|Conversations and Correspondences for Cooperation
INRIA Sophia Antipolis
Conversations for Cooperation refer to the perspectives on design of cooperative systems which rely on the today prevalent conversational analogyÑ i.e., viewing communication with and through computers as a conversation. Among those perspectives are the ones based on the Òmodels of language useÓ of Clark and his colleagues. From my experience of Clark and colleaguesÕ models of the referring process when applied to computer-mediated communication (CMC), I try to show that the conversational prevalence (a) leaves aside aspects of other communication modes, in particular asynchronicity and Òwritten-ness,Ó which are important in some CMC sites (e.g., Usenet News), and (b) leads to reject or neglect related communication models, especially ÒliteraryÓ models. I hence claim to also value ÒliteraryÓ models and analogies, especially correspondence, and suggest to adapt Clark and colleaguesÕ models to take more account of literary uses. So doing I suggest that designers adopt also the Correspondences for Cooperation perspective, to design Òtools for correspondenceÓ complementing the current Òtools for conversation.Ó
Conversation, Correspondence, Computer-Mediated Communication, Cooperative Work.
Conversation commonly means an ÒÊoral exchange of sentiments, observations, opinions, or ideasÊÓ (Stanford KSL Network Services Webster Gateway). Conversation is currently the prevalent reference model of communication in human-computer cooperation and human-computer-human cooperation. This prevalence has been crystallized as follows by McCarthy and Monk (1994, pp. 35-36)Ê: ÒÊThe analogy between a person conversing with another and a person operating a computer is now well established in human-computer interaction. [...] More recently computer applications have been explicitely designed to facilitate or guide communication within a group. [...] Here again [in computer-mediated communication or cooperative work] the conversational analogyÊhas been influential.ÊÓ
Such expressions as ÒÊtools for conversationÓ (Winograd and Flores, 1986), ÒÊtechnologies that support conversationsÊÓ (Olson and Olson, 1997), Òdesign for conversationÓ (Tatar, Foster and Bobrow, 1991), ÒÊconversation space designÊÓ (Miles et al., 1993; Mc Carthy, Miles and Monk, 1991), ÒÊmultimodal conversationÊÓ (Nagao and Takeuchi, 1994), ÒÊconversant assistantsÊÓ (Gruen and Sidner, 1997)Ê, etc., show clearly how determinant is the conversational prevalence on system design perspective.
Let us call Conversations for Cooperation (by reference to the Conversations for Action perspective proposed by Winograd and Flores, 1986) the perspectives on design of cooperative systems which rely on the conversational analogy. Among those perspectives are the ones based on the ÒÊmodels of language useÊÓ of Clark and his colleagues (see, e.g., Clark, 1992, 1996). Such a perspective is illustrated by BrennanÕs (1991) paper, Conversation with and through Computers.
In this paper I try to show, from my experience of applying some ClarkÕs models of the referring process to computer-mediated communication, that the prevalence of the conversational model (a) leaves aside aspects of other communication modes, in particular asynchronicity and Òwritten-ness,Ó which are important in computer-mediated communication, and (b) leads to reject or neglect related communication models, especially ÒliteraryÓ models, which could be useful for designing cooperative systems and interpreting the behavior of these systems' users.
I hence claim to also value ÒliteraryÓ models and analogies, like correspondence (or Òcommunication by lettersÓ, Stanford KSL Network Services Webster Gateway), and suggest to adapt Clark and colleaguesÕ models to take more account of literary uses. So doing I suggest also to complement the Òanalogy-beltÓ of the designers with the Correspondences for Cooperation perspective, which will lead to the design of Òtools for correspondenceÓ supplementing current tools for conversation.
2. CONVERSATIONS FOR COOPERATION
2.1. Reasons for the Prevalence of the Conversational Model
To explain why some communication models are not satisfactory, Clark and his colleagues steadily hunt down the tacit behind the modelling behavior of language specialists (see example below). I think that we can also search for the implicit, or the Ògiven,Ó to find reasons for the prevalence of the conversational model; it will give light on the negative implications of this prevalence. In the following I just initiate such a search for reasons.
What is given is that (a) computers are interactive tools, and (b) one of the main features of face-to-face conversation is interactivity. Hence bringing conversation and electronic communication together seems evident: moreover the ÒÊconversational analogyÊÓ is acknowledged by common language which also defines conversation as ÒÊan exchange similar to conversation; esp: real-time interaction with a computer esp. through a keyboardÊÓ (Stanford KSL Network Services Webster Gateway).
What is given too is that conversation prevails over other forms of communication also in the modelling of communication behaviors. This is obvious in Clark and colleaguesÕ works. Here are examples of how this prevalence is uttered in these works:Ê ÒÊÊConversation is the fundamental site of language use. [...] other arenas of language useÑnovels, newspapers, lectures, street signs, ritualsÑare derivative or secondary;ÊÓÊ ÒÊif conversation is fundamental, its processes are likely to underlie or shape processes in other uses of language as wellÓ (Clark and Wilkes-Gibbs, 1986, pp.Ê1 and 3). In more recent wordsÊ: ÒÊface-to-face conversation is the basic setting for language useÊ;Ó other settings are ÒdeficientÊÓ (my term): they Òlack the immediacy, medium, or control of face-to-face conversation, so they require special techniques or practices;ÊÓ accordingly the following point must be taken for grantedÊ: ÒÊ If we are ever to characterize language use in all its settings, the one setting that should take priority is face-to-face conversationÊÓÊ (Clark, 1996, p. 11).
A ÒdeeperÓ reason for the conversational prevalence seems to be what I will call the Òonce-and-for-all rejectionÓ attitude, i.e., having rejected once a model of communication because of its inappropriateness, considering implicitely that it has been rejected Òonce and for all.Ó This attitude, I think, would date back to the time of the well-known Clark and Wilkes-GibbsÕ (1986) paper, ÊReferring as a collaborative process. In this paper, Clark and Wilkes-Gibbs proposed a conversational model of definite reference, in which speakers and addressees are viewed as collaborating in the making of references; in the model, referring is thus described as an Òacceptance processÓ governed by a Òprinciple of mutual responsibilityÓ that participants take for granted. Clark and Wilkes-Gibbs called their model of referring a Òconversational modelÓ in contrast with the Òliterary modelÓ of reference, a model Òtraditionally presupposedÓ (at the time) by philosophers, linguists and psychologists accounting for definite reference in conversation. This model was called ÒliteraryÓ because it is grounded on the Òliterary sites of language useÓ such as novels, newspapers and letters, and not on the Òconversational site.Ó
The problem with the literary model, Clark and Wilkes-Gibbs said, is that it embodies ÒÊtacitÊÓ assumptions and ÒidealizationsÓ incompatible with the reference process actually carried on in conversation. Assumptions are mainly that Òspeakers refer as if they were writing to distant writers,Ó and Òretain complete responsibility and control overÓ the referring process, and that addresseesÕ actions Òhave no bearing onÓ the speakers in reference. Clark and Wilkes-Gibbs contested the idealizations and assumptions of the literary model. They deservedly argue that a conversational model of the reference process Òought to look quite differentÓ from the literary model. They demonstrated this from analyses of conversations occurring in natural situations and in an experimental situation (the Tangram task).
Let us call Òmis-situatednessÓ the fact that literary models of the referring process are Òmis-situated,Ó i.e. that they are grounded on incorrect models of the communication situation, causing Òproblems for conversation.Ó If Clark and Wilkes-Gibbs rejected the literary models for their mis-situatedness, they however seemed to reject it once and for all: they have not reconsidered the possible appropriateness of literary models in other occasions.
2.2. Negative Implications of the Prevalence
If conversational models have a Òproven valueÓ for thinking about the design of cooperative systems (McCarthy and Monk, 1994), giving prevalence to these models may however have negative implications on system design, and upstream on the elaboration of the communication process models themselves. I will now present three of these implications, with respect to Clark and colleaguesÕ works.
First, because it considers other sites of language use as secondary or derivative, prevalence may lead to consider conversation as a kind of standard or ideal, and to relate other sites to it. This attitude orientates research and application questions (see, e.g., Whitakker, Brennan and Clark's (1991) question, ÒIn what ways is mediated communication likely to differ from face-to-face interaction?Ó), and answers (see, e.g., Olson and Olson, 1997, referring to Clark, 1996Ê, ÒÊConversation conducted face-to-face has some natural characteristics that are severely disturbed when we converse using technology at long distancesÓ; see also Clark claiming in its CSCW'96 keynote address that Òwhen we converse at a distance in time or space, we modify these [face-to-face conversation] techniques to deal with the constraints of the mediumÓ). The Òconversation-as-standardÓ attitude orientates also modelling; in particular, it may lead to reduce writing/reading to speaking/hearing, to see written or text-based communication as mainly a Òconversation between writers and readersÓ, Òa conversation etched in inkÓ (Meyer, 1985). If discourse in CMC systems (e.g., chat and bulletin board) Òexhibits many characteristics of an oral cultureÓ (December, 1993), this doesn't mean that characteristics of a written culture are absent, or have to be under-estimated.
Second, the conversational prevalence may limit the scope of the communication situations that are studied for modelling the communication process, and it also limits the scope of the CMC applications considered. For example, Clark and his colleagues mainly focus on conversation situations, or situations that ÒresembleÓ the most to conversation situations (see, e.g., Clark, 1996). The applications they consider are mainly synchronous applications, either speech-based (e.g., videoconference, Isaacs and Tang, 1994), or text-based (e.g., Òe-talkÓ situations, Whittaker, Brennan and Clark, 1991). As a consequence, some CMC situations and applications relevant to the designers (e.g., e-mail or Usenet News) are neglected.
Third, the prevalence seems to wreck modelling actions, and consequently deprives designers of relevant models. A good illustration of this can be found in Clark and Wilkes-Gibbs (1986). Questioning the generality of the principle of Òmutual responsibilityÓ in the conclusion of Referring as a collaborative process, Clark and Wilkes-Gibbs notice that the principle doesn't apply in fact to all the situations because partners Òmay have no full access to one another,Ó and that it is necessary to adapt it to account for the incompatible situations. They come to conclude that the principle Òmay get weakened or modified in various ways,Ó depending on languages uses. For literary uses (novels, newspapers, letters), lectures, radio broadcasts, and e-mail, participants are consequently assumed to adhere to the weakened version of the principle of mutual responsibility, the so-called principle of Òdistant responsibility.Ó As far as I know, this principle has not been studied systematically elsewhere, whereas it is considered as Òhighly relevantÓ by some designers of cooperative systems (see Tatar et al., 1991).
These negative implications would lead us to adjust the importance of conversation, and of its main dimensions, in particular orality and synchronicity. As McCarthy and Monk (1994, p. 46), pointed out, ÒText is not generally better or worse than video communication or face-to-face. Rather there is a ÔfitÕ between each of these media and a particular type of task and social context. It is this ÔfitÕ that the designer must be aware of.Ó So we would have to give their place to asynchronicity and Òwritten-ness,Ó in system design and in communication process modelling (or further modelling and re-modelling, if we consider the models of Clark and his colleagues). Hence the Correspondences for Cooperation perspective I would like to propose.
3. CORRESPONDENCES FOR COOPERATION
3.1. Origin of the Perspective : Asynchronous Text-based Communication and Usenet News
The perspective originated in two studies of referential practices of E(lectronic)-news writers, those users who post articles in Usenet News groups such as Òcomp.ai.nat-lang,Ó a newsgroup dealing with computational linguistics. In these studies I questionned the appropriateness of ClarkÕs conversational models of reference to account for the referential practices of E-news writers, and proposed instead a correspondence model. I assumed then that users refer to as ÒcorrespondentsÓ rather than as Òconversants.Ó
In the first study (Giboin, 1996a) I analyzed the correspondence hypothesis for definite reference, and by contrast with the conversational model of definite reference proposed by Clark and Wilkes-Gibbs (1986). In the model I proposed, E-news writers cooperate in referring by playing correspondence games: they exchange ÒlettersÓ (correspondence1), establishing and maintaining mapping relationships (correspondences2) between their representations and processes and the representations and processes of their partners (for a recent elaboration of the notion of correspondence2, see Giboin, 1998). In the model, referring Ñalso called reference1Ñ is directed by Òreference-taking,Ó or reference2. Besides bringing literary models back into favour, our study allowed to revisit some concepts present in the Clark and Wilkes-GibbsÕ model of collaborative reference. For example, I proposed to complement: (a) the notion of Òmutual acceptanceÓ with the notion of Òvirtual joint acceptanceÓ; and (b) the ÒsufficiencyÓ principle (Òto assure understanding to a criterion sufficient for current purposesÓ) with the distinction between the ÒsatisficingÓ and ÒoptimizingÓ principles, both headed by the ÒimportÓ principle (Òwhat is important to correspondents to achieve at the momentÓ). I illustrated possible design implications of the correspondence model with respect to the Netscape newsreader interface.
In a second study (Giboin, 1996b), I analyzed the correspondence hypothesis for demonstrative reference, and by contrast with the conversational model of demonstrative reference proposed by Clark, Schreuder and Buttrick (1983). I pointed out that a tacit assumption behind the Òoptimal designÓ principle governing demonstrative reference in face-to-face conversation is that the speaker has to design nothing but utterances (especially the linguistic expressions, or descriptors, used to refer). The study gave preliminary evidence that, in electronic correspondence, E-news writers perform other design activities than utterance or descriptor design; they frequently design demonstrations (i.e, the ÒgesturesÓ accompanying the descriptors) and sometimes demonstrata (i.e., the entities at which they point). Two examples of demonstration design are: (1) prefixing (manually or automatically) each line of a quoted article with typographical cues, or symbols, like Ò>Ó or Ò>>Ó; (2) elaborating typographical cues such as the Òhorizontal arrow-heads lineÓ (^^^^^^), or Ascii arrows. (1) is a way of orienting readers to appropriate referential sets. (2) is a way of helping readers identify a specific referent. Ascii Art Pictures, and High-Resolution Pictures are some of the that writers use for designing demonstrata.
Thus, what these two studies showed is that correspondence devices and processes are used by E-news writers, and that these devices and processes need also to be mimic or assisted by Òtechnologies of correspondence.Ó
3.2. Evolution of the Perspective : Asynchronous Graphical Communication and Corporate Memory
These two studies led me also to abandon the initial view that electronic correspondence Òis a kind of text-based computer-mediated communication. It would be better viewed as a kind of graphic-based computer-mediated communication. To me, now, E-news correspondence is a situation of graphical communication, in the sense of Alarcos Llorach (1968): it rests on graphical media (from the figurative representation reproducing the object of the communication, to typography), and it is performed through visionÓ (Giboin, 1996b).
To develop the ÒgraphicalÓ aspect within the Correspondences for Cooperation perspective, and in the related model of communication, I am relying in particular on: (a) WallerÕs (1982, 1985) Òtext-as-diagramÓ metaphor, and (b) GoodyÕs (1979) analyses of Ògraphical figures.Ó (a) Designed Òto focus on the written-ness of text,Ó the text-as-diagram metaphor Òproposes that readers who cannot see the wood for the trees may benefit from a mapÑthat typographic and spatial factors can be used to clarify the larger structural relationships in a text, easing the cognitive burden that long, featureless texts impose on the readerÓ (Waller, 1985, p. 107). (b) As opposed to Òrhetorical figures,Ó Ògraphical figuresÓ are devices such as lists, tables, and formulas which fill the two major functions of writing: (1) information storing, and (2) audition-to-vision transfer; (1) allows communicating through time and space, and provides humans with a method for marking, memorizing, and recording; (2) makes it possible to examine differently, to rearrange, and to rectify discourse (Goody, 1979).
Developing the graphical aspect of communication is strongly motivated by the aim of our research team (Acacia), i.e, corporate memory design (cf. Dieng et al., 1998): in its asynchronous form, corporate memory is obviously related to graphical communication (cf. design rationale). Graphical development is also strongly motivated by what interests me in corporate memory, i.e., long-time-distance communication, or communication between persons at different epochs. To analyze and modelize how Òasynchronous partners'' can cooperate using graphical communication, I intend to use theoretical elements present in Clark and colleaguesÕ models, e.g. elements on Òasynchronous joint actionsÓ (Clark, 1996), and elements on memory processes (Clark, 1992). Concerning memory processes, one of my goals is to study graphical communication from the point of view of the cooperation, or correspondence2, between encoding and retrieval processes (see Giboin, 1996a, 1998).
For system designers to understand how the users behave and why they behave the way they do, and to (re)design tools properly, it is important to have a Òrealistic model of the [communication] situationÓ their users (will) face (Tatar et al., 1991). For designers of models of the communication process, like Clark and his colleagues, to build an appropriate process model, it is important, as we noticed, to have a proper model of the situation within which the process occurs, i.e., to avoid Òmis-situatedness.Ó
If one has to select some existing process model for system design purposes, the question to ask seems thus to be: Is the underlying situation model appropriate? Does it account realistically for the situation the users (will) face? What I have attempted to show in this paper is that Clark and colleaguesÕ models of reference, because they give prevalence to conversation situations, so neglecting ÒcorrespondenceÓ situations, may lack important aspects of how the process is situated in some computer-mediated communication settings.
However, because it is the userÕs model of the situation which, in the last intance, will determine the success or the failure of the communication, the question to ask is not only: Is the situation model of the model and system designers appropriate? It is also: Is the situation model of the user itself appropriate? Easterbrook (1995; see also ÊPemberton, 1996), for example, showed that a problem of groupware applications such as email is that they mask occurrences of communication breakdowns, and that part of the problem Òis that expectations for communication through email are based on mental models of other forms of interaction, such as telephone, letter writing and face to face communication. As email does not provide the same cues as any of these other media, breakdowns go unnoticed, and the period of confusion is prolonged.Ó
Finally, it can be suggested that considering all the critical dimensions of the situation, and not only some of them (e.g., medium and purpose, as in Clark and Brennan, 1991) is necessary to get a more realistic model of the situation. It will remain however to determine which are the dimensions that define a situation properly (see, e.g., Van Dijk, 1998).
Alarcos Llorach, E. (1968). Communication orale et graphique. In A. Martinet (Ed), Linguistique, EncyclopŽdie de la PlŽiade, Paris, Gallimard, 515-568.
Brennan, S.E. (1991). Conversation with and through computers. User Modeling and User-Adapted Interaction, 1, 67-86.
Clark, H.H. (1992). Arenas of Language Use. Chicago: The University of Chicago Press.
Clark, H.H. (1996). Using Language. Cambridge: Cambridge University Press.
Clark, H.H. and Brennan, S.E. (1991). Grounding in communication. In L.B. Resnick, J.M. Levine, and S.D. Teasley (Eds), Perspectives on Socially Shared Cognition. APA Books, Washington D.C., pp. 127-149.
Clark, H.H., Schreuder, R. and Buttrick, S.(1983). Common ground and the understanding of demonstrative reference. Journal of Verbal Learning and Verbal Behavior, 22, 245-258. (Also published as a chapter inÊ: H.H. Clark (1992), Arenas of Language Use, pp. 78-99.)
Clark, H.H. and Wilkes-Gibbs, D. (1986). Referring as a collaborative process. Cognition, 22, 1-39. (Also published as a chapter inÊ: (1) P.R. Cohen, J. Morgan and M. Pollack (eds.) (1990), Intentions in Communication, CambridgeÊ: The MIT Press, pp. 463-493 [Curiously, section ÒPerspective and change in perspectiveÓ has been omitted in this edition.], and (2) H.H. Clark (1992), Arenas of Language Use, pp. 107-143.)
December, J. (1993). Characteristics of oral culture in discourse on the Net. Paper presented at the Twelfth Annual Penn State Conference on Rhetoric and Composition, University Park, Pennsylvania, July 8, 1993. http://www.december.com/john/papers/pscrc93.txt
Dieng, R., Giboin, A., AmergŽ, C., Corby, O., DesprŽs, S., Alpay, L. Labidi, S. and Lapalut. S. (1998). Building of a Corporate Memory for Traffic Accident Analysis, AI Magazine, in press.
Easterbrook, S. (1995). Coordination breakdowns: why groupware is so difficult to design. Proceedings, 28th Hawaii International Conference on Systems Sciences (HICSS-28), Maui, Hawaii, 3-6 January, 1995, IEEE Press, pp 191-199.
Giboin, A. (1996a). How E-news writers cooperate in referring: correspondence games. In Proceedings of COOP'96. Second International Conference on the Design of Cooperative Systems (June 12-14, Juan-les-Pins, France), INRIA Press, pp. 37-56.
Giboin, A. (1996b). How E-news writers (do not) help E-news readers identify referents. In Proceedings of UCIS'96. Using Complex Information Systems: Cognitive, Ergonomic, Educational Aspects, Language and Communication Laboratory (University of Poitiers, France) and Research and Development Center for Information Technology in Education (University of Joensuu, Finland), pp. 21-30.
Giboin, A. (1998). DŽfauts de correspondance et conflits argumentatifs dans des dialogues homme-(pseudo)machine sur des diagnostics mŽdicaux. Interaction & Cognitions, in press.
Goody, J. (1979). La raison graphique. La domestication de la pensŽe sauvage. ParisÊ: Editions de Minuit. (French translation ofÊ: The Domestication of the Savage Mind, CambridgeÊ: Cambridge University Press, 1977.)
Gruen, D. and Sidner, C. (1997). Conversant Assistants. Lotus Corporation Research Center, Web page, Last update: 08/13/97, See Lotus Web Server, http://www.lotus.com/ (Click successively on the menu items ÒCommunitiesÓand ÒResearchÓ)
Isaacs, E.A. & Tang, J.C. (1994). What video can and cannot do for collaboration: a case study, Multimedia Systems, Vol. 2, 63-73. (Previously published as: Isaacs, E.A. & Tang, J.C. (1993). What video can and canÕt do for collaboration: a case study, Proceedings ACM Multimedia, Anaheim, CA: ACM, 199-206.)
Mc Carthy, J.C., Miles, V.C., and Monk, A.F. (1991). An experimental study of common ground in text-based communication. In Proceedings of CHI'91 Human Factors in Computing Systems (April 27-May 2, New Orleans, LA), ACM/SIGCHI, NY, 1991, pp. 209-215.
McCarthy, J.C. and Monk, A.F. (1994). Channels, conversation, cooperation and relevanceÊ: all you wanted to know about communication but were afraid to ask. Collaborative Computing, 1, 35-60.
Meyer, B.J.F. (1985). Signaling the structure of text. In D.H. Jonassen (ed.), The Technology of Text. Volume 2Ê: Principles for Structuring, Designing, and Displaying Text, Englewoods CliffsÊ: Educational Technology Publications, 64-89.
Miles, V.C., McCarthy, J.C., Dix, A.J., Harrison, M.D. and Monk, A.F. (1993). Reviewing designs for a synchronous-asynchronous group editing environment. In M. Sharples (Ed.), Computer Supported Collaborative Writing, Springer Verlag, 137-160.
Nagao, K. and Takeuchi, A. (1994). Social interaction: multimodal conversation with social agents. Proceedings of the 12th National Conference on Artficial Intelligence (AAAIÕ94), Vol. 1, pp. 22-28.
Olson, G.M. and Olson J.S. (1997). Research on computer supported cooperative work. In M. Helander, T.K. Landauer, and P. Prabhu (eds.), Handbook of Human-Computer Interaction, Second completely revised edition, AmsterdamÊ: Elsevier Science B.V, pp. 1433-1456.
Pemberton, L. (1996). Telltales and overhearersÊ: participant roles in electronic mail communication. In J.H. Connolly and L. Pemberton (Eds), Linguistic Concepts and Methods in CSCW, LondonÊ: Springer Verlag, pp. 144-161.
Tatar, D.G., Foster, G., and Bobrow, D.G. (1991). Design for conversation: lessons from Cognoter. International Journal of Man-Machine Studies, 34, 185-209.
van Dijk. T.A. (1998) Towards a theory of context and experience models in discourse processing. In H. van Oostendorp & S. Goldman (Eds.), The Construction of Mental Models during Reading. Hillsdale, NJ: Erlbaum, (in press).
Waller, R.H.W. (1982). Text as diagram: using typography to improve access and understanding. In D.H. Jonassen (ed.), The Technology of TextÊ: Principles for Structuring, Designing, and Displaying Text, Englewoods CliffsÊ: Educational Technology Publications, 137-166.
Waller, R.H.W. (1985). Using typography to structure argumentsÊ: a critical analysis of some examples. In D.H. Jonassen (ed.), The Technology of Text. Volume 2Ê: Principles for Structuring, Designing, and Displaying Text, Englewoods CliffsÊ: Educational Technology Publications, 105-125.
Winograd, T. and Flores, F. (1986). Understanding Computers and Cognition. A New Foundation for Design, Reading, Addison-Wesley Publishing Company.
Whittaker, S. Brennan, S.E., and Clark, H.H (1991). Co-ordinating activity: an analysis of interaction in computer-supported co-operative work. In Proceedings of CHI'91 Human Factors in Computing Systems (April 27-May 2, New Orleans, LA), ACM/SIGCHI, NY, 1991, pp. 361-367.