Boston,MassachusettsUSA* April24-28,1994 HumanFactorsinComputingSystems%?

INFORMAL WORKPLACE COMMUNICATION: WHAT IS ITLIKE AND HOW MIGHT WE SUPPORT IT?

Steve Whittaker7David Frohlich and Owen Daly -JonestOffice Appliances Dept.

Hewlett-Packard Laboratories

Filton Road, Bristol,BS126QZ, UK.

steve.whittaker@ crd.lotus.tom, dmf@hplb.hpl.hp.tom, odj@hplb.hpl.hp. com

ABSTRACTWe present new findings about the nature of informat com-munications, derived from a naturalistic study of people’severyday working activities. We identify why such interac-tions are so common, and valuable and how they are achievedin the workplace. We also address weaknesses in currentsystems that support such interactions remotely and identifyfurther requirements for better support. We also discuss theimplications of this work for conversational theories.

KEYWORDS: Informal communication, audio, video, CSCW,workplace activity, ethnography.

THE NATURE AND IMPORTANCE OF INFORMAL COM-MUNICATIONSInformal communications we define as taking place syn-chronously in face-to-face settings. They are distinct fromother methods of office communication such as phone, doc-uments, memos, email, FAX and voicemail. For most officeworkers, informal communication is a frequent workplaceactivity and for many jobs such as management it representsthe most frequent workplace activity. Questionnaire data pro-duce estimates of between 25% and 70% of time being spentin face-to-face interaction, with these figures depending on

job type[8, 12, 14].

What then is the nature and function of informat communi-cation? One of the few observational studies, showed it tobe brief, unplanned, and frequent[8]. Informal communica-tion supports a number of different functions: the executionof work-related tasks; co-ordination of group activity; trans-mission of office cultur~ and sociat functions such as teambuilding[5, 8]. Its importance is shown by research into sci-entific collaboration demonstrating that physical distance is astrong predictor of whether scientists will co-publish, because

*Now at Lotus Development Corporation, 1 Rogers St., Cambridge, MA02142

t Also at Dept. of Psychology, U. of York, York YOI 5DD, UK.odjl@york.uk.a;

Permission to copy without fee all or part of this material is

granted provided that the copies are not made or distributed for

direct commercial advantage, the ACM copyright notice and the

title of the publication and its date appear, and notice is given

that copying is by permission of the Association for Computing

MachinerV. To copy otherwise, or to republish, raquires a fee

and/or epecific permission.

CH194-4/94 Boston, Massachusetts USA@ 1994 ACM 0.89791-650.6/94/0131 . ..$.3.50

people who are physically collocated are more likely to com-municate frequently and informally. Questionnaire studiesatso suggest that physicat proximity supports frequent op-portunistic conversations which are vitat to the planning anddefinitionat phases of projects[7]. Other questionnaire studiessupport the effects of proximity and hence informal commu-nications on social and cultural knowledge: Researchers aremore likely to be familiar with, and to respect the work ofcolleagues who sit close to them[8].

This work indicates that physicat proximity is crucial for in-format communication, but trends towards telework, mobilework and the globalisation of business are geographicallyseparating workers. Given its importance, what systemshave been built to support informal interaction at distance?In contrast to other methods of remote communication, in-format interactions are poorly supported by technology[8].The two main types of prototype system that have been builtboth involve synchronous audio and video. The tirst pro-vides a permanently open link between the commons areasof two geographically separated sites, with the aim of fa-cilitating unplanned conversations between workers at thetwo locations [3, 4]. Commons area connections have beenmoderately successful in promoting brief social interactions,with 70$Z0of the Xerox Portland link conversations beingshort remote “drop-ins”in which greetings are exchanged[l].Although these interactions would not have occurred with-out the link, the general conclusion was that while the linkwas “barely adequate to promote shared context and cul-ture” it was “insufficient for accomplishing tasks (P51)’’[5].Bellcore’s Videowindow generated similar conclusions, withunplanned interactions between remote sites atso occurringover the link. However there was evidence that brieflyseeing someone over the videolink was less likely to pro-mote an extended interaction, than an equivalent face-to-facesighting[4].

The second class of system for remote informal communica-tions involves Desktop video, either point-to-point[51 or be-tween multiple locations [9]. Some of these systems provideadditional visual information about the communication statusof the call recipient, using a “glance” feature. Here callers cancheck the availability of their recipient, by opening a visuatlink for a few seconds, prior to initiating the conversation.An evahtation of one such system showed, however, that itdoes not effectively replicate face-to-face interaction[5]. Thesystem was rated as closer to the phone than face-to-face in-

131

a HumanFactorsinComputingSystems CHI’94 * “Ce/ebratin(qlnterdepende}tce”__,.,_, ,

teraction, and less useful than face-to-face for doing work orlearning about their conversational partner. Furthermore, anattempt to duplicate the process of opportunistically “bump-ing into” someone in the corridor, which was implementedby arbitrarily connecting two users was highly unsuccess-ful with 97% of such links being terminated immediately[5].One positive result. however, was that people atways choseto use videophone over phone atone. Another study foundthat Desktop videoconferencing partiatly replaced the use ofother media such ax email or brief face-to-face meetings. Itatso increased the use of Shared Workspaces and wax alsoperceived to reduce phone use[17].

Informat communications also raise numerous questions forinteraction theories. Informal conversations are mainly op-portunistic, so how do participants co-ordinate and initiatethem, given that they are unanticipated? There are also of-ten time lags between informal interactions: how do peoplemaintain and re-establish context, given such lags? Finatly,informal communications often occur between frequent inter-actants who often share large amounts of background knowl-edge. What affect does this shared knowledge have on initi-ating conversations and managing conversation context?

We provide data on the basic properties of informal com-munications such as their frequency, duration, whether theyare pre-arranged, and the role of documents in such interac-tions. For interactions away from desk, we also look at wherethey occurred. Next we look at their structural properties,examining how conversations are opened and closed and theways that participants introduce and agree on conversationcontext. We also test predictions about frequency: frequentinteractants should share background context as well as be-ing familiar with each other. We therefore expected frequentinteractants to show less formatity in opening and closinginteractions as well as reduced need for context setting.

METHODMany previous studies of workplace communication have re-lied on self-report techniques such as interviews, diaries andquestionnaires [12]. Unfortunately these methods are foundto be extremely inaccurate when reported estimates of ac-tivity have been compared with actual activity measures [12].More recent observationat studies of workplaces have be-gun to provide a more accurate picture of communicationbehaviour, usually by audio or video recording of activitywithin a single medium such as face-to-face meetings[l 1, 16],videoconferences [l O], or a single setting such as controlrooms[ 15]. The main problem with these studies, however,is that they do not allow an anatysis of the evolution of longerterm communication patterns across media and settings.

To overcome these problems, we developed a new method of

workplace observation called “remote shadowing” in which

we tracked and recorded the activities of individual office

workers in the absence of an observer. We used a portableSony camcorder (CCD-TR303E), with a wide angle lens anda radio microphone receiver, fixed in the corner of each sub-ject’s office. Participants wore a radio microphone transmitterwhich captured atl their spoken conversations around the of-fice site, even when they were out of camera shot. Wheneverpossible, arrangements were made to move the camcorder to

off-site locations and to special meeting rooms on-site whencommunication happened there, but for the majority of outof office communications, we have only audio data. Copiesof written correspondence were also obtained, together withnotes made in visually recorded meetings. Brief discussions

about work practice also took place every day between ob-server and participant.

We recorded a full working week for both participants, ex-cluding lunch hours, and confidential meetings. An inherentlimitation of our method is that it produces huge amounts ofvideo and audio data, which are time-consuming to process

and analyse. We therefore have only two main participantsin the study, but this is compensated for by the fact that theytalked to 97 other people, in a total of 294 different syn-chronous interactions. We started with a total of 55 hours oforiginal recordings, for both participants combined. After weexcluded confidential data, interviews, solitary activity andequipment failure there were 29 hours of conversational data.

The first p~ticipant B., is Communications Manager for anindustrial research laboratory. She manages public relationsand marketing communication, and facilitates collaborationbetween research and manufacturing. She is often away fromher desk interacting with others mainly on site. During thetarget week she was engaged in four main activities; organ-izing a Lab visit involving 20 internal and external people,identifying the marketing and communication requirementsfor a research project, organizing a press conference, andplanning a communications strategy presentation.

The second participant, R., is a surveyor for a medium sizedconsultancy dealing with commercial property letting andvaluation. He works as a professional negotiator for clientsfacing rent reviews or lease renewals. He is often absent fromhis desk, both within the office site and on frequent trips toclient premises within the city. He interacts frequently withothers, mainly using his office phone for business negotiationswith clients who are off-site. During the target week hewas engaged in about 30 ongoing cases. Three demandedparticul~ attention: a written response to an opposition reportabout a rent review undergoing arbimation; arranging to actfor a new tenant client in a rent review cas~ and completingthree lease renewals.

We chose these participants because they are mobile profes-sionals for whom communication is a central part of theirjob. They exhibit two main forms of local area roaming:in building roaming (B.) and out of building/metropolitanarea roaming (R.). Together these forms account for 87% ofthe “mobility” exhibited by mobile professionals in the US,in Contrast to wide area roaming outside the city or coun-try base[2]. Future work should contirm how these resultsgeneralise to larger numbers of participants in different pro-fessions.

Boston,MassachusettsUSAo April24-28,1994 HumanFactorsinComputingSys(ems%?

RESULTS

The Nature Of Informal Communications

Below we present some typicat examples indicating the char-acter and function of these interactions:

Example 1: R. and F.

R. IS STOOD UP READING A DOCUMENT BEHINDHIS DESK WHEN HIS COLLEAGUE F. WALKS INTOVIEW EN ROUTE TO HIS OWN DESK FROM ANOTHEROFFICE.

1. R: ( LOOKS AT DOCUMENT)

‘ ‘F. can you reath read (--) this:s

report for me? ‘ ‘

(1. 5s WALKING TOWARDS EACH

LOOKING AT THE DOCUMENT)

2. F : (WALKING AND LOOKING AT

DOCUMENT),, E:::: rrh (.) now?”

(1.0)

3. R: (LOOKING AT DOCUMENT)

OTHER BOTH

THE

“.hhhhhhh “Aiy if you’ve got a

minute’ ‘

4. F: (LOOKING AT DOCUMENT)

“Yeah - ‘r

In this 8 second interaction, R. sees that F. is moving aroundthe office and hence is not currently directly engaged in work.Hetherefore opportunistically engages F.’s help. Note theimportance of a shared visuat environment in affording R.this information about F.>savailability and allowing F to lookat the document.

Example 2: R. and E

R. IS DICTATING AT HIS DESK. F. IS ON THE PHONEACROSS THE OFFICE

1. F: ‘ ‘A:lri_ghk thank you bye bye’ ‘

(F PUTS DOWN PHONE)

R IS REPLAYING HIS DICTAPHONE

(0.7)

2. R: (TO F.)

‘ ‘Is he alright?”

3. F: .“Yeah”

4a. R: “Which one’s he-he’s got-”

(0.7)

4b.R: ‘ ‘There’s a restaurant’ ‘

(0.5)

5. F: “I said that 1’11 do this one

initially and then further afield”

6. R: ‘ ‘Which one’s that?’ ‘

7. F: “That’s: eighdy two it’s the

offices’ ‘

8.R: “>Oh< th-(.) hhh (.) Yeah::h

we act for the landlord on that one. I

did a rent review against him on that”

. . . . . . . . . . . .

This longer interaction fragment lastingl 5s, also showsanunpkumedconversation. Itarosebeeause R.heardF.’sphonecall and wanted to monitor the outcome. It finishes byR.offering unprompted advice and assistance. The interactiondisplays the implicit shared context between the participants.Withoutbeingtold, R. knowstheidentity ofF.’scaller(L2),anddetails of thecase(L4a,4b). Unprompted, R, proceedsto supply background information which F. may notalreadyhaveknown(L8). The sharedcontext results inacondensedandcrypticconversational style. Thisconversation continuedfor several more minutes after the extract. R. gave moredetails and offered a warning about acting for both client andtenant. Thus an unplanned conversation led to a detailedtask oriented discussion. A final example is an interactionconsisting of a single utterance, interrupting a person on thephone “It’stwen~fivep astfour’’, withanacknowledgement

“okay”. This served as a reminder that materials had to be

posted that evening to meet an important deadline. This againindicates the brevity and context-dependence of this type ofinteraction.

Figure 1 presents general data on: Own Office communi-cation (face-to-face with others in one’s own office), andRoaming (Out of Office: in others’ oftices, meeting rooms,offsite, public areas, including communication episodes intransit between these). A communication event was definedas a synchronous face-to-face verbal interaction, over andabove a greeting. It excludes asynchronous or technologymediated communications such as phone, email, voicemailand FAX, as well as non-communicative activity, e.g. soli-tary actions at one’s desk, or watking around the building.Arranged meetings were defined as being explicitly agreedand scheduled by participants.

Figure 1 shows that informat communication accounts for alarge proportion (31%) of work time. This overall figure,however, is mainly composed of a large number of brief, un-planned, dyadic interactions. Extended, arranged, multipartyinteractions are highly unusual. The brevity of the interac-tions is striking: in the Own Office case, 50% of interactionslast less than 38s.

We also looked at where Roaming conversations occurredand their mean duration. The majority of these (67%) werein another person’s office (mean duration, 1.94 reins), 1570in public areas (mean, 1.06 reins), and 17% whilst on themove (mean, 0.82 reins), A few longer interactions (1%)took place in dedicated meeting rooms (mean, 13.13 reins).These data show that the location of a roaming interactioninfluenced its duration. Consistent with other work[13], weatso found that 62% of outgoing phone calls by our partic-ipants failed to connect with their intended recipient. Thismay be explained by the large amounts of time spend roam-ing, and it presents major problems for people trying to make

synchronous connections, especially from offsite.

StructureDue to the lack of mobility of the camcorder, we only havecomplete visuat information about the activities of both par-ticipants in the Own Office case. Given the importance ofvisual information, for our analysis, we therefore only lookedat conversation structure for the 152 Own Office interactions.

133

Q‘–-lHumanFactorsinComputiigSystems CM’94 ~ “Ce[ebrati\~jinterdependence”

Own office Roaming Overall(N= 152) (N=142) (N= 294)

%Total Work 14 17 31

Mean duration 2.37 1.38 1.89

(reins)

Mean frequency 11.57 12.38 5.98

(reins)

% Unscheduled 89.47 96.48 92.86

% Dyadic 82.23 84.50 83.32

Figure 1: Informal communication characteristics, forown office, and roaming

In a typicat scenario here, the recipient is engaged in prior

activity which is interrupted to initiate the interaction. Wenoted the cues that the caller used to determine when to ini-

tiate. Very few of these interactions(11 %) were prefaced bygreetings (e.g. “hello”, “hi”) by either party. When the recip-ient was already engaged in activity, the callers only waitedfor a verbal or visual orienting sign of attention from the re-cipient on 32% of occasions, other times they just launchedinto the interaction. Vkual orientation was defined as a phys-ical body movement towards the caller, and such movements

were obvious in the majority of cases, because attention hadpreviously been directed at the desktop (reading, writing,phone) or at another person (talking), so that gaze at thecaller produced a distinct head movement. There was onlyone instance of verbal orientation (“Take a seat”).

The recipient’s prior activity determined interruptibility. Ac-

tivities involving talking were less interruptible than silentreading or writing, with callers being more likely to wait foran orienting cue before commencing (X2 = 11.21, p <0.001,df = 1). There were no differences in interruptibility, how-ever, between the different forms of verbal activity, namelytalking, dictating and phoning.

We also analysed interaction closings. Only 3’% of occasionsended with formal farewells (“bye”, “see you”). There werealso few verbal or visuaf orienting signs (e.g. turning away)given to indicate conversations were about to close: theseonly occurred on 21 YO of occasions.

To determine how participants managed context we analysedthe first utterance of each interaction, after the greeting. Wenoted whether or not this tirst utterance assumes a commonreference to a previous interaction or issue, and prior knowl-edge on the part of the recipient. On 7570 of occasions, par-ticipants assumed prior context, this was often accompaniedby deixis, e.g.

‘ ‘mm – this invoice have you got the

original quote and also the quote from

Triangle? ‘ ‘

‘ ‘ I have taken a detailed look at this

now, and the problem was . . . ‘ ‘

Frequent Infrequent Chi-Squared

Duration (sees) 37.8 219.0

%Greetings 2 23 12.06*

%Farewells o 10 5.13*

%Intermption 91 64 11.90*

%Final 21 43 5.86*arrangement

%Previous context II 24 1 19 I 0.37 IFigure 2: Impact of interaction frequency

Contrast this with interactions which have no prior history orcontext.

“Hello B., we’re doing a bit of an

audit around the site . . . . ‘ ‘

,, B I ‘m taking a couple of days leave,

Mon&y and Tuesday. ‘ ‘

Context management also involves planning the next time toconverse. Given the opportunistic nature of most of theseconversations, we examined when people formally made ar-rangements to meet again. We looked at the final utterance,or when there were farewells, the penultimate utterance. Ar-rangements were defined as agreed future directed actions orscheduling meetings. Formal arrangements were made on28% of occasions only.

The Effect Of Interaction Frequency On StructureNext we examined how Own Office communications dif-fered depending on interaction frequency. Frequency af-fects the interactants’ familiarity with the subject materialand each other. We identified interactions between peoplewho talked infrequently. This analysis was conducted onlyfor the dyadic interactions.

As we expected, more frequent interactants had briefer con-versations (Pearson r = -0.24, p <0.05, df = 103). Frequentinteractants were much less likely to be format, producingfewer greetings and farewells and more interruptions (definedas an interaction that begins in the absence of a recipient vi-sual/verbal attention cue). Asterisks in the Figure indicatesignificance levels at p < 0.05. Frequent interactants atsomade fewer future-directed arrangements within the meet-ing. Arranged meetings were also much less likely to occurbetween frequent interactants (X2 = 10.58, p < 0.001, df= 1). Contrary to our expectations, however, frequency hadlittle effect on whether participants chose to preface their con-versations with remarks that reintroduced context. We hadexpected familiar interactants to have little need for explicit

1Dem~d as th~,e Whoover the whole week averaged fewer than tWO

interactions per day. Thus we categorised 55% of conversations asoccurring

between infrequent interactarrts, and 45% between frequent interactants.

134

Boston,MassachusettsUSA* April24-28,1994 HumanFaciorsinComputingSystems%!!

Annotate Cue Answer Prop

Report 56.3 73.2 83.3 85.7 I

Memo 43.8 2.4 4.2 0

File o 14.6 8.3 0

Other o 9.8 4.2 14.3

% Joint 25 41.5 50 46.4 I

Figure 3: Document usage: type and function

context reintroduction, but it maybe that frequent interactants

share many potential contexts and it is therefore necessary forthem to be explicit about context in order to distinguish these.

The Role Of DocumentsDocuments were involved in 53% of atl Own Office inter-actions. We anatysed the episodes containing documents.The four main types of documents were reports (typed ma-terials); memos (personal notes/postits/handwritten notes);files (material accessed from an archive); and other (includ-ing photos, viewgraphs, books). We also looked at how thedocument was used during the conversation: for annotationor signing; to ask a question; to answer a question; or as aconversational prop, where participants used the document tostructure the conversation by repeated visual attention, refer-ence and gesture at the document[16, 19]. When a documentserved multiple functions we noted each of these. Figure 3shows that reports were the most common type of documentinvolved in conversations, and that most of the time thesewere talked about (for question, cue or prop). Annotationof reports was less common, and atmost hatf of annotationwas personat notes. We also looked at when the documentwas shared (defined as joint visual orientation). Informationcue, answer, and prop functions, normally involved sharingdocuments. Annotation in contrast, was normally a solitaryactivity.

SYSTEM IMPLICATIONSHow might we support informaf interactions when groupsare geographically distributed? Again we add the provisothat these findings are based on a very small sample and maybe sensitive to job specification, physicaf layout and officeculture. We should also test our analyses against people’sperceptions of the structure and function of these interac-tions. This study did not determine, for example, the relativeimportance of office versus meeting room conversations forparticipants, or the perceived intrusiveness of different waysof initiating conversations. Nevertheless, our study docu-ments the frequency of informal communications: they ac-count for 31% of office activity. Taken together with otherstudies showing that removing such interactions significantlydecreases effective collaboration [7], this reinforces claimsthat this is a major area for technology support.

Our results show that such interactions are generally dyadic,very brief, with few format greetings or farewells, and callersoften began without waiting for visible signs of recipientreadiness. The absence of openings and closings and their

interruptive nature suggests a view of informat communi-cations as one long, intermittent communication comprisingmultiple brief related fragments with an open channel. Figure2 also shows that infrequent interactants tend to have longerconversations and be slightly more formal.

How might we support this in software? One possibility ispersistent attdio/video links to the set of frequent interactants,offering minimat cost in initiating conversation, and readyaccess to the current workgroup. For infrequent interactants,a connection model may be more appropriate but connectionmust still be quick and effortless: if the whole interaction

only lasts a few minutes, then startup and close must be briefcompared with this[5, 18].

How should we achieve initiation? Overall the data suggestthat frequent callers begin almost regardless of recipient’svisual activity. Thus current implementations of “glance”facilities [5, 9], in allowing the caller information about thepresence/absence of the recipient, and the recipient an audiocue that they are being observed, may be sufficient here.It does not seem to be necessary to supplement this withmore specific information about recipient’s visual attention,e.g. whether they are reading or writing. Recipient’s verbalactivity was important in initiation, however, so that a brief“eavesdrop”, giving an indication of whether the recipientis talking, may be useful here2. More format methods ofinitiation requiring recipient feedback may be needed withless frequent catlers, given that they were more likely towait for a recipient cue before beginning, Visual recipientfeedback has been shown to be important for certain classesof video mediated interactions[6].

Over half the interactions involved documents, suggestingthe need for an integrated Shared Workspace. Document useindicates a requirement for simple systems rather than full-blown shared editors. A system that allowed mutual viewing

of documents, with the ability to point at and possibly makesimple annotations, may be atl that is required here[19].

Most interactions were dyadic, arguing for videophone ratherthan multipartys ystems, atthough 88% of interactions wereterminated by a third party joining an existing conversation.There may therefore be a need to support this type of inter-jection. Conversations in offices were more frequent thancommons area conversations, with only 15% of conversa-tions occurring in public areas. This implies that we shouldfocus on desktop communications rather than commons-areaapplications[4], although further data should be collectedabout the relative importance to the participants of the differ-ent classes of conversation.

A relatively large number of roaming interactions take placein other people’s offices. This indicates that the callers weredeliberately looking for the recipients for aparticularpurpose,rather than “falling into” conversation with others they hap-pened to bump into while out of the office. Thus interactionsappeared purposeful even though they were not generallyscheduled to take place in advance. This was suggested bythe character of opening lines of interactions, and by the fact

2~is may have to be masked to preserve privacy, but the fiPowt

information to transmit is that the recipient is engaged in conversation

135

E!?!HumanFactorsinComputiigSystems CHI’94 * “Ce/ebwitgInterdependence”

that the majority of roaming interactions were self-initiated(75%) whereas most own office interactions were initiated byothers (60%). This again argues for desktop connections andexplains the lack of success of attempts to connect arbitrayparticipants via videophones[4].

Consistent with other studies, our participants experiencedproblems in achieving connection outside the building[13].People are often away from their desks, explaining the ob-

served lack of success in making phone calls. Combinationsof communication technologies are needed to address thisproblem. We need increased integration of synchronoustechnologies with those that do not require temporat co-ordination, eg. voicemail or electronic postits [1 8], for non-urgent messages. For urgent communications, that requireimmediate response, the requirement might be for a mobilephone or pager. Further studies about the urgency, and hencethe requirement for synchrony for different types of inter-personal communications are needed to determine the appro-priate technologies here. We also need to understand how topresent these different communication options, eg. postit ver-

sus pager, to the caller, and to provide effective filters for therecipient to prevent intrusions in non-urgent circumstances.T’his integration should allow asynchronous technologies toplay an important role in co-ordinating remote interactants inthe absence of a shared physical environment.

THEORETICAL ISSUESOur data also have ramifications for conversational theory.Conversations are traditionally regarded as having a begin-ning and ending, with conversational context evolving throughthe discussion. In contrast, we found that informal commu-nications seem to consist of one long intermittent conversa-tion consisting of multiple unplanned fragments often lackingopenings and closings. This presents two major problems forparticipants that do not occur in standard conversations ac-counts (a) co-ordination to achieve co-presence, given theunplanned nature of the fragments; (b) regenerating contextbecause of time lags and intervening activity between inter-mittent interactions.

How is co-ordination achieved given the amount of time thatpeople spend away from their offices and in conversation withothers? Currently co-ordination works most effectively forphysically collocated people because they have multiple op-portunities to find the other people available for conversation.The challenge for future technology is in supplying this typeof information for geographically distributed groups. Com-munication frequency atso affected the planning of such inter-

actions: people who met often, had little need to pre-arrangemeetings, because the frequency of interaction guaranteedmultiple future opportunities to interact. The opportunisticnature atso explains another finding. Two person conversa-tions are more likely than multipamy because the chances oftwo people being at the same location at a given time aregreater than for three.

There may also be degrees of opportunism and another issueconcerns the differences between interactions that are: (a)pre-arranged by both participants; (b) brought about when justone participant seeks another out; (c) the result of genuinelychance encounters, where neither intended the interaction

to take place. Our findings were that interactions “on themove” are much shorter than those in meeting rooms, whichmay support the view that genuinely chance encounters arebriefer than pre-arranged ones, but we need to validate thisagainst participant’s accounts of how different interactionscome about.

Another related variable is where the interaction occurred,

and we found differences in duration relating to location.Why is this? Conversational resources such as documentsmay be limited at out-of-office locations, such as commons-

areas or “on the move”, and this may reduce conversationalduration, because of the absence of these resources. Lo-cations may also differ in their perceived intrusiveness: incommons-areas or “on the move” the caller is not interruptingthe recipient in mid-activity, so that even infrequent interac-tants may show little initiation behaviour.

How is context management achieved? Our initial viewwas that it would be closely related to interaction frequency,with frequent conversations among familiar people servingto preserve context across interactions. We expected morecontext-dependent, more cryptic and hence briefer interac-tions among frequent interactants. Although conversationsamong frequent interactants were shorter, however, we foundno effect of frequency on context management. Future workshould carry out more detailed analyses of other measuresof context-dependence, e.g. anaphors, deixis, personal pro-nouns to find out whether these are prevalent among frequentinteractants. We also only looked at context management atbeginnings and ends of conversations, and future work shouldanalyse the whole conversation. A further possibility is thatdocuments carry context[16, 19], so we would predict morecontext-dependence when conversations include documents.

A final related issue is the sheer brevity of informal commu-nications. The absence of openings, closings, and the reducedneed to make arrangements for future meetings clearly reduceconversational length, but are other factors at work? The pos-sibility that shared context allows participants to be crypticremains to be substantiated. Another hypothesis relates toplanning: if participants are uncertain that they will meetfrequently they may condense multiple issues into a singleinteraction. In contrast, familiar interactants know that futureconversations are guaranteed, so that only the most pressingones need be done now. Again however, this needs to betested.

ACKNOWLEDGEMENTS

Thanks to Brid O’Conaill and Phil Stenton for discussionsand help with data analysis and to all the participants takingpart in the study.

REFERENCES

1.

2.

M. Abel. Experiences in an exploratory distributed or-ganization. In J. Galeger, R. Kraut, and C. Egido, ed-itors, Intellectual Teamwork.” Social and Technologic-al Foundations of Co-operative Work, pages 489–5 10.Lawrence Erlbaum Associates, Hillsda.k NJ, 1990.

W. Ablondi and T. Elliot. Mobile professional marketsegmentation study. BIS Strategic Decisions, 1993.

136

Boston,MassachusefisUSAo April24-28,1994 HumanFactorsinComputingSystems%?

3. S. Bly, S. Harrison, and S. Irwin. Media spaces: Bring-ing people together in a video, audio and computingenvironment. Communications of the ACM, 36:2845,1993.

4. R. Fish, R, Kraut, and B. Chatfonte. The videowindow

system in informal communication. In Proceedings ofthe Conference on Computer Supported Co-operativeWork, pages 1-12,1990.

5. R. Fish, R. Kraut, R. Root, and R. Rice. Video as a tech-nolog y for informat communication. Communications

of the ACM, 36:48-61, 1993.

6. C. Heath and P. Luff. Disembodied conduce Commu-nication through video in a multi-media environment.In Proceedings of Conference on Computer Human In-teraction, pages 99–103, 1991.

7. R. Kraut, C. Egido, and J. Gategher. Patterns of contactand communication in scientific research collaboration.In J. Gategher, R. Kraut, and C. Egido, editors, Intel-lectual Teamwork, pages 149–173. Lawrence ErlbaumPress, Hillsdate, N.J., 1990.

8. R. Kraut, R. Fish, R. Root, and B. Chalfonte. Infor-mat communication in organizations: form, functionand technology. In R. Baecker, editor, Groupware andComputer-Supported Co-operative Work, pages 287–314. Morgan Kaufmann, 1993.

9, M. Mantei, R. Baecker, A. Sellen, W. Buxton, T. Milli-gan, and B. Wellman. Experiences in the use of a mediaspace. In Proceedings of the Conference on ComputerHuman Interaction, pages 203–209, 1991.

10. B. O’Conaill, S. Whittaker, and S. Wilbur. Conversa-tions over video-conferences: An evaluation of video-medlated interaction. To appear in Human ComputerInteraction, 1993.

11. G. Olson, J. Olson, M. Carter, and M. Storrosten. Smallgroup design meetings: An anatysis of collaboration.Human Computer Interaction, 7;347–374, 1992.

12. R. Panko. Managerial communication patterns. Journalof Organisational Computing, 1992.

13. R. Rice and D. Shook. Voice messaging, co-ordinationand communication. In J. Galegher, R. Kraut, andC. Egido, editors, Intellectual Teamwork, pages 32’7–350. Lawrence Erlbaum Press, Hillsdate, N.J., 1990.

14. L. Sproull. The nature of managerial attention. InL. Sproull and J. Larkey, editors, Advances in infor-mation Processing in Organizations. JAI Press, 1984.

15. L. Suchman. Constituting shared workspaces. InD. Middleton and Y. Engestrom, editors, Cognition andcommunication at work. Sage, 1992.

17. J. Tang and E. Isaacs. Why do users like video: Stud-ies of multimedia-supported collaboration. TechnicalReport SMLI TR-92-5, SUN Microsystems Lab, 1992.

18. J. Tang and M. Rua. Montage Providing teleproximityfor distributed groups. In Proceedings of the Conference

on Computer Human Irueraction, 1994.

19. S. Whittaker, E. Geelhoed, and E. Robinson. Sharedworkspaces: How do they work and when are theyuseful? InternationalJournal of Man-Machine Studies,1993.

16. J. Tang. Findings from observationat studies of collab-orative work. International Journal of Man-MachineStudies, 34:143-160,1991.

137