Developing an Affective Evaluation Tool - Katherine Isbister
Developing an Affective Evaluation Tool - Katherine Isbister
Developing an Affective Evaluation Tool - Katherine Isbister
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
The Sensual <strong>Evaluation</strong> Instrument:<br />
<strong>Developing</strong> <strong>an</strong> <strong>Affective</strong> <strong>Evaluation</strong> <strong>Tool</strong><br />
<strong>Katherine</strong> <strong>Isbister</strong> 1 , Kia Höök 2 , Michael Sharp 1 , Jarmo Laaksolahti 2<br />
1 Rensselaer Polytechnic Institute<br />
2 SICS<br />
110 8 th Street, Sage 4208<br />
Box 1263<br />
Troy, New York 12180 U.S.A.<br />
S-164 28 Kista, Sweden<br />
{isbisk,sharpd}@rpi.edu<br />
{kia,jarmo}@sics.se<br />
+1 518 276 8262<br />
ABSTRACT<br />
In this paper we describe the development <strong>an</strong>d initial testing<br />
of a tool for self-assessment of affect while interacting with<br />
computer systems: the Sensual <strong>Evaluation</strong> Instrument. We<br />
discuss our research approach within the context of existing<br />
affective <strong>an</strong>d HCI theory, <strong>an</strong>d describe stages of evolution<br />
of the tool, <strong>an</strong>d initial testing of its effectiveness.<br />
Author Keywords<br />
<strong>Affective</strong> interaction, User-centered design, ambiguity,<br />
gesture-based interaction, embodiment.<br />
ACM Classification Keywords<br />
H.5.2 User interfaces<br />
INTRODUCTION<br />
Practitioners in the CHI community have become<br />
increasingly convinced of the import<strong>an</strong>ce of affect—both in<br />
terms of designing experiences for users which take it into<br />
account [18, 16], <strong>an</strong>d also in terms of developing measures<br />
for user satisfaction with such systems [10].<br />
There are various modes of approach to measuring affect.<br />
Traditionally, affect has been ascertained in two primary<br />
ways: using questionnaires administered after <strong>an</strong><br />
experience, which ask the user to rate his/her feelings about<br />
what occurred, <strong>an</strong>d <strong>an</strong>alysis of videotaped sessions with<br />
users that typically combine interpretation of think-aloud<br />
commentary with deciphering of other cues of emotion<br />
(smiling, gestures <strong>an</strong>d the like) to develop <strong>an</strong> impression of<br />
user’s affective reactions. In recent years, additional tools<br />
based upon biometrics have evolved—measuring galv<strong>an</strong>ic<br />
skin response, detecting small movements of the muscles of<br />
the face, tracking pressure on the mouse [19]. These signals<br />
Permission to make digital or hard copies of all or part of this work for<br />
personal or classroom use is gr<strong>an</strong>ted without fee provided that copies are<br />
not made or distributed for profit or commercial adv<strong>an</strong>tage <strong>an</strong>d that copies<br />
bear this notice <strong>an</strong>d the full citation on the first page. To copy otherwise,<br />
or republish, to post on servers or to redistribute to lists, requires prior<br />
specific permission <strong>an</strong>d/or a fee.<br />
CHI 2006, April 22-27, 2006, Montréal, Québec, C<strong>an</strong>ada.<br />
Copyright 2006 ACM 1-59593-178-3/06/0004...$5.00.<br />
provide a complex trace of the user’s affect which m<strong>an</strong>y in<br />
our community (<strong>an</strong>d in the psychological <strong>an</strong>d physiological<br />
communities) are still working to underst<strong>an</strong>d adequately.<br />
Our own efforts in developing <strong>an</strong> affective assessment tool<br />
have been shaped by our past work in underst<strong>an</strong>ding <strong>an</strong>d<br />
enh<strong>an</strong>cing nonverbal communication, our commitment to<br />
user-centered design, <strong>an</strong>d our interest in tr<strong>an</strong>s-cultural<br />
methods to support global design practice. Our research<br />
approach arises from several baseline principles: 1) a belief<br />
that freeing the person giving us affective feedback from<br />
words, <strong>an</strong>d incorporating the use of their body to give<br />
feedback, will be beneficial; 2) the belief that self-report is<br />
a valuable practice, as part of a user-centered design<br />
approach; 3) <strong>an</strong> interest in creating a flexible <strong>an</strong>d portable<br />
tool that c<strong>an</strong> be widely used, that could tr<strong>an</strong>scend l<strong>an</strong>guage<br />
<strong>an</strong>d culture; 4) the desire to create <strong>an</strong> expressive, openended<br />
tool that does not remove all ambiguity from<br />
communication by the user but rather offers a r<strong>an</strong>ge of<br />
expressive possibilities.<br />
Our development process was a user-centered one,<br />
conducted in several stages, with input along the way: <strong>an</strong><br />
initial brainstorming stage; early, open-ended concept<br />
testing; refinement of the research instrument; <strong>an</strong>d openended<br />
testing of the prototype tool.<br />
CONTEXT<br />
This project emerged from a Europe<strong>an</strong> Network of<br />
Excellence initiative called Humaine (http://emotionresearch.net/),<br />
focused on the study of affective systems.<br />
Our work group in the project is dedicated to exploring<br />
ways <strong>an</strong>d me<strong>an</strong>s for evaluating a system’s affective impact<br />
on users. As we gathered sources <strong>an</strong>d learned about existing<br />
methods, we found that there was still much work to do in<br />
refining methods for eliciting <strong>an</strong>d underst<strong>an</strong>ding user affect.<br />
In particular, there seemed to be a gap in terms of easy to<br />
use self-report measures. System designers were either<br />
relying on questionnaires given after the interaction was<br />
complete, were laboriously coding data from videotape of<br />
sessions, or were using biometric equipment that requires<br />
relatively complex <strong>an</strong>alysis in order to deliver reliable<br />
affective data. Was it possible to create a light-weight, selfreport<br />
measure that could give reasonable in-line feedback<br />
1163
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
from the user as the experience was unfolding We felt this<br />
was worth exploring, especially insofar as it could enh<strong>an</strong>ce<br />
the design process by enabling more rapid test-<strong>an</strong>d-iterate<br />
cycles based upon affective responses to a system in<br />
development.<br />
It is import<strong>an</strong>t to reiterate here that we w<strong>an</strong>ted this tool to<br />
allow flexibility for users to adapt it to their own style of<br />
expressing affect, toward creating a dialog between<br />
designers <strong>an</strong>d users that enh<strong>an</strong>ces design. While we did aim<br />
for some level of consistency in response to the<br />
instrument’s qualities, we were not solely focused on<br />
consistent use patterns nor would we claim that this is a<br />
highly accurate tool for measuring affect. In this regard the<br />
aims of the project differ radically from other self-report<br />
measures, <strong>an</strong>d our instrument should be viewed as<br />
complementary to rather th<strong>an</strong> as <strong>an</strong> alternative to other<br />
measures.<br />
In Support of a Body-based, Nonverbal Approach<br />
Neurophysiologists <strong>an</strong>d psychologists have in recent years<br />
proposed that our brains, rather th<strong>an</strong> operating in a wholly<br />
logical, conscious verbal m<strong>an</strong>ner, actually process<br />
information <strong>an</strong>d make decisions using various layers<br />
working in parallel, complementary ways. They have<br />
demonstrated, for example, that we c<strong>an</strong> learn something<br />
new <strong>an</strong>d ‘intuitively’ put it into action before we are able to<br />
consciously verbalize it [15]. <strong>Affective</strong> processing, in<br />
particular, has been shown to occur at levels other th<strong>an</strong> the<br />
cognitive/word-oriented level of the brain (e.g. the primal<br />
nature of fear: see 14). Emotions are experienced by both<br />
body <strong>an</strong>d mind. Often, they are evoked by sub-symbolic<br />
stimuli, such as colors, shapes, gestures, or music.<br />
If we rely on the verbal ch<strong>an</strong>nel for self-report of affect<br />
during interaction with a system, we may be missing out on<br />
much of this information, as it is filtered through the<br />
person’s verbal system. The authors [6, 11] have explored<br />
the role of nonverbal expression in interface in the past, <strong>an</strong>d<br />
have found the body to be <strong>an</strong> import<strong>an</strong>t ch<strong>an</strong>nel for<br />
communicating affect <strong>an</strong>d other qualities. We decided that<br />
our affective instrument would be sensual <strong>an</strong>d physical in<br />
nature, to allow us to access this part of a person’s<br />
experience more directly. In addition, we hoped that<br />
avoiding verbalization would make it more likely that users<br />
could give in-line feedback without their task being as<br />
disrupted as if they had to explicitly verbalize their current<br />
affect.<br />
User-centered Design Values<br />
Höök in particular has had a long-st<strong>an</strong>ding commitment to<br />
user-centered design practice, <strong>an</strong>d finding ways to bring<br />
users into the design process earlier in development cycles<br />
to strengthen final design outcomes. To the extent that a<br />
usability practice is intuitive <strong>an</strong>d simple to operate <strong>an</strong>d<br />
<strong>an</strong>alyze, it becomes more likely to be adopted for use<br />
during development, especially during early stages when<br />
signific<strong>an</strong>t ch<strong>an</strong>ge is still possible. Höök’s experience has<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
been that tools which have a hope of being used well in<br />
early design practice need to be flexible—allowing for<br />
vari<strong>an</strong>ce in user’s approaches <strong>an</strong>d mental models of what<br />
they are doing, <strong>an</strong>d r<strong>an</strong>ges of mode of expression. A<br />
modicum of ambiguity in the tool at this stage is good<br />
rather th<strong>an</strong> bad, as it allows users to co-define the me<strong>an</strong>ing<br />
of their feedback with the designer, to reach a richer shared<br />
picture of what is going on [7]. This is especially<br />
appropriate for a tool to measure affect—models of emotion<br />
from the realm of psychology are not always fully<br />
descriptive of (or useful in the context of) measuring<br />
affective reaction to a system. Some emotions rarely come<br />
into play when engaging with computer systems, <strong>an</strong>d the<br />
ebb <strong>an</strong>d flow of reaction to a system is not necessarily the<br />
same as that which occurs in face to face interaction among<br />
people. Part of the hidden work of this project, then, was<br />
learning about which emotions are truly relev<strong>an</strong>t in the<br />
context of evolving successful interaction with computer<br />
systems, <strong>an</strong>d offering ways to express those emotions with<br />
our research instrument.<br />
Practical Considerations: A Portable, Flexible, Crossculturally<br />
Valid Instrument<br />
By avoiding verbalization, we also realized that we had a<br />
ch<strong>an</strong>ce of developing a less culturally dependent research<br />
instrument. Questionnaires must normally be tr<strong>an</strong>slated <strong>an</strong>d<br />
counter-tr<strong>an</strong>slated to insure that they are truly measuring<br />
the target emotion, even though there is strong evidence for<br />
tr<strong>an</strong>s-cultural similarities in basic emotional response.<br />
Perhaps a nonverbal, body-based approach could tap more<br />
directly into these shared responses, saving designers time<br />
<strong>an</strong>d energy, <strong>an</strong>d again, leading to greater likelihood of early<br />
user participation in design. We were inspired by the work<br />
of Liz S<strong>an</strong>ders [21], Bill Gaver [7, 8, 4], <strong>an</strong>d others, in<br />
conducting inquiries that move freely beyond the lab <strong>an</strong>d<br />
into various cultural settings. Thus from the beginning, one<br />
end goal of our project was the creation of a ‘kit’ that could<br />
be easily shipped to others around the world for use.<br />
And Finally, in Support of Fun<br />
We subscribe to the notion of ‘Functional Aesthetics’ [4],<br />
<strong>an</strong>d would like to extend this notion to the design of<br />
methodologies as well as end products. Particularly in the<br />
study of affect, we feel the experience of the feedbackgiving<br />
should be pleas<strong>an</strong>t in <strong>an</strong>d of itself.<br />
APPROACH<br />
We developed the prototype of the Sensual <strong>Evaluation</strong><br />
Instrument after some initial exploratory study of the<br />
possibility space.<br />
Stage One: Exploratory Design <strong>an</strong>d Iteration<br />
We beg<strong>an</strong> by exploring work-to-date on nonverbal selfreport<br />
measures. There was some history of the<br />
st<strong>an</strong>dardization <strong>an</strong>d use of nonverbal scales in psychology<br />
(e.g. PONS (the Profile of Nonverbal Sensitivity) [3], from<br />
which we could draw lessons.<br />
1164
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
Most work done on nonverbal systems of evaluation has<br />
involved <strong>an</strong>thropomorphic imagery (likenesses of hum<strong>an</strong><br />
faces <strong>an</strong>d/or bodies). For example, researchers who work<br />
with children have established the value <strong>an</strong>d reliability of<br />
face-based Likert scales for determining emotional<br />
responses to systems <strong>an</strong>d situations (e.g. Wong <strong>an</strong>d Baker’s<br />
work on children’s subjective evaluation of pain—see<br />
Figure 1) [24].<br />
Figure 1. A pain scale used to help children indicate their level<br />
of discomfort.<br />
There are also popular uses of nonverbal affective scales<br />
(thumbs-up <strong>an</strong>d down movie ratings, the ‘little m<strong>an</strong>’ on the<br />
S<strong>an</strong> Fr<strong>an</strong>cisco Chronicle movie review page<br />
(http://www.sfgate.com/eguide/movies/reviews/ see Figure<br />
2), indicating that calibration <strong>an</strong>d use of nonverbal scales is<br />
possible <strong>an</strong>d appealing in everyday contexts.<br />
Figure 2. The S<strong>an</strong> Fr<strong>an</strong>cisco Chronicle’s movie review system<br />
uses facial expression <strong>an</strong>d body posture to indicate a movie’s<br />
quality.<br />
Finally, there has been some work in the product design<br />
community on mapping product qualities to affective<br />
reactions, for example this facial wheel used by Wensveen<br />
<strong>an</strong>d colleagues in developing <strong>an</strong> affectively appropriate<br />
alarm clock (Figure 3, see [22,23]).<br />
Our wish in this project was to move away from discrete,<br />
pre-defined emotional labels such as faces, figures, or<br />
names, <strong>an</strong>d move toward some form of nonverbal code that<br />
allowed more open-ended interpretation but that still<br />
evoked emotion without explicitly representing the hum<strong>an</strong><br />
form. We also hoped to extend the sensory experience of<br />
this instrument beyond the purely visual.<br />
Figure 3. Facial expressions arr<strong>an</strong>ged along the axes of arousal<br />
<strong>an</strong>d valence, used to help product designers gauge user<br />
emotions about designs.<br />
There is some nonrepresentational work on sensing emotion<br />
in alternate sensory ch<strong>an</strong>nels. For example in his book<br />
Sentics, Clynes describes characteristic movement patterns<br />
on a touch pad when users are asked to ‘perform’ a<br />
particular emotion with their finger [1]. Fagerberg et al.’s<br />
work on eMoto, a non-representational system for adding<br />
affective content to SMS messages on mobile phones, is <strong>an</strong><br />
example of the use of gesture <strong>an</strong>d touch to generate <strong>an</strong><br />
emotional response [6]. Product designers know that<br />
surface materials <strong>an</strong>d their tactile qualities c<strong>an</strong> profoundly<br />
impact users’ emotional response to products [9], but there<br />
has been little systematic work done up to know to asses the<br />
specific emotional effects of various materials.<br />
Method<br />
Exploration One: Personal Home Objects<br />
To begin the project, we decided to apply one traditional<br />
model of emotion to everyday household objects, to see<br />
what sort of physical properties caused them to be arrayed<br />
in which places in the emotional spectrum. We worked<br />
from the Russell circle [20], which arrays emotions along<br />
two axes: arousal <strong>an</strong>d valence. A high arousal, high valence<br />
emotion such as ecstatic joy would be located in the upper<br />
right quadr<strong>an</strong>t, for example, whereas a low arousal, low<br />
valence emotion such as depression would be located in the<br />
lower left quadr<strong>an</strong>t. We had several researchers bring in<br />
objects that had emotional me<strong>an</strong>ing for them, <strong>an</strong>d array<br />
them upon a circle projected onto the floor (see Figure 4)<br />
Findings<br />
This early experiment revealed m<strong>an</strong>y properties of objects<br />
that help to evoke emotion: material, shape, ‘give’ in the<br />
h<strong>an</strong>d, color, size, as well as memories associated with the<br />
object.<br />
1165
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
Figure 4. Household objects arrayed on the Russell circle<br />
Exploration Two: Colored Cylinders<br />
We decided to isolate one dimension of the m<strong>an</strong>y object<br />
properties we found in step one, <strong>an</strong>d test how people would<br />
use this dimension to convey their emotions while<br />
interacting with a system. We chose color as our first<br />
variable, as color had strong connotations in the first test<br />
<strong>an</strong>d with other projects by Höök <strong>an</strong>d colleagues. We crafted<br />
colored objects that could be held in the h<strong>an</strong>d, <strong>an</strong>d which<br />
were similar in weight, texture <strong>an</strong>d other properties,<br />
differing only in color (see Figure 5). Then we brought in<br />
several people not familiar with our research, <strong>an</strong>d asked<br />
them to do three things:<br />
1. Create their own taxonomy of the objects, using the<br />
projected Russell circle (as we did in the initial test).<br />
2. Interact with a computer game that evoked strong<br />
emotions, <strong>an</strong>d use the objects to convey how they felt.<br />
3. Discuss with us how it was to use the objects to convey<br />
emotion.<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
everyday sense, so they hadn’t really pl<strong>an</strong>ned how they<br />
would express them.<br />
Based upon these results, we made two key design<br />
decisions:<br />
1. We would move to a more biomorphic framework upon<br />
which to vary the objects, one that we hoped was less<br />
subject to narrative overlays. We drew from work by<br />
Disney <strong>an</strong>imators in crafting less explicitly<br />
<strong>an</strong>thropomorphic forms that still m<strong>an</strong>aged to strong convey<br />
emotion (see Figure 6). We were also influenced by prior<br />
investigation into qualities of movement <strong>an</strong>d its <strong>an</strong>alysis<br />
(e.g. the work of Lab<strong>an</strong>; see [6] for further discussion of<br />
movement <strong>an</strong>alysis <strong>an</strong>d expression of affect.)<br />
2. We would take note of the emotions that users expressed<br />
during this test, <strong>an</strong>d incorporate system-use-based emotions<br />
into our subsequent design thinking. How might one<br />
express mild frustration, confusion, <strong>an</strong>d the like These<br />
were the sorts of feedback we w<strong>an</strong>ted to support with our<br />
instrument.<br />
Figure 6. The famous Disney flour sack, <strong>an</strong> illustration of<br />
emotions conveyed by <strong>an</strong> object despite the absence of a<br />
clearly articulated hum<strong>an</strong> figure [12].<br />
Figure 5. Early sensual evaluation objects arrayed on the<br />
Russell emotional circle.<br />
Findings<br />
Our testing revealed that some people used the colored<br />
objects to create a personal taxonomy of affect, but others<br />
found that their own narrative connotations for colors (e.g.<br />
red <strong>an</strong>d green remind me of Christmas) got in the way of<br />
crafting <strong>an</strong>d re-using a personal affective taxonomy during<br />
the game play. We found that users were experiencing<br />
emotions that didn’t match neatly to their initial taxonomy,<br />
<strong>an</strong>d were not sure how to indicate them—such as mild<br />
frustration, confusion, <strong>an</strong>ticipation, <strong>an</strong>d the like. These were<br />
not what they typically thought of as ‘emotions’ in the<br />
Iteration<br />
At this point we solicited the support of a professional<br />
sculptor who had experience in crafting biomorphic forms.<br />
Rainey Straus (see www.raineystraus.com for more<br />
examples of her work) crafted a set of objects that had<br />
biomorphic qualities, working from ongoing discussions<br />
with us, <strong>an</strong>d from a list of emotions we provided that was<br />
based upon our exploratory studies. The initial list was<br />
confusion, frustration, fear, happiness, surprise, satisfaction,<br />
contentment, stress, <strong>an</strong>d flow. She crafted a total of eight<br />
objects, not necessarily me<strong>an</strong>t to be a one-to-one mapping<br />
to these emotions, but rather a successful set of tools for<br />
evoking/expressing this r<strong>an</strong>ge of emotions (see Figure 7 <strong>an</strong>d<br />
8).<br />
1166
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
Figure 7. The Sensual <strong>Evaluation</strong> Instrument objects.<br />
Straus created the objects in clay, then we had them cast in<br />
plastic in order to have a durable surface <strong>an</strong>d multiples for<br />
use in different lab settings. Internally, we assigned names<br />
to the objects to make it easier to code data, as follows<br />
(follow figure 7): back row—spiky, pseudopod; next row—<br />
<strong>an</strong>teater, bubbly; next row—stone, doubleball, ball; front—<br />
barba papa (named after a figure from a French <strong>an</strong>imated<br />
cartoon popular in Sweden). We did not use these or <strong>an</strong>y<br />
other names for the objects with particip<strong>an</strong>ts—these names<br />
are introduced only to aid you, the reader, in following<br />
discussion of object use in this paper.<br />
Figure 8. An alternate view to give the reader additional visual<br />
information about the contours of the objects.<br />
Stage Two: Open-ended testing of prototype objects<br />
With the cast objects in h<strong>an</strong>d, we designed a study to assess<br />
whether this new instrument would allow people to provide<br />
me<strong>an</strong>ingful affective feedback while they engaged in<br />
interaction with a computer system. The aim of the study<br />
was both to explore use potential for the object set, <strong>an</strong>d also<br />
to engage the students <strong>an</strong>d faculty in a discussion of the act<br />
of using the objects, <strong>an</strong>d potential interactions/innovations<br />
toward a refined design.<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
• Expl<strong>an</strong>ation of research purpose <strong>an</strong>d orientation to the<br />
objects. In this phase, the particip<strong>an</strong>t was shown the<br />
objects <strong>an</strong>d encouraged to h<strong>an</strong>dle them, <strong>an</strong>d the<br />
experimenter described the purpose of the research <strong>an</strong>d<br />
emphasized that <strong>an</strong>y type of usage of the objects to<br />
convey affective state was fine (multiple objects at once,<br />
movement of objects, <strong>an</strong>d so forth).<br />
• Use of objects with a subset of the IAPS (International<br />
<strong>Affective</strong> Picture Set). In this phase, the particip<strong>an</strong>t was<br />
shown a series of images taken from the IAPS [13], <strong>an</strong>d<br />
asked to use the SEI objects to indicate affective response<br />
to each picture. (See the Findings section for examples of<br />
the images used). The images were selected because their<br />
arousal/valence scores mapped roughly to the feelings<br />
that the artist initially intended to convey with the<br />
objects.<br />
• Use of objects with a computer game. In this phase, the<br />
particip<strong>an</strong>t played through the first puzzle in a PC<br />
adventure game, The Curse of Monkey Isl<strong>an</strong>d, <strong>an</strong>d was<br />
instructed to use the objects to indicate affect during play.<br />
The experimenter was present during this phase, <strong>an</strong>d was<br />
available to offer hints/tips on using the game during<br />
play.<br />
• Use of objects during a chat. After the game, the<br />
particip<strong>an</strong>t was asked to chat using AIM inst<strong>an</strong>t<br />
messaging with the experimenter’s assist<strong>an</strong>t, to discuss<br />
how it was to play the game. The experimenter left the<br />
room during this portion, after instructing the particip<strong>an</strong>t<br />
to use the objects to indicate affect while chatting.<br />
• A discussion of how it was to use the objects to give<br />
affective feedback. At the end, the experimenter returned<br />
<strong>an</strong>d walked the particip<strong>an</strong>t through a series of questions<br />
about what it was like to use the objects, including<br />
solicitation of suggestions for improvement of the SEI.<br />
Questions included: What was it like to use the objects to<br />
express emotion Did you find that you had a consistent<br />
set of mappings How hard or easy was it to use them<br />
Any other thoughts Suggestions for ch<strong>an</strong>ges or<br />
alternatives<br />
The sessions were digitally recorded (with particip<strong>an</strong>ts’<br />
consent), using a picture within picture format, so that we<br />
could later <strong>an</strong>alyze usage in conjunction with what the<br />
particip<strong>an</strong>t was seeing on-screen (see Figure 9).<br />
Method<br />
In this stage, we took 12 individual particip<strong>an</strong>ts (10 male, 2<br />
female) through <strong>an</strong> hour-long session, which consisted of<br />
five phases:<br />
1167
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
Figure 9. Particip<strong>an</strong>ts were recorded so that we could see how<br />
they m<strong>an</strong>ipulated the objects, as well as what they were seeing.<br />
Analysis <strong>an</strong>d Findings<br />
Calibration with Images<br />
Considering the images arrayed by the primary dimensions<br />
used in the IAPS (arousal <strong>an</strong>d valence, see Figure 10 for<br />
sample images), we found that there were some visible<br />
trends in the use of objects.<br />
Although there were definitely individual differences in<br />
particip<strong>an</strong>t reactions to the images (for example, one<br />
student liked the shark image <strong>an</strong>d felt very calm <strong>an</strong>d<br />
positive about it as he loves sharks), looking at the object<br />
use reveals some relation to the valence <strong>an</strong>d arousal metrics<br />
compiled for the IAPS (see figures 11 <strong>an</strong>d 12).<br />
Figure 10. Sample high arousal/low valence (above) <strong>an</strong>d lower<br />
arousal/higher valence (below) images selected from the IAPS.<br />
If we consider the IAPS images arr<strong>an</strong>ged from most<br />
negative to most positive valence, we c<strong>an</strong> see that the<br />
objects with the sharp edges tended to be used in<br />
association with the negative valence imagery (bottom of<br />
Figure 11). Particip<strong>an</strong>ts used spiky <strong>an</strong>d <strong>an</strong>teater more<br />
frequently in response to these images. Objects with smooth<br />
edges <strong>an</strong>d few protrusions (ball, stone, barba papa, bubbly)<br />
tended to be used in conjunction with images with more<br />
positive valence (top of Figure 11).<br />
Figure 11. Objects used in response to the IAPS images, arr<strong>an</strong>ged by typical image valence response.<br />
1168
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
Figure 12. Objects used in response to the IAPS images, arr<strong>an</strong>ged by typical image arousal response.<br />
The trends are less clear when it comes to the arousal level<br />
associated with the IAPS images—those with the highest<br />
arousal level as well as negative valence: a gun pointed<br />
toward the viewer or a dog barking at the viewer, were<br />
associated with the sharper edged objects. Mid-level arousal<br />
<strong>an</strong>d positive valence images: the bride with arms out, the<br />
bride with child, were associated with rounded objects.<br />
There was not a clear pattern for low-arousal images (see<br />
Figure 12).<br />
Individual Taxonomies <strong>an</strong>d Use Patterns<br />
in post-discussion. Spiky <strong>an</strong>d <strong>an</strong>teater both got fear <strong>an</strong>d<br />
<strong>an</strong>ger attributions from most particip<strong>an</strong>ts (though one<br />
particip<strong>an</strong>t felt <strong>an</strong>teater should me<strong>an</strong> interest). Barba papa<br />
was often used to indicate humor. Bubbly was used for both<br />
humor <strong>an</strong>d for confusion <strong>an</strong>d frustration. M<strong>an</strong>y particip<strong>an</strong>ts<br />
commented that the bubbly form seemed chaotic <strong>an</strong>d<br />
suggested confusion or indeterminacy. Pseudopod got some<br />
similar reactions—one particip<strong>an</strong>t used this to indicate that<br />
he had a goal in mind but it wasn’t wholly resolved—a<br />
feeling of directed <strong>an</strong>ticipation.<br />
Despite these generally similar tendencies in associating the<br />
objects with particular affective states, particip<strong>an</strong>ts used<br />
them in action in quite different ways.<br />
Figure 13. The Sensual <strong>Evaluation</strong> Instrument objects, for<br />
reference when reading this section.<br />
There were also patterns in usage during the session, <strong>an</strong>d in<br />
discussion afterward that echoed <strong>an</strong>d also extended what<br />
was seen in the object calibration exercise outside the<br />
valence/arousal model. Particip<strong>an</strong>ts tended to describe the<br />
smooth, rounded objects with fewer protrusions as happy,<br />
or calm, or fun in the post-discussion, <strong>an</strong>d to use them when<br />
in such situations during play or chat. The ball was<br />
frequently given a higher energy positive attribution th<strong>an</strong><br />
the stone, which was described as bored or as zenlike calm<br />
Figure 14. A particip<strong>an</strong>t using multiple objects.<br />
This particip<strong>an</strong>t (Figure 14) used multiple objects arrayed in<br />
a tri<strong>an</strong>gular formation to express his reaction. His use of the<br />
objects was limited to those that were most rounded in form<br />
(ball, stone, doubleball, barba papa), <strong>an</strong>d he said he<br />
probably would not use the sharper forms (such as spiky)<br />
unless he saw the ‘blue screen of death’.<br />
1169
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
Use of objects during the session (total time in the game<br />
<strong>an</strong>d chat tasks was typically ~30 minutes total) r<strong>an</strong>ged from<br />
4 to 53 times. Particip<strong>an</strong>ts used the objects to express<br />
emotion both when playing the game with the experimenter<br />
present, <strong>an</strong>d when inst<strong>an</strong>t message chatting with the<br />
experimenter not present in the room, although usage was<br />
less frequent in the latter case (see Figure 17).<br />
Figure 15. A particip<strong>an</strong>t stacking objects.<br />
This particip<strong>an</strong>t (Figure 15) stacked objects in various ways<br />
during his session. He used every object except spiky in his<br />
session.<br />
Figure 18. Pattern of multiple object usage inst<strong>an</strong>ces, per<br />
particip<strong>an</strong>t.<br />
Use of several objects at once varied from person to person,<br />
but overall, multiple use accounted for about 21% of total<br />
inst<strong>an</strong>ces of object selection (see Figure 18).<br />
Figure 16. A particip<strong>an</strong>t gesturing with <strong>an</strong> object.<br />
This particip<strong>an</strong>t (Figure 16) shook the spiked object in<br />
response to a lag problem in the game during play (<strong>an</strong>d at<br />
other times). He tended to hold the objects in his h<strong>an</strong>ds<br />
when they were ‘active’. He made frequent <strong>an</strong>d dramatic<br />
use of the objects including broad gestures—over 50<br />
inst<strong>an</strong>ces in the 30-minute session.<br />
Figure 17. Pattern of object usage inst<strong>an</strong>ces in the two active<br />
tasks.<br />
Anecdotal Discussion<br />
Particip<strong>an</strong>ts gave us both positive <strong>an</strong>d negative feedback<br />
about this method of providing affective information about<br />
using a system. Positive comments included:<br />
“I’d do this over surveys <strong>an</strong>y day. I hate surveys!” (The<br />
particip<strong>an</strong>t explained that he liked being able to give<br />
feedback during as opposed to after, <strong>an</strong>d that he liked<br />
having more th<strong>an</strong> one object <strong>an</strong>d being able to ch<strong>an</strong>ge their<br />
orientation).<br />
“They were fun to play with, just to roll them around in<br />
your h<strong>an</strong>d.” (This particip<strong>an</strong>t had <strong>an</strong> object in h<strong>an</strong>d almost<br />
const<strong>an</strong>tly during the session).<br />
“They were novel <strong>an</strong>d interesting.”<br />
“I like the concept.”<br />
“I think a survey form would be as good, but you’d have to<br />
stop <strong>an</strong>d think about it.”<br />
“I felt like I was very involved.”<br />
“It was fun. Much better th<strong>an</strong> filling out a<br />
survey…feedback is more immediate.”<br />
“In retrospect I suppose they helped me to see when<br />
something wasn’t working because I was like hmm which<br />
object should I use to convey why this won’t work. By the<br />
same token, I knew that I had a good strategy when I was<br />
able to keep the happy object out <strong>an</strong>d I could see the game<br />
was adv<strong>an</strong>cing.”<br />
1170
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
Negative comments:<br />
“I didn’t feel like I got a lot of leverage out of that one<br />
(doubleball).” (Doubleball <strong>an</strong>d pseudopod were objects that<br />
had less reson<strong>an</strong>ce for several particip<strong>an</strong>ts).<br />
“It felt like there could have been more… like vastness. I<br />
would make something larger th<strong>an</strong> the others.” (Several<br />
particip<strong>an</strong>ts indicated ‘missing’ emotions, <strong>an</strong>d a desire for<br />
more objects. Missing emotions included: mel<strong>an</strong>choly, a<br />
sense of accomplishment, hatefulness beyond <strong>an</strong>ger,<br />
disingenuousness, <strong>an</strong>d ever-so-slight frustration).<br />
“This is ambiguous.” (Some expressed concern that the<br />
results would not be clear or definitive enough to use as<br />
data).<br />
Particip<strong>an</strong>ts also made some very interesting suggestions<br />
for evolving the instrument, which included:<br />
• adding the ability to squeeze or shape the objects<br />
somehow with one’s own grip, to further indicate<br />
emotion.<br />
• adding droopy objects to indicate despair or low energy<br />
level.<br />
• introducing scale <strong>an</strong>d texture as variables.<br />
One other factor for consideration in the object use was the<br />
tendency for particip<strong>an</strong>ts sometimes to match the shape of<br />
the object with the shape of the thing they were reacting to.<br />
This was especially so with the IAPS image calibration<br />
(some subjects used the pseudopod or barba papa in<br />
response to a gun image, or the doubleball or bubbly<br />
because it looked like the rounded mountains, <strong>an</strong>d so forth).<br />
This tendency seemed to become less prevalent with the<br />
active system feedback sessions, but is a concern in further<br />
evolution of the shapes themselves.<br />
GENERAL DISCUSSION OF RESULTS<br />
This testing exercise seems to indicate that we are on the<br />
right track in our design of the Sensual <strong>Evaluation</strong><br />
Instrument. Specifically:<br />
• It seems to be fun <strong>an</strong>d engaging for people, in part<br />
because it makes use of the body <strong>an</strong>d sense of touch, <strong>an</strong>d<br />
also because it allows for flexibility in response.<br />
• There are indications (through use patterns <strong>an</strong>d comments<br />
afterward) that this is a feasible way to give in-process<br />
feedback without too much disruption, as we had hoped.<br />
• Use patterns <strong>an</strong>d verbal descriptions seem to indicate that<br />
this could be evolved into a reasonably consistent<br />
instrument in terms of general affective dimensions,<br />
while maintaining a flexibility of expression for<br />
particip<strong>an</strong>ts.<br />
We believe this approach holds promise <strong>an</strong>d are pursuing it<br />
further.<br />
NEXT STEPS<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
Effectiveness for Designers<br />
The first <strong>an</strong>d most import<strong>an</strong>t follow-up step is to establish<br />
that this instrument c<strong>an</strong> provide helpful feedback to a<br />
designer of a system. We are pl<strong>an</strong>ning to conduct user<br />
feedback sessions with designs-in-progress in the coming<br />
year, <strong>an</strong>d are seeking developers interested in using this<br />
method during their design process. In particular, we<br />
suspect that being able to see the nonverbal gestural<br />
feedback as it unfolds will help designers in different <strong>an</strong>d<br />
interesting ways th<strong>an</strong> getting summarized results from other<br />
methods for measuring user affect.<br />
Cultural Tr<strong>an</strong>sfer<br />
One import<strong>an</strong>t claim we’ve made in this paper is that this<br />
instrument could be used across cultures. We have<br />
completed a study replicating the method used in the U.S.<br />
in Sweden, <strong>an</strong>d are <strong>an</strong>alyzing those results, <strong>an</strong>d would like<br />
to solicit interested groups from other cultural backgrounds<br />
to conduct replication studies. We c<strong>an</strong> provide <strong>an</strong> object set<br />
on lo<strong>an</strong> for such studies.<br />
Further Exploration of <strong>Affective</strong> Dimensions in Shape<br />
The patterns in response to the objects created by our<br />
sculptor suggested the following sculptural dimensions of<br />
interest:<br />
• rounded versus spiky (positive to negative valence)<br />
• smooth versus bubbly or protruding surface (low versus<br />
high arousal)<br />
• symmetrical versus asymmetrical (calmer or more<br />
directed/resolved versus confused/chaotic)<br />
We are pl<strong>an</strong>ning to explore iterations of objects based upon<br />
these dimensions (with the addition of the ‘droopiness’<br />
dimension suggested by one particip<strong>an</strong>t), perhaps even<br />
crafting <strong>an</strong> interface that allows people to use sliders to<br />
craft their own affective objects, to give us a better<br />
underst<strong>an</strong>ding of these shape variables <strong>an</strong>d how they relate<br />
to affect (the objects could then be ‘printed’ using a 3-d<br />
printing device <strong>an</strong>d tested in further studies).<br />
Multiple Method Tri<strong>an</strong>gulation<br />
We’ve also made the claim in this paper that this instrument<br />
could act as a helpful supplement to other modes of<br />
detecting/requesting affective state information. We are<br />
working toward a replication of our study that includes<br />
biometric monitoring as well as a post-survey, to compare<br />
results <strong>an</strong>d get a more complete picture of what each<br />
method exposes so far as particip<strong>an</strong>t affect.<br />
Dynamics<br />
Finally, we are also considering embedding sensors to<br />
allow us to record <strong>an</strong>d underst<strong>an</strong>d particip<strong>an</strong>t’s gestures<br />
with the objects (as with the eMoto system [6]).<br />
1171
CHI 2006 Proceedings • Beliefs <strong>an</strong>d Affect<br />
CONCLUSION<br />
Through iteration <strong>an</strong>d prototype testing, we believe we have<br />
demonstrated the promise <strong>an</strong>d potential of the Sensual<br />
<strong>Evaluation</strong> Instrument as a real-time, self-report method for<br />
eliciting affective responses to a computer system, toward<br />
creating <strong>an</strong> additional form of exch<strong>an</strong>ge between users <strong>an</strong>d<br />
designers. We are excited about the possibilities for the<br />
instrument in further enabling design teams to engage in<br />
productive <strong>an</strong>d reasonably-scaled user testing that improves<br />
the emotional experience for end users. We hope that other<br />
researchers <strong>an</strong>d designers in the CHI community will take<br />
up our offer of using the SEI in other cultural contexts, <strong>an</strong>d<br />
toward the development of systems, so that we c<strong>an</strong> further<br />
refine the instrument toward a final stage of having a ‘kit’<br />
that c<strong>an</strong> be used by <strong>an</strong>yone, <strong>an</strong>ywhere—a helpful addition<br />
to the r<strong>an</strong>ge of methods we have to measure <strong>an</strong>d underst<strong>an</strong>d<br />
affect in action.<br />
ACKNOWLEDGMENTS<br />
We th<strong>an</strong>k the Europe<strong>an</strong> Union for the funding of<br />
researchers dedicated to this project; we th<strong>an</strong>k Rainey<br />
Straus for creating the objects, <strong>an</strong>d for donating time to the<br />
project; <strong>an</strong>d we th<strong>an</strong>k those in the U.S. <strong>an</strong>d Sweden who<br />
participated in our studies. Th<strong>an</strong>ks also to those affective<br />
presence researchers who inform <strong>an</strong>d inspire our work:<br />
Phoebe Sengers, Bill Gaver, Michael Mateas, <strong>an</strong>d Geri Gay.<br />
REFERENCES<br />
1. Clynes, M. 1989. Sentics: The Touch of Emotions. MIT<br />
Press.<br />
2. Desmet, P.M.A. 2003. Measuring emotion: development<br />
<strong>an</strong>d application of <strong>an</strong> instrument to measure emotional<br />
re-sponses to products. In Blythe, MA, Overbeeke, K,<br />
Monk, AF & Wright, PC (Ed.), Funology: from<br />
usability to en-joyment. (Hum<strong>an</strong>-computer interaction<br />
series, 3, pp. 111-123). Dordrecht: Kluwer.<br />
3. DiPaulo, B.M., Rosenthal, R. 1979. The Structure of<br />
Nonverbal Decoding Skills. Journal of Personality<br />
47(3), 506-517.<br />
4. Djajadiningrat, W., Gaver, W., <strong>an</strong>d Fres, J.W. 2000.<br />
Interaction Relabelling <strong>an</strong>d Extreme Characters:<br />
Methods for Exploring Aesthetic Interactions.<br />
Proceedings of the conference on Designing interactive<br />
systems: processes, practices, methods, <strong>an</strong>d techniques,<br />
August 2000.<br />
5. Ekm<strong>an</strong>, P. 1972. Emotion in the Hum<strong>an</strong> face: Guidelines<br />
for Research <strong>an</strong>d <strong>an</strong> Integration of Findings,<br />
Pergammon Press.<br />
6. Fagerberg, P, Ståhl, A., <strong>an</strong>d Höök, K. 2004. eMoto –<br />
emotionally engaging interaction, Journal of Personal<br />
<strong>an</strong>d Ubiquitous Computing, 8(5), 377-381.<br />
7. Gaver, W., Beaver, J., Benford, S. 2003. Designing<br />
Design: Ambiguity as a Resource for Design.<br />
Proceedings of CHI 2003, 233-240.<br />
April 22-27, 2006 • Montréal, Québec, C<strong>an</strong>ada<br />
8. Gaver, W., Dunne, T., <strong>an</strong>d Pacenti, E. 1999. Design:<br />
Cultural Probes. Interactions 6(1), 21-29.<br />
9. Green, W.S. <strong>an</strong>d Jord<strong>an</strong>, P.M. 2002. Pleasure with<br />
Products: Beyond Usability. London: Taylor <strong>an</strong>d<br />
Fr<strong>an</strong>cis.<br />
10. <strong>Isbister</strong>, K., <strong>an</strong>d Höök, K. Evaluating <strong>Affective</strong><br />
Interfaces: Innovative Approaches. CHI 2005.<br />
http://www.sics.se/~kia/evaluating_affective_interfaces/<br />
11. <strong>Isbister</strong>, K., <strong>an</strong>d Nass, C. 2000. Consistency of<br />
personality in interactive characters: Verbal cues, nonverbal<br />
cues, <strong>an</strong>d user characteristics. International<br />
journal of hum<strong>an</strong>-computer studies 53(2), 251-267.<br />
12. Johnston, O. <strong>an</strong>d Thomas, F. 1995. The Illusion of Life:<br />
Disney Animation, Disney Editions.<br />
13. L<strong>an</strong>g, P.J., Bradley, M.M., & Cuthbert, B.N. 2005.<br />
Interational <strong>Affective</strong> Picture System (IAPS): Digitized<br />
Photographs, Instruction M<strong>an</strong>ual <strong>an</strong>d <strong>Affective</strong> Ratings.<br />
Technical Report A-6. Gainesville, FL. The Center for<br />
Research in Psychophysiology, University of Florida.<br />
14. LeDoux, J.E. 1996. The Emotional Brain. New York,<br />
Simon <strong>an</strong>d Schuster.<br />
15. Myers, David G. 2002. Intuition: Its Powers <strong>an</strong>d Perils.<br />
Yale University Press.<br />
16. Norm<strong>an</strong>, D.A. 2004. Emotional Design: Why We Love<br />
(or Hate) Everyday Things. New York: Basic Books.<br />
17. Pashler, H.E. 1998. Central Processing Limitations in<br />
Sen-sorimotor Tasks, Pp. 265-317 in The Psychology of<br />
Atten-tion, Cambridge, MA: MIT Press.<br />
18. Picard, R. 1997. <strong>Affective</strong> Computing. Boston, MA:<br />
MIT Press.<br />
19. Picard, R., <strong>an</strong>d Daily, S.B. 2005. Evaluating affective<br />
interactions: Alternatives to asking what users feel.<br />
Presented at the 2005 CHI Workshop ‘Evaluating<br />
<strong>Affective</strong> Interfaces’.<br />
20. Russell, J.A. 1980. A Circumplex Model of Affect,<br />
Journal of Personality <strong>an</strong>d Social Psychology 39(6), pp.<br />
1161-1178, Americ<strong>an</strong> Psychological Association.<br />
21. S<strong>an</strong>ders, L. (faculty web page)<br />
http://arts.osu.edu/2faculty/a_faculty_profiles/design_fa<br />
c_profiles/s<strong>an</strong>ders_liz.html.<br />
22. Wensveen, S.A.G. 1999. Probing experiences. Proceedings<br />
of the conference Design <strong>an</strong>d Emotion, November<br />
3 - 5 1999, Delft University of Technology, 23-29<br />
23. Wensveen, S.A.G., Overbeeke, C.J., & Djajadiningrat,<br />
J.P. 2000. Touch me, hit me <strong>an</strong>d I know how you feel. A<br />
design approach to emotionally rich interaction.<br />
Proceedings of DIS'00, Designing Interactive Systems.<br />
ACM, New York, 48-53.<br />
24. Wong, D., <strong>an</strong>d Baker, C. 1998. Pain in Children: Comparison<br />
of Assessment Scales, Pediatric Nurse 14(1),<br />
9017.<br />
1172