Discourse and Information
Structure
GREGORY WARD AND BETTY J. BIRNER
0 Introduction
In addition to deciding what to say, the speaker must decide
how to say it. the central premise of the study of the relationship between
syntax and function of discourse is that the use of the speaker of the
particular structural options limited by certain aspects of the context of the
speech. Working in discourse has found a certain range of discourse functions
served by the individual syntax constructions.1 More recently, in Birner and
Ward (1998) we examine generalizations that apply throughout the construction,
identify ways in which a given functional principles embodied in the various
similar but different construction.
1 Theoretical Framework
English, like many other languages,
shows a tendency to order “given” information before “new” information in an
utterance. Indeed, Prince (1981a: 247) posits a “conspiracy of syntactic
constructions” designed to prevent NPs that represent relatively unfamiliar
information from occupying subject position (see also Kuno 1971, interalia).
Chafe (1976) defines given information as “that knowledge which the speaker assumes
to be in the consciousness of the addressee at the time of the utterance,” while
new information is defined as “what the speaker assumes he is introducing into
the addressee’s consciousness by what he says” (1976: 30). Other notions of
given information have relied on such notions as predictability and shared
knowledge, or assumed familiarity (see Prince 1981a). In reviewing the
literature on givenness in discourse, Prince (1992) finds that three basic
approaches may be distinguished, which she terms focus/presupposition,
hearer-old/hearer-new, and discourse-old/discoursenew.
1.1 Focus/Presupposition
A focused constituent is realized
intonationally with some kind of prosodic prominence, generally unclear accent.
Presupposed information is the complement of focus: it represents the
information that the speaker assumes is already part of the common ground, i.e.
either salient or inferable in context. A presupposition is a proposition
that is presupposed in this way.Because utterances are intended to be
informative, the presupposition typically does not exhaust the information in
the utterance; instead, the proposition being presupposed is “open” – that is,
lacking certain information. Such a proposition is represented with a variable
in place of one or more constituents. For example, the utterance in (1a) would
give rise to the presupposed open proposition (OP) in (1b), in the sense that a
person hearing (1a) would immediately thereafter be licenced to treat (1b) as
part of the common ground: (1) a. Pat brought those cookies to the BBQ. b. Pat
brought X to the BBQ.
Although only a single word, or
syllable, of the focus bears nuclear accent, the focusitself can be indefinitely
large; consider (2): (2) Pat brought a bag of those yummy cookies from Treasure
Island to the BBQ. In a context in which the speaker has been asked What did
Pat bring?, the focus in (2) would be a bag of those yummy cookies from
Treasure Island. It is also possible for a clause to have more than one focus,
as in the exchange in (3): (3) A: Who brought what to the BBQ? B:
Pat brought cookies. The presupposition in this case is X brought Y, and
Pat and cookies are foci. Notice that Pat need not
represent entirely new information in order to count as new in this context.
Even if Pat is salient in the discourse, Pat here is new as an
instantiation of the variable in the presupposition. In effect, to say that Pat
represents new information in this way is to say that the proposition Pat
brought cookies is (believed to be) absent from the hearer’s mental store
of propositions, despite the presence of the proposition X brought Y. Not
all utterances involve presuppositions; for example, (2) may felicitously be uttered
in a context in which it is not presupposed that anyone brought anything. In such
a context, the entire utterance may be considered the focus (often called
“broad focus”).
1.2 “New
to the discourse” vs. “new to the hearer”
Noting that a two-way division of
information into given and new is inadequate, Prince (1992) offers a pair of
cross-cutting dichotomies which classify information as, on the one hand,
either “discourse-old” or “discourse-new” and, on the other hand, either
“hearer-old” or “hearer-new.” Discourse-old information is that which has been
evoked in the prior discourse, while hearer-old information is that which the
speaker believes to be present within the hearer’s knowledge store.2 This
distinction captures the fact that what is new to the discourse need not be new
to the hearer (cf. Firbas 1966; Chafe 1976; Lambrecht 1994); that is, an entity
may be familiar to the hearer yet new to the discourse. Thus, consider a simple
discourse-initial utterance such as (4): (4) Last night the moon was so pretty
that I called a friend on the phone and told him to go outside and look. Here, the moon represents information that
is discourse-new but hearer-old, denoting an entity that has not been evoked in
the prior discourse but which can be assumed to be known to the hearer; a friend represents information that
is both discourse-new and hearer-new, having not been previously evoked and
also being (presumably) unknown to the hearer; and him represents information that is discourse-old and (therefore)
hearer-old, having been explicitly evoked in the previous clause (as a friend). The status of what Prince calls “inferable” information
(e.g. the phone in (4), since
people are typically assumed to have telephones) is left unresolved in Prince(1992)
and will be discussed below.
We will call the linguistic or
situational material that licences the inference to the anchor the trigger (Hawkins 1978).6 As we have
seen, this inference may be based on a poset relation (as in (6a)), but it need
not be (as in (7)). The inference may be triggered by one or more items, one of
which may be the link itself. Thus, in (6a),
mention of the kitchen alone does not give rise
to the poset {elements of a house}, since, if it did, every utterance of an NP
would give rise to a cognitive explosion of instantaneously constructed
part/whole relations in which the referent participates (Fraurud 1990). Rather,
it is not until the speaker utters on
the counter that mention of
the kitchen and the counter
combine to evoke the poset that relates the two. Notice, finally, that it is
entirely possible for the trigger, anchor, and link to all represent the same
information, as in (8):(8) On one of September’s last blast furnace days, Emil
Peterson parked his car along a quiet street in the tiny Delaware County burg
of Eddystone and pulled a yellow plastic bucket from the back seat. In it he had expertly wedged an assortment
of brushes and cans of cleanser, a
hollyberry room deodorizer, knives, scissors, a couple of no-slip no-crease pants hangers and a box containing a
boulder-sized zircon ring.
(Philadelphia Inquirer, October
2, 1983) Here, the trigger a yellow
plastic bucket evokes a singleton set containing the bucket as its only
member. This set is the anchor, which in turn is related (trivially) to the
link it via a linking relation
of identity.7 Thus, even cases where the machinery of posets Gregory Ward and Betty J. Birner and
linking relations may not seem necessary are nonetheless consistent with this account, allowing the development of
a unified theory. With these
theoretical primitives in hand, we can now proceed to see how they apply to some of the noncanonical
constructions of English. Our analysis is based on a combined corpus consisting of several thousand naturally
occurring tokens collected over
a period of approximately ten years. The data can be described as more or less standard American English and
were drawn from a wide range of sources. Whenever possible, the prior and subsequent context was noted for each
token. Data were collected from
both speech and writing; the written sources include newspapers, magazines, novels, nonfiction books,
academic prose, and portions of the Brown
Corpus (Kucera and Francis 1967). Spoken data were drawn from personal
conversations, films,
interviews from Working (Terkel
1974), transcripts of the 1986 Challenger
Commission meetings,8 and a variety of television and radio programs.
2
Preposing
As we use the term, a “preposing”
is a sentence in which a lexically governed phrasal constitutent appears to the
left of its canonical position, typically sentence-initially (Ward 1988).9 Extending
the theory of preposing presented in Ward (1988), we claim that felicitous
preposing in English requires the referent or denotation of the preposed
constituent to be anaphorically
linked to the preceding discourse (see Prince 1981b, 1984; Reinhart 1981;
Vallduvà 1992). The information conveyed by the preposed constituent can be
related to the preceding discourse in a number of ways, including such
relations as type/subtype, entity/attribute, part/whole, identity, etc. These
relations can all be defined as partial orderings, and in Ward (1988) it is
argued that the range of relations that can support preposing are all poset
relations:(9) Customer: Can I get a bagel? Waitress: No, sorry.
We’re out of bagels. A bran muffin I can give you. (serviceen counter) Here,
the link (a bran muffin) and trigger (bagels) stand in a poset
relation as alternate
members of the inferred anchor
set {breakfast baked goods}. The link could also have been explicitly mentioned
in the prior discourse, as in (10):(10)
A: Can I get a bagel?
B: Sorry – all out.
A: How about a bran muffin?
B: A bran muffin I can give you.
Here, although the link a bran
muffin is coreferential with the trigger explicitly evoked in A’s second
query, the salient linking relation is not identity. Rather, the link is related
via a type/subtype relation to the anchoring set {breakfast baked goods}, of which
both bagels and bran muffins are members. Some types of preposing also
permit links to anchors with a
single member.
In addition, Ward (1988) shows
that certain types of preposing constructions require a salient or inferable
open proposition in the discourse (see also Prince1981b, 1984). The variable in
the OP is instantiated with the focus, which must be a member of a contextually
licenced poset. Preposings can be classified into two major types based on
their intonation and information structure: “focus preposing” and “topicalization.”
The preposed constituent of focus preposing contains the focus of the
utterance, and bears nuclear accent; the rest of the clause is typically deaccented.10
Topicalization, on the other
hand, involves a preposed constituent other than the focus and bears
multiple pitch accents: at least one on the preposed constituent and at least
one on the (nonpreposed) focus.11 Nonetheless, both types of preposing
require a salient or inferable OP at the time of utterance for felicity.
Consider first the focus preposing in (12), where the focus is contained within
the preposed constituent: (12) A: Where can I get the
reading packet?
B: In Steinberg. [Gives directions]
Six dollars it costs. (two students in conversation) The preposed
constituent in this example, six dollars, contains the nuclear accent, which
identifies it as the focus of the utterance:
(13) OP = It costs X, where X is
a member of the poset {prices}.“It costs some amount of money.” Focus = six
dollars
Here, six dollars serves
as the link to the preceding discourse. Its referent is a member of the poset
{prices}, which is part of the inferable OP in (13). The OP can be inferred on
the basis of the prior context; from mention of a reading packet, one is licenced
to infer that the packet costs some amount of money. While the anchoring
poset {prices} is discourse-old,
the preposed constituent itself represents information that has not been
explicitly evoked in the prior discourse. In the case of focus preposing, then,
since the anchoring poset must be discourse-old yet the link is the focus (and therefore
new), it follows that the poset must contain at least one other member in
addition to the link.
3
Postposing
As used here, the term
“postposing” denotes any construction in which a lexically governed phrasal
constituent appears to the right of its canonical position, typically but not
exclusively in sentence-final position, leaving its canonical position either empty
or else occupied by an expletive (Birner and Ward 1996). The postposing constructions
we will concentrate on are those in which the logical subject is postposed and
the expletive there appears in
the canonical subject position – i.e. what have
traditionally been known as
existential and presentational there-sentences,
as in (16a) and (16b), respectively: (16) a. “There’s a warm relationship, a great respect and trust” between
[United Air
Lines]’s chairman, Stephen M.
Wolf, and Sir Colin Marshall, British Air’s chief executive officer, according
to a person familiar with both sides. (Wall
Street Journal, August 23,
1989) b. Not far from Avenue de
Villiers there lived a foreign doctor, a specialist, I understood, in midwifery and gynecology. He was a
coarse and cynical fellow who had called me in consultation a couple of times,
not so much to be enlightened by my superior knowledge as to shift some of his
responsibility on my shoulders. (Munthe, A. 1929: 143. The Story of San Michele. London: John Murray).
3.1 English
existential there-sentences
As noted by Prince (1988, 1992)
and Ward and Birner (1995), the postverbal NP of existential there-sentences is constrained to
represent entities that the speaker believes are not familiar to the hearer:
(17) What can happen is a hangup
such as Rocky Smith ran into, as the independent hauler was traversing Chicago
with a load of machinery that just had to get to a factory by morning. “There was this truck in front of me
carrying giant steel coils, and
potholes all over the place,” he remembers. (Wall Street Journal, August 30,1989).a. I have some news you’re
going to find very interesting. #There
was on the panel your good friend
Jim Alterman. b. President Clinton appeared at the podium accompanied by
three senators and the Speaker of the House. #There was behind him the vice president.
The PVNPs in these examples
represent entities that are new to the discourse, but presumably familiar to
the hearer, and the existential there-sentences
are unacceptable. Now consider there-sentences
whose PVNPs are not only hearer-old but also discourse-old.
.
4
Argument
Reversal
While preposing involves the
noncanonical leftward placement of a constituent, and postposing involves the
noncanonical rightward placement of a constituent, argument reversal
incorporates both.
.
4.1 Inversion
In inversion, the logical subject
appears in postverbal position while some other, canonically postverbal,
constituent appears in preverbal position (Birner 1994), excluding cases where
expletive there occupies syntactic subject position (which are both
formally and functionally distinct). We will refer to the noncanonically
positioned constituents as the “preposed” and “postposed” constituents for
convenience, although again we wish to remain neutral with respect to the
syntactic analysis of the
construction.
Thus, even in cases where both
constituents have been previously evoked, the post posed constituent
nonetheless represents less familiar information, where familiarity is defined
by prior evocation, inferability, and recency of mention. Therefore, what is
relevant for the felicity of inversion in discourse is the relative
discoursefamiliarity of the information represented by these two constituents.
4.2 Passivization
Like inversion, English by-phrase
passives reverse the canonical order of two constituents, and like inversion,
they are also constrained pragmatically in that the syntactic subject must not
represent newer information within the discourse than does the NP in the by-phrase
(Birner 1996). We claim that passivization and inversion represent distinct
syntactic means for performing the same discourse function in different
syntactic environments.
Based on an examination of the
first 200 by-phrase passives appearing in the Brown Corpus, Birner
(1996) shows that the syntactic subject of such passives consistently represents
information that is at least as familiar within the discourse as that
represented by the by-phrase NP. Moreover, when the information status
of the relevant NPs is reversed, infelicity results. Consider again example
(26), repeated here as Given that passivization, like inversion, places
relatively familiar information before
relatively unfamiliar
information, it too can be viewed as performing a linking function (see section
1.3). That is, in passivization as in inversion, the information represented by
the preverbal constituent generally stands in a poset relationship with a
previously evoked or inferable anchor.
5
Left-dislocation
Here, the direct object pronoun them
is coreferential with the sentence-initial constituent gallstones.
Left-dislocation is also functionally distinct from preposing. As we have seen,
preposing constructions constitute a functionally unified class in that the preposed
constituent consistently represents information standing in a contextually licenced
poset relationship with information evoked in or inferable from the prior context.
No such requirement holds for left-dislocation, however.
Prince (1997) argues that there
are three types of left-dislocation (LD), distinguishable on functional
grounds. Type I LD is what Prince calls “simplifying LDs”: A “simplifying”
Left-Dislocation serves to simplify the discourse processing of Discourse-new
entities by removing them from a syntactic position disfavored for Discourse-new
entities and creating a separate processing unit for them. Once tha unit is
processed and they have become Discourse-old, they may comfortably occur in
their positions within the clause as pronouns. (1997: 124).
6
Right-dislocation
Like existential and
presentational there-insertion, right-dislocation involves the noncanonical
placement of an argument of the verb in postverbal position. However, in
contrast to both existential and presentational there-insertion,
right-dislocation (RD) does not require the postverbal NP to represent new
information. Consider the rightdislocations in : a. Below the waterfall (and
this was the most astonishing sight of all), a whole mass of enormous glass
pipes were dangling down into the river from somewhere high up in the ceiling! They
really were enormous, those pipes. There must have been a dozen of
them at least, and they were sucking up the brownish muddy water from the river
and carrying it away to goodness knows where. (Dahl, R. 1964: 74–5 Charlie
and the Chocolate Factory. New York: Knopf.)
In each of these examples, the
sentence-final constituent represents information that has been evoked, either
explicitly or implicitly, in the prior discourse. The functions that previous
researchers have posited for RD, in fact, have generally assumed that the
dislocated NP must represent information that is given or inferable within the discourse.
For example, Davison (1984) argues that RD marks the referent of the dislocated
NP as a topic, and thus also as having a “discourse antecedent” (1984: 802).
Similarly, Ziv and Grosz (1994)
argue that RD identifies a situationally or textually evoked entity as the most
salient entity available for subsequent reference.18 Indeed, our corpus-based
study shows that, in every case, the dislocated NP represents information that
is both hearer-old and discourse-old. Thus, right-dislocation cannot be viewed
as marking information that is new in any sense, and in this way differs from existential
and presentational there-insertion on functional grounds. As we argued
in previous work (Birner and Ward 1996), the difference in function can be
attributed to the anaphoric pronoun of right-dislocation. Given that the marked
NP in a right-dislocation is coreferential with the pronoun, and that the
pronoun is anaphoric and therefore represents a discourse-old entity, it
follows that the marked NP must also represent this same discourse-old entity.
Thus, it is not accidental that
right-dislocation does not
require the marked NP to represent new information; the presence of the pronoun
in fact precludes such a possibility.
7. Conclusion
We have suggested that a complete
functional account of the noncanonical constructions of English requires
reference to open propositions, discourse- and hearerfamiliarity, and linking
relations. By now it should be clear that these constraints are not randomly
assigned to the various construction types, but rather that broad generalizations
can be made regarding the correlation of syntax and discourse function.
Specifically, we have argued that: • preposing constructions require the preposed
constituent to represent information that is old in some sense, while
postposing constructions require the postposed constituent to represent
information that is new in some sense; • the constraints on preposing and
postposing are absolute, while those placed on argument reversal are relative; •
the functional constraints observed for the classes of preposing and postposing
constructions do not hold for superficially similar constructions in which the marked
constituent’s canonical position is filled by a referential pronoun (i.e.right-
and left-dislocation). Although we have found no necessary correspondence
between particular constructions and specific functional constraints, discourse
functions nonetheless correlate with syntactic constructions in a principled
way. Our research indicates that the range of discourse functions a given
construction may serve is constrained by the
form of the construction; within
that range, however, there is room for arbitrary variation. This approach
reconciles both the strong correlations we have found among construction types
and function types and the equally strong evidence of variation inthe
correlation between form and function.
Although the various studies utilizing
these terms have by and large used them in very simila ways, these studies have
failed to draw the (in our view) crucial distinctions among the linguistic
items being related, the poset relation connecting the information represented
by these items, and the poset itself.
7 In this example the preposition
in does not constitute part of the link, unlike the
preposition in (7a). The difference between the two types of
links correlates with distinct preposing constructions; see
Ward (1988) for discussion. 8 This
corpus consists of over 1.3 million words of transcribed oral data drawn from
the official transcripts of the Presidential Commission on the Space Shuttle
Challenger Accident (1986). We are grateful to Julia Hirschberg for making an
on-line version of these transcripts available to us. 9 For convenience, we
will use terms like “preposing” and “postposing” to
refer to the noncanonical
placement of syntactic constituents, although we wish to remain neutral with
respect to their actual syntactic analysis. 10 By “accent,” we mean
“intonational prominence” in the sense of Terken and Hirschberg (1994): “a
conspicuous pitch change in or near the lexically stressed syllable of the
word” (1994:126); see also Pierrehumbert (1980). 11 Of course for both
topicalization and focus preposing, other constituents may bear pitch accents.
Intonationally speaking, the difference between focus preposing and
topicalization is that only the former requires that the nuclear accent be on
the preposed constituent.12 As noted in Ward (1988), there is one preposing
construction – “locative
1. We use the term “construction”
in the conventional sense, to refer to each of the various grammatical configurations
of constituents within a particular language. See Fillmore (1988), Prince (1994),
and Goldberg (1995), inter alia, for alternative views of what constitutes a
linguistic
construction.
2. What is relevant here is the presence of
information within the hearer’s knowledge store, not the hearer’s beliefs
regarding its truth (in the case of a proposition), existence (in the case of
an entity), attributes, etc. That is, what matters for hearer-status is the
hearer’s knowledge of, rather than about, the information.
3. Strictly speaking it is the
information itself that possesses some information status (and not the
constituent representing that information), but where no confusion will result
we will speak of constituents as being discourse-old, discourse-new, evoked, etc.
for convenience.
4 Thus, the “discourse-old” link
need not itself have been explicitly evoked within the prior discourse; as long
as it stands in an appropriate relationship with previously evoked information,
it is treated by speakers as discourse-old.
5 Higher-value preposings are
actually quite rare, and are usually explicitly designated as such, as with the
quantifier.
REFERENCES
Aissen, J. (1975). Presentational-there
