18Some Interactions of Pragmatics and Grammar
GEORGIA M. GREEN
It has been recognized in generative grammar since the 1960s1 that the acceptability of sentences depends on the referential and predicative intents imputed to the speaker. Of course, this fact has not always been represented in those baldly pragmatic terms; for a long time it was (and in some quarters still is) socially or tactically unwise to refer to speakers’ mental states in describing syntactic knowledge, so the problem was reframed in terms of the properties of the syntactic or semantic properties of the sentence (see e.g. Karttunen 1977, Karttunen and Peters 1979). Now, however, a deeper understanding of the relation of knowledge of grammar and lexicon to knowledge of the principles of language use, both universal (e.g. Grice’s (1967) Cooperative Principle2) and language particular (cf. e.g. Morgan 1978), allows a direct and straightforward description of linguistic competence (Green 2000). The problem is not an isolated one restricted to a few troublesome constructions. Many or most of the constraints that have been proposed by generative grammarians (e.g. binding constraints, constraints on the reference of unexpressed subjects of infinitives) must either be stated in ultimately pragmatic terms or describe constructions whose use conveys pragmatic information about the beliefs of the speaker – beliefs about the world (presuppositions), about the propositional attitudes of the addressee, or about the structure of the ongoing discourse.3 This chapter reviews a broad selection of syntactic phenomena that have been observed to have pragmatic values, and sketches the properties of a description of competence that straightforwardly reflects them.
1 What Kind of Information is Pragmatic?
In a sense, all pragmatic information is ultimately indexical information, that is, related to indices for speaker, hearer, time, and location of an act of uttering something that the sign represents (Bar-Hillel 1954, Nunberg 1993, Levinson this volume). It is important to be clear that it is not linguistic forms (words,
408 Georgia M. Green
morphemes, expressions) that carry pragmatic information (though informal descriptions often suggest this, in formulations like “this form expresses/marks the speakers belief/intention that . . .”), but the facts of their utterance. Pragmatic information is information about the relation between the user of the form and the act of using the form. First and foremost, pragmatic information is information about mental models: speaker’s and addressee’s mental models of each other. Linguistic pragmatics irreducibly involves the speaker’s model of the addressee, and the hearer’s4 model of the speaker (potentially recursively). For George to understand Martha’s utterance of “Xxx” to him, he must not only recognize (speech perception, parsing) that she has said “Xxx,” he must have beliefs about her which allow him to infer what her purpose was in uttering “Xxx,’’ which means that he has beliefs about her model of him, including her model of his model of her, and so on, as illustrated in figure 18.1. Thus, when Martha says “Xxx” to George, meaning by it “p”, she does so believing that George believes “not p”. And when George hears her speak to him, he recognizes that she has said Xxx, and understands that she believed that he believed the negation of what she meant by it. Any of these beliefs could be incorrect at one level of detail or another. That is why, in figure 18.1, we see that George’s image of Martha, and his of her, as well as his belief about what she meant, are not quite accurate. Martha’s image of George as having less, longer, and straighter hair than we can see he has, and his image of her as being smaller and having longer hair than she in fact does are visual metaphors for the fact that we have imperfect knowledge of each other. Significantly, the participants’ images of each other are also successively less detailed, illustrating that their knowledge of each other is incomplete as well as not quite accurate. And most important, illustrated by the different fonts for the letter “p” that represents Martha’s meaning, what Martha meant by Xxx is distinct from what George believes she meant. As a New Yorker cartoon caption once had it: “I know you believe you understand what you thought I said, but I’m not sure that you realize that what you heard was not what I meant.” Since acts are interpreted at multiple levels of granularity, this built-in indeterminacy or margin of error is present for acts involved in choosing words and construction types, as well as for acts of uttering sentences containing or instantiating them.
These are background assumptions against which all discussion of linguistic pragmatics must take place, and which in fact motivate a pragmatic load for both “meaningless” discourse particles like um, well, and like (Schourup 1985, Kose 1997, Schwenter and Traugott 2000, Green 2001, Fukada-Karlin (in preparation), Blakemore this volume) and truth-conditionally equivalent alternative constructions to express the same proposition.
2 Some Illustrative Phenomena
In English and probably all other natural languages, there are truth-condition- ally equivalent alternatives to practically every describable construction, and
Some Interactions of Pragmatics and Grammar 409
~p
~P
Xxx.
Xxx.
Figure 18.1 Illustration of speaker’s and addressee’s mental models of each other
to the extent that this is true, the alternatives turn out to have different pragmatic values.5 Horn (1984a, 1993) offers a detailed and convincing explanation of why this is inevitable.
The factors which might enter into the choice between or among truthconditionally equivalent constructions are numerous. To take one of the most familiar examples, choosing a passive construction over an active counterpart might be motivated by an intention to represent the patient as the topic,
410 Georgia M. Green
and/or defer information about the agent to the end of the sentence, as in (1a), where it will be more perceptually prominent, naturally receiving sentence stress. On the other hand, using the passive allows expression of the agent to be entirely suppressed, enabling a speaker to accommodate the fact that it is unknown (1b) or irrelevant (1c) who the agent is, or to just avoid saying who the agent is, even if he does know, as in (1d).
(1)a. The bank was robbed by two young men with extensive facial scars.
b.My bike was taken between 3:00 and 5:30 on Monday.
c.Over 20,000 copies of the book were sold before it was discovered that pages 285 and 286 were missing.
d.[Do you know where the February Scientific American is?] It was thrown out.
Using a passive also commonly implies a belief that the event described had a particular effect on some contextually salient sentient individual (R. Lakoff 1971a, Davison 1980, Fukada 1986). Often the affected individual is the subject (as in (2a, 2b)), but it can be any contextually salient legal person, including, but not limited to, the speaker or addressee, as in (2c, 2d).
(2)a. He was interrogated for three hours.
b.He was awarded a Pulitzer Prize for that photograph.
c.The evidence was destroyed in the fire.
d.This idea has been attacked as simplistic and naïve.
The effect can be negative, as in (2a), or positive, as in (2b); only the details of the context in which the sentence is uttered could tell us which it is in (2c). The jarring effect of (3a) is attributable to the fact that the referent of a term is clearly affected, but the construction does not represent that term as a subject (cf. 3b).
(3)a. #A car hit your dog, but he’s OK now.
b.Your dog was hit by a car this afternoon, but he’s OK now.
There is even suggestive evidence that details of complementizer choice have pragmatic implications associated with them (cf. Bolinger 1972a, Borkin 1974, Riddle 1975, 1978), reflecting different assumptions of the speakers as in (4) and (5), where the (b) examples imply a stronger conviction on the part of the subject.
(4)a. He expects that he will win.
b.He expects to win.
(5)a. I know that it’s raining.
b.I know it’s raining.
In the cases that are the focus of this chapter, the connection between the way something is said and what is intended to be conveyed is automatic,
Some Interactions of Pragmatics and Grammar 411
suggesting an analysis as fossilized conversational implicature, according to the following logic. Since implicatures exploit (assumed) literal meanings, with “live” conversational implicature, and even with short-circuited conversational implicature (Morgan 1978), it is often reasonable to deny that an implicature was intended. Thus, denying the causal implicature of the embedded conjunction in (6) is acceptable.
(6)The committee money disappeared from the safe that day, and Lee came home with a new jacket that night, but I don’t believe or intend to imply that the two events have anything to do with each other.
In contrast, the cases discussed here involve use conditions which refer to beliefs or attitudes of the speaker and amount to presuppositions; they are so strongly linked to a syntactic construction that it sounds irrational to use that construction and then deny that the conditions hold. For example, the present tense can be used to refer to future time as long as the event referred to is assumed to be prearranged, and there is an adverbial expression indicating a future time indicated explicitly or in ellipsis (G. Lakoff 1971b, Prince 1982). Thus, (7a) can be used in many of the same situations as (7b).
(7)a. The Celtics play the Bucks tomorrow.
b.The Celtics are going to play the Bucks tomorrow.
If the event is not (mutually) understood to be prearranged or scheduled, simple present tense cannot be used for future time, so using (8) would imply that the speaker (and the addressee, in the speaker’s estimate) either (a) knew that the game was fixed, or (b) had a firm belief in predestination, and believed that the speaker had an inside line on the Omnipotent’s plans for basketball games.
(8) The Celtics crush the Bucks tomorrow.
The time adverbial can be either explicit as in (7) and (9a), or implicit, as in (9b).
(9)a. Sandy arrives tomorrow, so we’ll have to clean up the guest-room.
b.A: What’s on the docket for tomorrow?
B:Well, Sandy arrives, so I have to go to the airport.
But when a shared presumption that the event is scheduled cannot be assumed, as in (10), people won’t use a present tense form to refer to a future event.
(10)A: What’s new?
B:#My sister comes in from Seattle, so I’m getting ready to go to the airport.
412 Georgia M. Green
The bizarreness of (11a) shows that this belief requirement cannot be denied like a conversational implicature, but can be suspended, like a presupposition (11b).
(11)a. #The Celtics win tomorrow, but it’s not preordained or fixed; there’s an off chance the Pistons might beat them.
b.The Celtics win tomorrow, if Benny got the fix in and explained to the players on the other team what would happen to them if they didn’t lose.
The main focus of this chapter is how the construction used to say something reflects the speaker’s attitude toward and beliefs about the topics and referents in the ongoing discourse. Other aspects of how what is said is said that have been shown to affect what is conveyed include intonational choices (Schmerling 1976, Cutler 1977, Olsen 1986, Ward 1988, Hirschberg 1991) and more segmental phonological choices (e.g. sarcastic nasalization, Cutler 1974: 117), as well as the choice of which language to use (Gumperz 1976, Burt 1994). In the cases to be discussed, the focus is more on what the speaker intends (or is content) to convey than on an appreciation of the addressee’s limitations in processing language “on-line,” though of course, a rational speaker will take the hearer’s needs into consideration in deciding how to say what must be said. This topic is addressed briefly in section 5. However, a primary goal of this chapter is to draw attention to a variety of cases where potential inferences from the way something is said are grammaticalized as part of the syntactic repertoire. The following sections describe a selection of other constructions that differ from their truth-conditionally equivalent counterparts in various ways. Most reflect different beliefs about or attitudes toward referents of linguistic expressions that are part of the utterance. Of course, truth-conditionally equivalent sentences may also differ from each other in rhetorical function (i.e., in what gets asserted and what is presupposed), and these are treated in section 3. Section 4 treats syntactic devices that reflect the speaker’s assumptions about the structure of the discourse. Section 5 addresses processing-related issues: syntactic constructions that enable a speaker to compensate for (perceived) difficulties in producing or parsing a complex utterance. Many of the constructions have more than one use or function, and show up in more than one category.
3 Belief/Attitude/Value Cases
Another instance of “stylistic variants” exhibiting a difference in rhetorical value involves constructions with sentential complements as in (12) and (13), or adjuncts as in (14). The (a) sentences in these sets differ from the (b) sentences in that the italicized subordinate clause represents a presupposed or otherwise subordinate proposition in the (a) sentences, but has its own declarative illocutionary force in the (b) sentences.
Some Interactions of Pragmatics and Grammar 413
(12)a. I bet it’ll float if you throw it in the lake.
b.It’ll float if you throw it in the lake, I bet. [SLIFTING]6
(13)a. That Sandy thought it was Tuesday is obvious/clear.
b.It’s obvious/clear that Sandy thought it was Tuesday. [EXTRAPOSITION]
(14)a. Someone who said the girls were supposed to bring two quarts of potato salad called.
b.Someone called who said the girls were supposed to bring two quarts
of potato salad. [RELATIVE CLAUSE EXTRAPOSITION]
Thus, depending on the sense intended for bet in (12a), (12a) is either a wager or a speculation, but (12b) can only be a speculation – bet does not have a performative interpretation in that construction (Ross 1975, Horn 1986). And while both (13a) and (13b) could be used to assert something about the claim that Sandy had some belief about the identity of a day, only (13b) could be used to make the claim that Sandy had that belief (Morgan 1975b, Horn 1986). In the case of (14), the (a) sentence reports who called, while the (b) sentence reports what someone said the girls were supposed to bring (Ziv 1976).
Other constructions reflect particular kinds of beliefs speakers have about the objects of their discourse. For example, use of the INTERNAL DATIVE construction (Green 1974) in (15b) implies that the speaker believes that the referents of the subject and beneficiary noun phrases were alive at the same time.
(15)a. Win this one for the Gipper/me.
b.Win me/#the Gipper this one.
Wierzbicka has argued (1986) that use of the internal dative construction reflects more generally the speaker’s greater interest in the referent of the indirect object noun phrase.
The RAISED SUBJECT constructions in (16b, d–f) (Borkin 1974, Postal 1974, Steever 1977, Schmerling 1978) reflect the speaker’s assumption of the possibility of interaction between the (implied) experiencer (MacArthur in (16b)), or agent (Eks in (16d–f)) and the referent of the raised subject (Caesar, Sandy, Dale) at the time referred to by the raising verb.
(16)a. It seemed to MacArthur that Patton/Julius Caesar was the greatest general in history.
b.Patton/#Julius Caesar seemed to MacArthur to be the greatest general in history.
c.Eks asked that Sandy leave.
d.Eks asked Sandy to leave.
e.Eks allowed Dale to examine Dana.
f.Eks allowed Dale to be examined by Dana.
414 Georgia M. Green
A similar construction with finite complements and predicates including looks like and appears as if (Rogers 1971, Postal 1974: 356–68) has related properties, so that sentences like (17a) provide no information about the nature of the evidence for the claimed resemblance, while ones like (17b) reflect the speaker’s ability to perceive the referent of the subject displaying the predicated property at the time of the speech act.
(17)a. It looks like Stalin’s been dead for years.
b.#Stalin looks like he’s been dead for years.
R. Lakoff (1969a) and Horn (1971, 1978b, 1989) have described the NEGATIVE TRANSPORTATION phenomenon illustrated in (18a) where, with a certain class of verbs and adjectives, a negative occurs one or more clauses above the clause it conversationally negates. Thus (18a) would communicate what (18b) straightforwardly asserts, and (19a) would be conversationally ambiguous between a report that Dana lacks the desire to wash dishes, and a report like (19b) that Dana desires not to wash dishes.
(18)a. I don’t think Sandy will arrive until Monday.
b.I think Sandy won’t arrive until Monday.
(19)a. Dana doesn’t want to wash dishes.
b.Dana wants to not wash dishes.
The difference between the (a) sentences and the (b) sentences in (18–19) is that the (a) sentences, with transported negatives, are hedged – they represent weaker claims, apparently by implicating rather than asserting the relevant negative proposition (cf. Horn 1978b: 131–6, 177–216 and Horn 1989, Chapter 5 for discussion). A similar phenomenon is evident in the fact that the morphologically incorporated negative in (20b) is pragmatically stronger than the unincorporated negative in (20a) (Sheintuch and Wise 1976), though of course, as (20a,b) are truth-conditionally identical, they have the same entailments.
(20)a. I didn’t see anyone there.
b.I saw no one there.
The use of some or any described by R. Lakoff (1969b) provides a very clear reflection of speakers’ attitudes. As Lakoff showed, in the class of interrogative, conditional, and hypothetical constructions where some and any are truthconditionally equivalent, the use of some indicates a positive attitude toward the situation described by the proposition it is part of, while any reflects a neutral or negative attitude. Thus, the condition in (21a) is satisfied by the same state of affairs as the one in (21b), but (21b) implies an assumption that it is likely that there are no apples on the table.
Some Interactions of Pragmatics and Grammar 415
(21)a. if there are some apples on the table
b.if there are any apples on the table
Similarly, the question in (22a) reflects the hope that Bill wants spinach, while (22b) may reflect the hope that he does not.
(22)a. Does Bill want some spinach?
b.Does Bill want any spinach?
In the same vein, (23a) could be a bribe, intended to get the addressee to eat bread (treating cooking hamburgers all week as a reward), while (23b) would be a threat, intended to keep the addressee from eating bread (and treating cooking hamburgers all week as an undesirable event).
(23)a. If you eat some bread, I’ll cook hamburgers all week.
b.If you eat any bread, I’ll cook hamburgers all week.
A whole host of other NEGATIVE POLARITY ITEMS (cf. Baker 1970, Horn 1971, 1989, Schmerling 1971a, Israel this volume) reflect attitudes similar to those which any reflects (Ladusaw 1980). The examples in (24) expose the speaker’s suspicion that Bo spent, ate, and knew nothing, and did not bother to RSVP, respectively.
(24) Do you think Bo 1 spent a red cent on that? |
5 |
4 |
4 |
2 ate a bite? |
6 |
4 knows bupkes? (“nothing” [ Yid.]) |
4 |
3 bothered to RSVP? |
7 |
In addition to the negative polarity sensitivity of syntax illustrated just above, languages may display sensitivity to ignorance. Both the inversion of subject and auxiliary verb in embedded questions that is common in many dialects of English and truncation of embedded questions (SLUICING) imply that the individual to whom the answer is implied or assumed to be relevant is in fact ignorant of the answer. Thus the (a) sentences in (25) and (26) are more acceptable than the (b) sentences, which have contradictory implications which their acceptable and uninverted or unreduced (c) counterparts lack (cf. Ross 1975).
(25)a. She wants to know who did I appoint.
b.#She already knows who did I appoint.
c.She already knows who I appointed.
(26)a. John broke something, but he won’t say what.
b.#John broke something, and he said what.
c.John broke something, and he said what he broke.
416 Georgia M. Green
This property of subject–auxiliary inversion and sluicing explains why the examples in (27) induce the implicature that the speaker does not know the answer to the question.
(27)a. It never occurred to me to wonder who did she appoint.
b.John went somewhere with my car, and you know where.
Although the examples cited here are all from English, it would be surprising to find a language where custom didn’t link beliefs or attitudes to the use of particular words or constructions. Sakakibara (1995) and Kose (1997) give detailed examples of the speaker beliefs associated with the use of Japanese long-distance reflexives and sentence-final particles, respectively; cf. Huang (this volume) for additional discussion of the pragmatics of anaphora.
4 Reflections of Discourse Structure
Language scholars have long recognized that there are correlations between the order of syntactic constituents in a sentence and the discourse function of the information which a particular constituent references (Mathesius 1928, Firbas 1964, Halliday 1967, Kuno 1972, Gundel and Fretheim, this volume; Ward and Birner, this volume, inter alia).7 In general, and all other things being equal, the first phrase in a sentence tends to be intended to denote familiar (or TOPICAL, or GIVEN, or OLD, or presupposed, or predictable, or THEMATIC) material, while phrases toward the end of the sentence tend to denote NEW (or FOCUSED, or asserted, or RHEMATIC) material. Other things are not always equal, however. Sentence stress or intonational accent (higher pitch which falls off rapidly and is perceived as louder) also correlates with information being treated as new (Schmerling 1976), and new information may be expressed in phrases that occur toward or at the beginning of a sentence if they bear the main sentence stress, as in (28) (Olsen 1986).
(28) John ate the cookies.
Furthermore as Prince (1981a) demonstrated, familiar, predictable, given, old, theme, and sentence topic do not denote interchangeable notions, and different writers have used the same term to refer to rather different categories. Still, the various writers seem to have been addressing the same point, summarized by Horn’s (1986) observation that the initial slot in a sentence tends to be reserved for material taken to refer to the discourse theme or sentence topic (i.e., what the sentence is about). Typically, this is material that the speaker (reflexively) assumes to be familiar to the addressee, and preferentially, it is material which is either salient (assumed by the speaker to be in the addressee’s consciousness) or presupposed (taken as non-controversial) (Horn 1986: 171). It is not surprising, then, that syntactic rules of languages provide for numerous alternative
Some Interactions of Pragmatics and Grammar 417
constructions which differ in the order of phrases while preserving truthconditional semantics and illocutionary force. This is true even in a “fixed word order” language like English,8 as illustrated by the incomplete list of options for English declarative sentences given in examples (29–32):
(29)a. Eks delivered a rug to Aitchberg.
b.A rug was delivered to Aitchberg by Eks. [PASSIVE]
c.There was a rug delivered (to Aitchberg) (by Eks). [THERE-INSERTION]
d.A rug, Eks delivered to Aitchberg. [TOPICALIZATION]
e.It was a rug that Eks delivered to Aitchberg. [CLEFT]
f.What Eks delivered to Aitchberg was a rug. [PSEUDO-CLEFT]
g. . . . and deliver a rug to Aitchberg, Eks did. [VERB PHRASE PREPOSING]
(30)a. Finding typographical errors is never simple.
b.It is never simple to find typographical errors. [EXTRAPOSITION]
c.Typographical errors are never simple to find. [TOUGH-MOVEMENT]
d.Simple to find, typographical errors are not. [ADJECTIVE PHRASE
PREPOSING]
(31)a. Eks met a woman who said she was the Princess Anastasia’s governess at Treno’s.
b.At Treno’s, Eks met a woman who said she was the Princess Anastasia’s
governess. [ADVERB PREPOSING]
c.Eks met at Treno’s a woman who said she was the Princess Anastasia’s governess. [HEAVY NP SHIFT]
d.Eks met her at Treno’s, that woman who said she was the Princess Anastasia’s governess. [RIGHT DISLOCATION]
e.That woman who said she was the Princess Anastasia’s governess, Eks met her at Treno’s. [LEFT DISLOCATION]
(32)a. The little bunny scampered into its hole.
b.Into its hole, the little bunny scampered. [LOCATIVE PREPOSING]
c.Into its hole scampered the little bunny. [INVERSION]
Of course, old information does not tend to go first just because it is old, or become old just because it is first. Sometimes none of the material in a sentence represents “old information,” and as noted above, new information sometimes goes first; generally speakers have more particular reasons (not necessarily conscious reasons) for making a particular constituent first or last in a sentence (cf. Green 1982a). Such functions of word order have been explored in some detail for a number of constructions. Two are described below.
4.1 Preposing
Ward’s (1988) analysis of preposings like those in (29d), (29g), and (30d) indicated that while they may serve a variety of discourse functions, which he
418 Georgia M. Green
described in detail, they have two properties in common. In Ward’s terminology, they first of all mark the preposed element as referring to an entity which is related in a certain way (as a BACKWARD LOOKING CENTER) to entities previously evoked in the discourse (the set of FORWARD LOOKING CENTERS).9 Second, they mark the presupposed open proposition of the unstressed part of the sentence as salient in the discourse. Viewed from a different perspective, the referent of the presupposed element must function as a backward-looking center and the open proposition must be salient in the discourse for the utterance of a sentence with a preposed phrase to be acceptable in its context. Thus, a sentence like (29d) might be used in a context like (33), where the referent of one of these rugs is very obviously in a subset relation to the previously mentioned set {rugs to be given as rewards} and the open proposition is “Eks deliver a rug to someone”.
(33). An Eastern bloc embassy official gave Eks six full-size oriental rugs, and directed him to give them to the senators who had been most cooperative. One of these rugs Eks delivered to Sen. Aitchberg.10
It might also be used in a context like (34), where one of these rugs is a member of the set {rugs concealing cocaine}, and the open proposition “Eks delivered something to Aitchberg” is salient in the discourse.
(34). FBI agents suspected both Eks and Aitchberg of trafficking in cocaine, and had been tailing them for months. In March, they learned from an informant that six oriental rugs concealing 20 pounds of cocaine each had come through JFK airport, and as Exhibit B indicates, one of these rugs Eks delivered to Aitchberg.
But a sentence like (29d) could not be used in a context like (35), where neither of these conditions holds.
(35). Eks and Aitchberg played golf together regularly. ??An oriental rug Eks delivered to Aitchberg one day.
4.2Main-verb inversion
English is graced with a number of inversion constructions which, as in (32c), allow the subject noun phrase to appear after the main verb instead of before it (Green 1980, 1982b, 1985, Birner 1992, 1994, Ward and Birner, this volume). The inversions begin with a preposed adjective phrase, participial phrase, or locative or directional adverbial phrase. Because of this, they can be used to serve any of a number of functions which exploit that phrase order (Green 1980, Birner 1992, 1994). For example, a writer11 may use an inversion with a preposed phrase which refers to a previously established or implied referent to describe how information following it relates to previous discourse, as in
Some Interactions of Pragmatics and Grammar 419
the examples in (36), where the initial phrase contains something explicitly ((36a,b)) or implicitly ((36c)) anaphoric to something preceding in the discourse.
(36)a. [ . . . new license . . . ] Attached to it, as always, is an application blank for next year’s license.
b.Jerome and Rita Arkoff and Tom and Fanny Irwin were in the front row . . . Back of the Arkoffs and Irwins were William Lesser and Patrick Degan, and between them and slightly to the rear was Saul Panzer.
[Rex Stout, Might as Well Be Dead, p. 180. (New York: Viking Books, 1956)]
c.At issue is Section 1401(a) of the Controlled Substances Act.
In other instances, what is exploited is the fact that the inversion construction puts what would otherwise be a subject noun phrase in the sentence final position, which is typically reserved for focused, new information. This enables a writer to introduce a new discourse element (e.g. an important character or object or an element of the setting) in a focused position, as in (37).12 Travelogstyle descriptions exploit this extensively, as in (37c).
(37)a. In a little white house lived two rabbits.
[Dick Bruna, Miffy. (New York: Two Continents, 1975)]
b.Competing with the screamers for popularity are the phone-in programs, an adaptation of two rural American pastimes – listening in on the party-line and speaking at the town meeting.
[Robert Dye, “The Death of Silence,” Journal of Broadcasting, 12, 3 (1968). Reprinted in Subject and Strategy, ed. Paul Escholz and Alfred Rosa, 169–72, p. 170. (New York: St. Martins, 1978)]
c.The grounds were lavishly furnished with ceramic, stone, and wroughtmetal sculpture. There were an enormous stainless steel frog and two tiny elves in the foyer of the guest house, and outside stood a little angel.
A related use of inversions is to describe an event or locative relationship which resolves a salient indeterminacy in a narrative as it has been established up to that point. It might be the whereabouts of an important character, as in the second inversion in (38a), or the identity of the previously unknown agent of some significant action, as in (38b), or an event significant in the protagonist’s execution of his plans, as in the second inversion in (38c) (the first introduces a new discourse element).
(38)a. Then at the darkest hour dawned deliverance. Through the revolving doors swept Tom Pulsifer.
[S. J. Perelman, “The Customer is Always Wrong,” The Most of S. J. Perelman, p. 227. (New York: Simon and Schuster, 1958)]
420Georgia M. Green
b.One night there was a tap on the window. Mrs. Rabbit peeped through the window. Outside stood a little angel.
[Bruna]
c.Dumble vanished and in his place rose a dark, angry cloud of bees. They flew straight at the soldiers’ faces, and from the soldiers came yells of anguish, of sorrow, and of despair.
[Jay Williams, The King with Six Friends (New York: Parents’ Magazine Press, 1968)]
The particular discourse values of several other of the constructions in (29– 31) have been explored in some detail.13 For Japanese, Makino (2001) describes the conditions on the use of the formal nouns no and koto.
5 Reflections of Perceived Difficulty
Finally, speakers may take advantage of constructions like EXTRAPOSITION (39b) and HEAVY NP SHIFT (40b), which allow a constituent to appear at the end of the sentence to put the longest or most conversationally significant constituent last.
(39)a. Whether Kim will visit museums in France and Dana will go to concerts in Vienna, or Dana will visit museums in France and Kim will go to concerts in Vienna is unclear.
b.It is unclear whether Kim will visit museums in France and Dana will go to concerts in Vienna, or . . .
(40)a. Dana attributed a poem in which intuitions were compared to anemones and academic theories were described as battlements to Coleridge.
b.Dana attributed to Coleridge a poem in which intuitions were compared to anemones and academic theories were described as battlements.
It is not clear whether this option serves to make the sentence easier to articulate (cf. Olsen 1986) or simply easier to keep track of, or whether the motivation is altruistic – accommodating the addressee’s likely strategies or difficulties in parsing, or some combination of these. Length and discourse significance seem to be at least partially independent factors. Longer postposed noun phrases tend to sound better, as in (41a), even if they have no more semantic content, as in (41b), but of two noun phrases of equal length the more significantsounding sounds better, as (41c) shows. (See Arnold et al. 2000 for related discussion.)
(41)a. But they attributed to Blake The Rise and Fall of the Roman Empire/ ?Typee.
Some Interactions of Pragmatics and Grammar 421
b.The district attorney considers indictable Montgomery J. Jingleheimer-Smith III/?Rose Budd.
c.The committee has attributed to Margaret Thatcher an extraordinary poem/?the 27-line poem.
6Representing the Pragmatic Value of Syntactic Constructions
In the early years of generative grammar, the distribution of linguistic expressions in sentences was taken to be an exclusively formal matter, and the relevant notion of identity among expressions was assumed to be identity of form. However, as early as 1965, attempts were made to incorporate various kinds of pragmatic conditions into the framework then available for syntactic description. For example, G. Lakoff (1965), assuming that restrictions on distribution were all syntactic in nature, claimed that the unacceptability of beware in certain constructions (e.g. in (42)) reflected the fact that beware bore a syntactic RULE FEATURE which indicated that auxiliary inversion could not apply in clauses where beware was the main verb.
(42) #Did you beware of John?
Of course, Lakoff’s system would break down when confronted with the fact that pragmatic factors influence the acceptability of using syntactic constructions, such as inversion in embedded interrogatives. It cannot be claimed that know has a (different) rule feature which precludes its complement from appearing inverted, to account for the unacceptability of (25b), because inversion is fine in the complement of know in (25a). Lakoff (1965) also proposed STRUCTURAL DESCRIPTION FEATURES to constrain a structure from meeting the structural description of a transformational rule, but these would be equally inadequate to the task of excluding (43a), while allowing (43b).14
(43)a. #We need a registrar who can beware of phony addresses.
b.We need a registrar who will beware of phony addresses.
At the same time, addressing other issues, Chomsky (1965) proposed indexing nodes in constituent structure trees for coreference so that the kind of identity required for personal and relative pronouns could be represented as syntactic information.15
Approaches to accommodating other kinds of pragmatic conditions into syntactic descriptions involved representing contingent assumptions about realworld relations among situations as syntactic information (e.g. Green 1968). For example, to distinguish between the appropriateness of (44a) and (44b), one would have to have access to the proposition that being interested in sports (but not being interested in knitting) entails or implies the ability to tell a zone press from a fast break.
422 Georgia M. Green
(44)a. Jo isn’t interested in sports, and Bo couldn’t tell a zone press from a fast break either.
b.#Jo isn’t interested in knitting, and Bo couldn’t tell a zone press from a fast break either.
This proposal was taken seriously despite the fact that it clearly required making the rules of the grammar (insertion of the particles too and either) sensitive to properties of speakers (specifically, to whether they believed that some proposition implied some other proposition), and this involves a gross category error. A later proposal (Green 1973) to encode the implication as part of the deep structure – as was then being done at every turn to account for the syntactic constraints imposed by various illocutionary forces (e.g. R. Lakoff 1968, Ross 1970a) – was equally doomed. Morgan (1973a) showed that this approach led to theory-internal logical contradictions.
Meanwhile, linguists’ interpretation of Grice’s paper “Logic and Conversation” (Grice 1989: Chapter 2), which had been circulating underground for several years, prompted them to begin to describe relations between grammaticality and usage. Although Morgan (1975b) had demonstrated that making a strict separation in grammatical descriptions between constraints on form and constraints on usage wasn’t going to be as simple as it looked, given the fact that certain forms (like those in (45)) induced implicatures which are not induced by semantically equivalent forms (cf. (46)), this warning went largely unheeded.
(45)a. Do you have any idea how much that cost?
b.Why paint your house purple?
(46)a. Any idea how much that cost?
b.Why do you paint your house purple?
Gordon and Lakoff (1971) had interpreted Grice’s proposal as sanctioning the codification of likely implicatures into “conversational postulates” and the incorporation of speech act participants’ beliefs and intentions into syntactic derivations in the guise of constraints on derivations that referred to other possible derivations (a sort of precursor of today’s optimality syntax (e.g. Grimshaw 1997) and pragmatics (Blutner, this volume)). Gordon and Lakoff claimed that a rule of you-TENSE DELETION derived sentences like (45a) from structures similar to that of (46a) if and only if the logical structure L of (46a), taken in conjunction with CONI, a class of contexts, and the set of conversational postulates entails (47).
(47)Unless you have some good reason for doing VP′, you should not do VP′.
Gordon and Lakoff’s CONI encoded speakers’ intentions and beliefs as if they were semantic matters of truth. But that obscures the fact that the relation
Some Interactions of Pragmatics and Grammar 423
between the use of a form and its interpretation in context depends on the speaker’s and addressee’s beliefs and intentions about each other’s beliefs and intentions (Cohen and Perrault 1979, Cohen and Levesque 1990, 1991, Green 1996a) and not on any other kinds of contingent facts.
An approach that is more consistent with these facts and with the better understanding of implicature now available to us will minimize the number of grammatical constraints on the syntactic combination of grammatical categories and unify them with lexical and syntactic constraints on their semantics, and with construction-specific pragmatic constraints of the sorts discussed here. Such an approach would be, in effect, a complex function on a theory of communication which entailed the integration of such more or less universal principles as Grice’s Cooperative Principle and the strategies of relevance and quantity (cf. Horn 1984a) that derive from it (strategies for referring, predicating, focusing, etc.), with culture-specific interpretations (or implementations) of politeness principles (cf. Brown and Levinson 1987, Green 1993b). All of these aspects of pragmatics refer directly to language users’ intentions and beliefs, linking them to the conventions of usage (Morgan 1978) that the constructionspecific constraints encode. Such a treatment, in contrast to known predecessors, would not claim that sentences like (25b) are ungrammatical.
(25)b. #She already knows who did I appoint.
Rather, it would predict (a) that the use of such sentences will cause hearers to make certain inferences about their speaker, (b) that some of these may result in the sentence being considered inappropriate, given what else the hearer knows about the speaker and the subject matter, or contradictory, or ineffective for the purpose the hearer imputes to the speaker, or, in plain language, dumb, and (c) that the speaker is aware at some level of (a) and (b).
Thus, while the approach of the 1960s entailed claiming that a sentence like (25b) was ungrammatical because the auxiliary-inversion rule had applied in the complement of a verb constraining it from applying there, and the approach of the 1970s entailed claiming such a sentence was ungrammatical because inversion in the complement implied that the speaker believed that the referent of the subject of the embedding verb didn’t know the answer to the evoked question, contradicting the assertion of the whole sentence (that the referent of the subject has it figured out, and does know), the approach sketched here claims that sentences like (25b) are perfectly grammatical: they conform in every respect to the rules of syntactic combination that comprise the grammar. It claims that such sentences are nonetheless inappropriate, ineffective, or dumb because of the contradiction between what is asserted and what is implicated by the choice to use an inversion in an interrogative complement.
Pollard and Sag (1994) take a first step toward such an approach, treating speaker’s presuppositions and other categories of propositional attitudes as part of the representation for lexical and phrasal expressions. As discussed in
424 Georgia M. Green
Green (1994, 2000), such representations might be very detailed, incorporating much of the same sort of information as might be expressed in a Discourse Representation Theory representation (Kamp 1981, Kamp and Reyle 1993). One concern of Pollard and Sag (1994: 332–5), however, is that while background presuppositions have to be projected from lexical items to phrases containing them, it has been known since the early 1970s that the projection is not a function of tree geometry (Morgan 1973b, Karttunen and Peters 1979), or even of the semantic class of predicates and operators in the structural projection path. Morgan showed that neither the problem nor the solution is strictly linguistic, but depends instead on beliefs attributed by the interpreter to the speaker and agents and experiencers of propositional attitude verbs in the sentence. Morgan’s account, and Gazdar’s (1979) formalization of it, show that conversational implicatures of the utterance of the sentence limit the presuppositions of a sentence uttered in context to the subset of presuppositions associated with the lexical items in it that are consistent with the speaker’s assumptions and intended implicatures.
Conversational implicature, of course, is a function of a theory of human behavior generally, not something specifically linguistic (Grice 1989, Green 1993a), because it is based on inference of intentions for actions generally, not on properties of the artifacts (sentence and utterance tokens) that are the result of linguistic actions. Conversational implicatures arise from the assumption that it is reasonable (under the particular circumstances of the speech event in question) to expect the addressee to infer that the speaker intended the addressee to recognize the speaker’s intention from the fact that the speaker uttered whatever the speaker uttered. Thus, it would be naïve to anticipate that the filtering in the projection of presuppositions or other associated propositional attitudes could be represented as a constraint or set of constraints on values of some attribute of linguistic expressions and therefore as of the same character as, say, the constraints on the projection of agreement or subcategorization or unbound dependency information, precisely because conversational implicature is inherently indeterminate (Morgan 1973b, Gazdar 1979, Grice 1989).
This does not necessarily mean that a projection principle for pragmatic information is logically impossible. Background propositions of a phrase can be computed as a conjunction of the background propositions of all the daughters, along the lines suggested by Pollard and Sag (1994: 333) and Wilcock (1999). This sort of context inheritance principle would be completely consistent with the inherently indeterminate character of Gricean conversational implicature. If that conjunction should happen to contain predications that are inconsistent with each other, or predications that are inconsistent with what is predicated by the sentence as a whole, that does not pose a logical problem, or a problem for a formal theory. It is not even a linguistic problem. It is a practical sort of problem for a human being who wants to construe the speaker’s behavior in uttering the sentence as rational.16 Doing that requires using knowledge of principles of sense and/or reference transfer (Nunberg 1995, this volume) and
Some Interactions of Pragmatics and Grammar 425
lexical rules, as well as beliefs about what is sensible and what is silly. For example, one way of resolving a conflict involving lexical presuppositions would be to interpret one or more of the presupposition-bound phrases involved as figuratively intended in a way that allows propositions intended to be conveyed to be regarded as all true. In any case, the resolution of such contradictions is precisely what the Cooperative Principle was invented for (Grice 1989, Green 1996b) and what the computation of implicatures is about, as Morgan and Gazdar have demonstrated.
ACKNOWLEDGMENT
This work owes much to Jerry Morgan, whose deep understanding of these issues has influenced my articulation of them in uncountable ways. Some of the discussion in this chapter is reworked from material that has appeared elsewhere (e.g. Green 1982a, 1996a, 2000).
NOTES
1Cf. Lees and Klima (1963).
2Cf. Green (1990, 1996a) for discussion.
3An extensive, and still incomplete, catalog may be viewed at http:// mccawley.cogsci.uiuc.edu/~green.
4The choice of addressee and hearer in this sentence is not accidental. Speakers plan speech with a particular audience in mind, but everyone who hears it has access to the same rules for interpreting it.
5There are also well-known cases where the number agreement morpheme on a verb with an apparently plural subject induces inferences from the fact that one choice was made rather than another, inferences that, in fact, affect the truth conditions, not just “mere” pragmatic differences (Morgan 1972a, 1972b, Pollard and Sag 1994). Thus, (i) would be used to extol the virtues of two foods,
while (ii) is about a single unusual concoction.
(i)Pickles and ice cream taste good.
(ii)Pickles and ice cream tastes good.
6The construction names in SMALL CAPITALS are those familiar in modern generative grammar. See Green and Morgan (2001: Appendix) for further references.
7For some more recent treatments, see Prince (1981a), Zaenen (1982), and Horn (1986).
8It has generally not been claimed that the Old Information First principle is a universal principle, though it may be a universal tendency driven by the interpersonal nature of discourse. Most of the illustrations have come from Czech, English, French, and Japanese.
426 Georgia M. Green
9Specifically, the backward-looking center must stand in a salient scalar relation to the partially ordered set constituting the forward-looking centers. A partially ordered set is a set whose members are all related by some ordering relation which is transitive, and either reflexive and antisymmetric, like “is as tall as or taller than,” or irreflexive and asymmetric, like “is taller than” (cf. Ward 1988 52ff., Hirschberg 1991). Salient scalar relationships include any where one element
is higher or lower than another on some scale, or they are incomparable alternative values,
but there are values higher or lower than both. Some examples are the set/subset relation, part/whole, type/subtype, entity/attribute,
and of course, greater than and less than.
10The example is adapted from a passage in Nixon’s Six Crises cited by Ward (1988: 57).
11Or speaker; cf. Green (1982b) for discussion of the genres and
registers where these constructions are found.
12This common function of inversions may be what has misled some writers (e.g. Longuet-Higgins 1976) into thinking that inversions after directional phrases must describe a character coming into (the narrator’s) view. Inversions have other uses, and we do find such inversions as (i)–(iii), which describe a character going out
of view.
(i)Then off marched the little tailor, cocky as could be, with his thumbs thrust through his boasting belt.
(ii)Into the forest ran the four, and soon they could be seen no more.
(iii)Off across the grass ran the three little girls.
Examples (i) and (iii) are from children’s books whose exact titles I cannot locate.
13Cf. for example Prince 1981c,
1984 (TOPICALIZATION and LEFT DISLOCATION); Prince 1978 (CLEFT
and PSEUDO-CLEFT); Milsark 1977, Napoli and Rando 1978, Aissen 1975 (THERE-INSERTION). Further references may be found in Green and Morgan (2001: Appendix).
14The pragmatic condition on beware, still mysterious, seems to involve (not all that surprisingly, given its meaning) reference to awareness of a threat. Cf. Green (1981) for some preliminary observations. The absence of inflected forms explains the unacceptability of such forms as *We bewore of the bandersnatch, *We bewared of the dog.
15McCawley (1968) argued that referential indices have the structure of sets rather than being discrete units; Postal (1967) outlines some reasons to be skeptical about Chomsky’s notion. Morgan (1968, 1970) provided early demonstrations that the kind of identity required varies from syntactic construction
to syntactic construction.
16Obviously, it is also a practical problem of enormous dimensions for any automated natural language processing system that seeks to interpret natural language input,
if for no other reason than its inescapable dependence on the encyclopedic knowledge that human natural language users take for granted.