See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220888824

Comparing two approaches to context: Realism and constructivism

Conference Paper · August 2005

DOI: 10.1145/1094562.1094597 · Source: DBLP

CITATIONS READS

19 163

3 authors:

Antti Oulasvirta Sakari Tamminen

Aalto University University of Helsinki
310 PUBLICATIONS 12,026 CITATIONS 34 PUBLICATIONS 1,506 CITATIONS

SEE PROFILE SEE PROFILE

Kristina Höök
KTH Royal Institute of Technology
216 PUBLICATIONS 6,944 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Somaesthetic Design View project

Ethnographic Praxis in Industry Conference Proceedings View project

All content following this page was uploaded by Kristina Höök on 21 February 2014.

The user has requested enhancement of the downloaded file.

 Comparing Two Approaches to Context:
Realism and Constructivism
Antti Oulasvirta Sakari Tamminen Kristina Höök
Helsinki Institute for Information Helsinki University of Stockholm University / KTH
Technology Technology Forum 100
P.O.Box 9800 P.O.Box 9210 SE 164-40 Kista, Sweden
FIN-02015 HUT, Finland FIN-02015 HUT, Finland
ABSTRACT “recognized” and adapted to. More recently, inspired by
During the last few years, there have been debates over human and social sciences, a new approach has emerged
what is context and how computers should act upon it. Two that emphasizes the role of people in creating and utilizing
disparate camps of thought can be recognized. First, Real- contexts with the help of computers.
ism, having its roots in natural sciences, believes that con-
texts exist out there and that, if properly instrumented and At this point, then, the field seems to be divided into two
programmed, computers can correctly recognize and adapt camps holding disparate views on what context is and how
to them. Second, Constructivism, having its roots in human it should be harnessed. The first, called here Realism, be-
and social sciences, believes that contexts are human crea- lieves that contexts exist, ontologically, and that, if properly
tions, mental and social, and that computers ought to pro- instrumented and programmed, computers can correctly
vide resources for managing them. We reveal some funda- recognize and adapt to them. The second, called here Con-
mental differences between the two in three different appli- structivism, believes that contexts are human creations,
cation domains. We show that despite the deep-going con- mental and social, and that computers ought to provide re-
troversies, both camps benefit from considering the alterna- sources for managing these contexts.
tive approach and a middle ground can be found. In this paper, our goal is to analyze the two approaches in
order to explicate and compare their underlying assump-
Author Keywords tions, examine practical implications to context-awareness,
Context, constructivism, realism, context-awareneness, so- and discuss possibilities for reconciliation.
cial navigation, ethnomethodology, ubiquitous computing.
REALISM
ACM Classification Keywords Realism holds that there are properties that can be applied
H5.2. User Interfaces: Theory and methods. to many things, rather than denoting a single specific indi-
vidual. According to realists, these properties, so called
INTRODUCTION universals, really exist. Naïve Realism, particularly, be-
Innovation of use scenarios and applications for context- lieves that objects and universals are as we perceive them,
aware computing has been both inspired and constrained by independently of our observation. It is naïve in the sense
developments regarding the concept of context; that is, what that scientific descriptions and models (here: of contexts)
is being conceived of being context [5-7,18-22]. Although are taken at face value. Its ontological and epistemological
Weiser [22] never intended it to be interpreted narrowly as assumptions and knowledge interest have their roots in
static, immediately observable features of user’s surround- positivistic philosophies, which provide a basis for natural
ings, his followers thought context mainly as location. After sciences (e.g. naïve realism of perception).
roughly five years, the restrictiveness of this view was real-
ized and broader views were proposed (e.g., [5,6,18,20]). The following set of beliefs can be recognized:

Some called this the context debate, a debate still going on. • Ontological. Context is real, it is structured, and the
However, context was and is still mainly pursued by com- structure can be modeled by a learner. Contexts share
puter scientists and regarded as something that must be properties (or universals) that exist independent of hu-
man interpretation
• Epistemological. Computers can perceive (i.e., recog-
nize based on sensor data) context universals
• Pragmatic. If correctly recognized, computers can
adapt their behavior to the requirements of the context.
 for a contextual state and the markers of transition between
Applications
The main promise of applications that adhere to Realism is contexts, which should be recognized before the actual in-
the delivery of context-sensitive automation. teraction happens. Thus, the challenges related to recogniz-
ing and repairing interaction are the result of contradictory
Ubiquitous computing is the key area that has taken on the presuppositions about the nature and relationship between
realistic perspective. According to Tennenhouse [21] there the concepts of context and interaction.
will be thousands of processors per each person in the envi-
ronment, and these processors will be communicating, try- For example, emotions are not stable states to be recognized
ing to build up patterns of recognition of what each one of but part of on-going interactions. People’s affective interac-
us is doing and from those predict what we are going to do tion consists of much more than what can be understood
next. Consider, for example, a pillow that inflates before we from simplistic local measurements of their bodily reac-
realize that we need it, or an agent that appoints meetings to tions. Significant emotions (beyond elemental experiences
a medical doctor, or a video projector that prevents slides such as of surprise, disorientation, or disgust) are to a large
recognized as being “sensitive” from being shown. extent social phenomena that take place in specific cultural
settings, taking on particular expressions colored by the
In affective computing, we find several examples of appli- culture and the group of people at a particular place. The
cations building on the assumption that it is possible to cor- meaning and expression of emotions like guilt, shame, are
rectly recognize and then adapt to user’s emotional state. It given both by their local social as well as by their cultural
is assumed that it will be possible to recognize users’ inner context. They are part of a communication process where
emotional states from for example face recognition systems we are trying to make ourselves understood to others, where
or biosignals. Many application ideas build on results such their reciprocal reactions will modify our own emotional
as Ekman’s six basic emotions that can be seen and recog- processes in turn, and where initial bodily reactions are re-
nized irrespective of culture, or on findings that biosignal flected upon consciously, filtered through our attitudes and
data recognizing heart rate or sweat can be used as stable personality, and thereby continuously modified (see [8]).
indications of arousal. For example, Expression Glasses
[17] measure the movement of facial muscles and classify CONSTRUCTIVISM
the resulting expression into a small, discrete set of emo- An alternative to Realism is Constructivism. Constructivism
tions. Emotion Mouse [1] extends a normal computer recognizes that all our knowledge is “constructed”, it does
mouse to deduce users’ emotional states from physiological not reflect any external "transcendent" realities, but is con-
information such as pulse and galvanic skin response. tingent on convention, human perception, and social ex-
perience. It gives credit to different kinds actors as active
Similarly, in interruptablity research, some have proposed
constructors of contexts, not just passive reactive agents to
a secretary-like agent that estimates your level of interrup-
environmental aspects and new technologies. In fact, it pro-
tability and based on that blocks or permits incoming phone
vides a standpoint from which multiple interpretations and
calls (see [9]). A presumption is that people have a mental
understandings of contexts are legitimized and from which
state of interruptability that is independent of the intentions
these understandings can be imported into the design proc-
of the person who is calling you, and of your representa-
ess. In addition, it values the member’s point of view to the
tions of her, and, more importantly, that can be reliably
concrete use situations without this being as problematic to
detected from immediately observable data (e.g., phone call
the epistemic-ontologic configuration as it is for the realists.
activity, body posture etc.).
It also frees the researches from "one truth" to do critical,
reflective evaluations of research/design processes in terms
Challenges and Limitations
of power relations, etc. (cf. [10]).
In general, Realism has a zero tolerance for multiple simul-
taneous interpretations of context as it tries to attain the The following set of beliefs can be recognized:
"truth". There is no room for epistomological pluralism in • Ontological. Contexts are constructed socially, in in-
this culture of thought. This is a particularly difficult as- teraction with other agents in the world, and psycho-
sumption when the to-be-recognized context is mental or logically, in making sense of sense data.
social state. A consequent challenge is related to the ab- • Epistemological. Interpretation of context is always
sence of ways for repairing or renegotiating the computa- constituted within a frame of reference.
tional suggestions of contexts once they have gone wrong, • Pragmatic. Instead of labeling contexts, computers can
as Suchman [19] has argued. provide resources for people themselves to create and
Because realism holds that successful interaction consists of maintain contexts in their action.
communication of information between the computer and Historical Background
the user, interaction is defined simply as information ex- Constructivism is a general category including various re-
change. This is why "context" is presupposed to exist and to search strategies and philosophical orientations. It has its
surround interaction, rather than seen as an evolving out- intellectual roots in many social scientific and philosophical
come of it. For realists, then, an interlinked challenge is to branches, of which the larger HCI community knows
define the markers of context: the start and end conditions mainly phenomenology and ethnomethodology. They have
offered alternative perspectives in HCI and influenced the we have shown that the dichotomy is not just hand waving
move from single-user systems, ergonomics, and cognitive but manifests itself in the applications in fundamentally
sciences. Curiously enough, anthropological studies of different ways of thinking about uses of contexts.
computing systems also inspired Weiser’s vision of ubiqui-
It seems that realists view "context" simply as a new buzz-
tous computing, but these origins were soon forgotten in the
word for "environment"—something that can be dealt with
main current of the field.
old, natural sciences based, concepts, methods, and tech-
niques. Constructivists believe that context entails more
Applications
The design of systems that follows a constructivist stance than just a location or task. Context is all about situational
will not infer the inner psychological state of the user (e.g., sense-making, both by humans and interactive systems.
“interruptability”), or outer social state (e.g., “meeting”), or Thus, context could be viewed as a sum of situational ac-
in general label sensor data into particular higher-level tions and resources (plans, emotions, technologies, other
“meaningful” states to which the system can adapt. Rather, people) about how different resources integrate to the fab-
such a system will be designed to offer its state to users for rics of given interactions between actors. Contexts emerge
interpretation and, through their individually constructed from actions; they are emergent properties (achievements)
understanding, make it useful in their life, empowering negotiated in reciprocal relationships between actors (inter-
them to do a multitude of different things – most of which subjectivities and -objectivities) in complex interactions
may not be anticipated at the design moment. Meaning is among humans and computers (see also [14]).
not something that resides in our brain as a result of our We conclude the paper by analyzing the relationship of the
conscious processing, but is created and re-created through dichotomy proposed here to previous ones, its relationship
being in and interacting with the world [6]. to user-centered design. We then discuss how one might
In affective computing, a constructively oriented system find a balance between the two approaches.
may well use biosensors and similar, but the interpretation
“Constructivism” or “Phenomenology”?
is offered to the user who can use it to construct new con-
Dourish has recently presented similar arguments and a
texts [2]. For example, Fagerberg et al.’s [8] SMS input
divided approaches to context to three different camps [7].
system recognizes pen gesture patterns and maps those as
A critical reader might ask why we picked different terms.
affectively loaded background pictures (e.g., red for ex-
cited) that are shown to the receiver. If the image suggested Our Realism maps loosely to his Positivism that views con-
by the system does not suit the sender, it can be changed. text as a representational problem (how can one best repre-
Thus, no “semantic” category is imposed to the pattern of sent context to a computer). On the contrary, our concept of
gestures, rather the color, texture, and shape of the back- Constructivism wider than Dourish’ Phenomenological
ground image is provided as an extra “context” in con- approach that maybe overly emphasizes the role of individ-
structing an interpretation of sender’s mood or intentions. ual, first-person experiences and simple human-computer
interactions in the construction of contexts. Dourish´s
In interruptability research, instead of agents that block
analysis deduces all attempts to see contexts as con-
incoming notifications, users can be supported in how they
structed, or as "interactional achievements", to phenome-
manage their time and tasks themselves. For example, Mi-
nological roots. However, this analysis does not account
croSoft has explored the possibility of visualizing incoming
adequately for different strands and philosophical under-
notifications by their urgency and type (e.g., email, IM) in a
pinnings of constructivism both within ethnomethodology
radar-like presentation and giving the user several tools for
(e.g. [12]) and constructivism at large (for an excellent
selecting and organizing the view [4].
overview see [11]). By choosing the term Constructivism,
Yet another example is social navigation where the social we want to highlight that contexts are constructed in com-
trails of others are left in order to allow for richer interpre- plex interaction and interpretation chains that include not
tation of the meaning of the system. These trails also only computers but also other resources and people.
change the system functionality so that the original inten-
tions created by the designer will not be what the system User-Centered Design and the Two Approaches
ends up being. As users walk certain paths more frequently, One of the observations we did is that the most representa-
they get more “worn”, and increase in importance. Thus, tive applications examples of Realism are some sort of
the system changes with its usage, reflecting its usage, but autonomous agents that make decisions on behalf of the
without any real interpretation of what this might mean user. This naturally triggers UCD issues like user veridical-
besides what the trails “are” in themselves [13]. ity (correctness of interpretations), transparency, control,
privacy, trust, and empowerment.
DISCUSSION
Interestingly, veridicality of context inference is a non-issue
The goal of this paper has been to bear in a significant way
for Constructivism, almost by definition, as context infor-
on the context debate. Constructivism contradicts Realism,
mation is a resource for human action and has to be ac-
and explicating and analyzing the differences is crucial to
counted by their users in social situations. This shifts the
bring unity to the otherwise disparate research. Importantly,
 focus from improving the “correctness” of the machine in- nology, Atlanta, GA.
ference to their repairability in social interaction. 6.Dourish, P. Where the Action Is: The Foundations of Em-
bodied Interaction. Cambridge, MIT Press (2001).
Towards Reconcilication
Labeling contexts does not have to imply obtrusive behav- 7.Dourish, P. What we talk about when we talk about con-
iors, however. Salovaara and Oulasvirta [14] have distin- text. Personal and Ubiquitous Computing, 8 (1), 2004,
guished between six types of proactive behaviors according 19-30.
to a resource management scheme. Proactive behaviors can 8.Fagerberg, P., Ståhl, A., and Höök, K. Designing gestures
prepare, optimize, suggest, manipulate, inhibit, or finalize for affective input: an analysis of shape, effort, and va-
user’s resources. Given that categorizing and labeling con- lence. Proc. MUM 2003, Norrköping, Sweden.
texts often fails, we believe that the labeling approach
9.Fogarty, J., Hudson, S.E., and Lai, J. Examining the ro-
might be especially suitable for the three first types opera-
bustness of sensor-based statistical models of human in-
tions as they are potentially unobtrusive to the user. Prepar-
terruptibility. Proc. ACM CHI 2004, Vienna, Austria,
ing new resources, optimizing on-going ones, or suggesting
ACM Press (2004), pp. 207-214.
new ones to the user can be harmless, be it based on realis-
tic or constructivistic idea of context. 10.Friedman, B. and Kahn, P.H. Jr. New directions: a
value-sensitive design approach to augmented reality.
A good example of “resource preparation” is “best guesses” Proc. Designing Augmented Reality Environments,
in a context-aware annotation system. Sarvas et al. [16] (2000), pp. 163-164.
developed a system where context-interpretations fill in
mobile image metadata annotation fields (e.g., name of the 11.Hacking, I. The Social Construction of What? Cam-
location) that can be easily overridden by the user if they bridge, MA, Harvard University Press (1999).
are wrong. There are naturally other design remedies to the 12.Mehan, H, and Wood, H. The reality of ethnomethodol-
design challenges—for example, context can be faked by ogy. New York, Wiley (1975).
the user [3]. This has the potential of turning labels into 13.Munro A., Höök, K., and Benyon, D. Social Navigation
resources for users themselves to construct interaction. of Information Space. London, UK, Springer (1999).
We conclude the paper by noting that not all researchers fall 14.Oulasvirta, A., Tamminen, S., Roto, V., Kuorelahti, J..
into either of the two categories—our goal has been to stir Interaction in 4-second bursts: The fragmented nature of
up a theoretical debate over the foundations of the field, yet attentional resources in mobile HCI. Proceedings of
keeping in mind its practical interest. We regard it of ut- CHI’05, Portland, OR, ACM Press (2004), pp. 919-928.
most importance that a balance can be found between the
15.Salovaara, A. and Oulasvirta, A. Six modes of proactive
two approaches. This begins by admitting the fact that
resource management: A user-centric typology for proac-
automatic context-interpretations are often arbitrary and
tive behaviors. Ext. Abstr. NordiCHI’04 (2004).
wrong, especially when they concern mental and social
states. As we have argued, automatic actions based on sen- 16.Sarvas, R., Herrarte, E., Wilhelm, A., and Davis., M.
sor data should be negotiable and repairable, and only pro- Metadata creation system for mobile images. Proc. Mobi-
vided as a controllable and transparent resource, starting Sys 2004, Boston, MA, USA. ACM Press (2004).
point, or option for users to amplify their action. 17.Scheirer, J., Fernandez, J., and Picard, R. Expression
glasses: a wearable device for facial expression recogni-
REFERENCES tion. Proc. CHI '99, Pittsburgh, PA (1999).
1.Ark, W., Dryer, D., and Lu, D. The emotion mouse. Proc.
HCI International 1999, Munich, Germany. 18.Schmidt, A., Beigl, M. and Gellersen, H.-W. There is
more to context than location. Computers & Graphics, 23
2.Höök, K. User-centred design and evaluation of affective (6), 1998, 893-901.
interfaces. In Pelachaud, C. and Ruttkay, Z. (Eds.),
Evaluating ECAs, Kluwer (forthcoming). 19.Suchman, L. Plans and situated actions: The problem of
human-machine communication. Cambridge, Cambridge
3.Brown, P.J., and Jones, G.J.F. Context-aware retrieval: University Press (1987).
exploring a new environment for information retrieval
and information filtering. Personal and Ubiquitous Com- 20.Tamminen, S., Oulasvirta, A., Toiskallio, K., and
puting, 5(4), 2001, 253-263. Kankainen, A. Understanding mobile contexts. Personal
and Ubiquitous Computing 8 (3), 2004, 135-143.
4.van Dantzich, M., Robbins, D., Horvitz, E. & Czerwinski,
M. Scope: Providing awareness of multiple notifications 21.Tennenhouse, D. Proactive computing. Communications
at a glance. Proc Advanced Visual Interfaces 2002, of the ACM 43, 5 (2000), pp. 43–50.
Trento, Italy, 2002. 22.Weiser, M. The Computer for the 21st Century. Scien-
5.Dey, A. K. and Abowd, G. D. Towards a better under- tific American, 265 (3), 1991, 66-75.
standing of context and context–awareness. Technical re-
port GIT-GVU-99-22, 1999, Georgia Institute of Tech-

View publication stats