Integrating awareness incooperative applicationsthrough the reaction-diffusionmetaphor

Carla SimoneUniversity of Torino, simone@di.unito.it

Stefania Bandini University of Milano, bandini@disco.unimib.it

AbstractThe paper discusses the notion of awareness from the point of view of thedesign of a supportive technology. This perspective requires a deeperunderstanding of the ways and means people adopt to deal with awarenessinformation as well as considering the integration of awareness tools with toolssupporting other forms of coordination. First, we suggest to consider two typesof awareness: by-product awareness that is generated in the course of theactivities people must do in order to accomplish their cooperative tasks; andadd-on awareness that is the outcome of an additional activity, which is a neatcost for the cooperating actors in relation to what they must do and isdiscretional in that it depends on actors' evaluation of the contingentsituation. Secondly, we propose a reaction-diffusion metaphor to describe theawareness phenomenology and to take into account the two above-mentioned typesof awareness and integration. The model of awareness derived from the metaphormakes visible and accessible by different types of users a set of elementalprimitives whose flexible composition allows them to construct the awarenessmechanisms they dynamically need. These primitives are incorporated in asoftware module that can be used in combination with coordinative applicationsfor sake of promoting awareness information. The main architecture of themodule is presented together with its interoperability with the targetapplication; moreover, a simple example illustrates how the incorporatedprimitives can be used to build awareness mechanisms.

Keywords: cooperation, metaphors, awareness model, CSCW architecture

1. Introduction

In the last years the theme of awareness became very popularin the CSCW literature. A series of field studies elaboratedon the concept by highlighting how coordination achieved inreal situations is based on different forms of awareness.From the standpoint of the design of a supportivetechnology, different approaches have been proposed toprovide various applications with awareness capabilities bymeans of different technical solutions. Despite thishuge amount of work, the notion of awareness didn't reach anunderstanding that goes beyond a common sense concept. Forexample, there is a lack of a precise definition of thisconcept in the framework of CSCW. To our knowledge, asystematic analysis of the concept is contained in (Schmidt,1998) starting from the definition of the term awarenessgiven in English dictionaries. Specifically, two mainmeanings are reported: awareness as 'having knowledge of'and awareness as 'sentience to, elementary consciousnessof'; in addition, the author says that 'Within CSCW the term'awareness' is generally been used in the second sense ...'.There is of course a difficulty in dealing with this term(also in our native language) because of the many facetsimplied by a definition involving 'knowledge' and'consciousness'. Moreover, we believe that within CSCW thetwo meanings are used in a mixed way. Consequently, as theterm awareness is becoming very popular, its not clarifiedusage can easily overload its meaning and, because of this,generate confusion when a supportive technology has to bedesigned. Our contribution to the clarification of theawareness concept is in the light of reducing this risk.Focusing on the design of a supportive technology requiresconsidering awareness as a mode of coordination and, namely,to recognize different types of actors' ability: the abilityto perceive a stimulus (e.g., a sound); the ability toanalyze it (e.g., identify its source); and finally, theability to interpret it in the given context (e.g., as analarm or as an expected state of a resource). Theseabilities characterize who is 'aware of' the stimulus, thatis, of the information pertaining to awareness (hereafter,awareness information). In speaking about whom provides theawareness information the term 'awareness promotion' isusually adopted. If the above abilities make sense, thenpromoting awareness involves: acting on the perception(e.g., strengthening the stimulus); enhancing its analysis (

e.g., 'presenting' the information in a suitable way); andfinally, supporting its interpretation (e.g., includinginformation about the context where the interpretation hasto be performed). These three types of 'promotion' raisedifferent requirements for a technological support, not onlyat the user interface level but also in terms of medium andcontent. In this respect, this paper focuses on the firstform of promotion, and marginally on the other ones.More recently, the 'awareness cycle' identified by the

above three forms of ability/promotion has been enriched byconsidering awareness as a means to establish conventionsamong the cooperating actors. This aspect is discussed in(Mark, 2000) through the notions of stimulus, conventions,commitment, learning process and memory. We refer to thatpaper for an illustration of the problems and requirementsthese interrelated notions raise for the supportivetechnology, and as an example of how interdisciplinarystudies can bring new insights on the complexity of thephenomena related to the issue of awareness in cooperation.The relevant point to our paper is the emphasis on theoverhead activities required to promote and maintain theawareness necessary to identify and preserve thoseconventions. The latter become a fundamental component ofthe background where to interpret, iteratively, awarenessinformation (together with information pertaining to othermodes of coordination). In this paper, we analyze thenotion of awareness from the point of view of the additionaleffort these activities require to the involved actors bydistinguishing between what we call by-product and add-onawareness in the light of identifying the implications on thedesign of a supportive technology. The software module wepropose incorporates a model of awareness, based on the reaction-diffusion metaphor, and is able to mesh by-product and add-onawareness behaviors in a natural and innovative way.

Finally, there is a shared consensus that the discovery ofthe awareness phenomenology was an important step in CSCWevolution as it highlighted a new perspective in looking atcooperation, especially when its coordinative aspects areconsidered. Any discovery brings with it the necessity tounderstand how its outcomes are to be integrated in theresults already achieved in the related field. Thisapplies also to the case of awareness. An effort is neededto understand how the new perspective, and the related formsof coordination, can be integrated with other forms ofcoordination, possibly supported by specialized cooperativeapplications. The paper aims to contribute to this effortby elaborating on the implications of taking into accountawareness in the construction of an integrated technologicalsupport of cooperation.

The paper is organized as follows. The next section analyzesthe notion of awareness in order to get a deeper insight onits meaning and use whereas Section 3 presents aninterpretation of the role of awareness in coordination.Section 4 derives from the previous sections somerequirements for a technology supporting the integration ofawareness capabilities with other supports of cooperation.Then the paper illustrates the proposed approach in terms ofunderlying metaphor and awareness model (Sections 5 and 6),and describes the implementation of the model in a softwaremodule devoted to support awareness promotion in cooperativeapplications (Sections 7 and 8). The concluding sectiondiscusses achievements and limits of the approach, andhighlights aspects that deserve additional investigations.

2. Ways and means of promoting awarenessSeveral empirical studies, and our every day experience,show how cooperative actors promote and use awarenessinformation in various ways and through different means tocoordinate their actions. These ways and means are stronglyinfluenced by the physical and logical layout of thecooperative work setting. Examples of factors influencingawareness promotion are the availability of physical orlogical shared spaces, the synchronous or asynchronousnature of the cooperative interactions, the degree of

visibility of other actors and of the common field of work.Moreover, the various ways and means are used seamlesslyso that their study is made difficult without a sort ofanalytical distinction providing the observer with a coarseclassification of ways and means. Coarse means here thatthe borderline between the classes is fuzzy on the onehand, and on the other hand that the related elements aresomehow fictitiously distinguished. In order to discuss therole of awareness in cooperation and the issues concerningthe design of a supporting technology, we found it fruitfulto adopt a distinction based on two classes that we denoteas by-product 1 and add-on awareness. In the following, weillustrate these two classes in turn.

1 The term 'by-product' is used in (Schmidt, 1998) with a similarmeaning.

2.1 By-product awareness

By-product awareness refers to stimuli conveying informationabout the state of the work setting and are generated in thecourse of the activities people must do in order toaccomplish their cooperative tasks. These obligations donot refer to rigid prescriptions but to the normative contextgoverning the activities to be performed in the consideredwork setting. Roughly speaking, the latter is characterizedby (possibly empty) classes of entities: physical objects,physical and logical artefacts, actors dynamically entering,moving and leaving the cooperative setting, and by theactions these actors can perform to change the state of thecommon field of work in order to make their cooperativeactivity possible. Each entity is characterized by itsfeatures and location in the 'space of cooperation'identified by the common field of work. These features arein principle different for each specific class of entity:for example, an actor has a posture, a look, a mood andpossibly is performing some recognizable action; a document,or a display, shows a structure, its content, its support;physical entities show a state. Locations define themutual position and distance between entities: position anddistance can express both logical and physical relationshipsamong entities. If we consider a snapshot of the abovepartial description of a cooperative work setting, we canidentify a first class of awareness information generated bythe entities through their simple being and acting in aspace, and possibly perceived by the other entities bysimply 'looking' at them. In other words, this informationis "what is there for picking" (Schmidt, 1998). Examplesare: the structure and contents of a form (as in the case ofthe Bug Form Report (Carstensen and Sørensen, 1996); theordering and marking of strips (as in the case of AirTraffic Control (Hugues et al., 1992); or finally, themutual orientation of the people (as in the Reuter's case(Heath et al., 1995)). Notice that the action of 'looking'can be mediated by some technology (e.g., a video link as inthe case of the steel plant reported in (Schmidt, 1998)). In addition, if the setting is observed in its dynamic

evolution, that is movements and sequences of actions(altogether, events) are considered, then additional

awareness information can be identified. Using a widelyadopted discretization of time, awareness information mayconcern the past: the history of events, the sequence ofsubsequent states and locations; or the present: the currentstate, location and active actions; or finally the future:possible or expected states, positions and actions to beactivated. This idea is common to the path model discussedin (Chalmers, 2000). Here awareness does not refer toinformation stored in any sort of repository rather to theoutcome of dynamically reconstructing information pertinentto the current or future action. For example, if acolleague did switch on her computer and then left heroffice, it is likely that she will be back quite soon. In sum, the awareness information considered so far refersto a context in which physical and logical as well as timeand space aspects are to be considered. Moreover, being aby-product of the activity people must do in order toaccomplish their cooperative tasks, the awarenessinformation considered so far is such that the

related cost - in terms of energy and intention- ispractically irrelevant in relation to the cost of the on-going activity. In fact, awareness information is producedand consumed just by living and acting in the space wherecooperation takes place, by doing the things actors must doand by using the 'tools' supporting them. Actually, thereal additional cost is about the acquisition, definitionand maintenance of some (silently) established mutualconventions about performing the actions and using the'tools' and the space themselves. These conventions, rootedin the given normative context, are necessary to produce andinterpret correctly awareness information (again, this pointis raised also in (Mark, 2000)). A typical example is theimplicit semantics associated to the layout of entities in ashared and possibly loosely structured workspace. Anotheradditional cost, which is surely more demanding, is theeffort to recover the sequence of past events, if this typeof awareness information is to be considered to act in thepresent and in the future.

2.2 Add-on awareness

Other studies, and again our everyday experience, show thatthis relatively 'cheap' form of awareness promotion andconsumption is just one of two possibilities. In specialwork settings and in special contingencies, people investpart of their energy, in a more or less explicit way, inpromoting awareness on top of the information they get 'forfree' in the sense illustrated above. This type of effortcan lead to (combination of) actions like: addingannotations to a form; amplify gestures; using voice in aunexpected ton and inflexion; moving to unplanned placesto highlight some specific fact (an interest, an intention,an indifference, a danger and so on); and finally, notifyingfacts about unforeseen or not perceivable situations to besure that the other actors know about them, wherever theyare or whatever they are doing. For example, in (Gutwin etal., 1996) "verbally shadowing their own behavior" andmarking objects to notify future use of them are taken as"examples of information that cannot be passively gatheredby the groupware system and suggests that some awarenessinformation can only be generated explicitly by theparticipants". In the same direction go the terms

"designing" and "configuring" used in (Heath et al., 2000)to describe how people "reveal particular events orinformation, without demanding that anyone should respond oreven listen", and the final claim in (Chalmers, 2000) that "designers give up progressively more control of how userclassify, interpret and work with their own information".In this phenomenology, two aspects are fundamental anddistinguish it from the previously illustrated case. Theabove actions constitute an additional activity, that is, aneat cost for the cooperating actors in relation to theirjob: a cost for both planning and performing it. Let'sillustrate the point by making a simple analogy with atrivial programming situation. Keeping a sequence ordered isnot mandatory to search an element but it helps a lot inreducing its cost. However, the overall cost is not reducedin general since insertion of an element in an ordered listis more expensive than in the unsorted case. The convenienceof this additional effort depends on the context of use:e.g., the ratio of searches and insertions in the givenapplication. This consideration leads to the second aspect:

the additional effort is not mandatory but discretional, thatis, depending on the actor's evaluation of the contingentsituation. By consequence, the other cooperating actorscannot be sure, in the general case, that the information isproduced. On the contrary, a possibly reciprocal awarenessis the product of a social behavior generatinginterpersonal or intra/inter-group conventions. The latterare in turn the source of an additional cost. This fact isfor example documented in (Mark et al., 1997) where theeffort and problems of defining naming conventions to denotedocuments is widely illustrated. Another aspect of actors'discretion concerns the 'sign' of the conveyed information.It can be positive, showing willingness to make cooperationsmooth, or negative, showing resistance, irritation,dislike. A similar reasoning can be made on the side of theconsumers of the promoted awareness information. They haveto be attentive to stimuli that they may not believe theyare useful for them, plan the access to them as well astheir interpretation and the interpretation of any deviationfrom the established conventions. And again, they may(openly) decide to disregard this information. All this isnot part of people's job, in a strict sense. It depends onpeople's willingness, on their more or less cooperativenature, on the level of criticality of the undergoingcooperative activity, on various forms of operational orsocial pressure.

We conclude this section by recalling that the above waysand means can be difficult to separate and distinguish asthey are seamlessly used and are all fundamental to improvethe articulation of activities among cooperating actors. Incooperation everything is dynamic (people, actions, spaces,artifacts, motivations) depending on varying physical,social and critical conditions. Hence, the course of actionrequires actors to determine a dynamic mix of ways and meansof promoting and consuming awareness information on thebasis of the dynamic evaluation of their performances inenhancing the effectiveness of cooperation in the givensituation2. However, the approach we are going to proposeis characterized by its capability to manage add-on2 These aspects can be summarized by saying that cooperating actors

have to operate in a complex dynamical system. This connotation influencedthe choice of the proposed metaphor .

awareness as the design of a technological support is morecomprehensive in this case. In fact, the resultingtechnology enhances by-product promotion of awareness byproviding virtual co-location in space and (present andpast) time as required by the distributed nature ofcooperation. Moreover, it also tries to reduce theadditional costs of add-on promotion and, in so doing,improves awareness promotion proactiveness of the actors aswell as of the artefacts they use. This twofold goalrequires the integration of technologies supportingdifferent modes of articulation work and, before this, adeeper understanding of the role of awareness incoordination

3. The role of awareness in coordination In the introduction we mentioned that the discovery of theuse of awareness as a means of articulation work played arelevant role in the debate on the nature of cooperation andon the capabilities of the technology supporting it. Morespecifically, we refer here to the controversial issues ofstructured versus unstructured flow of work, of proceduresversus exceptions, of regularity and ad-hocness incooperation. The elaboration on the notion of awarenessprovided the appropriate lexicon to speak about the secondcomponent of the above contrasting pairs. The emergence ofthe awareness of the awareness phenomenology together with thereflections derived from Suchman's seminal intuition ofaction situatedness (Suchman, 1987) led to reinforce theinterpretation of cooperation along the new perspective.And at the same time, it led to reinforce the gap betweenwho claims that regularity 'overtakes' ad-hocness or viceversa. For example, awareness is considered when sharedworkspaces are the focus of attention (e.g., (Syri, 1997) )but in this case little support is given to regularity; onthe other hand, the research oriented to the new generationof workflow systems interpret ad-hocness just in terms ofincremental design capabilities from a pure linguistic,modeling, technical point of view (Klein et al., 2000). Inthis case, awareness is out of the scope of the underlyingconceptual development and technological support. Similarly,empirical studies almost focus on either one of them (inaddition to the already mentioned ones, e.g., (Bowers etal., 1995)).

Even in approaches that consider the two perspectives asequally important, both at the empirical and design levels,they remain separated, not really integrated in ahomogeneous (conceptual or technological) framework. Inother words, the theme of understanding the empiricaloutcomes and the technological implication of an integratedview of regularity and ad-hocness is rarely addressed.Recently, the notion of activity awareness has been proposed(Nomura et al., 1998) in relation to potentially structured

activities. However, it refers more to the interplay ofindividual and collaborative views of the work settings byconsidering the first one as primary and the second one as asort of abstraction of the individual threads of actions.Initial elaborations of the integration we are claiming forcan be found in the empirical analysis of the use of akanban system (Schmidt, 1997) and in the on-goingdevelopment of an architecture for CSCW applications(Agostini et al., 1997).In order to contribute to the effort toward integration, inthe following we discuss a way to see the interplay ofregularity and ad-hocness, that is, to make more precise(or simply more explicit) the role of awareness incoordination.First of all, recurrence is captured by coordinationprotocols3 that are a fundamental portion of theorganization memory as they materialize procedures andconventions on which cooperating actors can rely upon toarticulate their activities

3 By coordination protocols we refer to "formal organizationalconstructs - procedures, workflows, process models, etc. " (Schmidt, 1997)regulating recurrent coordinative activities.

in the given normative context. For these reasons, suchprotocols represent a precomputation of coordination patterns(Schmidt, 1997). As such, they are not only supporting theexecution and accountability of recurrent actions but alsoproviding, as special kind of artefacts, awarenessinformation about the work practices characterizing acooperative work setting: this is particularly useful, e.g.,when newcomers join a cooperative ensemble. On the one hand,protocols can be used in different ways according to thecurrent situation: this point has been accurately elaboratedin (Schmidt, 1997) through the metaphor of 'maps' and'scripts' and we refer to this paper for any additionalargumentation. On the other hand, action situatednesssuggests that coordination protocols cannot be fullyspecified a priori as their actual execution requirescompleting their definition in the current situation. Thisis the only way to guarantee that coordination protocols areflexible enough to be used in an effective way avoiding, forexample, that actors spend effort to circumvent theirprescriptive nature or to manage exceptions that necessarilyoccur when they are supposed (and hence designed so as) toincorporate an actually open-ended space of possibility.This point is one of the main focuses of the current trendsin workflow development (Klein et al., 2000), also in thespecial domain of software processes (Cugola, 1998).Secondly, the under-specification of coordination

protocols implies that actors have to perform a continuousproblem solving activity in order to select out of the spaceof possibility the most appropriate strategy. Here is whereawareness promotion comes to the scene as that specific partof the articulation work that provides cooperating actorswith pieces of information about their current context.Therefore, they can perform the above mentioned problemsolving activity in a more 'aware' way. In this view,awareness receives a quite specific connotation: it is not ageneric source of information to (re-)compute patterns ofbehavior incorporated in established procedures andconventions. Rather, it is a complementary source ofinformation that is not part of them. For example, aprotocol and the related distribution of control andexecution among the cooperating actors tell them what theyare expected to do next. Awareness about the current statusof the protocol execution allows them to prepare themselves

to act in a more effective way and to establish adequatepriorities of which action to do next among the set ofpredefined possibilities. Moreover, they can decide if andwhen information about their current action is worthwhile tobe propagated to others so that the latter can be preparedin advance to deal with possibly unforeseen conditions(early involvement, lack of inputs, delays, and so on). Inthis view, awareness goes beyond the connotation of control(in the cybernetic sense of monitoring and feedback) andassumes the more active role of improving the creativity andeffectiveness of actors' behavior.In consequence of the previous considerations, the

technology supporting the execution of procedures andconventions capturing regularity as well as the promotion ofawareness of the current context where they are executedhave to be carefully designed in order to augment and makemore effective the smooth integration of regularity and ad-hocness of cooperation.

4. The implications on the supporting technology The type of integration we claimed before impliesrequirements of a technological framework where coordinationprotocols and awareness information (as basic means ofarticulation work) are equally supported in the light of theanalytical distinction proposed in section 3. Therequirements of the functionality supporting protocols havebeen discussed in previous work about the design ofcoordination mechanisms (Schmidt and Simone, 1996). They canbe summarized by two keywords: malleability and visibility. Thefirst one refers to the possibility for the cooperatingactors to define 'precomputed' protocols according to themodeling approach most suitable to their culture andattitudes and to the nature of the interaction protocol athand, in an incremental way. The second keyword refers tothe possibility for the cooperating actors to have a fullaccess to the linguistic features constituting the abovemodel. Visibility is based on the identification of a setof linguistic features that can be flexibly combined toachieve the desired expressiveness and malleability.Moreover, these features have to be at the semantic level ofarticulation work, that is, with an associated semanticsthat makes sense to the actors in their activity of definingthe needed coordination mechanisms.The point we want to make is that malleability and

visibility are required also for the features actors need todesign tools supporting awareness promotion. This is truefor at least two reasons. First of all, they are requiredto deal with add-on awareness. As illustrated in section2, add-on awareness is a part of the voluntary effortdevoted to articulation work and has to be supported in thesame way as the definition of protocols, its complementaryactivity in articulation work. Users have to be able todefine the related functionality in a flexible andincremental way. Secondly, malleability and visibility arerequired by the by-product awareness too. In fact, althoughthis type of awareness can be primarily in charge of thesystem underlying the current application, users have to be

able to see, if and when necessary, what information thesystem provides them with and how, and to be able to filteror modify this information according to their dynamicneeds. Moreover, since add-on awareness is constructed ontop of by-product awareness their supportive technology hasto share the same requirements. Finally, malleability andvisibility are fundamental to define in a more flexible waythe allocation of awareness functionality between the system(by-product awareness) and users (add-on awareness): thisflexibility is required again by the dynamic nature ofcooperation.

4.1 Metaphors to build models of awareness

As for the case of precomputed protocols, the requirementsof malleability and visibility suggest that models areneeded to represent the awareness phenomenology. Moreover,these models have to identify a finite set of linguisticfeatures actors can access to design the pertinent tools. Ofcourse, this idea is not new. The work on the Spatial Modelof Awareness (originated by (Benford and Fahlén, 1993) anddeveloped in (Greenhalg and Benford, 1995; Benford andGreenhalgh, 1997; Sandor et al., 1997)) is a milestone inthe development of CSCW infrastructures supportingcooperation. This seminal work proved the possibility tomake awareness a 'computational concept'. In fact, it led toseveral implementations, some experimentation, and to afirst effort to formalize it (Rodden, 1996): the latter canbe seen as an identification of an initial set of linguisticfeatures to promote awareness. Unlikely in the case of precomputed protocols, awareness

models can be hardly derived from the analysis of real casesin a direct way. In fact, when actors articulate their workby using precomputed protocols, they 'speak' and 'create'the pertinent categories. Hence these categories naturallybecome the basic constituent of the articulation workmodel.4 This is not the general case when awareness isconcerned: ways and means have to be described throughdifferent 'tools' since they are verbal and non-verbal, andthe contents through which awareness is promoted are notnecessarily verbal too. The transition from thedescriptions of the awareness phenomenology provided by theempirical studies to a model can be supported by theselection of a suitable conceptual metaphor. As widelydiscussed in (Lakoff and Johnson, 1999), conceptualmetaphors are cognitive mechanisms that allow conventionalmental imaginery from sensorimotor domains to be used for(observing) new/different domains of subjective experiences.Moreover, although not all conceptual metaphors aremanifested in the words of a language (Black, 1962) but alsoin gestures, symbolic structures or rituals, neverthelessnon linguistic metaphors can be expressed through a languageor other symbolic systems. The strength of metaphors as areasoning tool is that they allow mappings of inferencesabout a conceptual domain onto inferences about another4 This was the way in which the Model of Articulation Work proposed in

(Schmidt and Simone, 1996) was conceived of.

conceptual domain, possibly uncovering inference patterns inthe target domain (Nehaniv, 1998). More specifically,metaphors allow one to speak about awareness at a semanticlevel that is independent of the technological tools used toimplement it.For all the above-mentioned reasons, metaphors are a

powerful tool when awareness is concerned. How good theselected metaphor is depends on the extent to what therelated model captures the awareness phenomenology andidentifies linguistic constructs an actor can use toachieve awareness promotion.

5. Why a new metaphor

Spatial Models are based on a metaphor comprising thenotions of space, medium and adapter, and the notions ofaura, focus and nimbus associated to the objects populatingthe target application. The mutual awareness of any twoentities moving in a space is the outcome of a negotiationprocess between pairs of entities playing the role ofobserver and observed object, respectively. The negotiationis

mainly based on two types of information: the focus, which"describes the observer's allocation of attention" and thenimbus , which "describes the observed object'smanifestation or observability" (Greenhalg and Benford,1995). Nimbus and focus allow for the definition of a'measure of awareness', which is a computable value of themutual interest of entities for sake of awareness. We fully appreciate the idea that the metaphor takes the

points of view of who generates and perceives awarenessinformation, and combines these points of view to take intoaccount the awareness phenomenology. A reflection on theoriginal metaphor and the way in which this combination isachieved triggered a new point of view that could improveboth of them. A disclaimer is in order before going deeperin this topic. When metaphors are used to describe patternsof behavior, and not to infer formal properties, they arestill useful and suggestive but their usage can become lessrigorous: in fact, one can use them in different ways andpush them behind their original meaning. Hence, discussingabout metaphors does not involve expressiveness in a strictsense (a metaphor can be often 'emulated' by another one),rather it is about the (slightly) different interpretationof the phenomenology it wants to account for. Said that, wepresent the motivations leading us to consider analternative metaphor to formulate the awareness model weare proposing. In the above spatial metaphor, the notion of focus seems

to partially account for the intuition behind the notion of'peripheral awareness', or at least a possibleinterpretation of it. Consider the following scenario. Anactor is concentrated on a specific activity to perform herduty. Suddenly, someone enters her office and moves aroundlooking for something. The noise, the smell, possibly someuttered word make her aware of who the visitor is andpossibly his intentions. All this has little to do with theobject of her current focus, which is oriented on the tasksshe has to accomplish by using a computer laid on her desklocated away from the entrance door. She can just ignorethe presence or react in different ways (just tellingsomething while continuing her activity, or stopping it andorienting her focus toward the visitor). The case showsthat peripheral awareness does not involve only entities

that are somehow related to the current focus, but also tofully unrelated ones. In other terms, it is not always thecase that focus fully dictates the peripheral attention. Theattitude of the listener is what defines her awareness ofthe current situation. In the same scenario, she could beso concentrated on her work that she not just ignores buteven does not hear or perceive the 'nimbus' of the visitor.In another situation, even if her perception can be just'peripheral', it could be sufficient to stimulate a shiftof attention from the current focus to the new visitor(e.g., because he is a person relevant to her, for somereasons). This brings to the second, more substantial point.Although the notions of focus and nimbus allow for a

flexible computation of the 'measure of awareness', they donot account for the effects of mutual awareness betweenentities. Some of the illustrated applications show howthis measure can be used (for example, to tune the bandwidthof a communication support). First of all, this aspect isnot part of the model as it does not provide any explicitmeans to express it. Secondly, the 'measure of awareness'seems more oriented to provide

information to an external observer of the currentsituation than on the cooperating entities. In our view,they can be interested more in how they can use awarenessinformation than in an absolute value expressing how muchthey are aware of some other entity. Although the measurecan be a starting point for them to select a future behavioror a valuable information for an observer (typically, theunderlying system) to provide ad-hoc functionality, themeasure is too elemental to support actors in designingtheir own awareness mechanisms. Again, the point is aboutwhat the metaphor wants to express as basic phenomenology.In this respect, the spatial metaphor is perhaps notpowerful enough and a richer one is required. This need canbe rephrased in terms of the distinction between primary andcomplex metaphors proposed in (Lakoff and Johnson, 1999).Primary metaphors are those regarding a large part ofcognitive unconscious derived and learned from sensorimotorexperiences: spatial metaphors are a typical example.Complex metaphors are composition of primary metaphors andderive their greater expressive power from the combinationof the descriptions obtained through the primary metaphorsconstituting them 5. The metaphor we propose in the nextsection to formulate a new Model of Awareness, belongs tothe class of complex metaphors. It incorporates aspects ofthe spatial metaphor as basic components in combination withother features to obtain the expressive power necessary torepresent add-on awareness.

A final motivation for a new metaphor is based on anadditional requirement raised by a full integration of toolssupporting precomputed protocols and awareness promotion.Namely, that forms of awareness are called to operate acrossdifferent and interoperating applications, as made possibleby the general purpose CSCW infrastructures envisaged bycurrent research efforts. This challenge puts an additionalrequirement to the desired model of awareness: that is, howto handle this integration in terms of (logical) spaces andinteractions among them. This aspect is not considered in

5 For this reason, complex metaphors form a large part of theconceptual system human beings use to structure their everyday language andtheir reasoning activity (up to the extreme of the growing of scientifictheories (Boyd and Kuhn, 1979)) as they allow one to map rich linguisticand inferential capability across conceptual domains.

the metaphor underlying the Spatial Model of Awareness:however, the proposed measure of awareness is hardlyadaptable to this new requirement as it would imply aconcomitant accessibility to the spaces of differentapplications in order to define and compute it. Thiscontrasts the principle of encapsulation of information thatis basic to guarantee distributedness and localmalleability of these applications.

6. The proposed model of awareness

The requirements of an awareness model discussed in theprevious sections led to consider the Reaction-Diffusionmetaphor as a candidate complex metaphor supporting thedesired model of awareness. This section presents themetaphor, the derived model and a comparison of the latterwith the Spatial Model of Awareness.

6.1 The metaphor underlying the modelMany fundamental structures and dynamical behaviors inphysics, chemistry and biology are described in terms ofreaction-diffusion metaphors, which are rooted in the workof Alan Turing (Turing, 1952). Applications of this metaphorrange from pattern formation or epidemic spreads to naturalselection through ecological systems and the functioning ofthe immune system. Reaction refers to phenomena where twoor more entities become in contact and modify their state inconsequence of this fact. Diffusion implies the existence ofa space where the involved entities are situated and canmove. While reaction is univocally interpreted, diffusiontakes different meanings in different disciplines.Specifically, in the disciplines considering bioticentities, diffusion is interpreted as movement of entitiesin a given space. For example, in animal population dynamics(Okubo, 1980), entities move to concentrate together or toscatter according to behavioral patterns and mutualinteractions (e.g., some entities form a group according toan opportunistic behavior so as to optimize the use ofresources). This interpretation captures the essence of thenotion of diffusion we are interested in dealing withawareness. As such, reaction and diffusion describe thebehavior of (biotic) entities in terms of changes of stateand position, respectively; however, they are not enough todescribe that some factor, possibly independent of theinvolved entities, can influence the way in which theyperform both reaction and diffusion behaviors. Thisinfluence is captured by the notion of field as used inphysics. There are fields whose sources are outside theconsidered space. For example, the heat, light, gravitygenerated by the sun; and fields that are originated by someentities located in the space: for example, the emission ofad hoc substances in the case of animal species (likepheromone in ants) or the heat and light generated by a

lamp. Specific distribution laws that determine field valuesat each point in the space characterize fields of bothkinds. Moreover, entities are characterized by theircapability of being sensitive to those fields at differentdegrees in relation to their current state.The reaction-diffusion metaphor is based on principles

that seem to fit the awareness phenomenology. First of all,the control is fully distributed. The entity behavior isdetermined by a local 'computation' based on its positionand sensitivity to fields as well as on reaction anddiffusion patterns characterizing its type. In this way,both emission and consumption of awareness information canbe represented, respectively. Hence, the metaphor containsthe features needed to represent both by-product and add-onawareness. Moreover, the notion of field sensitivity fitswell the idea of peripheral awareness as it can be definedin terms of the dynamic attitude of an entity rather than onits current action. Finally, field propagation fits wellthe requirement of interoperability across applications inorder to support awareness promotion across them. In fact,this feature allows each application to encapsulate its owninformation and to minimize the amount and type ofinformation crossing interoperating applications.

6.2 Details of the model The features of the reaction-diffusion metaphor describedabove are incorporated in a Model of Modulated Awareness(MoMA) (Simone and Bandini, 1997). MoMA is a constellationof interoperable layers: hence, it is a non-hierarchical,multi-layered model. Each layer encapsulates its ownstructural and behavioral information and exports/importsinformation to/from other layers in a standardized way.Before going into the details of each layer, we give an

intuitive idea of its basic features (see Figure 1). Eachlayer contains a space populated by a set of heterogeneousentities, which behave by using diffusion and reactioncapabilities.

Figure 1. (a) The space is populated by three heterogeneous entities(‘white’ and ‘black’ types are shown). Each entity is characterized by adifferent state (A, B and C). The entity in the state A can send fields1 and 2, and can receive fields 3 and 4. The entity in the state B canemit field 1, and can perceive fields 2 and 3. The entity in the state Ccan receive fields 1, 3 and 4. (b) After receiving field 2 emitted bythe entity in state A, the entity in state B changes its location,state (D) and it is no longer sensitive to field 2.

As for diffusion, each entity can move in the space and isendowed with a receiving set and a sender, both able tosimultaneously receive/send signals of different nature anddifferent intensity according to the type and currentsituation (state and location) of the entity carrying them.

Signals may cause the receiving entity to change its stateor location. As for reaction, groups of entitiessufficiently close and holding specific states, cansynchronously modify their own state. The mix of diffusionand reaction behaviors is due to the changeable states andlocations of entities, which allow reaction to influence theconditions of diffusion, and vice-versa.

More precisely, each layer contains a space , which isdefined as a set of sites together with an adjacency relation :the latter is represented by a dynamic, connected graph,which is used to evaluate when entities come in contact andto express how fields propagate in the space. The space canbe partitioned in sub-spaces .The space is populated by entities; a region is a sub-spacewith an assignment of entities to its sites. Entities aredefined in terms of their type that specifies the values ofsome parameters governing their behavior in given states.These parameters are as follows. The field sensitivity functiontells in which states the related entity is sensible towhich fields in accordance with a vector of thresholds, onevalue for each field. The source function tells in which statesthe related entity is the source of which field togetherwith its initial value. And finally, the composition functioncombines the different values of the same field at the samesite (this is required when the topology is not regular orthere is more than one source of the same field). A field is characterized by its set of values and distributionfunction and by functions to compare field values,sensitivity values and thresholds. Notice that each entitycan be source of, and sensitive to, different fields at thesame time. The perceived fields can be generated by entitiesbelonging to any layer. If the emitter and the receiverbelong to different layers, then the received value is, bydefinition, the initial value.Type, current state and position characterize eachindividual entity. A general law states that each sitecontains at most one entity. A configuration of a layer is described in terms of its regionsand of the values of the fields at its sites. A configurationof a MoMA is the constellation of the configurations of thelayers it contains.The dynamic part of the model is described in terms of ruleschemes , which define the transitions between configurations(as in standard transition systems). There are four mainrule schemes based on both state and position of theinvolved entities: • Field diffusion rules define when a field is generated:IF an entity of a specific type, in a specific state, is a

source of a field THEN the field is propagated according to its distributionfunction and initial value . • Trigger rules define how a field affects the state of anentity:IF an entity is sensitive to a field in relation to itsstate , threshold and sensitivity function THEN the entity changes its state according to a predefinedstate transition function.• Transport rules define how a field effects the position of anentity:IF an entity is sensitive to a specific field (as fortrigger rules)THEN it changes its location according to a predefined sitetransition function.

• Reaction rules define the synchronous state change of entitiesconstituting a vicinity, that is, they are all pairwiseadjacent:IF entities constituting a vicinity possess specific types andspecific states THEN they change their state according to a predefined statetransition function.

In addition, there are rule schemes managing the dynamicnature of the space and the number of instantiated entities. To this aimthe type of an entity specifies in which states it canmodify the graph underlying the space, with the only constraintthat occupied sites cannot be deleted; and in which statesan entity can generate and locate in the space a new entity of the sametype. Moreover, an entity reaching the special dead stateis deleted by a clear rule .

Since each entity can be defined so as to be sensitive to asub-set of fields generated outside the layer it belongs to,the behavior the entities belonging to one layer can beinfluenced by the other layers and react through appropriatetrigger or transport rules so as to change their state orposition in consequence of this sensitivity. The rationaleof this choice is mainly based on the requirement of keepingthe different layers as much independent as possible, inaccordance with the requirement of local malleability ofeach layer.

We conclude this section by emphasizing that MoMA (whosecomplete formalization is given in (Bandini and Simone,1999)) contains a small and orthogonal set of parametrizedrule schemes. Instantiated rules can be composed intoscripts to allow the user of the model to constructcompound behaviors combining by-product and add-on awarenesscapabilities. In this respect, MoMA can be seen as a 'shell'allowing one to define patterns of generation andconsumption of (awareness) information in response to themultifarious requirements of the target reality. Section 7will illustrate this aspect in detail.

6.3 Comparing models A this point it is in order to discuss both similarities

and differences between the Spatial Model and MoMA, in thelight of the disclaimer mentioned in section 5 (Table 1).The first difference concerns the degree of complexity of

the two models. The first one is quite parsimonious: itcontains few concepts to define an awareness measure. Webelieve that part of its simplicity comes from the fact manyadditional features, just apparently out of a workablemodel, have to be added when the model is used to constructawareness mechanisms in a real implementation. In general,it is not immediate to consider this model without makingreference to specific implementations or contexts of use. Onthe contrary, MoMA aims at being self-contained to support ageneral purpose 'shell' whose features are visible to theuser and therefore is more articulated. Specifically, ourgoal is to explicitly define,

within the model, all the features needed to handle bothby-product and add-on awareness. In any case, the realquestion is whether these features are all useful for aflexible management of awareness. We are quite convincedthat this is the case: in fact, in our experience, theelimination of whichever feature reduces the expressivepower of the model. A greater complexity could be questionedagainst usability. Users are not constraint by the use ofall the features to construct useful mechanisms. We willshow how different classes of users, with different skillsand responsibilities, have available different sets offunctionalities: the latter range from the simple tuning ofsome parameter, to the customization of predefinedmechanisms, or finally to their construction fromscratch.

Spatial Model MoMAproduction nimbus fieldperception focus sensitivity functionnegotiation various forms of

intersections of focusand nimbus that are defined accordingly

comparison of field values against thresholds and sensitivity functions

medium, persistency

in some implementations

in the model, as properties of subspaces and fields

collective entities

Third Party Objects NONE

adaptability surface - tuning deep - full visibility

distributed control

in some implementations

inherent to the model

induced action NONE add-on awareness through scripts of behavioral rules

Table 1 The main distinguishing features of the two space-based models

The second evident difference is that MoMA can be structuredin non-hierarchical layers, each one containing a space withits fields, typed entities and related behavior. A similaridea can be introduced in the Spatial Model too, with theproblems of adapting the definition of measure of awareness

we discussed in section 5. The idea of structuring thelogical space to better frame in a context awarenessinformation has been proposed also in (Mansfield et al.,1997) through the notion of locale (supporting reciprocalawareness) and in (Prinz, 1999), a framework promotingawareness on the base of an event handling model.

If we concentrate on a single layer, the main differencesare generated by three primary sources: the sense that istaken as main reference in the two approaches, the choice ofthe underlying metaphor and the capability to deal withcollective entities. As for the first aspect, in theSpatial Model the notion of focus naturally refers to'sight' whereas in MoMA the notion of field naturally refersto 'hearing'. As discussed in Section 5, this leads to adifferent interpretation of peripheral awareness: in thefirst case, the interpretation is similar to the one widelydiscussed in (Yamaashi et al., 1996); in the second case,peripheral awareness is a full-circle capability. As for themetaphor, MoMA incorporates the main features of the spatialone and combines them with features coming from otherprimary metaphors that are related to the evolution of theinvolved entities: primarily, the notions of state, full-circle sensitivity, and reacting behaviors. In addition, themeasure of awareness proposed in the Spatial Model (Rodden,1996) does not imply, at the model level, the locality thathas been taken as a primary concern in MoMA in order toobtain the maximum degree of distributedness not only acrosslayers (applications) but also within each single layer.Finally, the extension of the Spatial Model with thenotion of Third Party Object allows the management ofawareness behavior of collective entities. MoMA does notcontain any feature able to express such behavior. Somelimited capabilities can be emulated when the behavior ofcollective entities do not imply their movement (e.g.,bounded rooms and buildings, floor control objects, commonfoci (Benford et al., 1997)).

Since they share the spatial metaphor, the two models showsome similarities too. For example, the two main notions offocus and nimbus characterizing the Spatial Model have theircounterpart in MoMA in the notions for modeling thereception and emission of fields by entities, respectively.The notions of field distribution and field sensitivityallow for a quite flexible and compact way of representingvarious types of nimbi and foci. Defining the level ofpersistency of a field allows one to express some of thetime features characterizing Aether (Sandor et al., 1997).In addition, Aether shares with MoMA the idea to spread

awareness through an ether defined by a semantic network ofobjects. According to the reaction-diffusion metaphor,'allocation of attention' is described as 'increase ofsensitivity' to incoming fields, while 'manifestation orobservability' is described as 'increase of strength' of theemitted fields. Since both emitters and receivers canconsider different fields simultaneously, then the'negotiation' can be performed taking into account differentkinds of awareness information. In MoMA nimbus and focusare of a different nature: the former is a field while thelatter is a sensitivity to fields. In the Spatial Modelfocus and nimbus are of the same nature (at least in theexamples and formalization): in fact, here measures ofawareness are computed as various forms of 'overlapping' or'intersection' of the set of objects constituting them. InMoMA it is quite natural to express sensitivity to(specific types of) information, independently of theidentity and position of their possible sources. The SpatialModel naturally expresses focus orientation toward entities.Both approaches are able to represent the aspects moredirectly represented in the other model by using acombination of features. For example, in MoMA focusorientation can be obtained by modulating sensitivityaccording to some privileged directions in the topologicalneighborhood of the considered entity.

7. The AW-Manager: from the model to the functionalityThe features of MoMA identify primitives6 that have beenimplemented in a software module, called AW-Manager, as partof a CSCW infrastructure. This module provides awarenessservices to generic cooperative applications whose on-goingbehavior (facts) is visible to its environment, as shown inFigure 2.

AW-Manager Component

Awareness framework Application framework

CooperativeApplicationandUser Interfaces

facts

aw-infoto actors/agents

* definition* activation

Figure 2 The AW-Manager interface with the target application

The primitives are devoted to: the construction of a dynamicspace, the creation of entities together with their types,the definition of 'fields' together with their parameters,and finally the definition of the behavior of the entitiesin terms of displacements in the space, individual andsynchronized state changes, and generation of 'fields'.First, the primitives can be used by the application engineer to definemechanisms promoting what we denoted as by-product awareness, thelatter being coupled with the application itself as anadditional functionality. Second, the primitives can be usedby the application users too in order to tune the above awarenessmechanisms by modifying the variable part of the model(functions, thresholds, and so on). Finally, users can accessthese primitives to define mechanisms promoting what we denoted as add-on

6 The syntax of these primitives can be naturally derived from theformalization of the rules they are associated to, and is out of the scopeof this paper. The relevant point here is that this syntax allows thefulfillment of the visibility requirement of the primitives to constructawareness mechanisms through a standard syntax-driven user interfacesupporting their usage.

awareness. In this way, we aim at what is called surface anddeep customization in (Bentley and Dourish, 1995). In thefollowing we illustrate how the above three modalities ofusage can be realized by considering first how factsoriginated from the application can be associated tosuitable primitives to build the associated awarenessfunctionality.

7.1 Defining by-product and add-on awareness

A first class of facts concerns the construction of theawareness space out of the (implicit) logical space of theapplication. It is likely that the two spaces share arelevant part of their topological structure: however, inthe general case, they are not identical. For example, theassignment of an actor to an activity is a logical link thatcan be implicit in the space identified by the causalrelation among activities. In the awareness space, this linkis likely to become explicit in terms of the dynamicadjacency relation between the two sites where the twoentities (actor and activity) are located in order toexpress their logical proximity. Since the targetapplication can be 'closed' at various degrees, the mostgeneral case requires to catch the facts notifying theconstruction of the logical space of the application and toassociate to them the corresponding primitive to constructthe awareness space, in order to reproduce the samestructure. Obviously, this strategy can be non-optimal inthe case of a very large space, which requires reducingduplications. However, these cases can have ad-hocsolutions in the definition step of the awareness space,which in this case shows a quite different topology(typically, much more coarse-grained). The same holds if theapplication is 'open' enough to allow for the re-use of theinformation contained in its logical space (that is, withoutreproducing it), possibly decorating it with additionaltopological structure. In all cases, the primitive forspace construction is used when the initially defined spaceis modified at run-time by the application and the same facthas to be reproduced in the awareness space, accordingly.A similar reasoning holds for facts concerning the

creation of entities. The entities populating anapplication usually own a type as well a set of statesgoverning their behavior through a state transition diagram.This is typically the case when states are used to realizeapplication adaptability in order to define if and when themodification of an entity can occur in relation to itscurrent state. The states used in the application are the(initial) set of states of an entity, that is one of thecomponent of its type. Again, if additional states have tobe introduced to express awareness behavior, then thedesigner can access the primitives to define entity types.

The last class of facts, namely those related to theaction entities can perform in the application framework,can be grouped into two main classes: actions producing a(logical) displacement and actions producing a change ofstate of the involved entities. In this case, these factsare 'translated' into the primitives corresponding to thetransport and reaction rules of the model, respectively, toreproduce the same modifications. So far, awareness isconveyed just in terms of information related to positionsand states of entities and is conveyed just to entitiesliving/acting in the same (portion of) space, that is,having direct visibility of them. In the case of humanagents, this opens the problem of how to present awarenessinformation to them. We will come to the theme of awarenessvisualization later on.

The capability to propagate awareness information to 'lessclosed' entities is based on the notion of field. Thefeatures related to field handling are typically the onesnecessary to construct, in combination with the other ones,the mechanisms promoting awareness in a more powerful way.They do not have any counterpart in the application: theyhave to be 'programmed' by the awareness mechanism designer(a user or the application engineer) according to what thisperson thinks can be a valuable support for the applicationusers. These features concern the definition of fields,sensitivity functions to them, field source rules andfinally the field parameters in rules premises.Defining a field requires to specify two types of

information: its content (values) in terms of an optionalmessage and a mandatory value expressing the strength of themessage; and its gradient (distribution function) thatspecifies how the strength is modulated during fieldpropagation in the awareness space starting from an initialvalue at the source site. Moreover, a field plays a role inawareness promotion if there is a field distribution rulespecifying which entity, in which state becomes a source ofit with a specific strength. From the side of a receivingentity, its sensitivity function specifies the degree ofinterest in the field content according to the entitystate: that is, a dynamic subscription to and filtering ofthe conveyed information. Hence, sensitivity is acoefficient that can amplify weak messages or weaken strongmessages in relation to the current context expressed by thestate of the entity. The value expressing the strength is afundamental parameter to govern the degree of intrusion ofthe related message for the receiving entity: e.g., a changeof priority among messages in the case of artificialcomponents or a different presentation strategy to humanactors. The possibility that a user decides to define a new field

(through the modification of the entity playing as herassistant for sake of awareness promotion) opens the problemof how another entity can know about it. By default, anentity is sensitive to a new field having a non-null valueat the site it occupies, which is greater than a (default)threshold (the user can specify) for new fields. Later on,the user can tune her receptivity to these fields by

modifying her sensitivity and threshold according to herneeds. Moreover, she can use the new information theyconvey to generate ad-hoc awareness. This brings to the last feature of the model. The

possibility to specify field parameters in rules premises toconstruct add-on awareness promotion allows the users todefine awareness behavior of the kind: if I receivespecific pieces of awareness information, then I modify mybehavior so as to play a different role in awarenesspromotion. In other words, actors can plan, in advance oron the fly, the effect of receiving awareness information soas to modify their interest in awareness informationreaching them in the future or their willingness to act morepro-actively in relation to the other actors. Basically,this is achieved through the definition of the awarenessbehavior of the entities playing the role of assistant ofthe actors themselves, by using the primitives constructingthe pertinent rules. Notice that the resulting rules act onthe awareness space and entities populating it without anydirect connection with facts happening in the application,as it was the case in the previously discussed mechanisms.This is exactly what add-on awareness is about: a voluntaryeffort to shape individual awareness behavior, in relationto, but independently of, the behavior generated by thecooperative application.The use of the above range of possibilities is illustrated

in the next section through a simple example. To conclude,we want to emphasize that, since all types

of awareness mechanisms are programmed by using a fullyvisible set of primitives, the three modalities of usage (todesign awareness capability to be associated to theapplication either by the application designer, or tuned orfully defined by the user) can be dynamically meshed in asmooth way. On the one hand, once an awareness mechanism is(re-)defined, it becomes immediately active as a newspecification the run-time support has to consider. On theother hand, the uniformity of the language used to defineany kind of awareness mechanism allows both designers andusers to inspect its current structure to adapt it in afully 'aware' mode. In this way, the analytical distinctionwe made to structure the argumentation about awarenessmechanisms construction practically disappears in thesoftware module, in accordance with the claim that ways andmeans of awareness promotion are seamlessly defined andused.

7.2 A simple exampleAs an example to illustrate a possible instantiation of thevarious model components (without claiming the validity ofthe particular instantiation), we consider a simplifiedworkflow management system (WFMS). It describes workprocesses in terms of tasks connected by causal relations:Figure 3 sketches a portion of the related AND/OR graphwhere the responsible role is associated to the relatedtask. This example contains both explicit (causal relations)and implicit (responsibility) topological links.

.

T0

T1T2

T3

T4

R0

R1R2

R3

R4

T5and

or R5

......

Figure 3: The logical space of the workflow management system.

The same structure can be reproduced in the awareness space

by associating to the facts constructing the diagram therules to create the corresponding sites and adjacencies, aswell as entities of type task and role, and to locateentities in the appropriate sites. Roles occupydisconnected sites. Similarly, disconnected sites can becreated to locate entities of type actor, which will beallocated to tasks at the enactment time (they are not shownin the diagram). The space of awareness reflecting thelogic of the application can be enriched with additionallinks in order to support the propagation of awarenessinformation: this operation is performed through theprimitives available in the AW-Manager interface allowingone to add

new space information. A possible choice is shown in Figure4. Roles are linked to the tasks they are responsible of,and concurrent tasks are made adjacent since theirexecutions, although independent, must be all concludedbefore the execution of the following tasks.

T0

T1T2

T3

T4

R0

R1R2

R3

R4

and

or

T5R5

.....

Figure 4: The awareness space associated to the workflow

Typical facts that the WFMS can make visible are related tothe selection of a path in the WF diagram (OR node) and tothe task life-cycle (for example, its enactment,activation, conclusion); and facts concerning the state ofits execution (for example, delays or other events that canbe critical for the environment). Again through the AW-Manager interface, the designer initializes the set ofstates of the entities with their states in the WFMS. As forthe type task the set states becomes: future, enacted,concluded. With a similar reasoning she can assign to thetype role the set of states: assigned to, in-charge, done; and tothe type actor the set of states: available, doing, done.

As examples of how rules can be associated to facts comingfrom the application, let us consider (α) the execution ofan OR node, (β) the enactment of a task by its responsiblerole and (γ) the critical situation where a task is goingbeyond its deadline. These facts, and their possibleimplications, allows us to illustrate the usage of almostall features of the model.• Fact (α) can be codified as follows:

fact α: select the left branch of an OR nodeFact α has associated a rule disconnecting the right branchof the node so that it becomes unreachable from now on, forany awareness purpose.

• Fact (β) can be codified as follows:

fact β: a task Ti is enacted by Ri ; actors Ai,1, ...,Ai,k are allocated to Ti The designer decides that the enactment of a generic task Tiby Ri with the assignment of actors Ai,1, ..., Ai,kactivates the following awareness behavior. First of all,Ai,1, ..., Ai,k become adjacent to the task Ti ; then, Ti,Ri and Ai,1, ..., Ai,k change their state: if before theenactment they are in the state future, assigned to andavailable, respectively, after the enactment their statebecomes enacted, in-charge of and doing, respectively.Hence, fact β has associated a script composed of thefollowing rules: the transport rule: IF Ai,1, ..., Ai,k available THEN Ai,1, ..., Ai,k , Ti and

Ri form a vicinity (the consequence informally summarizes the changes of theawareness space topology ).and the reaction rule:

IF future Ti AND Ri assigned to Ti AND Ai,1, ..., Ai,kavailable

THEN enacted Ti AND Ri in-charge of Ti AND Ai,1, ..., Ai,kdoing Ti

Figure 5 shows the configuration of the awareness spaceafter the application of the above rules in a situationwhere T1 and T3 are currently enacted while T2 and T0 areconcluded and T4 is future. Obviously, a task can be concludedafter it has been enacted.

.

T0

T1T2

T3

T4

R0

R1R2

R3

R4

and

A01 A02

A21 A22

A31 A32

A31A32

Figure 5: The awareness space controlled by the AW-Manager: afterthe activation of facts (α) and (β) for T0, ..., T3. T1 and T3 areenacted; T2 and T0 are concluded whereas T4 is future. Circlesdenote the vicinities created by the transport rule associated tofacts (β) .

The fact concerning task conclusion, not described here,can be treated analogously as the fact of a task enactment.The only point relevant here is that the actors involved ina concluded task are in the state done as well as theresponsible role, and all keep their adjacency to it. Hence,they can be the targets of some future awarenesscommunication, if needed, as we shall see later on. Let us suppose that, afterwards, the designer decides that

the enactment of a task and the related assignment of actorshave to be notified to the role responsible of the HumanResources (called HRR). This role is supported by a specificapplication managing the organizational structure of thecompany where the WFMS operates: this application has itsown awareness space that, for sake of simplicity, is notdescribed here. We want just to focus on the aspectsconcerning the propagation of awareness information crossingdifferent applications. We suppose that the HRR role in thestate active is sensitive to the external field RA (ResourceAllocation). Hence, its sensitivity function sens-HRR isnon-null in the state active as far as RA is concerned: e.g.,sens-HRR(RA,active) = 2. RA is defined in the awareness spaceof the WFMS so as to have value: <message about resourceallocation, strength = 3 at the source>, and to beoriginated directly by the enacted task through thefollowing field source rule:

IF enacted Ti THEN propagate RAwhere the conclusion informally describes the activation ofthe field RA according to its distribution function and thesource function associated to the type task. In this casethe distribution function is irrelevant, as RA is anexported field: therefore, the only sensible value is itssource value.This is the very simple mechanism that makes awarenessinformation cross the various AW-Manager layers, stillkeeping each of them fully autonomous. In fact, how HRRreacts to RA is managed inside the space it belongs to in afully transparent way in relation to the awareness spaceassociated to the WFMS. In addition, if for any reason HRRbecomes insensitive to RA , this fact is not perceived fromoutside, unless explicitly notified.

• We consider now fact (γ), i.e., a task goes beyond its

deadline:fact γ: task Ti is going beyond its deadline: its

state becomes critical.A new state, let's call it in-trouble, has to be added to theset of states of the type task in order to use thisinformation in the awareness space. To this aim, let'ssuppose that the information about this critical situationshould reach the actors closely related to the problematictask and the HRR who can use it to select appropriatestrategies in future assignments of human resources. Hence, fact γ has associated the following field diffusionrule

IF in-trouble Ti THEH propagate TRwhere TR is a field to be defined. Through the AW-Managerinterface the designer defines the distribution function ofTR, denoted by DFTR, so that its gradient decreases inproportion to the distance. Let s be any site of the spaceand dist be the distance function between any two sites:

DFTR(s) = if s is a source node then 2 else 2 x 1

dist(source,s).

In the situation described in Figure 4, if the task T1 isin-trouble and hence becomes the source of a field TR, thevalues of TR are propagated as follows:DFTR(T1) = 2; DFTR(R1) = 2; DFTR(A11) = DFTR(A12) = 2for j = 0, 2, 3, 4: DFTR(Tj) = 2; DFTR(Rj) = 1; DFTR(Aj1) =DFTR(Aj2) = 1.

Now, each entity in the space can be differently interestedin the information conveyed by TR. This aspect is managed bythe threshold and sensitivity functions characterizing theirtypes. These functions take into consideration the state ofthe entity under concern. As described in the previoussection, this tuning activity can be performed, locally, byeach user being sensitive to this new field. As a possiblealternative, the WFMS designer can specify defaults valuesfor those functions in order to promote awareness accordingto the logic of application itself (not personalized add-onawareness) and leave to the users a possible finer tuning ofthe proposed awareness mechanism. Let's consider the secondcase and describe the actions the application designerperforms to construct the mechanism.For sake of simplicity, let's suppose that the designerassociates to all entity types the one constant as thresholdfor TR: i.e., any value of TR greater than or equal to onecould in principle be perceived in any state, while valuessmaller than one are neglected. Notice that thresholds allowone to circumscribe the actual propagation of fields in thespace, thus limiting their computational cost. The realstrength of the perception is however controlled by thesensitivity functions. Again, for sake of simplicity, thesensitivity functions of all entity types take three values:0 (insensitive); 1 (sensitive); 2 (highly sensitive). Then,the designer decides, according to the logic of the WFMS, todefine sensitivity functions to TR as follows:for the type task: sens-task(TR, enacted) = 2; sens-task(TR, future) = 1,

sens-task(TR, concluded) = 0;for the type role: sens-role(TR, in-charge) = 2;

sens-role(TR, x) = 1, for x ∈{assigned, done};

for the type actor: sens-actor(TR, doing) = 2; sens-actor(TR, done) = 1,

sens-actor(TR, allocated) = 0;for the type HRR: sens-hrr(TR, active) = 1 (in thepertinent space).The above definition of the sensitivity functions "modulate"the perception of the field TR by the entities of thevarious types: in fact, they cancel, maintain or amplify itsvalues by playing as a coefficient for the values of TR artthe sites where entities are located.

As a final example, let us suppose that the actors decidethat, when they are involved in a task that is in-trouble,they synchronously reduce their interest in awarenessinformation not concerning the critical task, so as toconcentrate on it and bring it again on schedule. This is atypical example of an explicitly stated convention. Asimilar convention has been reported by (De Michelis et al.,1997) in the case of a group of designers who assume aparticular position in the physical space when they have toaccomplish a crucial task. They sit in a protected area andturn their back to the other people so as to send thefollowing warning: "Please, do not disturb! We will not careof what you say". The above situations describe a typicaladd-on awareness that users want to construct by using theawareness generated by the target application. In the caseof the simple WFMS, the awareness mechanism defined above bythe designer makes them aware of TR whereas they can definethe following reaction rule in order to construct the add-on awareness mechanisms they agree upon:IF Ai,1, ..., Ai,k doing Ti AND Ai,1, ..., Ai,k sensitive

to TRTHEN alerted Ai,1, ..., Ai,k

together with the insertion of the new state alerted in theset of actor states and the modification of the sens-actorfunction so as that the sensitivity to all fields differentfrom TR is null. As expected, the promotion of add-onawareness is obtained with the effort to construct thepertinent mechanism. How much this effort is worthwhile forthe users depends on the reward they can receive from it: ahighly discretionary evaluation.In any case, the promotion of add-on awareness requires

specifying in the premises of the rules the sensitivity tospecific fields. This is never the case of by-productawareness promotion since, in this case, the users decidehow to react to the awareness information by doing someaction in the application framework and the awareness spaceconfiguration just reproduces the effects of this behavior.This difference is managed by the Execution Module of theAW-Manager, which is described in the next section.

8. The architecture of the AW-ManagerAs alluded to earlier (Figure 2), the AW-Manager has beenconceived of as part of a CSCW infrastructure to providegeneric cooperative applications with awareness services. Asimilar approach was taken in (Mariani, 1997), in thespecific case of database technology, more recently in(Prinz, 1999) within a broader scope and in Elvin(Fitzpatrick and Kaplan, 2000).

The awareness services supported by AW-Manager depend onthe level of visibility of the application behavior: from asimple log of facts up to a greater openness in terms ofaccessibility to its constituent components. In the formercase, it is likely that awareness information is managedjust by human actors cooperating through the application; inthe latter case, awareness information could be promotedalso taking into consideration the behavior of theapplication components. For example, in ABACO (Divitini et al.,1996), a multi-agent architecture realizing the Ariadnenotation (Divitini and Simone, 2000) to define coordinativeprotocols, each agent (internal agents as well as userinterface agents playing the role of assistant of the actorsinvolved in the protocol) contains communication primitivesdevoted to send and receive awareness information. Whenused in isolation, ABACO promotes awareness on an event-based approach (as in (Syri, 1997), (Fuchs et al., 1995)).If combined with AW-Manager, ABACOs agents could directawareness communication to it in order to transmit relevantfacts and receive the related awareness information. Thisarchitecture shows two advantages. First, since thecommunication primitives of ABACO's agents are accessible tothe users, the latter can specify facts that are relevantfor their working context. Secondly, ABACO can use aricher way to propagate awareness information due to thefeatures of the model incorporated in AW-Manager (typically,those exploiting space). Moreover, this happens withoutloosing the full tailorability of the awareness mechanismsby the users since AW-Manager guarantee a full visibility ofthe primitives derived from the model: this approach is inthe line of the incremental customization proposed in (Bentleyand Dourish, 1995). Moreover, the multilayered structure ofMoMA on which the AW-Manager is built allows the mechanismsfor awareness promotion to elaborate facts coming fromdifferent applications and to give back their results to theuser interfaces for presentation purposes.The AW-Manager contains a Definition and a Configuration

Module that are devoted to the definition , management andadaptation of the awareness mechanisms according to MoMA,and an Execution Module cooperating with them (Figure 6).Moreover, the Execution Module interacts with the targetapplication. It receives from it the notification of factsrelevant from the awareness point of view, activates the

awareness mechanisms accordingly, and sends back to theapplication the awareness information to be given to theappropriate actors/components.

RulesRepository

DEFINITION MOD.

* Interface to define MoMAcomponents

EXECUTION MOD.

* Engine to deal with rules and facts

CONFIGUR. MOD.

° space° entities° fields

add-on awarenessdefinition (RULES)

scripts associated to facts

configurationinformation(from designer)

configurationinformation(from application)

run-time modifications(by-product +add-on awareness)

facts

aw-info

AW-MANAGER

* Engine to manage

current configuration

exports fields values

imports fields values

Figure 6: The overall architecture of the AW-Manager

The Definition Module allows the designer/user toconstruct the various components of MoMA: basically, itmanages the user interface that makes available theprimitives illustrated in Section 7.1. The relatedinformation is passed to the Configuration Module when itconcerns the definition of space properties, entities andfields, and is stored in the Rules Repository when itconcerns the definition of (scripts of) rules to constructadd-on awareness mechanisms or the 'translation' of factsinto MoMA's primitives. Hence, the Configuration Module isresponsible for all information concerning the currentconfiguration of the awareness space populated by entitiesand internal/external fields values. In addition, itrealizes the transition to the next configuration: thistransition implements the modifications requested by theExecution Module on the basis of the interaction with thetarget cooperative application and the rules applicable tothe current configuration.Then, the core engine of the AW-Manager is the ExecutionModule. It is split into sub-components (Figure 7) that areconcurrently active to deal with the need to receive factsfrom the application, elaborate them and give it back theappropriate awareness information in a timely way. Let usfollow the cycle from the incoming facts to the communicatedawareness information by considering the behavior of eachcomponent.

Rules/FactsRep.

CONFIG.MODULE

Facts/RulesInterpreter

PriorityManager

Conflict Solver

Applicat.InterfaceManager

EXECUTION MODULEfacts

aw-info factsaw-info

exports fields values

Figure 7: The architecture of the Execution Module

The Application Interface Manager is listening and recordingfacts to be interpreted by the Facts/Rules Interpreter. Avirtual clock governs the behavior of the latter: at eachstep it verifies the existence of facts to be interpreted by

scripts or rules to be applied. The Interpreter performsthis verification by accessing the rule repository, thecurrent configuration and the imported fields values. Thene,it selects the appropriate rules on the basis of theirpriorities and conflict resolution strategies by interactingwith the pertinent modules (Priority Manager and ConflictSolver) and finally applies them in order to modify thecurrent configuration and obtain the next one. To this aim,the Interpreter triggers the interaction with the othercomponents in order to give the Application InterfaceManager the awareness information to be transmitted to theapplication, to request the appropriate modifications to theConfiguration Module (see Figure 5) and to export fieldsvalues, if necessary.

Priorities are specified when the AW-Manager isinitialized to define the policy by which the rules have tobe verified and used by the Priority Manager: for example,field diffusion rules have to be applied before any otherone. This possibility is useful to tune the AW-Manager tothe target application, which might raise differentrequirements according, e.g., to the distribution of thevarious types of incoming facts. Moreover, the PriorityManager contains strategies to 'traverse the space' in orderto guarantee a fair verification of rule premises on theentities. The verification of rule premises can generate conflict

sets arising from different situations. Typically, when asingle entity is involved in alternative rules in the samestep or entities are competing for the same site. Forexample, a conflict arises when in a given state, an entitycan participate in both a trigger and reaction rule.Conflicts are managed by the Conflict Solver, which has toselect one rule out of each conflict set. By default, thehighest priority is given to the execution of the scriptsassociated to the incoming facts since they represent whatdid really happen in the application. Once these areselected, the other resolution strategies are applied tosolve the remaining conflicts. Resolution strategies aredefined again during the initialization of the AW-Manager toreflect different types of behavior. For example, in thecase of the above-mentioned conflicts between trigger andreaction rule the selection in favor of trigger rulespromotes a more individualistic behavior, while theselection in favor of reaction rules promotes a morecooperative behavior. In addition, the resolutionstrategies can be specified taking into account the type ofthe involved entities by recognizing that some of them aremore influential than others. When conflicts are solved theFacts/Rules Interpreter elaborates the consequences of theselected rules and behaves as described above.In order to deal with the promotion of awareness across

different applications, it is possible to define in aseparated and autonomous way several instances of the AW-Manager, each one with its configuration, rules, facts,priorities and resolution policies. The only commoninformation concerns external fields: they are easily

identified at definition time since they are defined andgenerated in a layer and possibly perceived in differentones (that is, they participate in the definition of thesensitivity function of some entity type). As previouslymentioned, the Rule/Facts Interpreter of each layer accessesthe information about field values to be imported from theConfiguration Manager of its layer and tells field values tobe exported from the current layer to the ConfigurationManager(s) of the pertinent layer(s). The idea isillustrated in Figure 8.

The AW-Manager is implemented in the JAVA™ language toguarantee portability across different platforms and theconcurrent activation of its modules to support a smoothintegration of the definition and execution functionality.The basic principle governing the design of the AW-Manager'sarchitecture is modularity for two basic reasons: first ofall, it allows for the distribution of AW-Manager'sfunctionality on potentially networked machines (e.g., tosatisfy different computational demands); secondly, eachcomponent is tailored to a specialized functionalityrequiring ad-hoc algorithms. The architecture allows forthe improvement of single modules performance andfunctionality without effecting, in principle, the othermodules.

layer J

AW-Manager J

Application J

R/F Interpreter J

Config.Manager J

accesses imported field values from

layer IAW-Manager I

Application I

R/F Interpreter I

Config.Manager I

accesses imported field values from

exports field values to

Figure 8. The flux of external field values across layers

9. Summary and open issues The notion of awareness has been discussed from the point

of view of the design of a supportive technology focusing ontwo basic requirements: integration with tools supportingother forms of coordination and visibility by differenttypes of users. Obviously, visibility was the leadingrequirement since what is visible has to make sense to theusers. Integration has simply to deal with compositionalaspects that fully respects the choices made to fulfillvisibility. Both integration and visibility can beachieved through the definition of a language whoseprimitives have to fulfill, in turn, two basic requirements:they make sense to, and can be flexibly composed by, theusers to construct the awareness mechanisms they dynamicallyneed; and have a well defined semantics, so that users havea precise consciousness of their capabilities on the onehand and on the other hand other software components caninteroperate with the awareness support at the logical aswell as at the technical level.

In order to make sensible what users can access it wasnecessary to go through some intermediate steps. First, wehad to give awareness a more precise connotation. In tryingto get a deeper understanding of the ways and means peopleadopt to deal with awareness information, we suggestedconsidering two types of awareness: by-product awarenessthat is generated in the course of the activities peoplemust do in order to accomplish their cooperative tasks; andadd-on awareness that is the outcome of an additionalactivity being a neat cost for the cooperating actors inrelation to what they must do, and is discretional in thatit depends on actors' evaluation of the contingentsituation. Hence, we proposed to give awareness the activerole of improving the creativity and effectiveness ofactors' behavior in coordination beyond a pure role ofcontrol in terms of monitoring and feedback. Thisconnotation makes it possible to avoid the risk toimproperly overload the meaning of awareness and design aconfusing support. Secondly, and in consequence of that, weproposed a new metaphor, the reaction-diffusion metaphor, todescribe the awareness phenomenology, in order to take intoaccount the two above-mentioned types of awareness. Third,we derived from the metaphor a model of awareness (MoMA) inwhich awareness control is fully distributed within eachsingle space as well as across different awareness spaces.Finally, we derived from the model the primitives of alanguage, fully accessible by the users for sake ofconstructing by-product and add-on awareness mechanisms.In order to achieve integration, these primitives have

been incorporated in a software module that can be used incombination with coordinative applications for sake ofpromoting awareness information. The main architecture ofthe module was presented together with its interoperabilitywith the target application; moreover, a simple example wasused to illustrate how the incorporated primitives can beused to build awareness mechanisms. There are several aspects that need a deeper

understanding and improvement. Among the most obvious onesare the technical aspects of scalability and performance(Ramduny et al., 1998). Although we claimed for a modulethat can be integrated in generic cooperative applications,our implicit focus is on scenarios that can be described by

multiple spaces of limited size, as typically identified bycoordination mechanisms as defined in (Schmidt and Simone,1996). This view is fully coherent with the layeredstructure of our awareness model and reduces the problem ofscalability, iteratively, to small groups of entities. Inaddition, the highly modular structure of the model and ofthe related software architecture opens some possibilitiesto improve performance of each functionality. In any case,both aspects have to be further investigated once theapproach can be considered as convincing.

Therefore we focus here on the implications of ourapproach. MoMA is based on a 'complex' metaphor andtherefore brings this complexity into the derived language.Is this complexity sustainable by the users? We have noexperience of use of the implemented system but we derivedits primitives from the ways and means people use to dealwith awareness information as demonstrated by field studiesand individual experiences. These primitives give thelanguage the needed expressive power. It is interesting tonotice that this approach goes in the opposite direction inrelation to Elvin (Fitzpatrick and Kaplan, 2000): there,simplicity and experience of usage were taken as startingpoint while more complex constructs are envisaged as anevolution from the initial ones. Also for Elvin there isthe challenge to be successful when its complexity willincrease. User acceptability of a complex artifact is based on how

it is structured and presented. We paid attention to allowfor an incremental usage of the language to defineawareness mechanisms by different classes of users, indifferent steps of the related learning process (Mark etal., 1997), possibly in differently demanding situations.However, since the main goal of the implementation was toprove the conceptual and technical feasibility of the model,little effort has been paid to the strategies governing thepresentation, to the users and at run time, of the awarenessinformation associated to the target application. In ouropinion the problem of presentation is common to almost allCSCW applications, and becomes particularly critical whenawareness support is concerned since user interfaces have tosensibly combine awareness information with the informationrelated to the current focus of action. Presentation strategies delimit, per se, a research area

where experts of different disciplines could collaborate inorder to discover new ways of presenting information tocooperating actors or letting them organize it, beyond thelimits of the current technology, so as to inform thedevelopment of more effective user interfaces. Someinteresting experiments have been proposed (e.g., in therecent CSCW conferences). They are oriented to challenge thecurrent technology in its capability to combine features(e.g., virtual reality and multimedia) or to provide

different presentation capabilities (e.g., manipulatingimages in various ways) or finally to use different physicalsupports (e.g., screen and walls). Perhaps, another waycould be to start again from the study of how people managethe complexity of the world they live in (awareness wasuncovered in this way!) to highlight the limits of thecurrent solutions and more importantly to get new insightsand breaking ideas. Otherwise, what looks like a solutionfor some aspects can open new problems. For example, asalluded to earlier, various forms of segmentation have beenproposed to cope with the complexity of the awarenessinformation space. But they all run the risk to increase theneed of re-composition, perhaps at a different level or froma different perspective, as discussed in (Grinter, 1998). Our emphasis on the logical integration of cooperation

supports makes the capability to keep the whole manageablethe crucial problem in the design of interfaces. In thisrespect abstraction becomes a key concept. From the modelingpoint of view, and therefore from the point of view of thedesign of a supportive technology, abstraction techniquesare quite consolidated and refer to different ways to 'hideinformation' by means of hierarchical constructs (e.g.,refinement or inheritance) or logical links (as in hypertexttechnology). These approaches showed to be very powerfulfor many purposes, but are likely to be poor whenabstractions and detailed information have to coexists and'interoperate' in the same view, in a dynamic way, possiblyunder user's control. Some recent proposals have beendiscussed in (Donatelli et al., 2000) as a possible sourceof inspiration of an alternative way to look at the problemin order to overcome the above limits: however, theirapplicability still requires a deeper understanding.Abstraction is perhaps only an example of a concept thatshould be investigated before to attack the problem ofsupporting the various abilities and promotion needsconcerning awareness, that have been discussed in theintroduction. This understanding could contribute tointerpret the information supporting cooperation as'knowledge' and 'consciousness' in a mixed and coherent way,and consequently to achieve a smooth integration of therelated supportive technologies.

Acknowledgments The authors want to thank Gloria Mark for her careful

reading of an early version of the paper and Kjeld Schmidt for the manydiscussions on the relation between awareness and articulation work.

References

Agostini, A., G. De Michelis, and A. Grasso (1997): Rethinking CSCWsystems: the architecture of MILANO. In 5th European Conference onCSCW, Lancaster- UK, ed. J. Hugues, W. Prinz, T. Rodden, and K.Schmidt. Kluwer Academic, pp. 33-48.

Bandini, S. and C. Simone (1999): Integrating forms of interaction in adistributed coordination model. Fundamenta Informaticae, . - (toappear).

Benford, S. and L. Fahlén (1993): A Spatial Model of Interaction inLarge Virtual Environments. In 3th European Conference on CSCW,Milano - IT, ed. G. De Michelis, C. Simone, and K. Schmidt. KluwerAcdemic Publishers, pp. 109-124.

Benford, S.., C. Greenhalg, and D. Lloyd (1997): Crowded collaborativevirtual environments. In CHI'97. ACM-Electronic Proceedings. -(http://info.acm.org/sighchi/chi97).

Benford, S. and C. Greenhalgh (1997): Introducing third party objectsinto the Spatial Model of Awareness. In 5th European Conference onCSCW, Lancaster - UK, ed. W. Prinz, T. Rodden, and K. Schmidt. KluwerAcademic Publishers, pp. 189-204.

Bentley, R. and P. Dourish (1995): Medium versus mechanisms: supportingcollaboration through customization. In 4th European Conference onCSCW, Stockholm - Sweden, ed. H. Marmolin, Y. Sunblad, and K.Schmidt. Kluwer Academic, pp. 133-148.

Black, M. (1962): Models and Archetypes. In Studies in Language andPhilosophy. London: Ithaca, pp. 1-16.

Bowers, J., G. Button, and W. Sharrock (1995): Workflow from Within andWithout: Technology and Cooperative Work on the Print IndustryShopfloor. In 4th European Conference on CSCW, Stockholm - Sweden, ed.H. Marmolin, Y. Sundblad, and K. Schmidt. Kluwer Academic Publishers,pp. 51-66.

Boyd, R. and T. Kuhn (1979): Metaphor in science. In Metaphor andThought, ed. A. Ortony. Cambridge: Cambridge University Press.

Carstensen, P. and C. Sørensen (1996): From the social to thesystematic: mechanisms supporting coordination in design. CSCWJournal, vol. 5, no. 4, pp. 387-413.

Chalmers, M. (2000): Information Awareness and Representation. CSCWJournal, . - (in this issue).

Cugola, G. (1998): Tolerating deviations in Process Support Systems viaflexible enactment of Process Models. IEEE-Trans. on SoftwareEngineering, vol. 24, no. 11, pp. 982-1001.

De Michelis, G., E. Leiva-Lobos, and E. Covarrubias Gatica (1997):Augmenting and multiplying spaces for creative design. In ACM-GROUP'97, Phoenix (AZ), ed. S. C. Hayne and W. Prinz. ACM Press, pp.177-186.

Divitini, M. and C. Simone (2000): Supporting different dimensions ofadaptability in workflow modeling. CSCW Journal- Special Issue on'Adaptive Worflow Systems', vol. 9, no. 3-4, pp. 365-397.

Divitini, M., C. Simone, and K. Schmidt (1996): ABACO: coordinationmechanisms in a multi-agent perspective. In Proc. Second InternationalConference on the Design of Cooperative Systems (COOP96), Antibes-Juan-les-Pins, France, 19--22 June1996, pp. 103-122.

Donatelli, S., M. Sarini, and C. Simone (2000): Towards a ContextualInformation Service supporting adaptability and awareness in CSCWsystems. In COOP 2000, Sophia-Antipolis (Fr), ed. R. Dieng, A. Giboin,G. De Michelis, and L. Karsenty, pp. 83-98.

Fitzpatrick, G. and S. Kaplan (2000): Supporting public availability andaccessibility with Elvin. CSCW Journal, . - (in this issue).

Fuchs, L., U. Pankoke-Babbatz, and W. Prinz (1995): Supportingcooperative awareness with local event mechanisms: the Groupdesksystem. In 4th European Conference on CSCW, Stockholm - Sweden, ed. H.Marmolin, Y. Sunblad, and K. Schmidt. Kluwer Academic Publishers, pp.247-262.

Greenhalg, C. and S. Benford (1995): MASSIVE: a collaborative VE forTele-conferencing. ACM TOCHI, vol. 2, no. 3, pp. 239-261.

Grinter, R.E. (1998): Recomposition: put it all back together again. InCSCW'98, Seattle - WA, ed. S. Poltrock and J. Grudin. ACM Press, pp.393-402.

Gutwin, C., M. Roseman, and S. Greenberg (1996): A usability study ofawareness widgets in a shared workspace groupware system. InCSCW96, Boston - MA, ed. M. S. Ackermann. ACM Press, pp. 258-267.

Heath, C., J. Hindmarsh, P. Luff, and D. vom Lehn (2000): Configuringawareness. CSCW Journal, . - (in this issue).

Heath, C., P. Luff, and G.M. Nicholls (1995): The collaborativeproduction of the document: context, genre and the borderline indesign. In COOP95, Juan-les-Pins, ed. M. Zackland. INRIA, pp. 203-218.

Hugues, J., D. Randal, and D. Shapiro (1992): Faltering from ethnographyto design. In CSCW92, Toronto - CA, ed. M. Mantei and R. Baecker. ACMPress, pp. 115-122.

Klein, M. , C. Dellarocas, and A. (eds) Benstein (2000): Special Issueon 'Adaptive Workflow Systems'. CSCW Journal, vol. 9, no. 3-4.

Lakoff, G. and M. Johnson (1999): Philosophy in the flesh. New York: BasicBooks.

Mansfield, T., S. Kaplan, G. Fitzpatrick, T. Phelps, M. Fitzpatrick, andR. Taylor (1997): Evolving Orbit:a progress report on buildinglocales. In ACM-GROUP'97, Phoenix (AZ), ed. S. C. Hayne and W. Prinz.ACM Press, pp. 241-250.

Mariani, J. (1997): SISCO: Providing a cooperation filter for a SharedInformation Space. In ACM-GROUP'97, Phoenix (AZ), ed. S. C. Hayne andW. Prinz. ACM Press, pp. 376-384.

Mark, G. (2000): Conventions and commitments in distributed groups. . -(in this issue).

Mark, G., L. Fuchs, and M. Sohlenkamp (1997): Supporting groupwareconventions trough contextual awareness. In ECSCW97, Lancaster, ed.J. Hugues, W. Prinz, T. Rodden, and K. Schmidt. Kluwer Academic, pp.253-268.

Nehaniv, C. L., ed. (1998): Computation for Metaphors, Analogy, andAgents. LNAI 1562. Berlin: Springer Verlag,.

Nomura, T., K. Hayashi, T. Hazama, and S. Gudmunson (1998): Interlocus:workspace configuration mechanisms for activity awareness. InCSCW'98, Seattle - WA, ed. S. Poltrock and J. Grudin. ACM Press, pp.29-38.

Okubo, A. (1980): Diffusion and Ecological Systems: Mathematical Models.New York: Springer Verlag.

Prinz, W. (1999): Nessie: An awareness environment for cooperativesettings. In 6th European Conference on CSCW-ECSCW99, Copenhagen,DK, ed. S. Bødker, M. Kying, and K. Schmidt. Kluwer Academic, pp.391-410.

Ramduny, D., A. Dix, and T. Rodden (1998): Exploring the design spacefor notification servers. In CSCW'98, Seattle - WA, ed. S. Poltrockand J. Grudin. ACM Press, pp. 227-335.

Rodden, T. (1996): Populating the Application: A Model of Awareness forCooperative Applications. In CSCW ’96. Proceedings of the Conference onComputer-Supported Cooperative Work, Boston, MA, November 16 -20, 1994.New York: ACM Press, pp. 87-96.

Sandor, O., C. Bodgan, and J. Bowers (1997): Aether: an awareness enginefor CSCW. In 5th European Conference on CSCW, Lancaster, UK, ed. W.Prinz, T. Rodden, and K. Schmidt. Kluwer Academi Publishers, pp. 221-236.

Schmidt, K. (1997): Of maps and scripts -The status of formal constructsin cooperative work. In ACM-GROUP'97, Phoenix (AZ), ed. S. C. Hayneand W. Prinz. ACM Press, pp. 138-147.

Schmidt, K. (1998): Some notes on mutual awareness. COTCOS-ReportTechnical University of Denmark.

Schmidt, K. and C. Simone (1996): Coordination Mechanisms: towards aconceptual foundation for CSCW systems design. CSCW Journal, vol. 5,no. 2/3.

Simone, C. and S. Bandini (1997): Compositional features for promotingawareness within and across cooperative applications. In ACM-GROUP'97, Phoenix (AZ), ed. S. C. Hayne and W. Prinz. ACM Press, pp.358-367.

Suchman, L. (1987): Plans and situated actions. The problem of human-machine communication. Cambridge: Cambridge University Press.

Syri, A. (1997): Tailoring cooperation support through mediators. In5th European Conference on CSCW, Lancaster - UK, ed. J. Hughes, W.Prinz, T. Rodden, and K. Schmidt. Kluwer Academic Publishers, pp.157-172.

Turing, A. (1952): The Chemical Basis of Morphogenesis. Philos. Trans. R.Society, vol. 237.

Yamaashi, K., J.R. Coperstock, T. Narine, and W. Buxton (1996): Beatingthe limitations of camera-monitor mediated telepresence with Extra-Eyes. In CHI'96. ACM-Electronic Proceedings. -(http://info.acm.org/sighchi/chi96).