Expectations in Incremental Discourse Processing 
Dan Cristea 
Faculty of Computer Science 
University "A.I. Cuza" 
16, Berthelot Street 
6600 - Iasi, Romania 
dcristea@infoiasi, ro 
Bonnie Webber 
Dept. of Computer L: Information Science 
University of Pennsylvania 
200 South 33rd Street 
Philadelphia PA 19104-6389 USA 
bonnie©central, cis. upenn, edu 
Abstract 
The way in which discourse features ex- 
press connections back to the previous dis- 
course has been described in the literature 
in terms of adjoining at the right frontier 
of discourse structure. But this does not 
allow for discourse features that express ez- 
pectations about what is to come in the 
subsequent discourse. After characterizing 
these expectations and their distribution in 
text, we show how an approach that makes 
use of substitution as well as adjoining on a 
suitably defined right frontier, can be used 
to both process expectations and constrain 
discouse processing in general. 
1 Introduction 
Discourse processing subsumes several distinguish- 
able but interlinked processes. These include refer- 
ence and ellipsis resolution, inference (e.g., inferen- 
tial processes associated with focus particles such as, 
in English, "even" and "only"), and identification of 
those structures underlying a discourse that are as- 
sociated with coherence relations between its units. 
In the course of developing an incremental approach 
to the latter, we noticed a variety of constructions 
in discourse that raise expectations about its future 
structural features. We found that we could rep- 
resent such expectations by adopting a lexical vari- 
ant of TAG - LTAG (Schabes, 1990) - and using 
its substitution operation as a complement to ad- 
joining. Perhaps more interesting was that these 
expectations appeared to constrain the subsequent 
discourse until they were resolved. This we found 
we could model in terms of constraints on adjoining 
and substitution with respect to a suitably defined 
Right Frontier. This short paper focuesses on the 
phenomenon of these expectations in discourse and 
their expression in a discourse-level LTAG. We con- 
elude the paper with some thoughts on incremental 
discourse processing in light of these expectations. 
The following examples illustrate the creation of 
expectations through discourse markers: 
Example 1 
a. On the one hand, John is very generous. 
b. On the other, he is extremely difficult to find. 
Example 2 
a. On the one hand, John is very generous. 
b. On the other, suppose you needed some money. 
c. You'd see that he's very difficult to find. 
Example 3 
a. On the one hand, John is very generous. 
b. For example, suppose you needed some money. 
c. You would just have to ask him for it. 
b. On the other hand, he is very difficult to find. 
Example 1 illustrates the expectation that, follow- 
ing a clause marked "on the one hand", the discourse 
will express a constrasting situation (here marked 
by "on the other"). Examples 2 and 3 illustrate 
that such an expectation need not be satisfied im- 
mediately by the next clause: In Example 2, clause 
(b) partially resolves the expectation set up in (a), 
but introduces an expectation that the subsequent 
discourse will indicate what happens in such cases. 
That expectation is then resolved in clause (c). In 
Example 3, the next two clauses do nothing to sat- 
isfy the expectation raised in clause (a): rather, they 
give evidence for the claim made in (a). The expec- 
tation raised in (a) is not resolved until clause (d). 
These examples show expectations raised by sen- 
tential adverbs and the imperative use of the verb 
"suppose". Subordinate conjunctions (e.g., "just 
as", "although", "when", etc.) can lead to similar 
expectations when they appear in a preposed subor- 
dinate clause - eg. 
Example 4 
a. Although John is very generous, 
b. if you should need some money, 
c. you'd see that he's difficult to find. 
As in Example 2, clause 4(a) raises the expectation 
of learning what is nevertheless the case. Clause 4(b) 
partially satisfies that expectation by raising a hy- 
88 
pothetical situation, along with the expectation of 
learning what is true in such a situation. This latter 
expectation is then satisfied in clause 4(c). 
In summary, these expectations can be charac- 
terized as follows: (1) once raised, an expectation 
must be resolved, but its resolvant can be a clause 
that raises its own expectations; (2) a clause rais- 
ing an expectation can itelf be elaborated before 
that expectation is resolved, including elaboration 
by clauses that raise their own expectations; and (3) 
the most deeply "embedded" expectations must al- 
ways be resolved first. 
Now these are very likely not the only kinds of 
expectations to be found in discourse: Whenever 
events or behavior follow fairly regular patterns over 
time, observers develop expectations about what will 
come next or at least eventually. For example, a di- 
alogue model may embody the expectation that a suggestion 
made by one dialogue participant would 
eventually be followed by an explicit or implicit re- 
jection, acceptance or tabling by the other. Other di- 
alogue actions such as clarifications or justifications 
may intervene, but there is a sense of an expectation 
being resolved when the suggestion is responded to. 
Here we are focussed on discourse at the level of 
individual monologue or turn within a larger dis- 
course: what we show is that discourse manifests cer- 
tain forward-looking patterns that have similar con- 
straints to those of sentence-level syntax and can be 
handled by similar means. One possible reason that 
these particualr kinds of expressions may not have 
been noticed before is that in non-incremental ap- 
proaches to discourse processing (Mann and Thomp- 
son, 1988; Marcu, 1996), they don't stand out as 
obviously different. 
The labels for discourse coherence relations used 
here are similar to those of RST (Mann and Thomp- 
son, 1988), but for simplicity, are treated as binary. 
Since any multi-branching tree can be converted to a 
binary tree, no representational power is lost. In do- 
ing this, we follow several recent converging compu- 
tational approaches to discourse analysis, which are 
also couched in binary terms (Gardent, 1997; Marcu, 
1996; Polanyi and van den Berg, 1996; Schilder, 
1997; van den Berg, 1996). 
Implicit in our discussion is the view that in 
processing a discourse incrementally, its semantics 
and pragmatics are computed compositionally from 
the structure reflected in the coherence relations 
between its units. In the figures presented here, 
non-terminal nodes in a discourse structure are la- 
belled with coherence relations merely to indicate 
the functions that project appropriate content, be- 
liefs and other side effects into the recipient's dis- 
course model. This view is, we believe, consistent 
with the more detailed formal interfaces to discourse 
semantics/pragmatics presented in (Gardent, 1997; 
Schilder, 1997; van den Berg, 1996), and also allows 
for multiple discourse relations (intentional and in- 
formational) to hold between discourse units (Moore 
and Pollack, 1992; Moser and Moore, 1995; Moser 
and Moore, 1996) and contribute to the seman- 
tic/pragmatics effects on the recipient's discourse 
model. 
2 Expectations in Corpora 
The examples given in the Introduction were all 
"minimal pairs" created to illustrate the relevant 
phenomenon as succinctly as possible. Empirical 
questions thus include: (1) the range of lexico- 
syntactic constructions that raise expectations with 
the specific properties mentioned above; (2) the fre- 
quency of expectation-raising constructions in text; 
(3) the frequency with which expectations are sat- 
isfied immediately, as opposed to being delayed by 
material that elaborates the unit raising the expec- 
tation; (4) the frequency of embedded expectations; 
and (5) features that provide evidence for an expec- 
tation being satisfied. 
While we do not have answers to all these ques- 
tions, a very preliminary analysis of the Brown Cor- 
pus, a corpus of approximately 1600 email messages, 
and a short Romanian text by T. Vianu (approx. 
5000 words) has yielded some interesting results. 
First, reviewing the 270 constructions that Knott 
has identified as potential cue phrases in the Brown 
Corpus 1, one finds 15 adverbial phrases (such as 
"initially", "at first", "to start with", etc.) whose 
presence in a clause would lead to an expectation 
being raised. All left-extraposed clauses in English 
raise expectations (as in Example 4) so all the sub- 
ordinate conjunctions in Knott's list would be in- 
cluded as well. Outside of cue phrases, we have iden- 
tified imperative forms of "suppose" and "consider" 
as raising expectations, but currently lack a more 
systematic procedure for identifying expectation- 
raising constructions in text than hand-combing text 
for them. 
With respect to how often expectation-raising 
constructions appear in text, we have Brown Cor- 
pus data on two specific types - imperative "sup- 
pose" and adverbial "on the one hand" - as well as 
a detailed analysis of the Romanian text by Vianu 
mentioned earlier. 
There are approximately 54K sentences in the 
Brown Corpus. Of these, 37 contain imperative 
"suppose" or "let us suppose". Twelve of these cor- 
respond to "what if" questions or negotiation moves 
which do not raise expectations: 
Suppose -just suppose this guy was really 
what he said he was! A retired professional 
killer If he was just a nut, no harm was 
done. But if he was the real thing, he could 
do something about Lolly. (c123) 
1 Personal communication, but also see (Knott, 1996) 
89 
Alec leaned on the desk, holding the clerk's 
eyes with his. "Suppose you tell me the 
real reason", he drawled. "There might be 
a story in it". (c121) 
The remaining 25 sentences constitute only about 
0.05% of the Brown Corpus. Of these, 22 have their 
expectations satisfied immediately (88%) - for ex- 
ample, 
Suppose John Jones, who, for 1960, filed 
on the basis of a calendar year, died June 
20, 1961. His return for the period January 
1 to June 20, 1961, is due April 16, 1962. 
One is followed by a single sentence elaborating the 
original supposition (also flagged by "suppose") - 
"Suppose it was not us that killed these 
aliens. Suppose it is something right on the 
planet, native to it. I just hope it doesn't 
work on Earthmen too. These critters went 
real sudden". (cmO~) 
while the remaining two contain multi-sentence elab- 
orations of the original supposition. None of the ex- 
amples in the Brown Corpus contains an embedded 
expectation. 
The adverbial "on the one hand" is used to pose 
a contrast either phrasally - 
Both plans also prohibited common direc- 
tors, officers, or employees between Du 
Pont, Christiana, and Delaware, on the one 
hand, and General Motors on the other. (ch16) 
You couldn't on the one hand decry the 
arts and at the same time practice them, 
could you? (ck08) 
or clausally. It is only the latter that are of interest 
from the point of discourse expectations. 
The Brown Corpus contains only 7 examples of 
adverbial "on the one hand". In three cases, the 
expectation is satisfied immediately by a clause cued 
by "but" or "or" -e.g. 
On the one hand, the Public Health Ser- 
vice declared as recently as October 26 that 
present radiation levels resulting from the 
Soviet shots "do not warrant undue public 
concern" or any action to limit the intake 
of radioactive substances by individuals or 
large population groups anywhere in the 
Aj. But the PHS conceded that the new 
radioactive particles "will add to the risk 
of genetic effects in succeeding generations, 
and possibly to the risk of health damage to 
some people in the United States".(cb21) 
In the remaining four cases, satisfaction of the ex- 
pectation (the "target" contrast item) is delayed by 
2-3 sentences elaborating the "source" contrast item 
-- e.g. 
Brooklyn College students have an ambiva- 
lent attitude toward their school. On the 
one hand, there is a sense of not having 
moved beyond the ambiance of their high 
school. This is particularly acute for those 
who attended Midwood High School di- 
rectly across the street from Brooklyn Col- 
lege. They have a sense of marginality at 
being denied that special badge of status, 
the out-of-town school. At the same time, 
there is a good deal of self-congratulation 
at attending a good college . .. (cf25) 
In these cases, the target contrast item is cued by 
"on the other hand" in three cases and "at the same 
time" in the case given above. Again, none of the 
examples contains an embedded expectation. 
(The much smaller email corpus contained six ex- 
amples of clausal "on the one hand", with the target 
contrast cued by "on the other hand","on the other" 
or "at the other extreme". In one case, there was no 
explicit target contrast and the expectation raised 
by "on the one hand" was never satisfied. We will 
continue to monitor for such examples.) 
Before concluding with a close analysis of the Ro- 
manian text, we should note that in both the Brown 
Corpus and the email corpus, clausal adverbial "on 
the other hand" occurs more frequently without an 
expectation-raising "on the one hand" than it does 
with one. (Our attention was called to this by a 
frequency analysis of potential cue phrase instances 
in the Brown Corpus compiled for us by Alistair 
Knott and Andrei Mikheev, HCRC, University of 
Edinburgh.) We found 53 instances of clausal "on 
the other hand" occuring without an explicit source 
contrast cued earlier. Although one can only specu- 
late now on the reason for this phenomenon, it does 
make a difference to incremental analysis, as we try 
to show in Section 3.3. 
The Romanian text that has been closely anal- 
ysed for explicit expectation-raising constructions is 
T. Vianu's Aesthetics. It contains 5160 words and 
382 discourse units (primarily clauses). Counting 
preposed gerunds as raising expectations as well as 
counting the constructions noted previously, 39 in- 
stances of expectation-raising discourse units were 
identified (10.2%). In 11 of these cases, 1-16 dis- 
course units intervened before the raised expectation 
was satisfied. One example follows: 
Dar de~i trebuie s~-l parcurgem in 
intregime, pentru a orienta cercetarea este 
nevoie s~. incerc~m inc~ de pe acum o pre- 
cizare a obiectului lui. 
(But although we must cover it entirely, in 
order to guide the research we need to try 
already an explanation of its subject mat- 
ter.) 
90 
3 A Grammar for Discourse 
The intuitive appeal of Tree-adjoining Grammar 
(TAG) (Joshi, 1987) for discourse processing (Gar- 
dent, 1997; Polanyi and van den Berg, 1996; 
Schilder, 1997; van den Berg, 1996; Webber, 1991) 
follows from the fact that TAG's adjoining operation 
allows one to directly analyse the current discourse 
unit as a sister to previous discourse material that 
it stands in a particular relation to. The new in- 
tuition presented here - that expectations convey a 
dependency between the current discourse unit and 
future discourse material, a dependency that can 
be "stretched" long-distance by intervening mate- 
rial - more fully exploits TAG's ability to express 
dependencies. By expressing in an elementary TAG 
tree, a dependency betwen the current discourse unit 
and future discourse material and using substitu- 
tion (Schabes, 1990) when the expected material is 
found, our TAG-based approach to discourse pro- 
cessing allows expectations to be both raised and 
resolved. 
3.1 Categories and Operations 
The categories of our TAG-based approach consist 
of nodes and binary trees. We follow (Gardent, 
1997) in associating nodes with feature structures 
that may hold various sorts of information, including 
information about the semantic interpretations pro- 
jected through the nodes, constraints on the specific 
operations a node may participate in, etc. A non- 
terminal node represents a discourse relation holding 
between its two daughter nodes. A terminal node 
can be either non-empty (Figure la), corresponding 
to a basic discourse unit (usually a clause), or empty. 
A node is "empty" only in not having an associated 
discourse unit or relation: it can still have an asso- 
ciated feature structure. Empty nodes play a role 
in adjoining and substitution, as explained below, 
and hence in building the derived binary tree that 
represents the structure of the discourse. 
Adjoining adds to the discourse structure an aux- 
iliary tree consisting of a root labelled with a dis- 
course relation, an empty foot node (labelled *), and 
at least one non-empty node (Figures lc and ld). In 
our approach, the foot node of an auxiliary tree must 
be its leftmost terminal because all adjoining oper- 
ations take place on a suitably defined right frontier 
(i.e., the path from the root of a tree to its rightmost 
leaf node) - such that all newly introduced mate- 
rial lies to the right of the adjunction site. (This is 
discussed in Section 3.2 in more detail.) Adjoining 
corresponds to identifying a discourse relation be- 
tween the new material and material in the previous 
discourse that is still open for elaboration. 
Figure 2(a) illustrates adjoining midway down the 
RF of tree a, while Figure 2(b) illustrates adjoining 
at the root of a's RF. Figure 2(c) shows adjoining 
at the "degenerate" case of a tree that consists only 
of its root. Figure 2(d) will be explained shortly. 
Substitution unifies the root of a substitution 
structure with an empty node in the discourse tree 
that serves as a substitution site. We currently 
use two kinds of substitution structures: non-empty 
nodes (Figure la) and elementary trees with substi- 
tution sites (Figure lb). The latter are one way by 
which a substitution site may be introduced into a 
tree. As will be argued shortly, substitution sites can 
only appear on the right of an elementary tree, al- 
though any number of them may appear there (Fig- 
ure lb). Figure 2(e) illustrates substitution of a non- 
empty node at ~, and Figure 2(f) illustrates substitu- 
tion of an elementary tree with its own substitution 
site at ~1 
Since in a clause with two discourse markers (as 
in Example 3b) one may look backwards ("for exam- 
ple") while the other looks forwards ("suppose"), we 
also need a way of introducing expectations in the 
context of adjoining. This we do by allowing an aux- 
iliary tree to contain substitution sites (Figure ld) 
which, as above, can only appear on its right. 2 An- 
other term we use for auxiliary trees is adjoining 
structures. 
3.2 Constraints 
Earlier we noted that in a discourse structure with 
no substitution sites, adjoining is limited to the right 
frontier (RF). This is true of all existing TAG-based 
approaches to discourse processing (Gardent, 1997; 
Hinrichs and Polanyi, 1986; Polanyi and van den 
Berg, 1996; Schilder, 1997; Webber, 1991), whose 
structures correspond to trees that lack substitution 
sites. One reason for this RF restriction is to main- 
tain a strict correspondence between a left-to-right 
reading of the terminal nodes of a discourse struc- 
ture and the text it analyses - i.e., 
Principle of Sequentiality: A left-to- 
right reading of the terminal frontier of the 
tree associated with a discourse must cor- 
respond to the span of text it analyses in 
that same left-to-right order. 
Formal proof that this principle leads to the restric- 
tion of adjoining to the right frontier is given in 
(Cristea and Webber, June 1997). 
The Principle of Sequentiality leads to additional 
constraints on where adjoining and substitution can 
occur in trees with substitution sites. Consider the 
tree in Figure 3(i), which has two such sites, and an 
adjoining operation on the right frontier at node Rj 
or above. Figure 3(it) shows that this would intro- 
duce a non-empty node (uk) above and to the right 
of the substitution sites. This would mean that later 
substitution at either of them would lead to a viola- 
tion of the Principle of Sequentiality, since the newly 
~We currently have no linguistic evidence for the 
structure labelled ~ in Figure ld, but are open to its 
possibility. 
9\] 
U 
a. One-node tree 
(Non-empty node) 
{2 , U 
b. Elementary trees c. Auxiliary trees 
with substitution sites 
U*~I * U 
d. Aux trees with 
substitution sites 
Figure 1: Grammatical Categories. (* marks the foot of an auxiliary tree, and l, a substitution site.) 
R i Rk R i R2 
(a) Adjoining at Ri+2 on the RF of a (b) Adjoining at the root (R1) of 
.............................. 
u , 11 
(c) Adjoining at root of single node tree ¢x 
RI R 1 
R 3 
1 1 
(d) Adjoining at R3 on the right frontier of ot 
~ ~'~ ~ ~-- R 2 
u 1 
(e) Substituting a material node at I (f) Substituting at elementary pn 
tree with substitution site 2 
Figure 2: Examples of Adjoining and Substitution 
92 
/~Rjl ~k dRj'l 
(ii) (iii) 
Figure 3: Adjoining is constrained to nodes the inner_RF, indicated by the dashed arrow. 
substituted node u~+t would then appear to the left 
of uk in the terminal frontier, but to the right of 
it in the original discourse. Adjoining at any node 
above Rj+2 - the left sister of the most deeply em- 
bedded substitution site - leads to the same problem 
(Figure 3iii). Thus in a tree with substitution sites, 
adjoining must be limited to nodes on the path from 
the left sister of the most embedded site to that sis- 
ter's rightmost descendent. But this is just a right 
frontier (RF) rooted at that left sister. Thus, ad- 
joining is always limited to a RF: the presence of a 
substitution site just changes what node that RF is 
rooted at. We can call a RF rooted at the left sister 
of the most embedded substitution site, the inner 
right frontier or "inner_RF". (In Figure 3(i), the in- 
ner_RF is indicated by a dashed arrow.) In contrast, 
we sometimes call the RF of a tree without substi- 
tution sites, the outer right frontier or "outer_RF". 
Figure 2(d) illustrates adjoining on the inner_RF of 
a, a tree with a substitution site labelled h. 
Another consequence of the Principle of Sequen- 
tiality is that the only node at which substitution 
is allowed in a tree with substitution sites is at the 
most embedded one. Any other substitution would 
violate the principle. (Formal proof of these claims 
are given in (Cristea and Webber, June 1997). 
3.3 Examples 
Because we have not yet implemented a parser that 
embodies the ideas presented so far, we give here 
an idealized analysis of Examples 2 and 3, to show 
how an ideal incremental monotonic algorithm that 
admitted expectations would work. 
Figure 4A illustrates the incremental analysis of 
Example 2. Figure 4A(i) shows the elementary tree 
corresponding to sentence 2a ("On the one hand 
..."): the interpretation of "John is very generous" 
I 
corresponds to the left daughter labelled "a". The 
adverbial "On the one hand" is taken as signalling a 
coherence relation of Contrast with something ex- 
pected later in the discourse. 
In sentence 2b ("On the other hand, suppose 
..."), the adverbial "On the other hand" signals 
the expected contrast item. Because it is al- 
ready expected, the adverbial does not lead to the 
creation of a separate elementary tree (but see 
the next example). The imperative verb "sup- 
pose", however, signals a coherence relation of an- 
tecedent/consequent (A/C) with a consequence 
expected later in the discourse. The elementary 
tree corresponding to "suppose ..." is shown in 
Figure 4A(ii), with the interpretation of "you need 
money" corresponding to the left daughter labelled 
"b". Figure 4A(iii) shows this elementary tree sub- 
stituted at ~1, satisfying that expectation. Fig- 
ure 4A(iv) shows the interpretation of sentence 2c 
("You'd see he's very difficult to find") substituted 
at 12, satisfying that remaining expectation. 
Before moving on to Example 3, notice that if Sen- 
tence 2a were not explicitly cued with "On the other 
hand", the analysis would proceed somewhat differ- 
ently. 
Example 5 
a. John is very generous. 
b. On the other hand, suppose you needed money. 
c. You'd see that he's very difficult to find. 
Here, the interpretation of sentence 5(a) would cor- 
respond to the degenerate case of a tree consisting of 
a single non-empty node shown in Figure 4B(i). The 
contrast introduced by "On the other hand" in sen- 
tence 5(b) leads to the auxiliary tree shown in Fig- 
ure 4B(ii), where T stands for the elementary tree 
corresponding to the interpretation of "suppose...". 
93 
Contrast A/C Contrast Contrast 
b ~2 b 
(i) (ii) (iii) (iv) 
Contrast Contrast Contrast Contrast • a 
a • T C ~" a A/C 
b ~l b ~! b 
(i) (ii) (iii) (iv) (v) 
Contrast 
(i) 
Contrast 
Evid~Xx ~" 
C 
OD 
Contrast Contrast 
b b c 
(iii) (iv) 
Figure 4: Analyses of Examples 2, 3 and 4. 
A. Example 2 
B. Example 5 
C. Example 3 
The entire structure associated with sentence 5(b) 
is shown in Figure 4B(iii). This is adjoined to the 
single node tree in Figure 4B(i), yielding the tree 
shown in Figure 4B(iv). The analysis then contin- 
ues exactly as in that of Example 2 above. 
Moving on to Example 3, Figure 4C(i) shows the 
same elementary tree as in Figure 4A(i) correspond- 
ing to clause 3a. Next, Figure 4C(ii) shows the aux- 
iliary tree with substitution site ~2 corresponding to 
clause 3b being adjoined as a sister to the interpre- 
tation of clause 3a, as evidence for the claim made 
there. The right daughter of the node labelled "Ev- 
idence" is, as in Example 2b, an elementary tree 
expecting the consequence of the supposition "you 
need money". Figure 4C(iii) shows the interpreta- 
tion of clause 3c substituted at ~2, satisfying that 
expectation. Finally, Figure 4C(iv) shows the inter- 
pretation of clause 3d substituted at 11, satisfying 
the remaining expectation. 
4 Sources of Uncertainty 
The idealized analysis presented above could lead 
to a simple deterministic incremental algorithm, if 
there were no uncertainty due to local or global am- 
biguity. But there is. We can identify three separate 
sources of uncertainty that would affect incremental 
processing according to the grammar just presented: 
• the identity of the discourse relation that is 
meant to hold between two discourse units; 
• the operation (adjoining or substitution) to be 
used in adding one discourse unit onto another; 
• if that operation is adjoining, the site in the 
target unit at which the operation should take 
place - that is, the other argument to the dis- 
course relation associated with the root of the 
auxiliary tree. 
It may not be obvious that there could be uncer- 
tainty as to whether the current discourse unit sat- 
isfies an expectation and therefore substitutes into 
the discourse structure, or elaborates something in 
the previous discourse, and therefore adjoins into 
it. 3 But the evidence clarifying this local ambiguity 
may not be available until later in the discourse. In 
the following variation of Example 4, the fact that 
clause (b) participates in elaborating the interpreta- 
tion of clause (a) rather than in satisfying the expec- 
tation it raises (which it does in Example 4) may not 
be unambiguously clear until the discourse marker 
"for example" in clause (c) is processed. 
Example 6 
a. Because John is such a generous man - 
b. whenever he is asked for money, 
c. he will give whatever he has, for example - 
d. he deserves the "Citizen of the Year" award. 
The other point is that, even if a forward-looking 
cue phrase signals only a substitution structure as 
3This is not the same as shift-reduce uncert~nty. 
94 
in Figure 4A(i) and 4A(ii), if there are no pending 
subsitution sites such as ~1 in 4A(i) against which to 
unify such a structure, then the substitution struc- 
ture must be coerced to an auxiliary tree as in Fig- 
ure ld (with some as yet unspecified cohesion rela- 
tion) in order to adjoin it somewhere in the current 
discourse structure. 
5 Speculations and Conclusions 
In this paper, we have focussed on discourse expec- 
tations associated with forward-looking clausal con- 
nectives, sentential adverbs and the imperative verbs 
("suppose" and "consider"). There is clearly more 
to be done, including a more complete characteri- 
zation of the phenomenon and development of an 
incremental discourse processor based on the ideas 
presented above. The latter would, we believe, have 
to be coupled with incremental sentence-level pro- 
cessing. As the previous examples have shown, the 
same phenomenon that occurs inter-sententially in 
Examples 1-3 occurs intra-sententially in Examples 4 
and 6, suggesting that the two processors may be 
based on identical principles. In addition, carrying 
out sentence-level processing in parallel with dis- 
course processing and allowing each to inform the 
other would allow co-reference interpretation to fol- 
low from decisions about discourse relations and vice 
versa. 
6 Acknowledgements 
Support for this work has come from the De- 
partment of Computer Science, Universiti Sains 
Malaysia (Penang, Malaysia), the Department of 
Computer Science, University "A.I.Cuza" (Iasi, Ro- 
mania) and the Advanced Research Project Agency 
(ARPA) under grant N6600194C6-043 and the 
Army Research Organization (ARO) under grant 
DAAHO494GO426. Thanks go to both the anony- 
mous reviewers and the following colleagues for their 
helpful comments: Michael Collins, Claire Gardent, 
Udo Hahn, Joseph Rosenzweig, Donia Scott, Mark 
Steedman, Matthew Stone, Michael Strube, and 
Michael Zock. Thanks also to Alistair Knott and 
Andrei Mikheev for giving us a rough count of cue 
phrases in the Brown Corpus. 

References 
Cristea, Dan and Bonnie Webber. June 1997. Ex- 
pectations in incremental discourse processing. 
Technical report, University A.I. Cuza, Iasi, Ro- 
mania. 
Gardent, Claire. 1997. Discourse tree adjoining 
grammars. Claus report nr.89, University of the 
Saarland, Saarbriicken. 
ttinrichs, Erhard.and Livia Polanyi. 1986. Pointing 
the way: A unified treatment of referential ges- 
ture in interactive discourse. In CLS 22, Part 
2: Papers from the Parasession on Pragmatics 
and Grammatical Theory, pages 298-314, Chicago 
Linguistic Society. 
Joshi, Aravind. 1987. An introduction to Tree Ad- 
joining Grammar. In Alexis Manaster-Ramer, ed- 
itor, Mathematics of Language. John Benjamins, 
Amsterdam. 
Knott, Alistair. 1996. A Data-driven Methodol- 
ogy for Motivating a Set of Coherence Relations. 
Ph.D. thesis, Department of Artificial Intelligence, 
University of Edinburgh. 
Mann, William and Sandra Thompson. 1988. 
Rhetorical structure theory: Toward a functional 
theory of text organization. Text, 8(3):243-281. 
Marcu, Daniel. 1996. Building up rhetorical struc- 
ture trees. In Proceedings of AAAI-96, pages 
1069-1074, Portland OR. 
Moore, Johanna and Martha Pollack. 1992. A prob- 
lem for rst: The need for multi-level discouse anal- 
ysis. Computational Linguistics, 18(4):537-544. 
Moser, Megan and Johanna Moore. 1995. Inves- 
tigating cue selection and placement in tutorial 
discourse. In Proc. 33rd Annual Meeting, Asso- 
ciation for Computational Linguistics, pages 130- 
135, MIT, Boston MA. 
Moser, Megan and Johanna Moore. 1996. Toward 
a synthesis of two accounts of discourse structure. 
Computational Linguistics, 22(2):TBA. 
Polanyi, Livia and Martin H. van den Berg. 1996. 
Discourse structure and discourse interpretation. 
In P. Dekker and M. Stokhof, editors, Proceedings 
of the Tenth Amsterdam Colloquium, pages 113- 
131, ILLC/Department of Philosophy, University 
of Amsterdam. 
Schabes, Yves. 1990. Mathematical and Compu- 
tational Aspects of Lexicalized Grammars. Ph.D. 
thesis, Department of Computer and Information 
Science, University of Pennsylvania. Technical 
Report MS-CIS-90-48, LINC Lab 179. 
Schilder, Frank. 1997. Tree discourse grammar, or 
how to get attached to a discourse. In Proceedings 
of the Tilburg Conference on Formal Semantics, 
Tilburg, Netherlands, January. 
van den Berg, Martin H. 1996. Discourse grammar 
and dynamic logic. In P. Dekker and M. Stokhof, 
editors, Proceedings of the Tenth Amsterdam Col- 
loquium, pages 93-111, ILLC/Department of Phi- 
losophy, University of Amsterdam. 
Webber, Bonnie. 1991. Structure and ostension 
in the interpretation of discourse deixis. Natural 
Language and Cognitive Processes, 6(2):107-135. 
