A Framework for Comparing Language Experiences 
(with particular emphasis on: 
The Effect of Audience on Discourse Models) 
Andee Rubin 
Bolt Beranek and Newman Inc. 
Cambridge, Massachusetts 
Those of us who have been involved in the 
study of language have often thrown up our 
hands in dismay at the complexity of the 
problem (at least I have, almost daily) and 
tried somewhat desperately to find some facet 
of the many-faced gem we confront which 
appears manageable. This desire to focus - to 
train the flashlights we use to illuminate the 
problem on a well-circumscribed area - led, 
for example, to the early transformational 
grammarians' separation of syntax from the 
rest of language. Many researchers have since 
discarded that particular focus and attempted 
to integrate syntax, semantics and pragmatics 
in a single theory. Yet, even they end up 
focusing on a particular kind of linguistic 
interaction. Many study oral conversations, 
some look at computer-person dialogues, some 
study newspaper articles. These limitations 
in scope are often not explicit, but are 
reflected in the examples they choose and 
discuss. Very few people study more than one 
communicative situation and even if they do, 
they do not usually analyze the similarities 
and differences among them. 
Just as the early transformational focus 
on syntax resulted in a model which missed 
many crucial insights about language, so does 
our current research risk formulating 
incomplete and even inaccurate models by 
focusing on certain communicative situations 
without adequate insight into their 
relationship to others. My own focus in an 
attempt to point out what such a narrow view 
might miss, and to provide a framework in 
which to examine a wide variety of language 
experiences and discover what effect the 
differences among them have on theories of 
language. This paper first presents the 
framework, then examines further one dimension 
of language experience -- the audience for an 
utterance -- as an example of the kind of 
considerations such an analysis suggests. 
I. Building the Space of Language Experiences 
Consider first two language experiences 
commonly studied by present-day investigators: 
face-to-face oral conversations and 
computer-person dialogues. These two 
situations differ in at least two ways: the 
modality of the interaction (current 
computer-person dialogues are written) and the 
lack of possibility of communicating with 
extra-linguistic devices such as gestures and 
facial expressions. The kinds of questions we 
want to be able to address are: 
What are the effects of these two distinctions 
on the language used in each of these 
situations? What are the effects on the 
models of language use which we thereby 
formulate? 
In order to capture these kinds of 
differences in a way which will enable us to 
approach these questions, I have used the 
metaphor of a multi-dimensional space. Each 
language experience lies at a point in the 
space defined by its position along several 
dimensions of the linguistic medium. The 
medium of a language experience is defined in 
contrast to its message: in as much as they 
can be separated, the message is what is 
communicated, while the medium is how it is 
communicated. Further, the medium here is 
expressed in experiential terms and does not 
represent simply the vehicle for the message; 
for example, the contrast is made between 
being in a conversation and watching a play, 
rather than between a conversation and a play. 
(I) 
Consider as a starting point in building 
the space the following message: an 
invitation and directions to a party. One 
common linguistic situation in which this 
might be communicated is a face-to-face oral 
conversation. Conversations, however, do not 
need to be oral; it is quite possible to 
maintain all the aspects of a conversation 
while writing it down by, for example, passing 
notes. These two language experiences form a 
"minimal pair"; that is, they differ along 
(I) I have omitted from this discussion any 
consideration of the message communicated in a 
language experience. In Rubin (1978) I 
identify three message-related dimensions 
structure, function and touic - and discuss 
their interactions with the medium-related 
dimensions introduced here. 
133 
only one dimension. We might represent the 
situation graphically by labelling the line 
connecting the two experiences with the 
dimension along which they differ. The 
relevant dimension in this diagram is mod~li~ 
i.e. whether a language experience is oral or 
written. (Modality will be further dissected 
below.) 
HAVING A GARDEN-VARIETY 
CONVERSATION 
MODALITY 
Figure i. 
HAVING A CONVERSATION 
BY PASSING NOTES 
Let us now look at two other pairs of 
language experiences which illustrate another 
dimension. Consider communicating the same 
message over the telephone compared to using a 
tape cassette. In the first case, it is 
possible for the two participants to interact, 
for the listener to express confusion and ask 
for additional information, for the speaker to 
monitor the listener's reactions and provide a 
more complete explanation. In the cassette 
situation, the speaker must decide once and 
for all how to give the directions without the 
benefit of intermediate feedback; any feedback 
which might occur would happen after the 
listener had heard the tape all the way 
through and would be temporally removed from 
the time the speaker composed the tape. I 
have termed this dimension of language 
experience ~ntgrac~io~, as Figure 2 
illustrates. 
The other minimal pair in Figure 2, which 
also illustrates the interaction dimension, is 
communicating by letter versus communicating 
by a conversation over teletypes (while this 
is a somewhat unusual communicative setting, 
many of the people reading this paper have 
probably participated in it). Here again the 
crucial difference between the two is the 
possibility of feedback. In this particular 
task, for example, the speaker might want to 
ask of the listener, "Do you know the corner 
of Lewis and Fairview?" and base her further 
explanation on the response. Such an exchange 
would be impossible in the case of a letter. 
i TALKING ON THE TELEPHONE INTERACTION l LISTENING I TO A TAPE I 
CASSET~ 
h I CON~IERSING 
I OVER LINKED 
L TELETYPES 
~INTERACTION t 
Figure 2. 
READING A LETTER 
Notice now that we can connect the two minimal 
pairs in Figure 2 by lines labelled 
"modality". Reading a letter and listening to 
a cassette form a minimal pair which differ 
only in modality; the same is true for 
teletype and telephone conversations. The 
modality and interaction axes together form a 
plane in which we can place these four 
language experiences. 
l TALKING I ON THE TELEPHONE 
I MODALITY 
I CONVERSING OVER LINKED 
TELETYPES 
INTERACTION \[ TOcAsSETTELISTENINGA TAPE 
I MODALITY 
INTERACTION HEADING A LETTEH 
Figure 3. 
Other dimensions can similarly be added 
to this space, by first discovering a minimal 
pair which focuses on a particular dimension, 
then attempting to pinpoint each language 
experience already in the space on that 
dimension and finally filling in the holes 
which exist because of the added axis. As an 
example of one step in building the space, 
consider the dimension of gxtra-linguistic 
communication, that is, communication by 
gestures, facial expressions, etc. For the 
message we are considering, gestures would be 
particularly useful to indicate spatial 
features such as "right" and "left" and the 
relative location of objects and landmarks. 
None of the four media in Figure 3 admit this 
type of interaction, but for each of them it 
is possible to construct an experience which 
differs from it along only this new dimension. 
For example, garden-variety conversations 
differ from telephone conversations because 
they allow this extra dimension, and passing 
notes differs from conversing over a teletype 
link in the same way. We now see where the 
pair of language experiences illustrated in 
Figure I comes in and by adding two more nodes 
we get the following cube. 
MO~LITY i 
MODALITY 
Figure 4. 
134 
The language experiences in this subspace 
differ in their degrees of naturalness; 
writing on the blackboard in such a way as to 
allow extra-linguistic communication but 
prohibit interaction, for example, seems 
contrived. This awkwardness is due primarily 
to the fact that certain dimensions generally 
covary and that the independence implied by a 
dimensional analysis doesn't really hold. 
Section 3 below discusses these 
interdependencies in more detail. 
2. The Dimensions of Language Experience 
In a similar fashion, we can add more 
dimensions to the space of language 
experiences. I have so far identified eight 
separate dimensions along which the medium of 
a language experience may vary. The 
dimensions are at least semi-independent; my 
informal criterion for listing a dimension 
separately was the existence of some minimal 
pair of language experiences whose media 
differed only along that dimension. The 
medium-related dimensions of language 
experience are: modality, interaction, 
extra-linguistic communication, spatial 
commonality, temporal commonality, 
concreteness of referents, separability of 
characters and specificity of audience. 
Below, each dimension is explicated by means 
of the question one would have to ask about a 
language experience to correctly place it on 
that dimension and additional details about 
its substructure and ramifications are given. 
I. MODALITY - Is the message written or 
spoken? Even in this seemingly simple 
dimension are hidden at least two different 
characteristics which affect the communicative 
situation: prosody and permanence. I will 
briefly discuss these here, but a more 
extensive discussion of the components of 
modality may be found in Schallert, Kleiman & 
Rubin (1977) and Sticht, Beck, Hauke, Kleiman 
& James (1974) provide a review of the 
literature comparing auding (comprehension of 
oral language) and reading. 
An obvious difference between an oral 
utterance and a text is the availability of 
prosodic cues. Temporal characteristics of 
speech such as pauses and changes in speed 
provide clues for the chunking of words into 
larger constituents. In general, pauses and 
breaths occur at syntactic boundaries and a 
more quickly spoken set of words may indicate 
an appositive phrase or an aside which is not 
germane to the top-level structure of the 
sentence. We rely on stress in oral language 
as an indicator of such discourse organizing 
topics as given/new, contrast and focus, as 
well as to aid in the disambiguation of 
pronominal references. Intonation is often 
used as an indication of the illocutionary 
force of an utterance or to communicate 
affective qualities of language such as humor 
or sarcasm. 
While written text clearly lacks these 
properties, it has some compensating features. 
Punctuation and other textual devices provide 
a partial analogue of many prosodic features, 
including illocutionary force (. ? !), pauses 
(;), lists (, : ;), related statements (;) and 
contrast and emphasis (underlining and 
italicizing). A written message also provides 
the recipient with concretely indicated 
segments both on the lower levels of word and 
sentence and on the more abstract level of 
paragraph and section structure. 
The second major distinction included in 
modality is the permanence of written text in 
contrast to the transitory nature of oral 
language. This permanence makes possible 
various "good reading" techniques such as 
skimming ahead to look at chapter headings, 
re-reading an entire paragraph whose point 
became clear only at the last sentence, or 
just re-reading a sentence which was misparsed 
the first time around. In oral language 
situations, such heuristics for dealing with 
misunderstanding are often replaced by an 
appeal to another (independent) characteristic 
of language experiences - interaction. 
2. INTERACTION - Are the participants able to 
interact? In an interactive language 
experience, participants have the opportunity 
to indicate that they have not understood a 
previous utterance, that a pronominal or other 
reference is ambiguous or that they wish to 
change the topic. Keenan and Schieffelin 
(1976) in particular have represented the 
establishment of discourse topic as a dynamic 
process which includes input from all 
participants. This possibility for 
interaction means that there is less necessity 
for a participant to entertain and maintain a 
set of competing hypotheses about the meaning 
of some part of the message. 
3. EXTRA-LINGUISTIC COMMUNICATION - Can the 
participants communicate via extra-linguistic 
means which require visual or tactile 
interaction? (This communication may 
sometimes be one-way, as in the case of a 
lecturer speaking to a large class.) 
Gestures, facial expressions and even body 
positions are all powerful in their 
communicative potential. In situations where 
emotions or spatial attributes are being 
communicated, these extra-linguistic means may 
be especially relevant. Children's early 
language experiences are especially dependent 
on this aspect of communication; deLaguna 
(1927) describes one developmental thread in 
children's language use as "a progressive 
freeing of speech from dependence on the 
perceived conditions under which it is uttered 
and heard, and from the behavior which 
accompanies it." 
13.5 
4. SPATIAL COMMONALITY - Can the participants 
interpret spatial deictie words such as "here" 
and "there" with reference to their own 
location? One indication that situations in 
which this condition is not met are difficult 
is the well-known situation in which two 
people, having arranged over the telephone to 
meet "here", discover that they had two 
different places in mind. Because the 
listener had to interpret the speaker's use of 
"here" relative to the speaker's location, it 
was necessary for her to know where the 
speaker was; incorrect information in this 
situation can have serious consequences. 
Young children may actually interpret "here" 
and "there" relative to their own position, 
rather than the speaker's. (see Tanz (1976) 
for details) 
5. TEMPORAL COMMONALITY - Do the participants 
share a temporal context which allows for 
simple interpretation of temporal deictic 
terms such as "now", "today" and "last 
Sunday"? The correct interpretation of such 
words, as well as verb tense markers, requires 
the reader/listener to take the temporal point 
of view of the speaker/writer. 
6. CONCRETENESS OF REFERENTS - Are the 
objects and events referred to visually 
present for the participants? If an object or 
event is concrete, many of its details are 
immediately apparent to the reader/listener 
besides the ones linguistically described in 
the message. Reading or hearing about an 
• object ' which is not present often requires 
remembering a partial, incomplete description 
and then reformulating it as more information 
becomes available. Objects (or pictures) also 
provide an external "memory" for their 
existence and properties. 
7. SEPARABILITY OF CHARACTERS - Are the 
distinctions among different people's 
statements and points of view clearly 
indicated? In face-to-face conversations, 
such distinctions are obvious, as each person 
makes his own statements and each point of 
view has a physical "anchor". In reading a 
play, characters' lines are clearly marked, 
although there is no physical object to which 
to attach each character. In a book, the 
reader must parcel out comments, feelings and 
motivations to characters on the basis of more 
subtle clues: punctuation, paragraph structure 
and inferences based on some consistent model 
of each of the characters. 
8. SPECIFICITY OF AUDIENCE - How complete and 
specific is the speaker's model of the 
audience for her message? Two extremes which 
illustrate this dimension are garden- variety 
conversations in which the speaker and hearer 
know each other well and books, which are 
written for wide non-specific audiences. In 
the former case, references to shared 
knowledge are possible, such as "The man 
looked like Uncle Joe," while in the latter, 
such an attempt would surely miss a large 
portion of the audience. To make matters 
worse, often a writer (or speaker) does not 
know who the audience is likely to be and in 
the case of books which are several hundred 
years old, the intended audience differs in 
significant ways from current readers. 
Now that we have these eight dimensions, 
we can use them to generate new language 
experiences which begin to fill up the space. 
Watching a play, reading a book with pictures, 
viewing a movie with subtitles, reading a 
comic book - all these fit in the 
eight-dimensional space we have defined. (In 
Rubin (1978) I discuss quite a number of 
"intermediate" language experiences and show 
how they fit into a multi-dimensional space.) 
However, some areas of the space are only 
sparsely filled. These relatively empty 
sections are indications of interactions 
between dimensions; given a particular 
position on one dimension, the choices for 
certain other dimensions may be sharply 
constrained. Descriptions of some of these 
interactions follow. 
3. Interactions Among the Dimensions 
One fairly obvious interdependency is 
between spatial commonality and 
extra-linguistic communication. Since both 
rely primarily on the participants being in 
the same place, it is not surprising that most 
language experiences which exhibit the 
potential for extra-linguistic communication 
also allow participants straight-forward use 
of spatial deictic terms. (In fact, in Rubin 
(1978), I treat the two as a single 
dimension.) However, in a note left, for 
example, on the kitchen table, the writer may 
use "here" and "there", but cannot use 
gestures or facial expressions, so the two 
dimensions do not always co-occur. 
Extra-linguistic communication is also 
most commonly found in oral language 
situations. The situation with the most 
potential for combining extra-linguistic 
communication with a written message is that 
of two people passing notes in what amounts to 
a written conversation. Although it is 
theoretically possible for them to point and 
grimace, it would be difficult for them to 
coordinate these gestures with the words in 
the written text. 
Interaction and temporal commonality also 
appear closely linked. If the participants 
are not communicating in "real time" - that 
is, if the sending and receiving take place at 
different times - then it might even seen 
impossible for them to interact. However, if 
we allow the kind of attenuated interaction 
that takes place in, for example, an exchange 
of letters, we can maintain these dimensions 
as at least semi-independent. 
136 
Finally, we note that specificity of 
audience and interaction have an interesting 
relationship. Less well-defined audiences 
tend to occur in situations in which 
interaction is difficult, if not impossible. 
In lectures, for example, the speaker has only 
a vague idea of the audience's beliefs and 
interaction between them is limited. This 
covariance reflects two different facts. One 
is that in a large (and therefore 
poorly-specified) audience, interaction is 
restricted simply because of its size. The 
other is that interaction is one device by 
which speakers construct better models of 
their audiences; thus, a lacM of interaction 
would lead to less well-specified audiences. 
An obvious question at this point is: Why 
bother to separate dimensions that are so 
closely related? There are really two 
answers: the first methodological, the second 
historical. In terms of getting a clean model 
of the complex tangle of language experiences, 
it is better to postulate a large number of 
dimensions and specify how they interact than 
to identify only a small number but talk about 
subdimensions. Having a larger number of 
dimensions also inspires a wider range of 
language experiences when the process of 
filling in the space is carried out. Without 
the separation of temporal commonality and 
interaction, for example, we would have missed 
the subtle notion of temporally attenuated 
interaction. 
The historical explanation derives from 
the original motivation for this work, which 
was an attempt to assess the relevance of 
children's early language experiences to their 
learning to read. Even if the dimensions 
identified here interact significantly, each 
still represents a cognitive skill which a 
child must learn in making the transition from 
garden-variety conversations to reading a 
text. In this framework, interactions among 
the dimensions are interesting because they 
represent pairs of skills which the child may 
have to learn together, rather than being able 
to separate them and learn one at a time. 
Now that we have a notion where 
garden-variety dialogues fit into the 
framework of language experiences, it is 
possible to see what kinds of considerations 
we are liable to leave out if we focus only on 
conversations. While all of the dimensions 
identified above point out areas which deserve 
attention, I want to focus here on specificity 
of audience as an example of the ways our 
models must be stretched to account for the 
diversity of language experience. 
4. Limitations and Compensation in Language 
Experiences: Non-Specific Audiences 
Certain language experiences present 
problems for the participants, especially in 
comparison with garden-variety conversations, 
which have many communication-facilitating 
features. Lack of spatial commonality, for 
example, poses extra difficulty in the 
interpretation of certain deitic words, and 
the absence of non-verbal communication in 
telephone conversations makes expressing 
emotion especially hard. In some cases, an 
aspect of the medium itself provides 
compensation for the limitations. Written 
text, for example, partially compensates for 
its lack of prosodic cues to structure by its 
permanence, which allows the reader to make 
several attempts at parsing and understanding 
the words on the page. In other cases, we 
ourselves take into account the limitations of 
the medium by expressing our message 
differently. In talking on the phone, for 
example, we express our emotions more 
explicitly, rather than relying on facial 
expressions to communicate them in more subtle 
ways. 
An important facilitating aspect of most 
garden-variety conversations is that they take 
place between people who have fairly good 
models of one another and who share a large 
set of beliefs. (See Cohen (1977) and Clark & 
Marshall (1978) for details on shared 
beliefs.) The disappearance of this feature 
in other language experiences can cause 
difficulties which require special attention 
from both speaker and listener. To get a 
feeling for the effect of an audience, 
consider the task of explaining the difference 
between analog and digital computers. Talking 
to a technically unsophisticated person makes 
the task hard enough, but it would be even 
more difficult if the audience were a large 
number of people with widely varying technical 
backgrounds. When one is faced with 
communicating with a person about whom one 
knows very little or, worse yet, an audience 
made up of many people with different beliefs 
it becomes necessary to use several 
compensating techniques to ensure that the 
message gets across. I will describe below 
some of the heuristics both speaker/writers 
and listener/readers use to deal with complex 
and poorlyspecified audiences. (From here on, 
I will use the words "speaker" and "listener" 
to refer to speaker/writers and 
listener/readers, respectively.) 
5. Speakers' Heuristics for Complex Audiences 
The audience for an utterance may be 
poorly specified in two different ways: it may 
be a single person about whom the speaker has 
less than complete knowledge or a group of 
people, each of whom the speaker knows more or 
less well. The speaker's task is to construct 
an utterance which is comprehensible to those 
who perceive that they are part of the 
intended audience. The following are 
techniques by which speakers may accomplish 
this task. 
Identify the Audignce: In some cases, the 
speaker really wants to address her remarks to 
a single person, even though several people 
are physically present and "available" as 
audience. Straightforward techniques exist 
for identifying the audience in these 
situations: a speaker may simply look in the 
direction of the intended audience or address 
137 
her by name. In giving a technical talk to a 
large and varied audience, a speaker may 
analogously select a subset of those present 
as the audience for a particular remark by 
using a phrase like "for you linguists in the 
audience". A more subtle and interesting 
method for accomplishing the same goal is to 
include in the remark a reference which only 
some of the audience understands, thereby 
clueing the others in to the fact that they 
may not get anything out of the utterance. At 
a recent conference attended by linguists, 
computer scientists and psychologists, a 
speaker, in answering a question from a 
computer scientist, resorted to some technical 
language (what he actually said was, "It's EQ, 
not EQUAL"), even though he knew two-thirds of 
the audience would be lost. Afterwards, he 
remarked that he had also realized that it was 
precisely the computer scientists who were 
confused about the point he was trying to 
make, so the remark was doubly appropriate: it 
selected exactly the audience who needed to 
comprehend it. 
Pla~ It Safa: Qover the Audience: If he 
realizes that the audience consists of two or 
more definable subgroups, a speaker may choose 
to include several descriptions of the same 
topic, one for each set of people. In 
addressing an audience made up of computer 
scientists and psychologists, a speaker might 
refer to the same concept by two different 
terms, e.g. "cache memory" for the computer 
scientists and "working memory" for the 
psychologists. In this case, most members of 
the audience understand only one of the two 
descriptions so both are necessary. A 
slightly different situation exists when a 
speaker makes a statement such as, "He had 
eyes just like Paul Newman...deep, dark blue." 
Here the elaboration may be seen as a comment 
to those in the audience who don't know what 
Paul Newman's eyes look like. Those who do 
must realize that, in some sense, the 
elaboration was not directed at them since it 
was planned for listeners with different 
knowledge. A third example of the "play it 
safe" strategy is this conversation which took 
place in front of an audience: 
B: Jerry has been studying the same 
thing. 
M: That's right...that's Jerry Fodor...I 
read his paper. 
Had B and M been conversing in private, 
the explanation that the Jerry being referred 
to was Jerry Fodor would have been 
unnecessary. However, M was aware that some 
people in the audience might not be able to 
figure out who was being referred to, so he 
played it safe and made the reference clear. 
Embrace Ambiguity: Sometimes, the speaker is 
aware that an utterance may be interpreted in 
two different ways, but decides that the 
ambiguity is acceptable or even desirable. An 
example from a personal letter: "The weather 
has been beautiful...perfect for riding a 
motorcycle in the country." What the writer 
had in mind here was a particular time she and 
the addressee had taken a motorcycle ride; she 
wanted to allude to that event. If, however, 
he didn't remember the ride, the sentence 
still communicated a coherent, if less 
specific, message. The writer actually 
considered both of these possibilities and 
decided that either reading of the sentence 
was acceptable. 
The following sentence taken from an 
advertising brochure shows a different kind of 
ambiguity: "The cane seats of a Mad River 
canoe provide excellent ventilation and 
drainage." The ad will be read both by people 
who already know that Mad River canoes have 
cane seats and by those for whom this is a new 
fact. For the latter group, the word "cane" 
obviously conveys new information; for the 
former group, perhaps, it focuses on that 
aspect of the seat which is relevant to the 
properties discussed. (I use the word 
"focuses" here in an informal sense, but it is 
similar to its use in Grosz (1977) and Sidner 
(1978). Again, the ambiguity exists because 
of the non-specificity of the audience, but 
both readings are acceptable and, in fact, 
desirable. 
Rel~ on Interaction: The standard way speakers 
check whether or not they are being understood 
and modify their utterances appropriately is 
by interacting with the hearer. In the Paul 
Newman example above, the speaker could have 
watched for signs of recognition (a smile or a 
nod) from the listener which would have made 
the elaboration unnecessary. In an 
interactive situation, the motorcycle example 
could have been followed by "Do you remember 
that time?" Unfortunately, language 
experiences in which the audience is unknown 
or unknowable, such as books and lectures, are 
often those in which interaction is difficult. 
The presence of both a specific audience and 
interaction is a "positive feedback" situation 
in which communication is greatly facilitated. 
The absence of both, however, necessitates the 
adoption of some of the heuristics mentioned 
here - and even then, communication may be 
impaired. 
The heuristics identified above assume 
that the speaker takes the responsibility for 
the efficacy of the communication. Speakers 
in general, though, can assume that they share 
with the listener certain communicative 
principles of the type explicated by Grice 
(1975). They can similarly assume that 
listeners have certain heuristics for 
determining what assumptions the speaker is 
making about the audience and whether or not 
they as listeners fit those assumptions. 
Because they have this faith in their 
listeners, speakers sometimes just "broadcast" 
a remark, leaving it to the listener to decide 
who the intended audience is. Some of these 
listener's heuristics for interpreting 
broadcast utterances are described in the next 
section. 
138 
6. Hearers' Heuristics for "Broadcast" 
Utterances 
Integral to a listener's understanding of 
any utterance is a model of the speaker's 
model of the hearer. Where there is some 
mismatch between this model and the listener 
himself, it may be difficult for him to figure 
out what the speaker really intended to 
communicate. In language experiences where 
the listener suspects that the speaker is 
broadcasting -- that is, not being careful 
about specifying an audience - he must 
"broadreceive", using some heuristics for 
deciding whether or not he should consider the 
remark to be addressed to him. 
One general technique a listener might 
use is to compare the intended effect of the 
utterance with his current state. If he has 
already fulfilled the effect, he can consider 
the remark to be addressed elsewhere. 
Commands have clear intended effects, so it is 
relatively simple for a listener to use this 
heuristic with such speech acts. A member of 
a congregation who is already standing when 
the minister says "Please stand up" will 
understand that the utterance is not meant for 
him, rather than yelling out, "I already am!". 
Signs are another medium in which this 
broadcast behavior is apparent. A sign asking 
patients to PLEASE REGISTER WITH THE 
RECEPTIONIST is clearly meant only for those 
who haven't already done so; if patients 
didn't use this heuristic, they would find 
themselves in an infinite loop of registering. 
A somewhat more complex example is the 
familiar airport announcement: Extinguish all 
smoking materials and have your boarding pass 
ready." which selects two different subsets of 
the audience which hears it - those who are 
smoking and those who do not yet have their 
boarding passes ready. 
Deciding on the intended effect of an 
utterance is no mean trick, of course; the 
speech act literature, (e.g., Austin (1962), 
Searle (1969)), makes this clear. One 
interesting example of the interaction of 
these considerations with those of audience is 
a subway sign proclaiming SMOKING IS DANGEROUS 
TO YOUR HEALTH. The intended effect may be 
seen either as getting smokers to give up 
their habit or as telling smokers and 
non-smokers alike that smoking is not 
healthful. In the first case, the intended 
audience i3 smaller than in the second. 
Although this more restricted audience is 
implied by the use of "you" on the sign, it 
seems plausible that the informational effect 
and, therefore, the larger audience is 
intended as well. A non-smoker's 
interpretation of the sign is then a complex 
process, involving an awareness of the two 
possible audiences for the message. 
Finally, and most obviously, speakers 
assume that hearers will make use of pragmatic 
clues to determine whether they are part of 
the intended audience. A volleyball player 
yelling "I'll set" assumes that only the 
members of her team will attend to the remark; 
she doesn't need to preface it with a direct 
address to her teammates. Similarly, the same 
player calling "You hit" hopes that the 
pragmatic context is strong enough that the 
one person who is the intended audience can 
identify herself. 
7 • Summary 
This admittedly brief discussion of the 
specificity of audiences is meant to 
illustrate the kinds of considerations a 
narrow focus on a single language experience 
might overlook. It is clear from just these 
few examples that the process of planning a 
speech act must utilize heuristics like those 
listed above and that speakers, models of 
listeners must contain some explicit 
representation of the size and specificity of 
the audience. These insights would not have 
arisen had we restricted ourselves to 
two-person conversations. The 
multi-dimensional space developed in this 
paper provides eight dimensions which can 
provoke similar investigations and a framework 
in which to integrate the results. 
A Qkn owledgement s 
I would like to thank Chip Bruce for comments 
on earlier versions of this paper, Mitch 
Marcus, Barbara Grosz and A1 Stevens for help 
with examples, and Jill O'Brien for preparing 
the final document. 
This research was supported by the National 
Institute of Education under Contract No. 
US-NIE-C-400-76-0116. 
139 
Referenee~ 
Austin, J.~. How to do _ Things 
Harvard University Press, 
Massachusetts, 1962. 
wi~ Wozds. 
Cambridge, 
Clark, Herbert H. and Marshall, Catherine. 
Definite ReferenDe and M~tual Knowledge. Paper 
presented at the Sloan Workshop on 
Computational Aspects of Linguistic Structure 
and Discourse Setting, University of 
Pennsylvania, May 1978. 
Cohen, Philip R. Qn Knowing What ..to.. Soy: 
Planning S~ech Acts. PhD Thesis, University 
of Toronto, Computer Science Department, 1978. 
deLaguna, G. Soeech: .... Its function ~a~d 
develo~ent. College Park, Maryland: McGrath 
Co., 1970 (Reprint of 1927 edition). 
Grice, H.P., Logic and Conversation. In D. 
Davidson and G. Harman (Eds.) The Logic of 
Grammar, Encino, California: Dickenson 
Publishing Company, 1975. 
Grosz, Barbara J. The reores~ntation and use 
of focus i~ dialogue understanding, PhD 
Thesis, University of California at Berkeley, 
Computer Science Department, May 1977. 
Keenan, E., & Scheffelin, B. Topic as a 
discourse notion. In hi (Ed.), Subject ~nd 
topic. Academic Press, 1976. 
Rubin, Ann D. A Theoretical Taxonomy of the 
Differences Between 0ral and Written Language. 
To appear in R. Spiro, B. Bruce and W. Brewer 
(Eds.), Theoretical Is~ue~ in ~ding 
e~, Hillsdale, N.J.: Lawrence 
Erlbaum, 1978, in press. Also as Center for 
the Study of Reading, Technical Report No. 35, 
1978. 
Schallert, D.L., Kleiman, G.M., & Rubin, A.D. 
Analyses of differences between written and 
oral language. Center for the Study of 
Reading, Technical Report No. 29, April 1977. 
Searle, J.R. SDeech acts: An Essay in the 
ph£L~soDhv Qf language. Cambridge: Cambridge 
University Press, 1969. 
Sidner, Candace L. A Progress Report Qn the 
DfseQur~e ~d Reference Compo~enhs ok PAL. AI 
Memo 463, Artificial Intelligence Laboratory, 
Massachusetts Institute of Technology, 
Cambridge, Massachusetts, 1978. 
Sticht, Thomas G., Beck, Lawrence J., Hauke, 
Robert N., Kleiman, Glenn M., and James, James 
H. &uding ~nd ReaHing: A Developmental Model. 
Human Resources Research Organization, 
Alexandria, Virginia, 1974. 
Tanz, C. Studies in the aoQu£sit%on of deictic 
~erms, PhD Thesis, University of Chicago, 
Department of Psychology, August, 1976. 
140 
