1 Reference Diaries 
Herbert H. Clark 
and 
Catherine Marshall 
Stanford University 
When we make a definite reference to a thing, 
we normally make sure that our audience "shares" 
with us certain knowledge about that thing. To 
refer to a woman as she, the woman, or Nancy, we 
usually have good evidence that our audience knows 
about her too. But exactly what "shared" knowledge 
is required? This question is critical if we are 
ever to discover how people make or interpret def- 
inite reference--how they represent knowledge in 
memory and consult it in uttering and interpreting 
expressions like she, the woman, or Nancy. The ques- 
tion is critical if we are ever to characterize 
the mental archive people have for storing the 
facts they need to know for definite reference. 
We will argue that this archive has to be another 
detailed diary, or reference diary, supplemented 
by atlases, histories, and certain other reference 
texts. To make this argument, and to see what goes 
into the archive, we will examine the prior ques- 
tion, What "shared" knowledge is required for def- 
inite reference? As it happens, this question 
leads directly to a puzzle we will call the mutual 
knowledge paradox. It is in the solution of this 
puzzle that we get our best clues as to what the 
reference diary must be like. 
The Mutual Knowledge Paradox 
Imagine that there is a Marx brothers retro- 
spective on at the local theater for which there 
are two or three movies a night for several even- 
ings. Against this background consider the fol- 
lowing scenario: 
Version i. On Wednesday night Ann and Bob 
go to see Monkey Business. The next morn- 
ing Ann meets Bob and asks, "what did you 
think of the movie?" 
What we are interested in is Ann's definite ref- 
erence the movie, which she intends to refer to 
Monkey Business. What facts does Ann have to 
assure herself of before she can felicitously 
make this reference? Our interest here is in only 
those facts that are involved in "shared" knowl- 
edge. As a first condition, for example, Ann must 
herself have a certain awareness of Monkey Busi- 
ness.. For now we will express that awareness as 
"knowing about R" (where R stands for the referent 
Monkey Business). Thus one fact Ann must assure 
herself of is this: 
(i) Ann knows about R. 
But is this enough? Of course not, for (i) 
provides no assurance that Bob knows about Monkey 
Business. The way it fails can Be made clear in 
a variation on the original scenario that goes like 
this: 
Version 2: On Wednesday night Ann and Bob go 
to see Monkey Business, but neither knows 
that the other went too. The next morning 
Ann meets Bob and asks, "What did you think 
of the movie?" 
Although version 2 satisfies condition (i), Ann 
has clearly made her definite reference without the 
right assurances. If (I) were all that had to be 
satisfied, version 2 would lead to a felicitous 
definite reference. Since it does not, we must 
add another condition, and the obvious one is that 
Ann must also assure herself that Bob knows about 
the movie, condition (2): 
(2) Ann knows that Bob knows about R. 
(If it seems too strong to require knowledge in- 
stead of belief, each know can be replaced by 
believe; without legislating on the argument, we 
will stick with know). 
At first, conditions (i) and (2) together seem 
enough, but it is easy to show that they are not. 
Consider this version of the original scenario: 
Version 3: On Wednesday night Ann goes to see 
Monkey Busine@s , and there she sees Bob. 
But he doesn't see her, and she realizes 
this. Furthermore, she realizes that Bob, 
unlike herself, might have seen A Day at the 
Races and A Night at the Opera, which are 
also showing that night. The next morning 
Ann meets Bob and asks, "What did you think 
of the movie?" 
Although Ann has satisfied conditions (i) and (2)-- 
she knows about Monkey Business and she knows that 
Bob knows about Monkey Business--she has not yet 
assured herself of enough. She cannot be sure Bob 
won't take the movie as referring to A Day at the 
Races or A Night at the Opera or even some other 
movie. Why? Because he couldn't be sure, uniquely, 
which movie she had in mind that he knew about. 
Bob must know not only about Monkey Business, but 
also that Ann knows about Monkey Business. At 
least, this is something Ann must try to assure 
herself of. This leads directly to the next 
57 
condition: 
(3) Ann knows that Bob knows that Ann knows 
about R. 
With condition (3) we must surely have strong 
enough conditions for the success of Ann's definite 
reference. But that isn't so, as we can show in 
still another variation on the original scenario: 
Version 4: On Wednesday night Ann goes to see 
Monkey Business, and there she sees Bob. As 
she walks down the aisle, she notices that 
he sees her, but as she is about to wave he 
turns and moves to another part of the 
theater. So she does not believe that he 
realizes that she has seen him. The next 
morning Ann meets Bob and asks, "what did 
you think of the movie?" 
This version satisfies conditions (i), (2), and (3). 
Ann knows about Monkey Business; she knows that 
Bob knows about it; and she knows that he knows 
that she knows about it. But Ann doesn't believe 
that he knows that she knows that he knows about 
it. This piece of negative knowledge should be 
enough to keep Ann from using her definite ref- 
erence. What if Bob had gone to A Day at the 
Races and A Night at the Opera too?, she should 
ask herself. He might think that while he is sure 
she didn't see him at Monkey Business, she might 
have seen him at one of the other two. If so, she 
might be referring to one of the other two. He 
couldn't be sure. According to Ann's reasoning, 
therefore, she must assure herself of something 
more--that Bob realizes that Ann realizes that he 
had been to see Monkey Business. That is, she must 
satisfy the following condition too: 
(4) Ann knows that Bob knows that Ann knows 
that Bob knows about R. 
With condition (4) it looks as if we have gone 
far enough (see Kempson, 1975, p. 165; Stalnaker, 
1977, p. 137), but can we be sure? Only if we can- 
not dream up another variation that satisfies con- 
ditions (i) through (4) but still doesn't work. 
Indeed, with a little difficulty, we can: 
Version 5: On Wednesday night Ann goes to see 
Monkey Business and there she sees Bob and 
Charles. Because she sits down a few rows 
in front of them, she believes that they 
see her there, but because she doesn't 
turn around while they are there, she be- 
lieves that they don't realize that she has 
realized that they have seen her there. On 
the way home, however, she meets Charles, 
who tells her that Bob did realize that she 
had seen them there, but because she hadn't 
waved at them, Bob was certain that she 
didn't realize that they had seen her 
notice that they were there too. The next 
morning Ann meets Bob and asks him, "What 
did you think of the movie?" 
Complicated as this version is, we realize that 
Ann in good conscience shouldn't have made this 
definite reference. Although conditions (i) 
through (4) are all satisfied, Ann should have con- 
sidered this possible reasoning on Bob's part. 
What if Bob had seen A Day at the Races and 
Night at the Opera too. He might think that she 
had seen him at, say, A Day at the Races and that 
she thought he had seen her there too, even though 
he hadn't. He would then have reason to think she 
was referring to A Day at the Races, since to have 
referred to Monkey Business she would have been 
sure that he knew that she knew that he knew that 
she was there (Bob's equivalent to condition (4)). 
So despite all of the conditions she has already 
assured herself of, she must add one more: 
(5) Ann knows that Bob knows that Ann knows 
that Bob knows that Ann knows about R. 
Is condition (5) enough? Hardly. What these 
versions show is that there is a way in principle 
of demonstrating that the last piece of iterated 
knowledge is insufficient. The method is this. 
Corresponding to Ann's condition (i) is an analo- 
gous condition that Bob must assure himself of if 
he is to uniquely identify the referent for Ann's 
definite reference, and it is this: 
(i') Bob knows about R. 
For Ann to be sure that her reference goes through, 
she must put herself in Bob's shoes, reason as Bob 
would, and make sure that he would identify the 
intended referent uniquely. What we did in con- 
structing version 2 was create a scenario in which 
(1) and (i') held, but Ann couldn't know that (i') 
held. This led us to add condition (2), Ann knows 
that Bob knows about R, the equivalent of Ann 
knows that (i'). But just as Ann needs to assure 
herself of (2), Bob needs to assure himself of 
(2'): 
(2') Bob knows that Ann knows about R. 
But then (2') is something else Ann needs to know, 
as we showed in creating version 3 of our scenario, 
and this led to condition (3). Corresponding to 
(3), however, is Bob's (3'), which we used in creat- 
ing version 4. In principle, we could use this 
procedure to construct countermanding versions ad 
infinitum. 
The successive versions and the conditions they 
give rise to eventually become absurdly complicate~ 
but they do bring out a general point. In princi- 
ple, one must satisfy oneself of an infinite num- 
ber of conditions either to make or to interpret 
a definite reference. Hence the mutual knowledge 
paradox. If each condition takes a finite amount 
of time to check, no matter how small, and if these 
checks cannot all be made in parallel, then making 
or interpreting a definite reference like the 
movie should take an infinite amount of time. 
Mutual Knowledge 
In common parlance, "shared knowledge" has sev- 
eral definitions. Ask your aunt what it means for 
the two of you to share knowledge that the mayor 
is an embezzler, andshe would probably say, "It 
means that you know he is an embezzler, and that I 
do too." If we express the proposition that the 
mayor is an embezzler as ~, then the first defini- 
tion of shared knowledge comes out like this: 
58 
A and B share I knowledge that p =def. heuristics. 
(1) A knows that p. 
(i') B knows that p. 
However, she might give you a more complicated 
answer: "It means that both of us know that he is 
an embezzler, and furthermore, I know that you know 
he is, and you know that I know he is." This leads 
us to a second definition of shared knowledge: 
A and B share 2 knowledge that p =def. 
(1) A knows that p. 
(l') B knows that p. 
(2) A knows that B knows that p. 
(2') B knows that A knows that p. 
Indeed, we can define a series of types of "shared" 
knowledge merely by extending the list of state- 
ments. None of these finite definitions, of 
course, describes the "shared" knowledge required 
of Ann and Bob in her reference to Monkey Business. 
For that we need something more. 
What is required, apparently, is the technical 
notion of mutual knowledge. It has been defined 
and exploited by Lewis (1969) and Schiffer (1972) 
for dealing with close cousins of the problem we 
have raised here. Mutual knowledge is Schiffer's 
term, while Lewis' term for the same thing is com- 
mon knowledge. We have opted for Schiffer's term 
since it seems more transparent and less open to 
misinterpretation. In any case, mutual knowledge 
is defined as follows: 
A and B mutually know that p =def. 
(i) A knows that p. 
(l') B knows that p. 
(2) A knows that B knows that p. 
(2') B knows that A knows that p. 
(3) A knows that B knows that A knows that p. 
(3') B knows that A knows that B knows that p. 
et cetera ad infinitum. 
Heuristics for Assessing Mutual Knowledge 
So far two conclusions seem firm. First, defi- 
nite reference requires a certain amount of mutual 
knowledge. Other simpler notions of "shared" 
knowledge will not do. Second, it is unthinkable 
that speakers and listeners assess mutual knowledge 
by working serially, statement by statement, 
through the infinity of statements that make up 
mutual knowledge. But they surely assess it some- 
how, as the first conclusion seems to require. The 
inevitable conclusion is that they use some sort of 
heuristics. We will consider two families of such 
heuristics--truncation heuristics and co-presence 
Truncation Heuristics 
The stickler in assessing mutual knowledge 
statements is that there is an infinity of such 
statements, and that is too many to check. What if 
people checked only a few of them--like the first 
four? The task could then be carried out in a fin- 
ite, even short, amount of time. There would be 
errors, of course, but they would probably be 
neither very serious nor very frequent. If Ann 
has verified the statement (4), Ann knows that Bob 
knows that Ann knows that Bob knows that p, it is 
extremely likely, on actuarial grounds, that the 
higher order statements would check out too. And 
when she does make an error, Bob will often look 
puzzled or ask for clarification, which will allow 
her to repair her reference. Indeed, repairs are 
quite frequent in spontaneous speech as if speakers 
might be doing just that. So people could assess 
only a truncated part of mutual knowledge. Heuris- 
tics of this kind will be called truncation heuris- 
tics. 
Are these heuristics plausible as the way 
people normally assess mutual knowledge? We 
believe not. Our doubts lie in two areas. First, 
it is not easy to deal with statements as compli- 
cated as (4). It is implausible that people check 
these statements per se. Second, the evidence 
needed to verify such statements anyway suggests a 
radically different family of heuristics. 
In version 4 of our movie scenario, Ann didn't 
believe that Bob knew that she knew that he knew 
about Monkey Business, a violation of knowledge 
statement (4). Version 4 is complicated. Not only 
did we have a hard time creating it, but people 
have a hard time grasping it, for it is difficult 
to keep track of who knows what. Statements like 
(4) are difficult not because of their syntactic 
form, but because they describe reciprocal rela- 
tions between two people. Whereas John Dean knew 
that Nixon knew that Haldeman knew that Magruder 
knew that McCord had burgled O'Brien's office is 
fairly comprehensible, John Dean knew that Nixon 
knew that John Dean knew that Nixon knew that 
McCord had burgled O'Brien's office is not. 
Although when we need to we can figure out fourth 
order reciprocal relations--not just the statements 
themselves, it seems highly implausible that we do 
so routinely. 
But what counts as evidence for the truth of 
statements like (i), (2), (3), and (4)? Take 
statement (3), Ann knows that Bob knows that Ann 
knows about R. Obviously, Ann won't have this 
statement per se already stored in memory. She 
doesn!t go through life creating statements like 
this for every object she or anyone else might want 
to refer to. Rather, what she needs to verify (3) 
is a piece of evidence from which she can deduc- 
tively or inductively infer it. Imagine that she 
and Bob had gone to Monkey Business together. It 
is hard to think of better evidence than this that 
she could appeal to for the truth of (3). Of 
course, the inductive rules by which she infers (3) 
from this evidence need to be spelled out, but 
that doesn't sound impossible. 
59 
The fact that Ann and Bob saw the movie together, 
however, is more useful evidence even than that. 
It is also about the best evidence we could imagine 
for the truth of (1), and of (2), and of (4), and 
so on ad infinitum. It is a piece of evidence 
that allows Ann, in one quick jump, to be sure of 
the truth of all the statements. Why, then, would 
she want to check the statements one by one--even 
a truncated list of them? She would be better off 
looking for that single piece of evidence that 
could in principle confirm them all. Indeed, that 
is the foundation assumption of the next family of 
heuristics we will take up, the co-presence heuris- 
tics. 
Consider the following strategy. When people 
make or interpret a definite reference, they try 
to assure themselves of mutual knowledge of the 
referent by searching for evidence of what we will 
call triple co-presence. This is evidence of a 
particular event in which the speaker, listener, 
and referent are "co-present," i.e., are "present" 
simultaneously, as when Ann, Bob, and Monkey Busi- 
ness are openly "present" together on Wednesday 
night. Strategies like this will be called co- 
presence heuristics. To see how they are reason- 
able, we will look at first principles. 
When Lewis and Schiffer hit on the notion of 
mutual knowledge, both recognized the need for a 
finite means of handling the infinity of state- 
ments. Their solutions were essentially the same. 
If A and B make certain assumptions about each 
other's rationality, they can use certain kinds of 
evidence, or states of affairs, to infer that each 
one of the infinite number of statements in mutual 
knowledge is true. But how? We get some hints 
from a concrete illustration of mutual knowledge 
devised by Schiffer. 
The scene: Ann and Bob are sitting across a 
table from each other, and there is a single 
candle between them. Both are looking at the 
candle, and both see the other looking at it too. 
The proposition ~ is that there is a candle on the 
table. Consider the scene from Ann's point of 
view. Clearly, she has direct evidence for the 
truth of (i): 
(i) Ann knows that p. 
But she also sees that Bob has his eyes open and 
is looking simultaneously at her and the candle. 
That is, she has evidence that she and Bob are 
looking at each other and the candle simultaneously. 
We will call this the simultaneity assumption. 
Indeed, she assumes that he is not only looking at 
her and the candle, but also attending to them. 
We will call this the attention assumption. 
Finally, she assumes that Bob is normal and in her 
shoes he would be drawing the same conclusions 
she is. We will call this the rationality assump- 
tion. But if Bob is attending to the candle and 
is rational, he has evidence for (i'): 
(i') Bob knows that p. 
This, however, is Ann's conclusion, and so she has 
evidence for (2): 
(2) ANn knows that Bob knows that p. 
But if Bob is rational, he will be drawing the 
inference that corresponds to hers--his equivalent 
of (2)--namely (2'): 
(2') Bob knows that Ann knows that p. 
Once again, this is Ann's conclusion, and so she 
has evidence for (3): 
(3) Ann knows that Bob knows that Ann knows 
that p. 
In like fashion, Ann would be justified in iterat- 
ing this process through the remaining knowledge 
statements (4) through infinity, and Bob would be 
justified in doing the same for his. 
So Ann has reason to b~lieve that she and Bob 
mutually know that there is a candle on the table. 
First, there is the "direct" evidence. She 
directly perceives that there is a candle o~ the 
table and that Bob is simultaneously looking at 
both her and the candle. Second, there are her 
assumptions about the situation. She assumes that 
Bob is consciously attending to her and the candle, 
that he is doing so at the same time she is, and 
that he is rational. The upshot is that she has 
no reason to believe that she couldn't confirm the 
knowledge statements as far down the list as she 
wanted to go. She is therefore justified in claim- 
ing mutual knowledge. Indeed, since nothing she 
doesn't know herself can be mutual knowledge, and 
since she can assume Bob is chronically rational, 
all she needs to do normally is search for evidence 
Of her and Bob simultaneously attending to each 
other and the candle on the table. With this we 
have the essence of the co-presence heuristics: 
To assess mutual knowledge, people search for evi- 
dence of triple co-presence--an event in which A 
and B are simultaneously attending to each other 
noting the same evidence for p. In equation form: 
Co-presence + Assumptions = Mutual knowledge 
The co-presence heuristics both solve the 
mutual knowledge paradox and make intuitive sense. 
When we assure ourselves of mutual knowledge, it is 
unlikely that we check for a series of pieces of 
evidence, even as few as the truncation heuristics 
might let us get away with. More likely, we check 
for a single piece of evidence of just the right 
kind. The candle example suggests that what we 
check for is evidence of triple co-presence. 
Varieties of Triple Co-presence 
There are many different kinds of evidence 
people may use for the triple co-presence of the 
speaker, listener, and referent. Some of these 
constitute strong evidence for triple co-presence, 
and others constitute weak evidence. That is, 
some kinds rightly give people a lot of confidence 
that the referent is mutually known, whereas other 
kinds do not. As reflected in our equation, there 
is a trade-off between the evidence and the assump- 
tions. The stronger the evidence is, the fewer 
assumptions are needed to infer mutual knowledge. 
Conversely, the fewer assumptions that are needed, 
the stronger the evidence is considered to be. The 
strongest evidence requires the fewest, or weakest, 
assumptions. 
60 
The cornerstone of our argument is this. The 
prototypical kind of evidence for mutual knowledge 
is physical co-presence, very much as illustrated 
in Schiffer's candle example. It is the strongest 
possible evidence, the one requiring the fewest 
auxiliary assumptions, and all other kinds are 
weaker in one way or another. What follows is a 
tentative classification of these varieties of 
triple co-presence. 
i° Physical co-presence. Ann, Bob, and the 
candle are an example par excellence of physical 
co-presence. Not only are the three of them phys- 
ically present together, but Ann can readily 
assume that Bob is attending to this fact, is doing 
so at the same time she is, and is rational. All 
three auxiliary assumptions are necessary. If she 
believed Bob was catatonic, or hypnotized the 
right way, or very near-sighted, for example, she 
wouldn't want to assume physical co-presence. 
Once Ann has assured herself of the direct evidence 
and these assumptions, she is warranted in infer- 
ring mutual knowledge of the candle and can refer 
to it as the candle. 
There are two distinct types of physical co- 
presence. Ann may refer to the candle while it is 
still physically co-present with them, as in The 
candle is romantic~ isn't it? Or she may refer to 
the candle some time after it has been co-present 
with them, as in The candle was romantic~ wasn't 
it? These two types could be called immediate 
and delayed physical co-presence. The first kind, 
on the face of it, is the stronger evidence. When 
physical co-presence is synchronous with the def- 
inite reference, Ann can be sure that she and Bob 
mutually know about the candle at the time she is 
referring to it. She doesn't have to count on 
Bob's remembering the past incident of physical 
co-presence, as she does in the delayed kind. 
The assumptions Ann would need in order to in- 
fer mutual knowledge from immediate physical co- 
presence are these: simultaneity, attention, and 
rationality. She would need an additional one for 
the delayed case: simultaneity, attention, 
rationality, and memory. Simultaneity, attention, 
and rationality refer to the assumptions we have 
described earlier. Memory refers to the addi- 
tional assumption for delayed physical co- 
presence: Ann must assume that Bob can and will 
recall the earlier incident of their physical co- 
presence. So far so good. The stronger the evi- 
dence, the fewer assumptions Ann needs in order to 
make her definite reference, l~ediate physical 
co-presence has one fewer requirement than delayed 
physical co-presence. 
2. Linguistic co-presence. Many things we 
refer to have never been physically co-present. 
They are often things we or someone else has men- 
tioned in conversation. Imagine Ann saying to Bob 
I bought a candle yesterday. Her utterance of 
candle is a locutionary act that posits for Bob the 
existence of a particular candle in the real world. 
If Bob hears and understands a candle correctly, 
he knows about the candle's existence at the very 
same time as she posits it. It is as if Ann has 
placed the candle on the stage before the two of 
them so that it would be physically co-present. 
So when Ann utters a candle and Bob simultaneously 
understands it, the two of them can be said to be 
in the linguistic co-presence of the candle. Once 
Ann has established this, of course, she can make 
a definite reference to it, as in The candle cost 
me plenty. 
Linguistic co-presence is weaker evidence for 
mutual knowledge than physical co-presence. Seeing 
is believing--hearing about something isn't. To 
begin with, linguistic co-presence requires the 
assumptions of simultaneity, attention, and ration- 
l ality. Ann and Bob must be attending to Ann's utterance of a candle simultaneously, and both must be ration--a~. ~d like delayed physical co- 
presence, linguistic co-presence requires memory. 
For Ann to refer to the candle, she has to count 
on Bob's recalling the earlier incident of linguis- 
'tic co-presence with her uttering of a candle. 
But there is an additional assumption we will call 
understandability. Ann must assume that Bob will 
penetrate her indefinite reference, a candle, and 
understand that she is sincerely positing the 
candle's existence. She must assume that Bob 
understands her, and he must assume that she be- 
lieves he does. 
3. Indirect co-presence. Imagine Ann saying 
to Bob I bought a candle yesterday; the wick is 
made of cotton. In uttering a candle Ann has 
established the linguistic co-presence of him, her, 
and the candle, but not of him, her, and the wick. 
How, then, can she refer to the wick? She has to 
assume that when Bob accepts the existence of the 
candle, he will also accept the existence of its 
wick. This way, by uttering a candle, Ann has 
established what we will call the indirect co- 
presence of her, Bob, and the wick. 
The inferences required in indirect co-presence 
are often much stronger than those needed for wick 
(see Clark, 1977; Clark & Haviland, 1977). Ann 
can refer to something that is only likely to 
be associated with a thing she has already estab- 
lished, or even only possibly associated with it. 
She can tell Bob: I bought a candle yesterday~ 
but the wrapper was torn; or I bought a candle yes- 
terday, and the bayberry smelled great. Candles 
don't necessarily come in wrappers nor are they 
often made of bayberries, yet these are parts she 
expects Bob to infer on the basis of her definite 
references to them. So what is established may be 
only the likelihood or possibility of a thing being 
co-present with the speaker and listener. Its 
certain existence is established only with the def- 
inite reference itself. 
Indirect co-presence is parasitic. It has to 
be established via some other type of co-presence-- 
for example, physical or linguistic co-presence. 
Before Ann can say The price was $3 of a candle, 
she must already have established the candle's 
co-presence. She and Bob could be looking at it, 
for physical co-presence, or she could have just 
mentioned it, for linguistic co-presence. For the 
moment we will assume that indirect co-presence 
is always established via either physical or lin- 
guistic co-presence. 
There is both a strong and a weak case of in- 
direct co-presence. Instead of saying The price 
was $3, Ann could have said The price of the candle 
61 
was $3, providing a much more certain reference. 
She would have made it explicit that the price 
referred to is that of the candle and not of 
something else. Bob would then have had no trouble 
inferring that there was one and only one price 
associated with the candle. They both could then 
assume that they mutually knew about the price. 
This case may be so direct that it ought to be 
placed in a separate category. For now we will 
treat it as a very strong kind of indirect co- 
presence. 
To infer mutual knowledge from indirect co- 
presence, Ann and Bob need all the assumptions 
of physical or linguistic co-presence, whichever 
is the parasite's host, plus one we will call 
associativity. They have to assume that each other 
is capable of entertaining the certainty, likeli- 
hood, or possibility of a particular part or role 
being associated with the thing whose co-presence 
has already been established. The hierarchy still 
works as expected. Indirect co-presence, because 
of its added assumption, is weaker evidence for 
mutual knowledge than either physical or linguistic 
co-presence. 
4. Cultural co-presence. Even when Ann is not 
acquainted with Bob, she can assume there are par- 
ticulars the two of them mutually know. The basic 
idea is that there are things everyone in a cul- 
tureknows about. She reads newspapers, and so 
does everyone else in her culture. So Bob and she 
can mutually assume that they both read newspapers. 
Ann can then take the fact that John Dean, Michael 
Doonesbury, and Billy Jean King have been promi- 
nently mentioned in the newspaper as good evidence 
that she and Bob mutually know about these people. 
This is an instance of what we will call cultural 
co-presence. Certain particulars are assumed to 
be universally known in a cultural milieu--they 
are culturally co-present for everyone in it-- 
and that is taken as evidence that everyone in the 
milieu knows about them. 
The trick, of course, is to judge cultural mil- 
ieus. Ann may think that she and Bob mutually 
realize that they are both high school graduates, 
or drug dealers, or nineteenth century history 
buffs, or New Yorkers, or telephone operators, or 
some combination of these, and her assumptions 
about cultural co-presence will change accordingly. 
If her assessments are accurate, her definite 
reference is likely to succeed, and if not, it 
isn't. 
Cultural co-presence doesn't appear to belong 
to the same hierarchy as the previous three types 
of co-presence. For one thing, it is relatively 
permanent, whereas the other three are relatively 
transitory. Culturally known particulars take 
time to become familiar and to lose familiarity. 
Teddy Roosevelt is familiar to Americans today, 
just as he was 75 years ago. Particulars known by 
physical, linguistic, or indirect co-presence have 
only fleeting familiarity and then only to specific 
pairs of people. Mutual knowledge about these 
particulars is easily established, but also easily 
lost. For another thing, cultural co-presence is 
parasitic on other forms of mutual knowledge. For 
ANn to establish that she and Bob mutually know 
that they belong to the same cultural subgroup, 
she must find evidence of triple co-presence of 
that fact. She might establish it, for example, 
through linguistic co-presence, as in, What do you 
know--we're both New Yorkers. 
To infer mutual knowledge from cultural co- 
presence, therefore, people need assumptions that 
are not comparable with those of the other three 
types. Take Ann's reference to Hoover Tower in a 
conversation with Bob. First, she must assume 
that she and Bob mutually know that they belong 
to a particular cultural subgroup, say Stanford 
University students. We will call this assumption 
cultural membership. How Ann justifies this 
assumption, however, will not be simple. Like 
other types of mutual knowledge, it must be based 
on evidence of some kind of co-presence. Second, 
she must assume that virtually everyone in this 
cultural milieu takes it for granted that they all 
know about Hoover Tower. We will call this assump- 
tion universality of knowledge. The paucity of 
these assumptions should not fool us into thinking 
that cultural co-presence is strong, for they hide 
a tangle of complex justifications based on other 
pieces of evidence and other assumptions. It is 
best to treat cultural co-presence as incommensu- 
rate with the other three. 
With cultural co-presence we have come to the 
last of the major kinds of co-presence. Not every 
kind of evidence for mutual knowledge, however, 
can be neatly classified as one of these four 
types. Some appear to require a complex comhina- 
tion of them, and not surprisingly, they provide 
intuitively weaker evidence for mutual knowledge. 
Reference Diaries 
If people assess mutual knowledge via triple 
co-presence, they must have a memory full of facts 
about triple co-presence. What do these facts 
look like? How are they represented? How are 
they assessed? If mutual knowledge is critical 
to definite reference--as we have suggested--then 
questions like these ought to be central to any 
theory of speaking, listening, or memory. Indeed, 
the arguments we have offered lead to a rather pro- 
vocative conception of memory representation and 
memory search. It is provocative in that some of 
its critical properties are absent from most cur- 
rent models of comprehension and memory. 
Most investigators have assumed that in pro- 
cessing definite reference people search memory 
for the particulars actually referred to. Take 
Ann's reference to Monkey Business. On bearing 
this Bob would search memory for a referential 
index to the intended referent M0nkey Business. 
This index is a stand-in, so to speak, for the 
movie itself. Although the current models of 
comprehension differ in their specifics, virtually 
all of them assume this kind of search for the 
intended referent. That includes Anderson (1976), 
Clark and Haviland (1977), Kintsch (1974), Rumel- 
hart, Lindsay, and Norman (1972), Schank and 
Abelson (1977), and Winograd (1972), to name just 
a few. 
But if people use some kind of co-presence 
heuristics, then all of these models are incor- 
rect--or at least incomplete. The point is that 
62 
Bob cannot search memory for the referent alone. 
That would hardly guarantee that it was mutually 
known to him and Ann, as it must be for her refer- 
ence to be legitimate. Rather, he must search for 
an event that involves not only the referent but 
also Ann and him. That is, it must be an event of 
triple co-presence--of physical, linguistic, 
indirect, or cultural co-presence, or of some com- 
bination of these four types. In none of the cur- 
rent models Just mentioned does the listener search 
for such an event. 
Previous models of comprehension have treated 
search through memory as if it were a search 
through a telephone book. In a definite reference 
like the man in the red shirt we are told the name 
and address of the individual we want to get hold 
of. Our task is to search the telephone book for 
his number, our direct connection to him, his ref- 
erential index. With the co-presence heuristics, 
memory must be more like a diary, more like the 
personal log Nixon kept of everything he did and 
experienced during his years at the White House. 
As before, in the man in the red shirt we are told 
the name and address of the individual we want to 
get hold of. But to find him we must search our 
diary for an entry that provides evidence of the 
co-presence of the speaker (say, Gertrude), us, 
and an individual of that description. The diary 
entry must show that we were physically or linguis- 
tically co-present, or that we were co-present in 
some other sense. That is, we must search in every 
case for an event. This is far more complicated 
than searching the telephone book, with or without 
yellow pages, for the right number. 
The diary, of course, cannot be used alone. We 
also need histories and atlases to refer to John 
Dean, the Second World War, the decline and fall 
of the Roman Empire, and China, particulars that 
are culturally co-present. And for indirect co- 
presence we will also need texts on science, med- 
icine, engineering, and law. To know that candles 
have wicks we need to look up facts about the 
engineering of candles. 
What we need, in summary, is a diary of the sig- 
nificant events in our own personal experience, 
supplemented by cultural histories and atlases 
for cultural co-presence and by various reference 
texts for indirect co-presence. Such a diary con- 
tains a record of the events we will need for 
assessing co-presence. Anything less than a diary 
will be too little. 
Footnote 
iThis paper is an abbreviated version of 
"Definite reference and mutual knowledge," pre- 
sented at the Sloan Workshop on Computational 
Aspects of Linguistic Structure and Discourse 
Setting, University of Pennsylvania, May 1978. 
We thank Eve V. Clark for her helpful comments on 
the manuscript. 
References 
Anderson, J.R. Language, memory and thought. 
Hillsdale, N.J.: Erlbaum, 1976. 
Clark, H. H., & Haviland, S. E. Comprehension 
and the given-new contract. In R. O. Freedle 
(Ed.), Discourse production and comprehension. 
Norwood, N.J.: Ablex Publishing, 1977, pp. 1- 
40. 
Clark, H. H. Inferences in comprehension. In D. 
LaBerge & S. J. Samuels (Eds.), Basic processes 
in reading: Perception and comprehension. 
Hillsdale, N.J.: Erlbaum, 1977. 
Clark, H. H., & Marshall, C. Definite reference 
and mutual knowledge. Paper presented at the 
Sloan Workshop on Computational Aspects of 
Linguistic Structure and Discourse Setting, 
University of Pennsylvania, May 1978. 
Kempson, R. M. Presupposition and the delimitation 
of semantics. Cambridge: Cambridge University 
Press, 1975. 
Kintsch, W. The representation of meaning in mem- 
ory. Hillsdale, N.J.: Erlbaum, 1974. 
Lewis, D. K. Convention: A philosophical study. 
Cambridge, Mass.: Harvard University Press, 
1969. 
Rumelhart, D. E., Lindsay, P. H., & Norman, D. A. 
A process model for long-term memory. In E. 
Tulving & W. Donaldson (Eds.), Organization 
of memory. New York: Academic Press, 1972, 
pp. 197-246. 
Schank, R., & Abelson, R. Scripts, plans, goals 
and understanding: An inquiry into human knowl- 
edge structures. Hillsdale, N.J.: Erlbaum, 
1977. 
Schiffer, S. R. Meaning. Oxford: Oxford Univer- 
sity Press, 1972. 
Stalnaker, R. Pragmatic presuppositions. In 
A. Rogers, B. Wall, and J. P. Murphy (Eds.), 
Proceedings of the Texas conference on per~ 
formatives, presuppositions, and implicatures. 
Arlington, VA: Center for Applied Linguistics, 
1977, pp. 135-148. 
Winograd, T. Understanding natural language. 
New York: Academic Press, 1972. 
63 
