8
Information structure
Suppose you’re sitting in your living room with a good book—say, the collected stories of Edgar Allan Poe. And you encounter the following:
(1) 
It might occur to you that Poe could instead have written:
(2) 
In fact, there are a great number of options from among which he could have chosen, such as (simplifying a bit for clarity):
(3) 
We could go on for quite a while; here I’ve used only a small number of the syntactic constructions available to a speaker of English (canonical word order, preposing, postposing, dislocations, and clefts). There are two things to notice at this point: First, all of these options are in fact grammatical. This might not be immediately obvious, because you may feel that you’d never encounter a sentence like, say, (3k). But consider (4):
(4) 
It should come as no surprise that context makes all the difference in whether a given construction will sound acceptable.
The second thing to notice is that all of the examples in (3) are semantically equivalent. So why would a language provide so many different ways of saying exactly the same thing? And we see this even in the most common constructions:
(5) 
(6) 
In (5a) the sentence is in canonical word order—the default, basic word order of English, essentially Subject-Verb-Object (or SVO). In (5b) we see the passive variant of the same sentence, which is semantically equivalent. Likewise for (6), which has the ditransitive verb gave: The variant in (6a), with a prepositional phrase, and the variant in (6b), with an indirect object, have the same truth-conditions. (We could argue about some cases—e.g., whether Everyone read three books has the same truth-conditions as Three books were read by everyone—but certainly the examples in (3)–(6) do not involve differences in truth.)
So why bother having the options? We’ve already hinted at the answer: Although two options may be semantically equivalent, they may nonetheless be pragmatically distinct—and therefore felicitous in distinct contexts. We’ll see in this chapter that having constructions that are felicitous in different contexts gives us a way to help the hearer to process the discourse more easily, in (at least) two ways: First, by putting familiar information before unfamiliar information; and second, by telling the hearer whether some expression presents information as already known or presupposed, or as new and informative.
Given and new information
Many languages, including English, tend to present ‘given’ information before ‘new’ information, presumably because it’s easier for a hearer to process new information by means of its connection to what’s already known. There has been disagreement over the years regarding exactly what ‘given’ and ‘new’ mean, but in general ‘given’ information is taken to be what is already known by or familiar to both interlocutors, whereas ‘new’ information constitutes what is not already known and is therefore new and informative, and therefore usually the focus or main point of the utterance. The extent to which something is already (believed to be) known or presupposed, or to which it is being presented as new, informative, or focused, is called its information status. A speaker’s choice of structures to manipulate the order in which information is presented is called information packaging (Chafe 1976). And the resulting order of information in a sentence is its information structure. We’ve seen that the canonical word order in English is SVO; constructions that let the speaker change this order are called noncanonical-word-order constructions.
Prague School linguists in the 1960s and 1970s proposed a notion of Communicative Dynamism (CD)—roughly, the informativeness of an expression, or the extent to which it moves the communication forward—in which the level of CD was argued to increase from the beginning of the sentence to the end. This principle was formulated as the Given-New Contract:
Given-New Contract: Given information tends to appear closer to the beginning of a sentence, while new information tends to appear closer to the end of a sentence. (Halliday 1967, Halliday and Hasan 1976)
Canonical word order (CWO) is, in general, unconstrained with respect to information status; with rare exceptions, a CWO sentence will be felicitous in any information-structural context. Noncanonical-word-order (NWO) constructions, on the other hand, are typically constrained in their distribution and serve some discourse-functional (i.e., information-structural) purpose.
Various dichotomies have been proposed to distinguish between given and new information; but Prince (1981) argues that givenness is not a dichotomy at all, and describes and exemplifies seven different levels of givenness. In Prince 1992 she improves on this taxonomy by distinguishing three basic types of givenness: Information, she says, may be presupposed or focused, it may be discourse-old or discourse-new, and it may be hearer-old or hearer-new. Her notion of presupposition is closely related to that discussed in the last chapter and will be discussed in more detail in this one; for the moment we will focus on her distinction between discourse-status and hearer-status. Information is discourse-old if it has previously been evoked in the discourse, or if it is inferentially related to such information. Otherwise, it’s discourse-new. Information is hearer-old if the speaker believes it is previously known to the hearer; if not, it’s hearer-new.
What’s discourse-old isn’t necessarily hearer-old, and vice versa. For example, if I mention right now that Chicago has had a really rough winter, Chicago represents information that is discourse-new (Chicago not having been previously mentioned in this discourse) but hearer-old (since I assume you have prior knowledge of Chicago). The opposite situation—information that’s discourse-old but hearer-new—is a little trickier to imagine, and in fact Prince (1992) suggests that this may be an impossible status, on the grounds that if the hearer is paying attention, then anything that’s been mentioned in the discourse should be known to the hearer. But later research (Birner 1994, 2006) shows that information that hasn’t been explicitly mentioned in the discourse, but which instead can be inferred from what has been mentioned (what Prince 1981 terms inferrable information) is also treated as discourse-old: Both explicitly evoked information and inferrable information show up in the same contexts, and are treated alike with respect to their felicity in NWO constructions. So, consider (7), in which the direct object the combs is preposed to the front of the sentence:
(7) 
Preposing requires the preposed constituent to represent discourse-old information (Ward 1988, Birner and Ward 1998). In (7a), the combs have been mentioned in the first sentence, so the noun phrase representing them in the second sentence is easily preposed. In (7b), the combs haven’t been previously mentioned, but they stand in an inferential relationship with something that has been—the set of grooming tools, of which they are a member. If you replace the combs with something that cannot possibly be part of the set of grooming tools (or the set of items being put away), the infelicity is hopeless:
(8) 
In short:
So let’s consider some of the NWO constructions of English and how they satisfy the Given-New Contract.
Preposing and postposing constructions
A preposing construction is a structure in which some subcategorized element in the sentence appears in a noncanonical, preverbal position. (A subcategorized element is essentially one that’s ‘called for’ by some other element, in the sense that the verb put calls for a direct object and a prepositional phrase.) All of the examples in (9a)–(9c) are preposings:
(9) 
As we have seen, a preposed phrase is required to be discourse-old. In (9a)–(9c), her stuffed animals, into the closet, and the closet have all been evoked in the preceding sentence, so any of them can be felicitously preposed. Because discourse-old information also includes information that stands in some inferential relation (essentially any set-based relation) with prior information, examples such as those in (10) are equally felicitous:
(10) 
The sentences in (10a)–(10c) are preposings as well, but here the preposed information hasn’t been explicitly mentioned in the preceding discourse, but stands in a set-based relationship with information that has been: A closet is a member of the set of things typically found in a bedroom, and stuffed animals are a member of the set of toys.
It’s not always an object of the verb that’s preposed; in fact, the verb itself can be:
(11) 
Survived they had is a preposed variant of the CWO they had survived, with the verb preposed.
Postposing is a bit more complicated, as it involves placing the logical subject in postverbal position. Since English syntax requires a subject, a ‘dummy’ or semantically empty there is pressed into service:
(12) 
In (12a), the canonical ordering of the last clause would be an almost open inquiry into…began. Here, however, the logical subject an almost open inquiry…is postposed, and the subject position is filled with a dummy there. Notice that by calling this a ‘dummy’ element, I mean that it has no semantic content; the author isn’t saying that the inquiry began in some particular place. Likewise, in (12b), the sentence-initial there in the second sentence doesn’t mean ‘in that location’; there’s a second there in the sentence that serves that purpose.
The other thing to notice is that the postposings in (12a) and (12b) differ in their main verb. In (12a) this verb is began; in (12b) it’s the copula was. (A copula is a form of be.) A postposing with an intransitive verb, such as that in (12a), is called a presentational sentence; a postposing with a copula, such as that in (12b), is called an existential sentence. The two differ subtly in their information-structural constraints: Existentials require that the postposed noun phrase represent hearer-new information, while presentationals require only that they represent discourse-new information (Birner and Ward 1998).
We see, then, how these constructions serve to maintain the Given-New Contract, with preposing giving speakers a way to place ‘given’ information at the front of the sentence and postposing giving them a way to place ‘new’ information at the back. It’s also possible to do both within the same sentence, in argument-reversing structures like ‘long’ passives (13a) and inversion (13b):
(13) 
Long passives are those that include a by-phrase (e.g., by a U.S.-Russian team of scientists), and they occur with transitive verbs like discover. Inversions are cases in which the canonical subject and some canonically postverbal phrase have essentially switched places (as in Above his head hung a massive seagull); they occur with intransitive verbs such as hang and lounge and also with copulas. Both structures reverse the order of the subject and some canonically postverbal phrase. And for both structures, either the preposed constituent must be discourse-old or the postposed constituent must be discourse-new (or both); the one combination that is infelicitous is discourse-new preposed information with discourse-old postposed information, as in (14):
(14) 
Here, the preposed information (a pair of hedgehogs) is discourse-new and the postposed information (the seagull) is discourse-old, and the inversion is infelicitous. The same holds true for passivization:
(15) 
In (15a), the passive it was told by Egyptian officials…contains a preposed discourse-old phrase (it) and a postposed discourse-new phrase (Egyptian judicial officials), and the passive is felicitous. If we swap the status of these two phrases, as in (15b), the passive is infelicitous.
So the preposed constituent in an inversion or passivization is not required to be old, nor is the postposed constituent required to be new; but one or the other must be the case. That is, either the constraint on preposing or the constraint on postposing must be satisfied. In essence, inversion and passivization can be seen as variants, or alloforms, of both preposing and postposing, and will therefore always satisfy one or the other constraint in any given case, depending which construction it is serving as a variant of (Birner 2018). Regardless, all of these structures—preposings, postposings, long passives, and inversions—give the speaker a way of preserving the Given-New Contract.
Presupposition/focus constructions
We saw in the last chapter that among the triggers for presupposition are cleft constructions, as in (16):
(16) 
Here we see an it-cleft (16a), a wh-cleft (16b), and a reverse wh-cleft (16c)—also known as a cleft, a pseudocleft, and reverse pseudocleft, respectively. They’re called ‘clefts’ because they ‘cleave’ the canonical sentence (17) into two parts, putting one into the foreground and one into the background.
(17) 
By using one of the clefts in (16), the speaker backgrounds, or presupposes, the proposition ‘you need X’, where X is a variable; meanwhile, they foreground, or focus, the thing needed, the instantiation of X, i.e., ‘a good night’s sleep’. These are therefore considered presupposition/focus constructions. The presupposition ‘you need X’ is an open proposition (OP), i.e., a proposition missing one or more elements (Prince 1986); the variable represents the missing element, which is then provided by the focus (here, a good night’s sleep). In short, then, in a context in which ‘you need X’ is salient, the cleft gives the speaker a way to presuppose that OP and provide the instantiation of X, which is to say, the focus.
Another possible way to think of clefts is in terms of the Question Under Discussion (QUD; Roberts 2012). Here the idea is that at any point in a given discourse, there are certain questions that are not only salient or common ground, but actually at issue. For the clefts in (16) to be felicitous, the most salient QUD would be essentially ‘what do you need?’ There may be many other topics that are salient or constitute common ground at the moment, but the utterances in (16) take this question to be the one that’s currently at issue, the QUD. Note that the QUD isn’t the same thing as a presupposition as discussed in the previous chapter; when I say The King of France is bald, it presupposes that there’s a King of France, but it doesn’t assume that the existence or nonexistence of the King of France is the QUD. Certainly if I tell you My sister just called, I may be presupposing that I have a sister, but whether I have a sister isn’t the primary issue at hand, the QUD. In short, there are many ways to think about what’s given and new, what’s backgrounded and foregrounded, what’s at issue or under discussion or being asserted; these concepts are not at all the same (far from it), and linguists are hard at work trying to determine which concepts are most relevant to which phenomena.
What’s interesting1 is that even though all clefts ‘cleave’ a sentence into a presupposition and a focus, they’re not all equally felicitous in all contexts:
(18) 
And of course, as we saw in the last chapter, not all of the presuppositions in clefts constitute common ground. For example, continuing with your Poe reading, you might encounter this:
(19) 
Technically, the fact that the Prince Prospero entertained his friends at a masked ball constitutes the presupposition, while the initial information concerning its timing, the seclusion, and the pestilence constitutes the focus—even though the masked ball is new information and the pestilence is already known in the context of the story. You’ll recall that such cases are known as ‘informative-presupposition it-clefts’, and they’re a curious usage compared to other uses of the presupposition in a cleft; for example, in (18a) it’s presupposed that the speaker would like something to drink, and Pepsi is the new and informative focus. And of course it’s not the case that all it-clefts are of the informative-presupposition type; in (16a) the presupposition ‘you need X’ is presupposed and ‘a good night’s sleep’ is focused.
Interestingly, preposing and inversion, which we became acquainted with in the previous section, are also presupposition/focus constructions, in that they generally require the presence of a salient open proposition in the discourse. Ward (1988) shows that such an OP is necessary for preposing, as in (20):
(20) 
In (20a), the preposing is pass I did, which preposes the verb pass; the canonical variant would be I did pass. Here, the presupposed OP is ‘I {did/didn’t} pass’, with the final did supplying the focus, i.e., filling in the missing value. This is made salient by It was necessary to pass, which raises the issue of whether I did in fact pass. In the absence of this OP, as in (20b), the preposing is infelicitous. We see a similar effect in inversion:
(21) 
In (21a), the inversion is in the second sentence, where the CWO variant would be Cameraman Alain Debos, 45, and soundman Nick Follows, 24, were wounded yesterday. The OP is ‘X were wounded yesterday’, and this is clearly rendered salient by the first sentence. In (21b), where this OP is not rendered salient by the first sentence, the inversion is infelicitous.
There are two exceptions to this generalization. The first is that the OP requirement is lifted for both preposing and inversion when the preposed phrase is locative:
(22) 
In (22a), there’s no salient OP to the effect that someone dispenses something, nor is there in (22b) a salient OP that something flows somewhere. It’s curious that preposing and inversion share not only an OP requirement but also this amnestying in locative contexts, until you remember that inversion is itself a variant of both preposing and postposing. So when it’s serving as a variant of preposing, it wouldn’t be at all surprising that it would share both the constraints on preposing and the conditions under which those constraints are lifted.
And that, in turn, brings us to the second exception to the OP constraint. Since inversion sometimes serves only as a variant of postposing, and postposing is subject to no such OP requirement, it would make sense that in those cases, inversion would have no such requirement either. How does one recognize those cases? One way to identify such a case is to look for instances of inversion in which the preposed and postposed constituents are both discourse-new. In that case, the constraint on preposing is not met, though the constraint on postposing is, so the inversion must be serving as a postposing. And in such cases, we do indeed find that there is no OP requirement. For example, it’s perfectly acceptable to start a news story with an OP-less inversion such as that in (23):
(23) 
In short, viewing inversion through its role as a variant of preposing and postposing constructions can help to explain otherwise puzzling aspects of its distribution in discourse.
Other constructions and constraints
Although obviously there isn’t space to consider every noncanonical construction in English, much less in other languages, a couple of points bear mentioning. First, it is not the case that every construction that places information in a noncanonical position to the left or the right of its canonical position counts as a ‘preposing’ or ‘postposing’ construction. For example, picking up where we left Mr. Poe, we find that the it-cleft is immediately followed by another construction:
(24) 
The canonical variant of the final sentence would be That masquerade was a voluptuous scene. Here, that masquerade is instead placed in sentence-final position, which might make you think it’s a postposing. But there are two things worth noticing: First, here the subject position is not filled by a semantically empty ‘dummy’ element, but rather by a referential pronoun. That is, it is coreferential with that masquerade. Second, that masquerade doesn’t satisfy the requirement on postposed constituents, i.e., that they represent new information. The masquerade has been evoked in the prior sentence, not to mention in the subject of the current sentence (by the coreferential pronoun it). Since this structure is both structurally and functionally distinct from the postposings we discussed earlier, it’s safe to say that it constitutes an entirely different construction. This construction is called right-dislocation (RD), and not only does it permit discourse-old information in the right-dislocated position, it appears to require it:
(25) 
Here, the right-dislocated phrase the food he served represents information that has not been evoked in the prior discourse, and the RD is infelicitous. Clearly, then, RD doesn’t share postposing’s requirement of new information, nor its structure, and it is therefore a distinct construction.
Just as a right-dislocation construction exists that is both functionally and structurally distinct from postposing, English also provides a left-dislocation (LD) construction that is both functionally and structurally distinct from preposing:
(26) 
Here, both of the final two sentences are LDs. Like preposing, they place a phrase in a noncanonical position at the beginning of the clause; so, the CWO variant of gallstones, you have them out and they’re out would be You have gallstones out and they’re out. But unlike preposing (and like RD), the noncanonically positioned phrase’s canonical position in LD is filled by a coreferential pronoun: In short, them replaces gallstones in you have gallstones out. LD also differs from preposing in that it does not require the initial constituent to represent discourse-old information:
(27) 
Here, the final sentence is a left-dislocation, and in this case, the landlady is discourse-new. Prince (1997) argues that there are three distinct categories of LD, with three distinct functions. The LD in (27) exemplifies her Type I, ‘simplifying’ LDs, which serve to take discourse-new information and create a separate processing unit for it, rendering it discourse-old, which in turn allows it to appear in its normal sentence position as a pronoun. Prince’s Type II and Type III LDs are illustrated in (28a) and (28b), respectively:
(28) 
Prince’s Type II LDs trigger an inference to a partially ordered set (see Hirschberg 1991); in (28a), the second sentence evokes a set of three groups of mice, and the left-dislocated one in the third sentence is a member of that set. And finally, Prince’s Type III LDs are cases in which a coreferential pronoun is inserted to rescue an otherwise ungrammatical preposing, such as in (28b), where the preposing without the pronoun would be the ungrammatical *My copy of Anttila I don’t know who has.
In short, while the Given-New Contract and presupposition/focus structure account for the distribution of a wide range of constructions in English, they do not account for all NWO constructions. Other factors that influence word order in English include grammatical ‘weight’ (relative length and/or complexity), semantic connectedness, and avoidance of ambiguity (see Wasow 2002, Wasow and Arnold 2011).
Finally, it should be noted that just as different languages have different basic word orders, different languages offer different sets of noncanonical word orders, as well as (of course) different constraints on their use. This brings up the interesting question of crosslinguistic research (and possible crosslinguistic generalizations) concerning information structure and noncanonical word order.
Kaiser and Trueswell 2004, for example, in a study of Finnish noncanonical constructions, shows that readers experience greater processing difficulties when reading noncanonical sentences outside of a supportive discourse context (that is, a context that renders the necessary constituents discourse-old, discourse-new, etc.). Ward 1999 examines two postposing constructions in Italian and finds that ci-sentences have a syntactic structure parallel to the English existential but are subject to the same constraint as the English presentational, showing that the mapping of constraints onto constructions is language-specific.
This language-specificity in turn has some interesting ramifications, and also raises some interesting questions. For example, Birner and Mahootian 1996 examines a structure in Farsi that corresponds in terms of word order to English preposing; that is, it has a preposed prepositional phrase followed by the subject and finally the verb. Thus, it corresponds to an English sentence like On the table a candle burned. Functionally, however, it corresponds to English inversion, requiring that either the preposed constituent be discourse-old or the postposed constituent be discourse-new; that is, the only disallowed combination is new-before-old. What’s interesting is that Farsi has a basic word order of SOV (subject-object-verb), with the ‘O’ slot being where that prepositional phrase canonically resides. So it’s unclear whether the noncanonical structure whose ordering is OSV (the equivalent of our On the table a candle burned) should count as simply a preposing of the prepositional phrase or an inversion of the prepositional phrase and the subject; either one, applied to the canonical SOV, would result in a noncanonical OSV. The fact that the noncanonical OSV variant corresponds functionally to English inversion rather than to English preposing might suggest that the Farsi construction is an inversion rather than a preposing—but this assumes that the construction-function mapping holds crosslinguistically, which as we just saw for English and Italian cannot be assumed. In short, there’s a great deal of research still to be done to determine the relationship between structures and functions crosslinguistically.