Eavesdropping On Electronic Guidebooks: Observing Learning .

on her device, she always hears her own audio clip. If A isnot currently playing an audio clip, but her companion B is,then B’s audio clip can be heard on A’s device. In otherwords, audio clips are never mixed, and A’s device alwaysplays a personal clip (selected by A) in preference to aneavesdropped clip (selected by B). Audio playback on thepaired devices is synchronized; if A and B are bothlistening to their own clips and A’s clip ends first, A willthen hear the remainder of B’s clip as if it had “started inthe middle.” To control a device’s eavesdropping volume(i.e., the volume at which A hears B’s clips), the interfaceincludes three option buttons: “Off,” “Quiet” and “Loud”(Figure 1, top left). “Loud” is the same as the volume forpersonal clips.In abstract terms, eavesdropping provides a relativelysimple audio space model (Mackay, 1999). We didconsider other options, such as a telephony-like connectionmodel in which visitors would independently initiate andterminate audio sharing sessions with their companions.We also considered email-like asynchronous models inwhich visitors would send and receive audio clips at theirconvenience. We rejected more complex abstractions ccept/reject, etc.) because we believed that thenecessary interface gestures would distract visitors fromtheir experience with the environment and theircompanions. In the audio space model, sharing requires nogestures of its own. To “receive,” a visitor merely sets theeavesdropping volume. To “send,” a visitor simply selectsan object; playing a description has the side effect ofsharing it, if the companion chooses to eavesdrop. Theaudio space model has the further advantage that it supportssimultaneous listening, which enhances social interaction bycreating the feeling that the content is part of a sharedconversation (Woodruff, Szymanski et al., 2001).Audio delivery.Visitors hear descriptions throughheadsets. We conducted a small study (n 8) to identifyheadsets that would allow visitors to converse and thatvisitors would readily accept (Grinter & Woodruff, 2002).Based on this study, we chose commercial single-eartelephone headsets, locally modified by the removal of theboom microphone (Figure 1, right). This configurationleaves one ear available to hear sounds from the externalenvironment, and visitors find the over-the-head designdesirable because it is familiar and gives them the sense thatthe headset is securely attached.Audio content. The prototype contains descriptions of 51objects in three rooms of the house. In most regards, thedescriptions are recorded along principles described in(Woodruff, Szymanski et al., 2001). The audio clips varyin length between 5.5 and 27 seconds, with the exception ofone story that runs for 59 seconds. The clip length is muchshorter than conventional audio tour clips, which often runto 180 seconds, and is intended to facilitate conversation byproviding frequent opportunities for visitors to take aconversational turn.Since we use single-ear headsets, both personal andeavesdropped audio content are necessarily presented in thesame ear. We distinguish the two types of content usingtwo mechanisms. First, we apply a small amount ofreverberation to the eavesdropped audio.A singleearphone cannot effectively deliver spatialized audio(Blauert, 1997), but can support other sound effects; wechose reverberation after conducting user tests (n 6)involving scenario-based tasks using the guidebook.Second, the default eavesdropping volume (“Quiet”), whichis most frequently used by visitors, is softer than thepersonal volume.METHODWe have conducted three major user studies at Filoli, aGeorgian Revival historic house located in Woodside,California (http://www.filoli.org/). Study 1 used an earlierversion of Sotto Voce that supported open air audio,whereas Studies 2/3 used the current version of Sotto Vocethat supports eavesdropping as described in the designsection of this paper. Study 1 and Study 2 involvedpreviously recruited participants on days the house wasclosed to the general public, whereas Study 3 involved 47visitors recruited on-site on days the house was open to thegeneral public. (Again, these studies are summarized inTable 1.)Because the participants and procedures for Study 1 andStudy 2 have been reported previously, below we reportonly the participants and procedure for Study 3. We thendiscuss our analytic methods, which were the same in allstudies.Participants. In Study 3, we observed 20 pairs, one groupof three, and one group of four using the guidebooks.These pairs and larger groups were comprised of visitorswho had come to Filoli together, e.g., mother/daughter orfriend/friend pairs. The majority of visitors had notpreviously used a handheld device. The visitors covered awide range of ages: the youngest visitors were in the “1829” age range, and seven visitors who used the guidebookwere “over 70.” (While we had several children test SottoVoce in the first and second studies, visitors from the agesof approximately 5-17 are quite rare at Filoli unless they arevisiting with a school group.)Procedure. Visitors to the house were recruited at theentrance to the Library, the first room discussed in theguidebook. After signing consent forms, visitors were fittedwith a wireless microphone, given guidebooks, and trainedin their use. Next, they visited the three rooms for whichthe guidebook had content. When they finished using theguidebooks, they participated in a semi-structuredinterview.The visitors’ conversation and comments during theinterview were recorded using the wireless microphones,

the visitors were videotaped by fixed cameras while usingthe guidebooks (all visitors to the house were notified thatvideotaping was in progress), and the visitors’ use of theguidebooks was logged by the device.visitors then had the options of dis-engaging (resulting inindependent activity), remaining engaged in shared activity,or maintaining a nascent engagement in expectation ofsubsequent re-engagement (Szymanski, 1999).Visitors typically spent about 15 minutes using theelectronic guidebooks. Their participation in the study tookapproximately 30-45 minutes; no time limits were imposedduring any portion of the procedure.With open air audio, visitor interactions tended to focus onchoosing individual objects and coordinating with theircompanions to listen to the descriptions. This setup,repeated for each sequence, focused more attention oncoordination activity than seems necessary or desirable.However, the open air audio did afford the opportunity toparticipate in shared responses to the “story,” motivatingthe visitors to begin setup for another sequence.Analysis. We analyzed the data in several ways. Forexample, we transcribed and analyzed the interview data toexamine the visitors’ attitudes and feelings about thetechnology and their experience. The majority of thefindings presented in this paper are based on anothermethod we used, conversation analysis (Sacks, 1984).Conversation analysis is a sociological method used toexamine naturally occurring social interaction to revealorganized patterns. To find such patterns, conversationanalysts study collections of interactive encounters andidentify sequences of actions that were recurrently made bythe participants. Actions in our context might includemaking a verbal utterance, pointing at an object, orselecting a description.To this end, we create a composite video of visitors andtheir guidebook screens and audio (re-created from theguidebook activity logs). We then transcribe the actionstaken by visitors, including dialogue, and look for recurringpatterns to identify visitors’ systematic practices.FROM OPEN AIR TO EAVESDROPPING:CHANGES IN VISITOR BEHAVIORIn this section, we compare the behavior of the pairs whochose to use mutual eavesdropping in Studies 2/3 to that ofsimilarly engaged pairs who used open air audio in Study 1.Specifically, we discuss the structure of the visitors’interactions and their physical mobility. The effect of theseaspects can be identified in the visitors’ learning-relatedbehavior, which is the subject of the following section.Changed Activity StructureVisitor activity was structured very differently witheavesdropped audio than with open air audio. The newstructure had a lower coordination cost, demanding lessattention. The decreased attention burden was reflected inthe visitors’ interactions.In all of the studies, a single overall structure pervaded theinteractions. Specifically, they exhibited the sequential,multi-phase organization known as storytelling in theconversation analytic literature (Sacks, 1974); as part of thisorganization, visitors created a conversational role for theaudio descriptions, i.e., they treated the guidebook like a“third party” taking an extended conversational turn (Aokiet al., 2002; Woodruff, Szymanski et al., 2001). Pairedvisitors entered a state of engagement at the beginning of agiven storytelling sequence; levels of engagement generallyrose and then fell over the course of a given sequence; andBy contrast, participation in mutual eavesdropping createdan ongoing assumption that the couple would continue inthe shared activity. This supposition of continuing sharedactivity meant that setup tended to be cursory. Further,while open air audio was primarily conducive to follow-updiscussions that related directly to descriptions, mutuallyeavesdropped audio was conducive to many diverse typesof follow-up sequences such as discussion of objects notdescribed in the guidebook.The change in activity structure had at least two beneficialeffects. First, by reducing the effort needed to choose andlisten to descriptions, mutual eavesdropping freed visitorsto direct more attention to meaningful interactions withtheir environment and their companions (i.e., away from theguidebook and routine coordination). In other words, thereduction in low-quality coordination talk meant thatvisitors had more time to investigate the room and itscontents and that a higher proportion of talk tended to focuson topics of substance. Second, since the new activitystructure supported more diverse types of sequences,visitors were more likely to pursue new topics or investigateobjects not described in the guidebook.Increased MobilityVisitors in Studies 2/3 were noticeably more mobile duringperiods of engagement. In Study 1, the open air audio wasplayed at a low volume, so any movement that changed therelative position of the visitors could cause significantsound attenuation due to distance or blockage (e.g., due tointerposed obstacles – even changes in body orientationcould cause the audio to be blocked). As a result, couplestended to remain close together and stationary while sharingaudio descriptions. See Figure 2a, in which a grandmotheris bending over to listen to the audio description that hergranddaughter is playing of the portrait over the fireplace.Note how this position prevents her from examining thepainting while she listens. In Studies 2/3, visitors were l

(L. Mann, Antenna Audio, personal communication). We describe an alternative mechanism for sharing audio. This mechanism, which we call eavesdropping, preserves the social interaction enabled by open air audio while avoiding the audio “clutter” that open air audio necessarily entails. In our system, visitors independently select objects