Добавил:
Upload Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
Скачиваний:
192
Добавлен:
08.06.2015
Размер:
3.63 Mб
Скачать

538 Elizabeth Closs Traugott

24 Historical Pragmatics

ELIZABETH CLOSS TRAUGOTT

1 The Field

During the last two decades, the study of meaning change based in theoretical pragmatics has come to play an important part in our understanding not only of semantic change and lexicalization, but also of the relationship between structure and use in general, most especially the nature of contextual meaning.1

Historical pragmatics is a usage-based approach to language change2 which came to be identified and institutionalized as a field of study largely owing to the work represented in Jucker (1995) and in the Journal of Historical Pragmatics. Jacobs and Jucker (1995) present an overview of historical pragmatics from various perspectives, and characterize it as being essentially of two types, which correspond roughly to the distinction between “external” and “internal” language change. The first they call “pragmaphilology.” This is a primarily a “macro-approach” (Arnovick 1999), and the focus is on the changing social conditions in which linguistic change occurs, for example changes in the “aims, motives, interests, public and private behaviour, institutions, formulae and rituals” (Jacobs and Jucker 1995: 5). The prime data are text types that are written or spoken: monologues, conversations, etc. seen in terms of religious, legal, pedagogical, and other norms of text production and reproduction. Research along these lines poses anthropological and cross-cultural or intercultural questions. Jacobs and Jucker call the second type of work on historical pragmatics “diachronic pragmatics.” This approach is typically a “microapproach.” The focus is on the interface of linguistic structure and use, and on “what types of rules, conditions, and functions of social acts were effective in earlier stages or processes of language change” (1995: 5). Data are textual evidence for the development of, for example, honorifics, focus particles, discourse markers, or performative uses of locutionary verbs. Within diachronic pragmatics, Jacobs and Jucker further distinguish two well-known approaches that will be important in discussion below (for fuller details see Geeraerts

Historical Pragmatics 539

1997: 17–18). One involves “form-to-function mapping” and is “semasiological.” The dominant question is: What are the constraints on ways in which a meaning can change while form remains constant (modulo independent phonological changes)? For example, what are the constraints on the ways in which may developed polysemies over time? The other approach involves “function-to- form mapping” and is “onomasiological.” The dominant question is: What constraints are there on recruitment of extant terms to express a semantic category? For example, what constraints are there on development of lexical resources for expressing epistemic possibility?

Although some very important theoretical work has been based on dictionaries, claims made by scholars like Bréal (1900) and Ullmann (1959), or introspection (see especially Horn 1984a et passim), much recent historical pragmatics is based in textual data. Some text-based work has investigated changes in patterns of foregrounding and backgrounding of material in the narrative story-line (e.g. Hopper 1979, Fleischman 1990) or of markers that signal narrative structure (e.g. Brinton 1996). Hardly surprisingly, there is considerable overlap between text-based historical pragmatics and historical discourse analysis; whether there is any significant difference between the two is debatable. Brinton (2001) provides a threefold classification of historical

discourse analysis: HISTORICAL DISCOURSE ANALYSIS PROPER, DIACHRONICALLY ORIENTED DISCOURSE ANALYSIS, and DISCOURSE-ORIENTED HISTORICAL LIN-

GUISTICS. These distinctions are based primarily on the methodological perspective of the work. According to Brinton, historical discourse analysis proper is essentially synchronic: the study of pragmatic factors such as orality, text type, or narrative markers, at a particular language stage; for example, the use of narrative markers in Middle English (this mode of work include aspects of what Jacobs and Jucker call PRAGMAPHILOLOGY). Diachronically oriented discourse analysis examines the evolution of forms or systems that have a discourse function, such as the development of particular adverbials or verbal phrases into narrative markers, or changes in the systems of narrative marking over time. Discourse-oriented historical linguistics is an approach to historical linguistics that seeks to find the origins of or motivations for change in discourse, for example the origins of semantic change in the conventionalizing of implicatures.

The present essay will discuss issues primarily from the perspective of diachronic pragmatics and diachronically oriented discourse analysis, taking pragmatics to be non-literal meaning that arises in language use. Section 2 outlines some proposals regarding discourse pragmatic origins and motivations for semantic and lexical change from a primarily synchronic “neo-Gricean” perspective, since this has been the most influential and widely used theoretical approach in historical pragmatics. Data used from this perspective are usually presented out of context, as individual lexical items, or constructions. Section 3 introduces proposals embedded in historical linguistics and based on textual data from historical corpora. Section 4 provides a case study using data of this type.

540 Elizabeth Closs Traugott

2From the Synchronic Perspective on Pragmatics

Much historical pragmatics builds on work in the early 1970s. An early and seminal suggestion was made in a short paper by Geis and Zwicky:

It seems to be the case that an invited inference can, historically, become part of semantic representation in the strict sense; thus, the development of the English conjunction since from a purely temporal word to a marker of causation can be interpreted as a change from a principle of invited inference associated with since (by virtue of the temporal meaning) to a piece of the semantic content of since. (Geis and Zwicky 1971: 565–6)

In this idea they echo Grice (1967), who tentatively remarked, “it may not be impossible for what starts life, so to speak, as a conversational implicature to become conventionalized” (Grice 1989: 39; for an early follow-up see Brown and Levinson 1987 on the development of honorifics).

Two questions have been of central concern since the mid-1970s:

1Do different conversational maxims motivate different types of semantic change?

2Does Grice’s distinction between particularized and generalized conversational implicatures help account for how semantic change occurs?

A further question has been posed primarily within historical linguistics:

3Are there additional important factors that need to be considered in accounting for frequently observed types of semantic change?

The first two questions are the topic of the next subsections, the third of section 3.

2.1The role of conversational maxims in semantic change

A central issue in the debate around Gricean pragmatics has been discussion of the validity of his maxims. These were reconceptualized by Horn as “principles.” Levinson further reconceptualized them as design features of communication or “heuristics,” available to speakers and hearers when they attempt to solve the problem of converting thought into speech (“heuristics” is the term adopted here).3 In neo-Gricean pragmatics, as exemplified by, for example, Atlas and Levinson (1981), Horn (1984a) and later works, some kind of division of labor has been maintained between what Grice initially identified as Quantity1: “Make your contribution as informative as is required (for

Historical Pragmatics 541

the current purposes of the exchange)” and Quantity2 (“Do not make your contribution more informative than is required”) (Grice 1989: 26). Among reasons given in Horn (1984a) and Levinson (2000a) for retaining the division of labor, despite objections from other research paradigms, especially Relevance Theory (e.g. Sperber and Wilson 1986a), is semantic change.

Since Horn’s and Levinson’s proposals, although related, suggest slightly different issues for semantic change, I treat each separately.

2.2 Horn’s proposals

Invoking Zipf’s (1949) recognition that much of language use can be accounted for in terms of the competing forces of speaker economy vs. hearer economy, Horn collapsed Grice’s Maxims into two principles, Q(uantity) and R(elation):

(1)a. The Q Principle (hearer-based):

MAKE YOUR CONTRIBUTION SUFFICIENT (cf. Quantity1). SAY AS MUCH AS YOU CAN (given R).

Lower-bounding principle, inducing upper-bounding implicata.

b.The R Principle (speaker-based):

MAKE YOUR CONTRIBUTION NECESSARY (cf. Relation, Quantity2, Manner).

SAY NO MORE THAN YOU MUST (given Q).

Upper-bounding principle, inducing lower-bounded implicata (Horn 1984a: 13).

Q-based implicature is “typically negative in that its calculation [by the hearer] refers crucially to what could have been said, but wasn’t,” and is therefore linguistically motivated, as exemplified by scalar implicatures such as hold between the members of the pair all, some . On the other hand, R-based implicature “typically involves social rather than purely linguistic motivation,” as exemplified by indirect speech acts (Horn 1996a: 313). Nevertheless, as discussed below, since the division of labor between Q and R principles often concerns lexical distribution of less vs. more complex forms, R-based implicature is often also linguistically motivated.

Horn relates the Q- and R-based implicatures to two types of semantic change that are well known from the work of Bréal and Ullmann: broadening and narrowing. He suggests that broadening is always uniquely R-based, e.g. xerox, kleenex (Horn 1984a: 35). In the case of xerox, a salient exemplar of a wider class, e.g. copy-machines, is generalized to denote that wider class. This is a case of form-to-function or semasiological change.

Where narrowing is concerned, the issues are more complex. Horn (1984a) mentions various types of semasiological narrowing in which a superordinate term comes to be interpreted as the complement of a hyponym, e.g. in certain

542 Elizabeth Closs Traugott

circumstances, finger is interpreted to exclude its hyponym thumb (e.g. when one says I hurt my finger), or rectangle is interpreted to exclude square. He calls this AUTOHYPONOMY. As he argues in Horn (1984b: 117), contra Kempson (1980), such narrowings tend to be highly irregular (“an ornery array of disparate cases”). The examples appear to be motivated by euphemism (cf. stink – smell), restriction of technical terms (e.g. rectangle), association with particular contexts (e.g. drink “alcoholic beverage”), and no generalization seems possible other than: “Diachronically, implicated autohyponymy leads to systematic polysemy” (Levinson 2000a: 103).

The second type of narrowing is not semasiological but onomasiological since it involves alignments among the meanings of lexical resources given a pre-existing set of “closely related meanings.” Synchronically there is often a “briefer and/or more lexicalized” form that coexists with a “linguistically complex or more prolix” expression (Horn 1996a: 314). The pair will typically reflect a pragmatic division of labor: “Given two co-extensive expressions, the more specialized form – briefer and/or more lexicalized – will tend to become R-associated with a particular unmarked, stereotypical meaning, use, or situation, while the use of the periphrastic or less lexicalized expression, typically (but not always) linguistically more complex or prolix, will tend to be Q- restricted to those situations outside the stereotype, for which the unmarked expression could not have been used appropriately” (Horn 1996a: 314). The less complex term is synchronically narrowed by R-based inferencing that crystallizes “unmarked” meanings such as kill, or will (future). By contrast, the more complex term is Q-restricted: cause to die implicates that direct causation does not obtain, or that the speaker does not have adequate information to vouch for it (Horn 1984a, 1996a, citing McCawley 1978; see also e.g. Langacker 1987, vol. 2); be going to “blocks the indirect speech act function of promising” conveyed by will (Horn 1996a: 314).

Horn’s claim about the division of labor is understood as motivating the principles variously referred to as BLOCKING (Aronoff 1976) or the PRINCIPLE OF CONTRAST (E. Clark 1993). It appears to be generally, perhaps universally, true that there are no true synonyms4 (see Haiman 1980a on iconic isomorphism), and that in general, given two or more semantically related lexemes, the more complex form (morphologically derived or periphrastic) represents the more specialized or less stereotypical meaning. From a historical point of view, meaning change and the development of new lexical resources are clearly constrained by “Avoid Synonymy” (Kiparsky 1982, Horn 1996a) and the principle of the division of labor. We see this repeatedly in grammaticalization, the stereotypic examples of which involved the recruitment of a prolix expression (often a construction such as be going to) into an extant lexical field in certain highly constrained contexts, followed by a realignment of the members of the extant set, and often the replacement of the earlier by the later construction (see Hopper and Traugott 1993). We also see it when synonymous lexemes appear or are borrowed (even though semantically synonymous the latter will always be pragmatically differentiated, precisely because they are borrowings).

Historical Pragmatics 543

Before looking at some striking cases of diachronically operative constraints on lexicalization that have been argued to be motivated by the division of labor between more marked and less marked members of a lexical pairing, several caveats deserve mention:

1As Horn has often pointed out, the principles are tendencies only. Indeed this is expected, given that change is never wholly predictable.

2Discussions of synonymy-avoidance tend to be made in terms of pairs, e.g. cook–cooker, drill–driller, kill–cause to die. However, these pairs have been extrapolated from a larger lexical set, making “synonymy-avoidance” less easy to pinpoint. In a recent study of the complexities of investigating the meanings of competing forms for various types of theft in Old and Middle English, Roberts (2001) shows that in Old English there was a large set of words for what we now think of as robbery (involving violence) and theft

(not involving violence). These included several lexemes based on stal- “steal,” ðief- cf. “thief,” reaf cf. “rape.” When rob- forms (< Lat. robaria “robbery”) appeared in Middle English around 1200, they were apparently used synonymously with native forms, but over the next 300 years ðief- forms gradually narrowed, stal-forms were restricted to e.g. stealth, and a variety of other lexical alignments took place. Lexical pairs are, therefore, often only part of a larger, complex story, in which competition, collocation, and specialized (in this case legal) use may play a part, clouding the attractiveness of claims about Q vs. R-narrowing made out of the context of the whole lexical field in question.

3What “synonymy” is may be theory dependent. To what extent should driller or cooker really be considered to be “affixally derived form[s] synonymous with” (Horn 1984a: 25, my emphasis) the simple forms drill and cook?

4Although “different form implies (partially) different meaning” is often regarded as crucial to an explanation of how the lexicon is constructed and maintained or realigned, its complement “same form implies (partially) same meaning” is often treated differently. Such a principle is often regarded with suspicion, and has been interpreted either as evidence for monosemy (one core meaning – see Relevance Theoretic approaches), or homonymy (see generative theories of semantics). However, from a historical perspective, a theory of semantics encompassing polysemy is crucial if an adequate account is to be given of semasiological change. Once the validity of polysemy is accepted, then we begin to see that some of the examples cited as “synonymy-avoidance,” such as the development of cooker beside cook, driller beside drill, may be thought of from a historical perspective as avoidance of polysemy across noun-verb doublets, i.e. innovation to construct contrast, rather than as synonymy-avoidance. However, this is a very weak tendency, as can be seen from productivity of noun-verb doublets with no morphological or stress pattern difference between them such as blanket, butcher, sanction.

544 Elizabeth Closs Traugott

5A “simple” form may historically derive from a complex one by loss of morphological and semantic compositionality or grammaticalization (development of grammatical out of lexical or constructional meaning), with the result that over time what was originally complex and marked will become the unmarked term, and a new marked one will enter the system. For example, will, itself originally a marked periphrastic future that developed to signal future meaning in Middle English, is being replaced by be gonna at least in spoken language (Krug 2000).

6The productivity of sets of related lexical items as well as the meanings associated with them may change over time, as may the extent to which there

is evidence for division of labor. An example is the development of full lexical verb vs. “light verb” + deverbal noun (e.g. advise/give advice, help/give help, answer/give an answer) (Brinton and Akimoto 1999). Old English (c.600– 1150) provides relatively little evidence for such pairings. The few examples that exist are firmly based in larger lexical sets and appear to show only part of the division of labor found in Modern English (primarily correlation with transitive–intransitive, e.g. cyðan “to make X known” – cyððe habban “have knowledge”; wrecan “to avenge X” – wracu don “exact revenge”). During Middle English (c.1150–1500) periphrastic constructions of various

types developed, e.g. prepositional phrases, complex verbal phrases (preauxiliaries + main verb constructions), and also article + noun phrase; all these syntactic changes together with borrowings from French correlated with increasing expansion of the lexicon, and by Early Modern English (c.1500–1700) the pairings in question had not only become highly productive but were also developing new semantic significance. In Modern English “the complex verb is an important means of making situations telic, that is, of converting activities into accomplishments or achievements, yet without the necessity of stating an explicit goal (e.g. dream/have a dream, nibble/have a nibble . . .)” (Brinton and Akimoto 1999: 6). R-based narrowing of the simple form appears to have occurred only after the complex form became entrenched in competition with the simpler one.

7Semantic and lexical change have sometimes been considered to be the same thing (see Householder 1992) and Horn and Levinson appear to equate them. Unquestionably there is considerable overlap: word formation involves meaning. This truism does not, however, entail that all word formation involves change in the same way. When speakers semanticized the causal implicature from temporal since, they developed a new polysemy; this had distributional consequences for the older lexeme, only secondarily for the lexical field of causal connectives. But when speakers innovated cooker, they developed a new form and also a new meaning that had consequences for the lexical field of instruments for cooking, only secondarily for the lexeme cook. The distinction being made here is once again the distinction between semasiological and onomasiological change. Because the consequences of these innovations are different, inferences do not necessarily function in the same way on the two dimensions.

Historical Pragmatics 545

All these caveats aside, the hypothesis that Q- vs. R-based inferencing motivates lexicalization is a powerful one. Horn (1989: chapters 5 and 6; this volume) makes a particularly interesting hypothesis regarding the relevance of the Q principle in an attempt to predict the (near) universal lack of lexical items that denote the negation of the weaker member of a scaled pair. He points out that in the logical square of oppositions O (“particular negative”) is typically not lexicalized as a monomorphemic form, although periphrastic expressions may occur. Thus we find some, all, none, and not all, but not *nall. Likewise there is no *nalways, or *noth of them. Horn proposes that the reason for such structural gaps is the conventionalization of (defeasible) scalar implicatures. The generalization is that historically a lexicalized expression of negative scalar O will not occur since I and O co-implicate each other and pre-existing I blocks O.

Horn has also shown that the lexicalization of modals has similar restrictions: alethic (or in natural language, more properly epistemic) modality typically allows only complex forms in the O position. For example, in contemporary English there is no monomorphemic form meaning “possible not” or “permit not” in the O corner, and in American English no reduced form mayn’t occurs for “permit not.” Instead, we find may not, or more complex forms yet, e.g. You are not allowed to go. However, the constraints are considerably weaker than in the case of quantifiers (or logical conjunctions, cf. not and for *nand). In modals O can in fact be expressed (cf. needn’t, British English mayn’t) and older forms may acquire polysemies by R-narrowing to E. In other words, diachronically there is an attested drift from O E: “The outer negation associated with a necessity predicate often seems to develop an inner negation reading” (Horn 1989: 261, citing Tobler 1882).

In a study of 29 languages of Europe and India, van der Auwera (2001) confirmed that while quantifiers and conjunctions do not lexicalize negative O, or appear in reduced form, modals sometimes do. Like Horn, he hypothesized that “implicatures can conventionalize, make the expression vague between the original literal meaning and the conventionalizing implicature and later oust the original meaning” (van der Auwera 2001: 32). One of the factors that motivates the frequent renewal and sometimes reduction of the negative modal O, according to him, is the fact that there is a great deal of vagueness in the modal domain.

In sum, Horn’s approach provides valuable insights into the ways in which pragmatic principles may give rise to change, especially in the realm of lexicalization and the onomasiological realignments associated with it. It remains to be seen, however, to what extent some of his findings (and equally those of Levinson, to be discussed immediately below) depend on the artifact of selecting pairs from larger lexical sets.

2.3 Levinson’s proposals

A related but somewhat different neo-Gricean account of Q- and R-based phenomena appears in some detail in Levinson (2000a). He cites Horn’s work

546 Elizabeth Closs Traugott

at considerable length on Q- and R-based inferencing, but does not restrict the division of labor to two principles. Instead, he works with three “heuristics” (2000a: 35–9; cf. Huang, this volume):

1THE Q HEURISTIC: “What isn’t said, isn’t.”

2THE I HEURISTIC: “What is expressed simply is stereotypically exemplified” (the “I” stands for “informativeness”).

3THE M HEURISTIC: “What’s said in an abnormal way isn’t normal”).5

He suggests that where inconsistent implicatures arise, they are (synchronically) “systematically resolved by an ordered set of priorities” (2000a: 39), among them:

(2) Q > M > I (read > as “defeats inconsistent”) (Levinson 2000a: 39)

Levinson proposes that Q and M have priority over I because I inferences are “based primarily on stereotypical presumptions about the world” and Q and M can be used to show that these I-inferences do not hold (2000a: 40). I implicatures are grosso modo Horn’s R implicatures, but Relation is downplayed in favor of Information. Levinson goes on to say that “Q relies on sets of alternates of essentially similar form with contrastive content, whereas M relies on sets of alternatives that contrast in form but not in inherent semantic content” (40); in other words, Q operates on contrasting sets of lexemes of similar morphological complexity (e.g. all, some), M operates on word-form(ation)s (e.g. drill–driller, advise–give advice). Therefore Q and M implicatures overlap with Horn’s Q implicatures and in particular his Division of Pragmatic Labor.

To what extent can (2) be projected onto historical pragmatics? As Levinson has pointed out for over two decades: “One major source for new grammatical constructions is what we have called I-implicatures” (2000a: 263), e.g. in the development of be going to, or of morphological markers of honorification in Japanese; I implicatures are also the major source for new polysemies, e.g. the temporal and causal polysemies of since. According to Levinson, M implicatures “seem to be essentially parasitic on corresponding I implicatures: whatever an unmarked expression U would I-implicate, the marked alternative (denotational synonym) will implicate the complement of U’s denotation” (2000a: 137; italics original). Historically, examples are said to include autohyponyms like Horn’s examples of rectangle and informant. Q-based implicatures affect semantic change primarily in the very restricted domains of the negation of quantifiers, logical connectives and modals, most particularly as they enter into the relationships postulated to exist in the Aristotelian square of oppositions. Given the complementarity of the domains of M- and Q-based heuristics in Levinson’s system, the need for both heuristics is unclear synchronically. Diachronically it is unnecessary (see Traugott, in press). Rather than invoke the M heuristic, it would appear to be preferable to invoke a general heuristic of the type called PRINCIPLE OF CONTRAST, a constraint that appears to be strong for children

Historical Pragmatics 547

(E. Clark 1993), but weak for adults, who presumably innovated such forms as cooker, driller, informant, etc. For adults the principle operates primarily on already extant competing forms, but is no doubt also drawn on when new words are deliberately constructed, e.g. by advertising agencies, or are deliberately restricted, e.g. by practitioners of professional discourse.

Levinson’s major contribution to historical pragmatics is that, having earlier suggested that the development of honorifics is the result of the conventionalizing of conversational implicatures (see Brown and Levinson 1987), he further proposed that “it is possible to argue that there is a sequence from particularized through generalized conversational implicatures to conventional implicatures” (Levinson 1979a: 216). The argument for the importance of distinguishing token implicatures that arise on the fly in conversation (particularized conversational implicatures or PCIs) from preferred type-implicatures (generalized conversational implicatures or GCIs) was expanded in Levinson (1995 and especially 2000a). It can be abbreviated as (3), where SM means conventionalized or “semanticized” (i.e. coded) meaning:

(3) PCI GCI SM (read as “may become”)

The key claims may be stated as follows:

1PCIs are highly context-dependent, and are not stably associated with any linguistic form.

2GCIs are normally, even stably, associated with certain linguistic forms.

Both types are defeasible in the sense that they can be defeated by the addition of premises.

From a historical point of view, provided we have a rich enough database, we can see how virtually every example of semasiological innovation (whatever its heuristic base) must have started with a PCI, i.e. a possible meaning in context. Typically the numbers of examples that qualify are minimal (one or two within a number of texts at a particular time or in a particular author’s work). Over time the number of eligible readings increases, perhaps at differential rates in different text types, and can be hypothesized to have become a preferred, if not default, reading, as the result of the operation of appropriate heuristics. Eventually this preferred reading may become conventionalized, or “semanticized,” as a polysemy of the item in question. Levinson gives the example of the development of the reflexive marker -self in English. Originally used as an emphatic as in Hi hiene selfne gefengon “They captured even him” (lit. “They him himself captured”), it could on occasion be construed reflexively; by Middle English, it came to be used reflexively with dramatically increasing frequency, and eventually became grammaticalized as a (largely) obligatory pronoun under specific binding conditions (Levinson 2000a: 338–59 and references therein; see also Huang, this volume for more on neo-Gricean approaches to anaphora).6

548 Elizabeth Closs Traugott

Despite Levinson’s insights that token meanings develop by pragmatic inferencing into type meanings, and ultimately may become semanticized, and that the “engine” for many of these changes is I-inferencing. (Levinson 2000a: 370), we are left with a problem: that of directionality of change. Honorifics develop out of non-honorifics, not the other way round, and addressee honorifics out of referent honorifics, not the other way round (Brown and Levinson 1987: 276–7). Reflexives arise out of emphatics and not the other way around (Faltz 1989, Kemmer 1993). Subsequent work in a large number of domains has pointed to more and more examples of unidirectionality in semantic change, e.g. from part to whole (Wilkins 1996), from deontic to epistemic (Traugott 1989, Bybee et al. 1994), from conditional to concessive (König 1985), from meanings based in the concrete sociophysical world to those based in the world of the speech event (Traugott 1982, Sweetser 1990). Additional ingredients are therefore needed to account for semantic change.

3 From the Perspective of Language Change

In Horn and Levinson’s work, a historical perspective on pragmatics is called upon primarily as additional support for synchronic analysis. In much recent historical work on semantic change, the role of pragmatics plays center stage. For a historical linguist key questions concern what the “path” of change is, what motivates it, and what the mechanism for the change is.7 The first of these questions has been called the TRANSITION question. Mechanisms and motivations are part of the ACTUATION question (Weinreich et al. 1968).

With respect to mechanisms of change, from the beginning of the twentieth century on, two have been recognized as being of crucial importance in the field of morphosyntax: reanalysis and analogy (see e.g. Meillet 1912, Harris and Campbell 1995); a third is borrowing. Reanalysis modifies underlying representations, whether semantic, syntactic, or morphological. For example, the development of auxiliaries is a case of reanalysis (category, and hence distributional status, changes). In semantics we may think of reanalysis as involving change in the status of implicatures associated with lexemes. Studies of a large number of instances of grammaticalization (e.g. Fleischman 1982 on the development of future markers in Romance languages, Traugott 1982 on the development of connectives in English, Bybee et al. 1994 on the development of tense, aspect, and modality in the languages of the world) and of semantic change independent of grammaticalization (e.g. Traugott 1996, on the development of promise and threaten), led to the hypothesis that the shift in

(3) is one, perhaps the most important, way in which reanalysis operates in semasiological change. By contrast, in the domain of lexicalization by innovative word-formation, this shift does not appear to be a factor. For example, in cases of zero-derivation such as are found in the creation of denominal verbs in English, e.g. calendar (n) to calendar “enter in one’s calendar” (v), or in the

Historical Pragmatics 549

creation of deverbal nominals in Latin, e.g. praelud- (v) “play before” praelud-

(n) “prelude,” the reanalysis occurs instantaneously, with entirely predictable semantic consequences for the new word formation in question (a denominalized verb acquires event-structure, for example). Word formation by overt derivation, such as driller and cooker, is likewise instantaneous. Such changes may have consequences for the semantics of the root form and certainly for newly contrasting members of the same lexical field, but (3) is not at issue.

With respect to motivations, two are most frequently mentioned in the literature: language acquisition and strategic interaction by speakers/writers (SP/Ws) and addressees/readers (AD/Rs) in the dyadic speech event. Innovation of a new meaning, once replicated by the innovator, by definition involves acquisition of the innovation. Replication by others, i.e. spread to others, also by definition involves acquisition. In this process of acquisition, the prime type of reasoning has been shown to be ABDUCTION (see Andersen 1973, Anttila 1992, Levinson 2000a, Hobbs this volume). Abduction is “common-sense reasoning” by which language acquirers infer something to be the case, based on observation of data and invoking a regularity, e.g. hearing cooker, and knowing that in English -er usually marks an animate agent, the hearer might conclude that it meant “one who cooks” until further evidence showed that cook already exists in that meaning. While much of the generative literature on syntactic and phonological change has assumed that restructuring in grammar occurs as a result of child language learning (e.g. Halle 1964, Lightfoot 1979 et passim), there is growing evidence that attention needs to be paid to language learners of any age (see e.g. Andersen 1973, Ravid 1995, Eckert 1999). When the data for historical linguistics comes from periods prior to recording, claims about acquisition can be only speculative, since it is impossible to access evidence of child (or adult) language acquisition in any detail from written records of the past. In the case of historical pragmatics, we can only say that we are coming to understand that children pay attention to pragmatic factors from a very early age (e.g. E. Clark 1999, 2002a), but they are of course also learning which strategies to pay attention to (Slobin 1994). To date we regrettably have no firm empirical evidence for the extent to which small children, young adults, or older adults actually bring change about through innovations motivated by inferencing.

We can, however, observe strategic interaction by the SP/W–AD/R dyad in various discourse situations, such as conversations, legal and medical interviews, etc. Since we have written textual evidence for a variety of genres, including interactive genres as represented by drama, we have more adequate evidence for motivations in this area than in acquisition. By hypothesis, SP/ Ws and AD/Rs interacted in similar ways drawing on similar heuristics across time, at least within the 3,000-year history of languages known to us (such as Indo-European, Chinese, and Japanese). We assume therefore that interlocutors draw and have drawn on a number of cognitive abilities. One is the ability to maximize Relation/Relevance to that situation. Another is the ability to use various kinds of rhetorical strategies like METAPHOR, METONYMY, and SYNECDOCHE8

550 Elizabeth Closs Traugott

(see Nerlich and Clarke 1992, 1999). Metaphor can be thought of as a cognitive process involving associations mainly of the “equivalence” or “paradigmatic” type: “a mapping of a domain onto another domain, both being conventionally and consciously classified as separate domains” (Barcelona 2000: 9), e.g. body part, space and time, emotion. Metonymy can be thought of as a cognitive process in which “one conceptual entity . . . provides access to another conceptual entity . . . within the same domain” (Kövecses and Radden 1998: 38). While metonymy has until recently been given little attention, its importance is coming to be increasingly recognized as “probably even more basic [than metaphor] to cognition” (Barcelona 2000: 4). Metonymies of this kind are typically conceptual and can be construed as including the implicatures associated with linguistic expressions (Traugott and König 1991). The existence at a moment in time of culturally accepted and salient metonymies and metaphors, as well as cognitively more general ones without doubt both motivates and constrains change. However, the actual process of the development of conventionalized metonymies and metaphors (“metaphorization” and “metonymization”) is crucially tied up with the mechanisms of reanalysis in the case of metonymization, and with analogy in the case of metaphorization. Anttila (1992) usefully pairs metonymy with indexicality and syntagmatic relationships, metaphor with iconicity and paradigmatic relationships.9 Here I discuss only the kinds of semantic changes that arise out of regularly recurring metonymic inferencing in the syntagmatic flow of speech.

One of the strongest metonymies is attraction of meaning to the metalinguistic domain of the SP/W–AD/R dyad, a process that has been variously identified as the development of “expressive,” “subjective,” “metatextual,” “procedural,” or “speech act” meanings over time (e.g. Traugott 1982, 1989, Sweetser 1990, Nicolle 1998); as a historical process deriving from pragmatic inferencing it is widely known as SUBJECTICATION (e.g. papers in Stein and Wright 1995). Subjectivity as a synchronic (and partially diachronic) phenomenon plays a role in Bréal’s work (1900: Chap. 25) and was later elaborated on in several works including Kuroda’s (1973) classic study of Japanese expressions of physical sensation. The concept is, however, most frequently associated with Benveniste, who distinguished “sujet d’énoncé”/“syntactic subject” and “sujet d’énonciation”/“speaking subject” (Benveniste 1958). In Lyons’s words: “The term subjectivity refers to the way in which natural languages, in their structure and their normal manner of operation, provide for the locutionary agent’s expression of himself and his own attitudes and beliefs” (Lyons 1982: 102). This “expression of self” may be instantiated lexically or grammatically, for example, by deictics, performative uses of speech act verbs, choice of aspect, or of discourse markers like after all.

As a historical pragmatic phenomenon, subjectification is the mechanism whereby meanings tend to become increasingly based in the SP/W’s subjective belief state or attitude toward what is being said and how it is being said.10 It is metonymic to the SP/W–AD/R dyad, in other words, to the communicative situation, and strongly suggests that R-based inferencing (using the heuristic

Historical Pragmatics 551

of Relation/Relevance, not just Informativeness) is crucial to semantic change. On the assumption that the basic problem of communication is the mismatch between speaker’s thinking and articulation of that thought (speaking), then SP/W has a significantly more difficult task than AD/R. This task will encourage exploitation of R heuristics as meanings are used in novel utterances, since it minimizes production effort (“Say no more than you must,” in Horn’s terms). Since the pragmatic cues given and inferred are contextually negotiated in terms of “the accepted purpose or direction of the talk exchange” (Grice 1989: 26), the R relation will be construed with reference to the speech situation and the dyad in it. AD/R is not excessively burdened by the innovation and the violation of M heuristics, because experimental innovations on the semasiological dimension are typically introduced in ways that are harmonic with other meanings already available in the context, and are therefore informationally minimally problematic. This will be demonstrated in section 4. If AD/ R does not interpret the cue or interprets but does not replicate it, then the experiment will remain just that. If AD/R does interpret it and replicate it, then he or she will do so as SP/W, and over time relatively normative and stable generalized implicatures may arise that are enrichments of earlier meanings.

Once a lexeme has undergone subjectification it may be further grounded in the communicative dyad and intersubjectification may occur. That is, meanings may develop that encode intersubjectivity: awareness of each participant by the other (see Benveniste 1958). As a historical mechanism, intersubjectification motivates the semasiological shift of meanings over time to encode or externalize implicatures regarding SP/W’s attention to the “self” of AD/R. Although honorifics might seem to exemplify intersubjectification only, in fact, the changes in meaning that a lexeme or construction undergoes if it is pre-empted to honorific function always involve some degree of subjectification. This is because honorifics index social status from the point of view of SP/W. One example of such a shift is pre-tenth-century Old Japanese non-honorific samoraFu “wait on/for, be in attendance” eleventh-century humiliative referent honorific saburahu “HUMIL-be” (indexing SP/Ws assessment that the subject referent is lower in social status than some other referent) thirteenthcentury addressee honorific “polite” saburahu/soorau, indexing AD as being socially superior to SP (Traugott and Dasher 2002: §6.5.1).

Some of the changes outlined here involve primarily social-cultural motivation, for example the kinds of euphemisms that Horn and Levinson discuss; examples tend to be nominal, or to involve nominals (e.g. go to the bathroom), but some are verbal (e.g. sleep with) (see Horn 2000a); all have clearly referential properties. Many other changes involve primarily linguistic motivation, e.g. space to time, temporal to causal, conditional to concessive, deontic to epistemic; examples tend to be expressed by demonstratives, verbs, adverbials or connectives, and to defy referential interpretation. In so far as subjectification and intersubjectification involve the SP/W–AD/R dyad and crucially influence interpretation of deixis, modals, etc., they are on the interface between social

552 Elizabeth Closs Traugott

and grammatical meaning. It is therefore not the case that R-based implicatures “typically involve social rather than purely linguistic motivation,” as suggested by Horn (1996a: 313; see also Levinson 2000a on I implicatures), except in the broadest sense that all implicatures arise in the speech event situation, which is a social phenomenon. Most R-based implicatures involve both social and linguistic motivation.

3.1The Invited Inference Theory of Semantic Change

An attempt to provide a fuller account of the pragmatic factors involved in semantic change than is proposed in the work discussed in section 2 is the Invited Inferencing Theory of Semantic Change (IITSC) (e.g. Traugott 1999, Traugott and Dasher 2002). The term INVITED INFERENCE is borrowed from Geis and Zwicky (1971) but has been extended to conversational implicatures; it is not considered to be a “special class of ‘implicatures’” distinct from conversational implicatures as in that work (1971: 565). “Invited inferences” is preferred over “implicatures” in that it highlights the dual role of SP/Ws and AD/Rs in the dyadic speech event: SP/Ws strategically use implicatures and invite AD/Rs to infer a meaning. Therefore, in this model PCIs are renamed Invited Inferences (IINs), and GCIs are renamed “Generalized Invited Inferences” (GIINs). (3) is restated as (4):

(4) IIN GIIN SM

Implicatures are considered to be R-based rather than I-based, because of the widespread development of meanings based in the interactive dyad, not exclusively in information. They are also assumed to be motivated largely by linguistic rather than social factors.

The IITSC focuses on schemas that represent types of semasiological reanalysis that language-specific lexemes may (but do not have to) undergo, constrained by larger cross-linguistic and onomasiological conceptual categories such as causal, conditional, future epistemic, animate, etc. It also focuses on the way in which stereotypes emerge, a perspective missing in most work discussed in section 2. Wherever semantic change is of the type in (4), the IIN is by definition not (yet) stereotypic. As IINs become more salient in the community, i.e. as they become GIINs, the stereotype is actually being created for the form with which they are associated; it does not pre-exist. For example, while it may be that inferences from emphatic -self could be made to anaphoric identity with a prior NP, so long as this inference was unusual and only contextually accessible, it was not yet a stereotype; the stereotype inference to the anaphor developed over time.

In the IITSC the assumption is that every innovation at the GIIN stage (if not before) has the potential for violating the Manner maxim “be orderly,” insofar as that may be interpreted as pertaining to maintenance of norms. As we shall see in the case study below, most GIINs become crystallized in contexts where

Historical Pragmatics 553

the new meaning is redundant with other elements in the sentence, and so violation of “avoid ambiguity” is probably relatively minimal in context, albeit large-seeming out of context. However, when a new polysemy is semanticized, redundancy of context is no longer in play, and ambiguity can occur more readily. In many cases the new polysemy is distributionally different from the old one and a reanalysis has therefore occurred, e.g. epistemic must is available in past tense and stative contexts in a way in which the older deontic must is not, evidently as a sentential epistemic adverbial is available in clause-initial position which the older manner adverbial (meaning “clearly”) is not. These new polysemies, particularly if they are modal or connective, may be used to signal that there is some doubt is being cast on the proposition, and so can function to warn that a special interpretation is necessary, i.e. as violations of “avoid prolixity.”

4 A Case Study: after all11

I turn now to a brief case study exemplifying an approach to historical pragmatics based on corpora: the development of after all. A methodologically useful way to begin such a study is to investigate the synchronic situation after a set of polysemies has come into existence and then to seek evidence in a historical corpus for the transition from an earlier stage to the synchronic one studied, all the while watching for developments that may have dead-ended. Such dead-ended developments frequently show incipient conventionalization of a meaning that is not replicated, or not replicated for any considerable length of time, whereas it may become a highly salient meaning in another language. For example, there is some sporadic evidence of incipient “because”- meanings for while in Middle English; this IIN never became a GIIN in English, but it did for the cognate weil in German.

Since most of the material for historical linguistics is written, it is important to use corpora in the same mode, if not genre. This means that contemporary spoken language can be used as a comparison only with spoken material from an earlier era; and constructed data may be helpful in highlighting areas of inquiry but they need to be used with great caution since they cannot be verified by native speakers in earlier periods.

After all has been discussed in terms of Relevance Theory by Blakemore (1987, this volume), Blass (1990), Carston (1993), and others as a prime example of a non-truth-conditional PROCEDURAL MARKER. It has also been variously called a “discourse marker” (Schiffrin 1987), “pragmatic marker” (Fraser 1996), “connector,” “(discourse) connective,” and in much European work “argumentative operator” (e.g. Jakobson 1957, Ducrot 1972, Anscombre and Ducrot 1989). Using constructed examples of the type in (5):

(5)a. He is brave; he is after all an Englishman.

b.Tom has left. After all, his wife is not here.

554 Elizabeth Closs Traugott

Blakemore proposes that after all in p after all q constructions provides evidence to the hearer for the truth of p and serves as a reminder of q (Blakemore 1987: 81–2). Blass adds that q is known to the hearer (Blass 1990: 129). These accounts assume unitary meanings that are contextually interpreted. They are also primarily MONOLOGIC, in that they assume perspectival continuity between the clauses represented by p and q. Other approaches have, however, been very different. Roulet (1990), for example, highlights the multiplicity of uses of French après tout and its English counterpart after all (he says they behave identically), as a concessive (“nevertheless”), and as a connective meaning “of course” in some contexts, “because” in others. He points out that it is often used “dans une contexte refutative” (“in a refutational context”) (Roulet 1990: 342). On this account adverbial after all is strongly “dialogic” and “polyphonic” in some of its uses: it involves “crystallized” dialogues, and “incompatible viewpoints” (Nølke 1992: 191, Schwenter 1999a).12

The following account of the history of after all builds on these prior analyses, especially Roulet’s, to suggest a methodology for text-based historical pragmatics.

4.1The present-day situation

The following account of contemporary after all is based on a computerized corpus of top stories issued by United Press International (UPI) in the years 1990–2; these stories contain quotations of speeches and also newspaper reporting. In this corpus there are 45 tokens of after all; the usages appear to be entirely characteristic of American English.

Of the 45 tokens of after all, 10 involve the preposition after followed by all NP; together they form a temporal adverbial construction, as in after all these years, which is to be understood literally and compositionally.

Four tokens have connective adverbial properties that are epistemic, that is, they concern SP/W’s assessment of the truth of the proposition that follows (q). These four tokens are oriented away from some aspect of the prior discourse (p) and are therefore adversative. Two of them are concessive adverbials meaning “however, nonetheless, despite what was expected” by some locutor, who could be a third person, SP/W in another (dialectic) role, or (hypothetically) AD/R, as in (6):

(6)The federal fund that finances presidential campaigns should have enough money to pay for the 1992 race after all, but by 1996 the coffer could be depleted, a new report said Wednesday.

The Federal Election Commission issued a new projection on how much money will be available to finance the 1992 primaries . . . that is more optimistic than some previous estimates. (UPI, August 14, 1991)

This example, which occurs at the beginning of the news story, has no antecedent p, but after all evokes one (there won’t be enough money to pay for the 1992

Historical Pragmatics 555

race), and confirms it in the second paragraph. The concessive introduces an argument q (the proposition), and invites the conclusion r that p (which is implicit in this case) is not true, i.e. it is backward-looking and epistemic.

There are two other epistemic examples that share some of the features of (6), but they are not restricted to final position. Like the concessive, they signal epistemic counter-expectation, but they are more complex (both positive and negative aspects of p are brought into focus), and the logical connection between p and q is less obvious. An example is (7):

(7)Grodin [an actor] also captures the movie star’s obsessive concern with his image, which is, after all, what he has to sell. (UPI, June 19, 1990)

In (7) q (his image is what he has to sell) must be understood to be not only plausible but acceptable in a social world of norms. Rather than involving “locutor or addressee did not believe/expect q to be true,” after all invokes “locutor or addressee might not have thought of q because of some aspect a of p (obsessive).” On this view, after all marks q as an argument to the conclusion r: “so you shouldn’t be concerned about the star’s obsessive worries”. This use of after all is often associated with an implicature of reminder, because appeal is made to obvious, interpersonally recoverable, societal norms or known situations.

By far the most common use in contemporary American English of after all is as a different kind of adverbial connective: one in which the conclusion is oriented toward p. This is a true discourse marker (DM) in the sense of Fraser (1988), a “deictic discourse marker” (Schiffrin 1990) that is metatextual in function and signals the speaker’s discourse strategy. It serves as “a comment specifying the type of sequential discourse relationship that holds between the current utterance – the utterance of which the discourse marker is a part – and the prior discourse” (Fraser 1988: 21–2). The purpose of such markers is to “combine viewpoints into structure” (Nølke 1992: 197) and to express the discourse relationship intended by the speaker between two utterances. While richer than the constructed example (5a), this type of after all has some of its flavor. (There are no examples resembling (5b)).

There are 26 examples in the corpus of uses like (8):

(8)The notion that bombs can strike military targets without killing and maiming innocent women and children is absurd. If there is war, the Iraqi dictator can be expected to employ chemical weaponry. After all, he killed his own people, including Kurdish women and children, with poison gas. (UPI, October 8, 1990)

This kind of DM use has been said (see Blakemore 1987, Blass 1990) to serve to signal that q is a justification for p: the speaker’s reason for saying p. Although partially correct, this characterization misses an important point. The justification is actually not of p, but of some particular expression in the prior discourse

556 Elizabeth Closs Traugott

of the speaker’s subjective attitude (typically expressed by negation, epistemic modality, or an evaluative lexeme). In (8) after all points back anaphorically to the evaluative lexeme absurd, as well as the modal can be expected, at the same time as signaling cataphorically that q is a justification of those two evaluations. Thus while DM after all has some epistemic properties (it signals SP/ W’s belief) and adversative properties (it rejects aspects of p), it is different from epistemic adversative after all in that it is anaphoric as well as cataphoric, and in that it elaborates on the attitudinal components of p. Also, although DM after all may conversationally imply that information in q is known to the addressee, the strongest invited inference is “You might have thought of this in relation to p.”

An additional characteristic of DM after all is that it is a disjunct and follows an intonational break (reflected by a comma in writing). q is therefore syntactically paratactic, and pragmatically an “add-on,” conveying a sense of superiority and condescension, even dismissiveness or verbal play on SP/W’s part, either to the point of view expressed in p (minus the evaluation to be justified), or to AD/R (“Despite the fact that you can’t put these two things together, I can!” and “Aren’t I being funny/clever/cute?”), or both. These are among factors that differentiate it from because and since, which can be substituted for it but which convey none of the subjective attitude or metatextual stance of after all.

4.2The diachronic development

I turn now to the question of how these uses came into being (for related studies see Traugott 1989, Powell 1992, Brinton 1996, Rossari 1996, Jucker 1997 among others).

After was a spatial (“behind”) and temporal (“later”) preposition in Old English. In the context of certain nouns it could indicate “in conformity with” as in example (9) from Middle English:

(9)Hou that men schal the wordes pike after the forme of eloquence

“How people shall pick the words according to the form of eloquence” (c.1393 Gower, Confessio Amantis 4.2651 [MED after 8])

The first example in my database of after all as an adverbial is from the sixteenth century and it is temporal:

(10)[about the funeral of the Bishop of Winchester] and my lord bysshope Bonar of London did syng masse of requiem, and doctur Whyt bysshope of Lynkolne dyd pryche at the sam masse; and after all they whent to his plasse to dener.

“and my lord Bishop Bonar of London sang a requiem mass, and Dr. White, Bishop of Lincoln, preached that Mass; and after it was all over they went to his place to dinner.” (c.1560 Machyn, Diary, p. 97)

Historical Pragmatics 557

This non-adversative meaning of “finally,” “on top of all that” occurs sporadically in the data, with reference not only to time but to sequences of argument, but is always preceded by and. There is an implicature of normative sequence or “natural order,” perhaps derived or at least supported by and. A striking early example of after all being used in reference to a sequence of arguments comes from a long “apology” by Dryden concerning a play that he feels was misunderstood. He shows how earlier criticism of the play had been addressed, and goes on to say that the most recent charge is that one of the characters, Almanzor, “performs impossibilities.” After a summary of the plot Dryden finally claims:

(11)but, ’tis far from being impossible. Their King had made himself contemptible to his people, as the History of Granada tells us. And . . . And, after all, the greatness of the enterprise consisted only in the daring: for, he had the King’s guards to second him. (1672 Dryden, Dedication,

Conquest of Granada, p. 30)

Here and, after all, can only mean “and in the end” (in terms of the plot) or “finally” (in terms of Dryden’s argument). There may be a conversational implicature of adversativity from after all, but it is derived largely from the general discourse strategy and is an IIN from the argument that “’tis far from impossible.”

By the beginning of the seventeenth century a few uses can be found of the temporal preposition after followed by NP, in which the sequence of events can be interpreted as implying that the expected order of events has been violated and that the outcome is not normative:

(12)a. [The Duke has disguised himself to test Angelo, and has said treasonable things about himself]

Lucio: Do you remember what you said of the Duke? . . . was the Duke a flesh-monger, a fool, and a coward, as you then reported him to be? . . .

Duke: I protest I love the Duke as I love myself.

Angelo: Hark how the villain would close now, after his treasonable abuses!

(1604 Shakespeare, Measure for Measure, V.i.330)

b.Really, Madam, says Robin, I think ’tis hard you should Question me upon that Head, after all I have said.

(1722 Defoe, Moll Flanders, p. 54)

There is little evidence that the new meanings in (12a) and (12b) were stereotypic at the time, or that there were any significant sociocultural changes in the community to which they could be tied. Rather, the examples in (12)

558 Elizabeth Closs Traugott

appear to have been experiments with IINs arising from the temporal expression. Such IINs (and later GIINs) arise fairly frequently cross-linguistically (see Rudolph 1996). Indeed, the use of after all, with its generalized quantifier all and temporal after has lexical properties in common with two of the major cross-linguistic sources13 for concessives identified by König (1985: 10–11); cf. English al-be-it, although, however, nevertheless. From a temporal perspective, the move is from “at the end” to “in the end”; the quantifier specifies “reexamination” of all that preceded, not just the immediately prior p (Rossari 1994: 20).

Over the course of the seventeenth century, the kind of dialogic, adversative use of after all in (12) develops from a contingent conversational implicature into a generalized one. By the end of the century we find a conventionalized concessive sentential adverbial use as in (13) (see Modern English (6)). It is no longer contextually supported by the construction . . . that is said and done, but rather stands alone in a set with despite everything.

(13)[Mrs. Fainell, Lady Wishfort’s daughter, has been falsely accused]

Mrs. Fainell: I know my own innocence, and dare stand a trial. [Exit] Lady Wishfort: Why, if she should be innocent, if she should be wronged

after all? I don’t know what to think.

(1700 Congreve, The Way of the World, Act V)

This dialogic clause-final concessive meaning became entrenched, and the bare temporal adverbial ceased to be used.

At the same time as the clause-final concessive arose we also find after all used as an epistemic adversative much like adversative, actually, in fact (see (7)). Unlike the clause-final concessive, it is first used in a construction with an adversative connective such as but, as in:

(14)I have not made it my business either to quit or follow any authority in the ensuing discourse: Truth has been my only aim . . . Not that I want [“lack”] a due respect to other men’s opinions; but, after all, the greatest reverence is due to truth.

(1690 Locke, Essay Concerning Human Understanding, bk. 1, chapter 4, section 23)

Here after all is harmonic with, indeed redundant to, but and is strongly motivated by the dialogic context in which opposing positions are presented, including negative propositions (I have not made it my business . . . Not that I want respect). Early examples like (14) are more strongly truth-oriented than many later ones, which are weakened, partly by generalization to nonnegative contexts.

Justification also appears in the seventeenth century, again contextually derived, this time in the context of and or for “because” preceded by a negative as in:

Historical Pragmatics 559

(15)You need not be much concerned at it; for after all, this way of explaining things, as you called it, could never have satisfied any reasonable man.

(1713 Berkeley, Dialogue, 2, p. 210)

When, at the beginning of the eighteenth century, justificational after all can occur clause-initially without a preceding connective, it has become conventionalized.

Blass (1990: 129) suggested that the concessive and justificational after all’s derive from after all that has been said and done. Examples like (12b) show that such constructions are indeed likely sources of the concessive. However, since Blass cites the construction out of context, her proposal gives no insight into the strongly adversative discourse context that motivated the conventionalization of the adverbial. Furthermore, there is no obvious or natural connection in terms of implicatures between the temporal and the justificational meanings of after all. Since justification requires potential adversativity, it is implausible that it derives directly from the temporal construction. Although temporal sequence can give rise to causal conjunctions (cf. since), and, as we have seen, quantified temporals are well known to be the source of concessives, they do not give rise directly to metatextual justification markers. The corpus data shows that adversative after all arises in strongly adversative, largely dialogic contexts. Justificational after all does not emerge until after all had come to be associated with an adversative GIIN.

When we consider the changes outlined here in terms of Horn’s and Levinson’s proposals about pragmatic bases for semantic change, we see that each change involved initial experimentation with IINs in redundant contexts; the hypothesis in (4) is supported. With respect to the differing roles attached to Q, M, and R(= Levinson’s I), we see that R-based enrichment is central, with Relation to the SP/W–AD/R dyad, especially SP/W’s subjective rhetorical purposes motivating the shift to adversative procedural meaning. Insofar as I-based enrichment is “information-based” and not linked to any particular aspect of the communicative interaction, it accounts less well than R-based enrichment for the direction of change evidenced by after all and the many other expressions like it. Each change violates the M heuristic in some respect

– the new meanings engendered competition with extant members of the same conceptual set (e.g. in (13) after all competes with despite everything, in (14) with indeed, actually; and the new syntactic positions in which the expression came to be used violate norms of use. Apparently, however, such M violations (in the sense of violations of norms and orderliness) did not trigger new restrictions on the pre-existing lexemes as the innovations were taking place. Insofar as it is prolix (three syllables long) it certainly is more informative than the but and for with which the adversative and then the justificational after all cooccurred, but its presence does not appear to have triggered new restrictions on them. The eventual loss of clause-final temporal after all does not appear to have been triggered by the M heuristic either: it was lost to the full prepositional phrase with a determiner and noun, e.g. after all these months. Insofar as Levinson’s Q implicatures operate on scalar particles, they are not relevant to

560 Elizabeth Closs Traugott

the development under discussion here. The main principle at work appears to be the principle of contrast that motivates the maintenance of distinctions from other lexical items within the same general semantic space, but also allows some readjustments over time within it.

The historical data show that the contexts in which new uses of old formmeaning pairs arise are clearly linguistic, not primarily “sociocultural”: the new meanings (new token exploitation of implicatures) are reinforced by juxtaposition with connectives that sharply constrain the implicatures. Where individual lexemes or constructions are concerned, implicatures that arise in on-line verbal utterances or writings (IINs) are initially recruited redundantly to extant linguistic contexts for discourse purposes. Once the new usage has been accepted by the community (i.e. has become a GIIN), such redundant cues may be reduced or allowed to be less directly explicit, in which case conventionalization (semanticization) may occur. Absent direct collocation with redundant cues signaled by other connectives or DMs they “actively help to construct that very context” (Hansen 1996: 108); however, as we have seen from the contemporary data on after all, redundant cues tend to be very much part of the fabric of the discourse and of the context in which interpretation is to occur. Eventually some DMs like so, because, indeed have come to serve a whole turn (at least in conversation); those that can do so are relatively old; newer ones cannot do this.14

5 Conclusion

Historical pragmatics can yield results that bear on fundamental claims in pragmatic theory as well as semantic and lexical change. In answer to question

(1) posed in section 2, “Do different conversational maxims motivate different types of semantic change?,” the answer is clearly “Yes.” The R heuristic (and R inferences from it) accounts for most regularly attested semantic change, in the sense outlined here, excluding word-formation. The Q and M heuristics account primarily for constraints on word formation and realignments of meanings among lexical items, for example in the domain of Aristotelian squares. With respect to question (2), “Does Grice’s distinction between particularized and generalized conversational implicatures help account for how semantic change occurs?,” again the answer is “Yes.” Particularized conversational implicatures are factors in semasiological change only if they become generalized; only generalized implicatures play a role in regular semasiological change. Finally, with respect to question (3), “Are there additional important factors that need to be considered in accounting for frequently observed types of semantic change?,” once more the answer is “Yes.” R inferences to the SP/W–AD/R dyad in the speech event are prime motivators of change. Furthermore, historical pragmatics requires going beyond decontextualized examples of semantic change, and paying attention to the discourse contexts in which the changes occur. To achieve a full picture, a macro-approach taking external factors such as cultural changes into consideration is, of course, also necessary.

Historical Pragmatics 561

NOTES

1Thanks to David Beaver and members of his Winter 2001 seminar on radical pragmatics at Stanford for discussion of some of the points in this paper, and to Eve V. Clark and David Beaver for comments on an earlier draft. Larry Horn made many helpful suggestions.

2Recent broad-based frameworks for usage-based approaches to language change can be found in Keller (1994) and Croft (2000); however, neither focuses on historical pragmatics.

3See also Keller (1994) for a proposed modified set of maxims of action

to account for language change.

4Except perhaps in a very strict truth-conditional sense of denotational synonymy that is rarely in effect in language use.

5Levinson (2000a: 41) provides a useful table summarizing corresponding terminologies in Grice, Horn, and Levinson.

6In a detailed study of the development of -self reflexives in the history of English, with attention to anti-synonymy, Keenan (to appear) suggests that Levinson’s assumption that non-marking of pronouns will necessarily implicate disjoint reference (2000a: 348–9) is questionable for Old English. Reflexive marking may have been an opportunistic reanalysis based on the emphatic once the latter had spread to non-subject arguments.

7“Path” or “trajectory” of change is only a metaphor for the linguist’s analysis, a metalinguistic way of

referring to changes of the type A A ~ B (B), not to any kind of hard-wiring in the brain.

8Nerlich and Clarke (1999) define “synecdoche” in terms of moving up and down taxonomic space, i.e. as the issue of change of hyponym to super-category (hyperonym) status and vice versa.

9Despite the differences between them, however, metaphor and metonymy sometimes blend; see Goossens (1995).

10“Subjectification” has been construed in another line of research, associated with Langacker (1990, 1999), as perspectival shifts from a “syntactic subject” to “speaking subject,” as evidenced by the development of raising constructions from control verbs such as promise, be going to do X

(directional movement for a purpose).

11An earlier version of this section was presented at the 10th International Congress of Linguistics, Paris, July 1997. Thanks to Regina Blass and Scott Schwenter for comments on that paper.

12For the distinction between “dialogic” and “monologic” uses of language and approaches to them, see Bakhtin (1981), Ducrot (1984), Roulet (1984) among others.

13The others are: conditionals (Gm. ob-gleich), expressions of co-existence or similarity in space or time, (Fr. tout de même, Turk. bununla beraber

“together with this”), or obstinacy, spite (de-spite, Arab. ragman from “compel”).

14This shows that DMs like so, therefore, whereas that can initiate discourses should not be considered of a different type from those that cannot, contra e.g. Rouchota (1996).