order to appreciate (p.39) this aspect, it might be worth revisiting the metaphor approach for a moment: metaphors crucially rest on facts about the world, in particular an ontological source domain and a target domain, and the way in which the former can be mapped onto the latter. If such a constellation is given (and appreciated as salient in a given culture), any speaker can at any time decide to coin a metaphor which exploits this ontological similarity. Metaphors need not be licensed by previous discourse.

Pragmatic accounts of grammaticalization, in contrast, share the view that grammaticalization requires a preparatory phase, a characteristic type of previous uses of an item, which is a necessary prerequisite for grammaticalization to come under way. An early proponent is Gustaf Stern (Stern 1931: ch. XIII). Stern singles out a class of meaning changes which he calls ‘permutation’ and which he characterizes as follows:

(1) A word is used in a phrase where a notion in some way connected with its meaning is liable to form an element of the context. (2) By frequent use, the associated notion is associated also to the word. (3) The associated notion takes the place of the original meaning, in phrases of the type mentioned. (4) The word is used in the new, secondary, meaning, in phrases of other kinds, where the primary meaning is not possible. (Stern 1931: 353)

Stern was mainly concerned with meaning shifts that leave the grammatical category of an item unchanged. Changes like the famous beads ‘prayers’ > ‘pearls’ or fast ‘immovable’ > ‘at high speed’ are at the core of Stern's interest. Yet he points out the crucial dependency of these changes on earlier utterance contexts and situations of use: the connection between ‘prayers’ and ‘pearls’ occurs only in limited utterance contexts where one might talk about beads. These utterance situations give rise to reinterpretation. And, most interestingly, Stern assumed that the reinterpretation was from sentence to word rather than an immediate reinterpretation of the word beads or fast (see Stern 1931: 352; also Stern 1921). I take Stern as a first proponent of ‘semantic reanalysis’.

More recently, Elizabeth C. Traugott in cooperation with König, Hopper, Schwenter, Dasher, Tabor, and other colleagues proposed a theory of context based meaning change in pragmatic terms. An early programmatic defence of this kind of approach, as opposed to metaphor and bleaching, was given in Traugott and König (1991), and the position defined in that paper lies at the core of a wealth of other work by Traugott and her co-authors, e.g. Traugott, (1988), Hopper and Traugott (1993), Tabor and Traugott (1998), Schwenter and Traugott (2000), Traugott and Dasher (2002), but her views have also been adopted by many other researchers in grammaticalization (see Detges (p.40) 1999, Waltereit 1999). The success of the theory is moreover witnessed by the fact that it led to the foundation of the Journal of Historical Pragmatics.

The picture drawn by Traugott is strikingly in line with Stern's views. The following five-step process reflects the current version of the theory, the Invited Inferencing Theory of Semantic Change:

  1. 1. In a first stage, an item L possesses a coded meaning M1.

  2. 2. In concrete utterance situations, this item L can be used in sentences that give rise to certain pragmatic implicatures (called Invited Inferences, briefly IIN).

  3. 3. These inferences can be exploited innovatively in the associative stream of speech and are re-weighted.

  4. 4. These processes eventually lead to conventionalization of certain inferences for certain sentences that contain the item L. (These conventionalized inferences are also called Generalized Invited Inferences, or GIIN, by Traugott and Dasher.)

  5. 5. At the final stage II, the conventionalized invited inferences have given rise to a new coded meaning for item L which is then ambiguous between meaning M1 and (new) meaning M2.

(See Traugott and Dasher 2002: 38, fig. 1.3.) The later parts of the present book adopt this theory as the basic picture and propose to refine it in certain respects.

While Stern was mainly concerned with meaning change at the word level, meaning change in grammaticalization and reanalysis clearly require that the changes in question take place in the tension between sentence meaning and word meaning. Traugott demonstrates lucidly and with a wealth of examples that the crucial link is not one between two concepts but arises at the propositional level. A literal sentence meaning gives rise, by Gricean pragmatic inferencing, to further information (likewise propositions), and only via this detour will one or the other word in a sentence be ascribed a new meaning M2 by conventionalization (Hopper and Traugott 1993: ch. 4.2 explicate the links to pragmatics, which will also be taken up in Chapter 3). Stern likewise maintained a sentence-centred view of the process, an insight that may be somewhat obscured by the fact that he uses the term ‘phrase referent’ for what one would today call ‘sentence denotation’ or ‘proposition denoted’.

The final step from stage (4) to (5) has not been extensively investigated so far. In some sense, it is just clear that speakers/hearers can form a hypothesis about a word's meaning once they see the word occurring in a sentence, know what the sentence means, and know the meanings of all other parts of the sentence. Still, even if we know that every speaker can perform this operation, (p.41) this will not make a description in theoretical terms superfluous. The process has been described as hypothetico-deductive reasoning in Itkonen (2002),12 however at a very abstract level and with only sketchy hints about detailed semantic representation.

The ability to perform semantic composition lies at the heart of the crucial step from (4) to (5). In recent years, the laws that govern semantic composition have been studied in formal semantics, and concise representations of denotations and semantic-pragmatic composition can successfully mirror core aspects of language use with great precision. The present study offers a first attempt to put these insights to work in historical semantics. Compositional semantics allows us to represent how speakers reconstruct the meaning of an item from an overall sentence meaning, showing an impressive analytic potential. The resulting meanings are used as confidently as if they had been looked up in grammar books. Compositionality is hence operant not only bottom-up—in order to code and decode the meaning of new sentences—but also top-down in order to determine the meaning contribution of isolated words to a known whole.

This view influences our expectations as to which speakers first adopt a new meaning as part of their linguistic system. Case studies as well as the emerging theoretical picture strongly suggest that the changes in question occur primarily in the mental lexicon of adult speakers. Adult speakers experience the crucial kind of pragmatically loaded communicative situations and adopt the relevant conventionalizations that lead to the final stage in (5). Note that this assumption does not conflict with the claim, frequently made in formal literature on syntactic change, that ‘reanalysis’ occurs between one generation of speakers and the next, i.e. is a process initiated in first language acquisition (see Lightfoot 1991, 1999, Andersen 1973). The instances of grammaticalization that we are interested in here are local. They concern the lexical entry of items, not the overall grammar. I propose that local shifts can occur in adult grammars, while global reorganizations most likely occur in first language acquisition. This view can be reconciled with formal theories on first language acquisition; Hopper and Traugott (1993) adopt a very similar position.

Let me once more come back to the five stages of the Invited Inference Theory of Semantic Change and focus on the details of stage (3). This stage refers to those contexts of use of an item that lay the ground for reinterpretation in (4)/(5). These contexts have been described as ‘downtuning’ some (p.42) aspects of the literal meaning of sentences, and ‘highlighting’ aspects of the inferred meaning (Traugott and Dasher 2002: 1.3.2, also Danchev and Kytö 1994). In other words, the item in question, from a certain point on, tends to be used more and more in sentences where the pragmatic inference is more salient, more relevant, much more the real information contribution to discourse than the literal meaning of the sentence.

It is important to remember, though, that these contexts so far were witnessed only in retrospect. The experience shared by all those who conduct an empirical study in grammaticalization is that the item in question occurs for some time in sentences of an ambivalent nature: they make sense both under the older and the newer use of the item, and we sometimes can only guess which sense was intended by the speaker. Eventually, first instances of the item in an unambiguously modern use show up in source texts—proof that something must have happened in the preceding phase. We may assume that at least the first ambiguous uses must have been produced and interpreted by speakers and hearers on the basis of the old mental lexicon. For these speakers, there was nothing ambiguous about the respective sentences. There must have been something special in certain contexts in addition to a potential ambiguity, something that invited interlocutors to change their lexicon in accordance with this ambiguity. I will refer to contexts of use that offered the right kind of structural and semantic ambiguity plus additional instigating factors as onset contexts of reanalysis.

Additional instigating factors that start reanalysis (or grammaticalization) have been proposed in the literature. The communicative intentions of the interlocutors are frequently named as the reason for change, and more specifically the wish to express information beyond what the means of literal language use would allow. Grammaticalization occurs, as Hopper and Traugott (1993) suggest, as a problem solving strategy of the speaker in order to improve expressivity of language. It is left somewhat unclear whether speakers achieve these improvements by appropriate ‘upgrading’ and ‘downtuning’ (which they in turn do how?) or whether upgrading and downtuning themselves are witnesses of earlier stepwise achievements in language improvement.

Moreover note that the expressive needs cannot be so pressing, given that in the onset contexts—that is, those contexts where those changes occur that only later lead to improved expressivity in other sentences—these needs are perfectly met by the implications of the respective sentence. If the speaker had not been able to convey what he or she wanted by way of an implication, the possibility for reanalysis would never have arisen. But this means that speakers in fact could express what they wanted. Hence, poor (p.43) expressivity cannot be the driving factor alone. We need to assume that speakers, somewhat meticulously, hold the expressive need to convey some piece of information as the literal content of a sentence.

In a related vein, Traugott points out subjectification as another driving force in semantic change. Speakers use (conventionalized) pragmatic inferencing to convey information about their inner conception of the world and inter-subjective relations. In using this mode of information transport, they can express more about their inner states than literal language use would allow. Traugott (1989) demonstrates that the emergence of epistemic uses of modals in English is an instance of subjectification; the emergence of discourse markers is another typical field that witnesses this general tendency.

It seems that further empirical research in language use in both the past and the present is needed to reveal the nature of onset contexts. Only such research can show whether needs such as the desire to express something literally rather than indirectly are actually a driving force in language change. We will come back to onset contexts in section 2.6 below.

Let me end this section with a brief excursion into terminological issues. Meaning change under grammaticalization, described in terms of pragmatic inferencing and conventionalization, is commonly classed as metonymic by its proponents (Hopper and Traugott 1993: ch. 4.3, Traugott and Dasher 2002). This view successfully highlights that pragmatic inferencing and subsequent reinterpretation is not a typical metaphoric shift, and the metaphor view on grammaticalization that was reported in the previous section is therefore implicitly refuted. However, I would like to point out that meaning change under grammaticalization is also different in several respects from typical metonymy.

Metonymic meaning shift, like metaphor, is commonly viewed as a transfer that rests on the facts of the world. Kövecses and Radden (1998) offer a general theory of metonymy which rests on this assumption. They develop a typology of metonymies in terms of idealized conceptual models (ICM, subjective reflections of facts in the world) and posit that metonymy can arise (a) between the whole ICM and its parts, and (b) between different parts of the same ICM. The authors proceed to a list of about thirty-five instantiations of these two modes. These instantiations confirm that the authors are thinking of ICMs as reflections of comparatively stable facts and relations in the world. They offer no example where an ICM would arise only in certain communicative situations.13 This picture—comprehensive as it is—will evidently (p.44) require substantial extensions to cover even simple examples of meaning change in grammaticalization.14 Most importantly, their view does not include the possibility that metonymic links only arise because speakers used certain constructions with certain communicative intentions.

If we look into productive uses of metonymy (see Nunberg 1995), we will find that speakers commonly exploit existing factual links between entities. The waitress who refers to the customer as the ham sandwich, the doctor who is going to see the ulcer, the mechanic who states that the red key is parked in the backyard all make use of existing relations between various objects in their everyday surroundings. The hearer will use pragmatic inferencing to conclude that the speaker cannot literally assert that a key is parked somewhere. Yet, no pragmatic inferencing whatsoever is necessary in order to establish the link between key and car. To put it bluntly, the link between key and car is a result of engineering, not a result of communication.

Finally, traditional examples of metonymy are commonly based on links between ontologically comparatively simple objects like humans, artefacts, events, or at most simple properties. The semantic objects that play a role in grammaticalization, in contrast, are frequently of a much more complex nature. In particular the resultant meanings sometimes only seem to become conceptually salient in the grammaticalization process. We can once more take the venir de faire past tense in French as a schematic example. The resultant meaning will have to be something like that in (2.5):

  1. (2.5) λP∃e (P(e) & e < < now)

(2.5) paraphrases as something like ‘give me an event description and I will state that an event of the respective kind took place, and moreover in the recent past’ (after Chapter 3, the reader will perhaps be better equipped to confront terms like this). It may be doubted whether a concept like this would even be accessible prior to first encounters in onset contexts. Once more, we see that the crucial target meaning (here: of venir de) does not stand in an independent metonymic relation to the source meaning (here: directed movement). It may be questioned whether this object λP∃e (P(e) & e < < now) would ever have gained cognitive salience, had not people started to use movement venir de constructions in sentences with certain temporal implications. The example hence highlights once more the difference between traditional metonymy and meaning change under reanalysis. Traditional metonymy builds on links between independently existent concepts. Reanalysis establishes new concepts as possible denotations.

(p.45) In sum, I adopt a three-way distinction between metaphor, metonymy, and semantic reanalysis/meaning change by pragmatic inferencing. This rigid three-way distinction differs from the more holistic picture favoured in Hopper and Traugott (1993: ch. 3) and Traugott and Dasher (2002: ch. 1.3) where meaning change under reanalysis may ‘involve’ metaphoric links as well as, more prominently, metonymic shifts. I propose that all that matters for meaning change under reanalysis is the relation between linguistic form (‘what does the respective sentence look like?’), implied proposition (‘what does the speaker intend to tell me?’), and the possibilities to derive the proposition from the sentence in a literal way—meaning shifts granted. According to this view, it is a pure accident whether the resulting meaning shifts carry the flavour of metaphoric change or not. It would be a question for empirical linguistic research whether metaphoric possibilities enhance meaning change under reanalysis. At present, we can only say that metaphoric links cannot be a necessary prerequisite for meaning change under reanalysis because—as we have seen—there are a considerable number of instances of grammaticalization that do not exploit metaphor. On the other hand, it is a matter of taste whether one wants to call the required coincindence of form and implication metonymic in nature or not.

This view remains unaffected by recent proposals towards a unified theory of metaphor and metonymy. Barcelona (2000) argues that many instances of figurative speech cannot clearly be classed as either metaphoric or metonymic and suggests a mixed mode of transfer instead. He offers the example of body sensations for emotional states. The equation ‘Anger is hot’ (Barcelona 2000 quoting Goossens 1989) for instance would normally be classed as an instance of metaphor, and yet it derives from the cause–effect link between negative emotional state and a rise in blood pressure, causing the subjective experience of heat. Cause–effect based transfers are commonly classed as metonymy. Hence the figure seems to be both metaphoric and metonymic. Generally, Barcelona states, metaphors involve contiguity of two domains as well as similarities between two domains, hence favouring a unified account rather than a strict dichotomy. Dirven and Pörings (2002) argue in the same direction, stating that metonymy and metaphor should be viewed as the poles of a continuum of possible meaning changes rather than incommensurable differents. In particular the technique of domain blending (Fauconnier 1994, 1997) seems to capture typical aspects of metaphor as well as metonymy. Nevertheless, one central difference between all these kinds of non-literal language use and reanalysis/grammaticalization remains unaffected: While metaphors, metonymies, and blending relate different independently mastered conceptual domains and exploit this link in terms of language, meaning change under reanalysis can lead to (p.46) denotations that had no previous conceptual status and only became a desirable denotatum in the course of the reanalysis in question.

I dare to speculate that the forefather of historical pragmatics, Gustaf Stern, would presumably appreciate the three-way distinction. Interestingly, he never uses the term ‘metonymy’ in connection with the meaning changes in his Class VI ‘permutation’. Stern's consistent non-use of the term ‘metonymy’ is all the more relevant as he shows no hesitation in localizing metaphor as distributed evenly over his Class IV (‘Transfer’) and Class V (‘Nomination’). Most of his ‘permutation’ examples are traditionally discussed as metonymies (like the famous bead example), yet his central examples are not commonly mentioned as metonymies (like the shift from fast ‘immovable’ to fast ‘at high speed’). Beyond all terminological discussions, however, my interest lies in elucidating the nature of the change processes that can occur under reanalysis. It remains a secondary goal to locate them in any comprehensive classification of modes of meaning change.

2.5 On Discourse Markers

In this section, I will come back to the question how the emergence of discourse markers relates to the notion of grammaticalization. We will consider cases like the shift of German weil, obwohl, or bloβ from subordinative conjunction/focus particle to discourse marker, or the emergence of a discourse marker use of English indeed, in fact, and the conditional if-then.15

The German causal connective weil goes back to the temporal adverb die Weile (‘while’), as traced in Traugott and König (1991). The meaning shift was driven by a pragmatic inference something like ‘while, hence because of’. In recent years, however, as was clearly shown in Günthner (1996, 2001), the particle has acquired a textual function, connecting a sentence with a reason why the speaker has uttered it. In this function, weil is obligatorily followed by a verb-second (German: main) sentence. Minimal pairs are given in (2.6) to (2.9):






















‘Will you go to the swimming pool on Sunday? I'm asking because it will rain.’






















‘Will you go to the swimming pool on sunday because it will rain?’

























‘She must be tired (I infer this because) she has been sleeping for 12 hours.’
























‘She is certainly tired because she has been sleeping for 12 hours already.’

(strangely suggests that long sleeping is the cause of tiredness)

A similar textual function has been pointed out for obwohl (‘although’) in Günthner (2000). In (2.10), obwohl is used as a sentence connective. It has the same function as English even though, asserting the conjoined truth of conflicting facts. Intertextually as in (2.11), obwohl serves to withdraw a previous assertion.


















‘He is unhappy even though he earns a lot of money.’




















‘He is certainly unhappy. Or perhaps that is still not true: after all, he earns a lot of money’

The different uses of bloβ were outlined in Chapter 1 and are further described in Scheier (2002). The development of English indeed and in fact have been traced in great detail in Schwenter and Traugott (2000) and Traugott and Dasher (2002). Tabor and Traugott (1998) report the development of two more English discourse markers, instead and anyway. An inspiring overview of contemporary literature on discourse markers in English and German is put forward in Couper-Kuhlen and Kortmann (2000).

Discourse markers and similar items do not pattern very well with classical cases in grammaticalization. If we apply Lehmann's parameters of grammaticalization (see 2.1), we find that emerging discourse markers do not show the right behaviour in many respects.

Considering the syntagmatic criteria, discourse markers commonly have a higher scope (semantically, frequently also syntactically) than their ancestor items, thus contradicting the tendency towards decrease in scope as suggested by Lehmann. This point has been demonstrated in great detail in Tabor and Traugott (1998).16

(p.48) Discourse markers are tendentially less bonded, less closely integrated into the sentence(s) than their ancestor items. Traditional interpunctation in German even suggests that obwohl in (2.11) or weil in (2.6) and (2.8) should be interjections rather than coordinations. This contradicts Lehmann's criterion of increasing bondedness.

Moreover, discourse markers usually show a greater syntactic variability than their ancestor items. For instance, the German examples listed above commonly show the following pattern: if the item conjoins a main clause with a subordinate clause, it may be either an older conjunction or a newer discourse particle. If the item, however, conjoins two main clauses, it is used as a discourse marker. Hence the discourse markers allow for more freedom, contradicting Lehmann's tendency of decrease in syntagmatic variability.

Looking at Lehmann's paradigmatic criteria, the picture is less decisive. Discourse markers can certainly lose phonological integrity (witness the English in deede > indeed). In terms of semantic integrity, older and newer item are frequently at the same level. The best example to witness this fact may be German weil. It contributes exactly the same semantic content (CAUSAL, ‘because’) at two different propositional levels: ‘A happened because B happened’ versus ‘I say A, because of B’.

Items that give rise to discourse markers usually shift from one closed class to another (temporal conjunction > discourse marker, focus particle > discourse marker, etc.) and hence show no clear increase or decrease in paradigmaticity.

Discourse markers, finally, also do not fit into the traditional Lehmann picture with respect to paradigmatic variability, the degree to which a category becomes obligatory in certain contexts. Although speakers agree that conversations without any use of discourse markers sound unnatural or even deliberately rude, there is as yet no grammatical requirement for speakers to set discourse markers.

In summary, clines that lead from adverbs, conjunctions, or focus particles to discourse markers will tendentially contradict four of Lehmann's six tendencies for grammaticalization. Hence, we will want to conclude that discourse markers do not pattern with other examples of grammaticalization very well. This finding is supported by the observation that the realm of discourse markers seems to offer true two-way pathways of change under reanalysis (see Chapter, 1 n. 3). In spite of the apparent differences between classical instances of grammaticalization and classical emergent discourse (p.49) markers, scholars show a persistent inclination to subsume the rise of discourse markers under grammaticalization. In 1991, Werner Abraham published the rich and inspiring paper ‘The grammaticalization of the German modal particles’ where he describes the emergence of a great wealth of discourse markers. Generally, collections of papers and workshop publications use ‘grammaticalization’ in a wide sense, clearly embracing discourse markers. The above-mentioned article of Abraham is included in Traugott and Heine's Approaches to Grammaticalization, and the contribution by Traugott and König likewise addresses discourse markers alongside other examples. Anna Giacalone Ramat and Paul Hopper's 1998 volume ‘The Limits of Grammaticalization’ includes Hopper's paper on ‘the paradigm at the end of the universe’ where he shows that a vague class of grammatically unpronounced items—he dubbs them as ‘adverbs’ including, of course, discourse markers—seems to constitute one endpoint in grammaticalization. Stefania Giannini's contribution on Italian local deixis in pronoun systems (in the same volume) is another investigation in language change that ends in an indexical rather than a grammatical category (‘intersubjectification’, following Traugott and Dasher 2002). More recently, Wischer and Diewald (2002) state explicitly that they subsume discourse marker clines under grammaticalization even though, as they concede, others might want to draw the line more narrowly.

Traugott and Dasher (2002) suggest that discourse markers and other instances of meaning change under reanalysis should be viewed as a coherent field of data, in particular because discourse markers are prime examples of meaning change by subjectification. While the source items usually make a truth functional statement (temporal relation between two events, causal relation between two facts, scalar assertions, adverbial modifications, etc.) the target item—the discourse marker—makes a more subjective assertion about the speaker attitude and viewpoint, meta-comments on previous discourse, etc. Traugott and Dasher suggest that subjectification is a major driving force in language change in general, thereby implying that grammaticalization and discourse marker clines should be similar in at least this respect, even if in no other (Traugott and Dasher 2002: ch. 7).

Summarizing, we may state that discourse markers, like grammatically bound items, arise by a syntactic and semantic reorganization of previous material in language. In both cases, syntactic and semantic change are inseparably intertwined in a two-way dependency: the new syntactic structure drives the semantic environment, and hence in part the semantic contribution, of the newly emerging item. Conversely, the semantic contribution of the newly emerging item determines its combinatorial potential and therefore, to a certain extent, its future grammatical nature. In both cases, finally, meaning (p.50) change is driven by pragmatic inferencing and conventionalization (see Traugott and Dasher 2002: chs. 3–6). It will have to remain an exciting topic for future research to bring the tools of formal semantics to bear on diachronic and synchronic investigation of discourse markers.

2.6 Communication: The Onsets of Change

Meaning change by reanalysis is unlike metaphor and metonymy in that we only very rarely witness reanalysis in a volitional instance of creative language use. Accordingly, the notion of a ‘dead reanalysis’, mirroring the concept of a ‘dead metaphor’, does not make sense: reanalysis arises as a result of conventionalization; unlike metaphors it is not ‘killed’ by conventionalization. The question of the nature of onset contexts of change therefore poses a prime challenge in the investigation into meaning change under reanalysis. Under what circumstances do speakers feel inclined to use a certain construction so as to invite reanalysis? When do hearers feel the conscious or subconscious inclination to reanalyse a construction—provisionally or as a new construction in their language? How often must this happen in order for it to get established as a persisting change in language?

These questions have received surprisingly little attention in the literature. Janda (2001) draws attention to the fact that sociolinguistic research on grammaticalization is practically non-existent. He concludes that we do not know much about the sociolinguistic nature of the rise and spread of grammaticalization, and we can justifiably read this in the wider sense that we do not know much about the communicative nature of onset contexts in general. Janda refers to the universality hypothesis about language change (Labov 1972), which states that the nature and rate of language change must be approximately the same at all times. As a consequence, he points out, grammaticalization (and reanalysis) must in principle currently take place. Investigations into such current instances would offer richer and more detailed insights than fragmentary written records on language use in the past. Janda therefore advocates more sociolinguistic research in current grammaticalization processes. Yet he fails to name one obstacle to this enterprise, namely our lack of ability to spot onset contexts of grammaticalization, contexts where the transmission and use of new constructions could be investigated.

Still, some concise ideas about the nature of onset contexts of grammaticalization and reanalysis have been brought forward. According to one line of thinking, reanalysis takes place in the tension between holistic and analytic interpretation of language. Haiman (1994) proposes that ritualization is an omnipresent tendency in social interaction which finds its linguistic manifestation in (p.51) grammaticalization. In a similar vein, Lehmann (2002) contrasts holistic and analytic interpretation of language. He argues that holistic interpretations of a phrase or sentence can potentially tie together literal meaning, implied meaning, and linguistic form in a way that results in reanalysis as soon as any one speaker starts to build up the sentence in a literal word-by-word fashion.

Traugott and Dasher (2002) in contrast allow for speakers to intentionally shape their language and to instigate reanalysis. Particular stress is laid on the fact that speakers can invite certain pragmatic inferences, thereby volitionally implying the unsaid and making it part of the information conveyed. As soon as these invitations to understand the unsaid become a conventionalized part of language use, the way is free for reanalysis (see also section 2.4).

Heine (2002) and Diewald (2002) address the quest for the contexts of change under two slightly different, but basically similar perspectives. Heine (2002: 86) distinguishes four types of subsequent contexts of use of an item:

  1. (i) Initial stage

  2. (ii) Bridging contexts

  3. (iii) Switch contexts

  4. (iv) Conventionalization

The neutral initial stage is followed by use of an item in bridging contexts which give rise to pragmatic inferences in favour of the new meaning. In bridging contexts, so to speak, the potential new meaning enters the stage. These contexts are followed by switch contexts where the new meaning aspects are relevant to a degree so as to make the older, literal meaning practically superfluous, even though the use of the construction can still be understood in the literal fashion, accounting for all additional information by way of inferencing. Finally, unambiguous uses of the item in the new sense witness its conventionalization. These uses can no longer be accounted for on the basis of the item in its older meaning.

While Heine's four-stage model is intuitively convincing and coheres well with Traugott and Dasher's model, his applications to sample developments are somewhat less satisfying. Notably, his survey of the development of reflexive to passive marker in !Xun (Khosian language, Southern Angola) is a synchronic collection of formal reflexives of different verbs and their intepretation. Example (2.12) below is offered as an example of a ‘bridging context’ (for transcription conventions see Heine 2002). While it is certainly plausible that the reflexive form of a verb meaning ‘to bear, to give birth to’ in contemporary !Xun will in all likelihood be understood as a passive rather than a literal reflexive, we get no information as to why the respective sentence in !Xun (quoted as a stage (ii) use) is legitimate at all. (p.52)

(2.12) !Xun (North Khosian, Khosian)



gé- à











‘I bore myself in Angola.’

The English gloss would not be ‘bridging’ at all. It would at best be a sentence that signals that the speaker volitionally uses the English reflexive construction in a non-literal way—presumably in order to suggest that his place of birth was Angola. Yet this interpretation involves more than pragmatic inferencing. Importantly, it already presupposes a renovation of language rather than paving its way.

Likewise, Heine's example of a switch context is in fact a straight instance of a passive (formally cognate to a reflexive): ‘The money was stolen,’ lit.: ‘The money stole itself.’ It is not entirely clear why Heine only accepts a sentence as a stage (iv) instance of the passive once the agent is realized as an additional transitive object. He seems to suggest that a stage (ii) and stage (iii) sentence immediately turns into a stage (iv) sentence as soon as an agent like ‘my mother’ or ‘by a pickpocket’ is added. Notably, his examples do not specify the communicative intentions of the speaker, or the relevance of some piece of information in a given context. This contrasts with, for instance, the attempts made in Hopper and Traugott (1993: 82 f.) to speculate about the contexts of emergence of the English going-to future where they state that in a sentence like ‘He is going to be married’ the imminent marriage should be much more relevant for the interlocutors than the actual movement of the referent of ‘he’. No similar pragmatic reasoning is offered in Heine (2002).

Diewald (2002) bases her system on three subsequent types of contexts: atypical contexts of use, critical contexts of use, and isolating contexts of use. Diewald's typology shows visible resemblances to the four-way distinction of Heine (particularly once we add an initial stage of neutral contexts of use); however she focuses on the linguistic environment of an item. An item is used in an atypical context if it gives rise to pragmatic inferences in favour of the new meaning but is structurally an unambiguous exponent of the older grammatical construction. Critical contexts combine semantic and structural ambiguity. Phenomenologically speaking, these are the quotes in corpora where the researcher might wonder whether they show a use of the older or of the newer item. Isolating contexts, finally, are those uses that are clearly based on the new item structurally and semantically.17

(p.53) Both accounts reflect the perspective of the philologist, concentrating on the question ‘when do I—the hearer/reader—know that a language change must be under way?’ rather than on the much more challenging: ‘Which conscious motives or subconscious trends drive me as a speaker/writer to produce such utterances in the first place?’ Neither Heine nor Diewald considers the communicative intention of an utterance, the epistemic background of speaker and hearer, and the question of what might be the most relevant information in a given context. Particularly Heine's examples of ‘bridging contexts’ are clearly the product of language change rather than its onset. In that sense, both authors leave the question about the nature of onset contexts of reanalysis unanswered.

In order to illustrate this criticism, let us come back to the fact that French has a venir de faire immediate past tense, and English and German do not. Certainly, a speaker of present-day English or German can utter a sentence like (2.13):

  1. (2.13) I am (just) coming from doing the shopping.

    Ich komme (gerade) vom Einkaufen.

Such utterances take place from time to time, and all give rise to the implication ‘I have just done the shopping’—or more generally ‘I have just done X’, depending on the embedded verb X. Would such uses be bridging contexts in terms of Heine? Certainly yes, if we look at Heine's verbal characterization of ‘bridging context’. Not so clearly, however, if we take his examples of bridging contexts into consideration. While a reflexive of the verb to bear already requires a reinterpretation, nothing in the sentences in (2.13) shows imminent language change (and to my knowledge, neither English nor German shows any traceable tendency to develop an immediate past tense).

Would sentences (2.13) be critical contexts in terms of Diewald? Certainly yes, if we look at her verbal characterization of ‘critical context’. Yet it seems that such utterances miss exactly that critical bit that would turn them into onset contexts for reanalysis. More than mere pragmatic inference is required in order to force a construction's meaning to shift. In the sense of conservative language use, there is nothing critical or non-standard in sentences (2.13). It appears that Lehmann (2002), even though he operates in much more traditional terms than both Heine and Diewald, might still have something to say as to when sentences like (2.13) turn into ‘critical mass’ for language change. What is crucially missing in the accounts of Heine and Diewald is a consideration as to when and why a certain phrase or construction (the reflexive, a modal, the coming from doing construction) turns into a stereotypical way to convey holistically some piece of information. Under what circumstances would a speaker use the coming from doing construction as a (p.54) ritualized construction to inform the hearer about recent activities? What ‘critical contexts’ did speakers of French encounter that speakers of English and German did not? The difference between a neutral utterance of (2.13) and a ‘critical use’ of (2.13) cannot be captured in terms of pragmatic inferencing: the implications of (2.13) about the immediate past are inevitable. Taking up Lehmann's proposal, we may speculate that an implication becomes critical once speakers start to use (2.13) as a somewhat longer version of I have just done the shopping (German: Ich habe gerade eingekauft). Chapter 4 includes an investigation into the contexts of emergence of the going to future that is conducted under such leading questions as ‘when does conventionalization start?’ and ‘what kinds of contexts support conventionalization’. Even though the results have to remain incomplete—given that we have almost no information about spoken discourse at the time—they show that conventionalization can be traced even after several centuries.

Traugott and Dasher (2002) focus on speakers' attempts to increase the expressiveness of language. This suggests that they view creativity as the driving force in conventionalization. The views of Haiman (1994), in contrast, might be interpreted so as to lean towards the laziness explanation. The nature of available data will frequently make it impossible in a particular instance of language change to investigate the communicative nature of onset contexts. As pointed out by Janda (2001), the investigation of language change and variation in the recent past, and sociolinguistic investigation of present-day language use, can be more elucidating than linguistic palaeontology—which may be more spectacular, but also more speculative. Language changes of a more recent date offer an exciting field of investigation into issues where sociolinguistic, diachronic, interactional, and philological aspects of communication need to be considered in concord. The case studies in Chapters 5, 6, and 7 offer a first step towards investigating the communicative factors that instigate reanalysis. Even though they are necessarily fragmental, they may help to get a better understanding of the onset of language change.

2.7 Grammaticalization, Once Again

I reviewed previous accounts of meaning change under reanalysis, in the literature frequently addressed more narrowly as meaning change under grammaticalization. I surveyed approaches that treat the respective changes as bleaching, as meaning changes by metaphor, and approaches that view them as mainly pragmatically driven meaning change (also studied under the label metonymy). In the remainder of the book I will adopt and refine the last approach and develop a spellout in terms of truth conditional semantics.

(p.55) This basic position has repercussions on the question whether we view grammaticalization as occurring in discrete steps or as a gradual process. I will assume that the meaning changes under investigation occur in discrete steps. Even though the subsequent language stages may only be minimally different, it will be fruitful to view them still as distinct positions in language history. More specifically, speakers are conceived as passing through discrete stages in steps that reflect Heine's or Diewald's four-stage systems in section 2.6.

  1. (i) The speaker possesses a lexical entry for the traditional ‘old’ item, including information about grammatical behaviour, meaning, and conventions of use.

  2. (ii) The speaker still possesses only one lexical entry, but experiences that the conventions of use undergo a change so that, using the item in certain kinds of sentences, interlocutors can conventionally be expected to be more concerned with certain implications of the assertion than with the literal assertion itself.

  3. (iii) Experiencing such uses of the old item, the speaker/hearer hypothetically adopts a new syntactic and semantic analysis for the respective full sentences. What once was a salient implication is understood as the literal content under the new analysis. This reanalysis leads to a hypothetical new entry in the lexicon, pairing the phonological content of the older item with new grammatical and semantic information.

  4. (iv) If the hypothetical new entry is confirmed in more interaction (or if the speaker has an innovative temperament), it is adopted as a permanent entry in the mental lexicon. The speaker will henceforth use the new item trustfully (hence producing Diewald's ‘isolating contexts’).

Let me point out that the speaker at stage (iii) is not uncertain about the nature of the new entry nor does he have to develop the new entry ‘gradually’ from the older one. The only uncertainty that might be experienced is of a sociological nature: do other speakers actually possess and use this new entry, or not? While the speaker may be gradually more and more convinced that the new item is indeed part of the common language, he or she does not, according to this picture, gradually develop the meaning and grammar of the new entry.

This four-stage process can be viewed as a speaker-internal version of Traugott and Dasher's Invited Inferencing Theory of Semantic Change. The resulting picture does not stand in conflict to the observation, made by several scholars, that grammaticalization is epiphenomenologically a gradual process. The following quote from Lehmann (concerning the grammaticalization of auxiliaries) reflects the latter position: (p.56)

The dispute whether auxiliaries are main verbs or not … is fruitless. Two grammatical categories connected on a grammaticalization scale are neither the same nor distinct. The difference between them is gradual, and there is no clear-cut dividing line. (Lehmann 1982 [1995]: 33)

I agree with Lehmann (as well as others, see for instance Haspelmath 1998) that the major grammatical categories that are in use in syntactic theory will in all likelihood not be sufficient to trace the—perhaps small—steps along a grammaticalization cline. I volitionally used the term ‘grammatical information’ rather than ‘syntactic category’ in the characterization of lexical entries in (i) to (iv) above. Semantic information, on the other hand, can be fine-grained to any necessary degree such that the formal framework poses no lower limits to the discrete steps in semantic change. Nevertheless, the speaker will have to justify each single use of a form by appropriate lexical information.

I likewise agree that the uses of an item in source texts seem to support a gradual view on language change, an impression that the case studies in later chapters will amply confirm. This fact can however be reconciled with a model of meaning change in discrete steps. Specifically, speakers in stage (iii) above will produce sentences that can equally well be analysed on the basis of the old item as on the basis of the new item, the two analyses sometimes differing slightly in their pragmatic links to previous discourse. (The case studies in later chapters, particularly the development of selbst and lauter, will offer examples.) To the degree that speakers gain faith in the hypothetical new item, they will use it in more and more contexts, gradually manifesting its full range of potential uses in the data. This process was first described as actualization in Timberlake's (1977) ‘Renalysis and actualisation in syntactic change’. The recent collection of studies in actualization by Andersen (2001c) reflects a renewed interest in this notion. Andersen (2001a, 2001b) explores the possibility that further language internal factors might drive actualization, and specifically that markedness has an influence on the eventual outcome. In terms of discrete language change, we must hence in principle be aware of the possibility that the four stages (i) to (iv) can be passed anew even before full actualization of an intermediate stage has taken place. We might also want to allow for a correction of the hypothetical new item at stage (iii). I will in the following maintain the discrete picture of language change by reanalysis, allowing for these refinements.

The present chapter, offering but an incomplete overview of current research in grammaticalization, may help to relate it to the topic of the present monograph. Approaching the data in meaning change under reanalysis (p.57) with the tools of truth value based semantics, I aim at making a contribution to research in language change and variation in the following respects.

  • What is the nature of the contexts that instigate reanalysis? Perhaps surprisingly, it turns out that semantic analysis in terms of truth conditional approaches can shed new light on this question. We will see cases where reanalysis seems to have been enhanced by the fact that the traditional literal meaning of the sentence(s) was simply unverifiable and therefore pointless in certain contexts of use. A truth conditional semantic approach can highlight the consequences of this kind of communicative failure and capture the semantic defectiveness that can lead to language change.

  • What leads from sentence meaning to word meaning? It has never been doubted that speakers can derive single word meanings from sentence meanings, if most other parts of the sentence are known. This process at the syntax–semantics interface can be successfully captured with the tools of formal semantics, resulting in a semantic representation of meaning change under reanalysis that brings several facts about language into focus: the compositionality of human language not only allows its flexible use in synchronic communication but also ensures an automatic adaptation of language to new communicative needs and practices. And it does so with almost mathematical precision, thus ensuring that the linguistic system as a whole will work smoothly and reliably even if parts of it are under change.

  • To what extent can reanalysis predetermine the results of actualization? If we grant speakers the ability to store and process fine-grained grammatical and semantic representations of words, phrases, and sentences, we will predict that the results of language reorganization are determined with great precision. This view can help us to understand why speakers will often agree about fairly peculiar aspects in the use of a new item. The case studies in the present monograph suggest that there is much less accident and contingency in language change than one might expect: while it remains a matter of historical accident whether a certain kind of change gets instigated in a certain type of onset situation, the results of this change appear to be determined in most respects. This astonishing combination of flexibility and determinedness in the language system ensures its reliability as a tool for communication as well as its adaptiveness under constantly changing circumstances.

Some readers may feel that questions like these have already received ample treatment in the literature and need not be taken up in terms of yet another semantic paradigm. Yet the following quote suggests that at least the relations (p.58) between word meaning, sentence meaning, and semantic composition do require clarification. Haspelmath (1999: 1062) proposes that

One of the most widely discussed aspects of grammaticalization, the fairly dramatic semantic changes, has not been mentioned explicitly at all so far. The reason is that I am not sure that semantic grammaticalization is as central to the process as has generally been assumed…. For instance, the emphatic negation marker pas in older French has lost its pragmatic markedness and has become the normal negation marker, without any semantic changes in the narrow sense having taken place. (emphasis mine)

This quote suggests that there is no semantic side to grammaticalization and hence no interesting object for study. Haspelmath illustrates his claim with the development of French pas = STEP to pas = NEGATION PARTICLE, an issue which will be taken up in Chapter 5. While he correctly observes that the overall sentence meaning of the crucial examples does not change, he fails to acknowledge that the meaning change at the word level is considerable. An explicit account of meaning change under reanalysis will be of help to clarify distinctions of this kind.


(1) Exceptions granted, see Heine and Reh (1984: 95): ‘Reanalysis thus appears as a concept which is largely synonymous with our term “grammaticalization”.’

(2) Quoted after Lehmann (1982 [1995]: ch. 1).

(3) Haspelmath (1998) offers a provoking antithesis to this general assumption: ‘Does grammaticalization need reanalysis?’ Haspelmath's claim however rests crucially on an idiosyncratic use of core notions like ‘reanalysis’ and ‘grammatical structure’ the critical review of Haspelmath (1998, 1999) in Campbell (2001) will certainly come as an intellectual relief at least for some puzzled readers.

(4) Webelhuth (1999) however made an interesting proposal to analyse the development of the German werden passive in an HPSG framework.

(5) Two-way classifications notoriously bear the implicit suggestion that the values ‘good’ and ‘bad’ should also be attributed to the two classes.

(6) ‘What first was new and striking will soon become common, and it thereby loses strength, bleaches, and eventually turns into one of those abstract parts of speech which it initially was coined to improve and complement as a stronger expression…. What is true of functional words is likewise true of word forms. Wherever new ones were coined, they were periphrastic … fresher, newer colours covered the old, bleached ones.’

(7) ‘The weakening of meaning, and the weakening of the surface form of the respective word, go hand in hand; once one and the other have advanced far enough, the word in question can end as being nothing more than an item deprived of any real content.’

(8) Campbell (2001) rightly points out that the danger of circular argument lurks behind this statement. In languages without written records, the search for etymological origins of a grammatical word is frequently driven by very concise expectations as to what the source item might mean. Such cases should, Campbell states, not be used as independent evidence in favour of a certain pathway of change.

(9) This is not as much an armchair insight as it may look at first sight. We may faithfully assume that our current language is changing during our lifetime, as much as it did at any earlier time. Given this, we are certainly witness to some early stages of grammaticalization for some words, phrases, or constructions. Hence, it is telling that the scenario described in (2.3) does not constitute part of our own communicative experiences. We basically explore Labov's (1972) universality hypothesis here; see section 2.6.

(10) In making this blunt statement, I subscribe to a notion of metaphor that rests on the intuitive plausibility of an intended match between source domain and target domain. While most theories of metaphor rightly assume that homomorphic source and target domains are one important precondition for metaphor, none could so far offer a formal spellout of when the structural similarity may be explored by metaphor (e.g. Gentner 1983, Indurkhya 1992). The most striking evidence in favour of the claim that mere similarity is not enough is offered by the well-known observation that metaphors are not symmetrical, even though ‘similarity’ is practically always presented as a symmetric relation. ‘Bad metaphors’ are another example to the same end (see Bartsch 2002). I feel therefore justified in taking ‘speakers' intuition’ as a reputable empirical criterion to spot metaphors.

(11) While Heine, Claudi, and Hünnemeyer concede that emerging metaphor arises through pragmatic inferencing (thus anticipating the pragmatic approaches in section 2.4), we may doubt whether new meanings that arise in grammaticalization through pragmatic inferencing do indeed require any extra licensing by metaphor. The facts favour the contrary.

(12) Itkonen criticizes, rightly in my view, previous proposals and particularly the frequently quoted paper by Andersen (1973). He shows that even though the general idea was right, Andersen equated the steps of hypothetical-deductive reasoning with speaker's rationale in language change in a mistaken manner.


(14) Unless one were willing to read idealized conceptual model liberally as anything that you might want to think of.

(15) As in if you are hungry, there is beef in the fridge.

(16) They even go as far as to suggest increase in structural scope as a criterion for grammaticalization that should replace Lehmann's decrease in structural scope tendency. Still Tabor and Traugott criticize Lehmann's criteria in a conservative fashion, suggesting that one general tendency should be replaced by another.

(17) We might hence speculate that Heine's stage (ii) example would be classed as an isolating context in Diewald's system.