A language/act ion perspective on the design of cooperat ive work

Terry Winograd Stanford University

Abstract:

In designing computer-based systems, we work within a perspective that shapes the design questions that will be asked and the kinds of solutions that are sought. This paper introduces a perspective based on language as action, and explores its consequences for practical system design. The language/action perspective is contrasted to a number of other currently prominent perspectives, and is illustrated with an extended example based on studies of nursing work in a hospital ward. We show how it leads to particular analyses of that work, which reveal potentials for creating new designs that can make the work (and the workers) more effective.

1. Prel iminaries

This paper reflects an approach to the design and analysis of computer-based systems that is not yet prevalent within either academic computer science or practical data processing, but which has been widely discussed and justified (see, for example DeCindio et al., Ehn and Kyng, Flores, Gerson, Kaasb¢ll, Kling and Scacchi, Nygaard, Suchman, Winograd and Flores). Some of the major issues (not necessarily agreed upon by all) can be summarized:

1) In designing computer-based systems for practical situations, we need to consciously focus on the questions of context and application, rather than concentrating on the internal structure and operation of the computer system. The structure of a system determines the possibilities for action by the people who use it, and it is this action structure that is ultimately important. We can develop systematic methods and associated tools for design of that structure.

2) Design is ontological. That is, when we introduce a computer-based system, we are not just designing its structure and function, but are participating in the larger design of the organization and collection of practices in which it plays a role. We are designing (or re-designing) the work, not just the tool.

3) In describing or constructing any system, we are guided by a perspective. This perspective determines the kinds of questions that will be raised and the kinds of solutions that will be sought. Its

scope is not limited to the computer system itself--it is a perspective on the organization and the kinds of

actions people take within it. 4) One can consciously apply a perspective as a guide to

design. It will not provide answers to all of the

specific design questions, but serves to generate the questions that will demand answers.

Within the community concerned with the design of computer systems, there is a growing recognition of the importance of perspectives, and the need to become more explicitly aware of how they determine design. One does not have the choice of whether to have a perspective, but can choose to adopt or strengthen particular perspectives that can lead to better designs.

One body of research that has set out to explicitly articulate a perspective for system design is the one based on understanding language as the primary dimension of human cooperative activity. It draws on work developed by Flores and his colleagues at Stanford University, Logonet, and Action Technologies [see Flores (1981), Flores & Ludlow (1981), Winograd & Flores (1986)], and has been the basis for designing commercially successful computer systems.

This paper is an attempt to clarify and illustrate that perspective. It addresses concrete questions of system development, looking at examples drawn from the nursing work in a hospital ward, as studied by Kaasb¢ll (1986, 1987) and Fafchamps (in preparation). The examples were chosen because there was a body of careful description and analysis of the setting and the structure of work. No actual computer system has been

203

A language/action perspective on the design of cooperative work

Terry WinogradStanford University

Abstract:

In designing computer-based systems, we work within aperspective that shapes the design questions that will beasked and the kinds of solutions that are sought. This paperintroduces a perspective based on language as action, andexplores its consequences for practical system design. Thelanguage/action perspective is contrasted to a number ofothercurrently prominent perspectives, and is illustrated with anextended example based on studies of nursing work in ahospital ward. We show how it leads to particular analyses ofthat work, which reveal potentials for creating new designsthat can make the work (and the workers) more effective.

1. Preliminaries

This paper reflects an approach to the design andanalysis of computer-based systems that is not yetprevalent within either academic computer science orpractical data processing, but which has been widelydiscussed and justified (see, for example DeCindio et al.,Ehn and Kyng, Flores, Gerson, Kaasbl1l11, Kling andScacchi, Nygaard, Suchman, Winograd and Flores).Some of the major issues (not necessarily agreed uponby all) can be summarized:

1) In designing computer-based systems for practicalsituations, we need to consciously focus on thequestions of context and application, rather thanconcentrating on the internal structure andoperation of the computer system. The structure ofasystem determines the possibilities for action by thepeople who use it, and it is this action structure thatis ultimately important. We can develop systematicmethods and associated tools for design of thatstructure.

2) Design is ontological. That is, when we introduce acomputer-based system, we are not just designing itsstructure and function, but are participating in thelarger design of the organization and collection ofpractices in which it plays a role. We are designing(or re-designing) the work, not just the tool.

3) In describing or constructing any system, we areguided by a perspective. This perspectivedetermines the kinds of questions that will be raisedand the kinds of solutions that will be sought. Its

scope is not limited to the computer system itself-itis a perspective on the organization and the kinds ofactions people take within it.

4) One can consciously apply a perspective as a guide todesign. It will not provide answers to all of thespecific design questions, but serves to generate thequestions that will demand answers.

Within the community concerned with the design ofcomputer systems, there is a growing recognition of theimportance of perspectives, and the need to becomemore explicitly aware of how they determine design.One does not have the choice of whether to have aperspective, but can choose to adopt or strengthenparticular perspectives that can lead to better designs.

One body of research that has set out to explicitlyarticulate a perspective for system design is the onebased on understanding language as the primarydimension of human cooperative activity. It draws onwork developed by Flores and his colleagues at StanfordUniversity, Logonet, and Action Technologies [seeFlores (1981), Flores & Ludlow (1981), Winograd &Flores (1986)], and has been the basis for designingcommercially successful computer systems.

This paper is an attempt to clarify and illustratethat perspective. It addresses concrete questions ofsystem development, looking at examples drawn fromthe nursing work in a hospital ward, as studied byKaasbl1l11 (1986, 1987) and Fafchamps (in preparation).The examples were chosen because there was a body ofcareful description and analysis of the setting and thestructure of work. No actual computer system has been

203

introduced into the work or even designed in detail. The goal here is to illustrate the kind of analysis that is relevant to the broad design questions that would guide work on a computer-based system.

2. Perspectives

A number of writers have described perspectives that designers bring to computer-based systems. Nygaard (1985) argues that the perspective of a computer scientist (or systems analyst) differs in important ways from the perspectives of people within the organization. Ehn and Kyng (1984) propose a 'tool perspective' that takes the 'labour process' as its origin, rather than data or information flow. This shift leads to significant differences in attitudes towards skill and hierarchical control of work, with numerous implications for design. Kling and Scacchi (1982) distinguish a narrow system-oriented perspective from a broader 'web' perspective in which the computer system is situated in a complex of political, economic, and social factors that shape its structure, use, and effects.

This paper identifies a perspective by its declaration of what people do in an organization. The most prominent perspectives in the current literature suggest the following:

1) People process informat ion and make

decisions. This is the predominant perspective,

both in traditional data processing and in current artificial intelligence approaches. The organization is a collection of 'components' (some human, and some mechanical), each of which 'processes and transmits information.' Information processing includes 'decision making' based on a structure of goals, plans, and knowledge. The activity of the

organization is understood by analyzing how the information is entered, stored, and transmitted, and how decisions are made on the basis of it.

2) People ca r ry out funct ional roles, using col lect ions of materials , accord ing to stable

rules. In his development of 'coordination system

technology,' Holt (1983) presents a 'role/activity theory' based on primitive notions of role, action, and interaction. Rather than emphasizing the cognitive activity of individuals, it focusses on their roles (which specify sets of behaviors) and on the temporal and spatial structure of their potential interactions (behaviors in which more than one role takes part), using Petri nets as a theoretical tool. The goal is to specify the rules of interaction: "If it's

rules the computer can help; if it isn't, it can't" (Holt 1983). DeCindio et al. (1985) also characterize th~ activities in an office as the rules of a game anc propose a design methodology based on determinin{ these rules and the strategies that go along witl

following them. 3) People c rea te and main ta in a s t ruc tu re o

author i ty . Every organizational structure include

lines of responsibility and authority embodied in th patterns of activity. It has been pointed out (b Kling and Scacchi (1982), among others) th~ system designs based on idealized concepts c information and function may fail (throug non-acceptance or subversion) if they do not respec the existing authority structure or if they change: in politically unacceptable ways.

4) People negot ia te and promote compet in

interests . Most of the analyses above adopt sorv

sort of 'harmony perspective' in which shared goa and values determine the activities. Ciborra (1985 Sandberg (Nygaard, 1986), and others have argue that a 'conflict perspective' may be more relevant 1 settings in which the competing interests of t[ various parties play a prominent role. Systems i~ negotiation are no less amenable to automatic than those for harmonious action, but must respe different priorities.

5) Peop le enter into personal re lat ionships . ] every human setting, regardless of organization purpose, people form personal relationships al express them through their activities. This includ individual relationships and group identificati~ ('morale', ~corporate culture,' etc.). System desi[ can have a significant impact on the developme and expression of relationships. If we take t] individual components as information processo] these other dimensions of human interaction may ] neglected.

Each of these perspectives generates an importa domain of questions that lead to design. The focus this paper is another perspective on what people they act through language. This does not just me~ uttering and writing sentences in a natural langua such as English. As the following sections w demonstrate, there is a coherent broader view language activity, which includes interactions wi computers and covers the use of formal and technic languages (such as mathematics) as well as ordina language.

204

introduced into the work or even designed in detail.The goal here is to illustrate the kind of analysis that isrelevant to the broad design questions that would guidework on a computer-based system.

2. Perspectives

A number of writers have described perspectives thatdesigners bring to computer-based systems. Nygaard(1985) argues that the perspective of a computerscientist (or systems analyst) differs in important waysfrom the perspectives of people within the organization.Ehn and Kyng (1984) propose a 'tool perspective' thattakes the 'labour process' as its origin, rather than dataor information flow. This shift leads to significantdifferences in attitudes towards skill and hierarchicalcontrol of work, with numerous implications for design.Kling and Scacchi (1982) distinguish a narrowsystem-oriented perspective from a broader 'web'perspective in which the computer system is situated ina complex of political, economic, and social factors thatshape its structure, use, and effects.

This paper identifies a perspective by its declarationof what people do in an organization. The mostprominent perspectives in the current literaturesuggest the following:

1) People process information and make

deci!dons. This is the predominant perspective,

both in traditional data processing and in currentartificial intelligence approaches. The organizationis a collection of 'components' (some human, andsome mechanical), each of which 'processes andtransmits information.' Information processingincludes 'decision making' based on a structure ofgoals, plans, and knowledge. The activity of the

organization is understood by analyzing how theinformation is entered, stored, and transmitted, andhow decisions are made on the basis of it.

2) People carry out functional roles, usingcollections of materials, according to stable

rules. In his development of 'coordination system

technology,' Holt (1983) presents a 'role/activitytheory' based on primitive notions of role, action,and interaction. Rather than emphasizing thecognitive activity of individuals, it focusses on theirroles (which specify sets of behaviors) and on thetemporal and spatial structure of their potentialinteractions (behaviors in which more than one roletakes part), using Petri nets as a theoretical tool.The goal is to specify the rules of interaction: "If it's

rules the computer can help; ifit isn't, it can't" (Holt1983). DeCindio et al. (1985) also characterize thEactivities in an office as the rules of a game ampropose a design methodology based on determininlthese rules and the strategies that go along wit!following them.

3) People create and maintain a structure 0

authority. Every organizational structure include

lines of responsibility and authority embodied in thpatterns of activity. It has been pointed out (bKling and Scacchi (1982), among others) thasystem designs based on idealized concepts (information and function may fail (througnon-acceptance or subversion) if they do not respe<the existing authority structure or if they change:in politically unacceptable ways.

4) People negotiate and promote competin

interests. Most of the analyses above adopt som

sort of 'harmony perspective' in which shared goaand values determine the activities. Ciborra (1985Sandberg (Nygaard, 1986), and others have argmthat a 'conflict perspective' may be more relevant 1

settings in which the competing interests of tlvarious parties playa prominent role. Systems fi

negotiation are no less amenable to automaticthan those for harmonious action, but must respedifferent priorities.

5) People enter into personal relationships. ]

every human setting, regardless of organizationpurpose, people form personal relationships aIexpress them through their activities. This includindividual relationships and group identificati<('morale', 'corporate culture,' etc.). System desi!can have a significant impact on the developmeand expression of relationships. If we take tlindividual components as information processOlthese other dimensions of human interaction may Ineglected.

Each of these perspectives generates an importadomain of questions that lead to design. The focusthis paper is another perspective on what people cthey act through language. This does not just me:uttering and writing sentences in a natural languasuch as English. As the following sections wdemonstrate, there is a coherent broader viewlanguage activity, which includes interactions wicomputers and covers the use of formal and techniclanguages (such as mathematics) as well as ordinalanguage.

204

As a broad framework for outlining a language/action perspective, we will adopt the traditional subdivisions of linguistic theory: syntax,

semantics and pragmatics. We will use these terms in a more general sense than one finds in the technical linguistic l i terature. The point is not to delineate precise boundaries, but to make distinctions that help organize an understanding of what people do.

1) S y n t a x is the structure of the visible (or audible)

forms of language. The syntactic rules (or 'grammar') of a language determine the form of the

basic elements (letters, words, etc.) and the ways in which they can be combined. In an extended sense one can talk about the syntax of an equation, a spreadsheet, or an invoice, or even of an event such as but toning a menu item on a screen. What dist inguishes syntax from the other levels of analysis is that it does not take into account any interpretat ion or at tr ibution of meaning.

Fur ther distinctions among analyses of sound pat terns (phonetics and phonology), word-internal

s tructure (morphology) and phrase structure (syntax) are not relevant here, so 'syntax' is used to cover all of these aspects and their analogs.

2) S e m a n t i c s is the systematic relation between

structures in a language and a space of potential meanings. It includes the definitions of individual

elements (e.g., words) and the meaning that is generated by combining them (e.g., the meaning of "Jill sees Bill" as different from "Bill sees Jill"). In extension, one can talk about the semantics of a part icular blank on a form, or of an operating system command.

3) P r a g m a t i c s deals with issues of language use. A

classical example is "It's cold in here" spoken by a master to a servant. Although the literal meaning is a s ta tement about the temperature, the intent is to evoke an action by the servant. Two fundamental aspects of pragmatics can be distinguished, as described below: action and context.

The following sections will describe each of these areas in more detail. Section 7 recapitulates the distinctions in the context of the hospital example, i l lustrating how the theoretical framework applies to a part icular situation.

3. The pragmatics of language action

The language/action perspective emphasizes pragrnat ics--not the form of language, but what people

do with it. The theory of speech acts is a useful s tar t ing point for understanding. However, as the following sections will show, speech act theory is only one par t of a larger picture.

3.1 Speech act theory

Austin (1962) noted that not all ut terances are s ta tements whose t ruth or falsity is at stake. Performatives, such as "I pronounce you husband and wife" are actions, which can be made appropriately (felicitously) or not, but which are neither true nor false

in a simple sense. Similarly, the language actions of commands, questions, and apologies are not descriptions of a non-linguistic world.

Searle (1975) identified five fundamental ' i l locutionary points ' - - things you can do with an utterance:

Assertive: Commit the speaker (in varying degrees) to something's being the case-- to the truth of the expressed proposition.

Directive: Attempt (in varying degrees) to get the

hearer to do something. These include both questions (which can direct the hearer to make an assertive speech act in response) and commands (which direct the hearer to carry out some linguistic or non-linguistic act).

Commissive: Commit the speaker (again in varying

degrees) to some future course of action. Declaration: Bring about the correspondence between

the propositional content of the speech act and reali ty (e.g., pronouncing a couple married).

Expressive: Express a psychological state about a state of affairs (e.g., apologizing and praising).

Three points deserve note:

1) The illocutionary point of an ut terance is interpreted by speaker and hearer in a background. A commissive need not include the words "I promise" or "I will," but can be "I guess" or "a dollar" (in response to "Can you give me anything?") or jus t a facial gesture. The identification of a language act depends on the backgrounds of speaker and hearer, and is a lways open to differences of interpretation. In fact, the same utterance may be interpreted as different types of act. "It's time for lunch" might be interpreted as an assertion and as a directive, depending on who says it to whom in what circumstances.

2) Directives and commissives (which will informally be called 'requests ' and 'promises' here) always deal

205

As a broad framework for outlining alanguage/action perspective, we will adopt thetraditional subdivisions of linguistic theory: syntax,semantics and pragmatics. We will use these terms in amore general sense than one finds in the technicallinguistic literature. The point is not to delineateprecise boundaries, but to make distinctions that helporganize an understanding of what people do.

1) Syntax is the structure of the visible (or audible)

forms of language. The syntactic rules (or'grammar') of a language determine the form of thebasic elements (letters, words, etc.) and the ways inwhich they can be combined. In an extended senseone can talk about the syntax of an equation, aspreadsheet, or an invoice, or even of an event suchas buttoning a menu item on a screen. Whatdistinguishes syntax from the other levels ofanalysis is that it does not take into account anyinterpretation or attribution ofmeaning.

Further distinctions among analyses of soundpatterns (phonetics and phonology), word-internalstructure (morphology) and phrase structure(syntax) are not relevant here, so 'syntax' is used tocover all of these aspects and their analogs.

2) Semantics is the systematic relation between

structures in a language and a space of potentialmeanings. It includes the definitions of individualelements (e.g., words) and the meaning that isgenerated by combining them (e.g., the meaning of"Jill sees Bill" as different from "Bill sees Jill"). Inextension, one can talk about the semantics of aparticular blank on a form, or of an operatingsystem command.

3) Pragmatics deals with issues of language use. A

classical example is "It's cold in here" spoken by amaster to a servant. Although the literal meaningis a statement about the temperature, the intent isto evoke an action by the servant. Two fundamentalaspects of pragmatics can be distinguished, asdescribed below: action and context.

The following sections will describe each of these areasin more detail. Section 7 recapitulates the distinctionsin the context of the hospital example, illustrating howthe theoretical framework applies to a particularsituation.

3. The pragmatics of language action

The language/action perspective emphasizespragmatics-not the form oflanguage, but what people

205

do with it. The theory of speech acts is a useful startingpoint for understanding. However, as the followingsections will show, speech act theory is only one part ofa larger picture.

3.1 Speech act theory

Austin (1962) noted that not all utterances arestatements whose truth or falsity is at stake.Performatives, such as "I pronounce you husband andwife" are actions, which can be made appropriately(felicitously) or not, but which are neither true nor falsein a simple sense. Similarly, the language actions ofcommands, questions, and apologies are notdescriptions ofa non-linguistic world.

Searle (1975) identified five fundamental'illocutionary points'-things you can do with anutterance:

Assertive: Commit the speaker (in varying degrees) tosomething's being the case-to the truth of theexpressed proposi tion.

Directive: Attempt (in varying degrees) to get thehearer to do something. These include bothquestions (which can direct the hearer to make anassertive speech act in response) and commands(which direct the hearer to carry out some linguisticor non-linguistic act).

Commissive: Commit the speaker (again in varyingdegrees) to some future course of action.

Declaration: Bring about the correspondence betweenthe propositional content of the speech act andreality (e.g., pronouncing a couple married).

Expressive: Express a psychological state about a stateofaffairs (e.g., apologizing and praising).

Three points deserve note:

1) The illocutionary point of an utterance is interpretedby speaker and hearer in a background. Acommissive need not include the words "I promise"or "I will," but can be "I guess" or "a dollar" (inresponse to "Can you give me anything?") or just afacial gesture. The identification of a language actdepends on the backgrounds of speaker and hearer,and is always open to differences of interpretation.In fact, the same utterance may be interpreted asdifferent types of act. "It's time for lunch" might beinterpreted as an assertion and as a directive,depending on who says it to whom in whatcircumstances.

2) Directives and commissives (which will informally becalled 'requests' and 'promises' here) always deal

with a future action. They differ in whether the

action is to be taken by speaker or the hearer. 3) Speech acts take effect by virtue of public

declara t ion--by mutua l knowledge of hearer and

speaker tha t the act has been made. This is especially obvious in the case of declarations and expressives (an apology mut tered bu t not heard is not an apology), bu t is equally true of the others.

3.2 C o n v e r s a t i o n s for act ion

Speech acts are not individual unrelated events, but part icipate in larger conversational s t ructures (see

Flores (1981), Flores & Ludlow (1981)). In a simple conversation for action, one party (A) makes a request to another (B). The request is interpreted by each party as

having certain conditions of satisfaction, which characterize a future course of actions by B. After the

initial ut terance (the request), B can accept (and thereby commit to satisfy the conditions); reject (and thereby end the conversation); or counter-offer with al ternat ive conditions. Each of these in turn has its

own possible continuations (e.g., after a counter-offer, A can accept, reject, or counter-offer again). The overall s t ructure is d iagrammed in Figure 1 (reprinted from

Winograd & Flores, 1986), with circles representing conversation s tates and lines representing speech acts.

This diagram is not a model of the mental state of a

speaker or hearer, bu t shows the conversation as a 'dance,' in which the acts generate the structure of

completion of the conversation. Each par ty has a pat tern of expectations about wha t might happen next, as the conversation moves towards mutua l agreement

Q A: Request ~( v%

B: Reject A: Withdraw

tha t the action has been completed (or that it will not b completed). Of course there are other possibilities i which the conversational acts themselves are taken a topic, for example in questioning intelligibilfl CWhat, I didn't hear you") or legit imacy ("You can order me to do that!"). But the basic logic deals with tl~

progression of steps towards mutua l recognition completion.

In the 'normal' course of events, B at some poll asserts to A that the conditions of satisfaction hal been met (moving to state 4). If A declares tha t he , she is satisfied, the conversation reaches a successi~ completion (state 5). On the other hand, A may n interpret the si tuation in the same way and ms declare tha t the conditions have not been mc re turning the conversation to state 3. In this sta ei ther par ty may propose a change to the conditions satisfaction. In any state one or the other par ty m~

back out on the deal, moving to different s tates completion (states 7 and 9).

Several points about this conversation s t ructu deserve note:

1) We are using 'conversation' in a very general sense indicate a coordinated sequence of acts tha t can l

interpreted as having linguistic meaning. It ne, not be a spoken conversation, or even the use ordinary language. A doctor who writes treatme:

requests on a pat ient form is engaged in conversation with the nurse who will administer t] t reatments , even if they never speak face-to-face.

A: Declare

,~p~ B: Promise

\ \ ~ A: Accept ~ I:~Cenug~ / N~C°unter/ ~__ /

A: C°unt~er~~ ~ U / \ A: Withdraw

A: Withdraw A:/ Reject ~ ~

B: Withdraw

Declar~ 0

Figure 1. The Bas ic Conversa t ion for Act ion.

206

with a future action. They differ in whether theaction is to be taken by speaker or the hearer.

3) Speech acts take effect by virtue of publicdeclaration-by mutual knowledge of hearer andspeaker that the act has been made. This isespecially obvious in the case of declarations andexpressives (an apology muttered but not heard isnot an apology), but is equally true ofthe others.

3.2 Conversations for action

Speech acts are not individual unrelated events, butparticipate in larger conversational structures (seeFlores (1981), Flores & Ludlow (1981». In a simpleconversation for action, one party (A) makes a request toanother (B). The request is interpreted by each party ashaving certain conditions of satisfaction, whichcharacterize a future course of actions by B. After theinitial utterance (the request), B can accept (andthereby commit to satisfy the conditions); reject (andthereby end the conversation); or counter-offer withalternative conditions. Each of these in turn has itsown possible continuations (e.g., after a counter-offer, Acan accept, reject, or counter-offer again). The overallstructure is diagrammed in Figure 1 (reprinted fromWinograd & Flores, 1986), with circles representingconversation states and lines representing speech acts.

This diagram is not a model of the mental state of aspeaker or hearer, but shows the conversation as a'dance,' in which the acts generate the structure ofcompletion of the conversation. Each party has apattern of expectations about what might happen next,as the conversation moves towards mutual agreement

that the action has been completed (or that it will not bcompleted). Of course there are other possibilities iwhich the conversational acts themselves are taken E

a topic, for example in questioning intelligibilil("What, I didn't hear you'') or legitimacy ("You canorder me to do thatl"). But the basic logic deals with tbprogression of steps towards mutual recognition ,completion.

In the 'normal' course of events, B at some POilasserts to A that the conditions of satisfaction ha'been met (moving to state 4). If A declares that he ,she is satisfied, the conversation reaches a successficompletion (state 5). On the other hand, A may n,interpret the situation in the same way and mE

declare that the conditions have not been m~

returning the conversation to state 3. In this staeither party may propose a change to the conditionssatisfaction. In any state one or the other party mlback out on the deal, moving to different statescompletion (states 7 and 9).

Several points about this conversation structudeserve note:

1) We are using 'conversation' in a very general senseindicate a coordinated sequence of acts that can 1interpreted as having linguistic meaning. It nelnot be a spoken conversation, or even the useordinary language. A doctor who writes treatmerequests on a patient form is engaged inconversation with the nurse who will administer tJtreatments, even if they never speak face-to-face.

A: Declare

A: Request

B: Reject

A: Withdraw

B: Promise

A: Reject/

B: Withdraw

A: Declare 0

Figure 1. The Basic Conversation for Action.

206

2) I t is init iated by a request (a corresponding structure can also be init iated by a promise), and therefore is rooted in the anticipation of some future action.

3) At each point in the conversation, there is a small set of possible action types, determined by the previous history. Each type has unlimited possibilities for detailed content. For example, the action 'counter-offer' includes any number of possible new conditions of satisfaction.

4) All of the acts are l inguis t ic-- they represent ut terances by parties to the conversation (or silences that are listened to as s tanding for an act). The act tha t follows a promise is an assertion (an assertive speech act) from the original hearer to the requestor tha t the request has been satisfied. It is followed by a declaration by the requestor tha t it is satisfactory. The actual doing of whatever is needed to meet the conditions of satisfaction lies outside of the

conversation structure. 5) Many acts are 'listened to' without being explicit. If

the requestor can recognize satisfaction of the

request directly, there may be no explicit assertion of completion. Other acts, such as declaring satisfaction, may be taken for granted if some

amount of time goes by without a declaration to the contrary. What is not said is listened to as much as what is said.

6) Conditions of satisfaction are not objective realities, independent of interpretations. They exist in the listening, and there is always the potential for difference among the parties. This can lead to breakdowns (for example, when the promiser asserts that the commitment is satisfied and the

requestor does not agree) and to a subsequent conversation about the understanding of the conditions.

7) There are a few states of completion (the heavy circles in the figure) in which it is mutual ly recognized the neither par ty is wait ing for further action by the other. All other states represent an incomplete conversation. Completion does not guarantee satisfaction. For example, if the promiser takes the action of 'reneging,' the conversation is completed without the original request being satisfied.

8) The network does not say what people should do, or deal with consequences of the acts (such as backing out of a commitment). These are important human phenomena, but are not generated in the domain formalized in this network.

Conversations for action are the central action s tructure for human organizations. We work together

by making commitments so that we can successfully anticipate the actions of others and coordinate them with our own. The emphasis here is on language as an activity, not as the transmission of information or as

the expression of thought. Although people think when they use language, and often describe things in language, the relevant s t ructures for analysis are the

language acts and the conversations into which they are woven. In computer system design, we are not concerned with duplicating the knowledge or thought pat terns of people, but with the structure of their language interactions and the embedding of those interactions in computer systems. The rules of conversation structure (such as those represented in the simple diagram above) can be embodied in

computer-based tools, as long as they are sufficiently precise and recurrent.

This recurrence may be universal (as with the diagram of Figure 1) or part icular to a setting. For

example, certain kinds of requests are made implicitly on the basis of a long-term declaration. A boss does not explicitly request each worker to come to work each

morning, although the conversation proceeds (in those cases where there is a breakdown) as though he or she

had. The recurrent request is listened to as an effect of the declaration "You're hired" within a shared understanding of common practices.

3.3 Other c o n v e r s a t i o n s

Conversations for action (CfA) form the central fabric of cooperative work. However many language acts do not participate directly in the completion of a CfA. Remarks such as "They're planning to remodel the West Wing next summer" need not relate to any specific future actions of speaker or hearer. From a cognitive perspective, one might choose to characterize these as 'conveying information' without a part icular motivation in action. From the perspective of language as action, the pr imary concern is with the role tha t all conversations (or ut terances within conversations) play with respect to action and potentials for action. We distinguish several additional kinds of conversation that go along with conversations for action: conversation for clarification, conversation for possibilities, and conversation for orientation. There is no sharp line between them, but they are accompanied by different moods.

207

2) It is initiated by a request (a corresponding structurecan also be initiated by a promise), and therefore isrooted in the anticipation of some future action.

3) At each point in the conversation, there is a small setof possible action types, determined by the previoushistory. Each type has unlimited possibilities fordetailed content. For example, the action'counter-offer' includes any number of possible newconditions ofsatisfaction.

4) All of the acts are linguistic-they representutterances by parties to the conversation (or silencesthat are listened to as standing for an act). The actthat follows a promise is an assertion (an assertivespeech act) from the original hearer to the requestorthat the request has been satisfied. It is followed bya declaration by the requestor that it is satisfactory.The actual doing of whatever is needed to meet theconditions of satisfaction lies outside of theconversation structure.

5) Many acts are 'listened to' without being explicit. Ifthe requestor can recognize satisfaction of therequest directly, there may be no explicit assertionof completion. Other acts, such as declaringsatisfaction, may be taken for granted if someamount of time goes by without a declaration to thecontrary. What is not said is listened to as much aswhat is said.

6) Conditions of satisfaction are not objective realities,independent of interpretations. They exist in thelistening, and there is always the potential fordifference among the parties. This can lead tobreakdowns (for example, when the promiserasserts that the commitment is satisfied and therequestor does not agree) and to a subsequentconversation about the understanding of theconditions.

7) There are a few states of completion (the heavycircles in the figure) in which it is mutuallyrecognized the neither party is waiting for furtheraction by the other. All other states represent anincomplete conversation. Completion does notguarantee satisfaction. For example, if the promisertakes the action of 'reneging,' the conversation iscompleted without the original request beingsatisfied.

8) The network does not say what people should do, ordeal with consequences of the acts (such as backingout of a commitment). These are important humanphenomena, but are not generated in the domainformalized in this network.

Conversations for action are the central actionstructure for human organizations. We work togetherby making commitments so that we can successfullyanticipate the actions of others and coordinate themwith our own. The emphasis here is on language as anactivity, not as the transmission of information or asthe expression of thought. Although people think whenthey use language, and often describe things inlanguage, the relevant structures for analysis are thelanguage acts and the conversations into which theyare woven. In computer system design, we are notconcerned with duplicating the knowledge or thoughtpatterns of people, but with the structure of theirlanguage interactions and the embedding of thoseinteractions in computer systems. The rules ofconversation structure (such as those represented in thesimple diagram above) can be embodied incomputer-based tools, as long as they are sufficientlyprecise and recurrent.

This recurrence may be universal (as with thediagram of Figure 1) or particular to a setting. Forexample, certain kinds of requests are made implicitlyon the basis of a long-term declaration. A boss does notexplicitly request each worker to come to work eachmorning, although the conversation proceeds (in thosecases where there is a breakdown) as though he or shehad. The recurrent request is listened to as an effect ofthe declaration "You're hired" within a sharedunderstanding ofcommon practices.

3.3 Other conversations

Conversations for action (CfA) form the central fabric ofcooperative work. However many language acts do notparticipate directly in the completion of a CfA.Remarks such as "They're planning to remodel theWest Wing next summer" need not relate to any specificfuture actions of speaker or hearer. From a cognitiveperspective, one might choose to characterize these as'conveying information' without a particularmotivation in action. From the perspective oflanguageas action, the primary concern is with the role that allconversations (or utterances within conversations) playwith respect to action and potentials for action. Wedistinguish several additional kinds of conversationthat go along with conversations for action:conversation for clarification, conversation forpossibilities, and conversation for orientation. There isno sharp line between them, but they are accompaniedby different moods.

207

In a c o n v e r s a t i o n for c la r i f i ca t ion the participants

cope with or anticipate breakdowns concerning interpretat ions of the conditions of satisfaction for a

CfA. The conditions are always interpreted with respect to an implicit shared background, bu t the

sharing is partial and needs to be negotiated. As a

simple example, the request "Give the pat ient some diazine" might evoke responses such as "Right now, or with the morning meds?" or "What dosage?".

One can never guarantee that everything is totally precise. Precision is relative to each party's implicit anticipation that the other par ty will have a

sufficiently shared background to fill in the rest of the details appropriately.

In a c o n v e r s a t i o n fo r poss ib i l i t ies , the mood is one of

generat ing conversations for action. Specific conditions of satisfaction will emerge in the course

of the conversation, and associated conversations for action will be initiated. Many gatherings that are called 'meetings' are in this mood. The meet ing is a

failure if some action does not come out of the discussion. Some conversations for possibilities are highly routinized. For example, 'work rounds' on a

hospital ward is a routine conversation for possibilities, during which the medical team visits each pat ient and specific requests ('orders') and

commitments are generated. In a c o n v e r s a t i o n f o r or i en ta t ion , the mood is of

creating a shared background as a basis for interpret ing future conversations. This shared background includes specific knowledge, interpersonal relations, and general att i tudes. The

most obvious examples are meet ings labelled 'orientation, ' in which newcomers begin to develop the unders tanding that is required to function in the

organization. Conversations for orientation are prominent in less formal sett ings ('shooting the bull' or ' telling stories'). The mood here is not directed

towards action, but it is important to recognize how critical it is for people to develop shared orientation as the basis for future effective action and appropriate interpretat ion of language acts.

4. The pragmat ics of context

Many current linguistic theories deal in some way with context or 'situation' (see, for example, Barwise and Perry, 1983). They focus on 'deictic' expressions, whose meaning is rooted in the conversation situation (e.g., "you," "here," and "tomorrow") and on 'anaphoric'

expressions (such as " that old man" and "the othe one") whose reference depends on complex factors of th

previous conversation and shared sett ing of speak~ and hearer (e.g., what is mutua l ly visible).

But there is a deeper problem. The interpretat ion q

every term and every sentence depends on situation, carried in the background of speaker and hearer. Fc example, a phrase tha t would convey a specific reque in a face to face conversation might become ambiguol and improper in an on-line system where the eventu reader 's identi ty is not known to the wri ter (especiaI

if there is no opportuni ty for conversations clarification). Language tha t has one meaning to nurse on a hospital ward may mean complete different things to the patient, or even to tha t san nurse in a later conversation.

The meaning of a language act exists in t]

interpretat ions of a speaker and a hearer in a part icul si tuation with their part icular backgrounds. If eve: si tuation and background were incomparably uniqc

there would be no way to provide for regular pat terr But of course this is not the case. In every sett ing the are regular i t iesmor ' recurrences ' - - tha t can identified and/or declared.

I nd iv idua l s : The most obvious regular i ty is tha t

personal identity. We can identify individuals m expect their interpretat ions to remain relative stable: what I read today, I will unders tand in t: same way tomorrow (more or less). Of course the are ever-changing moods', and each pers, experiences learning and personal change. cannot assume the existence of an objecti ' individual ' tha t s tays stable. However, t differences are small compared to the differen¢ between the backgrounds of different individuals.

Roles : In organizational settings, the regulari ty

often not of specific individuals, but of declar roles. The role of 'head nurse' or 'at tendi

physician' may be filled by different people different days. A role can be defined as the potent to enter into certain conversations. For examt some conversations for action about pat ients can ini t iated only by the at tending physician, wl~ others can be ini t iated by a nurse. Often, roles '. associated with the potential to make declaratio The obvious cases include Austin 's "I pronounce man and wife," bu t even in more rounds circumstances, roles determine the potential declaring a domain of potential action. For exam]

208

In a conversation for clarification the participants

cope with or anticipate breakdowns concerninginterpretations of the conditions of satisfaction for aCfA. The conditions are always interpreted withrespect to an implicit shared background, but thesharing is partial and needs to be negotiated. As asimple example, the request "Give the patient somediazine" might evoke responses such as "Right now,or with the morning meds?" or "What dosage?".One can never guarantee that everything is totallyprecise. Precision is relative to each party's implicitanticipation that the other party will have asufficiently shared background to fill in the rest ofthe details appropriately.

In a conversation for possibilities, the mood is one of

generating conversations for action. Specificconditions of satisfaction will emerge in the courseof the conversation, and associated conversations foraction will be initiated. Many gatherings that arecalled 'meetings' are in this mood. The meeting is afailure if some action does not come out of thediscussion. Some conversations for possibilities arehighly routinized. For example, 'work rounds' on ahospital ward is a routine conversation forpossibilities, during which the medical team visitseach patient and specific requests ('orders') andcommitments are generated.

In a conversation for orientation, the mood is of

creating a shared background as a basis forinterpreting future conversations. This sharedbackground includes specific knowledge,interpersonal relations, and general attitudes. Themost obvious examples are meetings labelled'orientation,' in which newcomers begin to developthe understanding that is required to function in theorganization. Conversations for orientation areprominent in less formal settings ('shooting the bull'or 'telling stories'). The mood here is not directedtowards action, but it is important to recognize howcritical it is for people to develop shared orientationas the basis for future effective action andappropriate interpretation oflanguage acts.

4. The pragmatics of context

Many current linguistic theories deal in some way withcontext or 'situation' (see, for example, Barwise andPerry, 1983). They focus on 'deictic' expressions, whosemeaning is rooted in the conversation situation (e.g.,n "~th " d "to ") dyou, ere, an morrowan on 'anaphoric'

expressions (such as "that old man" and "the otheone") whose reference depends on complex factors of thprevious conversation and shared setting of speakEand hearer (e.g., what is mutually visible).

But there is a deeper problem. The interpretation,every term and every sentence depends on situation, !

carried in the background of speaker and hearer. F<example, a phrase that would convey a specific requein a face to face conversation might become ambiguOland improper in an on-line system where the eventureader's identity is not known to the writer (especialif there is no opportunity for conversations ~

clarification). Language that has one meaning tonurse on a hospital ward may mean completedifferent things to the patient, or even to that sannurse in a later conversation.

The meaning of a language act exists in tlinterpretations of a speaker and a hearer in a particul.situation with their particular backgrounds. If evesituation and background were incomparably uniquthere would be no way to provide for regular patten:But ofcourse this is not the case. In every setting theare regularities-or 'recurrences'-that canidentified and/or declared.

Individuals: The most obvious regularity is that

personal identity. We can identify individuals aJ

expect their interpretations to remain relativEstable: what I read today, I will understand in t:same way tomorrow (more or less). Of course theare ever-changing moods', and each pers,experiences learning and personal change. ~

cannot assume the existence of an objecti'individual' that stays stable. However, tdifferences are small compared to the differeD(between the backgrounds of different individuals.

Roles: In organizational settings, the regularity

often not of specific individuals, but of declarroles. The role of 'head nurse' or 'attendiphysician' may be filled by different peopledifferent days. A role can be defined as the potentto enter into certain conversations. For examIsome conversations for action about patients caninitiated only by the attending physician, wr.others can be initiated by a nurse. Often, roles:associated with the potential to make declaratioThe obvious cases include Austin's "I pronounce ~

man and wife," but even in more mundlcircumstances, roles determine the potentialdeclaring a domain of potential action. For exam)

208

many kinds of conversation in the hospital can only be ini t iated by doctors.

P r o f e s s i o n s : Although being a nurse or doctor does

not define a narrow role, it reflects regulari ty in the background a person brings to a language act. A crucial par t of professional t raining is learning a j a rgon- - the distinctions and associated terms that provide a basis for inventing and taking relevant actions. 'Profession-oriented languages' (Kaasb¢ll, 1986) are an a t tempt to integrate this specialized language structure into the design of computer-based systems.

Locat ions : The location of an ut terance in time and

place provides a regular s tructure that affects its interpretation, not only in the deictic and anaphoric

effects mentioned above, but in deeper ways as well. 'Resuscitation' means very different things at an emergency site and on an intensive care ward, even when the same people are involved. As with people, locations are often recurrent in terms of functions,

not individuals.

5. Semant i c s

Section 3 described the structure of conversations without regard to their particular conditions of satisfaction. These conditions reflect a 'propositional

content, ' which can be characterized in terms of semant ics- - the relation between linguistic forms and a world of action that they concern. There is no sharp boundary between semantics and pragmatics, or between language use and 'literal meaning. ' But there are different spheres of concern. Traditionally, semantics has been described as a correspondence between forms in a language and some kind of ' t ruth conditions' on the world of which it speaks. As a

simplified example, "The patient has a fever" is correlated with a part icular objective condition (identified as "having a fever") of some individual who is a "patient." Pragmatics deals with how such s ta tements are interpreted in situations (for example, knowing which patient the doctor has in mind) or are

used (for example, as a s ta tement to a nurse in order to evoke certain actions).

Most semantic theories concentrate on how meaning is derived from the way that different words and phrases are combined, and take for granted a collection of basic t e rms- - the nouns, verbs, adjectives and the l i ke - - tha t refer to identifiable objects, properties, relations, and events in the world.

From the perspective of language as action, meanings of words cannot be defined in isolation from a part icular conversational set t ing in which they are used. Calling someone a "patient" is contingent on a

complex set of medical practices and roles, and the property of "fever" depends on the context of language use, not jus t the thermometer reading. Distinctions that are reflected in the choice of words are neither fixed nor individual. They arise through recurrent pat terns of conversations, in which breakdowns of action lead to new distinctions (see Winograd (1985) and Chapter 5 of Winograd and Flores (1986)).

In everyday settings, we are largely unconscious of

language i tself--of how its s tructure emerges and shapes conversation. In an unfamiliar world (for example the hospital with its medical jargon) we notice

that words have 'different meanings, ' and from this we can look at the question of how all meanings arise. From the standpoint of design, several issues need to be addressed:

1) W h a t d o m a i n s o f d i s t i n c t i o n s are t a k e n as

b a c k g r o u n d ?

Everyone in a normal work sett ing shares a natura l language and a lifetime of cultural experience. The

everyday use of language takes this for granted, in applying ordinary natura l language terms and structures, as well as s tandard technical language forms, such as those of mathemat ics and the clock and calendar. Others will be specialized to a technical area such as medicine. The boundary between natural and

formalized domains is not sha rp - -many words are used in both informal and semi-formalized, or stylized ways. "The pat ient is in stable condition" has a technical

interpretat ion distinct from the natural one. In some cases, such distinctions may be set down by formal

rules; in others, learned through practice. As mentioned above, research on profession-

oriented languages a t tempts to identify such domains of meaning and build them into computer-based tools for supporting work. Kaasb¢ll (in preparation) points out problems, such as locally-used distinctions that are not s tandard to the field and not immediately available to system designers. For example, the nurses in his s tudy referred to a lung-function test apparatus as 'the Ohio,' which was its brand name, and they had no more

general term. In some cases such mat ters are of critical importance to conditions of satisfaction. In the medical profession two different kinds of terms are used to describe medications: brand names ('Tylenol') and

209

many kinds of conversation in the hospital can onlybe initiated by doctors.

Professions: Although being a nurse or doctor does

not define a narrow role, it reflects regularity in thebackground a person brings to a language act. Acrucial part of professional training is learning ajargon-the distinctions and associated terms thatprovide a basis for inventing and taking relevantactions. 'Profession-oriented languages' (Kaasblllll,1986) are an attempt to integrate this specializedlanguage structure into the design ofcomputer-based systems.

Locations: The location of an utterance in time and

place provides a regular structure that affects itsinterpretation, not only in the deictic and anaphoriceffects mentioned above, but in deeper ways as well.'Resuscitation' means very different things at anemergency site and on an intensive care ward, evenwhen the same people are involved. As with people,locations are often recurrent in terms of functions,not individuals.

5. Semantics

Section 3 described the structure of conversationswithout regard to their particular conditions ofsatisfaction. These conditions reflect a 'propositionalcontent,' which can be characterized in terms ofsemantics-the relation between linguistic forms and aworld of action that they concern. There is no sharpboundary between semantics and pragmatics, orbetween language use and 'literal meaning.' But thereare different spheres of concern. Traditionally,semantics has been described as a correspondencebetween forms in a language and some kind of 'truthconditions' on the world of which it speaks. As asimplified example, "The patient has a fever" iscorrelated with a particular objective condition(identified as "having a fever") of some individual whois a "patient." Pragmatics deals with how suchstatements are interpreted in situations (for example,knowing which patient the doctor has in mind) or areused (for example, as a statement to a nurse in order toevoke certain actions).

Most semantic theories concentrate on how meaningis derived from the way that different words andphrases are combined, and take for granted a collectionof basic terms-the nouns, verbs, adjectives and thelike-that refer to identifiable objects, properties,relations, and events in the world.

From the perspective of language as action,meanings of words cannot be defined in isolation from aparticular conversational setting in which they areused. Calling someone a "patient" is contingent on acomplex set of medical practices and roles, and theproperty of "fever" depends on the context oflanguageuse, not just the thermometer reading. Distinctionsthat are reflected in the choice of words are neitherfixed nor individual. They arise through recurrentpatterns of conversations, in which breakdowns ofaction lead to new distinctions (see Winograd (1985)and Chapter 5 ofWinograd and Flores (1986)).

In everyday settings, we are largely unconscious oflanguage itself-of how its structure emerges andshapes conversation. In an unfamiliar world (forexample the hospital with its medical jargon) we noticethat words have 'different meanings,' and from this wecan look at the question of how all meanings arise.From the standpoint ofdesign, several issues need to beaddressed:

1) What domains of distinctions are taken as

background?

Everyone in a normal work setting shares a naturallanguage and a lifetime of cultural experience. Theeveryday use of language takes this for granted, inapplying ordinary natural language terms andstructures, as well as standard technical languageforms, such as those of mathematics and the clock andcalendar. Others will be specialized to a technical areasuch as medicine. The boundary between natural andformalized domains is not sharp-many words are usedin both informal and semi-formalized, or stylized ways."The patient is in stable condition" has a technicalinterpretation distinct from the natural one. In somecases, such distinctions may be set down by formalrules; in others, learned through practice.

As mentioned above, research on profession­oriented languages attempts to identify such domainsof meaning and build them into computer-based toolsfor supporting work. Kaasblllll (in preparation) pointsout problems, such as locally-used distinctions that arenot standard to the field and not immediately availableto system designers. For example, the nurses in hisstudy referred to a lung-function test apparatus as 'theOhio,' which was its brand name, and they had no moregeneral term. In some cases such matters are ofcritical importance to conditions of satisfaction. In themedical profession two different kinds ofterms are usedto describe medications: brand names ('Tylenol') and

209

generic names ('Acetiminophen'), which apply to

anything of equivalent chemical composition. A

request made for a brand name may or may not be satisfied by an equivalent generic, depending on a complex interaction of s tandard practices and local

regulations. Suchman (forthcoming) describes the problems that

arise from failing to account for differences in semantic interpretat ions when designing user interfaces. In a 'user-friendly' interface for a copier, language about the machine and the user 's actions appeared in various forms on the screen, using distinctions and words that

made perfect sense to the copier-designers, bu t tha t led to serious breakdowns for users without the same background. In one case, failure to distinguish the

"document cover" from the "bound document aid" led to interpret ing help instructions in a way completely different from what the designers intended.

The point here is tha t the system designer cannot assume that the semant ics- - the mapping from words to distinctions of interest---will ei ther be 'natural ' or follow some existing formal specification of the domain. Meaning is relative to the actions made in language and as a result of language.

2) H o w do n e w d is t inct ions e m e r g e ?

New distinctions are always emerging because of new breakdowns or anticipation of them. This happens

slowly (relative to everyday activity), which makes it at t imes invisible. A frequent reason for the failure of computer systems is tha t they 'lock in' a set of distinctions that was observed by the designers, without leaving room for evolution. Gradual ly people find more and more need to 'work around' the system, leading to complexity and chaos. In such cases, the power to declare new distinctions falls to the system engineers, since they can make changes to the program. They may be the least well qualified to do this in a way that makes sense for the work.

For example, a patient 's diet might be recorded as one of a fixed set of choices, such as 'no salt', 'diabetic', etc., based on the current dietary forms. Imagine that a new form of diet is added, such as l imiting the cholesterol within the existing diets. The new distinction is not simply one more al ternat ive bu t a

potential modification of the pre-existing choices (except for 'no fat' which by definition has no cholesterol). We might add a collection of new diets such as 'no salt low cholesterol', 'diabetic low cholesterol', etc. bu t this is a work-around. If a further

qualification ('low potassium') is added, the system wil

begin to bog down. Wha t is needed is an interpretatioI

of diet as combining a set of separable dimensions instead of as a simple choice. The original systen

designer could not anticipate this need.

3) H o w a re dis t inc t ions indicated?

Much of the tradit ional work on natura l languag semantics adopts the idealization of a relat ive[ s t raightforward 'compositional' mapping from forms t meanings. Pu t simply, each basic term (word) has meaning, and each phrase or sentence has a meanin tha t is made up from the meaning of its parts in s tandard way. Some current research goes furthel

focussing on the effect of context on meaning. I t ] based on s t ructured analyses of contexts (both th linguistic context and the si tuation of the speaker an

hearer) and the relation between those s t ructures an the meanings of utterances. There has been sor~ interest in applying the result ing theoretic~

framework to non-natural languages, such programming languages and human-comput~ interfaces (CSLI, 1984). It is beyond the scope of th

paper to survey the re levant work. It has significal l imitations, as discussed by Winograd (1985).

4) W h e r e a r e m e t a p h o r s used?

Most semantic theories break down when faced wil metaphorical use of language. At first glance, it mig] seem that metaphors are not re levant to syste: designers. Dealing with "My love is a red red ros~ seems less important than "Give him half a Dispril But unconscious metaphor is a central par t of

language use, including language about and withi computer systems (see Lakoff and Johnson, 1980).

Andersen and Madsen (1986) have studied the use metaphors by l ibrary workers using a computer-base system. They note metaphors based on physical spa ("Here's how to leave this database and enter the oth~

one."), communication ("Then it asks whether you wa: to continue"), and in tent ion--which they call 'organism metaphor ' - - (" I t doesn't want to log you in.' These appear in the informal language of people usiJ the system, and are embodied in the system design aJ documentation. Some modern interfaces make caref explicit use of metaphors (e.g., represent ing comput files as objects on a physical desktop) to help structu interactions. In unsystemat ic cases, the results can confusing. For example, the physical metaph suggests the usual rules of physical bodies (e.g., y

210

generic names ('Acetiminophen'), which apply toanything of equivalent chemical composition. Arequest made for a brand name mayor may not besatisfied by an equivalent generic, depending on acomplex interaction of standard practices and localregulations.

Suchman (forthcoming) describes the problems thatarise from failing to account for differences in semanticinterpretations when designing user interfaces. In a'user-friendly' interface for a copier, language about themachine and the user's actions appeared in variousforms on the screen, using distinctions and words thatmade perfect sense to the copier-designers, but that ledto serious breakdowns for users without the samebackground. In one case, failure to distinguish the"document cover" from the "bound document aid" led tointerpreting help instructions in a way completelydifferent from what the designers intended.

The point here is that the system designer cannotassume that the semantics-the mapping from words to

distinctions of interest-will either be 'natural' orfollow some existing formal specification ofthe domain.Meaning is relative to the actions made in languageand as a result oflanguage.

2) How do new distinctions emerge?

New distinctions are always emerging because of newbreakdowns or anticipation of them. This happensslowly (relative to everyday activity), which makes it attimes invisible. A frequent reason for the failure ofcomputer systems is that they 'lock in' a set ofdistinctions that was observed by the designers,without leaving room for evolution. Gradually peoplefind more and more need to 'work around' the system,leading to complexity and chaos. In such cases, thepower to declare new distinctions falls to the systemengineers, since they can make changes to the program.They may be the least well qualified to do this in a waythat makes sense for the work.

For example, a patient's diet might be recorded asone of a fixed set of choices, such as 'no salt', 'diabetic',etc., based on the current dietary forms. Imagine that anew form of diet is added, such as limiting thecholesterol within the existing diets. The newdistinction is not simply one more alternative but apotential modification of the pre-existing choices(except for 'no fat' which by definition has nocholesterol). We might add a collection of new dietssuch as 'no salt low cholesterol', 'diabetic lowcholesterol', etc. but this is a work-around. If a further

qualification ('low potassium') is added, the system wilbegin to bog down. What is needed is an interpretatioIof diet as combining a set of separable dimensionsinstead of as a simple choice. The original systendesigner could not anticipate this need.

3) How are distinctions indicated?

Much of the traditional work on natural languagsemantics adopts the idealization of a relativeIstraightforward 'compositional' mapping from forms tmeanings. Put simply, each basic term (word) hasmeaning, and each phrase or sentence has a meaninthat is made up from the meaning of its parts instandard way. Some current research goes furthelfocussing on the effect of context on meaning. It ibased on structured analyses of contexts (both thlinguistic context and the situation of the speaker anhearer) and the relation between those structures anthe meanings of utterances. There has been sominterest in applying the resulting theoreticlframework to non-natural languages, suchprogramming languages and human-computlinterfaces (CSLI, 1984). It is beyond the scope of thpaper to survey the relevant work. It has significallimitations, as discussed by Winograd (1985).

4) Where are metaphors used?

Most semantic theories break down when faced wi1metaphorical use of language. At first glance, it migJseem that metaphors are not relevant to systedesigners. Dealing with "My love is a red red ros,seems less important than "Give him half a DispriJBut unconscious metaphor is a central part of a

language use, including language about and withicomputer systems (see Lakoffand Johnson, 1980).

Andersen and Madsen (1986) have studied the usemetaphors by library workers using a computer-baslsystem. They note metaphors based on physical spa,("Here's how to leave this database and enter the oth

one."), communication ("Then it asks whether you wato continue"), and intention-which they call 1

'organism metaphor'-("It doesn't want to log you in.'These appear in the informal language of people usilthe system, and are embodied in the system design aldocumentation. Some modern interfaces make carefexplicit use of metaphors (e.g., representing computfiles as objects on a physical desktop) to help structuinteractions. In unsystematic cases, the results canconfusing. For example, the physical metaphsuggests the usual rules of physical bodies (e.g., y,

210

can't be in two places at the same time) and leads to confusion when the actual situation does not obey them.

6. Syntax

Linguistic theories of syntax concentrate on the particular properties of natural language phrases and sentences, and on the rules that generate appropriate combinations. We use 'syntax' in a somewhat more general sense to stand for the regularities in form in any recurrent pattern of linguistic interaction, including those that do not involve natural language directly. What distinguishes a description as 'syntactic' is its focus on the relation of the forms, rather than on the meanings they convey.

In fact, much of standard systems analysis is best characterized as a syntactic analysis, dealing with the 'paper flow' or 'information flow' in terms of syntactic items (relations in a data base, forms to be filed, etc.). In analyses like the wallgraph described below, the emphasis on syntactic structure can obscure the underlying meanings, since it is so strongly rooted in the particular forms that happen to have evolved or been previously designed.

Natural language structure can be divided into two areas: sentential syntax of individual phrases and sentences; and discourse syntax of conversations. The same distinction can be applied to computer-mediated interactions. There are relatively natural smaller units of conversation (individual messages, forms, commands, etc.) whose internal structure can be analyzed as obeying a grammar. These are combined into larger structures (terminal sessions, protocols, etc.) in which issues such as turn-taking (the order in which participants can act) play a role.

Although there are many important issues to be discussed in the syntactic domain (including much of what goes under the name 'human-machine interaction'), they are not the ones addressed in this paper, since they have received substantial attention elsewhere. In a way, syntactic analysis starts from the wrong end. We want to understand the structure of the the work, and have the detailed interaction structure follow from it, not design the forms and let the work adapt around them.

7. An example

The scattered medical illustrations used above were motivated by the case study described by Kaasb¢ll (1986). Researchers in the Florence project of the

Scandinavian SYDPOL (System Development and Profession Oriented Languages) research program have studied a hospital in Norway, analyzing it from what they call a 'systems perspective.' They focused on the tasks associated with giving medications in a ward for children with respiratory problems. In collaboration with the nurses, they created a graphical system description reflecting the paperwork associated with medication. It took the form of a 'wallgraph,' with icons representing the various documents used on the ward,

and arrows representing the flow of information from one to another. Table 1 paraphrases Kaasb¢ll's verbal description of a part of the activities that were

analyzed.

Each nurse has a special responsibility as the team nurse for a small group of patients.

On the day shift, one nurse (called the treating nurse) has the task of giving medicines. Her working day may be characterized roughly by the sequence: attend the report meeting; give medicines; record the medicines given; take care of children in kindergarten

or the dining hall. The report meeting takes place from 7:45 to 8 am.

One nurse informs the other staff of the status,

changes and activities of each patient during the previous day and night. She has heard a vocal report

from the night shift, and she reads the Kardex while

reporting. The Kardex contains diagnosis, planning

and evaluation for each patient, and a form with fixed

main patient information. It is supposed to be up to

date. Other staff take notes on their program sheets. Medicines are prescribed by the doctors on

prescription forms in cures lasting several days or in daily doses. Cures are recorded on the main patient

information form and on the medicine card. Medicines are given in a t reatment room between 8

and 9:30 am. Patients enter after having been

examined by a doctor, carrying wi th them a scrap of paper on which the doctor has wri t ten today's dose

and possible changes in cures. Prescriptions for

medicines during the intervals between the regular

medicine hours are noted down on a premedlist hanging on the wall in the t reatment room.

Some simple lung function tests are also performed

in the t reatment room to monitor the effects of the medicines. Tests to be taken are wri t ten on the prescription form and on a scrap taped to the medicine

211

can't be in two places at the same time) and leads toconfusion when the actual situation does not obey them.

6. Syntax

Linguistic theories of syntax concentrate on theparticular properties of natural language phrases andsentences, and on the rules that generate appropriatecombinations. We use 'syntax' in a somewhat moregeneral sense to stand for the regularities in form inany recurrent pattern of linguistic interaction,including those that do not involve natural languagedirectly. What distinguishes a description as 'syntactic'is its focus on the relation of the forms, rather than onthe meanings they convey.

In fact, much of standard systems analysis is bestcharacterized as a syntactic analysis, dealing with the'paper flow' or 'information flow' in terms of syntacticitems (relations in a data base, forms to be filed, etc.).In analyses like the wallgraph described below, theemphasis on syntactic structure can obscure theunderlying meanings, since it is so strongly rooted inthe particular forms that happen to have evolved orbeen previously designed.

Natural language structure can be divided into twoareas: sentential syntax of individual phrases andsentences; and discourse syntax of conversations. Thesame distinction can be applied to computer-mediatedinteractions. There are relatively natural smaller unitsof conversation (individual messages, forms,commands, etc.) whose internal structure can beanalyzed as obeying a grammar. These are combinedinto larger structures (terminal sessions, protocols, etc.)in which issues such as turn-taking (the order in whichparticipants can act) playa role.

Although there are many important issues to bediscussed in the syntactic domain (including much ofwhat goes under the name 'human-machineinteraction'), they are not the ones addressed in thispaper, since they have received substantial attentionelsewhere. In a way, syntactic analysis starts from thewrong end. We want to understand the structure ofthethe work, and have the detailed interaction structurefollow from it, not design the forms and let the workadapt around them.

7. An example

The scattered medical illustrations used above weremotivated by the case study described by Kaasb¢ll(1986). Researchers in the Florence project of the

Scandinavian SYDPOL (System Development andProfession Oriented Languages) research program havestudied a hospital in Norway, analyzing it from whatthey call a 'systems perspective: They focused on thetasks associated with giving medications in a ward forchildren with respiratory problems. In collaborationwith the nurses, they created a graphical systemdescription reflecting the paperwork associated withmedication. It took the form of a 'wallgraph,' with iconsrepresenting the various documents used on the ward,and arrows representing the flow of information fromone to another. Table 1 paraphrases Kaasb¢ll's verbaldescription of a part of the activities that were

analyzed.

Each nurse has a special responsibility as the team

nurse for a small group of patients.On the day shift, one nurse (called the treating

nurse) has the task of giving medicines. Her workingday may be characterized roughly by the sequence:attend the report meeting; give medicines; record themedicines given; take care of children in kindergarten

or the dining hall.The report meeting takes place from 7:45 to 8 am.

One nurse informs the other staff of the status,

changes and activities of each patient during the

previous day and night. She has heard a vocal report

from the night shift, and she reads the Kardex while

reporting. The Kardex contains diagnosis, planning

and evaluation for each patient, and a form with fixed

main patient information. It is supposed to be up to

date. Other staff take notes on their program sheets.

Medicines are prescribed by the doctors on

prescription forms in cures lasting several days or in

daily doses. Cures are recorded on the main patient

information form and on the medicine card.

Medicines are given in a treatment room between 8

and 9:30 am. Patients enter after having been

examined by a doctor, carrying with them a scrap of

paper on which the doctor has written today's dose

and possible changes in cures. Prescriptions for

medicines during the intervals between the regular

medicine hours are noted down on a premedlist

hanging on the wall in the treatment room.

Some simple lung function tests are also performed

in the treatment room to monitor the effects of the

medicines. Tests to be taken are written on the

prescri ption form and on a scrap taped to the medicine

211

card or on the program sheet if there are changes. The

test results are recorded on special forms.

After having given medicines, the treating nurse

brings her papers to the ward office. Together with

each of the team nurses, one at a time, she examines

the papers. All medicines given are now registered on

the curve sheet. Changes in cures are recorded in the

Kardex, and on the medicine card and eventually on

the premedlist. In addit ion, all sheets are compared for

the sake of control.

The patients are processed one by one. The team

nurse reads from her papers which medicines are to be

given. The treating nurse answers by stating which are

actually given, and the state of the patient. One day,

when the load on the nurses was relatively low, this

activity lasted from 9:30 am to 13:25 pm. During these

4 hours, at most 30 minutes were 'effective paper

work.' The rest was delays, either because some of the

papers were used by others, or because one of the

nurses was engaged in handling interruptions. Only a

minor part of the 30 minutes was used for updating

and comparing. The rest of the time was spent on

small conversations, init iated by findings in the

information they were handling. These included:

reporting to each other about the patients' state and

activities

deciding what were facts when inconsistencies were

found

deciding changes in some medicines after small

negotiations

reminding the treating nurse of a test that had been

forgotten

investigating why a medicine was not delivered from the chemist's

f inding out why a patient had to take a specific test.

Table 1: Act iv i t i e s on a hosp i ta l w a r d ( a d a p t e d

f rom Kaasbol l , 1986, p. 3)

Although no particular computer applications were developed, it is easy to imagine designing a 'medication information system' with on-line terminals (or workstations) at various sites, including the ward, the examining rooms, and the pharmacy. Many of the activities would be replaced with corresponding actions on a terminal. Doctors would enter prescriptions and later could enter modifications on an examining-room

terminal instead of sending them on scraps of pap, Pharmacists would fill prescriptions presented on thq terminals, and so forth. Records that needed to be places where direct computer access was infeasii could be printed out and posted (for example t 'premedlist' of medications before activities, which posted on the wall). The 'information flow' could redesigned, so that redundancies would be eliminai and there would no need for manual copying or posti of information that was already available.

It is far beyond the scope of this paper to develo! comprehensive design for such a system. We interested in identifying the theoretical consideratic that would guide such a design, examining the c~ study from a language/action perspective. 'I following section reiterates the framework of t previous sections, examining the hospital work language action, with its particular problems pragmatics, semantics, and syntax. Our m~ emphasis is on the pragmatics of language as actil and its dependence on context.

First, two points are worth reiterating:

1) We are designing work, not just tools. If compu' systems are introduced, they will redefine t

possibilities for action of all the workers, not jt those who use them directly. They will change t: patterns of conversation and potentials for getti~ things done.

2) The system designer is not faced with the choice whether or not to adopt a perspective. It may conscious or unconscious, systematic or haphaza~ but having a perspective is a necessary condition t being able to observe, describe, and design at all . . Nygaard and Sorgaard (1985) point out, standa system development methods all adopt 'information system' perspective, dealing primar: with the syntax of specific conversations, and wi the maintenance of and access to records of the conversations.

The Florence researchers developed an informati flow diagram (the wallgraph) relating the differc forms and activities, and felt that it gave an overview the work of the ward. The nurses, however, respon~ to the graphs in a quite different way. Their reacti was that the graphs did not show 'real nursir Kaasb¢ll concludes (p. 6) that "Since wall graphs sh paper flow, and 'real nursing' involves direct pati~ interaction, this seems evident."

But from a language perspective, the conclusion different. The problem is not in the wallgraph's faih

212

card or on the program sheet if there are changes. The

test results are recorded on special forms.

After having given medicines, the treating nurse

brings her papers to the ward office. Together with

each of the team nurses, one at a time, she examines

the papers. All medicines given are now registered on

the curve sheet. Changes in cures are recorded in the

Kardex, and on the medicine card and eventually on

the premedlist. In addition, all sheets are compared for

the sake of control.

The patients are processed one by one. The team

nurse reads from her papers which medicines are to be

given. The treating nurse answers by stating which are

actually given, and the state of the patient. One day,

when the load on the nurses was relatively low, this

activity lasted from 9:30 am to 13:25 pm. During these

4 hours, at most 30 minutes were 'effective paper

work.' The rest was delays, either because some of the

papers were used by others, or because one of the

nurses was engaged in handling interruptions. Only a

minor part of the 30 minutes was used for updating

and comparing. The rest of the time was spent on

small conversations, initiated by findings in the

information they were handling. These included:

reporting to each other about the patients' state and

activities

deciding what were facts when inconsistencies were

found

deciding changes in some medicines after small

negotiations

reminding the treating nurse of a test that had been

forgotten

investigating why a medicine was not delivered from

the chemist's

finding out why a patient had to take a specific test.

Table 1: Activities on a hospital ward (adapted

from Kaasb011, 1986, p. 3)

Although no particular computer applications weredeveloped, it is easy to imagine designing a 'medicationinformation system' with on-line terminals (orworkstations) at various sites, including the ward, theexamining rooms, and the pharmacy. Many of theactivities would be replaced with corresponding actionson a terminal. Doctors would enter prescriptions andlater could enter modifications on an examining-room

212

terminal instead of sending them on scraps of pap'Pharmacists would fill prescriptions presented on th,terminals, and so forth. Records that needed to beplaces where direct computer access was infeasilcould be printed out and posted (for example t'premedlist' of medications before activities, whichposted on the wall). The 'information flow' couldredesigned, so that redundancies would be eliminatand there would no need for manual copying or postiof information that was already available.

It is far beyond the scope of this paper to develo)comprehensive design for such a system. We linterested in identifying the theoretical consideraticthat would guide such a design, examining the Clstudy from a language/action perspective. Tfollowing section reiterates the framework of tprevious sections, examining the hospital worklanguage action, with its particular problemspragmatics, semantics, and syntax. Our mlemphasis is on the pragmatics of language as acti<and its dependence on context.

First, two points are worth reiterating:

1) We are designing work, not just tools. If compu'systems are introduced, they will redefine t

possibilities for action of all the workers, not jt;those who use them directly. They will change l'patterns of conversation and potentials for gettilthings done.

2) The system designer is not faced with the choicewhether or not to adopt a perspective. It mayconscious or unconscious, systematic or haphazalbut having a perspective is a necessary condition j

being able to observe, describe, and design at all.Nygaard and Sorgaard (1985) point out, standasystem development methods all adopt'information system' perspective, dealing primar:willi the syntax of specific conversations, and withe maintenance of and access to records of theconversations.

The Florence researchers developed an informatiflow diagram (the wallgraph) relating the diffenforms and activities, and felt that it gave an overviewthe work of the ward. The nurses, however, responcto the graphs in a quite different way. Their reactiwas that the graphs did not show 'real nursirKaasb~ll concludes (p. 6) that "Since wall graphs shpaper flow, and 'real nursing' involves direct patiEinteraction, this seems evident."

But from a language perspective, the conclusiondifferent. The problem is not in the wallgraph's faih

to show real patients, with their complex emotional and physical particularities. Of course, no systematic

account can capture this. But the wallgraph does not capture the conversation structure either. The tangle of boxes and lines deals with the syntax of what gets wri t ten when and where, but reveals little about the structure of completion of actions that drive the

conversations. Further , it captures only the routine parts of the structure, omitt ing all of the 'small conversations' that seemed to fill so much time, and demand the application of nursing knowledge and

skills. To some extent, the omission of other conversations

results from the way the wallgraph was created. Kaasb011 notes "There are discrepancies between the

narrow intention expressed and the broad scope of the actual information processing going on. The nurses did not mention that other tasks might also be done

depending on the situation. Hence, they are either not conscious of it, or they feel it is unimportant and will decrease the importance fur ther by not mentioning it."

More significantly, there was no language in which to adequately describe all of the activities. They didn't fit into the language of forms and Kardexes, so the

information perspective did not allow them to be seen. The same is true of the apparent picture of substant ial wasted effort (which could perhaps be avoided by automation). The apparent waste covers a s tructure of conversation that is not revealed from the information

perspective. From a language/action perspective, the designer is

not interested in the structure of information or its flow, but in the design of the conversations that make up the work. Work does not become efficient jus t because it is on-line. In fact, the requirements of a system for data entry and access may have quite the opposite effect,

increasing the amount of record keeping without corresponding advantages. Questions of effectiveness and efficiency appear in the domain of action

(conditions of satisfaction, and their completion) and in domain of creating and using distinctions.

7A. Pragmatics: conversations

C o n v e r s a t i o n s for action: There are many different

conversations for action, with a variety of visible forms. Some are highly routinized, such as the pr imary CfA dealing with the administrat ion of medications. Requests are made by doctors (either as standing cures or on the patient-carried paper scraps), to the treat ing nurse. Assertion of completion is represented on the

curve sheet, and the declaration of satisfaction is implicit in the doctor's lack of questioning of the records on his or her next visit. As a precondition for satisfying

these requests, the nurse mus t receive the medicine, and there are CfAs (with the pharmacy) to get the medications, using prescription forms to make requests. In general, conditions of satisfaction are determined in a rigid way by the codes and blanks that are made available on the forms, perhaps with extra notations in na tura l language. Acceptance of an offer or request is assumed whenever it is not explicitly rejected. Completion is reported on a standard form, which, like

all of the other forms, is associated with standards for interpretation, which are learned as part of the

re levant professional training. In addition to these routinized CfAs, there are

unscheduled verbal conversations. For example, a request may be made by a doctor to a nurse at the

bedside, with immediate explicit acceptance, rejection, or counter-offer. Completion may be reported later via a note in the patient 's chart. In these, as in all

conversations, success depends on sufficiently similar interpretat ions by the different parties, and on a commitment to carry the conversation forward to

completion. Rigid forms and routines are an a t tempt to assure this in cases where personal acquaintance and contact are not sufficient. In the hospital setting, completion of conversations can be a life-or-death matter , and there is a highly regularized structure of checks and crosschecks to ensure it for the central conversations. The regularization is both in the form of interactions (the various checklists) and the strict temporal routine of what mus t be done at what time. The fact tha t a part icular action is done at a part icular time can be taken for granted by all on the basis of the daily schedule of activities. This could be augmented

(and part ial ly replaced) by the use of a coordination system, in which the monitoring of completion (and coaching towards completing conversations) is incorporated in a communication medium. It is not jus t a ma t te r of having the conversation on l ine--computer screens can be ignored jus t as easily as pieces of paper. The gain comes from being able to have confidence that conversations will be completed effectively, without being forced into unnecessary rigidity of routine. This is a clear example of potential for redesigning the work.

There is also the potential to replace routine CfAs with declarations of recurrent responses. For example, ra ther than responding to each drug request, a pharmacis t might establish an automated 'prescription

213

to show real patients, with their complex emotional andphysical particularities. Of course, no systematicaccount can capture this. But the wallgraph does notcapture the conversation structure either. The tangleof boxes and lines deals with the syntax of what getswritten when and where, but reveals little about thestructure of completion of actions that drive theconversations. Further, it captures only the routineparts of the structure, omitting all of the 'smallconversations' that seemed to fill so much time, anddemand the application of nursing knowledge and

skills.To some extent, the omission of other conversations

results from the way the wallgraph was created.Kaasb\1Sll notes "There are discrepancies between thenarrow intention expressed and the broad scope of theactual information processing going on. The nurses didnot mention that other tasks might also be donedepending on the situation. Hence, they are either notconscious of it, or they feel it is unimportant and willdecrease the importance further by not mentioning it."More significantly, there was no language in which toadequately describe all of the activities. They didn't fitinto the language of forms and Kardexes, so theinformation perspective did not allow them to be seen.The same is true of the apparent picture of substantialwasted effort (which could perhaps be avoided byautomation). The apparent waste covers a structure ofconversation that is not revealed from the informationperspective.

From a language/action perspective, the designer isnot interested in the structure of information or its flow,but in the design of the conversations that make up thework. Work does not become efficient just because it ison-line. In fact, the requirements of a system for dataentry and access may have quite the opposite effect,increasing the amount of record keeping withoutcorresponding advantages. Questions of effectivenessand efficiency appear in the domain of action(conditions of satisfaction, and their completion) and indomain ofcreating and using distinctions.

7A. Pragmatics: conversations

Conversations for action: There are many different

conversations for action, with a variety of visible forms.Some are highly routinized, such as the primary CfAdealing with the administration of medications.Requests are made by doctors (either as standing curesor on the patient-carried paper scraps), to the treatingnurse. Assertion of completion is represented on the

curve sheet, and the declaration of satisfaction isimplicit in the doctor's lack ofquestioning of the recordson his or her next visit. As a precondition for satisfyingthese requests, the nurse must receive the medicine,and there are CfAs (with the pharmacy) to get themedications, using prescription forms to make requests.In general, conditions of satisfaction are determined ina rigid way by the codes and blanks that are madeavailable on the forms, perhaps with extra notations innatural language. Acceptance of an offer or request isassumed whenever it is not explicitly rejected.Completion is reported on a standard form, which, likeall of the other forms, is associated with standards forinterpretation, which are learned as part of therelevant professional training.

In addition to these routinized CfAs, there areunscheduled verbal conversations. For example, arequest may be made by a doctor to a nurse at thebedside, with immediate explicit acceptance, rejection,or counter-offer. Completion may be reported later viaa note in the patient's chart. In these, as in allconversations, success depends on sufficiently similarinterpretations by the different parties, and on acommitment to carry the conversation forward tocompletion. Rigid forms and routines are an attempt toassure this in cases where personal acquaintance andcontact are not sufficient. In the hospital setting,completion of conversations can be a life-or-deathmatter, and there is a highly regularized structure ofchecks and crosschecks to ensure it for the centralconversations. The regularization is both in the form ofinteractions (the various checklists) and the stricttemporal routine of what must be done at what time.The fact that a particular action is done at a particulartime can be taken for granted by all on the basis of thedaily schedule of activities. This could be augmented(and partially replaced) by the use of a coordinationsystem, in which the monitoring of completion (andcoaching towards completing conversations) isincorporated in a communication medium. It is not justa matter of having the conversation on line-computerscreens. can be ignored just as easily as pieces of paper.The gam comes from being able to have confidence thatconversations will be completed effectively, withoutbeing forced into unnecessary rigidity of routine. Thisis a clear example of potential for redesigning the work.

There is also the potential to replace routine CfAswith declarations of recurrent responses. For example,rather than responding to each drug request apharmacist might establish an automated 'prescrip~ion

213

filling system,' which takes the data from the request and activates a mechanized dispensary. This is the general application of 'computerization,' in which computers take over those functions for which precise repetitive rules can be established. From a

language/action perspective, we need to look at what the conversations are, and what implicit commitments lie in their automation. Also, there needs to be the potential for the other kinds of conversations below. If the medication request does not match the standard form for designating medications, then who is involved in the conversation for clarification, and how? In conventional system design, there will always be answers to such questions (especially after experience in use has pointed out the potential breakdowns). Through a conversational analysis we can anticipate and design for them.

Conversa t ions for clarification: These are much less

regular (as we would expect), and appear mostly as 'small conversations' within the larger structure, often verbally. The crosschecking of the various forms "for control" also functions to trigger these conversations when the different forms are not directly contradictory, but are open to conflicting interpretations. In designing for conversations for clarification, it is important to recognize their relative lack of recurrence. Recurrent differences of interpretation will lead to the declaration of new distinctions or new forms for making requests and commitments that are clear. One goal in the development of profession-oriented languages is to facilitate and regularize this process. But there will always be irregular, unexpected cases, and computer-based systems often make it difficult or impossible to deal with them by providing only rigid forms. Nygaard (1986) has emphasized the need for systems to include the possibility of 'cross-contact'--means by which workers can enter into conversations with other workers across the established lines of conversation and previously declared domains of distinctions. Something as simple as a generalized electronic mail facility can be the basis for this, but an analysis of the conversation structure can help to integrate these additional conversations in more convenient ways.

Conversa t ions for possibilities: Looking carefully at

the various forms and activities, much of what seems to be useless copying or verification of redundant information is really a routine way of creating an opportunity for conversations for possibilities. For example, in the review of medications:

"Only a minor part of the 30 minutes was used fo

updating and comparing. The rest of the time wa spent on small conversations, initiated by findings b

the information they were handling. Some example

of what the nurses did: Reporting to each other abou

the patients' state and activities; deciding what wer

facts when inconsistencies were found; decidin~

changes in some medicines after small negotiations reminding the treating nurse of a test that had beei

forgotten; investigating why a medicine was nc

delivered from the chemist's; f inding out why d patient had to take a specific test."

Most of these involve coping with or anticipating breakdown in some CfA.

The opening for conversational opportunities need to be considered directly in system design. If t} medication review were replaced by an automate process, that opportunity would be lost. It could 1 reinstituted in other ways, but only if it we: recognized in the analysis, and the overall organizatff allowed for appropriate contact. Similarly, if the cur, sheet "is replaced with a table of drugs, dosages ar times, it is no longer possible for the nurses to use it as vehicle of communication (in the notes) for the "sme conversations" dealing with specific breakdowns ai subsidiary CfAs. It might well be possible to replace with a better vehicle for these conversations, once the importance is recognized.

Conversa t ions for or ientat ion: When asked

comment on the systems analysis, the nurses felt that did not capture the aspect of their activities that dea with the 'total picture' rather than with the specifics particular medications and tests. The existir structure provides explicit routines to allow for th kind of conversation, such as the morning report, whe: "One nurse is informing the other staff of the statu changes and performed activities of each patient duri~ the last day and night." In addition, the inform exchanges among the nurses during the day inclu, both explicit CfAs and more general orientil discussion about the 'total picture.' One of tl important forms these conversations take is the tellil of'stories,' whether in the direct line of work, or arout the coffeepot. Orr (1986) describes the importance relating 'war stories' as part of the training f servicing copying machines. One has only to spend short time in the company of medical workers to reali: how prevalent this activity is. It is interesting to no that various computer %boards' play this role with the community of computer researchers. They oft~

214

filling system,' which takes the data from the requestand activates a mechanized dispensary. This is thegeneral application of 'computerization,' in whichcomputers take over those functions for which preciserepetitive rules can be established. From alanguage/action perspective, we need to look at whatthe conversations are, and what implicit commitmentslie in their automation. Also, there needs to be thepotential for the other kinds of conversations below. Ifthe medication request does not match the standardform for designating medications, then who is involvedin the conversation for clarification, and how? Inconventional system design, there will always beanswers to such questions (especially after experiencein use has pointed out the potential breakdowns).Through a conversational analysis we can anticipateand design for them.

Conversations for clarification: These are much less

regular (as we would expect), and appear mostly as'small conversations' within the larger structure, oftenverbally. The crosschecking of the various forms "forcontrol" also functions to trigger these conversationswhen the different forms are not directly contradictory,but are open to conflicting interpretations. In designingfor conversations for clarification, it is important torecognize their relative lack of recurrence. Recurrentdifferences of interpretation will lead to the declarationof new distinctions or new forms for making requestsand commitments that are clear. One goal in thedevelopment of profession-oriented languages is tofacilitate and regularize this process. But there willalways be irregular, unexpected cases, andcomputer-based systems often make it difficult orimpossible to deal with them by providing only rigidforms. Nygaard (1986) has emphasized the need forsystems to include the possibility of'cross-contact'-means by which workers can enter intoconversations with other workers across the establishedlines of conversation and previously declared domainsof distinctions. Something as simple as a generalizedelectronic mail facility can be the basis for this, but ananalysis of the conversation structure can help tointegrate these additional conversations in moreconvenient ways.

Conversations for possibilities: Looking carefully at

the various forms and activities, much of what seems tobe useless copying or verification of redundantinformation is really a routine way of creating anopportunity for conversations for possibilities. Forexample, in the review of medications:

214

"Only a minor part of the 30 minutes was used foupdating and comparing. The rest of the time waspent on small conversations, initiated by findings ilthe information they were handling. Some exampleof what the nurses did: Reporting to each other abouthe patients' state and activities; deciding what wer,facts when inconsistencies were found; decidin,changes in some medicines after small negotiationsreminding the treating nurse of a test that had beelforgotten; investigating why a medicine was ncdelivered from the chemist's; finding out why I

patient had to take a specific test."

Most of these involve coping with or anticipatingbreakdown in some CfA.

The opening for conversational opportunities nee I

to be considered directly in system design. If tImedication review were replaced by an automateprocess, that opportunity would be lost. It could 1reinstituted in other ways, but only if it werecognized in the analysis, and the overall organizati(allowed for appropriate contact. Similarly, if the cuI"sheet 'is replaced with a table of drugs, dosages aItimes, it is no longer possible for the nurses to use it asvehicle of communication (in the notes) for the "smaconversations" dealing with specific breakdowns arsubsidiary CfAs. It might well be possible to replacewith a better vehicle for these conversations, once theimportance is recognized.

Conversations for orientation: When asked

comment on the systems analysis, the nurses felt thatdid not capture the aspect of their activities that deawith the 'total picture' rather than with the specificsparticular medications and tests. The existiIstructure provides explicit routines to allow for thkind ofconversation, such as the morning report, whe"One nurse is informing the other staff of the statuchanges and performed activities of each patient duriIthe last day and night." In addition, the informexchanges among the nurses during the day inclueboth explicit CfAs and more general orientildiscussion about the 'total picture.' One of tlimportant forms these conversations take is the tellilof 'stories,' whether in the direct line of work, or aroUithe coffeepot. Orr (1986) describes the importancerelating 'war stories' as part of the training fservicing copying machines. One has only to spendshort time in the company of medical workers to realihow prevalent this activity is. It is interesting to nothat various computer 'bboards' play this role withthe community of computer researchers. They oftl

contain extended stories, commentaries and other forms of conversation that serve for mutual orientation. This would not be directly applicable to the hospital setting, where the workers do not spend long hours in front of a computer terminal, but it illustrates a new potential for design.

The n e t w o r k of conversat ions: Finally, the

conversational analysis includes not only the immediately visible conversations, but also the larger network of conversations in which they are situated. The design can have tremendous impact on these conversations, which can lead to serious difficulties and even failure of the system (see Scacchi and Kling, 1982).

One obvious example in the hospital is the legal conversation about the quality of care. All written records are potential evidence in a malpractice suit, and the people who create and manipulate them are aware of this possibility. In addition to the legal conversation, there are ongoing conversations about the hiring, evaluation, and dismissal of employees. Kaasb011 notes (1986, p. 11): "If the nurses make mistakes, they may be sued by the patients, and they may be punished by the hospital administration. This gives an incentive for not recording mistakes in the Kardex."

Certain conversations will inevitably go on outside of the written (or electronically stored) system. Explicit records will never correspond to an objective reality, but are the result of declarations by individuals, with their own interpretations and purposes. No computer system can change these fundamental facts about how humans function in organizations, but an explicit understanding of the larger network of conversations can help to recognize the roles that language acts play in a variety of conversations, and to match expectations to those roles.

7B. Pragmatics: situation

Roles: Many of the conversations go on day in and day

out with participants changing roles. For example, the role of ' treating nurse' is defined by the central role in the conversations for action about giving medicines. The fact that after the treatments this nurse "takes care of children in kindergarten or the dining hall" is really the combination in one formally-defined role of two separate conversationally-defined roles. One goal of a language-action analysis is to identify the conversationally relevant roles, which can open the possibility of reorganizing or reassigning them as the

patterns of work change. Some roles may disappear, as the conversational potentials of other participants are expanded. As a simple example, the role of 'clerk' (providing records from the file) may be meaningless when the retrieval of records becomes available directly to the nurses. Other new roles may emerge, for example an 'interface consultant' who helps design computerized forms.

Locations: Different workers need to be aware of the

details and status of the conversations at different times and different locations. Nurses have the Kardex at the nursing station; change requests are on scraps carried by the patient; requests for medication between regular hours are posted on the premedlist on the wall; and so on. Certain people (roles) are expected to update these various representations on a regular basis. For example, the Kardex can be assumed to accurately reflect what was given except for the past few hours, while the main patient information form may lag behind. In designing the conversations, virtual locations will be created, independently of the physical ones (Andersen and Madsen's "Now we are entering the catalog data base..."). They will work better if they are developed within a coherent domain of regularity.

Professions: From Kaasb¢ll's description, we cannot

analyze this case beyond the obvious distinctions among doctors, nurses, pharmacists, etc. A more careful analysis would make finer distinctions (different specialties, levels of training, etc.) that play a crucial role in determining who can enter into what conversations (e.g., the different possibilities open to medical students, residents, and senior physicians).

7C. Semantics

D o m a i n s of b reakdown: To some extent, the domains

of possibilities are determined by a straightforward model of relevant activity. Patients need to be given various drugs because they prevent breakdowns of their physical condition. They need to be given various tests to anticipate and/or prevent breakdowns due to drug effects. The secondary conversations (e.g., with the pharmacy) allow people to satisfy the conditions of satisfaction in this primary domain. However, the existing system leaves openings for individuals (or groups) to declare other relevant domains, in particular in the form of'notes' attached to the various forms.

Along with the specific conversations and forms, the system designer participates in designing the professional language of the workplace. An

215

contain extended stories, commentaries and otherforms ofconversation that serve for mutual orientation.This would not be directly applicable to the hospitalsetting, where the workers do not spend long hours infront of a computer terminal, but it illustrates a newpotential for design.

The network of conversations: Finally, the

conversational analysis includes not only theimmediately visible conversations, but also the largernetwork of conversations in which they are situated.The design can have tremendous impact on theseconversations, which can lead to serious difficulties andeven failure of the system (see Scacchi and Kling,1982).

One obvious example in the hospital is the legalconversation about the quality of care. All writtenrecords are potential evidence in a malpractice suit, andthe people who create and manipulate them are awareofthis possibility. In addition to the legal conversation,there are ongoing conversations about the hiring,evaluation, and dismissal of employees. Kaasbliill notes(1986, p. 11): "Ifthe nurses make mistakes, they may besued by the patients, and they may be punished by thehospital administration. This gives an incentive for notrecording mistakes in the Kardex."

Certain conversations will inevitably go on outsideofthe written (or electronically stored) system. Explicitrecords will never correspond to an objective reality,but are the result of declarations by individuals, withtheir own interpretations and purposes. No computersystem can change these fundamental facts about howhumans function in organizations, but an explicitunderstanding of the larger network of conversationscan help to recognize the roles that language acts playin a variety of conversations, and to matchexpectations to those roles.

7B. Pragmatics: situation

Roles: Many of the conversations go on day in and day

out with participants changing roles. For example, therole of 'treating nurse' is defined by the central role inthe conversations for action about giving medicines.The fact that after the treatments this nurse "takescare of children in kindergarten or the dining hall" isreally the combination in one formally-defined role oftwo separate conversationally-defined roles. One goalof a language-action analysis is to identify theconversationally relevant roles, which can open thepossibility of reorganizing or reassigning them as the

215

patterns of work change. Some roles may disappear, asthe conversational potentials of other participants areexpanded. As a simple example, the role of 'clerk'(providing records from the file) may be meaninglesswhen the retrieval of records becomes available directlyto the nurses. Other new roles may emerge, forexample an 'interface consultant' who helps designcomputerized forms.

Locations: Different workers need to be aware of the

details and status of the conversations at differenttimes and different locations. Nurses have the Kardexat the nursing station; change requests are on scrapscarried by the patient; requests for medication betweenregular hours are posted on the premedlist on the wall;and so on. Certain people (roles) are expected to updatethese various representations on a regular basis. Forexample, the Kardex can be assumed to accuratelyreflect what was given except for the past few hours,while the main patient information form may lagbehind. In designing the conversations, virtuallocations will be created, independently of the physicalones (Andersen and Madsen's "Now we are entering thecatalog data base..."). They will work better if they aredeveloped within a coherent domain of regularity.

Professions: From Kaasbliill's description, we cannot

analyze this case beyond the obvious distinctionsamong doctors, nurses, pharmacists, etc. A morecareful analysis would make finer distinctions(different specialties, levels of training, etc.) that playacrucial role in determining who can enter into whatconversations (e.g., the different possibilities open tomedical students, residents, and senior physicians).

7C. Semantics

Domains of breakdown: To some extent, the domains

of possibilities are determined by a straightforwardmodel of relevant activity. Patients need to be givenvarious drugs because they prevent breakdowns oftheirphysical condition. They need to be given various teststo anticipate and/or prevent breakdowns due to drugeffects. The secondary conversations (e.g., with thepharmacy) allow people to satisfy the conditions ofsatisfaction in this primary domain. However, theexisting system leaves openings for individuals (orgroups) to declare other relevant domains, in particularin the form of,notes' attached to the various forms.

Along with the specific conversations and forms, thesystem designer participates in designing theprofessional language of the workplace. An

intervention in the form of a new computer system will

alter the language, both in interactions with the system, and in the work around it. As a simple example, Andersen and Madsen (1986) point out the

change in usage of the word 'document' when an indexing system was designed for a document collection, then extended to the whole library. All of the indexed i tems (including books, magazines, etc.) came to be called 'documents, ' contrary to prior usage. In new uses of language there is both a danger of

creating confusion and an opportunity to shape the conversations and the work itself. This is especially important in a sett ing like the hospital, where there are

conflicting languages already in use (the language of

doctors, nurses, ordinary language, etc.).

D e c l a r a t i o n o f d i s t inc t ions : Kaasb¢ll (1986, p. 10)

gives an example of a technical term whose meaning was subject to a rgument and the imposition of authority:

"During a doctor's visit, the doctor and two nurses

started a conversation of what T I ' means. P-values

are measures of obstructions in the lungs.

Doctor: It is the air in the lungs that counts, not

the sounds.

Nurse1: It is obvious that Peter (the chief

physician) has a different opinion of P1 than you.

Nurse2: One has to remember that there are

individual discrepancies between the children,

such that P1 does not mean the same for one child

as for another.

This discussion can be interpreted as a negotiation

over the semantics of'P1', and thus as a development

of the language at the ward. It can also be seen as

part of the power struggle between the two professions

involved."

The larger pragmatic analysis of conversations and

roles includes the conversations in which meaning is negotiated, and their reflection in a computer system and in the t raining of people who use it. In looking at the problem of system maintenance, we can design conversations by which new distinctions (perhaps emerging from the informal cross-contact described above) can find their way into the formal par t of the system. By creating this meta-conversation, we can cope with some of the breakdowns that come from

system rigidity. One general principle pointed out by Nygaard

(1986) is the importance of having distinctions that are

open to new interpretat ion by the workers. In practic~ terms, this may be as simple as having a 'Notes' field i some data record, tha t allows the worker to enter (an retrieve) ordinary natura l language text, as opposed l 'fixed fields,' in which the distinctions are fixed b system convention.

I n d i c a t i o n of d i s t inc t ions : As mentioned above, th

interpretat ion of a part icular item in a record : sensitive to situation. Consider the meaning of fillin

in a 'medicine card' listing a patient 's medication From a s tandard semantic view, each blank would t filled with a term tha t denoted a part icular medicatiol

and the card as a whole (analogous to a sentence) wou enumera te the medications to be given to the patien This is mostly the case, bu t according to Kaasb¢ll (J preparation, p. 4):

"The sheets were filled in properly during the

minutes, except for a couple of observed 'missir

medicines' on the medicine cards. When asked abo

the 'mistakes' the treating nurse replied: 'Oh, but

know he (the patient) is going to have the medicin

even if it is not written here. It is erased only becau

he has been under intensive care for some days. '"

The issue here is not the exact form of the cards. It cou

jus t as well have been a computer system with da base entries for medications. The si tuat ion-dependen of meaning is in the people who 'enter ' and 'access' t] data. Accuracy is not jus t a mat te r of having t] computer keep its records straight. In designing a~ install ing a system, it is critical to unders tand t] different potentials for interpretat ion and either co with them or modify them through training.

7D. Syntax

Discou r se : This is the level tha t was visible in t]

wallgraph. There is a specific collection of forms, ea with its own syntactic s t ructure (various fields, blant boxes for checking, etc). These are organized in larger s t ructured collections (the Kardex, the patie record files, etc.). An analysis of this s tructure w reflect all of the part iculari t ies of how they are ill: out, and ends up (as Kaasb¢ll r epor t s )qu i t e comp and confusing to the nurses themselves. Of course, a system we build will have such a syntax, bu t if we vi the i tems as reflecting part icular steps conversations, there is order in the structure.

S e n t e n t i a h The nurses use a wide var ie ty of synta(

forms, ranging from highly specialized codes specifying drug dosages C50mg acetiminophen P

216

intervention in the form of a new computer system willalter the language, both in interactions with thesystem, and in the work around it. As a simpleexample, Andersen and Madsen (1986) point out thechange in usage of the word 'document' when anindexing system was designed for a documentcollection, then extended to the whole library. All ofthe indexed items (including books, magazines, etc.)came to be called 'documents,' contrary to prior usage.In new uses of language there is both a danger ofcreating confusion and an opportunity to shape theconversations and the work itself. This is especiallyimportant in a setting like the hospital, where there areconflicting languages already in use (the language ofdoctors, nurses, ordinary language, etc.).

Declaration of distinctions: Kaasb~ll (1986, p. 10)

gives an example of a technical term whose meaningwas subject to argument and the imposition ofauthority:

~During a doctor's visit, the doctor and two nursesstarted a conversation of what 'PI' means. P-valuesare measures ofobstructions in the lungs.

Doctor: It is the air in the lungs that counts, notthe sounds.

Nurse1: It is obvious that Peter (the chiefphysician) has a different opinion ofPI than you.

Nurse2: One has to remember that there areindividual discrepancies between the children,such that PI does not mean the same for one childas for another.

This discussion can be interpreted as a negotiationover the semantics of 'PI', and thus as a developmentof the language at the ward. It can also be seen aspart of the power struggle between the two professionsinvolved."

The larger pragmatic analysis of conversations androles includes the conversations in which meaning isnegotiated, and their reflection in a computer systemand in the training of people who use it. In looking atthe problem of system maintenance, we can designconversations by which new distinctions (perhapsemerging from the informal cross-contact describedabove) can find their way into the formal part of thesystem. By creating this meta-conversation, we cancope with some of the breakdowns that come fromsystem rigidity.

One general principle pointed out by Nygaard(1986) is the importance of having distinctions that are

open to new interpretation by the workers. In practic!terms, this may be as simple as having a 'Notes' field isome data record, that allows the worker to enter (anretrieve) ordinary natural language text, as opposed 1

'fixed fields,' in which the distinctions are fixed bsystem convention.

Indication of distinctions: As mentioned above, tb

interpretation of a particular item in a recordsensitive to situation. Consider the meaning of fillinin a 'medicine card' listing a patient's medicationFrom a standard semantic view, each blank would tfilled with a term that denoted a particular medicatioland the card as a whole (analogous to a sentence) wouenumerate the medications to be given to the patienThis is mostly the case, but according to Kaasbpll (i

preparation, p. 4):

~The sheets were filled in properly during the ::minutes, except for a couple of observed 'missirmedicines' on the medicine cards. When asked abothe 'mistakes' the treating nurse replied: 'Oh, but ~

know he (the patient) is going to have the medicineven if it is not written here. It is erased only becauhe has been under intensive care for some days.' "

The issue here is not the exact form of the cards. It coujust as well have been a computer system with dabase entries for medications. The situation-dependenof meaning is in the people who 'enter' and 'access' tldata. Accuracy is not just a matter of having tlcomputer keep its records straight. In designing alinstalling a system, it is critical to understand tJdifferent potentials for interpretation and either co:with them or modify them through training.

7D. Syntax

Discourse: This is the level that was visible in tl

wallgraph. There is a specific collection of forms, eawith its own syntactic structure (various fields, blanlboxes for checking, etc). These are organized inlarger structured collections (the Kardex, the patierecord files, etc.). An analysis of this structure wreflect all of the particulari ties of how they are filout, and ends up (as Kaasb~ll reports) quite compand confusing to the nurses themselves. Of course, asystem we build will have such a syntax, but if we vithe items as reflecting particular stepsconversations, there is order in the structure.

Sentential: The nurses use a wide variety of synta<

forms, ranging from highly specialized codesspecifying drug dosages ("50mg acetiminophen P

216

T.I.D") to informal natura l language in notes. A full unders tanding would include a detailed account of the profession-oriented languages (there will be more than

one in a set t ing like this where several professions interact). Medicine is an obvious example, in tha t medical terminology is consciously distinct from ordinary language. But in every organizational sett ing there is some degree of specialized language, suited to the part icular conditions of satisfaction that are re levant to the conversations.

8. B l indnesse s of a l a n g u a g e / a c t i o n perspective

The impacts of a technological innovation cannot be fully understood from any one perspective. Each perspective brings forth some concerns and is 'blind' to others. The coherence of a design comes from its generation within one perspective, but its success will come from anticipating breakdowns that become visible from other perspectives as well. The following list i l lustrates how the perspectives mentioned in Section 2 might interact with a design generated from concerns of language action.

Information process ing : Traditional information

system perspectives have centered on implementat ion problems. Analyses deal with files and records and specific channels for accessing them. They can then be used to create designs for effective and relatively efficient configurations of software and hardware. The language perspective does not have a close mapping to equipment considerations, although it does have

consequences for design at this level. We cannot succeed if we design a conversation structure that makes unreasonable demands on computer resources, or tha t requires wait ing for long periods for the system to do something. Similarly, a designer must pay at tention to the specific capacities of human memory, attention, and reasoning. Our relative lack of attention to those issues here does not mean that they are unimportant . It reflects the fact tha t they dominate the

current l i terature on computer system design (including the currently active interest in 'expert systems'), and need to be balanced with other perspectives.

The information processing perspective does not jus t include the immediate issues of data entry, storage, manipulation, and access, but also involves a larger web of issues surrounding the computer system itself. The design, acquisition, installation, maintenance, and

hiring and]or t raining of people to use a system have their own domains of conversation, possibilities and breakdowns (see Kling and Scacchi, 1982).

Ro les a n d mate r ia l s : The language/action perspective

deals with some of the same issues that arise in a role/function perspective, in part icular in defining roles

as pat terns of potential for action. But it does not deal with the more physical issues of resources and locations associated with work. Holt (1986) notes that "What first s tands out in any work environment is its a rch i tec ture- - tha t is to say its spatial-functional organization .... Functional proximity is what relates work places to each other. It is the relation which constrains and organizes the movement of people and

materials. . ." In designing a conversational s tructure we cannot ignore the brute physical situation: who has physical access to what objects when. There may be a critical difference between put t ing a single terminal at the main nursing station, put t ing a terminal in each examining and activity room, or having one available

by every bedside. In looking at the role of an individual, we need to recognize that his or her body can only be in one place at one time, and is limited in its

abil i ty to move from place to place. In a way this sounds mundane, bu t it is all too easy to design a system that would work wonderful ly-- i f the nurse would walk over to the nursing station before giving each pat ient medica t ion- -but doesn't succeed in practice.

A u t h o r i t y : An important aspect of every human

organization is the distribution of authori ty and power, and the mechanisms by which it is maintained. The

introduction of a new technology can perturb this s t ructure in a variety of ways, such as: facili tating detailed monitoring of performance; making it possible

for subordinates to work in ways that are not understood by their superiors; and in general opening possibilities for communication that cross the previous lines of authority. An analysis of conversational roles will identify part icular individuals as having the abili ty to initiate or respond in certain conversations,

and this s tructure is the practical consequence of authority. But the mechanisms by which authori ty is established and mainta ined go beyond this, and need to be confronted explicitly. In the hospital, there are powerful constraints on the appropriate role behavior of doctors and nurses. The structure of interactions within the organization mainta ins this identi ty and changes can threaten it. In the nursing example, one

217

T.LD") to informal natural language in notes. A fullunderstanding would include a detailed account of theprofession-oriented languages (there will be more thanone in a setting like this where several professionsinteract). Medicine is an obvious example, in thatmedical terminology is consciously distinct fromordinary language. But in every organizational settingthere is some degree of specialized language, suited tothe particular conditions of satisfaction that arerelevant to the conversations.

8. Blindnesses of a language/actionperspective

The impacts of a technological innovation cannot befully understood from anyone perspective. Eachperspective brings forth some concerns and is 'blind' toothers. The coherence of a design comes from itsgeneration within one perspective, but its success willcome from anticipating breakdowns that become visiblefrom other perspectives as well. The following listillustrates how the perspectives mentioned in Section 2might interact with a design generated from concerns oflanguage action.

Information processing: Traditional information

system perspectives have centered on implementationproblems. Analyses deal with files and records andspecific channels for accessing them. They can then beused to create designs for effective and relativelyefficient configurations of software and hardware. Thelanguage perspective does not have a close mapping toequipment considerations, although it does haveconsequences for design at this level. We cannotsucceed if we design a conversation structure thatmakes unreasonable demands on computer resources,or that requires waiting for long periods for the systemto do something. Similarly, a designer must payattention to the specific capacities of human memory,attention, and reasoning. Our relative lack of attentionto those issues here does not mean that they areunimportant. It reflects the fact that they dominate thecurrent literature on computer system design(including the currently active interest in 'expertsystems'), and need to be balanced with otherperspectives.

The information processing perspective does not justinclude the immediate issues of data entry, storage,manipulation, and access, but also involves a largerweb of issues surrounding the computer system itself.The design, acquisition, installation, maintenance, and

hiring and/or training of people to use a system havetheir own domains of conversation, possibilities andbreakdowns (see Kling and Scacchi, 1982).

Roles and materials: The language/action perspective

deals with some of the same issues that arise in arole/function perspective, in particular in defining rolesas patterns of potential for action. But it does not dealwith the more physical issues of resources and locationsassociated with work. Holt (1986) notes that ''What firststands out 10 any work environment is itsarchitecture-that is to say its spatial-functionalorganization.... Functional proximity is what relateswork places to each other. It is the relation whichconstrains and organizes the movement of people andmaterials..." In designing a conversational structurewe cannot ignore the brute physical situation: who hasphysical access to what objects when. There may be acritical difference between putting a single terminal atthe main nursing station, putting a terminal in eachexamining and activity room, or having one availableby every bedside. In looking at the role of anindividual, we need to recognize that his or her bodycan only be in one place at one time, and is limited in itsability to move from place to place. In a way thissounds mundane, but it is all too easy to design asystem that would work wonderfully-if the nursewould walk over to the nursing station before givingeach patient medication-but doesn't succeed inpractice.

Authority: An important aspect of every human

organization is the distribution of authority and power,and the mechanisms by which it is maintained. Theintroduction of a new technology can perturb thisstructure in a variety of ways, such as: facilitatingdetailed monitoring of performance; making it possiblefor subordinates to work in ways that are notunderstood by their superiors; and in general openingpossibilities for communication that cross the previouslines of authority. An analysis of conversational roleswill identify particular individuals as having theability to initiate or respond in certain conversations,and this structure is the practical consequence ofauthority. But the mechanisms by which authority isestablished and maintained go beyond this, and need tobe confronted explicitly. In the hospital, there arepowerful constraints on the appropriate role behavior ofdoctors and nurses. The structure of interactionswithin the organization maintains this identity andchanges can threaten it. In the nursing example, one

217

might imagine merging the various records and thus eliminating the Kardex. But, according to Kaasb¢ll (in preparation):

"Nurses are traditionally a paraprofession

subordinate to the physicians and their medical

knowledge .... In the nurses' struggle for acceptance of nursing as a profession, the theoretical concept 'nursing process' and its practical documentation in

the Kardex is of central importance for developing

nursing as a science on its own. In this struggle, the

Kardex as a basis for nursing decisions may be seen

as the nurses" answer to the physicians' medical

records."

In careful system development studies, this kind of issue has often emerged, for example between journalists and typographers in newspaper-publishing systems (Howard, 1985) and between librarians and clerks in libraries (Andersen and Madsen, 1986). The redesign of work is a negotiation among the groups already doing and supervising the work, and the results will be shaped by the interests of these groups and the compromises among them. In contrasting the 'tool perspective' to a more traditional systems perspective, Ehn and Kyng (1984) focus on this issue, looking at ways to maintain the autonomy of workers in the face of computer-based changes that can potentially be used to expand centralized authority.

Conflict: Most of the organizational models applied to

information systems design are based on the assumption of shared goals among the participants. In real organizations there are always conflicts among competing goals held by different individuals and groups. In some cases this is institutionalized (as in contractual labor-management relations or internal market competition in a firm), but it is always present. A system that assumes idealized cooperation may easily fail as the result of behavior that the systems analyst might label as stupidity, sabotage, or just plain human stubbornness. An analysis that takes conflicting interests into account is not a vain attempt to dissolve them, but can channel them into explicit forms of mutually agreed-upon negotiation.

The language/action perspective establishes a structure for negotiation, based on a theory of

cooperation that assumes the willingness to enter into serious conversation, without assuming shared goals or agreement. A conversation for clarification, for example, might involve each party's negotiating to get a 'favorable deal,' but it can nevertheless result in a

mutual agreement. More work needs to be done in integrating a conflict perspective (see Ciborra, 1985, and Sandberg in Nygaard, 1986).

Interpersonal relations: One of the most obvious

effects of computer systems is the replacement of face-to-face verbal interaction with computer-mediated exchange. In extreme cases, a job such as data-entry clerk may involve no immediate human interaction at all. Some problems that can arise from this can be characterized in conversational terms. A face-to-face interaction that is identified as playing a particular conversational role (e.g., medication record entry) often has other components (conversations for possibilities) that are lost. However, as with the issue of authority, there is also the question of people's perceptions ot themselves and their relationships with others around them. This is not just a matter of %edside manner, ' but a fundamental issue in all cooperative work environments. In a study of the introduction of a production planning and control system into a factory, Schneider and Howard (1985, pp. 14-15) noted:

In the contributing areas, Production Suppor~

personnel are constantly engaged in informai discussions, promises, and agreements ....

Schedulers spend nearly hal f their time in meetings:

competing with their colleagues over shop capacit3

and priority (one likens the process to "butting

heads'). Thus, a major part of the productior

p lanning and control process involves the extremel3 social acts of persuasion, negotiation, and, at times

argument. As one Production Control expeditor put,,

it, 7 'm jus t one leaf on the tree. I try to go in any ant all directions in order to get a part out. It all depend: on developing working relationships with people i

other departments--purchasing, quality contro

manufacturing engineering. It's a matter of trm,

built up over time. Personalities play a big role in it

In their study, they show the pitfalls of trying t redesign the work to eliminate these interaction: Although the language/action perspective focusses o the conversations among individuals, it is structura not psychological. It asks us to look at the potentials f( interaction, but not the motivations and feelings th~ will lead to what people actually do. Questions of moot motivation, and personal satisfaction go far beyo~ anything that has been dealt with here, and al essential to successful design.

218

might imagine merging the various records and thuseliminating the Kardex. But, according to Kaasb(llll (inpreparation):

"Nurses are traditionally a paraprofessionsubordinate to the physicians and their medicalknowledge.... In the nurses' struggle for acceptance ofnursing as a profession, the theoretical concept'nursing process' and its practical documentation inthe Kardex is of central importance for developing

nursing as a science on its own. In this struggle, theKardex as a basis for nursing decisions may be seenas the nurses' answer to the physicians' medicalrecords."

In careful system development studies, this kind ofissue has often emerged, for example betweenjournalists and typographers in newspaper-publishingsystems (Howard, 1985) and between librarians andclerks in libraries (Andersen and Madsen, 1986). Theredesign of work is a negotiation among the groupsalready doing and supervising the work, and the resultswill be shaped by the interests of these groups and thecompromises among them. In contrasting the 'toolperspective' to a more traditional systems perspective,Ehn and Kyng (1984) focus on this issue, looking atways to maintain the autonomy of workers in the face ofcomputer-based changes that can potentially be used toexpand centralized authority.

Conflict: Most of the organizational models applied to

information systems design are based on theassumption of shared goals among the participants. Inreal organizations there are always conflicts amongcompeting goals held by different individuals andgroups. In some cases this is institutionalized (as incontractual labor-management relations or internalmarket competition in a firm), but it is always present.A system that assumes idealized cooperation mayeasily fail as the result of behavior that the systemsanalyst might label as stupidity, sabotage, or just plainhuman stubbornness. An analysis that takesconflicting interests into account is not a vain attemptto dissolve them, but can channel them into explicitforms of mutually agreed-upon negotiation.

The language/action perspective establishes astructure for negotiation, based on a theory of

cooperation that assumes the willingness to enter intoserious conversation, without assuming shared goals oragreement. A conversation for clarification, forexample, might involve each party's negotiating to geta 'favorable deal,' but it can nevertheless result in a

mutual agreement. More work needs to be done inintegrating a conflict perspective (see Ciborra, 1985,and Sandberg in Nygaard, 1986).

Interpersonal relations: One of the most obvious

effects of computer systems is the replacement offace-to-face verbal interaction with computer-mediatedexchange. In extreme cases, a job such as data-entryclerk may involve no immediate human interaction atall. Some problems that can arise from this can becharacterized in conversational terms. A face-to-faceinteraction that is identified as playing a particularconversational role (e.g., medication record entry) oftenhas other components (conversations for possibilities)that are lost. However, as with the issue of authority,there is also the question of people's perceptions ofthemselves and their relationships with others aroundthem. This is not just a matter of 'bedside manner,' buta fundamental issue in all cooperative workenvironments. In a study of the introduction of aproduction planning and control system into a factory,Schneider and Howard (1985, pp. 14-15) noted:

In the contributing areas, Production Supporlpersonnel are constantly engaged in informaldiscussions, promises, and agreements ....Schedulers spend nearly half their time in meetingscompeting with their colleagues over shop capacitJand priority (one likens the process to 'buttineheads'). Thus, a major part of the productior.planning and control process involves the extremelJsocial acts of persuasion, negotiation, and, at timesargument. As one Production Control expeditor putlit, '['mjust one leafon the tree. I try to go in any an,all directions in order to get a part out. It all depend,on developing working relationships with people i,other departments-purchasing, quality contromanufacturing engineering. It's a matter of trwbuilt up over time. Personalities playa big role in it"

In their study, they show the pitfalls of trying tredesign the work to eliminate these interaction:Although the language/action perspective focusses 0

the conversations among individuals, it is structuranot psychological. It asks us to look at the potentials f(interaction, but not the motivations and feelings thewill lead to what people actually do. Questions ofmooemotivation, and personal satisfaction go far beyollanything that has been dealt with here, and alessential to successful design.

218

9. C o n c l u s i o n

Throughout this paper, we have seen how particular analyses can lead to designs that shape the nature of the work people do in organizations. The system designer emerges as an activist--someone whose technical designs are based on a larger understanding of what could and should be done. At first sight, this seems to justify an imperialism that wise system designers would reject. The computer professional is not typically the person whose understanding of a work situation should be followed. In fact, training in computer science can potentially lead to systematic distortions of what is relevant and important--to what is popularly called a 'computer mentality.' But designers cannot avoid responsibility either. Every time a system is built and introduced into a work setting, the work is redesignedweither consciously or unconsciously. We cannot choose to have no impact, just as we cannot chose to be outside of a perspective. We can make conscious choices as to which ones we follow and what consequences we anticipate.

Of course, we do not have full power over what will happen. In all design we must recognize the ongoing process of adaptation and evolution. We cannot design work, or language, or even computer devices in a way that establishes 'once and for all' what they will be and how they will be used. Design leads to technology, the use of which in turn leads to new breakdowns and new understanding, and back to new design. Our perspectives will shape the directions in which we look when breakdown is encountered, but we cannot anticipate the forms it will take.

Once we recognize that design is an active intervention, it is also important to recognize that the 'designer' is not some outside data-processing expert who comes into a situation with his or her own agenda and prescriptions for change. One of the unifying threads of the SYDPOL projects has been a concern with design methods in which the workers--those who will be directly affected by the system--are involved from the earliest stages. In language/action terms, we would say that they must participate in the design of the basic structure of the conversations, not just in the detailed questions of syntax that typically emerge at late stages of implementation. The role of the professional is to make clear what the different questions and potentials are, and what their likely consequences will be.

This does not mean that the system designer is purely a facilitator, with no direction. System design is a means of change, and the designer, like every socially responsible person, has the potential to introduce directions he or she believes in. As an example, The Coordinator (see Winograd and Flores (1986) and Winograd (forthcoming)) is a computer-based system for the coordination of work, designed by Flores and others at Action Technologies. It is based on theories of language action, and is explicitly designed to 'coach' its users to take clearer actions in declaring and following up on their commitments. In doing so it is a conscious intervention in how organizations go about their business, backed up by explicit training of the users in conversation theory (through courses offered by Logonet, Inc.). Its philosophical directions are not hidden in the details of design, but are explicitly taught to prospective users. Judged by commercial experience, it is a productive intervention.

Ultimately then, the practical development of a language/action perspective will go hand-in-hand with attempts to do conscious system design--to anticipate the ways in which computer systems will change the structure of work, and to enter into conversation with the people who will be affected. Through this discourse we move towards a capacity for innovative and appropriate design.

Acknowledgments The work presented here was supported by the System Development Foundation under a grant to the Center for the Study of Language and Information. I wish to thank the participants in the CSLI working meetings on system development for their valuable contributions to the evolution of this work. Jens Kaasb¢ll and Kim Halskov Madsen were especially helpful in giving me access to and a better understanding of the work being done in some of the SYDPOL projects. My conversations with Fernando Flores have been the basis for my understanding, in the most fundamental ways. Without his teaching, my perspectives would have been far different. In addition to these people, Chauncey Bell, Bradley Hartfield, Franqoise Herrmann, Mary Holstege, and Liam Peyton gave insightful critiques of an earlier draft.

219

9. Conclusion

Throughout this paper, we have seen how particularanalyses can lead to designs that shape the nature ofthe work people do in organizations. The systemdesigner emerges as an activist-someone whosetechnical designs are based on a larger understandingof what could and should be done. At first sight, thisseems to justify an imperialism that wise systemdesigners would reject. The computer professional isnot typically the person whose understanding of a worksituation should be followed. In fact, training incomputer science can potentially lead to systematicdistortions of what is relevant and important-to whatis popularly called a 'computer mentality.' Butdesigners cannot avoid responsibility either. Everytime a system is built and introduced into a worksetting, the work is redesigned-either consciously orunconsciously. We cannot choose to have no impact,just as we cannot chose to be outside of a perspective.We can make conscious choices as to which ones wefollow and what consequences we anticipate.

Of course, we do not have full power over what willhappen. In all design we must recognize the ongoingprocess of adaptation and evolution. We cannot designwork, or language, or even computer devices in a waythat establishes 'once and for all' what they will be andhow they will be used. Design leads to technology, theuse of which in turn leads to new breakdowns and newunderstanding, and back to new design. Ourperspectives will shape the directions in which we lookwhen breakdown is encountered, but we cannotanticipate the forms it will take.

Once we recognize that design is an activeintervention, it is also important to recognize that the'designer' is not some outside data-processing expertwho comes into a situation with his or her own agendaand prescriptions for change. One of the unifyingthreads of the SYDPOL projects has been a concernwith design methods in which the workers-those whowill be directly affected by the system-are involvedfrom the earliest stages. In language/action terms, wewould say that they must participate in the design ofthe basic structure of the conversations, not just in thedetailed questions of syntax that typically emerge atlate stages of implementation. The role of theprofessional is to make clear what the differentquestions and potentials are, and what their likelyconsequences will be.

This does not mean that the system designer ispurely a facilitator, with no direction. System design isa means of change, and the designer, like every sociallyresponsible person, has the potential to introducedirections he or she believes in. As an example, TheCoordinator (see Winograd and Flores (1986) andWinograd (forthcoming» is a computer-based systemfor the coordination of work, designed by Flores andothers at Action Technologies. It is based on theories oflanguage action, and is explicitly designed to 'coach' itsusers to take clearer actions in declaring and followingup on their commitments. In doing so it is a consciousintervention in how organizations go about theirbusiness, backed up by explicit training of the users inconversation theory (through courses offered byLogonet, Inc.). Its philosophical directions are nothidden in the details ofdesign, but are explicitly taughtto prospective users. Judged by commercial experience,it is a productive intervention.

Ultimately then, the practical development of alanguage/action perspective will go hand-in-hand withattempts to do conscious system design-to anticipatethe ways in which computer systems will change thestructure of work, and to enter into conversation withthe people who will be affected. Through this discoursewe move towards a capacity for innovative andappropriate design.

Acknowledgments

The work presented here was supported by the SystemDevelopment Foundation under a grant to the Centerfor the Study of Language and Information. I wish tothank the participants in the CSLI working meetingson system development for their valuable contributionsto the evolution of this work. Jens Kaasbl/lll and KimHalskov Madsen were especially helpful in giving meaccess to and a better understanding of the work beingdone in some of the SYDPOL projects. Myconversations with Fernando Flores have been thebasis for my understanding, in the most fundamentalways. Without his teaching, my perspectives wouldhave been far different. In addition to these people,Chauncey Bell, Bradley Hartfield, FranlSoiseHerrmann, Mary Holstege, and Liam Peyton gaveinsightful critiques of an earlier draft.

219

R e f e r e n c e s Andersen, Peter Bogh and Madsen, Kim Halskov, How

to handle the intangible: Metaphors in design and use of computer system, draft report, Aarhus University 1986.

Austin, John, How to Do Things with Words, Cambridge, Massachusetts: Harvard University Press, 1962.

Barwise, Jon and John Perry, Situations and Attitudes, Cambridge, Massachusetts: Bradford/M.I.T. Press, 1983.

Ciborra, Claudio U., Reframing the role of computers in organizations, the transaction costs approach, Sixth Annual International Conference on Information Systems, Indianapolis, December, 1985.

Research Program on Situated Language, Report CSLI-1, Center for the Study of Language and Information, Stanford, 1984.

DeCindio, F., G. DeMichelis, and C. Simoni, Petri net models of office game rules for office automation, Instituto di Cibernetica, Universita di Milano, draft, 1985.

Ehn, Pelle and Morten Kyng, A tool perspective on the design of interactive computer support for skilled workers, in M. Saaksjarvi (ed.), Report of the Seventh Scandinavian Research Seminar on Systemeering, Helsinki, 1984.

Fafchamps, Danielle, Medical deliberation: an alternative approach to medical decision making, Stanford University dissertation, in preparation.

Flores, C. Fernando, Management and communication in the office of the future, Doctoral Dissertation, University of California at Berkeley, 1981.

Flores, C. Fernando and Juan Ludlow, Doing and speaking in the office, in G. Fick and R. Sprague (Eds.), DSS: Issues and Challenges, London: Pergamon Press, 1981.

Gerson, Elihu M., Information systems in complex organizations: Challenges for research, Tremont Research Institute, draft, 1985.

Holt, Anatol, Primitive man in the electronic work environment, Presented at the conference on electronic work, Milan, March 1986.

Holt, Anatol, H.R. Ramsey, and J.D. Grimes, Coordination system technology as the basis for a programming environment, Electrical Communication 57:4 (1983).

Howard, Robert, UTOPIA: Where workers craft new technology, Technology Review 88:3 (April, 1985), 43-49.

KaasbOll, Jens, Intentional development of professional language through computerization: A case study and some theoretical considerations, in Fuchs-Kittowski et al. (eds.) Proceedings of the IFIP Working Conference on System Design for Human Development and Productivity through Participation (Berlin, May 12-15, 1986), North Holland 1987.

Kaasb011, Jens, Observation of people working with information: A case study, draft dated Jan 7, 1986, submitted for publication.

Kling, Rob, and Walt Scacchi, The web of computing: Computing technology as social organization, in M. Yovits (Ed.), Advances in Computers, Vol. 21, 1982, 1-90.

Lakoff, George and Mark Johnson, Metaphors We Live By, Chicago: University of Chicago Press, 1980.

Nygaard, Kristen, Program development as a social activity, to be presented at IFIP 1986 World Conference, Dublin.

Nygaard, Kristen, and Pal Sorgaard, The perspective concept in informatics, Working Conference on Development and Use of Computer-based Systems and Tools, Aarhus, 1985.

Orr, Julian, Narratives at work, story telling a~ cooperative diagnostic activity, Conference on Computer-Supported Cooperative Work, Austin Texas, December 1986.

Schneider, Leslie, and Robert Howard, Offict automation in a manufacturing setting, stud ', prepared for the United States Office of Technolog~ Assessment, April, 1985.

Searle, John, Speech Acts, Cambridge: Cambridg University Press, 1969

Searle, John R., A taxonomy of illocutionary acts, in I~ Gunderson (Ed.), Language, Mind and Knowledge Minneapolis: University of Minnesota Press, 197~

344-369. Suchman, Lucy, Plans and Situated Actions: Th

Problem of Human-inachine Communicatiol Cambridge: Cambridge University Pres: forthcoming.

Winograd, Terry, Moving the semantic fulcrun Linguistics and Philosophy, 8:1 (1985), 91-104.

Winograd, Terry, The design of collaboration: contrastive case study of two tools for conversatio: forthcoming (submitted for publication).

Winograd, Terry and Fernando Flores, Understandir Computers and Cognition: A New Foundation f Design, Norwood, New Jersey: Ablex, 1986.

220

Plans and Situated Actions: ThHuman-machine CommunicatiOlCambridge University Pres:

ReferencesAndersen Peter Bogh and Madsen, Kim Halskov, How

to ha~dle the intangible: Metaphors in design anduse of computer system, draft report, AarhusUniversity 1986.

Austin, John, How to Do Things with Words,Cambridge, Massachusetts: Harvard UniversityPress, 1962.

Barwise, Jon and John Perry, Situations and Attitudes,Cambridge, Massachusetts: BradfordfM.I.T. Press,1983.

Ciborra, Claudio U., Reframing the role ofcomputers inorganizations, the transaction costs approach, SixthAnnual International Conference on InformationSystems, Indianapolis, December, 1985.

Research Program on Situated Language, ReportCSLI-l, Center for the Study of Language andInformation, Stanford, 1984.

DeCindio, F., G. DeMichelis, and C. Simoni, Petri netmodels of office game rules for office automation,Instituto di Cibernetica, Universita di Milano, draft,1985.

Ehn, Pelle and Morten Kyng, A tool perspective on thedesign of interactive computer support for skilledworkers, in M. Saaksjarvi (ed.), Report of theSeventh Scandinavian Research Seminar onSystemeering, Helsinki, 1984.

Fafchamps, Danielle, Medical deliberation: analternative approach to medical decision making,Stanford University dissertation, in preparation.

Flores, C. Fernando, Management and communicationin the office of the future, Doctoral Dissertation,University of California at Berkeley, 1981.

Flores, C. Fernando and Juan Ludlow, Doing andspeaking in the office, in G. Fick and R. Sprague(Eds.), DSS: Issues and Challenges, London:Pergamon Press, 1981.

Gerson, Elihu M., Information systems in complexorganizations: Challenges for research, TremontResearch Institute, draft, 1985.

Holt, Anatol, Primitive man in the electronic workenvironment, Presented at the conference onelectronic work, Milan, March 1986.

Holt Anatol, H.R. Ramsey, and J.D. Grimes,Coordination system technology as the basis for aprogramming environment, Electrical

Communication 57:4 (1983).Howard, Robert, UTOPIA: Where workers craft new

technology, Technology Review 88:3 (April, 1985),

43-49.

Kaasbl1ill, Jens, Intentional development of professionallanguage through computerization: A case studyand some theoretical considerations, inFuchs-Kittowski et al. (eds.) Proceedings of the IFIPWorking Conference on System Design for Human

Development and Productivity through Participation(Berlin, May 12-15, 1986), North Holland 1987.

Kaasbl1ill, Jens, Observation of people working withinformation: A case study, draft dated Jan 7, 1986,submitted for publication.

Kling, Rob, and Walt Scacchi, The web of computing:Computing technology as social organization, in M.Yovits (Ed.), Advances in Computers, Vol. 21, 1982,1-90.

Lakoff, George and Mark Johnson, Metaphors We LiveBy, Chicago: University ofChicago Press, 1980.

Nygaard, Kristen, Program development as a socialactivity, to be presented at IFIP 1986 WorldConference, Dublin.

Nygaard, Kristen, and Pal Sorgaard, The perspective

concept in informatics, Working Conference onDevelopment and Use of Computer-based System~

and Tools, Aarhus, 1985.Orr, Julian, Narratives at work, story telling a1

cooperative diagnostic activity, Conference OIl

Computer-Supported Cooperative Work, AustinTexas, December 1986.

Schneider, Leslie, and Robert Howard, OffiCIautomation in a manufacturing setting, stud:prepared for the United States Office of Technolofr.

Assessment, April, 1985.Searle, John, Speech Acts, Cambridge: Cambridg,

University Press, 1969Searle, John R., A taxonomy ofillocutionary acts, in E

Gunderson (Ed.), Language, Mind and Knowledg.Minneapolis: University of Minnesota Press, 1971

344-369.Suchman, Lucy,

Problem ofCambridge:forthcoming.

Winograd, Terry, Moving the semantic fulcrullLinguistics and Philosophy, 8:1 (1985), 91-104.

Winograd, Terry, The design of collaboration:.contrastive case study of two tools for conversatlO.forthcoming (submitted for publication).

Winograd, Terry and Fernando Flores, UnderstandirComputers and Cognition: A New Foundation fDesign, Norwood, New Jersey: Ablex, 1986.

220