Disambiguating Cue Phrases in Text and Speech 
Diane Litman and Julia Hirschberg* 
AT&T Bell Laboratories 
Murray Hill NJ 07974 USA 
diane@research.att.com, julia@research.att.com 
Abstract 
Cue phrases are linguistic expressions such as 'now' 
and 'welg that may explicitly mark the structure of 
a discourse. For example, while the cue phrase 'in- 
czdcntally' may be used SENTENTIALLY as an adver- 
bial, the DISCOUaSE use initiates a digression. In 
\[8\], we noted the ambiguity of cue phrases with re- 
spect to discourse and sentential usage and proposed 
an intonational model for their disambiguation. In 
this paper, we extend our previous characterization 
of cue phrases aald generalize its domain of coverage, 
based on a larger and more comprehensive empirical 
study: an examination of all cue phrases produced by 
a single ,~peaker in recorded natural speech. We also 
associate this prosodic model with orthographic and 
part-of-speech analyses of cue phrases in text. Such 
a dual model provides both theoretical justification 
for current computational models of discourse and 
practical application to the generation of synthetic 
speech. 
1 Introduction 
Words and phrases that may directly mark the 
structure of a discourse have been termed CUE 
PttR.ASES, CLUE WORDS, DISCOURSE MAI:tKERS~ arid 
DISCOURSE PARTICLES \[3, 4, 14, 17, 19\]. Some exarn- 
pies are 'now', which marks the introduction of a new 
subtopic or return to a previous one, 'incidentally' 
and 'by the way', which indicate the beginning of a di- 
gression, and 'anyway' and 'in any case', which indi- 
cate return from a digression. In a previous study\[8\], 
we noted that such terms are potentially ambiguous 
between DISCOURSE and SENTENTIAL uses\[18\]. So, 
'now' may be used as a temporal adverbial as well 
as a discourse marker, 'incidentally' may also func- 
tion as an adverbial, and other cue phrases similarly 
have one or more senses in addition to their func- 
tion as markers of discourse structure. Based upon 
an empiricM study of 'now' in recorded speech, we 
proposed that such discourse and sentential uses of 
cue phrases can be disambiguated intonationally. In 
particular, we proposed a prosodic model for this dis- 
ambiguation which discriminated all discourse from 
*We thank Bengt Altenberg, l=tichaa-d Omanson mid Jan 
van Santen for providing information and helpful comments 
on this work. 
sentential uses of tokens in our sample. This model 
provided not only a plausibility argument for the dis- 
ambiguation of cue phrases, but also the beginnings 
of a model for the generation of cue phrases in syn- 
thetic speech. 
In this paper, we show that our prosodic model 
generalizes to other cue phrases as well. We further 
propose an initial model for the disambiguation of 
cuc phrases in text. Wc base these claims upon a 
further empirical study: an examination of all cue 
phrases produced by a single speaker in part of a 
recorded, transcribed lecture. In Section 2 we review 
our own and other work on cue phrases, in Section 
3 we describe our current empirical studies, in Sec- 
tion 4 we present the results of our analysis, and in 
Section 5 we discuss theoretical and practical appli- 
cations of our findings. 
2 Previous Studies 
The important role that cue phrases play in under- 
standing and generating discourse has been well doc- 
umented in the computational linguistics literature. 
For example, by indicating the presence of a struc- 
tural boundary or a relationship between parts of a 
discourse, cue phrases caa assist in the resolution of 
anaphora\[5, 4, 17\] and in the identification of rhetor- 
ical relations \[10, 12, 17\]. Cue phrases have also 
been used to reduce the complexity of discourse pro- 
cessing and to increase textual coherence\[3, 11, 21\]. 
In Example (1) 1, interpretation of the anaphor 'it' 
as (correctly) co-indexed with THE SYSTEM is facil- 
itated by the presence of the cue phrases 'say' and 
'then', marking potential antecedents in '... as an 
EXPERT DATABASE for AN EXPERT SYSTEM ...' as 
structurally unavailable. 2 
(1) "If THE SYSTEM attenqpts to hold rules, say as 
AN EXPERT DATABASE for AN EXPERT SYSTEM, 
then we expect it not only to hold the rules but 
to in fact apply them for us in appropriate situ- 
ations." 
1The examples are taken from the corpus described in Sec- 
tion 3. 
2InformMly, 'say' indicates the beginning of a discourse 
subtopic and 'then' signals a return from that subtopic. 
251 1 
Previous attempts to define the set of cue phrases 
have typically been extensional, 3 with such lists of 
cue phrases then further classified as to their dis- 
course function. For example, Cohen \[3\] uses a tax- 
onomy of connectives based on that of Quirk \[16\] to 
associate with each class of cue phrases a semantic 
function with respect to a model of argument under- 
standing. Grosz and Sidner \[4\] classify cue phrases 
based on changes to the attentional stack and inten- 
tional structure found in their theory of discourse. 
Schiffrin \[18\] classifies cue phrases into groups based 
on their sentential usage (e.g. conjunctive, adver- 
bial, and clausal markers), while Reichman \[17\] and 
Ilobbs\[10\] associate groups of cue phrases with the 
rhetorical relationships they signal. Finally, Zuker- 
man \[21\] presents a taxonomy of cue phrases based 
on three functions relevant to her work in language 
generation: knowledge organization, knowledge ac- 
quisition, and affect maintenance. 
Once a cue phrase has been identified, however, it 
is not always clear whether to interpret it as a dis- 
course marker or not\[6, 4, 8, 18\]. The texts in Exam- 
pie (2) are potentially ambiguous between a temporal 
reading of 'now' and a discourse interpretation: 
(2) a. "Now in AI our approach is to look at a 
knowledge base as a set of symbolic items 
that represent something." 
b. "Now some of you may suspect from the 
title of this talk that this word is coming to 
you from Krypton or some other possible 
world." 
On the temporal reading, (2a), for example, would 
convey that 'at this moment the AI approach to 
knowledge bases has changed'; on the discourse read- 
ing, 'now' simply initiates the topic of 'the AI ap- 
proach to knowledge bases'. 
It has been suggested that this difference between 
discourse and sententiai use may be intonationally 
disambiguable. Halliday and Hassan \[6\] claim that, 
in general, items used COtIES1VELY -- i.e., to relate 
one part of a text to another \[6, p. 4\] -- tend to 
be intonationally non-prominent (to be unaccented 
and reduced) unless they are "definitely contrastive". 
Non-cohesive uses, on the other hand, are indicated 
by non-reduced, accented forms.\[6, p. 268\] ttalliday 
and llassan particularly note that intonation disam- 
biguates in this way between cohesive (discourse) and 
non-cohesive (sentential) uses of classes of items we 
term cue phrases, such as conjunctions and adver- 
bials. Empirical studies to date have tended to bear 
out their observations. Studies of portions of the 
London-Lund corpus such as \[1\] have provided into- 
national profiles of word classes including DISCOURSE 
ITEMS, conjunctions and adverbials which are at least 
compatible with these views. However, the notion of 
'discourse item' used appears much more restrictive 
3An exception to this is found in the socio-linguistic work 
of Schifl'rin\[18\]. 
than the notion of 'cue phrase', 4 so it is difficult to 
make comparative use of these results. 
In an earlier study \[8\], we examined the use of 
various intonational, syntactic, and orthographic fea- 
tures to distinguish between discourse and senten- 
tim readings of a single cue phrase ('now'). 5 While 
features such as tense, structural configuration, sur- 
face order, and orthographic indicators were some- 
times useful, we found that intonational features 
provided only only significant correlation with dis- 
course/sentential status. All of the tokens in our 
sample were disarnbiguable in terms of intonational 
phrasing and type of pitch accentfi 
In our study of now, we found that discourse 
uses were either uttered as a single intermediate 
phrase (or in a phrase containing only cue phrases) 
(Discourse Type A), or uttered at the beginning of 
a longer intermediate phrase (or preceded only by 
other cue phrases in the phrase) and with a L* pitch 
accent or without a pitch accent (Discourse Type B). 
Cue phrases judged to be of Sentential Type were 
never uttered as a single phrase; if first in interme- 
diate phrase they were nearly always uttered with a 
H* or complex pitch accent (Sentential Type A); if 
not first in phrase they could bear any type of pitch 
accent or be deaccented (Sentential Type B). These 
results are summarized in Figure I. 
Based on these findings, we proposed that listeners 
use prosodic information to disambiguate discourse 
from sentential uses of cue phrases. To investigate 
this possibility further, we conducted another multi- 
speaker study of discourse and sentential uses of the 
cue phrase 'welt. Our findings were alrnost identical 
to results for the earlier study; briefly, of the 52 in- 
4 For example, in the 48 minute text Altenberg examines, he 
finds only 23 discourse items, or about 17% of what our study 
of a similar corpus (described below) would have predicted. 
Our corpus consisted of recordings of four days of tile radio 
call-in program "The Harry Gross Show: Speaking of Your 
Money," recorded during the week of 1 February 1982115\]. The 
four shows provided approximately ten hours of conversation 
between expert(s) m~d callers. 
6For the earlier study as well as the current one, we assume 
Pierrehumbel~,'s\[13\] system of phonological description. In 
this system, intonational contours are described as sequences 
of low (L) and high (H) tones in the FuraDAM~NTAL errs. 
QUENCV (F0) CONTOUrt. Pitch accents, peaks or valleys in 
the F0 contour that fall on the stressed syllables of lexical 
items, signify intonational prominence. A pitch accent con- 
sists either of a single tone or an ordered pair of tones, such 
as L*+H. The tone aligned with the stressed syllable is in- 
dicated by a star *; thus, in an L*+H accent, the low tone 
L* is aligned with the stressed syUahle. There are six pitch 
accents in English: two simple tones -- H and L -- and four 
complex ones -- L*+H, L+H*, H*+L, and H+L*. A well- 
formed intermediate phrase consists of one or more pitch 
accents, and a simple lfigh H or low L tone that represents 
the phrase accent. The phrase accent controls the pitch be- 
tween the last pitch accent of the current intermediate plwase 
and the beginning of the next -- or the end of the utterance. 
Intonational phrases are larger phonological milts, com- 
posed of one of more intermediate phrases, plus a boundary 
tone which may also be H or L. The occurrence of phrase 
accents and boundary tones, together with other phrase-froM 
characteristics such as passes aald syllable lengthening, enable 
us to identify intermediate mad intonational phrases. 
2 252 
Figure 1: Prosodic Characteristics of Discourse and Sentential Uses 
Cue Phrases 
Sentential recourse 
B 
Initial in Larger Phrase Non-Initial in Alone in Initial in Larger Phrase 
H or Complex Accent Larger Phrase Phrase Deaccented or L Accent 
stances of 'well' we examined, all but one token fit 
the model depicted in Figure 1. 
To see whether these findings could be extended to 
cue phrases in general, we began a third study -- of 
all cue phrases produced by a single speaker during 
75 minutes of recorded speech. The remainder of this 
paper describes our first results from this study. 
3 The Data 
To test whether our prosodic model of discourse and 
sentential uses of 'now' and 'well' extended to cue 
phrases in general, we examined intonational chm'- 
acteristics of all single-word cue phrases 7 used in a 
keynote, address given by I~onald Brachman at the 
First lnlernalional Conference on Expert Database 
Syslems in 1986. The address provides approxi- 
mately 75 minutes of speech t¥om a single speaker. 
For our first sample, we examined the 211 cue 
phrases uttered during the first 17 minutes of the 
address. Our tokens had the following distribution: s 
actually (6), also (2), although (1), and (68), basically 
(1), because (2), but (12), finally (1), \]i,'sl (1), further 
(4), however (2), like (11), look (11), next (4), now 
(26), ok (1), or (19), say (12), second (1), see (5), 
since (1), so (9), then (3), therefore (1), well (7) . 
To determine the classification of each token (ms- 
COURSE, SENTENTIAL, or AMBIGUOUS), the authors 
separately .judged each token by listening to the 
taped address while marking a transcription. 9 
rWe exmnined o~fly single-word cue plu, asea in tiffs study 
since our current prosodic model applies only to such items. 
In future work we plan to develop additional models for dis- 
course a~nL(l aententiel uses of multl-word cue phrases, e.g 'that 
reminds me', 'first o\] all', 'speaking off and so on. 
8Our set of cue phrases was derived from extensional def- 
initions provided by ourselves and othel~\[3, 4, 17, 18, 21\]. 
Tim following lexicel items, although also cue phrases, are 
not present in the portion of the axlch-ess examined to date: 
' alright', 'alternatively', 'anyway', %oy', ~ conversely', ' exeepf , 
'fine', '\]urthermore', 'incidentally', 'indeed', 'listen', 'more- 
over', 'nah', 'nevertheless', 'no', 'oh', 'right', 'why', 'yeah', 
~yes'. 
9The address was transcribed independently of our study 
by a meraber of the text processing pool at AT&T Bell Lab- 
oratories. We found that 20 cite phrases had been omitted 
by the traalscriber: 'and', 'now', 'ok', 'so', and 'well'. Signif- 
icantly, ell but two of these were termed 'discourse' uses by 
In comparing our judgments, we were interested in 
areas of disagreement as well as agreement. The set 
of tokens whose classification as to discourse or sen- 
tential use we agreed upon provide a testbed for our 
continuing investigation of the intonational disam- 
biguation of cue phrases. The set of tokens we found 
difficult to classify (i.e. those tokens we both found 
ambiguous or those whose cla.ssification we disagreed 
upon), provide insight into possible intonational cor- 
relates of discourse/sentential ambiguity. "Fable 1 
presents the distribution of our judgments, where 
'classifiable' represent those tokens whose classifica- 
tion we agreed upon and 'unclassifiable' represents 
those we both found ambiguous or disagreed upon. 
Table 1: Judgments by Type of Cue Phrase, N=211 
I Type Unclassifiable All 78 Conj 57 Non-Conj 21 Classifial, le laa 2LELH 
Of the 211 tokens in this initial sample, we found 
only 133 cue phrases (63.03%) to be unambigu- 
ously discourse or sentential. When we looked more 
closely at the 'unclassifiable' cases, we found that 
fully 73.08% were coordinate conjunctions (and, or, 
and but). In fact, when we compare percent classifi- 
able for conjunctions with other cue phrases, we find 
that, while only 42.42% of conjunctions were found 
to be classifiable, fully 81.25% of non-conjunctions 
were classified. Thus, the case of coordinate con- 
junction appears to explain a large portion of the 
our difficulty in agreeing upon a classification. 
Once we had made these judgments, we analyzed 
the tokens for their prosodic and syntactic features 
as well ms their orthographic context, much as we had 
done with tokens for the earlier two studies./° We 
noted whether each token was accented or not and, 
if accented, we noted the type of accent employed. 
We also identified the composition of the intermedi- 
both judges. 
1°We used a pitch tracker written by David Telkin and 
Talkin's Waves speech analysis software\[20\] in our prosodic 
analysis. 
253 3 
ate phrase containing each token as to whether the 
token constituted a separate phrase (possibly with 
other cue phrases) or not. And we noted each token's 
position within its intermediate phrase -- first (in- 
cluding tokens preceded only by other cue phrases) or 
not. We also noted syntactic characteristics of each 
item, including part of speech and its immediately 
dominating constituent, n Finally, we noted ortho- 
graphic indicators in the transcript which might pro- 
vide disambiguation, such as immediately preceding 
and succeeding punctuation and paragraph bound- 
aries. In both the syntactic and orthographic analy- 
ses we were particularly interested in discovering how 
well non-prosodic features which might be obtained 
automatically from a text would do in differentiating 
discourse from sentential uses. 
4 The Single-Speaker/Multi- 
Cue Phrase Study 
Our findings from the classified data (133 tokens) in 
this pilot study confirmed our model of prosodic dis- 
tinction between discourse and sentential uses of cue 
phrases. The distribution of these judgments with 
respect to the prosodic model of discourse and sen- 
tential cue phrases depicted in Figure 1 is shown in 
Table 2. Recall that this model includes two intona- 
Table 2: Prosody of Classified Tokens, N=133 
Judgment Prosody 
Discourse Sentential 
Discourse .... 44 4 
Sentential 17 .... 68 
(X 2 = 63.46, df --- 1, p <.001) 
tional profiles for discourse uses: Discourse Type A, 
in which a cue phrase constitutes an entire interme- 
diate phrase (or is in a phrase containing only other 
cue phrases) and may have any type of pitch accent; 
Discourse Type B, in which a cue phrase occurs at 
the beginning of a larger intermediate phrase (or is 
preceded only by other cue phrases) and bears a L* 
pitch accent or is deaccented; Sentential Type A, in 
which the cue phrase occurs at the beginning of a 
larger phrase and bears a H* or complex pitch ac- 
cent; and Sentential Type B, in which the cue phrase 
occurs in non-initial position in a larger phrase. Now 
note in Table 2 that the ratio of discourse to senten- 
tial usage was about 1:2. Of the 44 tokens judged 
to represent discourse use and fitting our prosodic 
model, one third were of Discourse Type A and two- 
thirds of Discourse Type B. 
llWe used Hindie's parser Fidditch\[7\] to obtain con- 
stituent structure and Fidditch aaid Church's part-of-speech 
program\[2\] for part of speech assignment. 
While overall results are quite significant, the 17 
items judged sentential which nonetheless fit the dis- 
course prosodic model must be explained. Of these 
17, 14 (representing two thirds of the total error) 
are conjuncts (11 'and's and 3 '0r's) which fit the 
type (b) discourse prosodic model. While all are thus 
first in intermediate phrase -- and, in fact, in into- 
national phrase -- none are utterance-initial. Both 
judges found such items relatively difficult to distin- 
guish between discourse and sentential use. 12 In (3), 
for example, while the first and seems clearly senten- 
tial, the second seems much more problematic. 
(3) "But instead actually we are bringing some 
thoughts on expert databases from a place that 
is even stranger and further away and that of 
course is the magical world of artificial intelli- 
gence." 
The difficulty in such cases appears confined to in- 
stances ofsentential coordination where the conjunct 
is not utterance initial. Table 3 shows how judgments 
were distributed with respect to our prosodic model 
when coordinate conjunctions are removed from the 
sample. Our model thus predicts 93.4% of non- 
Table 3: Prosod.y of Classified Non-Conjuncts, N=91 
Judgment Prosody 
" Discourse Sentential 
Discourse 36 3 
Sentential 3' 49 
(X 2 = 68.15, df = 1, p <.0Ol) 
conjunct cue phrase distinctions, as opposed to the 
84.2% success rate shown in Table 2. 
Our prosodic model itself can of course be de- 
composed to examine the contributions of individual 
features to discourse/sentential judgments. Table 4 
shows the distribution of judgments by all possible 
feature complexes for all tokens) 3 This distribution 
reveals that there is considerable agreement when 
cue phrases appear alone in their intermediate phrase 
(OF*, corresponding to Discourse type A in Figure 
1); such items are most frequently judged to be dis- 
course uses. There is even more agreement when cue 
phrases appear in non-initial position in a larger in- 
termediate phrase (NONF* -- Sentential type B in 
l~See Section 3. Of the 99 conjuncts in thin study, both 
judges agreed on a discom~e/sentential distinction only 42.4% 
of the time, compared to 78.6~ agreement on non-conjtmcts. 
Conjunct tokens represented two-thirds of all tokens the 
judges disagreed on, and 68:9% of tokens at least one judge 
was unable to assign. 
13Feature complexes axe coded as follows: initial 'O' or 
'NO': consists of a single intermediate phrase or not; medial 
'F' or 'NF': appears first in intermediate phrase or not; Final 
'D', 'H', 'L', or 'C': deaccented, or bears a H*, L* or complex 
pitch accent. Note that four cells (ONFD, ONFH, ONFL, and 
ONFC) are empty, since all items alone in their intermediate 
phrase must perforce come frrst in it. 
254 4 
Table 4: Prosodic Feature Configurations and Judgments, N=211 
OFD 
OFII 
All Tokens 
8 
% Judged Discourse 
100.00 
50.00 
% Judged Sentential 
0 
37.50 
% Unchussifiable 
12.50 
OFL 30 60.00 0 40.00 
OFC 9 77.78 22.22 
ONFD 
ONFH 
ONFL 
ONFC 
NOFD 
NOFIt 
0 
0 
NA 
NA 
NA 
NA 
8.47 
11.11 
42.86 
..... 50.00 
0 
0 
NA 
59 
NOFL '21 
NOFC 4 
NONFD 
NONFII 
NA 
NA 
NA 
22.03 
55.56 
4.76 
50.00 
89.29 
94.44 
NA 
28 
36 
NONFL .... 4 
NON FC 2 
NA 
NA 
NA 
69.50 
33.33 
52.38 
10.71 
5.56 
25.00 0 75.00 
0 .... i00.00 ....... 0 
Figure 1); tlhese tend to be judged sentential. How- 
ever, tokens which fit Discourse type B in Figure 1 
(first in a larger phrase and deaccented (NOFD) or 
with a L* (NOFL)) appear more problematic: of the 
former, there was disagreement on fiflly two thirds) 4 
While there is more agreement that tokens charac- 
terized as NOFIt (first in a larger phrase with a H* 
accent) or NOFC (same with a complex pitch accent) 
-- Sentential type A in Figure 1 --- are sentential, 
this agreement is certainly less striMng than in the 
case of tokens characterized a,s NONF* (non-initial 
il~ a larger phrase with any type of pitch accent -- 
Sentential type B). Since Discourse type B and Sen- 
tcntial type A differ only in 'type of pitch accent', we 
wight conclude that the pitch accent feature is not 
as powerfid a discriminator as the phrasal features 
'alone in intermediate phrase' or 'first in phrase'. 
As in our previous study, we also examined poten- 
tial non-prosodic distinctions between discourse and 
sentential uses. Of the orthographic and syntactic 
fi:atures we examined, we found presence or absence 
of preceding punctuation and part-of-speech to be 
most successful in distinguishing discourse from sen- 
tential uses. For the 113 tokens on which both judges 
agreed a.s to discourse or sentential status, 1~ orthog- 
ral)hy distinguishes between discourse and sentential 
use in 101 (89.4%) of cases. Specifically, 21 of 30 
discourse uses are preceded by punctuation and only 
3 of 83 sentential items. 
We also tbund that part-of-speech distinguishes 
discourse from sentential use, although less success- 
fully than orthography. If we simply predict dis- 
course or se.ntential use by the assignment most fre- 
quently associated with a given part-of-speech, both 
14And note that 91.3% of items in these two cells m'e 
conjmlcts. 
15Thls figm~ excludes those items which the transcriber 
omitted. 
Church's part-of-speech algorithm and Hindle's Fid- 
ditch predict discourse or sentential use in approx- 
imately 75% of cases where both judges agreed on 
discourse/sentential assigmnent. For example, we 
assume that since the majority of conjunctions and 
verbs are judged sentential that these parts-of-speech 
are predictors ofsentential status, and since most ad- 
verbials are associated with discourse uses, these are 
predictors of discourse status, and so on. While part- 
of-speech thus might seem less useful than ortho- 
graphic distinctions for our corpus, the fact that it is 
not subject to transcriber idiosyncracy might make 
it a more reliable predictor than orthographic indica- 
tors in the general case. Too, for text-to-speech ap- 
plications, in which one would like to infer discourse 
or sentential use in order to employ the appropriate 
intonational features when synthesizing the item in 
question, these text-based results are encouraging. 
5 Discussion 
Our findings for the first stage of our single-speaker 
multi-cue phrase study support the intonational 
model of discourse/sentential characteristics of cue 
phrases which we proposed in \[8\]. Discourse uses 
of cue phrases fell into two groups: in one, the 
cue phrase was set apart as a separate intermedi- 
ate phrase (possibly with other cue phrases); in the 
other, the cue phrase was first in its intermediate 
phrase (possibly preceded by other cue phrases) and 
either was deaccented or bore a L* pitch accent. 
Sentential uses were in general part of a larger in- 
termediate phrase: if first in phrase, they bore a H* 
or complex pitch accent. The association between 
discourse/sentential models and discourse/sentential 
judgments is significant at the .0(/1 level. We also 
found that the tokens we found difficult to clas- 
255 5 
sify were those in which disambiguation relied solely 
upon pitch accent, rather than some combination of 
pitch accent and phrasing. Furthermore, we found 
that orthographic cues (from transcription) success- 
fully disarnbiguate between discourse and sentential 
usage in 89.4% of cases in our pilot study. Part- 
of-speech was less successful in distinguishing dis- 
course from sentential use, disambiguating only 75% 
of cases in the study. 
The disambiguating power of both our textual and 
our prosodic models has both theoretical and prac- 
tical import. From a practical point of view, the 
construction of both text-based and prosodic models 
permit improvement in the generation of synthetic 
speech from unrestricted text \[9\]. With a prosodic 
model, we know how to convey discourse/sentential 
distinctions; with a text-based model, we know 
when to convey such distinctions. From a theo- 
retical point of view, our findings demonstrate the 
feasibility of cue phrase disambiguation in both text 
and speech and provide a model for how that disam- 
biguation might be done. Furthermore, these results 
strengthen the claim that the discourse structures 
crucial to computational models of interaction can 
indeed be identified. 

References 

\[1\] Bengt Alteuberg. Prosodic Patterns in Spo- 
ken English: Studies in the Correlation between 
Prosody and Grammar for Text-to-Speech Con- 
version, volume 76 of Lund Studies in English. 
Lund University Press, Lund, 1987. 

\[2\] K. W. Church. A stochastic parts program and 
noun phrase parser for unrestricted text. In 
Proceedings of the Second Conference on Ap- 
plied Natural Langvage Processing, pages 136- 
143, Austin, 1988. Association for Computa- 
tional Linguistics. 

\[3\] Robin Cohen. A computational theory of the 
function of clue words in argument understand- 
ing. In Proceedings, pages 251-255, Stanford, 
1984. COLING 84. 

\[4\] B. Grosz and C. Sidner. Attention, intentions, 
and the structure of discourse. Computational 
Linguistics, 12(3):175-204, 1986. 

\[5\] Barbara J. Grosz. The representation and use 
of focus in dialogue understanding. Technical 
Report 151, SRI International, Menlo Park Ca., 
1977. University of California at Berkeley PhD 
Thesis. 

\[6\] M. A. K. Halliday and Ruquaiya tIassan, Cohe- 
sion in English. Longman, 1976. 

\[7\] D. M. ttindle. Acquiring disambiguation rules 
from text. In Proceedings of the 27th Annual 
Meeting, pages 118-125, Vancouver, 1989. As- 
sociation for Computational Linguistics. 

\[8\] J. Hirschberg and D. Litnaan. Now let's talk 
about now: identifying cue phrases intonation- 
ally. In Proceedings of the 25th Annual Meeting, 
Stanford, 1987. Association for Computational 
Linguistics. 

\[9\] Julia Hirschberg. Assigning pitch accent in syn- 
thetic speech: The given/new distinction and 
deaccentability. In Proceedings of the Seventh 
National Conference, Boston, 1990. American 
Association for Artificial Intelligence. 

\[10\] J. Hobbs. Coherence and coreference. Cognitive 
Science, 3(1):67-99, 1979. 

\[11\] Diane J. Litman and James F. Allen. A plan 
recogntion model for subdialogues in conversa- 
tion. Cognitive Science, 11:163-200, 1987. 

\[12\] W. C. Mann and S. A. Thompson. I~lational 
propositions in discourse. Technical Report 
ISI/RR-83-115, ISI/USC, November 1983. 

\[13\] Janet B. Pierrehumbert. The Phonology and 
Phonetics of English Intonation. PhD thesis, 
Massachusetts Institute of Technology, Septem- 
ber 1980. 

\[14\] L. Polanyi and R. Scha. A syntactic approach 
to discourse semantics. In Proceedings, pages 
413-419, Stanford, 1984. COLING84. 

\[15\] M. E. Pollack, J. Ilirschberg, and B. Webber. 
User participation in the reasoning processes of 
expert systems. Technical Report MS-CIS-82-9, 
University of Pennsylvania, July 1982. A shorter 
version appears in the AAAI Proceedings, 1982. 

\[16\] R. Quirk. A Grammar of contemporary English. 
Longmans, London, 1972. 

\[17\] R. Reichman. Getting computers to talk like 
you and me: discourse context, focus, and se- 
mantics. Bradford. MIT Press, Cambridge MA, 
1985. 

\[18\] Deborah Schiffrin. Discourse Markers: Seman- 
tic Resource for the Construction of Conversa- 
tion. PhD thesis, University of Pennsylvania, 
1982. 

\[19\] Lawrence Schourup. Common discourse par. 
tieles in Effglish conversation. Garland, New 
York, 1985. 

\[20\] David Talkin. Looking at speech. Speech Tech- 
nology, 4(4):74-77, April-May 1989. 

\[21\] Ingrid Zuckerman and Judea Pearl. Compre- 
hension-driven gencratlon of meta~technical ut- 
terances in math tutoring. In Proceedings of 
the Fifth National Conference, pages 606-611, 
Philadelphia, 1986. AAAI. 
