On the Generation and Interpretation of Demonstrative Expressions* 
Jeanette Gundel 
Department of Linguistics 
University of Minnesota 
Minneapolis, MN 55455 USA 
Nancy Hedberg 
Department of Linguistics 
University of Minnesota 
Minneapolis, MN 55455 USA 
Ron Zacharski 
Department of Computer Science 
University of Minnesota 
Minneapolis, MN 55455 USA 
This paper presents necessary and sufficient conditions for the 
use of demonstrative expressions in English and discusses impli- 
cations for current discourse processing algorithms. We examine 
a broad range of texts to show how the distribution of demon- 
strative forms and functions is genre dependent. This research is 
part of a larger study of anaphoric expressions, the results of 
which will be incorporated into a natural language generation 
system. 
1. INTRODUCTION 
The primary purpose of this paper is to present a set of necessary 
and sufficient conditions for the use of demonstrative expressions 
in English, based on a corpus of naturally occurring discourse 
from a variety of spoken and written genres. We propose a 
comprehensive set of constraints on demonstrative use and sug- 
gest how they can be incorporated into a computational pro- 
cessing model which integrates the local centering and global fo- 
cusing aspects of discourse structure. Finally we show how our 
proposed algorithm for demonstratives can account for stressed 
pronouns as well. 
Existing computational work on demonstratives has been 
based on examples from only three genres: experimentally- 
elicited apartment descriptions (Linde 1979), technical dialogues 
(Reichman-Adar 1984), and scientific textbooks (Sidner 1983). 
Testing computational algorithms against multiple genres of 
natural discourse is important, especially given the universal 
scope of current frameworks (cf. Grosz and Sidner 1986). We 
have chosen to systematically examine texts from a broad range 
of genres, which vary in spoken versus written medium, number 
of participants, degree of pre-planning, and formality of situa- 
tion. These genres include informal conversation, partially- 
spontaneous televised discussion, newspaper articles, and 
planning and technical documents. 
2. REVIEW OF THE LITERATURE 
Demonstratives have generally been considered to be one type of 
deictic expression. Most of the linguistic literature on deixis has 
concentrated on classifying deictic expressions according to 
various parameters such as the linguistic or extralinguistic status 
of the referent. For example, Buhler (1982) distinguishes be- 
tween demonstratio ad oculos, anaphora, and deixis at phan- 
tasma, whose intended referents are respectively found in the ex- 
tralinguistic context, the linguistic context, or in memory or 
imagination. Similarly, Fillmore (1975) distinguishes gestural, 
anaphorie and symbolic uses of deictic expressions; and Lakoff 
(1974) distinguishes spatio-temporal, discourse, and emotional 
deixis. Such distinctions have been useful as a basis for descrip- 
tive classification of various uses of demonstrative expressions, 
but we do not feel that they are relevant for an explanatory ac- 
count of demonstrative function and so shall not elaborate further 
on them here. For the same reason we also will not discuss at- 
tempts to show that discourse deixis and/or anaphora is derived 
• from or less prototypical than spatio4emporal deixis (of., Lyons 
1977; Fillmore 1982). Our account is more compatible with the 
*The work presented here was partially supported under Control Data 
Corporation Grant #86M102 and a grant from the University of Minnesota 
Graduate School (Jeanette Gundel, Larry Hutchinson and Michael Kac, 
Principal Investigators). Support was also provided by a Doctoral 
Dissortation Fellowship from the University of Minnesota Graduate School 
to Nancy Hedberg. We would like to thank Karen Frederickson for kindly 
pri~vidlng us with transcripts of the casual conversations, Walling Cyre for ~ 
oviding the CDC documents, and Suellen Rundquist for helping in the 
nitial stages of this research. 
216 
view that deictic expressions signal a change in focus of attention 
while anaphoric expressions signal focus continuation (cf, 
Ehlich 1982; Bosch 1983). It is this view which most closely 
reflects the assumptions underlying computationaUy explicit 
models of focus-constrained discourse processing (e.g., Linde 
1979; Reichman 1985; Grosz and Sidner 1986). 
We turn now to a presentation of specific claims about 
demonstratives which have been made in the literature. 
Focus shift. 
a. That but not it may be used to refer to an item no longer in 
focus (Linde 1979). That marks the end of a discourse 
section (Reichman 1984, 1985). This but not that signals 
focus movement, except when this occurs in the scope of 
a quantifier or when its head noun is not identical to the 
head noun of an existing immediate focus (Sidner 1983). 
b. That but not it may be used.to refer to a preceding state- 
ment taken as a statement (Linde 1979). 
c. Stressed that must change focus (Isard 1975). 
ContrasL 
a. That may be used to indicate conwast, even for items cur- 
rently in focus (Linde 1979). When this and that are used 
together contrastively ('copresent' use), this specifies the 
primary, continuing focus and that specifies a secondary, 
temporary focus (Sidner 1983). 
Proximity. 
a. This is used to denote objects relatively close to the 
speaker and that for objects relatively far from the speaker 
or relatively close to the hearer. (Fillmore 1975; Halliday 
and Hasan 1976; Lakoff 1974) 
b. Both this and that may corefer with elements in the pre- 
ceding linguistic context, but only th& may eorefer with 
elements in the following linguistic context. (Fillmore 
1975, 1982; HaUiday and Hasan 1976; Lakoff 1974). 
c. Both this and that can be used to comment upon a 
speaker's own prior remarks, with this often signalling 
greater speaker involvement or continued discussion. 
Only that may be used to comment upon the remarks of 
another speaker (Halliday and Hasan 1976; Lakoff 1974; 
Sidner 1983). 
d. The interval denoted by a determiner this used with a time 
expression includes the speech time. (Fillmore 1975). 
That tends to be associated with a past time referent and 
this with one in the present or future. (Fillmore 1975; 
Halliday and Hasan 1976). 
Special effects. 
a. Demonstratives often have subtle emotional effects such 
as solidarity, distancing, sympathy, anger, irony, etc. 
(Lakoff 1974, Fillmore 1982; Halliday mad Hasan 1976). 
b. Unstressed determiner this has an indefinite specific use 
in colloquial narrative speech, (Lakoff 1974); shared 
attention and experience (Halliday and Hasan 1976); 
topicality (Prince 1981a) 
c. Unstressed determiner that phrases have a first-mention 
use for expected shared referents. (Wald 1983). 
3. CONSTRAINTS ON DEMONSTRATIVE USE 
The basic concepts which we take to be crUcial for any adequate 
description of demonstxatives are ones which are central to a the- 
ory of reference in general. These are concepts such as topic, fo- 
cus and various types of givenness. Since these concepts con- 
cern the cognitive status of a referent, definitions have often been 
iough and iatuitive, bv.:dcatiy con'ect but ~tot sufflcieLitly precise 
for eonlput~;ikmal iulplcmctmttion. On the oilier hand, because of 
the conlpledty of these concepts, attempts to ihrnish ntore pre- 
cise opcrallonal definitions, e.g. on the basis of surlhce line.'n of 
dcr or gLm,L matical rclatiot~s~ have failed to capture lhcir cognitive 
basis. 'lh(; ccsult has heel, a icrminological and conceptual cort- 
Nsion ill tl~C literature which has led computational linguists to 
create nc.w t:onsh;ucts such as fc)cns arid center, in some cases 
without rcl:tting tllc;m to similar linguistic concepts (Itajicov:t 
1987 is an, ltablc exception). Our aim hcrc is t-L characterize as 
p~vcisely as possible what the relevant concepts for deruouslra- 
iives arc :.hid how they relate to one anoihcr. In a later section wc 
will attcnlpt tl~ show how they Lelate to more operational con- 
structs prop 3sed in the, computational literature. 
Defiuit~.'ne~So As has often been pointed out, the basic notion 
which dctcrntines ~q)prol)riateness of a given rcfi.:rential cxpres. 
sion is the; status of the refcrc,t of lhat expression vis.a-vis a 
cognitive. :;tale of the addressee (cf. Chafe 19'/6). hi the weakest 
c~\:;t:, the SFeakcr c:~\[Lt:~.is the itddtcssee to uffdcrstand what type 
of eniily is beittg described, but not to uniquely identify the entity 
in q/,cslion. Such c,~ities arc typically reiEl'cnccd with indefinite 
itotni pbra!;cs The nlost basic distinction in demonstrative 
\[tntciion i~; between tile indefinite use of lhe determiner 1his, as 
~:xemplified in (t) and all other uses of hoth determiner and 
prouomiflal denlonstratives, which arc definite. As has bccll 
pointed ot,~ by a number of researchers, indcfinite this occurs 
only in cat;~,al, n,@a,ncd discouLsc. Tilts obscrvalion was con-. 
firmed hy our own study, which found instances of indefinite 
this ot,ly in the inhlrmal coaversalkms. 
(1) i couldn't sleep last night. My neighbors have this 
dog that kept me awake. 
We will be primarily concerned hete with definite noun 
phrases, where the speaker expects the addressee to uniquely 
identify the mlin'cnt on tile basis of the description in question. 
Th,e rcfe,e~lts of such expressions have often been characterized 
as 'given'. tlowevcr, as pointed out by Gundcl (1978a, 1978b) 
and l'rince (1981h), givcnness is not a mfitary concept. There 
arc diil'ereat senses of givenness each of which is relevant to the 
Rn'm of reicc,'ittg expr,-ssions, but in different ways. 
hletttiliabitity. In the weakest sense of givcnness, the slmaker 
expects the addressee to uniquely identify the referent, but the 
basis for th: identification is iiTelevant. Not only can it be lin- 
guistic or extralingttistic, based on entities in immediate or long 
term memory, but it need not he based on any previous shared 
experience ;:it all (cf. lIawkins 1978). The basis/'or the identifi- 
cation may be enc~led in lhe form itselt, as in (2). 
(2) I couldn't sleep last night. The neigltboes dog 
kept me awake. 
This type of givenness, which we rctisr to as identitiability, is 
both necessary and Slffficient tot the use of deliniie articles. 
Shared t'alailiarity. Most identifiable entities are identifiable 
because of some shared experience between speaker and aft. 
dressec; again this may be linguistic or exlralinguistic, based on 
local colRext or hmg term menuLly associated with shared cultural 
or personal expt:rienct:. It is tiffs slattls which we claint is neces- 
sary for the use of demonstrative expressions (with the exception 
of indcfinit,~ this ). Thus, (3) unlike (2), is li:licitous only if the 
addressee has prior knowledge of the dog in question. 
(3) I couldn't sleep last night. That dog next door 
kept me awake. 
ActilzatioJii. Entities which are fantiliar on the basis of presence 
in the imnlcdiate discourse context (either linguistic or extralin- 
gnistic) are not only shared but 'in awareness.' This sense of 
pjvcmless, which wc reler to as 'activated,' (of. Chafe 1976, 
Gundel 1978b, Haji~fov;i and Vrbovfi 1982) is necessary tor 
pronominal reference, including pronominal demonstratives. 
Thus, that' in (4) could only r~:li:r to the bmkiog of a dog if this 
\[lad lrg;eu ac~ivaR;d b) r the hnnu-diatc discourse context: 
(4) 1 couldn't sleep last night. That kept me awake. 
Activation is also a necessary condition for determiner this. 
Demonstcative this, both pronominal and definite determiner, has 
the additional condition that it not only be activated but speaker- 
activated, either linguistically or extralinguistically by virtue of its 
inclusion in the speakers context space, as in (5): 
(5) A: Have you seen the neighbors new dog? 
B: Yes, and that dog kept me awake all night. 
B': ??Yes, and this dog kept me awake all night. 
\[r~ focus. Finally, the most highly activated entities are not only 
in the speaker's arid ltearer's awareness but are the center of 
attention at ;t particular point in the discourse. We refer to this 
status as 'in focus.' Entities in .focus 1 always include at least the 
topic of the sentence as well as any higher level discourse topics 
which may not be overtly represented in the sentence itself. Un- 
der certain conditions, they may also include other elements such 
as the reference of the linguistic focus. Thus a shift in focus al- 
ways entails a shift in topic but not vice versa. The status 'in to- 
otis' is a necessary condition for unstressed pronomlnals and also 
for zero anaphora (cf. Gundel 1978b). 
The rehtlionship between tile various (definite) referential 
statuses and the forms that colTelate with them is shown in (6). 
These statuses are in a unidirectional implication rclation such that 
any status on lhe hierarchy implies all statuses higher on the hier- 
urchy but not vice-.versa: 
(6) GIVENNESS IIIERARCIIY: 
in focus --> activated --> slmred --> identifiable 
it that, this fllat N the N 
this N 
40 CRITIQUE OF CLAIMS IN TIlE LITERATURE. 
Focus Shift. Observations regarding the focus shift functiou of 
demonstratives follow naturally from the theory outlined above 
given two additional, but uncontroversial assumptions -- that 
pronominal it is necessarily unstressed and that the overwhelming 
majority of definite noun phrases are not only uniquely identifi- 
ahle, but shared. As noted above, the prinlary distinction between 
demonstrative pronouns and unstressed personal pronouns is that 
the referents of the latter must not only be activated, but in focus. 
A pronominal whose referent is not currently in focus is 
necessarily stressed (cf. Hirschbcrg and Pierrehumbert 1986). 
Since it is necessarily nnstressed, the only third person singular 
neuter pronominals available for focus shift are the demonstra- 
tives that and this. This accounts for the distribution of that ver- 
sus it noted by Linde (1979) as well as the uses of that noted by 
Reichman. Use of that in referring to previous statements (also 
noted by both these authors) is just a special case of focus shift, 
since the focus of attention at the point "after a statement is made is 
typically not the statement itself. However, our data does not 
support Sidner's (1983) claim that this but not that is used for 
focus movement. Exatnplcs illustrating the focus shift function 
of pronominal this and that include the following: 
(7) K i: And..So what he DID was ...came in, set up the 
free... 
2: and then he nlade wassail, with rum in it'? 
3: And..made it in coffee cans and heated it on the 
stove in the graduate lounge. 
A 4: Oh, gee. 
K 5: And this was the solstice tree. 
lu (7), tile topic and hence the focus of attention in K5 is the tree, 
which is activated by its mention in K1 thus licensing tile 
lThere is some confusion in tile literature resulting from the fact that tile 
term 'focus' has been used in two distinct and at least partly opposite ways (cf 
Ilaji~.bv~ 1987). We use 'in focus' to refer to tile psychological notion of 
focus of attention (tlaji~ovgs focUSAl) and 'linguistic focus' to lefer to tile 
point of linguistic prominence in tim sentence (the comment). 
2117 
use of a pronominal. However, since it is not in focus in K2, 
K3, and A4, the reference to it in K5 constitutes a focus shift 
and thus requires a slxessed demonstrative form. Note that since 
the tree was speaker activated, either this or that could be used. 
(8) N 1: I like the poor dog who was buried six times in 
one day! 
K 2: Oh. That must, that must be a story that comes 
from the Second Minnesota history, 
3: because that didn't appear in the, in the ah di- 
ary, 
4: so it must have come from somewhere, 
In (8), the topic and hence focus of K2 is the story. 
Since the story is activated but not in focus in N1 (the focus of 
attention in N1 is stories that N likes), reference to the story in 
K2 constitutes a focus shift and thus requires a stressed 
demonstrative. Note that the story continues to be topic (in focus) 
in K3 and K4 and that in K3, an unstressed pronominal it could 
have been used instead of that. 
(9) N 1: "Bob loves Mary", 
2: and someone else wrote "Mary loves Jim" 
3: and I wrote "Jim loves Bob"! (laughter) 
4: It was three different handwritings, three differ- 
ent people. 
K 5: Yeah, that's good. 
(9) illustrates the use of that to refer to a previous statement or 
utterance. Since that refers to the topic of K5, the story activated 
in N1-4 (but not in focus), use of that constitutes a focus shift. 
The focus shift function of determiner that (cf. Reichman 
1984) can be explained as a consequence of Grice's maxim of 
quantity, specifically don't be more informative than necessary 
(cf. Grice, 1975). Since the overwhelming majority of definite 
noun phrases refer to entities that are shared, use of a demon- 
strative determiner as opposed to the less restrictive definite arti- 
cle in most instances carries little additional information. Thus a 
demonstrative determiner is generally used only when the sig- 
nalling of shared familiarity has some additional communicative 
function. This is the case when there is a shift in focus, as in 
(10) and (11), since a focus shift always entails a shift in topic 
and topics are necessarily shared (cf. Gundel 1985). 
In the particular case of (10), there is not enough infor- 
mation encoded in the noun phrase itself to allow the addressee to 
uniquely identify the referent. Thus a demonstrative (as opposed 
to a definite article) is required in order to link the referent up 
with entities shared as a result of immediate discourse context. In 
(1 lb) on the other hand, the demonstrative functions simply to 
signal a focus shift and therefore a definite article would be 
equally appropriate. 
(10) a. John, this speech was a magnificent triumph for the 
President. He showed he could stay awake for 
twelve whole minutes. He showed that he could 
speak every word off of his teleprompter, even the 
long ones. But the speech doesn't have any chance 
of putting the scandal behind him, because the 
scandal is not about mistakes, as he said, and it's 
not about mismanagement, as the Tower Commis- 
sion said. It is about a betrayal of principles, it's 
about lying, and it's about breaking the law. 
b. And those issues remain. \[McLaughlin 3/6/87\]. 
(11) a. These incredibly small magnetic bubbles are the 
vanguard of a new generation of ultradense mem- 
ory-storage systems. 
b. These systems are extremely rugged: 
c. they are resistant to radiation and are nonvolatile. 
Contrast. The contrastive function of demonstratives, like the 
focus shift function, is related to the fact that contrast is marked 
by stress and pronominal it cannot carry stress. Moreover, con- 
trast may be just a special case of focus shift, since a eontrastive 
noun phrase always brings into focus other entities with which it 
is being contrasted, as illustrated in (12): 
(12) a. In both cycle steal mode and DPC, the attachmeni 
feature...responds with a condition code. 
b. For commands that do not require interrupts (that is, 
commands executed under DPC), the eonditi~m 
code provides current device status information. 
c. For commands that require executiou in cycle steal 
mode followed by an inten'upt request, this first 
condition code provides information concerning 
acceptance of the command by the attachment 
feature. 
d. Upon interrupt servicing by the processor, the at 
tachment feature provides a second condition 
code and an interrupt word. 
Thus, the referent of thisfirst condition code in (12c) is already 
in focus since it is also the topic of (12b). However, ,,dnce the 
use of this noun phrase brings into tbcus contrasting condition 
codes (cf. a second condition code in (12d)) it is not only 
contrastive, but constitutes an implicit focus shift as well 
Proximity. The speaker-activated condition on this predicts 
correctly that both this and that can comment upon a speaker's 
own prior remark as in examples (7) and (10) respectively, but 
only that can be used to comment upon the remarks of another 
speaker, as illustrated by example (9). 
The same condition also explains why this is used for ex- 
tralinguistic objects relatively close to the speaker and that for 
those relatively further away as in (13), and why the interval de ~ 
noted by determiner this includes speech time, while that tends to 
be associated with some time prior to speech time as in (14) 
This is so because speaker-activated means not only linguistically 
activated but activated within the speaker's context space. 
(13) N: This tastes like water. This tastes like THAT! 
Waita minute--the stuff that's $1.99 for two quarts 
hastes a- the same as one that's $2.07 for a fifth. 
(14) K: There he was that hairy hound from Buo 
dapest/Never leaving us alone./Never have I ever 
known/a ruder pest 
Special Effects. As seen in the diagram in (6), each of the 
referential statuses is also correlated as a necessary condition for 
a different type of definite reference. Since the statuses are 
implicationally related, reference with a particular form will 
generally imply appropriateness of reference with all forms asso- 
ciated with a status higher on the hierarchy, but not vice-versa. 
Thus, pronominal that in (13) may be replaced with that wine and 
these systems in (llb) can be replaced with the systems. 
However, not all cases of the N are replaceable with that N, as 
illustrated in (2) and (3); and not all cases of that N are 
replaceable with that., as illustrated in (3) and (4). 
The maxim of quantity would dictate that speakers will 
use the strongest possible form, i.e. will signal the most infor- 
mation as is relevant regarding the givenness status of the entity 
in question. The same maxim predicts that speakers would not 
use a stronger form than necessary in a given situation, i.e. they 
will not signal more information than is appropriate. Violation in 
either direction will often result in a special effect or implicature, 
as in (15), where the use of this as opposed to the equally 
appropriate the conveys an effect of solidarity: 
(15) Gov. D (from videotapel: I've got the energy to ran 
this marathon, the slrength to run this country, 
the experience to manage our government, and the 
values to lead our people. \[McLaughlin, 3/20/871 
We already noted that use of a demonstrative determiner 
often has a special effect, such as signalling a topic shift, because 
virtually all definite NPs are also shared, and thus demonstrative 
determiners do not normally convey much more infommtion than 
would be conveyed by the definite article. Similarly, in cases 
where the referent is not activated, determiner that acts as an ove~ 
signal to the addressee to search long term memoxy for the 
referent, as in (16): 
218 
(16) A: Yet* can get those little magic fi~ger jnbbies 
ihat tell you when ~o water ~t. They're 
only $10 or soumthing. 
5° ~.,TENP.E DIIFkCERENC.'I~;I 
The distdUt~tion of demonstrative types difli.qs acco, ding to 
gem:e, as ~;cen in Table 1 (See lfedheyl,, flnthCOming, ibr more 
(m these g¢ we~.) Thus, ti>r example, the relative percentage of 
pronontinal that appears to be gJ,:atcr ill multi-pmticipant oral 
communication d~an in written comntuni~atkm. The casual co,l- 
versatiom; *,rid electronic specification documents represent the 
r,=sl~ctive c~.iremes on tiffs sca/e~ 
~ha~ ~hat N this this N Total 
~;POKEI~q G~C.NRE~ 
CasualCo~rersations 49% i5 % 17 % 19% 3t6 
Televisedl;iscussio~t 30 % 15 % 12 % 34% 1i22 
WRl'f'FElq GENI,~E,q 
Ncwspap~;r Articles 14 % 2l % 7 % 58 % 83 
Pltvtming l)(~;ttruoiit ! 1% 9/) % 7 % 53 % {'~ 3 
Sty;~,ei ficalioq l)ocuU,tmt 1% :i % 4 % 90 % t 02 
fj!121q .I:.. l~q),,W;t!.:~!iv ~ di~)yibutiiA~2 hj_ ~£!v.e K~,jre~ _. 
'|'he distribution of demonstrative fl)~ ms and flmctlons for 
each genre will be individually presented and discussed below. 
5.1 Cas~m~ co~wersatio~,so The casual conve, sations were 
taped fbr tm q}ose:; unrelated to linguistic analysis; participants are 
family members reunited lbr holiday gatherings: 
~.hat ttnat N this this N Total 
Same speak.~r 67 18 7 9 101\[ 
Other Sl~:ak :r 58 3 5 .... 6 6 
i ~,xh'alinguis tic 26 4 36 5 7 i 
Reminder - - 17 ....... 1 7 
Indefinite ........ 36 3 6 
Tinte ....... 2 10 12 
"orwmd ....... ! .... 4 
Other 4 4 1 .... 9 
Total i\[~S 46 55 60 316 
'l'ab~e 2. Demonstratives ik~ casual eo~versations 
Other Speaker's Remarks° A large proportion of pronom- 
inal that in the casual conversations is due to its use for refer.. 
ence to an immediately preceding contribution of another conver- 
sational pardcipm~t. For an example see 0) above. 
As noted above, pronominal and determiner this require 
that the referent be speaker-activated. Thus we would expect to 
find that used extensively in any nmltiple-participant discourse, 
such as casaal conversation. While, we do have five exceptions 
to this gene,:alization, where this is used to refer to an entity acti- 
vated by tt,: addressee, all of these are clarification questions re 
questing re:,'erent identification, perhaps conveying polite inten- 
tion to not iaten'upt, as in (17): 
(17) iq: So yesterday I finished up the day's work and put it 
in gear and nothing happened. The cable for the 
trmlsmission .... 
M Is this your car? 
N: No. Truck. The old, beat up, lousy van I have to 
drive. 
E~g~'aling~dstie ~'efereneeo While tile linguistic or ex- 
tralinguistk status of the referent is irrelevant for predicting the 
form of a 1err,ring expression, this status does differ across gen- 
res. For obvious reasons, extraiinguistic reference occurs pri- 
marily in face-to4ace interaction, (18) is an example showing the 
use of tha* ~ for shifthag tile focus onto an extralinguistic entity, 
fblk~wed by subsequent reference using it'. Note that activation is 
accomplish~.,d here by a gesture: 
(18) 1( Wt,at is that, Beethoven whaL on that teeshirL 
N 1 think iUs the Ninth, isn't it. 
K. Number Three. lafler reading it\] 
N. Oh it is. 
Renfinder that° Since determiner that does not require actiw~- 
tlon of the referent, but only that it be shared, it can be used to 
rote, to entities that art: not present in the irrmmdiate discourse 
context, as in (19). 
(19) K: I realized something that seems significant to me 
ahout George.. that in the, in the fall, he... as 
everyone else, he wears .... 
N: \[Clothes. 
A: \[!.eaters. 
K: No. \[Those ldnd of tennis shoes that are 
expensive,: 
A: \[Boots. 
N: Adidas. 
A: \[Adidas. 
K: \[Adidas, ok. 
As notcd above, that N is sometimes used as an explicit signal to 
the hearer to search memory for the referent. One would expect 
such signals to be most common in interactions between individ- 
uals with shm'ed personal experiences, so it is not smprising that 
they would occur fi'equently in castml conversations between 
family menlbers. Typically such phrases include a relative clause 
specifying additional information to aid in the search and are 
often embedded in a request for confirmation that the referent gas 
indeed been located.. Since the referent is shared but not acti- 
vated, it frequently occurs in left-dislocated constructions which 
have tile lunction of introducing or reintroducing a topic into the 
discourse. (eL Keenan and Schieffelin. 1976) 
5.2 Televised Discussions. The televised discussion 
was a videotaped episode of The McLaughlin Group (initial tran- 
script obtained from the Federal News Service). This genre is 
similar to the casual conversations in being spoken and multi. 
pmticipant, but differs in degiee of fi)rmality and spontaneity and 
m awareness of an audience. Four journalists participate in a 
sttucttned discussion about current affairs under tile control of a 
moderator As shown in Table 3, most of demonstratives are 
pronominal that, as extxzctcd in interactive discourse. 
that that N this flds N Total 
Same Speaker 21 6 -- 4 3 1 
Other Speaker 26 2 ..... 2 8 
Discourse Topic I 1 6 7 22 3 6 
Extralinguisfic -- 1 ...... 
Remhader -o 2 ..... 2 
Indefinite ....... 3 3 
T:\[me-Person-Place ........ 11 I 1 
Forward ...... 5 -,. 5 
Other -- 1 2 2 5 
Total 48 18 I 4 42 122 
Table 3. Demonstratives in televised discussions 
Discourse topic this. The televised discussions are 
distinguished from the casual conversations by the frequent use 
of this to refer to non-speaker activated entities. Such uses 
contradict our claim that this" must be speaker-activated. To 
account for such examples, we suggest that a distinction be made 
between inclusive and exclusive speaker space, in the case of 
local discourse segments, speaker space often excludes the ad- 
dressee, but with higher-level discourse topics, which are often 
shm'ed, speaker space includes the addressee. In such cases, 
which are characteristic of highly sl'uctured interactions, this' may 
be used for something which was not initially activated by the 
speaker. In (20), reference is made to the explicit discourse topic 
of' tlm segment, Gorbachev's decision not to cancel the smnnfit: 
1The categories Same Speaker and Other Speaker include only referents 
which are locally activated in the immediately preceding utterance. Thus, 
they ate rnutually exclusive with the category of discourse topic. 
219 
(20) G: The business of trying to play Kremlinologist on 
this decision, I think it's simpler than that. I re- 
ally think that Reagan looked very, very vulnerable 
at that time. It was the thing for Gorbacbev to do, 
and he just misread it. \[McLaughlin, tO/30/87,21\] 
5.3 Planned, non-interactive genres. Tables 4-6 show 
the distribution of demonstratives in newspaper stories (New 
York Times "Week in Review" section, 6-11/87); a University of 
Minnesota administrative planning document; and an electronic 
specification document supplied by Control Data Corporation. 
A characteristic of these non-interactive, written genres is 
a relatively high percentage of determiner this. This may be part- 
ly attributable to the fact that everything that is activated is speak- 
er activated. In addition, there are malay time expressions with 
this in the newspaper articles and metadiscourse references such 
as this document in the planning document. Unlike the 
discourse-topic use in the televised discussions, referents of 
determiner this in the written genres are typically activated in tile 
immediately preceding sentence and constitute a focus shift. 
that that N this this N Total 
n-0 t 3 1 1 3 8 
n-1 9 11 3 12 35 
n-2+ -- 2 1 15 18 
Time-Person-Place -- 2 1 17 2 0 
Forward ...... 1 1 
Reminder - - 1 .... 1 
Total 12 1 7 6 4 8 8 3 
Table 4. Demonstratives in newspaper stories 
that that N this this N Total 
n-0 9 9 .... 18 
n-1 -- 13 6 20 39 
n-2+ -- 2 -- 9 11 
Time ...... 4 4 
Metadiscourse ...... 11 11 
Total 9 2 4 6 44 8 3 
T_able 5. Demonstratives i n- planning d0c_ument 
that that N this this N Total 
n-0 1 5 -- 3 9 
n-1 .... 4 78 82 
n-2+ ...... 8 8 
Metadiscourse ...... 3 3 
Total 1 5 4 92 102 
'Fable 6. Demonstratives in specification document 
Informative this. A use of determiner this expressions which 
is found exclusively in the non-interactive genres is to informa- 
tively redescribe a referent. In the newspaper stories, these are 
typically redescriptions of topical referents (already in focus) 
which would have been specifiable with an unstressed pronoun if 
the extra material had not been included, as in (22). Although 
speakers generally use the strongest possible form of referring 
expression, here a weaker form is being used for a special 
reason, namely to introduce new information in the noun part. 
(22) Nearly lost in the polemic was Judge Kennedy 
himself. That was ironic, because in many ways 
this former small-city lawyer with the sta- 
ble marriage and three attractive children 
and the fine reputatiou appears to personify just 
those values that made the image of Ronald Reagan 
so attractive after the convulsions of the 1960's and 
1970's. \[New York Times, 11/15/87, 4:ll 
Informative this is used in the electronic specification 
document for obligatory demonstrative reference to the referent of 
a heading which is activated but not yet in focus, as in (24). 
Note that here an unstressed pronoun would not be possible. 
1The rows n-0, n-I and n-2+ refer to the distance between the demonstrative 
expression and its antecedent. 
(24) Pg_!I~ The attachment leaturc sends this 
inbound tag to the Series/1 channel controls to 
indicate a poll capture for interrupt se~wicing or non- 
burst cycle steal servicing. It is not used to signal a 
burst transfer. 
6. IMPLICATIONS FOR ALGORITItMS 
The previous sections Woposed constraints on demonstratives 
and discussed their use in different genres. This section outlines 
components of a natural language system that would capture the 
relevant notions of shared familiarity, activation and form; and 
explores possibilities for incorporating these into current dis~. 
course-processing algorithms. 1 
Shared familiarity. At minimum, a computational rrtodel of 
shared familiarity requires maintenance of a user discourse his-. 
tory in which a record is maintained of all entities referenced in 
conversations with a particular user. Thus, for any entity in its 
knowledge base, the natural language system knows whether/bat 
entity has been discussed before (shared familiarity) or not 
(familiarity unknown). Only in the former case can a definite 
demonstrative expression be used. The recognition of discourse 
units (e.g. speech acts of Allen 1983) and the aelations between 
them, e.g. conversational moves (Reichman 1985), rhetorical 
predicates (McKeown 1985, Mann and Thompson 1986) are also 
important in demonstrative resolution. These require a so- 
phisticated user model which keeps a record of beliefs and inten- 
tions of discourse participants. While such a model could be in- 
corporated into existing discourse structure frameworks (e.g. 
Grosz and Sidner 1986), no specific proposals to account for 
shared familiarity have yet been advanced (but see Sparck Jones 
1986). 
Activation. An adequate model of activation must isolate that 
subset of shared entities which is activated at any given point in 
the discourse. This subset includes entities referenced in the im- 
mediately preceding sentence, entities present in the immediate 
spatio-temporal context of the discourse, and beliefs and inten- 
tions relevant to the cmTent discourse segment. Many currem 
discourse algorithms which function at the local level of dis- 
course structure can be used to model activation due to the imme- 
diately preceding sentence. For example, Haji~ov~i (1987) points 
oat that elements in McKeown's (1985) potential focus list can be 
equated with activated elements. Similarly, in the current 
centering paradigm (Grosz, Joshi and Weinstein 1986, Brennan, 
Friedman and Pollard 1987) elements in the set of forward look- 
ing centers can also be considered activated. At the global level, 
Sidner and Grosz (1986) describe a model of discourse structure 
which indicates currently activated beliefs and intentions at any 
• given point in a discourse. 
In Focus. Elements in focus are those which are most highly 
activated. 2 These always include at least the topic of the sen- 
tence, any high-level topics (including those not overtly repre- 
sented in the sentence), and under certain as yet poorly under- 
stood conditions, the referent of the linguistic focus. In the cut'-. 
rent centering paradigm, the topic of the sentence is equated to the 
backward looking center (ef. Joshi and Weinstein 1981). In this 
l Kryk (1987) presents an algorithur for demo~lstrative interpretation in 
Polish and English based on Sidner's 1983 algorithms. Kryk's algorithm, 
bowever, does not appear to be explicit enough to serve as a basis for 
implementation. Moreover, some of the algorithm's predictions, for 
example that the demonstrative that is never used fbr non-copresent 
extralinguistic reference whereas non-copresent this is u,':ed only for 
exla'alinguistic reference, are inaccurate. 
2Although our notion of activation is similar to Haji~ovlt (1987), it differs 
in that she considers the linguistic f~us to be more highly activated tha~t the 
topic. We feel that our assumptions about tiegrces of activation me more 
compatible wifli the fact, noted also by llaji~"ovfi, that topic maintains iir~ 
activation longer than does the referents of the linguistic focus. 
'2 2 O 
paradigm nns~essed pronouns are viewed as preferring an ex- 
isting bacl~wards-lookhlg center as antecedent. We have sug- 
gested that use of a demonstrative expression or stressed pronoun 
signals that the option to shift the backwards-looking center has 
been selected over the otherwise-preferred option of continuing 
an existing backwards-looking center. Brennan, Friedman and 
Pollard (1987), in fact, mention in passing that stressed pronouns 
in oral discourse could be used to unambiguously signal one such 
type of cer~ter shift. (cf. Sgall (1984), who remarks that demon- 
stratives c~Jn be used to unambiguously specify reference to the 
linguistic fi ~cus of the irmnediately preceding sentence.) 
Genre differences. A final application of our constraints is in 
the atea of discourse genre variation. We have shown that both 
the form and function of demonstrative expressions varies be- 
tween diffi;rent spoken and written discourse genres. Our final 
suggestion is that these and other genre differences should be 
further explored, so that it will eventually be possible to design 
maximally-efficient discourse-processing algorithms which dif- 
ferentially e, xploit such genre distinctions. 

REFERENCES 

Allen, James. 1983. Recognizing intentions from natural 
langual;e utterances. Computational models of discourse, ed. 
by M. \]Irady and R. C. Berwick, 267-330. Cambridge, MA: 
MIT P~ess. 

Bosch, Peter. 1983. Agreement and anaphora. London: 
Academic Press. 

Brennan, ~usan E., Marilyn W. Friedman and Carl Pollard. 
1987. A centering approach to pronouns. Proceedings of the 
25th A,anual Meeting of file Association for Computational 
Lingui~,~tics, Stanford University, 155-162. 

Buhler, Karl. 1982. The deictic field of language and deictic 
words. Speech, place, and action, ed. by R. J. Jarvella and 
W. Klein, 9-.30. Chichester, NY: John Wiley & Sons. 

Chafe, Wallace. 1976. Givenness, contrastiveness, subjects, 
topics, and point of view. Subject and topic, ed. by C. N. 
Li, 25-:;5. New York: Academic Press. 

Ehlich, Koarad. 1982. Anaphora and deixis: same, similar or 
differei~t? Speech, place, and action, ed. by R. J. Jarvella 
and W~ Klein, 315-338. Chichester, NY: John Wiley & 
Sons. 

Fillmore, Charles J. 1975. Santa Cruz lectures on deixis. Indi- 
auaUn~versity I,inguistics Club. Bloomington, IN: Indiana 
University Linguistics Club. 

Fillmore, Charles J. 19;~2. Towards a descriptive framework for spatial 
deixis. Speech, place, and action, ed. by R. J. Jarvella and 
W. Kh,'~n, 31-59. Chiehester, NY: John Wiley & Sons. 

Grice, H.P. 1975. l~gic and conversation. Speech Acts, 
Syntax and Semantics 3, ed. by P. Cole and J. Morgan, 41- 
58. New York: Academic Press. 

Grosz, Barbara J., A..K. Joshi and S. Weinstein. 1986. To- 
wards a computational theory of discourse interpretation, ms. 

Grosz, Barbara J. anti Candace L. Sidner. 1986. Attention, in- 
tendons, and the structure of discourse. Computational Lin- 
guistics 12.175-204. 

Gundel, Jc~mette K. 1978a. Sa'ess, pronominalization and the 
given-.n,~w distinction. University of Haw:tii working papers 
in linguistics I0. 1-13. 

Gundel, Jc~mette K. 197~b. A universal constraint on deletion. Proceedings 
of the 12th International Congress of Linguistics,ed. by W. 
Dressier and W. Meid, Innsbrucker Beitriige zur Sprachwis- 
senscha ft, 532~536. 

Gundel, Jc~mette K. 1985. Shared knowledge and topicality. Journal of 
Pragmadcs 9. 83-107. 

Haji~ov~i, Eva. 1987. Focussing - A meeting point of linguistics 
and artificial intelligence. Artificial Intelligence 1I: methodol- 
ogy, systems, applications, ed. by Ph, Jorrand and V. 
Sgux~ev, 311-321. Amsterdam: North-Holland. 

Haji~ovtt, Eva and Jarka Vrbov~i. 1982. On the role of the hier- 
archy of activation in the process of natural language un- 
derstanding. Proceedings of the Ninth International Con- 
ference on Computational Linguistics, Pxague, 107-113. 

Halliday, M.A.K. and Ruquaiya Hasan. 1976. Cohesion in 
English. London: Longman. 

Hawkins, John. 1978. Definiteness and indefiniteness: a study 
in reference and grammaticality prediction. Atlantic High- 
lands, NJ: Humanities Press. 

Hedberg, Nancy. Forthcoming. The discourse function of cleft 
sentences. University of Minnesota dissetaation. 

Hirschberg, Julia and Janet Pierrehumbert. 1986. The intona- 
tional structuring of discourse. Proceedings of the 241h An- 
nual Meeting of the Association for Computational 
Linguistics, Columbia University, 136-144. 

Is,-u'd, Stephen. 1975. Changing the context. Formal semantics 
of natural language, ed. by E. L. Keenan, 287-296. Cam- 
bridge: Cambridge University Press. 

Joshi, A.K., and Weinstein, S. 1981. Control of inference: role 
of some aspects of discourse structure - centering. Proceed- 
ings of tile 1981 Meeting of the International Joint Confer- 
ence on Artificial Intelligence, Vancouver, 385-87. 

Keenan, E. Ochs and B. Schieffelin. 1976. Topic as a discourse 
notion: a study of topics in the conversations of children and 
adults. Subject and Topic, ed. by C. Li, 337-385. New 
York: Academic th'ess. 

Kryk, Barbara. 1987. On deixis in English ea~d Polish: the role 
of demonstrative pronouns. Frankfurt am Main: Verlag Pe- 
ter Lung GmbH. 

Lakoff, Robin. 1974. Remarks on this and that. Papers from 
the Tenth Regional Meeting of the Chicago Linguistics Soci- 
ety, ed. hy M. LeGaly, R. Fox, and A. Bruck, 345-356. 
Chicago, 1L: Chicago Linguistics Society. 

Linde, Charlotte. 1979. Focus of attention and the choice of 
pronouns in discourse. Discourse and syntax, ed. by T. 
Givon (Syntax and semantics, 12), 337-354. 

Lyons, John. 1977. Semantics, Volume 2. Cambridge: Cam- 
bridge University Press. 

Mann, William C. and Sandra A. Thompson. 1986. Rhetorical 
structure theory: description and consla~uction of text struc- 
tures. Reprint ISI/RS-86-174. Information Sciences Insti- 
tute. Marina del Rey, California. 

McKeown, Kathleen, R. 1985. Text generation. New York: 
Cambridge University Press. 

Prince, Ellen. 1981a. On the inferencing of indefinite-this NPs. 
Elements of discourse understanding, ed. by A. Joshi, B. 
Webber, and I. Sag, 231-250. Cambridge: Cambridge Uni~ 
versity Press. 

Prince, Ellen. 1981b. Towards a taxonomy of given and new informao 
tion. Radical pragmatics, ed. by P. Cole, 223-256. New 
York: Academic Press. 

Reiehman, Rachel. 1985. Getting computers to talk like you and 
me: discourse context, focus and semantics (art ATN model). 
Cambridge, MA: MIT Press. 

Reichmau-Adar, Rachel. 1984. Technical discourse: the present 
progressive tense, the deictic "that," and pronominalization. 
Discourse Processes 7.337-369. 

Sgall, Petr. 1984. A note on text and reference. Contributions 
to functional syntax, semantics and language comprehension, 
ed. by P. Sgall, 139-144. Amsterdam: John Benjrmfins. 

Sidner, Candace, L. 1983. Focusing in the comprehension of 
definite anaphora. Computational models of discourse, ed. 
by M. Brady and R. C. Berwick, 267-330. Cambridge, MA: 
MIT Press. 

Sparek Jones, K. 1986. Issues in user modelling for expert 
systems. Artificial intelligence and its applications. New 
York: John Wiley & Sons. 

Wald, Benji. 1983. Referents and topic across discourse units: 
observations from current vernacular English. Discourse 
perspectives on syntax, ed. by F. Klein-Andreu, 91-116. 
New York: Acadenlie Press.
