the effects of mutual location-awareness on group coordination

ARTICLE IN PRESS

1071-5819/$ - se

doi:10.1016/j.ijh

�Correspondfax: +41 21 69

E-mail addr

(F. Girardin), p

Int. J. Human-Computer Studies 68 (2010) 451–467

www.elsevier.com/locate/ijhcs

The effects of mutual location-awareness on group coordination

Nicolas Novaa,�, Fabien Girardina, Pierre Dillenbourgb

aLiftlab, 4 Avenue Jules Crosnier, 1206 Geneva, SwitzerlandbSchool of Computer Sciences and Communication, Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland

Received 25 August 2009; received in revised form 22 December 2009; accepted 31 December 2009

Communicated by S. Wiedenbeck

Available online 3 March 2010

Abstract

The importance of space and place in collaborative practices has been strengthened with the ubiquitous computing paradigm, which

aims at the integration of computation in physical objects and places. New location-based applications allow users to know where other

individuals are in the physical world. New collaborative applications engage users in geographically dispersed and mobile activities.

However, there is still a lack of information concerning how mutual location-awareness (i.e. knowing partners’ whereabouts) might

influence socio-cognitive processes involved in coordination. To address this issue, we conducted field experiments with a mobile and

collaborative game, running on Tablet PCs, and compared two interfaces. On the first interface, groups received automatic updates from

teammates’ whereabouts, while this automatic MLA tool was not provided by the second interface. In addition, all users could use their

Tablet PCs to communicate by annotating the map. We found no differences between the two conditions with regard to the task

performance. However, contrary to our expectations, players without automatic MLA had a better representation of their partners’

paths, wrote more messages and provided more elaborate explanations of their strategies. Additionally, automatic location-awareness

undermined the coordination process, leading participants to be less articulate about their strategy. The paper discusses these results and

the implications of such results.

& 2010 Elsevier Ltd. All rights reserved.

Keywords: Pervasive gaming; Collaborative game; Location-awareness; Location-based services; Field experiment.

1. Introduction

For several decades, researchers and engineers proposedto use spatial metaphors in order to support human–computer interaction using concepts such as the ‘‘desktop’’organization or web ‘‘sites’’ and ‘‘chatrooms’’. Theimportance of space and place has been made morerelevant with the ubiquitous computing paradigm, whichaims at the integration of computation in physical objectsand places (Weiser, 1991). Among other technologies inthis domain, there is a surge of location-based services,which is to say mobile applications that take advantage oflocation information in various contexts like supporting

e front matter & 2010 Elsevier Ltd. All rights reserved.

cs.2009.12.007

ing author. Tel.: +41 78 614 85 61;

3 60 70.

esses: [email protected] (N. Nova), [email protected]

[email protected] (P. Dillenbourg).

group coordination, playing games or engaging users inlearning activities (see Benford, 2005 for a review). Therecent democratization of these services leads to a newfeature called mutual location-awareness (MLA in theremainder of this paper): users can be informed of theirown and/or their teammates’ locations. MLA raisesinteresting issues already studied within the CSCWcommunity, such as how collaboration can benefit frominterfaces that convey awareness of others. Dourishand Bellotti (1992) defined awareness as: ‘‘an under-standing of the activities of others, which provides acontext for your own activity’’ (p. 107). Drawing on thisdefinition, location-awareness would be the understandingof the others’ position in the spatial environment.Gutwin and Greenberg (2002) stressed that awareness isknowledge about the state of the work environment in alimited portion of time and space. Since this informationwas not available in many collaborative applications,

www.elsevier.com/locate/ijhcs

dx.doi.org/10.1016/j.ijhcs.2009.12.007

mailto:[email protected].3d

mailto:[email protected]

ARTICLE IN PRESSN. Nova et al. / Int. J. Human-Computer Studies 68 (2010) 451–467452

designers provided users with so-called ‘‘awareness tools.’’These tools were expected to facilitate team collaborationby showing information about presence (is anyone in theworkspace?), about their identity (who is that?), about theirlocation (where is an individual?), about their actions (whatis somebody doing?), and so forth. MLA is a specificcategory of awareness tools that provide the ‘‘where’’information, i.e. the locations of each group member (Dyckand Gutwin, 2001). What is meant by ‘‘location’’ takesdifferent forms; it can be a text message that indicates aperson’s whereabouts or dots on a map. The adjective‘‘mutual’’ covers knowing both one’s own location and thelocation of partners. It implies a notion of reciprocity: ifperson A may see her partners B and C’s location, thenB and C also have access to A’s location (not necessarily inreal time).

Different MLA interfaces have been developed: theydiffer in terms of methods for capturing the location(automatic versus self-disclosed), functions for posting aquery (e.g. querying who is where and when) and how theinformation is rendered to the users. Most of the MLAinterfaces use a visual output, often on a map, sometimesas textual description and in a few cases with moreinnovative visualizations. Past research about the roleof MLA in collaborative virtual environments haveshown that it supported implicit coordination and divi-sion of labor (Dillenbourg and Traum, 1997), as well asmutual understanding of each other’s intents (Nova et al.,2007).

MLA also relates to the communication practice ofgiving or asking geographical information in cell phoneconversations. Various studies have shown that it helpspeople to mutually establish and share a spatio-temporalcontext (Laurier, 2001), allow group coordination formeetings (Weilenmann, 2003) or be an index of interac-tional availability (Arminen, 2006). Given the importanceof this phenomenon, Arminen claims that a technicalsolution to indicate location-awareness would have a wideapplicability for a majority of mobile users.

Although numerous mobile applications implementedand tested MLA interfaces, there have been few experi-ments to understand their effects on collaboration. Existingstudies put more emphasis on the design than onthe empirical investigation of how users’ behavior isinfluenced by knowing where the partners or the compe-titors are located. Therefore, we conducted an empiricalstudy aimed at investigating the influence of MLA ongroup coordination in mobile settings. More specifically,we compared two interfaces that provide users withlocation-awareness: on the first interface, groups receivedautomatic updates from teammates’ whereabouts, whilethis automatic MLA tool was not provided by the secondinterface. Using these two categories of interface, weinvestigated group cognitive processes such as commu-nication and the representation each team member makesabout one another, which allows groups to coordinatewhen performing a joint activity.

This paper first describes the related work about howsocio-cognitive processes are impacted by MLA interfaces.The second section presents the theoretical framework weused to investigate this issue. After a brief description ofour research scope, we detail the field experiment weconducted to understand the influence of MLA oncollaboration. Finally, the last section discusses the out-comes and their consequences for practitioners.

2. Related work

Given that the core of our study concerns thecomparison between automatic versus intentional loca-tion-awareness, it is necessary to state the role of MLA aswell as the main drawbacks of automation in the context ofcollaborative applications.

2.1. Roles of mutual location-awareness

Mostly technology-driven, the exploration of MLA’sroles and influences has targeted the evaluation oflocation-based applications. The majority of these studiesonly address synchronous and ‘‘position-based’’ MLAinterfaces (information about partners in space, shownon a map in real time), delineating different roles suchsystems play in collaboration.MLA influences different moments of verbal interactions

in pervasive and collaborative applications. For instance,Licoppe and Inada (2005) studied a location-based game,deployed in Japan called Mogi Mogi, in which players haveto collect virtual localized artifacts in Tokyo. The authorsnoticed that viewing the others’ positions on the cell phonescreen constituted an affordance for social encounters and ledto specific forms of conversational openings. Location-awareness in this context was often used to trigger aconversation. Furthermore, in a study of the pervasive game‘‘Uncle Roy All Around You’’, Benford et al. (2004)demonstrated that MLA (in the form of a map) providedcues that can be perceived as ‘‘deictical’’ linguistic elements togive directions or tell one’s location. In this context, MLAwas used to indicate direction in two ways: either by using theknown location to derive possible directions (‘‘you are veryclose now, stay on that side of the road’’) or by givingdirections using absolute geographical references (such aslandmark: ‘‘Go to 12 Waterloo Place’’). For Brown et al.(2003), it was rather the content of verbal interactions thatwas influenced by the presence of an MLA. They exploredhow a mixed reality system would augment visits to museumsby allowing voice communication and sharing informationconcerning location and orientation (represented as a map).Their experiment showed that MLA was a powerful resourcefor conversation, namely because it facilitated referentialcommunication, that is the understanding that an item whichis present in an individual’s vicinity is being referred to by theconversational partner. Knowing the partners’ locationindeed helped to establish deictic cues: it enabled participants

ARTICLE IN PRESSN. Nova et al. / Int. J. Human-Computer Studies 68 (2010) 451–467 453

to understand what the partner was looking at and thusencouraged to either talk about it or to look at it themselves.

Dearman et al. (2005) investigated how location-awaretechnology impacts social behavior within the context ofrendezvousing (meeting at an agreed upon time andlocation). They compared three different conditions: usinga mobile phone, using a PDA displaying MLA as ascrollable map and using both devices. All groups wereable to complete the rendezvous task without muchdifficulty but participants exhibited very different beha-viors depending on the technology used. The phoneappeared to be relevant for discussing actions (‘‘what areyou doing?’’) and intentions (‘‘where are you planning togo?’’) but the MLA displayed on the PDA was moreefficient to find partners and to provide users with a feelingof ‘‘ambient virtual copresence.’’ All the participants whohave both the phone and the PDA first employed the MLAand only a small portion then placed a call. MLA was, inthe authors’ words, a ‘‘background communication chan-nel to monitor their partner’’ (Dearman et al., 2005, p.561). Furthermore, by seeing in an unobtrusive mannerpeople’s position and movement, MLA provided informa-tion about partner’s contribution to the task, as well astheir progress. Slightly similar to this, Brown et al. (2007)described how MLA is sometimes less about communicat-ing people’s whereabouts but rather to support users inwhat they already know about each other and what oneexpects. In their study, they designed a clock situated in thehome that enables family members to see where othermembers of the family are using categories such as‘‘home’’, ‘‘work’’ or ‘‘school’’. Their results showed howMLA was a form of reassurance, the added value being lessin the physical position per se but in what location meansin interaction.

2.2. Automatic MLA drawbacks

Despite all the roles of MLA we described beforehand,various studies have brought forward three majordrawbacks of automating location-awareness. The mostimportant one is certainly that it raises privacy concerns.It weakens location privacy, i.e. ‘‘the ability to preventother parties from learning one’s current or past location’’(Beresford and Stajano, 2002, p. 1536). This generatesdifficulties in the social acceptance of MLA technologiesthat lead to user rejection or reluctance to employcertain features. A possible answer to these concerns is toprovide abilities to switch off MLA or to definedifferent levels of permissions to access to the locationinformation (i.e. spatial cloaking). This relates to thelong-term debate in the CSCW field about thebalance between awareness and privacy (Hudson andSmith, 1996): designers of multi-user applications face theproblem of providing enough information transmittedto others (so that they can benefit from it) withoutthreatening the protection of users’ privacy. This problemis closely related to the level of control users would

accept to leave to the system. Iachello et al. (2005) haveindeed shown the lack of value of automatic MLAthat does not support the possibility to lie or plausibledeniability in communications. Using a messagingapplication for mobile phone that allows the user torequest the location of others and tell his/her location tothem, they found a lack of trust and a loss of controlfrom users. The system indeed did not allow participantsto deceive, or deny replies, from time to time, for purposesthat are important to them. In addition, Consolvoet al. (2005) showed that the most important factorsfor MLA users were who was requesting the information,why the requester wanted the participant’s locationand what level of detail would be most useful to therequester.A second concern is that technological pitfalls can lead

to flaws or bad positioning accuracy (Benford et al., 2006,2005): ubiquitous computing is still a maturing field inwhich frequent problems arise like unreliable network,latency, bandwidth, security, unstable topology ornetwork heterogeneity. However, users acquire strategiesto adapt to the aforementioned system failures. One of thesolutions to cope with MLA discrepancies was to let usersmanually reveal their positions (Benford et al., 2004).These authors found that rather than reporting themselvesto be at a different place, the users were in fact reportingthemselves to be at a different time. Their results showedthat revealing one’s position was an act of communicationthat also revealed past or future intentions. However,self-reported positioning required users to know wherethey were and/or where they were heading, which is notalways the case. What is interesting though is that self-reported positioning is not neutral, but as the authorssay ‘‘imbued with meaning’’ at the moment the messageis generated, conveying more information than solelylocation.Finally, closer to our interest in MLA and group

collaboration there is the difficulty in interpreting theinformation conveyed by MLA tools. In the previouslymentioned study, Dearman et al. (2005) underline how thedisplay of location information provided little assistance tousers in interpreting the associated state of the person. As amatter of fact, when a user was lost or not making anyprogress, participants were disconcerted because there wasnot enough information to understand what the problemwas. The uncertainty in interpreting location informationcan lead to detrimental effects of MLA on users’ under-standing of the situation. As shown by Smith et al. (2005),the automatic disclosure MLA may suffer because theaccompanying knowledge of intended context for inter-pretation is lost.Another difficulty in interpreting MLA is that the

position offered or described by technology may notcorrespond to the positions people want to refer to whenthey are conversing (Rudstrom et al., 2005). Hightower andBorriello (2001) also raised this issue by saying thatlocation-based technologies have the problem of turning


geographical coordinates to ‘‘place’’ meaningful to theusers.

3. A psycholinguistic framework to address MLA influence

As we have already mentioned, our field experiment willdeal with the difference between automating location-awareness and letting users providing intentional messagesabout their whereabouts. It is therefore important toframe this study in a context that can account for thedifference between these sorts of signals in the context ofcollaboration.

As we stated in the introduction, awareness refers to theperception and the understanding of others’ interactions inthe environment. Awareness interfaces automate theexchange of information expected to ease coordinationbetween collaborators. Meeting at a particular time andplace or driving on a one-lane road is a simple example ofcoordination problems, i.e. situations where each person’sactions depend on the actions of the others (Schelling,1960). From the psycholinguistic point of view, awarenesscan be considered as what Schelling called a ‘‘key’’ or a‘‘focal point’’ or what Clark (1996), with Lewis (1969),termed a ‘‘coordination device1’’: a rationale for mutualexpectations that make partners believe that they willconverge on the same joint action. Coordination is thenframed as a continuous process in which people (1)exchange coordination devices and (2) solve coordinationproblems by punctually performing ‘‘mutual modeling’’acts (Dillenbourg, 1999), i.e. using shared information toinfer mutual expectations. Coordination devices such asdirect communication or the information conveyed byawareness interfaces create a shared basis to infer whatpartners will be doing in the near future and to choosewhich actions one should perform to contribute to the jointactivity.

Clark distinguished 4 types of coordination devices: (1)explicit agreements (which are occurrences of dialogues inwhich parties explicitly communicate their own intentions);(2) conventions are community’s solution to a recurrentcoordination problem such as red traffic light; (3)precedents are norms and expectations developed withinthe on-going experience of the joint activity; (4) manifestelements refer to situations in which the environment(or the available information) makes the next moveapparent within the many moves that could conceivablybe chosen. These four coordination devices refer to mentalrepresentations (conventions, precedent), perceptual ele-ments from the environment setting (manifestness) andcommunication (explicit agreement). The construction ofmutual expectations is consequently both cognitive (i.e.expresses the need to access to mental representations) andsituated, given the set of coordination devices people can

1These authors do not take the word ‘‘device’’ in the typical sense of a

mechanical or electronic appliance, but rather use it for ‘‘tricks or plans

with a particular aim’’.

rely on to produce mutual intelligibility between partici-pants. In this typology of coordination devices, awarenessconveyed by different tools can be considered as a‘‘manifest’’ sign: MLA tools for instance allow makingmanifest-specific phenomena such as people’s movement inspace. The production of MLA requires first to capture anindividual’s location (by technological means) and then toprovide this information to the partners. MLA is thenavailable to the participants as a shared basis to formwhat Clark defined as ‘‘mutual expectations’’: inferencesabout what their partners did and what individual actionsshould be carried out (mutual modeling). MLA wouldindeed be a shared basis to develop mutual expectationsabout the present (knowing B’s current position, A couldinfer what B is doing or looking at), the past (knowing B’spast locations, A could infer what B has achieved orcollected so far) or the future (knowing B’s direction,A could infer where B is heading and thus what are B’sintents). Multiple levels of mutuality of knowledgeand intents can indeed be inferred from the spatialpositions of dispersed team members, which eventuallyease collaboration.

4. Research questions

The previous sections reviewed examples of how andwhy MLA can enhance collaborative processes. Ourframework leads us to assume that (1) shared models ofeach other’s course of action contribute to the coordinationof joint activities and (2) a coordination device such asMLA, conveyed by technological tools, is expected tosustain shared models by supporting a more efficientmutual modeling process. As a consequence, MLA isexpected to facilitate solving coordination problems andhence improving collaborative performance. Hence ourmain research questions are: Do MLA interfaces modifygroup coordination? How does MLA modify participants’interactions and group coordination? For each of thesequestions we have a hypothesis, respectively: (H1) Mutuallocation-awareness improves the task performance of thegroup; (H2) Mutual location-awareness improves theaccuracy of the mutual modeling (MM).This paper reports the results from a controlled

experiment that aimed at verifying these hypotheses.However, since there is little research about MLA as acoordination device, our study also has an exploratoryflavor; beyond testing our hypotheses, we explored severalqualitative aspects of how MLA influences collaborativeprocesses. Namely, we investigated how informationabout others’ whereabouts was used, especially with regardto spatial coordination, how subjects maintained arepresentation of the joint space, and whether this modifiedthe mutual coordination process. We analyzed if spatialinformation influenced communication patterns and howMLA information was integrated, misinterpreted ormissed out. Our aim was to understand the collaborativemechanisms influenced by MLA interfaces.

ARTICLE IN PRESS

Fig. 1. CatchBob! interface as seen by one player. Location-awareness is

the fact that 2 players’ avatars are displayed. In the no-MLA condition,

the interface only displays one’s own avatar.

N. Nova et al. / Int. J. Human-Computer Studies 68 (2010) 451–467 455

5. Method and material

5.1. Methodological choices

Balancing the needs to have an ecological validity andour aim for generalization led us to adopt a ‘‘fieldexperiment’’ methodology (Goodman et al., 2004). Itallowed us to involve real users in an activity that is setup in the real world as well as to control some variablesand compare different experimental conditions.

We tested our hypotheses through quantitative mea-sures, respectively group performance (H1) and mutualmodeling accuracy (H2). We collected these data during anexperiments in which we had two experimental groups:with or without MLA. This approach has been completedby ethnographic investigations of how MLA is interpretedand used among group participants. This research istherefore grounded in a quantitative paradigm, enrichedwith qualitative techniques (Creswell, 1994): We conductedinterviews and self-confrontation techniques (Theureau,1992) to complement the observer’s analysis with theparticipants’ perceptions of what happened during colla-boration.

Following the proposition by Chalmers and Juhlin(2005), we chose to use a collaborative game as a testplatform for four reasons. First, computer games aremotivating and fun, which ease experimentation. Main-taining one’s undivided attention is certainly easier in videogames than in unpleasant experimental environments.Games engage participants in complex problem-solvingtasks while maintaining a high level of motivation (Nilsenet al., 2004). Second, pervasive games involve participantsin a real context (the physical world) with a certainecological validity. Third, a game in public space indeedcreates a certain kind of complexity with passers-by or real-world features. Finally, the game design can serve as anapproximation of locative collaborative task as we will seein Section 5.4.

5.2. Participants and design

Sixty students of the Ecole Polytechnique Federale deLausanne (age range: 19–28; mean: 23.1) participated inthis experiment. We selected groups who belonged to thesame school class and who previously knew each otherbecause different levels of knowledge between partners mayimpact the representation each of them has about theirteammates. Participants spoke either French or English butthey used the same language within a group. They wererecruited through announcements on the school mailinglists and they were remunerated 20 Swiss Francs for theirparticipation. We also checked that players were allfamiliar with the campus, studying there for at least ayear. We controlled participants gender so that eachcondition was made up of 25% of female and 75% ofmale. Within a group, the repartition was generally 2 malesor 2 females. The experiments were conducted on our

campus, one group at a time. We created 20 groups of3 persons separated in 2 experimental conditions: in theexperimental condition, 10 teams used a tool providingMLA (MLA condition) while in the control condition, theMLA function was not available (no-MLA condition) tothe 10 other teams. The MLA tool provided participantswith the position of their partner in the real space in realtime (synchronous).

5.3. The game environment

CatchBob! is a pervasive game running on Tablet PCs.Groups of 3 teammates have to catch ‘‘Bob’’, a virtualstatic character. Completing the game requires the playersto surround Bob with a virtual triangle formed by eachparticipant’s position in the real space. To reach this goal,they employ an application as depicted in Fig. 1.The software displays players as avatars in three colors

(blue, green and red). A refresh button (only in the MLAcondition) updates the avatars positions. We chose toprovide the users with a refresh button (as opposed toautomatic updates) to count the number of times theywould access it. On the upper part of the interface, anindividual proximity sensor indicates whether the user isclose to or far from the target (Bob) by coloring in red thebars at the top of the display. There is actually no object onthe field; Bob only appears on their display when the usersform the proper triangle configuration around it. Thistriangle should be equilateral with a side length of 20–35 m.When the players are close to Bob, the triangle appears ontheir display and they have to adjust it to the expecteddimension. The game took place on the EPFL campus,whose dimensions are an 850� 510 m field.The interface enables synchronous communication:

players can annotate the map with the stylus. Theseannotations are reproduced on the two other displays, usinga different color for each player. They slowly fade out untilthey become completely invisible (after 4 min). This leads tosimple acts of communication and dialogue: for instance aplayer asks his or her teammate to move to a specificlocation with an arrow with the message ‘‘go there’’ and thepartner acknowledges this advice by writing an ‘‘ok, I gothere’’. Audio communication between participants was not


permitted by the system on the field. (but they haveco-present discussions in a strategy planning phase beforethe game).

We tested two types of interface. In the Control

condition ‘‘NoMLA’’, players only see their own characteras an avatar on the campus map. In the MLA condition,players could update his or her teammates currentpositions by clicking on the refresh button. They then seethe location of the three avatars. Note that all users withinthese 2 conditions could communicate through written mapannotations, which is another type of coordination device.The mixed indoors and outdoors settings of CatchBob!led us to use WiFi for the positioning, with an accuracy of10–30 m. This accuracy was lower in a few campus areashaving fewer access points or lower connectivity.

All the players’ interactions with the application (e.g.positions, annotations, refresh of others’ positions, con-nection loss) were logged on each client and the server. Weimplemented a replay tool that reproduces the path of eachplayer and the messages they wrote. This applicationallowed us to confront the players to a replay of their pathafter the game, as well as the actions they performed.

5.4. The collaborative task

Our pre-experiments showed that the CatchBob! gamecan be segmented into 3 phases: exploration, rendezvousingand triangle formation. The first phase is about exploringthe campus: players spread over the campus and try to findthe approximate area where Bob is located, using theproximity sensor. They communicated their results (num-ber of bars) to their partners by using shared mapannotations. In the second phase, rendezvousing, oneplayer eventually finds where Bob is approximately locatedand invites his/her partners who join him/her by annotat-ing the map. The third phase requires players to form thetriangle around ‘‘Bob’’: players try to form differentconfigurations of the triangle depending on their positions.An efficient way to find the boundaries between the threephases was to look at the group dispersion, namely theevolution of the perimeter of a triangle formed by each ofthe three players all over the game: in the first part, theplayers spread over the field (the perimeter increases) andthen get closer to each other in the second phase (theperimeter decreases) with a final spread to form thetriangle.

Even though finding Bob could be carried out individu-ally, collaboration is required by the fact that players haveto form a triangle surrounding the virtual object. That way,we minimize the risk of having a free rider effect. Othercollaborative aspects of the game are:

�
The exploration of the environment is an individual actionbut systematic sharing of findings makes collaborativesearch more efficient. � Before the game, participants had to discuss a strategy
to complete the game: who will go where, which part of

the campus should be searched by whom, how tocommunicate, how to form the triangle. This strategyhad sometimes to be revised during the game.
� During the game, the players collaborated via
map annotations, via the MLA tool (in the experi-mental condition) and sometimes by face-to-faceencounters.

Such a game was intended to be an approximation ofcollaborative locative activities like field workers (land-mine location, archaeological prospecting), emergencyrescues or certain military tasks. Each of these has both aspatial component (locating something or someone, friendor foe) and a social part (coordination or collaborativepractice). These tasks are also characterized by the factthat they may happen in situations for which audiocommunication is not possible. Certain landmine locationsystems or archaeological tools do indeed rely on mapannotations and not on voice communication. Conversely,military operations sometimes avoid the productionof noises.

5.5. Procedure

The participants went through five steps within about1 h. First, the experimenter presented the instructions atour office. Participants were asked to find the virtual objectand surround it with one constraint in mind: they shouldtake the shortest path to complete the task. We also toldthem that the time necessary to find the object was not acriterion of success. We chose distance as a performanceindex because this measure reflects the quality of teamcoordination and strategy rather than the participants’physical condition. We also invited them to try a demo ofthe Tablet PC software, namely the annotation feature,and to ask questions about the interface. Then, in aplanning phase, players were then given 5 min to plan theirstrategy on a paper map, which was then left in the office.The third part is the game itself: the experimenter led thegroup of 3 players to the common starting point at thecenter of the campus. They had 30 min to completethe task, which – from the pretests we ran – was sufficientto achieve the goal without much time pressure. Aftercompleting the game (or reaching the 30 min limit), playersreturned to our lab and filled a paper-based questionnaireduring 10 min. This questionnaire included 3 maps of thecampus on which they had to draw their own path as wellas paths followed by each of their partners. Finally, playerswere together confronted with a replay of their activity.The replay tool displayed in real time the position of eachgroup member as well as their annotations. Those traces ofthe activity were presented to push players to explain whathappened, an interview technique known as self-confronta-tion (Theureau, 1992). The player or the experimenterselected episodes of the game replay and asked one playerto tell what happened then. In general, the player’s answerstriggered other comments by partners, turning the


interview into a discussion about each of these moments.Conversation was recorded and for each episode, theexperimenter also asked one player whether (s)he knewwhat one of his partners was doing (and then thus partnerwas asked if the provided answer was right). This interviewallowed us to collect evidence of the inferences performedby the participants. The experimenter also asked who wasthe first player who found ‘Bob’.

5.6. Data collection and analysis

The CatchBob! platform allowed us to collect a wide setof data ranging from quantitative measures to playerinterviews and accounts of the game.

We measured task performance as the group traveldistance, i.e. sum of the path lengths over all players in agroup, measured in meters. We categorized the differentgroup spatial behaviors. Looking at how players spreadover the campus in the first phase as well as how theyjoined the first one who found ‘‘Bob’’ in the second phaseallowed us to gain insights about the group strategy as wellas how the MLA interfaces impacted it. For each of thethree parts of the game, players demonstrated variousspatial behaviors. In the first phase, there were twopossibilities of spreading over the campus: (A) some teamsmaximized the exploration of the campus by going in threedifferent directions and (B) in other teams, players went allin the same direction, not grouped together but insteadheading out together one behind the other.

As explained in the procedure section, we asked playersto draw the path of each of their partners. This enabled usto calculate the number of errors players made whiledrawing the path of their partners after the game. Weused this ‘‘positions recall’’ measure as an indicator of

Fig. 2. Examples of messages with regard to the two coding sch

mutual modeling accuracy. Since players had to drawthe paths of their partners, we could compare thepath player A attributed to B with B’s real paths and thesame for A and C or B and C. This comparison, measuredby the number of errors, represents the accuracy of A’srepresentation of B and C’s behavior in space. Wecomputed the number of errors by comparing the trailsdrawn on the paper map and the real paths generated bythe replay tool. We counted as an error places where thepartner has not been and the omission of a place where he/she went. Two criteria have been defined to describe whatconstituted an error: distance (if the line was longer thanthe maximum size of our campus corridor) and presence ofan obstacle (door/wall/glass). We calculated accuracy indexfor each individual and each group (the sum of theindividual indexes). Mutual modeling accuracy (MM-accuracy) is the sum of errors made by a player abouthis/her two partners’ paths. We calculated MM-accuracybetween each individual (M(A, B), M(B, A), M(A, C),M(B, C), y) and for each group (the sum of the individualmeasures).We coded the content of shared map annotations along a

two-dimensional coding scheme: (1) the content of themessages (position, direction, strategy, proximity to theobject, off-task) and (2) their pragmatic status (announce-ment, order, question, acknowledgement, corrections). Fig. 2shows examples of messages with the correspondingcategories. Inter-judge reliability of the coding system(performed on 25% of the sample) has proven to be goodas shown by a Cohen’s (1968) kappa of 0.89 for the contentvariable, a kappa of 0.86 for the pragmatics variable. Thiscoding scheme allowed us to count the frequency of eachcategory as well as the total number of messages exchangedby each player and then each group.

eme categories: messages content and messages pragmatics.

ARTICLE IN PRESS

Table 1

Duration of the three phases for each group (in minutes:seconds) in the

three conditions.

Synchronous MLA Control Student t p

Mean sd Mean sd

Phase 1 7:27 3:44 6:23 2:42 �0.73 0.47

Phase 2 3:57 1:54 4:22 1:54 0.5 0.62

Phase 3 3:19 1:35 5:43 2:47 2.3 0.03�

Total 14:44 4:02 16:29 5:44 0.78 0.44

� shows where p is significant.

Fig. 3. Group travel distance in the two experimental conditions (circle

corresponds to the mean while the black line is the median).

N. Nova et al. / Int. J. Human-Computer Studies 68 (2010) 451–467458

5.7. Operational hypotheses

Our hypotheses described in the introduction can then beoperationalized as follows:

H1. The first hypothesis postulates that the presence ofMLA improves collaborative task performance. Thismeans that we expect players from the MLA conditionsto find the object by covering a shorter distance than teamsin the control condition.

H2. The second hypothesis is that MLA improves themutual modeling within the group. We therefore expectplayers from the MLA condition to make fewer mistakeswhen drawing their partners’ path after the game than inthe control condition.

6. Quantitative results

6.1. Collaborative task performance (H1)

Our first hypothesis posited that the MLA tool wouldimprove the task performance. Since it was a collaborativegame, we analyzed the task performance at the group level,which corresponds to the group travel distance. Asdepicted in Fig. 3, groups in the 2 conditions had a verysimilar performance (Control: m=4859, sd=1670; MLA:m=5061, sd=1568); the only difference lay in thedispersion that is higher for players without the MLA. Aone-way ANOVA test2 did not show significant differences(F=0.07, p=0.78) between the performances of the twoexperimental conditions. This result then invalidates ourfirst hypothesis: the MLA tools do not facilitate the taskperformance. Let us already mention that the largevariance in performance reveals two sub-populations inthe Control condition: groups who managed to compensatefor the lack of MLA and others who had more trouble withthis.

Even though we have not used time as a performancemeasure, we found interesting differences between theexperimental conditions, especially while looking at theduration of each phase (exploration–rendezvousing–trian-gle formation). Table 1 depicts the results of such ananalysis, showing the duration of the three phases forgroups in each experimental condition.

Results show that groups in the Control condition tooksignificantly more time in the last phase. Players in the twoconditions took the same amount of time for the first twophases, but not for the forming of the triangle in the thirdphase. It seemed to require more time for players withoutthe MLA to carry out the triangle formation. Having theMLA tool seemed to facilitate this last phase of the gamebecause it requires a finer coordination. We also verified

2We used non-parametric tests such as Wilcoxon test when data were

not distributed normally or when the variances of the distributions were

not equal. When the distribution was normal, we used regular one-way

ANOVA analysis.

that there is no correlation between the time factor and ourexperimental condition.Let us have a look at the spatial strategies adopted by the

groups for each phase of the task. In Section 5.6, wedescribed two strategies chosen by groups to perform thetask: either spreading out over the campus or going all inthe same direction. In terms of performance the secondstrategy could be successful only if players found (bychance) the proper direction, which was not the case forone of the groups. Actually, only three groups chose thissecond strategy (one in the MLA condition, two in theControl condition). There was no significant differencebetween the conditions (w=0.66, p=0.71).In the second phase of the game, players adopted 3

different roles: the caller, the follower or the explorer. Thecaller was the first participant who saw that he/she wasclose to Bob through his/her proximity sensor, andtherefore called his/her partners. The two partners hadthen two possibilities: the ‘‘followers’’ went back on theirpath to directly join the ‘‘caller’’ while the ‘‘explorers’’joined him/her by taking another path as shown in Fig. 4.Thus, in phase 2 there was a clear distinction in terms of

ARTICLE IN PRESS

No optimization of the return: one caller (on the right) and two followers

Optimization of the return: one caller (on the right) and two explorers.

Optimization of the return: one caller (on the right), one follower (above) and one explorer (below).

Fig. 4. Evolution of the strategy after one of the players found Bob’s area and called the others.

N. Nova et al. / Int. J. Human-Computer Studies 68 (2010) 451–467 459

division of labor. Roles emerged differently in the twoexperimental conditions. There were more ‘‘followers’’ inthe MLA condition (w=6.35, p=0.04). In the awarenesscondition, a group was generally made up of a caller andtwo followers, whereas no-MLA teams often included onecaller, one explorer and one follower. Being a follower isnot an efficient way to join the caller: in this second phaseof the game, the object is still not well localized, whichmeans that the caller needs to keep wandering around.Explorers, by visiting parts of the campus, contributed tosearch efficiency. Since groups with MLA included more‘‘followers’’, it may be concluded that the location-awareness tool was detrimental to an efficient strategy.This spatial behavior (i.e. explorer/follower) reflects theevolution of the group strategy. ‘‘Followers’’ correspond toa heavy reliance on the strategy decided during theplanning phase (spreading over the campus and joiningas soon as possible the first player who found the object),whereas the ‘‘explorer’’ reflects the reshuffling of thestrategy. The result we described above shows that

players in the Control condition reshaped their strategymore, by adopting ‘‘explorers’’ behavior, rather thanbacktracking. On the contrary, MLA players sufferedfrom strategy persistence, by sticking to what they decidedduring the planning phase.

1.
Since there were no clearly established categories ofspatial behavior in the third phase we did not report theresults regarding this issue here.
6.2. Partners’ trails recall (H2)

To decide whether we compute the MM-accuracy at theindividual or group level, we followed the suggestions ofKenny et al. (1998): we checked the non-independence ofthe results by computing the intraclass correlation(r=0.39), which is significant (p=0.01). This means thatthe number of errors made by a subject is dependent on thenumber of errors made by his or her partners. We therefore

ARTICLE IN PRESS

Fig. 5. Number of errors made by each group during the post-test (while

drawing the path of the partner) in the two experimental conditions (circle

corresponds to the mean while the black line is the median).

Fig. 6. Frequency of shared map annotations written on the Tablet PC by

each individual in the two experimental conditions (circle corresponds to

the mean while the black line is the median).


used the group as the unit of analysis and summed thenumber of errors made by each individual within a group.Fig. 5 shows the number of errors per group in eachcondition.

We surprisingly found that players in the Control condi-tion made two times fewer errors than those who had theMLA (Control: m=4.90, sd=4.12; MLA: m=10.60,sd=7.26). A non-parametric Wilcoxon test showed thatthe difference between the two conditions was significant(W=61, p=0.02). Our second hypothesis is clearlyrejected: people in groups without the MLA better recalledtheir partners’ trails. This result will be explained in thenext findings regarding the map annotations.

6.3. Communication through shared map annotations

Map annotations have been investigated both byquantitative measures like the frequency and by qualitativecontent categorization. The shared map annotation fre-quency has been studied at the individual level since theintraclass correlation within the group was not significant(r=�0.21, p=0.87).

Fig. 6 shows the frequency of messages sent by players ineach experimental condition. The frequency of messageswas higher in the Control condition (Control: m=0.5,sd=0.19; MLA: m=0.31, sd=0.16). A non-parametricWilcoxon statistical test showed that this difference issignificant (W=55.56, po0.01).

We coded the content of the messages and theirpragmatics (as described in Section 5.6) and found adifferent repartition of messages: players in the Control

condition sent more messages about position (W=203,po0.01), direction (W=292, p=0.01) and strategy(W=269, po0.01) than those with the MLA. They also

asked more questions (W=228.5, po0.01). There were nosignificant differences concerning the number of orders oracknowledgements.In addition, we found a negative correlation between the

frequency of messages about strategy and the number oferrors made by the individual when drawing their partners’path (Pearson bivariate correlation r=�0.51, po0.001).This result was also confirmed by a regression analysis thatshowed a negative and significant relation between thenumber of errors and the number of strategy messages(b=�0.72, po0.001). Thus, the more the players sentmessages about strategy, the fewer the mistakes about theirpartners’ trails. The MM-accuracy was significantlycorrelated with the strategy messages but not with thetotal number of messages: what matters in this task is notto send a lot of messages but those relevant to strategy. Ifstrategy messages are very important to model the others’paths, it raises the question of what is memorized: thepath or the strategy? What is measured by this mutualmodeling index might be the reconstruction of thepath based on either the memorization of the partners’path or the memorization of group strategy. Theavailable data do not allow us to draw a conclusionabout which of the two might account for this pathreconstruction.For participants in the Control group, there was an

important relation between the exchanges of strategymessages and the mutual modeling, which was not thecase for those in the MLA condition. These results supportthe idea that players in different experimental conditionsplayed differently. The difference between the frequenciesof messages about strategy sent by players without theMLA and those with the MLA might explain the widedispersion in their performance.


7. Qualitative results

7.1. Use of coordination devices

As we described in our theoretical framework, thecoordination process is achieved through mutual modelingacts that consist in making inferences based on coordina-tion devices: where partners are or will be, if they are closeto Bob or to the triangle, if the partners understood thestrategy, etc. The coordination devices are the concreteitems on which these inferences can be drawn. Messagetranscriptions and the replay interviews allowed us tobetter understand how players exchanged coordinationdevices as well as the way they had been mutuallyrecognized by participants within a team. We observedfive categories of coordination devices (the descriptionbelow included excerpts from the post-game interview, alltranslated from French to English):

(1)
The plan that players set before starting the game wasthe most often cited coordination device. During thesefew minutes players discussed both the strategy theywanted to apply and the communication conventionsthey would use: ‘‘We decided that we would begin tosend messages only when having some signal on theproximity sensor’’ (Group 1, Control condition); ‘‘Weplanned that the first who had some signal would sendit to the other with the figure written on the map, beforeanybody should communicate anything’’ (Group 3,Control condition).
(2)

3Pseudonym we used to avoid using the participants’ real name.4Pseudonym we used to avoid using the participants’ real name.5We intentionally use the term ‘‘mentions’’ instead of ‘‘usage’’ because

Communication acts were the second most prominentcoordination device cited by players. Even thoughcommunication was achieved through a narrow med-ium (map annotations), important discussions aboutstrategies discussion have been undertaken. The ex-cerpts below provide some examples of dialogues (asplayers recalled them): ‘‘I knew they were joining mebecause I drew a circle on the map where I thought Bobwas, and they said ok we’re coming’’ (Group 11, MLAcondition); ‘‘At that moment, I asked her what she wasdoing and she told me that she was backtracking’’(Group 18, MLA condition). Dialogue occurrences likethese were important for trajectory awareness anddivision of labor. However, most of the communicationacts were not proper dialogues but announcements ofproximity signal information (‘‘I was telling them myreadings, on a regular basis’’: Group 22, MLAcondition), self-declared positions (‘‘As soon as I gota reading I would put it and I would tell him whereI was’’: Group 18, MLA condition) or trajectories (‘‘Hedrew an arrow, he marked the reading he was givingand the general direction where he was heading’’:Group 5, Control condition). Messages about strategywere also important as confirmed by the quantitativeanalysis in Section 6.3.

we drew this conclusion about participants verbalizations after the game
(3) and not during it.
The mutual location-awareness tool has been cited as animportant coordinative device by players in the MLA

condition: ‘‘I saw that he was going in that direction,then I joined him’’ (Group 18, MLA condition), ‘‘I sawSandra3 heading to the CE building so I thought wehad to join her’’ (Group 21, MLA condition), ‘‘I knewthey were arriving through the IN building becauseI saw their last positions’’ (Group 5, Control condition).Section 7.2 explores the question of how MLA has beeninterpreted.

(4)
Knowledge about the partners was also a powerfulcoordination key for some groups as explained by thesetwo players: ‘‘I saw that Sandra4 was not moving, butI know her, she always moves and she’s not lazy, so shewas moving’’ (Group 26, MLA condition), ‘‘I saw hewas coming from downstairs but I know him, I thoughthe would take the lift instead of the stairs’’ (Group 3,Control condition). By knowing a partner’s habit andbehavior, players could infer some meaning about thecourse of action.
(5)
Knowledge about the environment was also a key tocoordination. Here are examples of how two playersexpressed this: ‘‘There is a bottleneck here between theCE and CM buildings, I knew he could only arrivefrom there’’ (Group 11, MLA condition), ‘‘I did notknow he were joining me or go elsewhere but I knewthere were chance that I would meet him at La Coupole[center of the campus]’’ (Group 6, Control condition).
This list of coordination devices fits with Clark’s model ofcoordination that we presented in Section 3 as theycorrespond to his typology of devices. Knowledge aboutpartners and the environment are ‘‘Precedent’’ (when aprecedent experience allows participants to form someexpectations about others’ behavior). Dialogue acts are‘‘Explicit Agreement’’ (when the participants explicitlyacknowledge the information exchanged). The groundedplan discussed at the beginning is a set of ‘‘Conventions’’(when conventional procedures are set by the participants).And finally, manifest elements from the environment;topology, network coverage, MLA information and non-acknowledged communication acts are ‘‘Manifest elements’’.The description of the players’ coordination devices

enables us to differentiate the coordination process in thetwo conditions. Since the results partially emerged fromprovoked verbalizations (answers to a semi-structuredinterview while showing a replay of their activity on amap of the campus), we did run statistics about them. Wesimply looked at how often coordination devices havebeen mentioned. Table 2 shows the different uses ofcoordination devices by the two experimental groups.Table 2 shows that players with the MLA mentioned5

the use of conventions decided before the game and the use

ARTICLE IN PRESS

Table 2

Description of the different kinds of coordination devices used by each group.

Coordination devices mutually recognized Experimental group: MLA Control group (no-MLA)

Conventions (set in planning phase) Cited by most groups Cited by most groups

Manifest (a): information shown by the awareness tool Cited by most groups Never mentioned

Manifest (b): manifest information due to environmental/topology constraints Cited by some groups Cited by most groups

Manifest (c): message sent but not acknowledged by the partner Cited by some groups Cited by most groups

Explicit agreement (messages acknowledged by partners) Cited by some groups Cited by most groups

Precedent (knowledge about the partners’ habits and behavior) Cited by some groups Cited by most groups


of the awareness tool to coordinate. More importantly,they employed them without discussing the use of othercoordination devices; the environment is seldom mentionedas being taken into account for inferring what the partnerswere up to. When looking at the players without MLA, theelements used for coordination were also the plan/convention decided before the game and informationabout location-awareness (explicitly described because theydid not have the MLA) but also the manifest elements ofthe environment and explicit agreements negotiated overthe course of the game. Therefore they better took intoaccount contextual and negotiated elements than playerswith the MLA.

We have seen in Section 6.1 that players in the MLAcondition had more behavioral consistency, resulting inmore followers. This strategy inertia can be retrieved inthese verbalizations. Players without the MLA had lessinertia by better reshaping their strategy. What is strikinghere is that automating the exchange of a coordination keycan be detrimental to task coordination since players withMLA were stuck on one device and did not exchangeothers. What is important for players with the MLA isthat location-awareness is the relevant coordination key forthe whole task; they do not have to renegotiate it for thedifferent phases. Section 6.1 revealed that the importanceof location-awareness depends on the game phase. It mightbe the case that groups with the MLA wasted cognitiveresources monitoring others’ positions and thus discussedless which coordination device would be valuable at thatmoment. This means that automating awareness can have anegative effect on coordination by decreasing the exchangeof useless coordination devices (at least during phases 1and 2).

7.2. Interpretations of mutual location-awareness

Analyses of post-game interviews brought forward threecategories of MLA interpretations and uses. First, MLAwas used for activity awareness (Group 24: ‘‘It helped to getwhat the others were up to’’; Group 11: ‘‘To see that thered was close to the end’’). Second, MLA was employed for

implicit communication: participants did not need tocommunicate their precise position around the triangle inthe final phase of the game given that the movement ofparticipants in space was captured and conveyed to eachother by the MLA. To some extent, the communication

between players was made implicit, as shown by thefollowing quotes: ‘‘Otherwise, we would have been forcedto make more annotations to orientate each other’’, ‘‘I sawthat he was going in that direction, then I did notcommunicate that much’’, ‘‘It’s convenient to see wherepartners are, even without communication we knew thatthe object was not in the South or in the East, we inferred itwas in the West’’. And third, MLA was a resource for joint

actions: knowing partners’ location was a resource forchoosing which individual action to undertake: eithermoving into a certain direction or contacting a player tohelp him or her. The person who found Bob first often usedMLA to indicate the shortest path to his partners: ‘‘I knewwhere they were so I was able to indicate to them theshortest path to me’’ (Group 11). Additionally, playersmentioned how MLA was a resource for forming thetriangle at the end of the game: ‘‘It was useful to form thetriangle around Bob, I knew where to find my ownposition’’ (Group 18).An interesting issue regarding MLA is how players

reacted to discrepancies caused by the system. As a matterof fact, the information conveyed by the MLA is some-times imperfect. The reasons for this are diverse: patchywireless coverage, uneven connectivity, outdoor conditions(e.g. walls, humans, rain) creating disturbances. Playersspontaneously mentioned how they dealt with theseuncertainties. When confronted by a discrepancy concern-ing their partner’s position, three types of reactions hadbeen mentioned: believing the system, saying that thesystem was wrong (as reported by those players fromGroup 11: ‘‘I saw that it was indicated that B waspositioned here but he was not,’’ ‘‘I saw that B moved onthe screen but I know he did not’’) or not understanding(‘‘I did not get why he was there’’ said a skepticalparticipant from Group 21 with the MLA). When weasked them why they questioned the information providedby the system, players said that this was contradictory towhat they had in mind. For instance, a participant told usthat ‘‘Judging from my experience, the network coverage inthe MX is low, so I thought that was why player A was notmoving’’ (Group 17, MLA). This shows how anothercoordination device, i.e. knowledge about the campusdrawn from a precedent experience or a discussion withothers, could make him doubtful of the MLA information.It also brings forward the fact that coordination devicescan be conflicting, by providing divergent information.


In summary, when facing a potential discrepancy, partici-pants used other coordination devices (knowledge aboutthe network, explicit agreement, conventions, precedentexperience with participants).

8. Discussion

8.1. Results summary

This field study has showed that MLA does not improveteam performance in the whole game: the difference in taskperformance between groups is not significant. However,we found that automatic MLA could be more effective fortasks that would require finer grained coordination (suchas the triangle formation). We also found that automaticMLA could lower the accuracy of mutual modeling.Groups without location information built a better mutualmodel: they made fewer errors when drawing the path oftheir partners after the game because the locationinformation were more explicit, and thus easier toremember. In addition, the automatic MLA not onlyundermined the exchange of messages about position butalso about other kinds of information such as strategy ordirection, which were not provided by the system. Theautomatic MLA also damaged the quality of collaborationas it increased strategy persistence in terms of communica-tion and spatial movements. Looking at the players’strategy, players with the MLA interface changed theirstrategy less often during the game and maintained thedivision of labor they decided before the game. The reasonwhy MLA increased inertia is probably that it reducedcommunication and hence subjects were less articulateabout their strategy. As we stated, the presence of MLAput more emphasis in this specific coordination device andless on the others such as explicit agreements. In otherwords, the automatic awareness tool seems to make userspassive with respect to spatial coordination.

Finally, our study extends the list of roles of MLA wedescribed in the literature (Brown et al., 2003; Benfordet al., 2004; Dearman et al., 2005; Brown et al., 2007).First, MLA was used to elaborate a shared understandingof the situation. Second, MLA allowed participants to gainan implicit awareness of their partners’ activities by makingthem manifest and comparable to previous actions. As aconsequence, the MLA tool regulated the division of labor.Though MLA reduced strategic discussions, it was none-theless used in verbal interactions to acknowledge anutterance or to give an order.

Furthermore, our results must be considered in light ofcertain limitations, especially concerning the settings(decentralized collaboration activity) and the interface(map annotations). Another limitation of our study is theshort duration of the experiment, which hence requiredsynchronous coordination between peers. Our groupsonly played one game, which might be an issue in termsof learning the different elements of the situation: theinterface, the task and its rules. One possible response to

see whether the results still hold over time is repeatedplay as described in Barkhuus et al. (2005) or a crossedexperiment in which players from one condition play asecond game in the other condition. Finally, the fifthlimit is our testing of awareness tools with triplets; in thecontext of multi-user systems with 4–50 users, the use ofawareness tools should change. Paying attention toawareness cues left by 50 users would be more complicatedand difficult than testing those left by groups of two orthree.

8.2. Two ways of playing

Our field study deepens the results we described in theliterature review concerning automating MLA versusself-disclosure (Benford et al., 2004; Iachello et al., 2005;Consolvo et al., 2005) or with current research aboutawareness and interruptions (Romero et al., 2007; Romeroand Markopoulos, 2009) as we revealed more detaileddifferences. Our results especially allow one to gobeyond the claim by Benford et al. (2004) that self-disclosure of location-awareness conveys more informationthan solely location. Let us have a look at the differencesbetween the Control condition and players in the MLA

conditions.

8.2.1. The inhibition of communication

One of the most striking differences was the commu-nication pattern. Players in the control groups exchangedmore map annotations about positions, direction, strategyand questions. This had been confirmed by the qualitativeanalysis in which we saw more explicit agreement anddiscussion for this group. The presence of the MLAinterface led players to take for granted that MLA wouldbe sufficient to achieve their joint activity. This can beexplained by a variant of the least collaborative effortprinciple (Clark, 1996): participants used the coordinationdevices that were available (mostly the plan and the MLA)and were less bothered to send more information throughshared map annotations. Players in the Control conditiononly had the plan as a coordination device. It also meansthat Control players anticipated the need for locationinformation: they not only sent messages about theirdirection but also about their strategy: their teammatescould then better infer a mutual model. What this implies isthat annotation around location is better than simplyinformation about whereabouts: it gives a social contextthat is helpful to interpret locational data. This contextindeed helps participants to draw inferences aboutpartners’ activities.

8.2.2. The importance of human agency

By agency, we refer to the capability of an individual todecide whether or not to act. In the context of coordina-tion, agency describes the process by which a partner doesor does not make manifest a sign of mutual intelligibility byexchanging a coordination device. This is where the main


difference lay between the different experimental condi-tions: automating MLA is different from sending one’sposition in space. Indeed, participants from the Control

condition chose the information they perceived as relevant(position, direction and strategy) for the time being andsent them to their partners. Compared to automaticpositioning in which location is just information, self-declared positioning is both an information and an act ofcommunication act, intentional by definition. If A tells Bwhere he or she is located, not only B knows A’s locationbut he or she also knows that A considers that it is usefulfor B to know it. This allows us to deepen the claim byBenford et al. (2004) about how self-reporting locationbecomes a communication practice; our study has indeedshowed how this self-disclosure could convey intentionalityor can be employed in dialogues. In addition, as describedby Consolvo et al. (2005), an important factor in MLAusage concerns the knowledge of who requested theinformation and what level of detail would be useful tothe requester. Self-reported MLA was more efficient thanautomatic MLA for that matter, since location-awarenesswas not only dots on a map but also strategy and questionsabout it. It then helped the players to give their partners theinformation they needed with the adequate granularity.This concern for intentionality that characterizes thereplaced communication acts has emerged from studies inthe field of availability awareness (Romero et al., 2007;Romero and Markopoulos, 2009). Our work extends thisconcern to location-awareness.

8.2.3. Fewer coordination devices

The qualitative analysis of the data showed the differentsets of coordination devices used by players from thevarious experimental conditions. The accumulation ofcoordination devices over time in Control groups wasricher than that of the MLA groups. Since players withoutthe MLA exchanged more map annotations, they collectedmore information about the situation, which would be whythe richness of communication had a positive effect on themutual modeling accuracy.

8.2.4. MLA is locally effective

Another factor of differentiation between the twodifferent ways of playing is to look at the performance ofeach group for each phase. This discrimination allows us tomoderate the critique of the location-awareness tool.Actually, we saw that when looking at the task perfor-mance for each phase, the MLA interfaces had differenteffects. For phases 1 and 2, the giving of one’s locationcould be replaced by verbal communication, which was notthe case in the last phase where people had needed a tightcoordination. Therefore, as in Espinosa et al. (2000), weconclude that awareness impacts differ depending upon thetask. This is obviously a starting point for new fieldexperiments in order to investigate the articulation betweentasks and levels of automation with awareness interfaces.

8.3. Design implications

Although this experiment was a semi-controlled studythat engaged participants in a basic task, we believe thatthese results can have relevant implications for the designof location-awareness tools for tasks requiring coordina-tion.

8.3.1. Awareness as a systemic process

These studies have shown that awareness should bethought as a global and systemic issue. The awareness toolsuch as the one we implemented is not simple channels ofinformation; the information they convey are part of acomplex system. A simple modification in the conditionsleads to tremendous changes in participants’ behavior: theautomation of MLA lowered the mutual modeling withingroups, inhibited group communication and made usersless active. This is related to the critical discourse aboutcognitive augmentation and so-called ‘‘cognitive pros-theses’’. MacLuhan (1964) used to say that ‘‘Any inventionor technology is an extension or self-amputation’’ (p. 45),which highlights how extending human capabilitiesthrough technology often undermines other skills. Thisphenomenon has also been discussed in the field of human–computer interaction by researchers such as Woods (1997),who advocates against ‘‘clumsy automation’’, which isexactly what we observe here. Reducing the effort toperform an action (being aware of others’ activities) is ofcourse important but when this effort reduction is toopronounced, there is a less acute self-awareness inperforming these activities. Our first take-away is then thatMLA is not a simple interface widget, it is an importantimplicit communication tool, by turning users’ movementsinto an awareness tool. Communication tools should bedesigned to support references to locations.

8.3.2. Self-disclosure versus automation

This study emphasized the trade-off between automationand human agency. This trade-off is indeed illustrated bythe difference between self-positioning and automatedpositioning. Letting people build their own representationof the spatial information appears to be more efficient thanbroadcasting mere location information. To some extent,not giving location-awareness was a way to supportcollaboration more effectively; since players communicatedmore and better explained their activity and intents, whichled interestingly to the reshaping of their strategy. This factis of particular importance because it shows the differencebetween automatic positioning in which location is justinformation versus self-declared positioning, which is bothinformation and an act of communication, intentional bydefinition. Our study showed that the intentional coordina-tion devices are more important to collaboration thanthose that are automated. Another advantage for self-disclosing one’s location is that it allows people to employthe location names that make sense for the participants.This is related to the distinction between ‘‘space’’ and


‘‘place’’ (Harrison and Dourish, 1996). The difficulty oflocation-based applications in conveying a meaningfulsemantic of places makes it more efficient to let usersexpress their location by using their own description, atopic already discussed by Persson and Fagerberg (2002).This finding, that it is better to let people express their ownlocation, is confirmed by what Benford et al. (2005)revealed: self-reported positioning could be a reliablelow-tech alternative to automated systems like GPS.However, our findings go further by proving that lettingusers declare their position themselves is better with regardto various processes like communication or the construc-tion of a mental model about the partners. Our second

take-away is then that the future of location-awarenessapplications can go beyond the opposition between self-disclosure of one’s location or automating MLA and lie inthe combination of both. For example, if we take thecurrent synchronous MLA interface of CatchBob! thatdisplays the partners’ location as dots, it would beinteresting to add a circle around these dots when a playerexplicitly gives his or her location. This way, the intentionalpart of the message about position is conveyed.

8.3.3. An optimal collaborative effort

Our experiment showed that effective collaborationrequires participants to be involved in an optimal collabora-tive effort. Depending on the task, the awareness tool canreduce this effort. For example, in CatchBob, forming thetriangle was more efficient with the MLA tool. But below acertain effort threshold, MLA had the opposite effect. In thetwo first phases of the game, self-disclosure was moreefficient than the MLA because it allowed players tocommunicate more information than simply location. Asdiscussed by Dillenbourg and Traum (2006), the minimalcost is not always the optimal cost for collaboration. Whatour experiment showed is that the reduction of thecollaboration cost, through the introduction of an awarenesstool like the MLA interfaces, was indeed not always optimal.Furthermore, this ‘‘optimal collaborative zone’’ that wedescribe reflects two efforts depending on who is considered:

�
The producer of a coordination device provides thepartners with elements that he or she thinks are relevantwith regard to the situation and the common ground ofthe group. For this player, the cost is lower when there isan awareness tool because he or she does not have toproduce coordination devices. � The addressee(s) perceives those coordination devices,
integrates them in their common ground and potentiallyperforms mutual modeling acts based on them. For anaddressee, it is preferable to have punctual andintentional coordination devices, and not a continuousbroadcast of information. This is because the intentionalmessages have more weight in the common ground; theyindeed convey the implicit cue that they have been sentout because they would help the partner(s) to havemutual expectations.

With intentional acts of communication, one can alsosay that the producer has assessed the importance of theinformation and that this lowers the effort for theaddressee. The optimal collaborative effort then reflectsthe effort asymmetry between the producer and theaddressee. What this thesis proves is that the intentionalityof the coordination devices sent by the producers is morerelevant than information from automated devices. Theyare more efficient from a collaborative standpoint becausethey facilitate the recognition of pertinent informationversus noise. That being said, it is necessary to keep inmind that these principles partly depend on the situationand the granularity of the joint activity carried out byparticipants.The last take-away is that designers should not aim for

the maximum reduction of the collaborative effort. Thechallenge is to provide producers with an interface thatdecreases the effort of sending a coordination device butthat lets addressees give the right weight to this informa-tion. For that matter, a timely exchange of coordinationdevices is better than a continuous one. For instance, inCatchBob, it is better to give a partners’ location when he/she undertakes a specific action (like sending a message)than continuously broadcasting it. This way, it reduces thenoise caused by the permanent positioning of others.

9. Conclusion

The field study of a pervasive game allowed us to deepenexisting research about how mutual location-awarenessinfluences coordination. Using a psycholinguistic frame-work, we examined in details the differences betweenautomating MLA and the self-reporting of users’ where-abouts. Our results have highlighted the importance ofintentionality and human agency, complementing theresults of Benford et al. (2004). It did so through aquantitative study that can complement their qualitativeanalysis and it went deeper in the analysis of why self-disclosure can indeed be more effective than automatictracking in a specific task. This study also described inmore details how MLA was interpreted and employed inthe context of coordination. Further research is nowaddressing how environmental features such as topographyand infrastructures were employed as coordination devicesin the game.

Acknowledgments

This work was partly funded by a grant from the SwissNational Research Foundation (Grant no. 102511-106940).The CatchBob! project also benefited from all the players andthe good inspiration of heavy dubstep tunes.

References

Arminen, I., 2006. The social functions of location in mobile telephony.

Personal and Ubiquitous Computing 10 (5), 319–323.


Barkhuus, L., Chalmers, M., Tennent, P., Hall, M., Bell, M., Brown, B.,

2005. Picking pockets on the lawn: the development of tactics

and strategies in a mobile game. In: Proceedings of the 2005

Ubiquitous Computing Conference (Ubicomp 2005), Tokyo, Japan,

pp. 358–374.

Benford, S., 2005. Future location-based experiences. JISC Technical

Report (TSW0501).

Benford, S., Crabtree, A., Flintham, M., Drozd, A., Anastasi, R., Paxton,

M., Tandavanitj, N., Adams, M., Row Farr, J., 2006. Can you see me

now? ACM Transactions on Computer–Human Interaction (TOCHI)

13 (1), 100–133.

Benford, S., Rowland, D., Flintham, D., Drozd, A., Hull, R., Reid, J.,

Morrison, J., Facer, J., 2005. Life on the edge: supporting collabora-

tion in location-based experiences. In: Proceedings of the 2005 ACM

Conference on Human Factors in Computing Systems (CHI’2005),

Portland, USA, 2005, pp. 721–730.

Benford, S., Seager, W., Flintham, M., Anastasi, R., Rowland, D.,

Humble, J., Stanton, D., Bowers, J., Tandavanitj, N., Adams, M.,

Row-Farr, J., Oldroyd, A., Sutton, J., 2004. The error of our ways: the

experience of self-reported position in a location-based game. In:

Proceedings of the 2004 Ubiquitous Computing Conference (Ubicomp

2004), Nottingham, UK, pp. 70–87.

Beresford, A.R., Stajano, F., 2002. Location privacy in pervasive

computing. IEEE Pervasive Computing 2 (1), 46–55.

Brown, B., McColl, I., Chalmers, M., Galani, A., Randell, C. and Steed,

A., 2003. Lessons from the lighthouse: collaboration in a shared mixed

reality system. In: Proceedings of the 2003 ACM Conference on

Human Factors in Computing Systems (CHI’2003), Fort Lauderdale,

Florida, ACM Press, NY, pp. 577–584.

Brown, B.A.T., Taylor, A.S., Izadi, S., Sellen, A., Kaye, J.J., Eardley, E.,

2007. Locating family values: a field trial of the whereabouts clock

(Best Paper Nominee). In: Proceedings of Ubicomp 2007, Innsbruck,

Austria, pp. 354–371.

Chalmers, M., Juhlin, O., 2005. New uses for mobile pervasive games—

lessons learned for CSCW systems to support collaboration in vast work

sites. Paper for the Workshop ‘‘Computer Games & CSCW’’ at the

European Conference on Computer-Supported Cooperative Work 2005.

Retrieved from the Internet (15 July 2007): /http://www.tii.se/mobility/

Files/NewUsesforMobilePervasiveGames050620.pdfS.Clark, H.H., 1996. Using Language. Cambridge University Press,

Cambridge.

Cohen, J., 1968. Weighted kappa: nominal scale agreement with provision

for scaled disagreement or partial credit. Psychological Bulletin 70,

213–220.

Consolvo, S., Smith, I., Matthews, T., LaMarca, A., Tabert, J., Powledge,

P., 2005. Location disclosure to social relations: why, when, & what

people want to share. In: Proceedings of the Conference on Human

Factors and Computing Systems: CHI 2005. ACM Press, pp. 81–90.

Newsletter 10 (6), September 2003.

Creswell, J.W., 1994. Research Design: Qualitative and Quantitative

Approaches. Sage, Thousand Oaks, CA.

Dearman, D., Hawkey, K., Inkpen, K.M., 2005. Effect of location-awareness

on rendezvous behaviour. CHI Extended Abstracts, 1929–1932.

Dillenbourg, P., 1999. What do you mean by ‘‘collaborative learning’’? In:

Dillenbourg, P. (Ed.), Collaborative Learning: Cognitive and Compu-

tational Approaches. Elsevier, Oxford, UK, pp. 1–19.

Dillenbourg, P., Traum, D., 1997. The role of a whiteboard in a

distributed cognitive system. Paper presented at the Swiss Workshop

on Collaborative and Distributed Systems, Lausanne, Switzerland,

1 May 1997. Retrieved on 15 January 2007 from: /http://agora.unige.

ch/conferences/Lausanne97/dillenbourg/S.

Dillenbourg, P., Traum, D., 2006. Sharing solutions: persistence and

grounding in multimodal collaborative problem solving. Journal of the

Learning Sciences 15 (1), 21–151.

Dourish, P., Bellotti, V., 1992. Awareness and coordination in shared

workspaces. In: Turner, J., Kraut, R. (Eds.), Proceedings of the ACM

Conference on Computer Supported Cooperative Work (CSCW’92),

Toronto, Canada, pp. 107–114.

Dyck, J., Gutwin, C., 2001. Where are you and what can you see?

Maintaining location-awareness in collaborative 3D workspaces.

Technical Report HCI-TR-2001-02.

Espinosa, A.C., Rico-Gutierrez, J.J., Kraut, R.E., Scherlis, W., Lauten-

bacher, G., 2000. Coming to the wrong decision quickly: why

awareness tools must be matched with appropriate tasks. In:

Proceedings of the ACM CHI 2000 Human Factors in Computing

Systems Conference, 1–6 April 2000, The Hague, The Netherlands,

pp. 392–399.

Goodman, J., Brewster, S., Gray, P., 2004. Using field experiments to

evaluate mobile guides. In: Schmidt-Belz, B., Cheverst, K. (Eds.),

Proceedings of the HCI in Mobile Guides. Workshop at Mobile HCI

2004, Glasgow, UK, September 2004.

Gutwin, C., Greenberg, S., 2002. A descriptive framework of workspace

awareness for real-time groupware. Computer Supported Cooperative

Work 11 (3), 411–446.

Harrison, S., Dourish, P., 1996. Replacing space: the roles of place and

space in collaborative systems. In: Jones, Q., Halverson, C. (Eds.),

Proceedings of the 1996 ACM Computer Supported Cooperative

Work (CSCW 96). ACM Press, Cambridge, MA, pp. 67–76.

Hightower, J., Borriello, G., 2001. Location systems for ubiquitous

computing. IEEE Computer 38 (8), 57–66.

Hudson, S.E., Smith, I., 1996. Techniques for addressing fundamental

privacy and disruption: tradeoffs in awareness support systems,

Proceedings of the CSCW’96: ACM Conference on Computer-

Supported Cooperative Work. ACM Press, Cambridge, MA

pp. 238–247.

Iachello, G., Smith, I., Consolvo, S., Abowd, G.D., Hughes, J., Howard,

J., Potter, F., Scott, J., Sohn, T., Hightower, J., LaMarca, A., 2005.

Control, deception, and communication: evaluating the deployment of

a location-enhanced messaging service. In: Proceedings of the

Ubicomp 2005, Tokyo, Japan, September 2005, pp. 213–231.

Kenny, D.A., Kashy, D., Bolger, N., 1998. Data analysis in social

psychology. In: Gilbert, D., Fiske, S., Lindzey, G. (Eds.), Handbook

of Social Psychology, vol. 1. McGraw-Hill, Boston, pp. 233–265.

Laurier, E., 2001. Why people say where they are during mobile

phone calls. Environment and Planning D: Society & Space 19 (4),

485–504.

Lewis, D.K., 1969. Convention: a Philosophical Study. Harvard

University Press, Cambridge, MA.

Licoppe, C., Inada, Y., 2005. ‘‘Seeing’’ one another onscreen and the

construction of social order in a mobile-based augmented public space:

the uses of a geo-localized mobile game in Japan. In: Proceedings of

the 2005 Learning in the Mobile Age Conference. Hungarian Academy

of Sciences, 28–30 April 2005.

MacLuhan, M., 1964. Understanding the Media: the Extensions of Man.

McGraw-Hill, New York.

Nova, N., Wehrle, T., Goslin, J., Bourquin, Y., Dillenbourg, P., 2007.

Collaboration in a video game: impacts of location awareness. Journal

of Multimedia, Tools and Applications 32, 161–183.

Nilsen, T., Linton, S., Looser, J., 2004. Motivations for AR gaming. In:

Proceedings of the Fuse 04: New Zealand Game Developers

Conference, Dunedin, New Zealand, pp. 86–93.

Persson, P., Fagerberg, P., 2002. GeoNotes: a real-use study of a public

location-aware community system. Technical Report T2002:270, SICS,

Stockholm.

Romero, N.A., Szostek-Matysiak, Al, Kaptein, M.C., Markopoulos, P.,

2007. Behaviours and preferences when coordinating mediated

interruptions: social and system influence. In: Proceedings of the

CSCW 2007, Springer LNCS, pp. 351–370.

Romero, N.A., Markopoulos, P., 2009. Grounding interpersonal

privacy in mediated settings. In: Proceedings of the ACM 2009

International Conference on Supporting Group Work, Sanibel Island,

FL, USA, 10–13 May 2009. GROUP ’09. ACM Press, New York,

pp. 263–272.

Rudstrom, A., Hook, K., Svensson, M., 2005. Social positioning:

designing the seams between social, physical and digital space. In:

First International Conference on Online Communities and Social

&lang;http://www.tii.se/mobility/Files/NewUsesforMobilePervasiveGames050620.pdf&rang;

&lang;http://www.tii.se/mobility/Files/NewUsesforMobilePervasiveGames050620.pdf&rang;

&lang;http://agora.unige.ch/conferences/Lausanne97/dillenbourg/&rang;

&lang;http://agora.unige.ch/conferences/Lausanne97/dillenbourg/&rang;


Computing, at HCII 2005. Lawrence Erlbaum Associates, Las Vegas,

USA, 24–27 July 2005.

Schelling, T., 1960. The Strategy of Conflict. Harvard University Press,

Cambridge.

Smith, I., Consolvo, S., Hightower, J., Hughes, J., Iachello, G., LaMarca,

A., Abowd, G.D., Scott, J., Sohn, T., 2005. Social disclosure of place:

from location technology to communication practice. In: Proceedings

of the Pervasive 2005, LNCS, vol. 3468. Springer Verlag, pp. 134–151.

Theureau, J., 1992. Le cours d’action: analyse semio-logique. Editions

Peter Lang, Berne.

Weilenmann, A., 2003. ‘‘I can’t talk now, I’m in a fitting room’’:

availability and location in mobile phone conversations. Environment

and Planning A 35 (9), 1589–1605 (special issue on Technology and

Mobility, Laurier, E. (Ed.)).

Weiser, M., 1991. The computer for the 21st century. Scientific American

265 (3), 94–104.

Woods, D.D., 1997. Human-centered software agents: lessons from clumsy

automation. In: Flanagan, J., Huang, T., Jones, P., Kasif, S. (Eds.),

Human Centered Systems: Information, Interactivity, and Intelligence.

National Science Foundation, Washington, DC, pp. 288–293.

the effects of mutual location-awareness on group coordination

Documents