Discourse Processes

ISSN: 0163-853X (Print) 1532-6950 (Online) Journal homepage: https://www.tandfonline.com/loi/hdsp20

Indexical understanding of instructions

Arthur M. Glenberg & David A. Robertson

To cite this article: Arthur M. Glenberg & David A. Robertson (1999) Indexical understanding of
instructions, Discourse Processes, 28:1, 1-26, DOI: 10.1080/01638539909545067

To link to this article: https://doi.org/10.1080/01638539909545067

Published online: 11 Nov 2009.

Submit your article to this journal

Article views: 910

View related articles

Citing articles: 21 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=hdsp20
DISCOURSE PROCESSES, 28(1), 1-26
Copyright © 1999, Lawrence Erlbaum Associates, Inc.

Indexical Understanding of Instructions

Arthur M. Glenberg and David A. Robertson
Department of Psychology
University of Wisconsin-Madison

Background knowledge is essential for understanding. Our question concerns the

nature of that knowledge: Is background knowledge solely descriptive and abstract,
that is, consisting of propositions, schemas, and rules, or is there room for expe-
riential and perceptual components? The indexical hypothesis suggests that expe-
riential components are crucial for language comprehension. On this hypothesis,
indexing, that is, referring words and phrases to objects (or analogical repre-
sentations of objects), is required for comprehension. Once a phrase is indexed to
an object, then affordances derived from the object are used to guide the interpre-
tation of the language. We demonstrate support for the indexical hypothesis by
manipulating the opportunity to index words to objects while acquiring background
information about how to use a compass and map to identify landmarks. The
participants acquired similar levels of abstract knowledge as assessed by a verbal
test. Nonetheless, participants given the opportunity to index, compared to those
prevented from indexing, read and followed new directions faster, referred to the
background information less frequently, and performed the task using the compass
and map more accurately. Discussion focuses on how these data are consistent
with the indexical hypothesis and how that hypothesis pertains to comprehension
of narratives.

Understanding language requires forming a mental model (Zwaan & Radvansky,

1998) based on the language used, background knowledge, and inferences. Much
of the research investigating mental models has examined the types of inferences
drawn and the situations in which they are sanctioned. Our goal is to characterize
the format of the background knowledge that enters into the construction of
mental models.
Background knowledge is critical for effective learning and comprehension.
This simple fact is supported by research and by common sense: One could hardly
expect to understand calculus without understanding algebra, and one could not

Correspondence and requests for reprints should be sent to Arthur M. Glenberg, Department of
Psychology, University of Wisconsin, 1202 West Johnson Street, Madison, WI 53706. E-mail:
glenberg @ facstaff.wisc.edu
2 GLENBERG AND ROBERTSON

hope to understand a text about the recent history of Vietnam without knowing
something about nations and war. Background knowledge has typically been
characterized as abstract semantic knowledge—for example, nodes, propositions,
rules, and schemas. By abstract, we mean that the representational units are
arbitrarily related to what is being represented. Thus, the numeral 5 is an abstract
symbol in that its size, shape, density, and so forth are unrelated to the numerosity
of five objects. Similarly, the nodes in a semantic network are arbitrarily related to
the objects they represent. In this article, we investigate the contribution of a
different aspect of background knowledge—namely, perceptual information. In
contrast to abstract knowledge, perceptual information is modal and analogical.
That is, the representational units are related to the objects through perceptual
experience, and the representation replicates at least part of that experience.
Barsalou (1999) called these sorts of representational units perceptual symbols.
Many accounts of background knowledge suggest that background knowledge
is solely (or most importantly) abstract knowledge. Chiesi, Spilich, and Voss
(1979) demonstrated how background knowledge of the rules and goals of base-
ball are essential for the interpretation of a passage describing a baseball game.
These rules and goals are abstract descriptions (usually using words) in that they
do not include perceptual or experiential information, although they are surely
derived from perception and experiences. Chi, Feltovich, and Glaser (1981) dis-
cussed the differences in background knowledge used by experts and novices in
solving physics problems. Novices are presumed to use a "naive" representation
that "contains the literal objects and their spatial relationships" (p. 134). Experts,
however, move beyond this perceptual knowledge by using a "representation
[that] contains the idealized objects and physical concepts, such as forces, mo-
ments, and energies which are necessary to generate the equations of the algebraic
representation" (p. 134). Later, Chi et al. noted that experts "base their selection
of the appropriate [problem solving] principle on the resulting second-order,
derived cues" (p. 150). Thus, the experts are presumed to utilize a more abstract
and "derived" representation. As a final example, McNamara and Kintsch (1996)
demonstrated how background knowledge affects the comprehension of a text
about Vietnam. They noted that this background knowledge may be derived from
"knowledge about the language, about the world in general . . . [and] also the
reader's personal experience" (p. 252). However, the form of this background
knowledge is taken to be propositional as implemented by the construction-in-
tegration model (Kintsch, 1988) used to explain the results.
In contrast to this research, consider an anecdote that suggests that some types
of background knowledge may have aspects that are perceptual and not abstract.
Recall the last time you attempted to construct an artifact such as a bookshelf
or to program a VCR. Upon initial reading, the instructions may well seem to
be incomprehensible, bizarre, or written by a nonnative user of the language.
After experience with pushing buttons, turning screws, and aligning boards, how-
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 3

ever, the instructions begin to make sense. In fact, after completing the task, the
difficulty that you had with the instructions may no longer be apparent: They
are written clearly.
How does the experience of pushing buttons and aligning boards facilitate the
comprehension of the instructions? One possibility is that the experience resulted
in encoding additional abstract knowledge, much like what one would gain from
further reading. That seems unlikely, however. Imagine that the instructions for
building the bookshelf contained descriptive footnotes and appendixes. For ex-
ample, one of the footnotes might provide a lengthy description of just how to
turn the screws and push the buttons, or an appendix might verbally describe the
exact dimensions of the boards and their colors, grains, and hefts. This morass
of words is more likely to engender confusion than enlightenment. A second
possibility is that the experience results in a type of knowledge that is not abstract.
For example, consider the instruction, "Turn the screw until it can be inserted
into the tunnel." 'Turning a screw" is a common enough phrase, and it seems
to be comprehended without difficulty. But why does the instruction read, "until
it can be inserted"? Wouldn't it be simpler to write, "Insert the screw into the
tunnel" or "Screw the screw into the tunnel"? The confusion arises because we
envision turning a screw by rotating it along its major axis. However, in building
a bookshelf, it may be necessary to place the screw into a dado on the bottom
of the shelf and turn the screw around its minor axis until the screw can be
inserted into a tunnel that allows the screw to connect to another board. (Under-
standing this last sentence will be facilitated by reference to Figure 1.) The point
is that this simple instruction about the screw may be difficult to comprehend
without reference to a physical situation or an analog representation of that
situation—such as Figure 1—or a perceptual symbol. That is, the proper sense
of turn must be selected based on constraints provided by the physical situation.

FIGURE 1 A context in which it makes sense to say, 'Turn the screw until it can be
inserted into the tunnel."
4 GLENBERG AND ROBERTSON

The indexical hypothesis helps to explain why reference to a physical situation,

not a description, is needed to understand some language. The standard linguistic
category of indexicals consists of words such as pronouns and deictics (e.g., this
and that have deictic functions) that take much of their meaning from the context
in which they are uttered. Indexing is related to anaphora. In both cases, inter-
pretation requires recovering information from a source outside of the immediate
words. Consider the sentence, "The faculty believe that they deserve a raise."
The word they might refer to faculty. In this case, they is anaphoric in that it
takes its meaning from the preceding linguistic context. However, they might
refer to a group of people, for example, a group of teaching assistants, to which
the speaker is pointing. In this case, they is indexical in that it takes its meaning
from the extralinguistic context. Thus, indexing provides a link between a word
and an extralinguistic context such as an object or an illustration of an object.
Here, we extend the term indexical in two ways. First, we assume that a word
or phrase can be indexed to a perceptual symbol (Barsalou, 1999) as well as a
physical object. Second, we propose that many components of language (e.g.,'
noun phrases, verbs such as turn, etc.) are indexical.
As with most hypotheses about language, the indexical hypothesis asserts that
comprehension requires the coherent combination of ideas, such as ideas about
screws, turning, tunnels, and so on. The difference is in the format of those ideas
and how they are combined. According to the indexical hypothesis, the combi-
nation of ideas depends on affordances (Gibson, 1979) of objects. Affordances
arise from the interaction of an observer and an object. They are ways in which
an observer can interact with an object, and they are presumed to arise, to a great
degree, from perceptual processes. For example, a floppy disk affords grasping,
and it affords back scratching. A thread also affords grasping, but it does not
afford back scratching. Our use of the term affordances differs from Gibson's
in that we do not wish to claim a direct perception of affordances, and we allow
for the mental representation of affordances. What is common between Gibson's
and our use of the term is that affordances are the causal properties of an object
or action relevant to the perceiver.
According to the indexical hypothesis, words and phrases are indexed to
objects, analog representations such as pictures, or internal analog representations
such as perceptual symbols (Barsalou, 1999). This indexing makes available the
affordances of the objects. The important claim is that it is the affordances of
the objects, not the words, that constrain the way ideas are combined. Because
we use affordances, a sentence such as "Danielle used the 3.5-inch floppy disk
to scratch her back" is sensible and understandable. Everyone knows that a floppy
disk can be used to scratch one's back even if no one has ever tried it or read
about it. Similarly, everyone knows that the sentence "Danielle used a 3.5-inch
piece of thread to scratch her back" does not make much sense, presumably
because the affordances of small lengths of thread cannot be successfully com-
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 5

bined with the actions of scratching one's back. Sentences such as these are
analyzed further in the Discussion section.
Consider another example to help make the indexical hypothesis a bit clearer.1
This example is taken from the compass learning task that forms the basis of the
experimental investigation reported later. In describing how to use a compass to
identify landmarks, one step is "Point the direction of travel arrows at the moun-
tain" (see Figure 2). Comprehension of this sentence requires an understanding
of what it means to point because pointing is quite different for different objects.
It means one thing to point an index finger, another to point a rifle, another to
point a car, and as we see later, quite another to point direction of travel arrows.
Do we have in memory a huge listing of what it means to point, and do we
select the appropriate meaning based on a search of memory? Perhaps, but con-
sider this alternative. Our knowledge consists of a general notion of what it
means to point, such as aligning the long axis of an object with a reference, or
to-be-pointed-at, object. Also, we have a repertoire of bodily skills, such as how
to grasp and turn objects of different sizes and hefts. To understand the sentence
about pointing the direction of travel arrow, we need to index the words direction
of travel arrow to the actual object so that its affordances are made available.
Then, we can mesh our general knowledge of pointing, the affordances of the
direction of travel arrow, and our repertoire of bodily skills to envision a coherent
set of actions that result in pointing the direction of travel arrow at the mountain.
In contrast, if we did not know the affordances of a direction of travel arrow,
then we could not comprehend the sentence: We would not know how to act on
it. Or, if a direction of travel arrow did not afford pointing, then the sentence
would be nonsense. For example, if a direction of travel arrow was a vector that
indicated the mean direction and duration in which one had been walking, it is
not clear what it would mean to point this vector at a mountain. Although the
vector has a direction, it is not something that can be changed by the usual
actions of pointing. Informally, the perceptual characteristics of the direction of
travel arrow flesh out what it means to point in this context; that is, the affordances
give meaning to the sentence.
The process of indexing a phrase to an object is not trivial. For example, a
compass has imprinted on it a variety of unfamiliar objects including at least
three arrows with different functions. As a first step in indexing, a reader might
parse a sentence to form a description of a to-be-indexed object (e.g., direction

1
The indexical hypothesis provides an explanation for how examples help comprehension. The
statement of a hypothesis often uses abstract language. According to the indexical hypothesis,
comprehension (especially in unfamiliar domains) requires indexing the abstract language to specific
objects and situations so that the affordances of the objects can constrain the combination of the
ideas. Examples use descriptions of specific, often concrete, situations. People can generate from
these descriptions analog perceptual symbols and mental models that make affordances of the objects
evident. The same explanation applies to how Figure 1 aids comprehension.
6 GLENBERG AND ROBERTSON

r -'•

FIGURE 2 Suunto Model M3 com-

pass. The direction of travel arrows
•It are the two parallel arrows imprinted
on the top half of the base of the
compass.

of travel arrows) and to determine characteristics of the object, such as color and
approximate location used to guide a search. Second, using this information, the
reader would scan the compass and locate the particular arrows. That is, he or
she would index the concept direction of travel arrow to the object. After locating
the object, affordances can be determined.
A skeptic might argue that indexing can be avoided by reading, perhaps in
an appendix, about the location, size, heft, and solidity of the object. But qualities
such as location, size, heft, and solidity are not easily described or remembered.
Even if the qualities were described, the description alone does not clearly convey
the implications for interaction that are required to know how to combine the
idea of pointing with the words direction of travel arrows. For example, although
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 7

the appendix might describe the dimensions of the object, some careful and tedious
thinking might be required to determine the major axis. Difficulty in accurately
describing physical characteristics, difficulty in remembering the descriptions,
and difficulty in deriving needed information from the descriptions are why
additional verbal information in appendixes and footnotes is not always helpful.
The point (!) is that understanding the meaning of "Point the direction of
travel arrows" may well require a background knowledge that is not verbal and
not abstract: knowledge of how to interact with an object by applying a repertoire
of bodily skills. That knowledge is acquired or made evident through indexing.
Another example of indexing, this time in the comprehension of brief narratives,
is presented in the Discussion section.
To summarize, the indexical hypothesis specifies three steps for understanding
sentences in context. The first step is to index words and phrases to objects,
analogical representations of the objects such as pictures, or to perceptual sym-
bols. This first step requires parsing the sentence, encoding descriptive informa-
tion used to guide a search, and finding the object. The second step is to derive
affordances from the indexed objects. The third step is to mesh those affordances
into a coherent (i.e., doable or envisionable) set of actions. This combination
process (or mesh; Glenberg, 1997) is explicated in the Discussion section.
We performed an experiment that provides a test of the indexical hypothesis.
All participants learned a new skill from verbal instructions. However, we ma-
nipulated the opportunity to index words to objects while acquiring background
knowledge. One group (listen and index [L&J]) saw a compass while listening
to background information; another group (listen and read [L&R]) was given
multiple opportunities to learn from verbal presentation of the background in-
formation. According to the indexical hypothesis, although two groups may have
similar abstract background knowledge, the group with greater perceptual knowl-
edge will comprehend a related text more fluently and effectively. In fact, the
L&R participants were marginally better on a test of abstract background knowl-
edge. Nonetheless, the L&I participants dramatically outperformed the L&R par-
ticipants on transfer tests: reading related information and using the compass.
At first glance, neither the indexical hypothesis nor the experimental results
may seem new. After all, we know that people often build mental models when
reading (e.g., Glenberg, 1997; Kintsch, 1988; Zwaan & Radvansky, 1998), and
we know that people gain enormously from reading when the text is accompanied
by pictures (e.g., Glenberg & Langston, 1992). Also, others have studied learning
from procedural or instructional texts accompanied by pictures (e.g., Mayer,
Bove, Bryman, Mars, & Tapangco, 1996; Mayer, Steinhoff, Bower, & Mars,
1995) or objects (e.g., Mills, Diehl, Birkmire, & Mou, 1995). What is new about
the hypothesis is that it makes a unique suggestion regarding the nature of mental
models: They are constructed using affordances. This suggestion is particularly
timely given recent evidence that mental models are not simple Euclidean spaces
(Langston, Kramer, & Glenberg, 1998; Rinck, Hahnel, Bower, & Glowalla, 1997).
8 GLENBERG AND ROBERTSON

Also, several features of the experiment and the data are new. First, we demon-
strate that indexing to pictures facilitates transfer, that is, the ability to read and
comprehend new texts. Second, we demonstrate that the effects of indexing can
be dramatic even when there are no differences among groups in abstract knowl-
edge. Third, we demonstrate effects not just on reading time but also in the
performance of a skill.

METHOD

Participants
The original group of participants consisted of 95 women enrolled in introductory
psychology classes at the University of Wisconsin-Madison. We selected women
as participants because of suggestions in the literature that women come to
orienteering tasks with less background knowledge than do men. Thus, constrain-
ing the sample to women should (a) reduce variability, (b) produce a sample
with low background knowledge to maximize our chances of manipulating it,
and (c) demonstrate the effectiveness of indexing for learning in a new domain.
As described later, the initial group of 95 women was pared to a final sample
of 76. All were native speakers of English. Participation in the experiment was
one of several options that could be used to fulfill a research requirement for the
course. In addition, participants could earn a cash bonus, which is described later.

Design and Procedure

Table 1 includes an outline of the design. The students participated in the ex-
periment individually. Participants were forewarned that they would be learning
about compasses and maps and that they would be asked to demonstrate their
knowledge by actually using a compass and map to identify landmarks. In the
first formal stage of the experiment (after signing consent forms), participants
were given a pretest (see Appendix A) to assess their knowledge of compasses
and maps. The test consisted of 17 four-alternative multiple-choice questions.
Next, participants were randomly assigned to one of four groups. The groups
were distinguished by the amount of opportunity to index while learning back-
ground information. The listen (L) group listened to the audiotrack of a videotape
(the script, see Appendix B) that described, in detail, parts of a compass (e.g.,
direction of travel arrows) and parts of a map (e.g., meridian lines). After listening
to the script, the participants read a filler text describing a backpacking trip. The
filler text referred to compasses and maps several times, but no details of their
use were provided. This condition affords few opportunities to index. Participants
in the L&R group also listened to the script. Instead of the filler text, however,
this group was provided with a written version of the script and asked to read
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 9

TABLE 1
Design of the Experiment

Group

Listen Listen and Index Listen and Read Listen and Picture

Pretest Pretest Pretest Pretest Pretest

Background videotape Audiotrack Audio and visual Audiotrack Audio and picture
only indexing only (no indexing)
Background text Filler Filler Written script Filler
Posttest Posttest Posttest Posttest Posttest

Transfer tasks
Written script plus written instructions provided to all.
Task: Use the compass and map to name a mountain.
Four mountains were named successively.

it. This second exposure to the script was intended to ensure a high level of
abstract background knowledge. The listen and picture (L&P) group listened to
the audiotrack presenting the script. While the script described parts of the com-
pass, the videotrack displayed an image of the compass used later in the experi-
ment, and while the script described parts of the map, the videotrack displayed
an image of the map used later. This condition provides the opportunity for
self-indexing. That is, to the extent that the participant could index the words to
components of the images, indexing would be accomplished. Following presen-
tation of the script, this group read the filler text. The L&I group listened to the
script while images of the compass or map were displayed on the screen. In
addition, when a term or operation was introduced, an actor's hands indexed the
term or operation. For example, when the script first mentioned "direction of
travel arrows," the actor pointed to those arrows on the compass. When the script
mentioned that the 360° dial can be rotated, the actor turned the dial. This
condition provided the greatest opportunity to accomplish indexing.2 Following
exposure to the script, the L&I participants read the filler text.
These conditions were chosen to allow several types of comparisons. The L
group forms a baseline condition. Comparison of the L group to the L&I group
allows for replication of a standard finding: Reading with pictorial support en-
hances comprehension. However, this comparison does not allow a precise speci-
fication of the nature of that enhancement. For example, the L&I participants

2
In pilot work, the participants indexed the script to a real compass (not a video image).
Unfortunately, this procedure had several drawbacks. First, participants would often ask the
experimenter if their indexing was correct. Second, participants would often index incorrectly, forcing
the experimenter to intervene. Third, with this procedure it was difficult to keep the experimenter
blind to a participant's condition. Thus, we opted for the artificiality of the videotape procedure to
preserve experimental control.
10 GLENBERG AND ROBERTSON

may learn more abstract information (derived from the video) than the L partici-
pants, or they may have learned perceptual information. Comparison of the L&I
with the L&R group provides a more precise specification of what is learned.
Namely, if the two groups perform similarly on a verbal test of abstract knowl-
edge, but the groups differentiate on other tests of knowledge, then we will have
evidence that the L&I group has excelled in more than learning abstract infor-
mation. Finally, comparison of the L&I and L&P groups allows us to compare
active indexing (L&I) to a picture-only (i.e., video-only) control.
The experimenters were kept blind to condition by taking the following pre-
cautions. First, the four videotapes corresponding to the four conditions were
arbitrarily labeled A through D. The correspondence of the labels to conditions
was known only to the first author, Arthur M. Glenberg (who did not act as an
experimenter), and the correspondence was changed weekly over the 12 weeks
of data collection. Second, the written filler text and written script read by the
L&R participants were enclosed in file folders that were also labeled A through
D. Third, when the participant viewed the videotape, the screen was oriented so
that it was not visible to the experimenter. Although the experimenter could hear
the audio portion, it was identical for all conditions. Thus, from the experimenter's
point of view, participants were randomly assigned to conditions A through D,
and the referents of these labels were not evident.
After reading the filler (or the script for the L&R group), all participants were
treated identically. In the next stage of the experiment, participants retook the
compass and map test, which was now a posttest. The purpose of the posttest
was to demonstrate (a) that there was significant learning accomplished by virtue
of exposure to the script, and (b) that the conditions differed in the amount of
abstract background knowledge acquired. In particular, it was important that
performance of the L&R group given two opportunities to learn from the script
not be exceeded by any of the other groups. Given this outcome, we can claim
that the abstract background knowledge of the L&R group was at least as great
as that of any other group.
Following the posttest, the participants were taken to the middle of a large
laboratory room where a small, waist-high cart was located. Taped to the top
surface of the cart was a picture of an "x-marks-the-spot" type of cross. In using
a compass to identify landmarks, it is convenient to know one's exact location
in the environment. The cross on the cart was used to designate the exact location,
and a similar but much smaller cross was printed on the map. Other materials
also located at the cart included a Suunto Model M3 compass, the map, the
written version of the script, and the critical instructional text. The compass and
map were those used in production of the videotapes. Printed on the map were
a compass rose; meridian lines slanted to reflect magnetic declination; a cross
corresponding to the participant's current location (at the cart), which was printed
in the middle of the map; and 19 small triangles corresponding to 19 named
mountains. The mountains were arranged irregularly around the edge of the map.
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 11

The script was identical in content to the material presented in the audio portion
of the videotape. It was printed on one side of a piece of paper.
The critical instructional text was printed on the other side of the paper con-
taining the script. The instructional text presented information on how to use a
map and compass together to determine the names of landmarks. These instruc-
tions are reproduced in Table 2. Remember, all of the technical terms used in
the instructions were carefully described in the script. The question of interest
was how well the participants could use information gained from the script (and
indexing) to understand and use these critical instructions.
Thirty large triangles, approximately 17 cm on a side, corresponding to the
to-be-named mountains, were taped to the walls of the room and were approxi-
mately equally spaced around the room. A dot marked the centroid of the triangle
to assist in precise localization when using the compass. The number and spacing
of these triangles did not correspond to the number and spacing of the triangles on
the map to force the participants to use the compass to identify mountains rather
than to look for groups of mountains that clustered identically in the environment
and on the map. This is the usual situation in the real environment. That is, one's
map does not correspond exactly to what is visually available. Some mountain
peaks will be hidden from view because of trees and other mountains.
When a participant was brought to the cart, the various materials (compass,
map, script, instructional text, cross, large mountain symbols) were pointed out.
The participant was further instructed that she would have to read the instructional
text and use the other materials to name four mountains, one at a time. Because
the task took some effort, we attempted to motivate the participant by explaining
that she would be awarded a cash bonus of $ 1 if she managed to correctly name
three of the four mountains. Each of the four naming trials began with the

TABLE 2
The Three-Step Instructions

1. While holding the compass over the large cross on the cart," face the mountain," hold the compass
level, and point the direction of travel arrows directly at the mountain." It is very important that
you hold the compass level and over the large cross on the cart.
2. Without moving yourself or the base plate (that is, keep the direction of travel arrows pointed at
the mountain), turn the 360 degree dial" until the red part of the magnetic needle and the orienting
arrow are aligned." Note the number of degrees indicated by the pointer. This number is the
bearing (direction in degrees) to the mountain. Once you have taken the bearing, be careful not
to turn the 360 degree dial. PLEASE TELL YOUR EXPERIMENTER THE BEARING.
3. Place the small cross on the left edge of the compass on top of the cross on the map." While
keeping the two crosses aligned, rotate the base plate (not the 360 degree dial) until two conditions
are met: (a) The orienting lines are parallel with the meridian lines," and (b) the orienting arrow
is pointing approximately toward North on the map." The front corner of the left edge now points
directly at the symbol on the map for the mountain;" just read off the name of the mountain.
PLEASE TELL YOUR EXPERIMENTER THE NAME OF THE MOUNTAIN,

"Components scored; see the Method section for an explanation.

12 GLENBERG AND ROBERTSON

participant closing her eyes. The experimenter designated a to-be-named mountain

by attaching a yellow self-stick note to the appropriate large triangle. The four
mountains were located at bearings (number of degrees from the magnetic north
to the mountain) of 124°, 210°, 287°, and 351°. The order in which they were
identified was counterbalanced over participants. The experimenter then returned
to the participant's side and directed her to open her eyes. The experimenter
pointed out the to-be-named mountain and started a stopwatch when the partici-
pant began to read the instructional text. The participant attempted to follow the
instructions, find the bearing of the mountain, and name the mountain. Upon
naming the mountain, the experimenter recorded the time, and the next trial began
with the participant closing her eyes while the experimenter designated the next
mountain. After the last mountain was named, the participant was debriefed and
paid any bonus that had been earned.

Dependent Variables

For each trial (i.e., naming a mountain), the experimenter recorded several de-
pendent variables. One dependent variable was the total time needed to read the
instructions and name a mountain. A second dependent variable was the bearing
indicated by the participant as part of the second step (see Table 2).
Three additional dependent variables were recorded for each step of the in-
structions.

1. We scored how accurately the participant performed the step. For example,
for Step 1, we recorded a score of 0 to 3 corresponding to whether or not the
participant held the compass over the cross (1 point), faced the mountain (1
point), and pointed the direction of travel arrows at the mountain (1 point). Up
to 2 points could be scored for Step 2 (see Table 2), and up to 4 points could
be scored for Step 3.
2. We noted how frequently the participant referred to the script. Remember,
the script was printed on the reverse side of the page containing the instructions.
Thus, it was relatively easy to note references to the script: The participant turned
over the page. We used the number of references to the script as a measure of
indexing. As described in the introduction to this article, indexing from text is
likely to require several steps, including encoding information from the text
needed to locate the to-be-indexed object. Thus, when a participant trying to
understand the instructions returns to read the script, that reading will be correlated
with indexing.
3. We also noted reference to the written instructions.

The experimenters were thoroughly trained in how to take these measurements:

They worked with Arthur M. Glenberg, they practiced on one another, and 17
pilot participants were used to ensure consistency of measurement.
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 13

RESULTS

Elimination of Data
The task we set for the participants required their cooperation. That is, they were
asked to read in an unfamiliar domain and to learn a set of high-level skills well
enough to perform them accurately. During the pilot testing, it became obvious
that some of the participants simply did not wish to cooperate to that extent. For
example, the median time needed to identify the first mountain was 249 s, but
some participants gave a response within 64 s. For the fourth mountain, the
median time was 73 s, whereas some responded in as short as 26 s. Accuracy
was similarly variable. For the first mountain, the median error in finding a
bearing was only 10°. However, the mean was 51°, and the third quartile was
120° (the maximum possible error was 180°).
We decided on the following criteria to distinguish between participants who
were trying to learn and those who were not. We kept a participant's data if that
participant was able to set the bearing (Step 2) for at least two mountains with
an error of less than 30°. This criterion resulted in the elimination of 18 partici-
pants, approximately equally distributed across the four conditions. By eliminat-
ing 1 additional participant, the last participant run in the experiment, we ended
up with 19 participants in each group. Only data from these 76 participants are
included in the following analyses.

Pretest
The means on the pretest (taken before exposure to the script) are presented in
Table 3. The groups did not differ significantly, F(3, 72) = 1.84, MSE = 3.95,
p = .15. The overall mean, 5.70 (SD = 2.02), is not much different from what
would be expected by chance, 4.25.

Posttest
The means on the posttest (taken after exposure to the script) are presented in
Table 3. The overall mean, 13.26 (JSD = 2.44), is significantly greater than the
mean of the pretest, t(72) = 25.03, SE = 0.30, p < .001. Clearly, the participants
learned from the script. Also, the groups differed in how much they learned, F(3,
72) = 5.72, MSE = 5.01, p < .001. Participants in the L&I and L&R groups
performed better than did the participants in the L and L&P groups, F(l, 72) =
16.81, p < .001, but there was no significant difference between the L&I and
L&R groups, F(l, 72) = 0.02. The failure to find a significant difference between
these two groups does not reflect a lack of statistical power. Assuming that the
mean square error from the analysis of variance (ANOVA) is a good estimate
for the population variance, then the experiment had a power of 0.8 to detect a
 14 GLENBERG AND ROBERTSON

TABLE 3
Data From the Experiment

Listen t Listen Listen and

Listen and Index Value* and Read Picture

Pretest (mean no. correct) 5.53 6.16 0.08 6.21 4.90

Posttest (mean no. correct) 12.00 14.26 0.25 14.37 12.42

Script references (no. on first trial) 1.95 0.42 4.23** 1.00 0.89
Script references (mean for four trials) 0.76 0.22 1.44 0.42 0.32
Instruction references (mean for four trials) 3.84 4.25 0.59 4.84 4.17

Performance time (s) 159.26 148.63 1.90* 186.07 159.60

Bearing error (median degrees) 9.00 6.00 0.24 6.50 10.00

Step 1 (mean correct out of three) 2.34 2.92 2.12** 2.54 2.68
Step 2 (mean correct out of two) 1.79 1.91 0.10 1.93 1.92
Step 3 (mean correct out of four) 2.33 2.80 0.37 2.92 2.51

"The t values are comparisons between the Listen and Index group and the Listen and Read group based
on the overall mean square error.
*p = .06. **p < .05.

real difference between the L&I and L&R posttest means as small as 2.03, a
difference comparable to the difference between the L&I group and the L group.
The posttest data are extremely important for interpreting the rest of the data.
First, they demonstrate quite a bit of learning. Second, with the data we demon-
strate that indexing helps: The L&I group outperformed both the L and L&P
groups. The difference replicates the oft-reported beneficial effects of pictures
on language comprehension. The question of interest is, What exactly did the
indexing accomplish? One response is that the indexing resulted in the acquisition
of additional abstract knowledge. For example, the participants might have
learned a precise description for the location of the direction of travel arrows.
Because the L&I and L&R groups demonstrated equivalent abstract knowledge,
this alternative predicts that the two groups will perform similarly on the tasks
that follow. A second alternative, the indexical hypothesis, predicts that the L&I
group learned something in addition to abstract knowledge. For example, by
indexing the words direction of travel arrows to the object, the L&I participants
learned the affordances of direction of travel arrows. For example, the L&I
participants will know that to point the direction of travel arrows requires holding
the compass in the palm of one's hand, even though this has not yet been described
in the text. Comparison of these two alternatives requires the comparison of
performance in the L&I and L&R groups, and analyses of that contrast are
reported in Table 3. The error term for these contrasts was formed from the mean
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 15

square in the relevant ANOVA using all four groups. Hence, the / statistic has
72 degrees of freedom. Contrasts involving the L and L&P groups are difficult
to interpret for two reasons. First, these groups differ from the L&I group in
abstract knowledge as well as opportunity for indexing. Hence, the comparison
is confounded. Second, there is some evidence for a speed-accuracy trade-off.
That is, whereas participants in the L and L&P groups read the material quickly
(see Table 3), they performed relatively poorly on tasks such as setting the
bearing.

References to the Script

The mean number of times a participant looked at the written script is presented
in Table 3. The first line presents the data for the first trial, and the second line
presents the data collapsed over the four trials. A two-factor ANOVA was used
to analyze these data. Group was a between-subjects factor with four levels (L,
L&I, L&R, and L&P), and trial was a within-subjects factor with four levels
(first through fourth mountain identified). There were significant effects for group,
F(3, 72) = 5.89, MSE = 0.72; trial, F(3, 216) = 14.06, MSE = 0.44, p < .001;
and the interaction, F(9, 216) = 3.37, p < .001. The interaction comes about
because of the substantial differences among the groups on the first trial (see
Table 3). After the first trial, however, there were far fewer references to the
script (M = 0.22) and consequently small differences among the groups. The
most important finding concerns the contrast between the L&I and L&R groups
on the first trial, r(72) = 4.23, p < .001.
These data provide the first hint that the L&I participants learned something
different from the L&R participants. Although the two groups were equivalent
on the posttest and the L&R group had actually read the script, the L&R group
referred to the script more than twice as often as the L&I group. As discussed
earlier, one interpretation of this behavior is that the participants in the L&R
group were attempting to use information in the script to accomplish indexing
now that the actual compass was available to them.

References to the Instructions

The three-step instructions (Table 2) were new for participants in all groups, and
hence we expected all to read the instructions carefully. Table 3 presents the
means computed over the four trials. The participants looked at the instructions
frequently, and there were no significant differences among the groups, F(3, 72)
= 1.39, MSE = 9.51./7 = .25. The frequency of looking at the instructions decreased
over trials (from 6.61 to 1.89), F(3, 216) = 144.58, MSE = 2.39, p < .001. The
interaction between the factors was not significant, F(9, 216) = 1.74, p = .08.
16 GLENBERG AND ROBERTSON

Time
The mean time to identify a mountain averaged over the four trials is presented
in Table 3. Because time to identify a mountain decreased substantially from the
first (283 s) to the fourth (82 s) mountains, the trials effect was significant, F(3,
216) = 139.67, MSE = 4,328, p < .001. Although the group effect was not
significant, the specific contrast between the L&I and L&R groups was marginally
significant, f(72) = 1.90, p = .06. It is notable that, although the participants in
the two groups performed similarly on the posttest, and the L&R participants
had actually read the script, the L&R participants took about 25% more time
than did the L&I participants. The Group x Trial interaction was not significant,
F(9, 216) = 1.01.
As with the references to the script, these data demonstrate that the L&I and
L&R groups have learned something different even though they performed simi-
larly on the posttest. In fact, it appears that the two findings are related: The
reason the L&R group takes longer to read and perform the task is because they
are referring to the script. Support for this hypothesis is provided by a series of
regression analyses (see Table 4). We attempted to predict reading and perform-
ance time for each trial separately using variables that are theoretically related
to time. Posttest performance measures abstract knowledge; hence, on the hy-
pothesis that abstract background knowledge affects comprehension, it should
contribute to reading time. References to the script is a measure of indexing;
hence, if indexing is an important component of the comprehension of instruc-
tions, it should account for a sizable proportion of variance. Finally, references
to the instructions, a rather direct measure of time spent reading, should also
enter into the regression equations.
Several effects reported in Table 4 are notable. First, performance on the
posttest was never a significant predictor of time. Second, the number of refer-
ences to the script was always a significant predictor of reading time. Third, the
number of references to the instructions is usually a significant predictor of
reading time. Fourth, overall, these variables account for a fair amount of the
between-subjects variance in reading time.

TABLE 4
Regression Analyses: Predicting Time to Read
the Instructions and Perform the Task

Coefficients

Script Instruction
Response Posttest References References R2

Trial 1 time 7.30 (/= 1.35) 38.69 (t = 3.44) 10.78 (t = 1.73) .18
Trial 2 time 6.33 (t= 1.67) 58.00 (r = 3.99) 11.12 (r = 2.64) .25
Trial 3 time -0.65 (t = -0.27) 41.26 (t = 4.06) 14.44 (t = 5.51) .45
Trial 4 time -0.16 (f = -0.09) 47.10 (/ = 4.34) 7.31 (f = 2.94) .33
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 17

Performance Measures

Accuracy is not as important a dependent measure as might first be expected

because there was no speed stress and because participants had unlimited time
to identify the mountains. Nonetheless, Table 3 reports several measures of ac-
curacy. The first is the median absolute error in the bearing set during Step 2
computed over the four trials (see Table 2). Medians are reported because of
positive skew in the data. There were no significant differences among the con-
ditions or across trials; remember, however, that there was a large decrease in
time across trials.
We also scored the number of components of each step performed correctly
for Steps 1 through 3 (see Table 2). For Step 1, which had a maximum possible
score of 3, there was a significant difference among the groups, F(3, 72) = 3.71,
MSE = 1.22, p < .02. The contrast between the L&I and L&R groups was
significant, f(72) = 2.12, p < .04. Although the two groups were statistically
indistinguishable on posttest performance, and the L&R group took 25% longer
to read and perform the instructions, the L&I group was, nonetheless, more
skillful in applying the instructions.
For Step 2, which had a maximum possible score of 2, there were no significant
differences among the groups, although there was a marginally significant im-
provement over trials (from 1.80 to 1.92), F(3, 216) = 2.52, MSE = 0.11, p =
.06. Similarly, for Step 3, which had a maximum possible score of 4, there were
no significant differences among the groups, although there was a significant
improvement over the trials (from 0.58 to 0.70), F(3, 216) = 5.63, MSE = 0.65,
p < .001.
The failure to find significant differences between the L&I and L&R groups
on several of the performance measures appears to be evidence contrary to the
indexical hypothesis. Remember, however, that participants were given unlimited
time to read the instructions and the script and to perform the task. In addition,
they were given a bonus for performing the task correctly. Thus, as in many
reaction time and reading experiments, the critical dependent variable is time.
Given that the L&I participants read the materials and performed the task more
quickly than did the L&R participants, it is particularly noteworthy that the L&I
participants outperformed the L&R participants on any measures of performance.

DISCUSSION

The major question concerns the nature of background knowledge. The data
suggest that perceptual knowledge plays a role in the comprehension of the
three-step instructions. As assessed by the posttest, the L&I and L&R groups
appeared to be very similar in the amount of abstract background knowledge
learned. Nonetheless, the groups diverged when asked to apply that knowledge
to further reading, comprehension, and performance. Thus, the L&R participants
18 GLENBERG AND ROBERTSON

took about 25% longer to read the instructional text than did the L&I participants,
the L&R participants looked back at the script twice as often as did the L&I
participants, and the L&R participants were less successful in completing the
components of Step 1 than were the L&I participants.
One interpretation of these data is that the L&I participants learned more
abstract information than did the L&R participants. Several aspects of the data
speak against this interpretation, however. First, this interpretation trades on the
assumption that the posttest failed to reveal differences between the groups in
abstract information. In contrast, the posttest was vetted by both naive and expert
confidants to ensure that it adequately tested the information presented in the
script.3 Also, performance on the posttest was not near the ceiling (17 correct),
which could have obscured any group differences. Second, this interpretation
does not help us to understand why participants in the L&R group referred back
to the script so frequently. Remember, this group was the only group to be
exposed to the scripted information twice. Why would these participants need to
look back at the script? The indexical hypothesis offers an answer. To understand
an instruction such as "point the direction of travel arrows at the mountain," the
reader must know how to physically manipulate the arrows. Of course, this
requires knowledge of the location of the arrows, their heft, size, and so forth.
Thus, a typical reader in the L&R group would reread the script to encode
information useful for locating the direction of travel arrows and use this infor-
mation to index the concept direction of travel arrow to the object. Second, the
reader would acquire the affordances of the object. Finally, the instruction could
be understood by meshing—that is, combining the affordances into a coherent
(i.e., envisionable and doable) set of actions.
If this indexing account is correct, we would expect the number of references
to the written script to be negatively related to the opportunity to index during
presentation of the background information. Thus, the fewest references are ex-
pected for the L&I group (M = 0.22 for four trials, M = 0.42 on the first trial)
and next fewest for the L&P group that may have been able to accomplish some
indexing using the video picture of the compass (Ms = 0.32 and 0.89, respec-
tively). Participants in the L&R group may have been able to accomplish some
indexing by creating an analog mental model from the multiple exposures to the
script (Ms = 0.42 and 1.00, respectively). The participants in the L group should
have accomplished the least indexing, and, in fact, they refer to the script the
most frequently (Ms = Q.I6 and 1.94, respectively).
An alternative hypothesis is that number of references to the script is inversely
related to amount of abstract background knowledge acquired. That is, those
participants who learned the least from the initial presentation of the script would
need to refer to it the most while reading the instructions. In fact, the correlation

3
Thanks are due to members of the Seminar in Cognitive Psychology, Spring 1996-1997, who
contributed to development of the posttest.
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 19

between posttest performance, used as a measure of abstract background knowl-

edge, and script references is —.34 (r2 = .11), which is consistent with this
alternative. However, when variability associated with the conditions (L, L&I,
L&R, and L&P) is removed, the relation between posttest and script references
is greatly reduced (r2 = .06). In contrast, the conditions variable (L, L&I, L&R,
and L&P) accounts for r2 = .20 of the variability in the number of references to
the script. When variability associated with the posttest is removed, the relation
between conditions and script references is still strong (r2 = .16), F(3, 72) = 4.39.
Thus, although there is modest support for the alternative that script references
are controlled by amount of abstract background knowledge, the data reveal a
stronger relation between references to the script and need to index during reading
of the instructions.
It is important to note that these results could only be found when using a
"rich" experimental environment. If we only tested learning using conventional
means, such as the verbal posttest, we might have wrongly concluded that the
L&I group and the L&R group had learned equivalent information. In contrast,
by noting references to the script and by using performance measures of learning,
we came to quite the contrary conclusion. This fact underscores the importance
of respecting naturalistic situations in studies of language and learning (see Clark,
1996, 1997).
Is indexing a special process that is used only when dealing with instructional
texts in unfamiliar domains? Although the data reported here speak solely to that
aspect of indexing, we believe that the indexical hypothesis is much broader.
First, the use of pictures in textbooks, manuals, encyclopedias, and so on, as well
as the beneficial effects of pictures on text comprehension (e.g., Glenberg &
Langston, 1992), are all consistent with the indexical hypothesis. The hypothesis
is broadly consistent with results reported by Mayer and colleagues (e.g., Mayer
et al., 1996; Mayer et al., 1995). They found that pictures accompanying scientific
texts facilitate performance on comprehension tests and abstract problem solving.
In addition, they noted the importance of the simultaneous occurrence of pictures
and texts. The indexical hypothesis explains this finding: It is easier to index
phrases to pictures (and thereby derive affordances) when the text and picture
are presented simultaneously rather than separately. Second, frequent use of ges-
ture in speech (e.g., Clark, 1996; McNeill, 1992) suggests a role for indexing.
Some gestures are used to point to objects so that indexing can be accomplished.
Other gestures may be iconic (e.g., to indicate the length of a fish caught) to
allow indexing to the gesture itself. Third, there are many instances in which
specificity is an important feature in language. Spatial descriptions are more
memorable the more definite they are (Mani & Johnson-Laird, 1982); specific
sentences are read more quickly than ambiguous sentences (Glenberg, Robertson,
Jansen, & Johnson-Glenberg, in press; Glenberg & Robertson, in press; see also
Robertson, Glenberg, & Members of the 1997/98 Honors Seminar in Cognitive
Psychology, 1999); and proper names, which carry definite reference, are more
20 GLENBERG AND ROBERTSON

memorable and receive preferred processing over common nouns (Miiller &
Kutas, 1996; Robertson, Gernsbacher, & Robertson, 1998).
Finally, other research from our laboratory illustrates indexing in the compre-
hension of short narratives (Glenberg, 1998; Glenberg & Robertson, in press).
In one experiment, the participants rated the sensibility of sentences such as
"After wading barefoot in the lake, Erik used his shirt to dry his feet," and "After
wading barefoot in the lake, Erik used his glasses to dry his feet." Of course,
people found the first sentence to be more sensible. The question is, Why? The
sentences are both grammatical. Coherent connected propositions can be formed
for both sentences. The concepts shirt and glasses are equally frequent and both
are unassociated with dry, so differences in abstract background knowledge are
not an issue. Furthermore, the participants experienced little or no difficulty in
understanding the first sentence. That is, in a separate experiment, people read
the first sentence as quickly as the control sentence, "After wading barefoot in
the lake, Erik used his towel to dry his feet." Thus, it is unlikely that people
needed to engage in a long string of inferences, such as shirts are made of cloth,
cloth is absorbent, absorbent things can be used to wipe up liquid, water is a
liquid, and so on, to understand the first sentence. Consider an explanation based
on the indexical hypothesis. In understanding a sentence, a reader indexes the
words to objects, or in the case of these well-known objects, to analog repre-
sentations of the objects such as perceptual symbols (Barsalou, 1999). The per-
ceptual symbols determine affordances—that is, how the reader can interact with
the objects. If the reader can use the affordances to envision a sequence of doable
actions, then the reader understands the sentence. It is easy to combine the
affordances of a shirt with the affordances of feet to accomplish drying; it is
difficult to do the same with the affordances of glasses. Similarly, in the compass
learning task, understanding "point the direction of travel arrows" requires con-
sideration of the particular affordances of those arrows, rather than features of
arrows and pointing in general.
There are clear precedents for the importance of indexical processes in lin-
guistics (e.g., Fillmore, 1997), cognitive linguistics (Duchan, Bruder, & Hewitt,
1995), and cognitive science (Ballard, Hayhoe, Pook, & Rao, 1997; Tanenhaus,
Spivey-Knowlton, Eberhard, & Sedivy, 1995). The indexical hypothesis is also
closely related to recently developed theories of embodied cognition. In brief,
this approach to cognition is based on the premise that cognition is a biological
phenomenon that coevolved with other bodily systems. Furthermore, cognition
depends on representations closely related to how the body works. For example,
Barsalou (1999) proposed that knowledge takes the form of perceptual symbol
systems. These perceptual symbols are derived from brain processes associated
with perception of the referents, and hence the perceptual symbols are analogically
related to their referents. This is in contrast to standard cognitive symbol systems
in which the symbols are arbitrarily related to their referents.
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 21

The indexical hypothesis results from a combination of Barsalou's (1999)

notion of a perceptual symbol and Glenberg's (1997) claim that meaning (of a
sentence, an object, or a situation) consists of the set of actions available to a
given individual in a given situation. This set of actions depends on the relation
between the affordances (derived from the environment or perceptual symbols),
the individual's goals, and the individual's learning history. According to Glen-
berg, affordances, goals, and learning history can be integrated, or meshed, in
the medium of bodily action. Mesh is a process of combing intrinsic constraints—
that is, constraints based on physics and biology—as interpreted by our perceptual
systems, rather than formal constraints such as those found in logic or mathe-
matics. For example, consider how we can manage to move our bodies, while
holding a compass, to point a direction of travel arrow. The sequence of actions
must respect built-in, or intrinsic, constraints on how bodies can interact with
the direction of travel arrows. Or consider the sentence, "Danielle used a 3.5-inch
piece of thread to scratch her back." There is nothing in the logic or statistics of
threads that disallow its use in back scratching. However, the intrinsic constraints
of physics and biology suggest that the affordances cannot be meshed: The thread
is not rigid enough to poke at one's back, and a person cannot tightly stretch a
short piece of thread (to make it rigid) while having her hands behind her back
in a way that allows scratching. Thus, a situation or text is perceived as coherent
and sensible if it can be used to guide or envision action, when that action
conforms to intrinsic constraints. Newton (1996) discussed how action can form
the basis of cognition in general, and MacWhinney (1998) discussed the relation
of embodiment to language acquisition.
In summary, consistent with the indexical hypothesis, we have demonstrated
that an important component of background knowledge is experiential and per-
ceptual, and thus background knowledge is not solely verbal and abstract. Per-
ceptual background knowledge results from indexing, and it is the source of
affordances. The affordances allow for faster reading, better comprehension, and
more skillful performance.

ACKNOWLEDGMENTS

Partial support for this research was provided by the University of Wisconsin
Graduate School Research Committee Grant 990288. We thank Adam Sadler for
his insightful and careful work on all phases of this research, Jennifer Dolland
and Brad Thiel for assisting in data collection, members of the Seminar in Cog-
nitive Psychology, Spring 1996-1997, for helping to develop the materials and
procedures, and Mina Johnson-Glenberg and Mike Rinck for their comments on
an earlier version of this article.
22 GLENBERG AND ROBERTSON

REFERENCES

Ballard, D. H., Hayhoe, M. M., Pook, P. K., & Rao, P. N. (1997). Deictic codes for the embodiment
of cognition. Behavioral and Brain Sciences, 20, 723-767.
Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577-660.
Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of physics
problems by experts and novices. Cognitive Science, 5, 121-152.
Chiesi, H. I., Spilich, G. J., & Voss, J. F. (1979). Acquisition of domain-related information in
relation to high and low domain knowledge. Journal of Verbal Learning and Verbal Behavior,
18, 275-290.
Clark, H. H. (1996). Using language. New York: Cambridge University Press.
Clark, H. H. (1997). Dogmas of understanding. Discourse Processes, 23, 567-598.
Duchan, J. F., Bruder, G. A., & Hewitt, L. E. (1995). Deixis in narrative: A cognitive science
perspective. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
Fillmore, C. J. (1997). Lectures on deixis. Stanford, CA: Center for the Study of Language and
Information.
Gibson, J. J. (1979). The ecological approach to visual perception. New York: Houghton Mifflin.
Glenberg, A. M. (1997). What memory is for. Behavioral and Brain Sciences, 20, 1-19.
Glenberg, A. M. (1998, August). Indexical understanding: An embodied approach to sentence
meaning. In N. Newton (Chair), Embodiment as a basis for cognition. Symposium conducted at
the meeting of the Cognitive Science Society, Madison, WI.
Glenberg, A. M., & Langston, W. E. (1992). Comprehension of illustrated text: Pictures help to build
mental models. Journal of Memory and Language, 31, 129-131.
Glenberg, A. M., & Robertson, D. A. (in press). Symbol grounding and meaning: A comparison of
high-dimensional and embodied theories of meaning. Journal of Memory and Language.
Glenberg, A. M., Robertson, D. A., Jansen, J. L., & Johnson-Glenberg, M. C. (in press). Not
propositions. Journal of Cognitive Systems Research.
Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration
model. Psychological Review, 95, 163-182.
Langston, W., Kramer, D. C , & Glenberg, A. M. (1998). The representation of space of mental
models derived from text. Memory & Cognition, 26, 246-262.
MacWhinney, B. (1998). The emergence of language from embodiment. In B. MacWhinney (Ed.),
The emergence of language (pp. 213-256). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Mani, K., & Johnson-Laird, P. N. (1982). The mental representation of spatial descriptions. Memory
& Cognition, 2, 181-187.
Mayer, R. E., Bove, W., Bryman, A., Mars, R., & Tapangco, L. (1996). When less is more: Meaningful
learning from visual and verbal summaries of science textbook lessons. Journal of Educational
Psychology, 88, 64-73.
Mayer, R. E., Steinhoff, K., Bower, G., & Mars, R. (1995). A generative theory of textbook design:
Using annotated illustrations to foster meaningful learning of science text. Educational Technology
Research and Development, 43, 31-43.
McNamara, D. S., & Kintsch, W. (1996). Learning from texts: Effects of prior knowledge and text
coherence. Discourse Processes, 22, 247-288.
McNeill, D. (1992). Hand and mind. Chicago: University of Chicago Press.
Mills, C. B., Diehl, V. A., Birkmire, D. P., & Mou, L.-C. (1995). Reading procedural texts: Effects
of purpose for reading and predictions of reading comprehension models. Discourse Processes,
20, 79-107.
Müller, H. M., & Kutas, M. (1996). What's in a name? Electrophysiological differences between
spoken nouns, proper nouns, and one's own name. NeuroReport, 8, 221-225.
Newton, N. (1996). Foundations of understanding. Philadelphia: Benjamins.
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 23

Rinck, M., Hähnel, A., Bower, G., & Glowalla, U. (1997). The metrics of spatial situation models.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 622-637.
Robertson, D. A., Gernsbacher, M. A., & Robertson, R. R. W. (1999). Advantage of referential
specificity, first mention, and suppression and enhancement affect memory for characters in
sentences. Manuscript submitted for publication.
Robertson, D. A., Glenberg, A. M., & Members of the 1997/98 Honors Seminar in Cognitive
Psychology. (1999). Grounding symbols and computing meaning: A supplement to Glenberg &
Robertson. Available: http://psych.wisc.edu/glenberg/supplement_jml_g&r.html
Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy, J. C. (1995). Integration of
visual and linguistic information in spoken language comprehension. Science, 268, 1632-1634.
Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and memory.
Psychological Bulletin, 123, 162-185.

APPENDIX A
Pretest and Posttest

1. A compass rose a. is on the base plate of a compass; b. indicates the

orientation of a map; c. indicates the orientation of a compass; d. rotates
when the 360 degree dial is turned.
2. When holding the compass in front of you, where do the direction of travel
arrows point? a. magnetic North; b. true North; c. towards where you are
facing; d. same direction as the orienting arrow.
3. Describe when orienting lines move. a. when the 360 degree dial is moved;
b. when the map is turned; c. never, when the compass is level; d. when the
compass rose is tilted.
4. The orienting arrow always points a. to the magnetic North; b. in the direction
you are traveling; c. to the North Pole; d. in a parallel direction to the
orienting lines.
5. A bearing is a. true North; b. a synonym for declination; c. the difference
between orienting lines and the orienting arrow; d. a direction.
6. The bearing is indicated by where the 360 degree dial crosses the a. direction
of travel arrows; b. left edge; c. orienting arrow; d. pointer.
7. Meridian lines a. are parallel lines on a map; b. never indicate declination;
c. show the bearing to a landmark; d. always run true North to true South.
8. Meridian lines a. are part of the base plate; b. adjusted by the magnetic
needle; c. may be associated with magnetic declination; d. are part of the
compass rose.
9. The 360 degree dial is turned a. to align the magnetic needle and the direction
of travel arrows; b. to adjust the direction of travel arrows; c. to determine
a bearing; d. to determine true North.
10. When the 360 degree dial is turned you affect a. the magnetic needle; b. the
orienting arrow; c. the compass rose; d. the direction of travel arrows.
11. Magnetic declination is a. the slant of a compass; b. the difference in degrees
between the compass rose and the direction of travel arrows; c. the difference
24 GLENBERG AND ROBERTSON

in degrees between the orienting arrow and the magnetic needle; d. the
difference in degrees between true North and magnetic North.
12. We adjust for magnetic declination by a. turning the 360 degree dial; b.
building the adjustment into the map; c. turning the orientation of the compass
rose; d. moving the magnetic needle to a specific direction.
13. In which direction does the magnetic needle point? a. true North; b. magnetic
North; c. depends on your declination; d. depends on your location and
altitude.
14. Which part of the base plate is used in conjunction with the map? a. the
direction of travel arrows; b. the orienting arrow; c. the left edge; d. the
middle front of the base plate.
15. How many mountains are indicated on our map? a. 19; b. 5; c. 27; d. 14.
16. Which is not part of the compass? a. magnetic needle; b. 360 degree dial;
c. compass rose; d. orienting lines.
17. Which direction corresponds to 270 degrees? a. North; b. East; c. South; d.
West.

APPENDIX B
Script

How to Use a Compass and Map to Name Landmarks

Here is the situation that you should imagine. You are hiking in the mountains
with a map and compass. You know your current location, and you can find that
location on the map. There are mountains all around you, but one is particularly
magnificent, and you want to know what it is called. The names of the mountains
are indicated on the map, but how do you know which mountain is which? You
are going to learn how to use a compass and map to name landmarks. First you
will learn about the parts of a compass and the parts of a map. Then, you will
be given a compass and map and asked to identify the names of mountains.

Parts of a Compass

Magnetic needle. When the compass is held level, the red part of the needle
always points to the magnetic North. If the compass is not level, the arrow will
not always point North.

360 degree dial, pointer, and bearing. Directions are divided into 360
degrees. Due North corresponds to 0 degrees, East is 90 degrees, South is 180
degrees, and West is 270 degrees. A compass has a movable round dial marked
from 0 to 360 degrees. This is the 360 degree dial. When you turn the dial, you
do not affect the magnetic needle, because it always points to the magnetic North.
 INDEXICAL UNDERSTANDING OF INSTRUCTIONS 25

The pointer is the small inverted triangle that is near the top of the 360 degree
dial. As the 360 degree dial is turned, the pointer points to different degrees.
A bearing is a direction in degrees. For example, a bearing of 38 degrees
means a direction of 38 degrees from North. The 360 degree dial and the pointer
will be used to determine bearings.

Orienting lines and orienting arrow. Underneath the magnetic needle are
a series of parallel orienting lines. One of the orienting lines is a thick red arrow.
This is the orienting arrow. The orienting lines (including the orienting arrow)
move when the 360 degree dial is turned.

Base plate, direction of travel arrows, and left edge. The main parts of
the compass (the needle, dial, and orienting lines) are attached to a rectangular
base plate. Two parts of the base plate are important. The first part consists of
two direction of travel arrows. They are the two, parallel red arrows imprinted
in the middle of the front of the base plate. When holding the compass in front
of you, the direction of travel arrows point toward where you are facing. Note:
The direction of travel arrows are different from the orienting arrow. The direction
of travel arrows do not move when the 360 degree dial is moved, but the orienting
arrow does move.
The other part of the base plate that will play an important role is the left
edge and the small cross near the back of the left edge, in particular. The left
edge and the small cross will be used in conjunction with the map.

Parts of the Map

Of course maps can have many features and symbols. Here we will be using a
very simple map that has only a few parts of interest.

Compass rose. The compass rose consists of the lines (near the top of the
map) labeled N, S, E, and W. The compass rose indicates how the map is oriented,
that is, which part of the map corresponds to true North in the environment. As
with most maps, the map that we will be using has true North on the top of the
map.

Meridian lines and magnetic declination. True North is at the North Pole.
Magnetic needles do not point to the North Pole, however, they point to the
magnetic North. The difference in degrees between true North and magnetic
North is called magnetic declination. It is important to correct for this difference
in degrees. We correct for declination by building the correction into the map
by slanting the meridian lines. Meridian lines are the parallel lines that typically
run vertically (true North to true South) on a map. On our map, however, the
meridian lines are slanted to run from the magnetic North to the magnetic South.
26 GLENBERG AND ROBERTSON

This way, North indicated by the meridian lines corresponds to North as indicated
by the compass needle.

Symbols. Maps have symbols that correspond to roads, cities, lakes, moun-
tains, etc. Our map will have two types of symbols. One indicates your current
location, the cross (*). The other type of symbol (a triangle with a circle in it)
corresponds to a mountain peak. There are 19 symbols for mountain peaks in-
dicated on the map.