2.1.1 The inferential pragmatics research paradigm

Grice's original insight about the crucial role of inferencing for conveying the speaker's intended message (at the expense of encoding) has been adopted by all inferential pragmatics theories, and inferencing is now seen as a major ingredient in utterance interpretation. In fact, whenever the linguist is debating whether to assign semantic or pragmatic status to some interpretation, the default preference is to treat it as pragmatically inferred. Following Grice's (Reference Grice1989: Chapter 3) Modified Occam's Razor Principle, it is more economical to assign a given interpretation to pragmatics than to semantics, since the former comes “for free”, speakers performing pragmatic inferences anyway. The latter requires grammatical stipulation, which is supposed to burden native speakers.

Inferential pragmatists leave formal grammar mostly intact, except for some “territory seizures”, where some potentially semantic accounts are relegated to pragmatics. As empirical basis for the analyses offered, inferential pragmatists traditionally, and very often even today, rely on their own intuitions, although recently, some corpus studies (Ariel, Reference Ariel2004) and some experimentation (Gibbs and Moise, Reference Gibbs and Moise1997; Noveck, Reference Noveck2001; Noveck and Chevaux, Reference Noveck, Chevaux and Skarabela2002; Papafragou and Musolino, Reference Papafragou and Musolino2003) have been adduced in support of researchers’ positions. The three variant inferential pragmatics theories proposed by Grice, neo-Griceans, and Relevance theoreticians are quite well known. Given the commonalities outlined above, emphasis will here be laid on the differences between them.

We start with Grice. Grice's theory simultaneously addresses two fundamental problems about linguistic interactions. The first one is, how can we distinguish between a natural and an unnatural discourse? In other words, what characterizes coherent discourse progression? Grice's answer is that interlocutors first and foremost abide by the Cooperative Principle, which dictates to them to “[m]ake your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged” (Grice, Reference Grice, Cole and Morgan1975: 45). Presuming that this agreement is adhered to, the interlocutors can further assume that four maxims and eight sub-maxims inform speakers’ utterances, and hence, addressees’ interpretations: Quantity (informativity), Quality (truth and reliability), Relation (relevance), and Manner (optimal choice of linguistic form).

The second question is, how can speakers convey more than they explicitly encode? Grice proposed that interlocutors need not always straightforwardly obey the maxims in order to be cooperative. He offered the mechanism of conversational implicature generation to account for such cases. The idea is that speakers guide their addressees to extract more meaning out of their utterances than they actually encode. Taking into consideration the encoded message, the maxims, as well as relevant contextual background, addressees must “read between the lines” in order to process the additional inferred meanings intended by the speaker. These conversational implicatures may be generated in order to make sure that some maxim is abided by, or in order to justify an apparent maxim violation.

The neo-Griceans (most notably, Horn, Reference Horn and Schiffrin1984b, Reference Horn1989; Levinson, Reference Levinson, Verschueren and Bertuccelli-Papi1987, Reference Levinson2000) accept the Gricean picture for the most part, but they don't have use for so many maxims and sub-maxims. They leave Quality intact, Relation is kept by Levinson but rejected by Horn. The neo-Griceans’ attention is mainly focused on reducing the six sub-maxims subsumed under Quantity and Manner. Horn proposes two principles instead, based on Grice's Quantity. Whereas Grice saw no particular problem in balancing between his Q₁ and Q₂, Horn and Levinson see the two as clashing, one instructing the speaker to be maximally informative, thereby blocking potentially inferable enrichments, the other instructing the speaker to be minimally informative, thereby encouraging the addressee to enrich the speaker's linguistic meaning with pragmatic inferences. They then explain how interlocutors resolve this clash, by reference to markedness: “The use of a marked expression when a corresponding unmarked alternate expression is available tends to be interpreted as a marked message (one which the unmarked alternative would not or could not have conveyed)” (Horn, Reference Horn and Schiffrin1984b: 22). In other words, enrichment to the stereotype (a derivative of Q₂) applies when unmarked forms are involved, and anti-enrichment (a derivative of Q₁) applies when marked forms are used.

Sperber and Wilson (Reference Sperber and Wilson1986) are even more reductive than the neo-Griceans. They have suggested that a single principle of optimal relevance can account for discourse coherence, as well as for how addressees go about reading more into speakers’ utterances. According to this theory, Relevant information necessarily hooks up with addressees’ contextual assumptions, yielding contextual implications based on the two computed together. Speakers aim for an adequate quantity of contextual implications (pragmatic inferences), which must be conveyed in a way that requires the least processing cost from the addressee.

In this way all three theories account for how discourse is coherent, and what triggers addressees’ inferences. Interestingly, while the theories are quite different, for the most part there is agreement between them on where to draw the grammar/pragmatics division of labor for specific linguistic phenomena. For example, they all agree that, e.g., and's semantic meaning is equivalent to its logical counterpart ∧, and that the additional interpretations often associated with it (e.g., ‘and then’, ‘and therefore’, etc.) are pragmatically derived. The same is true for (either) or and for scalar expressions (such as some, good). The semantic meanings of the latter are lower-bounded only (e.g., ‘at least some/good’), the upper bound (‘not all’/‘not excellent’) pragmatically provided. However, there can, theoretically, be disagreements within this inferential pragmatics research paradigm regarding the grammar/pragmatics division of labor.⁴ Moreover, the fact that we mostly find agreement regarding the semantics/pragmatics division of labor does not mean that the analyses are otherwise identical. Quite the contrary. (Neo-) Gricean and Relevance-theoretic theories classify pragmatic interpretations somewhat differently.

While there is universal agreement that PCIs play a major role in natural discourse, Sperber and Wilson have proposed an additional distinction between implicated assumptions and implicated conclusions. The former are assumptions the speaker intends the addressee to draw on because they are needed for deriving the implicated conclusions. Only the latter are actually a speaker's point in the discourse. Next, (neo-)Griceans assume a special status for conventional implicatures. These are like conversational implicatures in that they don't affect the truth conditions of the proposition they are in, but like semantic meanings, they are conventionally encoded for specific expressions. (Neo-)Griceans here prefer truth-conditionality as the criterion for the grammar/pragmatics divide. Since Relevance theoreticians (see Blakemore, Reference Blakemore1987 and onwards about but) take the code/inference distinction alone as the criterion distinguishing semantics from pragmatics, conventional implicatures fall squarely on the semantics side of the divide. In other words, there is no place for conventional implicatures as a distinct species of pragmatic interpretation under Relevance theory, since no inference is involved.

Last, neo-Griceans, much more than Grice himself, focus their attention on generalized conversational implicatures (GCIs). These are conversational implicatures, except that they are generated under normal circumstances (as a default, according to Levinson). It is these implicatures which under some contextual circumstances “intrude” on semantics, in that they contribute to the truth-conditional meaning of the relevant proposition, despite the fact that they are pragmatic implicatures. Relevance theoreticians have argued against assuming this additional type of conversational implicature. According to Carston (Reference Carston1990 and onwards), for example, to the extent that GCIs are conversational implicatures, they should be seen as PCIs (which happen to be generated rather often). But for many inferences treated as GCIs by the neo-Griceans, Relevance theoreticians have proposed a different status, that of explicated inferences, i.e., inferences which form part of the Relevance-theoretic concept of explicature.⁵

Relevance theoreticians are not surprised that pragmatics intrudes on semantics. They take it as given that pragmatic inferences contribute to the truth-conditional content of the proposition. While semanticists and pragmatists are in agreement these days that the linguistically encoded meaning falls short of the meaning actually communicated by the speaker, Griceans have more or less adhered to Grice's original “what is said,” which includes only a minimal quantity of pragmatically inferred interpretations, and they are happy to call this level truth-conditional semantics. Sperber and Wilson disagree. Any pragmatically inferred interpretation is a full-fledged pragmatic inference, which should count as one. The fact that many of these are needed to develop the encoded Logical Form of the speaker's utterance into a complete proposition doesn't alter their pragmatic status. Instead of the minimally enriched “what is said,” Sperber and Wilson offer the pragmatically richer explicature (the idea being that the pragmatic inferences needed to complete the speaker's encoded meaning into a full proposition count as part of the explicit message). Now, these differences mean that pragmatic interpretations associated with the same linguistic expressions may well receive different analyses by proponents of different inferential pragmatics theories. Indeed, whereas for (neo-)Griceans the pragmatic interpretations commonly associated with and are GCIs, they are explicated inferences for Relevance theory. And whereas the upper bound on, e.g., some is a GCI under the neo-Gricean analysis, it is sometimes a PCI, and sometimes part of the explicature for Relevance theoreticians.⁶

In sum, all three inferential pragmatics theories propose some overarching principle(s) in order to account for discourse coherence and for how pragmatic inferences are routinely drawn over and above the speaker's linguistically encoded message. Differences between the theories are found in the number of principles, and in the types of pragmatic interpretations researchers assume in general and for specific cases (and, some). But these don't alter the common focus within this research paradigm: drawing a grammar/pragmatics division of labor between semantic codes and pragmatic inferences.⁷

2.1.2 Inferential pragmatics and reference

Reference, just like any other speaker-intended interpretation, requires both a code and a set of inferences. The common goal shared by inferential pragmatics theories is the attempt to relegate as much as possible to pragmatics. In other words, preference is given to analyses involving small codes and big inferences. The role of the specific referential forms tends to be downplayed, while the role of the contextual inferences is upgraded.

Kempson (Reference Kempson, Newmeyer and Robins1988) is a typical Relevance-based proposal. Referring expressions, Kempson reminds us, have quite a variety of uses. For example, only some are referential. Some are impersonal, some involve bound variable anaphora, others require bridging. Kempson proposes a very “thin” code for all definite noun phrases: “Presumed accessibility.” It is then up to the addressee to determine both the type of reading and the referent intended by the speaker, based on Relevance theory (making sure there are sufficient contextual effects for a minimal processing cost). Wilson (Reference Wilson1992) emphasizes that reference assignment forms part of the overall process of utterance interpretation, which is Relevance-guided. She focuses on cases where more than one NP can serve as antecedent (e.g., Sean Penn_i attacked a photographer_j. The man_i/j was badly hurt). While both interpretations are conceivable, Relevance theory, she proposes, does not force the addressee to consider all options before he can choose the one intended by the speaker (a costly processing procedure). Interlocutors are tuned to specific coherence patterns, and can immediately choose the appropriate overall interpretation (the state here described in the second utterance is the consequence of the event in the first) based on the content of the utterances. This effortlessly points them to the appropriate reference.

Assuming that coreferential readings are more informative than disjoint readings, Levinson (Reference Levinson2000) constructs a Horn scale of the prototypical referring expressions: Lexical NP < Pronoun < Ø (see also Huang, Reference Huang2000). The idea is that less informative forms (on the right) trigger stronger (coreferential) readings (via Q₂ – enrichment to the stereotype). If so, using an informative expression when the grammar would allow for a coreferent reading with a less informative expression triggers a disjoint reading (via Q₁ – the anti-enrichment principle, e.g., He_iwent over there and approached the man_j).

As can be seen, inferential pragmatic analyses naturally assign a primary role to pragmatic inferencing in determining the use and interpretation of referring expressions. While Levinson does draw some formal distinctions between referring expressions, a much finer set of distinctions will be offered by form/function pragmatics theories (see section 2.2.2).

2.2.1 The form/function pragmatics research paradigm

Form/function pragmatics is concerned with a small subset of pragmatic meanings, those conventionally associated with specific linguistic expressions (constructions and discourse markers for the most part). Since these meanings are conventional, the pragmatic meanings offered are claimed to belong in grammar. Most, though not all, researchers use naturally occurring examples, but hardly any experimentation has been used to support pragmatic form/function claims.⁸

Here is one example. Originally, the generative account needed to stipulate that the subject (X) “removed” from sentence initial position in existential sentences (there is an X…) had to be an indefinite NP. This grammatical condition, however, is lacking in both descriptive and explanatory adequacies, it was argued, in the jargon of the period. The missing explanatory generalization is that existential constructions serve a specific discourse function, to introduce new entities into the discourse. Since indefinite NPs stand for new entities, no wonder this restriction applies to existential sentences. Hence, what seems to be an arbitrary grammatical stipulation turns out to be pragmatically motivated. Moreover, once the grammatical restriction (against definites) was replaced with a functional restriction (against Given entities), formally definite NPs no longer constitute counterexamples to the account, provided they are not Given (Rando and Napoli, Reference Rando and Napoli1978).

Now, grammar is responsible for form/function correlations, and the analyses here concerned clearly associate specific forms with specific functions. Why should this research be considered pragmatics then? Different researchers provide different reasons. The dilemma here is identical to the one concerning conventional implicatures – in fact, there is no reason not to view the form/function correlations here concerned as conventional implicatures. For some form/function pragmatists, their research is pragmatic, because the functions associated with the syntactic constructions and discourse markers all concern interpretations that do not contribute truth-conditional elements to the proposition expressed. In other words, if we were to present new information where old information is called for, or vice versa, the resulting sentence is not taken to convey a false proposition, nor is it an ungrammatical sentence. Since, especially early on, many pragmatists subscribed to the truth-conditionality criterion as distinguishing semantics from pragmatics, the functions associated with syntactic constructions were automatically classified as pragmatic (see Brinton, Reference Brinton2008 for a similar assumption). Later on, in view of the inner contradictions between various definitions for pragmatics, pragmatists tended to give up on a coherent definition for the field, and such research was seen as part of pragmatics, simply because a canonical list of pragmatics topics had been established, and functional syntax formed a (marginal) member on this list. Thus, for the most part, functional syntax, and certainly research on discourse markers, is considered pragmatic, rather than semantic.

At the same time, some researchers insisted that the form/function correlations they were analyzing were conventional, and hence part of grammar. Prince (most explicitly in Reference Prince, Newmeyer and Robins1988) treated the non-truth-conditional interpretations she was analyzing as grammatical, because they directly and conventionally associated specific linguistic forms with specific interpretations and/or use conditions.⁹ Now, although for Prince form/function pragmatics forms part of grammar, she assigns it to a special, discourse component within grammar. Blakemore took one step further, and treated conventional implicatures as part of semantics proper (see Ariel, Reference Ariel2010: Chapter 8 for arguments for and against this position). So, while the status of form/function pragmatic meanings within grammar is not settled, there is no doubt that the conventions involved (or at least a subset of these) are encoded for specific forms, and hence fall on the grammar side of the grammar/pragmatics divide.

Most classical form/function pragmatic analyses leave formal grammar intact. Rarely do they appropriate a grammatical phenomenon as pragmatic (as in the case of the definiteness restriction in existential constructions). Traditionally, attention was focused on so-called optional choices between variant forms (paraphrastic utterances). The argument made was that speakers’ preference for one rather than another of these syntactic paraphrases (or for using or not using some discourse marker) was not random: “every contrast a language permits to survive is relevant, some time or other” (Bolinger, Reference Bolinger1972: 71). Formal grammar cannot offer an account for such choices between semantic paraphrases (nor between zero discourse markers and explicit discourse markers), but form/function pragmatics can. In fact, what seems an “optional” choice between “free variants” in terms of grammar turns out to be a choice informed by form/function conventions pertaining to non-truth-conditional meanings and/or use conditions. Recently, there is a new line of research which incorporates many of the insights introduced by form/function pragmatists, but sees no reason whatsoever to (a) distinguish between + and – truth-conditional meanings and (b) accept formal syntactic analyses: Cognitive Linguistics (Lakoff, G., Reference Lakoff1987; Langacker, Reference Langacker1987, Reference Langacker1991) and Construction Grammar (Goldberg, Reference Goldberg1995). On these approaches, form/function correlations are part of grammar, no matter what their nature is.

A typical form/function pragmatics analysis is Prince (Reference Prince1978), where she not only distinguishes between unmarked non-clefted constructions and cleft sentences, she also points to the different discourse functions associated with the two cleft sentences in English. The presupposed component in it-clefts, she argued, introduces Given information in general. But the presupposed component in wh-clefts refers to a subset of the information Given to the addressee, that which can cooperatively be assumed to be currently accessible to him. In Ariel (Reference Ariel1983) I analyzed a specific appositive in Hebrew, which is dedicated to the introduction of (very) important persons, and Ward (Reference Ward1990) argued that English VP preposing is restricted to cases where the speaker intends to affirm her commitment to a proposition which has recently been evoked in the discourse. In each of these cases a specific discourse function is associated with a specific syntactic construction.

But these are marked constructions. Most constructions do not manifest a one-to-one form/function correspondence. In fact, even Prince analyzed a few functions for it-clefts, as did Ziv (Reference Ziv1982) for existential constructions (and see especially Kuzar, Reference Kuzar2009; Lakoff, G., Reference Lakoff1987). The idea is that each construction has multiple features, each of which is potentially appropriate for a certain discourse function. For example, Prince found that in addition to stressed-focus it-clefts as characterized above, there are also informative it-clefts, where the presupposed information is not actually Given to the addressee. Rather, it is presented as a fact known to some, knowledgeable people. It is not surprising that it-clefts, rather than wh-clefts, are used for this non-Given function. Note that unlike for wh-clefts, the presupposed information in it-clefts is presented in final sentence position, a position more appropriate for non-Given information. Different sentential positions for certain syntactic roles (non-initial position for subjects for existential constructions), different degrees of verb transitivity and semantic content (existential there takes low-transitivity and low-content verbs), different optional components in the construction (is there or is there not an additional embedded clause attached to the subject NP in the existential clause?) are all potentially mobilized for conveying various discourse functions. Hence the realization that syntactic constructions are not invariably reducible to single pragmatic “sign” functions. Rather, they constitute structure functions which allow for a variety of sign functions to be realized in them (Du Bois, Reference Du Bois and Tomasello2003).

Summing up, form/function pragmatic research associates discourse functions and/or use conditions with linguistically specified forms (morphemes, phrases, or whole syntactic constructions). The difference between classical semantic functions and these mostly information-status conditions is that although they are equally grammaticized, they involve extralinguistic factors which normally do not contribute to the truth-conditions of the proposition expressed. Since researchers within this paradigm are committed to the code/inference division of labor between grammar and pragmatics, these form/function correlations are considered grammatical. The result is that they see the pragmatic interpretations they analyze as separate in nature from other pragmatic interpretations, most notably, from conversational implicatures, analyzed by inferential pragmatics theories. But at the same time, many of these researchers also uphold the conviction that being non-truth-conditional, the interpretations at hand are pragmatic. Hence their self-classification as pragmatists.

2.2.2 Form/function pragmatics and reference

Although form/function pragmatists do appreciate the role of inferencing in referential acts, they see a larger role for the specific referring expressions. Both Ariel (Reference Ariel1990) and Gundel et al. (Reference Gundel, Hedberg and Zacharski1993) have proposed much richer scales of referring expressions than Levinson's, the idea being that there is a conventional form/function association between forms and referential interpretations.¹⁰ I have proposed that referring expressions each specialize for a different degree of accessibility for the mental representation the addressee is to retrieve. The form/function associations are far from arbitrary. Less informative expressions (she vs the woman), less rigid (uniquely referring) expressions (the woman vs Dana), and more phonetically attenuated forms (Ø vs she, USA vs the United States) are used to access relatively highly accessible referents. Conversely, the more informative, rigid, and phonetically large the form (e.g., June, the woman who just walked out, SBC: 008), the less accessible the mental representation is assumed to be.¹¹ Infinitely many referring expressions indicate various intermediate degrees of mental accessibility (e.g., June, the woman, demonstratives, pronouns, cliticized pronouns, etc.), all arranged on a scale of degrees of accessibility. Gundel et al. (Reference Gundel, Hedberg and Zacharski1993) are more ambitious and propose that each referring expression encodes a specific cognitive status (such as ‘in focus’, ‘activated’, ‘familiar’ etc…). For example, it is chosen in (1), because the colander is extremely accessible according to Ariel, and similarly, it is ‘in focus’ for Gundel et al. (Reference Gundel, Hedberg and Zacharski1993). That, on the other hand, is used in (2) because the propositional ‘reading you some’ is not highly accessible to justify the use of a pronoun according to Ariel (it is ‘activated’, not ‘in focus’ for Gundel et al., Reference Gundel, Hedberg and Zacharski1993):

1. (1) MARILYN:…There's a colander_i –…Oh. It_i's gone. Oh here it_i is. (SBC: 003)

2. (2) PAMELA:…I could read you some…I mean is that allowed? (SBC: 005)

As can be seen, a much more central role is assigned to specific linguistic forms under this approach, although both theories fully appreciate the need for pragmatic inferencing on top of decoding for proper reference resolutions (see section 2.5.1). Referring expressions are each directly associated with some cognitive concept in a grammatical manner.

2.3.1 The historical and typological pragmatics research paradigm

Much historical (grammaticization) and typological pragmatics research analyzes the current grammar as pragmatically motivated. As in inferential pragmatics thinking, the assumption here is that in order to best use their grammar to fulfill their interactional goals, interlocutors make heavy use of context. All newly formed form/function correlations are crucially context-bound, and hence the clear role of pragmatics in their initiation. Gradually, a consistent use of some form in some context, which contributes to the derivation of some extralinguistic reading, may bring about the entrenchment of the pragmatically derived form/function correlation into conventional codes. Once conventional, cancelability no longer applies, nor is contextual support needed. A piece of grammar has emerged.

We start with innovations. When meanings are intended, which are not straightforwardly codable in the current grammar, speakers must improvise. Uttering ungrammatical utterances is not an option. But speakers can mobilize current lexemes and constructions to convey their innovative messages relying on pragmatic inferencing as mediators. A wealth of relevant examples, universally attested, are provided by Heine and Kuteva (Reference Heine, Wischer and Diewald2002): ‘Alone’ can be mobilized to express ‘only’, body parts (e.g. ‘back’) can be mobilized to express spatial relations (‘in back of’), demonstratives can indicate a relative clause construction (that anaphorically referring to the head) or definiteness in general (that > the), ‘or’ can help mark (alternative) questions, ‘ability’ can trigger an interpretation of ‘possibility’, ‘all’ or ‘people’ can indicate ‘plurality’, possessive + ‘head’ or ‘body’ can help indicate an action performed on the self, etc. In all these cases, the assumption is that initially, the innovated meaning was only a conversational implicature derived with the help of a richly supportive context. But, as Grice (Reference Grice1989) originally noted, what is initially generated as a PCI may end up a semantic meaning.

Note that speakers rely on pragmatic implicatures not only when their grammar is incapable of expressing their intended meaning. We often prefer to convey our messages only implicitly. For example, instead of explicitly stating that x is the reason for y (using some because expression), speakers can implicate it. Now, we're here interested in the eventual development of a specialized form for expressing reasons. One form speakers can use in order to implicate a causal relation is a temporal adverbial such as after, because events preceding other events are easily construed as causing or explaining them (Hopper and Traugott, Reference Hopper and Traugott2003). Interestingly, cross-linguistically, many reason conjunctions are either ambiguous (since) or etymologically derived from temporal expressions (Hebrew ekev derived from be=ikvot ‘following’):

1. (3) And there's a lot of moves that we just know. After being…there for so long (LSAC).

Clearly, “being there for so long” is intended as an explanation for how it is that “there's a lot of moves that we just know.” But the speaker in (3) chose not to use an explicit ‘because’ adverbial, although quite a few of those are available in English: because, since, for, as, on account of. The reason is that speakers sometimes wish to convey their message indirectly. This can account for the fact that many languages possess multiple causal/reason adverbials, most of which are motivated diachronically since they can easily be seen as having originally triggered a causal implicature. Clearly grammatical constructions too can be similarly motivated, the creation of what are now called reflexive pronouns, for example. Initially, speakers mobilized an emphatic (complex: pronoun + self) referring expression to indicate what they rightly considered an unexpected/marked coreference relation (see section 2.3.2 below).

Finally, even an (initially) unintended meaning may become associated with some form, given our encyclopedic knowledge, which is automatically brought in when we interpret messages. Consider the reflexive construction, which often turns into a low transitivity construction (e.g., French se laver, ‘wash oneself’ = ‘wash’ – Kemmer, Reference Kemmer1993). It so happens that self-inflicted actions tend to be associated with a low degree of transitivity (as defined by Hopper and Thompson, Reference Hopper and Thompson1980). They are often unintended (Cf. I hit/cut myself with I hit/cut it), internal, rather than external (blame/consider oneself), etc. If that's the case, then even in the absence of an initial intention to convey a low degree of transitivity, interlocutors may come to associate the construction with low-transitivity activities.

The idea is that all contextual enrichments, no matter their source, may penetrate the grammar. The processes here briefly mentioned have one thing in common. Once many speakers start using the same explicit forms (e.g., back, after) as a basis for the same pragmatic inferences (a spatial relationship, a causal interpretation, respectively) a conventional code may be established between these forms and functions. The pragmatically derived meaning semanticizes, and becomes a coded meaning. Similarly, once the syntactic contexts in which reflexive forms tended to recur became a discourse pattern salient to speakers, a grammatical convention could set in, and in fact did set in in many languages. Since reflexive pronouns were very often used for a coreferent co-argument, this could very well explain the nature of Binding Condition A. And once the low transitivity often attributed to reflexive-marked actions becomes salient to speakers (not necessarily consciously) the construction may be reanalyzed as a low-transitivity or even an intransitive construction (cf. French se).

Researchers within this paradigm feel they can offer substantive explanations not only for why grammars of unrelated languages are so similar. They can also account for differences between languages, which the innateness assumption cannot (parametric variation is not enough to account for the rich, fine-grained variety found in the world's languages). The idea is that for the most part what we have are near universals, rather than precisely identical codes (Evans and Levinson, Reference Evans and Levinson2009). If grammar evolves when specific forms gradually come to be associated with specific functions, then differences between languages can be found where some language has already grammaticized some form/function correlation, but another hasn't (yet). Supporting evidence for this claim is the common finding that what is grammatically stipulated in one language is a possible, or sometimes even very common, discourse pattern in another (English after possibly triggers a ‘because’ interpretation, as in (3), but in Hebrew it has grammaticized). And if the Binding Conditions are setting in, they may only be optionally applied in some language (e.g., Old English, where personal pronouns regularly received bound readings), but be obligatory in another (Modern English). Another source of (slight) variation is the somewhat different translation of a pragmatic motivation into grammatical dress. For example, if speakers are indirectly conveying the ‘behind’ spatial relation using a body part, they may choose ‘back’ for triggering this initially pragmatic interpret-ation, but they may equally plausibly choose ‘buttocks’ for this purpose (this is the source of Hebrew ‘behind’). Such is also the variation between languages which use resumptive pronouns, as analyzed in Keenan and Comrie (Reference Keenan and Comrie1977).

Summing up, researchers within the historical and typological pragmatics paradigm deal with both codes and inferences, although their main interest lies in providing what they consider natural explanations for the universal tendencies in the typology and grammaticization of natural languages. The interest in pragmatic inferences is restricted to explaining the why and how of grammatical codes: the processes leading from inferences and represented discourse patterns to grammatical codes and the principles restricting typological variability. Core grammatical phenomena are here pragmatically motivated, and not only optional choices among equally acceptable forms (cf. form/function pragmatics).

2.3.2 Historical/typological pragmatics and reference

Binding relations are taken as a universal, in fact innate, linguistic concept by generative grammarians. However, not all languages have use for a dedicated marking of these relations. Historical/typological research seeks to explain how it is that such marking can evolve in real discourse. The assumption is that the pragmatic inference at work here is that co-arguments of the same verbs tend to stand for disjoint referents (e.g., Faltz, Reference Faltz1985, and see Ariel Reference Ariel2008a: 6.1.1 for supporting corpus statistics). If so, co-argument personal pronouns should preferably be interpreted as disjoint. However, since speakers do sometimes need to refer to coreferent co-arguments (e.g., Am I killing myself?, SBC: 015), more marked pronouns (what we now call reflexive pronouns) gradually evolved to indicate the marked coreference. Since it is co-argument coreference that is consistently marked, the resulting binding conditions grammaticized specifically around co-arguments (Reinhart and Reuland, Reference Reinhart and Reuland1993). Such an account can explain why the application of the binding rules was slow and gradual. Reflexive forms first appeared for more marked cases, where self actions (i.e., coreference) are less expected, as when destruction events are described (e.g., X destroyed herself). They appeared later for self actions which are more expected (e.g., X dressed her(self). Keenan, Reference Keenan, Moore and Polinsky2003).

Ariel (Reference Ariel, Barlow and Kemmer2000) has argued that the evolution of verbal person agreement markers out of personal pronouns (e.g., Hebrew shavar=t ‘broke=2nd pers. fem.’ from shavar at ‘broke you-fem.’) can be accounted for by Accessibility theory. Recall that Accessibility theory predicts that highly accessible mental representations are retrieved using highly accessible referring expressions, which are phonetically attenuated. It stands to reason that the speaker and the addressee, both highly accessible referents in face-to-face interactions, would frequently be referred to by high-accessibility markers, cliticized pronouns quite often. A consistent pattern whereby first and second person pronouns are reduced may lead to their cliticization to the verb. This may pave the way for a reanalysis whereby the original pronouns are taken as first/second person verbal bound morphemes. This pragmatic explanation can also account for why typologists have found a cross-linguistic difference between first/second person verbal agreements and third person verbal agreement. The latter are far less commonly marked on verbs. Accessibility theory can account for this asymmetric paradigm by noting that unlike the speaker and the addressee, third person referents are not consistently highly accessible in discourse (see the statistics in Ariel, Reference Ariel, Barlow and Kemmer2000). Hence, third person pronouns do not get reduced consistently enough to trigger a gradual process of cliticization.

In sum, typological/historical analyses set out from pragmatically motivated pressures (the need to mark a marked coreference, the tendency to reduce referring expressions denoting highly accessible referents) and explain the grammaticized pattern as resulting from an unintended series of small and gradual changes, starting with optional choices, going through preferred discourse patterns, and ending with an entrenchment whereby the discourse pattern is translated into an obligatory grammatical convention. The focus for this research paradigm is on the crossing from pragmatics to grammar.

2.4.1 Competition within paradigms

The different accounts for discourse reference by the two inferential pragmatics theories mentioned above (the Relevance and neo-Gricean accounts) follow straightforwardly from the general differences between the theories. Typically, the Relevance accounts assign a more significant role to the overall interpretation of the utterance, governed by the single Principle of Relevance. They naturally emphasize how reference resolution is a by-product of this interpretative process. Levinson focuses more on (some) actual forms, and relies on the interaction between the various neo-Gricean Principles (I, Q) to explain referential patterns. The specific context plays less of a role for him, the interpretations seen as GCIs.

Next, recall that both Ariel and Gundel Reference Gundel, Hedberg and Zacharskiet al. offer form/function accounts for the use and interpretation of referring expressions. While Ariel views referring expressions as arranged on a scale, each indicating a relatively higher (or lower) degree of mental accessibility, Gundel Reference Gundel, Hedberg and Zacharskiet al.'s theory is much more precise in that it associates each referring expression with a specific cognitive status. I have elsewhere argued that when we examine the definitions given by Gundel et al. (Reference Gundel, Hedberg and Zacharski1993) for each cognitive status, it's no longer clear that they are indeed distinct (Ariel, Reference Ariel, Sanders, Schilper-oord and Spooren2001). But a major problem with the Givenness hierarchy is that there simply aren't enough cognitive statuses to go around. Here's one such case:

1. (4)

   REBECCA: .. put the newspaper_i on his lap,

   RICKIE: Y[eah],

   REBECCA: Ø [mas]turbated,

   and then lifted the paper_i up, (SBC: 008)

It is no coincidence that the first mention of the newspaper is phonetically larger than the second one (this is a consistent finding). But under a six-category Givenness scale (Gundel Reference Gundel, Hedberg and Zacharskiet al.'s) there is no way to account for such delicate differences (between a full and a reduced definite NP, as well as between full and cliticized pronouns, etc.). An additional theory is needed, which could arguably be Accessibility theory. In fact, this is the point of Gundel Reference Gundel, Hedberg and Zacharskiet al. (forthcoming), where their Givenness hierarchy is shown to be orthogonal to Accessibility accounts (for the main part). If so, the fact that both theories offer form/function pragmatics accounts for referential forms does not automatically render either one of them redundant. It may well be that referential forms must meet both requirements, for example, that a referent marked by a demonstrative NP must be ‘familiar’, as well as relatively (but not maximally) accessible.

2.4.2 Competition across paradigms

I here cite two cases where the grammar/pragmatics division of labor is under debate. Both form/function pragmatic theories and historical/typological accounts compete with inferential pragmatics as to what to relegate to pragmatics and what to grammar. What is viewed as currently pragmatically inferable under an inferential pragmatics approach may be taken as a form/function code (Hebrew cliticized pronouns). Similarly, a distributional pattern may be analyzed as grammaticized by historical/typological theories, although it seems straightforwardly derivable by inferential pragmatic theories (some reflexive pronouns).

Consider the following Hebrew discourse, where the same entity (the press_i), is already very highly accessible (the first mention here is the sixteenth reference to it):

Given the availability of cliticized pronouns for Hebrew hem ‘they’, it is the avoidance of the shorter forms in the second and fourth references that is puzzling on any inferential pragmatic theory. Minimizing processing cost should have prompted the speaker to use the reduced pronouns throughout according to Relevance theory, for the same interpretation is made available by the two forms (see especially Reboul, Reference Reboul, Connolly, Vismans, Butler and Gatward1997). Levinson actually predicts a disjoint reading here, because the speaker avoided the most minimal form. Form/functionalists Ariel and Reference Gundel, Hedberg and ZacharskiGundelet al., however, can account for the alternating uses of full and cliticized pronouns by noting the points at which the speaker switches from one to the other. Discourse connectives, such as ‘but’ and ‘another thing’ here, signal that a potential topic change may occur, thus reducing the accessibility of the current topic. Reference Gundel, Hedberg and ZacharskiGundelet al. can similarly explain the data as a change from ‘in focus’ to ‘activated’.¹⁵ Hence, the differential preference for full and for cliticized pronouns in different contexts. In this case, it looks like the form/function approach is superior to the inferential approach.

Next, what is the status of the binding conditions regarding the use of reflexives? Levinson (2000: Chapter 4), König and Siemund (Reference König, Siemund, Frajzyngier and Curl2000) and Ariel (Reference Ariel2008b: Chapter 6) are all in agreement that a pragmatic interpretative pattern lies behind these grammaticized conventions (see again sections 2.3.1, 2.3.2). As an inferential pragmatist, Levinson points out that reflexives in different languages don't necessarily share the same grammatical/pragmatic status. For example, in some languages co-arguments can, but need not always be reflexive-marked. Obviously, the pattern is only pragmatic in these languages. But the picture is more complicated than that. Even for a single language, sub-patterns may be either pragmatic or grammatical. For example, according to Ariel's (Reference Ariel2008b: 6.3) grammaticization analysis, some adjuncts obligatorily take reflexive forms (despite herself, with himself) when coreferential with a clause-mate antecedent, although other adjuncts only manifest a pragmatic preference for the extension of the Binding convention beyond co-arguments (except for him(self), picture of her(self), jokes about him(self)). Other sub-constructions obligatorily take a pronoun, even though an obligatorily coreferential co-argument is involved (He didn't have any spots on him(*self/*her)). Although this pattern is an exception to the general Binding principle, and although it is clearly pragmatically motivated (according to the pragmatic motivation here, unmarked coreference is not in need of special marking), a grammatical convention is nonetheless involved (note the unacceptable reflexive form above). In other words, a competition between pragmatic and grammatical analyses is at work not just for the general distributional pattern applicable to the language as a whole, but also at much lower-level generalizations within the same language.

Different pragmatic theories certainly compete for the best account, whether within (2.4.1) or across (2.4.2) research paradigms. While researchers have mainly engaged in intra-paradigm debates, it's time for cross-paradigmatic debates and collaborations to take the stage in pragmatic research.

2.5.1 Reference in three keys

We've briefly looked at a number of competing pragmatic accounts for the use and interpretation of referring expressions. Although some accounts will ultimately have to be rejected, reality is that we need all three approaches in order to fully account for natural language reference systems. Each approach provides some of the relevant pieces needed to complete the great grammar/pragmatics puzzle. Form/function pragmatics codes (which may turn out to be neither Reference Gundel, Hedberg and ZacharskiGundelet al's nor Ariel's) are needed to account for the basic conventions informing reference marking and resolution. But, of course, codes never exhaust actual use. This is true for classical semantic codes, and it is equally true for grammatical pragmatic codes. This is where inferential pragmatic accounts must come in. For example, those two idiots (in (6)) seems to be too low an accessibility marker for an entity just now mentioned by pronominal they. The violation of Accessibility theory can, then, be explained as a special inferred use (epithet), where the speaker's goal is not just to refer, but at the same time to also predicate on the referent:

1. (6) .. So they_i go barging in on ∼Mar.

   .. So Mom felt obligated,

   to ask those two idiots_i to lunch (SBC: 006).

The same applies to the reference resolution cases analyzed by Kempson (Reference Kempson, Newmeyer and Robins1988) and Wilson (Reference Wilson1992) (see section 2.1.2). Finally, we need an account for why and how the pragmatics/grammar divide is crossed (e.g., how pronouns evolve into verbal person agreement markers). It is the discourse profiles analyzed by historical and typological pragmatists that constitute potential grammaticization paths. Such theories bridge the gap between temporary integrations between inferred and coded meanings (on-line conveyed meanings) and permanent (grammaticized) form/function correlations. To account for the full range of use and interpretation of referring expressions we therefore need the three different pragmatic approaches, even if some (or possibly all) current theories are factually incorrect. Different research paradigms are needed for handling the different aspects of the grammar and pragmatics of reference. The same applies to all other linguistic phenomena.

2.5.2 Conclusion: The value of multiple paradigms in pragmatics

Each of the pragmatic approaches here surveyed has offered a significant amendment to the limited classical code model of language. The main point of the (originally Gricean) inferential pragmatics critique of the code model was that truth-conditional codes cannot exhaust speakers’ intended on-line meanings, because ad-hoc contextual inferences are generated in addition. According to form/function pragmatists, the classical codes fall short of exhausting all grammatically specified conventions, because not all codes are truth-conditional. Finally, contra generative grammarians, historical/typological pragmatists argue that classical codes cannot account for possible versus impossible natural language grammars. Since each approach finds different flaws in the classical code model, no wonder they each enrich it in a different direction. But it cannot be emphasized enough that these different research directions are not at all contradictory. Quite the contrary. They complement each other. Here are some thoughts on how they could each gain from interactions with the other approaches.

The most basic pragmatic approach, inferential pragmatics, has a strong preference for maximizing pragmatics and minimizing grammar. Anything that can be analyzed as pragmatic inference must so be analyzed. However, once researchers consider form/function pragmatics findings regarding potentially conventional associations between constructions and non-truth-conditional meanings, some analyses in terms of inferences may be relegated to form/function pragmatics within the grammar. Why shouldn't perfectly plausible inferences get entrenched in an automatic, subconscious process? For example, perfectly general inferential pragmatics attempts to explain the use and interpretation of cleft constructions by reference to their compositional meaning accompanied by pragmatic inferences (Wilson and Sperber, Reference Wilson, Sperber, Oh and Dinneen1979) must give way to a form/function pragmatic analysis, where a conventional interpretation is associated with each of the constructions (see section 2.2.1 above). Indeed, Blakemore (Reference Blakemore2002) is a clear example of a Relevance theoretician who has consistently used form/function (procedural) analyses for discourse connectives (and see Ariel, Reference Ariel2010: Part III).

The same is true for the relevance of historical/typological pragmatics findings for inferential pragmatics, specifically, the recognition that inferences may gradually turn into grammatical codes. For example, if GCIs are normal or default inferences associated with specific forms, they are potential semanticization cases (Traugott and Dasher, Reference Traugott and Dasher2002). Indeed, Levinson's (Reference Levinson2000: Chapter 4) is an analysis in this spirit (and see Ariel, Reference Ariel2008b: Part II). At the same time, the fact that the GCIs most discussed in the field (e.g., scalar implicatures, and-associated implicatures) do not seem to semanticize in any language might mean that the inferences are not after all as “normal” or default. Why is that? Such questions will naturally arise once cross-paradigmatic discussion (and debate) become more commonplace.

Next, form/function pragmatists tend to automatically assume that the correlations they find between specific linguistic forms and non-truth-conditional meanings or use conditions are conventional, and hence, grammatical. But it's not clear that this is invariably the case. Such pragmatists would do well to integrate the code/inference division of labor with form/function analyses. While, no doubt, many of the correlations they discussed are encoded for specific constructions and forms, others may be better accounted for as further inferences (see Ariel, Reference Ariel2010: Chapter 7 for such analyses). Du Bois’ (Reference Du Bois and Tomasello2003) distinction between sign functions (codes) and structure functions, which only allow for certain functions to be fulfilled in them (with the help of inferencing), should then be useful. Especially now that less marked constructions are increasingly analyzed for their functions (following Construction Grammar), it is becoming rather clear that there is no one-to-one relationship between the added meaning of syntactic constructions and discourse functions. Often, a few functions are associated with a single construction, only some of which are encoded. Form/function pragmatists tend not to problematize such questions of code/inference division of labor (Goldberg, Reference Goldberg2006: Chapter 8).

Finally, historical/typological pragmatics research too could benefit from integrating inferential pragmatics theories and questions. Proponents must ascertain that they only assume very small steps of grammaticization, each of which is analyzable as depending on a reasonable on-line pragmatic inference. Etymological analyses can serve as excellent pointers to potential paths of grammaticization, but they cannot replace detailed analyses of the actual inferential steps, leading from one stage to the next. There is a danger in only noting common source-target pairs (Heine and Kuteva, Reference Heine and Kuteva2002), because these are more often than not connected by a whole series of very small changes. For example, it's a bit misleading to claim that expressions denoting small quantities, such as French pas ‘step’, Hebrew klum ‘something’, evolve into negators or negative polarity items, because the change never involves the initial and the endpoint directly. Indeed, this is why some of Sweetser's (Reference Sweetser1990) analyses of metaphoric changes have been reanalyzed by Bybee et al. (Reference Bybee, Perkins and Pagliuca1994) and Traugott and Dasher (Reference Traugott and Dasher2002) as processes where enrichment inferences have been semanticized. The same applies to the very valuable typological clines and semantic maps (Croft, Reference Croft2001; Haspelmath, Reference Haspelmath and Tomasello2003; Kemmer, Reference Kemmer1993). Semantic changes must be shown to have evolved out of theoretically justified on-line pragmatic inferences.

We can also envision further sophistication in explaining the processes leading from pragmatics to grammar by reference to inferential pragmatics theories. Ever since Grice proposed that conversational implicatures can semanticize, the assumption has been that implicatures sometimes turn semantic. Traugott has argued that it is specifically generalized invited inferences (she prefers this term over GCIs), rather than particularized inferences, that are potential semanticization cases. But what about the Relevance-theoretic explicated inferences (see the definition in section 2.1.1 above)? Given that explicated inferences are closer to semantic meanings than PCIs and GCIs (explicated inferences contribute truth-conditional aspects; they are inseparable from the linguistic meaning – Recanati, Reference Recanati and Davis1989), I have suggested that it may be explicated rather than implicated inferences that directly semanticize. PCIs must go through an explicated stage before they semanticize (Ariel, Reference Ariel2008a, Reference Ariel2008b).¹⁶ Last, some historical pragmatists have relied on the form/function pragmatics practice of distinguishing truth-conditional from non-truth-conditional meanings. Since some historical changes evolve non-truth-conditional functions (for many discourse markers), they have defined these as pragmaticization, rather than semanticization (Erman and Kostinas, Reference Erman and Kostinas1993). Given that the processes here concerned make a pragmatic pattern evolve into grammatical convention, the distinction is probably not justified. Greater attention should then be paid to the role of inferential pragmatics theories.

Summing up, while theories within the same research paradigm addressing the same linguistic phenomenon are in competition with each other (e.g., neo-Gricean and Relevance-theoretic analyses of scalars and and), theories from different research paradigms complement each other for the most part. Just because they each focus on a different aspect of linguistic form and use, all three paradigms are necessary for a complete picture of grammatical forms, and for what should be assigned pragmatic status. Our main (but all too brief) test cases were competing, as well as complementary, referential theories. I’ve argued that form/function pragmatic theories (e.g., Reference Gundel, Hedberg and ZacharskiGundelet al., Ariel) should account for the conventional interpretations associated with various referring expressions, inferential pragmatic theories (e.g., Relevance, neo-Gricean theories) account for the many referential interpretations mediated by inference (e.g., bridged coreference, type versus token readings), and historical/typological pragmatic theories account for dominant discourse tendencies regarding the actual use of referring expressions (e.g., marked coreference by marked forms), some of which turn grammatical, at least in some languages.

What the field of pragmatics, indeed of linguistics, needs now is not so much, or not only, inter-paradigmatic debates about who got it right and who got it wrong. More fruitful and exciting insights will be gained from cross-fertilization and cooperative research between proponents of different paradigms. It takes all pragmatic keys to fine-tune the grammar/pragmatics division of labor.¹⁷