A stroll with Carletto: adaptation in drama-based tours with virtual characters

A stroll with Carletto. Adaptation in drama-based, interactive

tours with virtual characters.

Rossana Damiano, Cristina Gena, Vincenzo Lombardo, FabrizioNunnari, Antonio PizzoCIRMA, Universita di Torino, Torino, ItalyVirtual Reality and Multi Media Park, Torino, ItalyE-mail: {{rossana,cgena,vincenzo,fnunnari}@di.unito.it}

Abstract.In this paper, we present an application for character-based, interactive guided

tours on mobile devices. The application is based on the DramaTour methodologyfor information presentation, which incorporates principles of drama authoring incharacter-based presentations. The application has been developed for an historicalsite and opened to the public. The presentation is performed by a virtual spider withan anthropomorphic aspect, “Carletto”, who engages in a dramatized interactionwith the user.

Content items are delivered in a location-aware fashion, relying on a wirelessnetwork infrastructure in the site, where the visitor can stroll freely. The selectionof contents keeps track of the interaction history, in order to deliver the appropriatequantity and type of informative items, and to manage the given/new distinction indiscourse. In order to enforce dramatization effects, the communicative strategies ofthe character are designed to keep it believable along the interaction with the user.The behavior units of the characters are tagged with metadata concerning content,communicative functions and interactional functions.

The description of the application is accompanied by an evaluation study basedon a random sample of about 300 users, carried out in April 2006 with real visitors.

Keywords: interactive drama, virtual characters, cultural heritage, mobile com-puting

1. Introduction

In this paper, we present an application for character-based, interac-tive guided tours on mobile devices. The application, developed for anhistorical site, is enacted by a virtual character, Carletto the spider,who engages in a dramatized interaction with the visitor.

The application has brought the DramaTour methodology for in-formation presentation (Damiano et al., 2006) to a real context withhundreds of visitors in an open-to-the-public installation. The Drama-Tour methodology incorporates principles of drama in character-basedpresentations. Drama is the combination of two main features: the dis-play of actions performed by characters while experiencing a relevantconflict related to their emotional values; drama typically exhibits a

umuai-personalization.tex; 9/05/2007; 2:36; p.1

2 Damiano et al.

Figure 1. The scenario of “Carletto the spider”.

movement, or direction, towards a solution of this conflict (Aristotle,1920). In the DramaTour methodology, dramatization emerges alongthe presentation as a result of the interaction with the user, thus adher-ing to the paradigm of interactive drama (Mateas, 2002). So, the systemmust be able to adapt to the behavior of the user, while maintainingthe typical development of drama.

The major novelty of the DramaTour methodology is the applica-tion of dramatization to the presentation of information by explicitlydriving the interactional and communicative strategies of the virtualcharacter who enacts the presentation. The working assumption is thata character, who acts in first person and shares the user’s present timeand space, yields a powerful effect of physical and emotional presence.The emotional conflict concerning the character, intrinsic to the no-tion of drama, improves the user’s engagement, bringing an affectiveconnotation to the her/his experience (Picard, 1997). Dramatic media,like television or cinema, add emotional qualities to the content theyconvey, for the very same reason of displaying live action (Esslin, 1987),and improve the effectiveness of content reception (Laurel, 1993). Thedirectional movement of drama depends on a continuous exchange be-tween the presentation carried out by the character and the responseof the user, who manifests different degrees of interest for the topicsaddressed by presentation through the input she/he provides to thesystem.

“Carletto the spider”, the application described in this paper (seeFigure 1), adapts to the behavior of the user in many ways. It relieson a wireless network to track the position of the visitor, and adaptsthe presentation to her/his current location. The visitor is free to strollinside the apartments; as she/he moves from a room to another, Car-

A stroll with Carletto. 3

Figure 2. “Carletto the spider”.

letto follows her/him as if he were observing her/him from a webcam,and modulating its presentation according to the topics concerning thenew space.

Before the actual visit starts, Carletto provides the visitor with somegeneral facts about himself (mainly his conflicts) and the location; thenstarts “following” the visitor in the rooms, providing information fromgeneral to specific and carrying on the conflicts that characterize him;such conflicts are likely to move toward a solution by the end of thepresentation. In every moment, Carletto remains aware of the sociallyrelevant steps in the interaction, and gives priority to them: for exam-ple, he welcomes his guests at the beginning of the presentation andapologizes when he makes mistakes in the exposition. Like a humanguide, Carletto does not let the visitor take the complete control of thevisit, but mixes the adaptation to the preferences of the visitor with ahint of initiative: from time to time, Carletto may invite the visitor tomove to a different room, if the visitor has been too long in the currentone. Sometimes, he may invite the visitor to move toward the centerof the room to regain the network signal that has been lost. Thanksto the techniques of dramatic writing employed in the application, allthese interactional aspects are accounted for in a way that is consistentwith the personality of Carletto.

The exposition delivered by Carletto is never flat or monotonous:beside talking, he moves across the screen of the device, interleaving thepresentation with some of the activities that characterize him (makingthe web, polishing his medals, etc. - see Figure 2). By doing so, he

4 Damiano et al.

entertains the visitor, contemporarily giving her/him time to orienteerin the room and assimilate the load of information received so far.

In this paper, we describe the Dramatour methodology, both under-lying rationale and system architecture, and the application “Carlettothe spider”, together with an evaluation of the approach on the usersof an open-to-the-public installation. In particular, we focus on twoelements: the process of interaction design, that adapts the presentationto the context determined by user’s input and a set of relevant exter-nal factors, and the authoring process, that covers both the dramaticaspects of the presentation and its content. The paper is organized asfollows: in the next section, we introduce the DramaTour methodology;then, we describe the application “Carletto the spider” by instantiatingthe generic issues posited by the methodology; then, we present anevaluation of the system in a real setting; finally, we discuss the resultsof the evaluation in the context of the interaction design, by focusingon adaptation aspects.

2. The DramaTour Methodology

The DramaTour methodology provides a framework for conceptuallydefining the character’s behavior through a factorization into seman-tically tagged elementary units, and a software architecture that gen-erates the presentation by assembling these units in real time. “First-person, present-time” action, as well as directionality of conflicts – bothrequired by the drama paradigm – are achieved through the authoringof the elementary units and the design and animation of the character,which includes its personality, values and conflicts (McKee, 1997; Seger,1990). Consequently, the methodology requires the author to create thepresentation by thinking in terms of elementary units that the systemwill subsequently shape into a coherent dramatic direction along theinteraction with the user.

The units of the character’s behavior (called behavior units) arescripted by the author (the writer proper) off-line and then fed to themultimedia production (that includes actor, animators, director) toyield the “audiovisual” units: the latter include the character’s utter-ances and gestures, the setting realization, the proper editing of audioand video tracks from the bare animation clips. Behavior units arethe atomic, fully-scripted, items of the presentation. They are taggedwith meta-data encoding their audio-visual features, their informa-tive content and their interactional functions. When the system isin execution, meta-data are exploited to assemble the behavior unitsaccording to the interaction strategy encoded in the interaction script,

that also accounts for the user’s input. Besides allowing flexible, on-the-fly, editing, meta-data are also crucial for content reuse across mediaor applications, provided a coherent semantic tagging.

The DramaTour methodology views the whole presentation as therealization of some communicative function (Austin, 1962; Searle, 1969).In line with Grice’s principle of cooperation (Grice, 1975), beside theinformative task itself, the character must address the social and in-teractional issues raised by the user during the presentation. All ofthe communicative functions are hierarchically organized, and prioritiesimpose a partial order over communicative functions (for example, thecharacter must introduce itself before providing any information). Com-municative functions are realized by behavior units: each behavior unitcorresponds to a communicative function, specified in the interactionalmeta-data that accompany it. Among the communicative functionsadmitted by the DramaTour methodology, the informative function isthe sole responsible for the advancement of the content presentation,and is tagged by semantic meta-data. The semantic specification ofan informative behavior unit is provided by relating its content to aseparate ontological representation of the presentation domain.

The interaction script is a “score” that describes the character’sstrategy in terms of communicative functions, and guarantees thatthe presentation acquires a dramatic directionality. It determines theadaptation to the user by allowing the author to specify how the presen-tation must be oriented in dependance of the interaction. For example,at each decision step, the system selects the communicative functionsand the informative content to be delivered by taking into account thevisitor’s location, the current advancement, and the time elapsed fromthe beginning.

In fact, differently from traditional authoring, the author of a Dram-aTour application – like the one described in this paper – is to beconsidered a “procedural author” (Murray, 1998). The notion of pro-cedural authoring encompasses the task of writing content for drama-based, interactive applications, in which the audience becomes part ofthe process of dramatic achievement. An important skill of the pro-cedural author consists in the ability of constraining the outcomes ofthe system-audience interaction within the boundaries prescribed bythe drama paradigm. For example, the system keeps the drama ad-vancement coherent even if the visitor strolls in the apartment withoutapparently caring of the information provided and changes locationquickly. Moreover, the author must tune the length of the behaviorunits (with their atomic, not further divisible, nature) to the rate ofthe user’s interactions: in fact, the system reaction can only occurat the unit boundaries, though the user may have introduced some

6 Damiano et al.

relevant inputs in the meantime. The author creates a tension alongthe presentation, by exploiting the alternation among different com-municative functions. If the strategy is carefully designed, this tensionshould evolve into a solution by the end of the presentation, followinga “dramatic arc”, a reversed U-shaped development of conflicts thatcharacterize drama (Freytag, 1883).

2.1. System Architecture

The system is structured according to a client-server schema. Thisschema is suitable for a number of possible deployments, given that thenotions of location awareness and content dimension (the informationto be presented) are present. In fact, we have applied the same schemato on-site tours, with mobile devices connected to the server through awireless network, and virtual tours, in which visitors explore a virtualreplica of the location by using a point+click map interface on the webbrowser and connect to the server through the internet.

The server executes a decision-execution-sensing loop: decision ad-dresses the selection of the next behavior unit; execution concerns thedelivery of the selected unit; sensing concerns the processing of newuser input. Although the functioning of the server is inspired by thebelief-desire-intention (BDI) model of rational agency (Bratman, 1987),according to which an agent forms intentions based on its beliefs andgoals and becomes committed to them, the system does not incorporatea full-fledged BDI agent. In particular, since the DramaTour paradigmonly requires a simplified, limited interaction, the system does notrepresent goals and intentions explicitly, and does not monitor theeffects of actions, differently from what prescribed by the propertiesof commitment (Cohen and Levesque, 1990).

The intelligent core of the server, which is responsible for the decisionphase, has a modular structure: the interaction manager accounts forthe interaction with the user, the presentation manager accounts forthe content organization, the delivery manager handles the ultimatedelivery of the presentation to the user in a well-edited, audiovisualcontinuum. A sketch of the control and data flow among these modulesis visible in Figure 3. In the following, we describe the architecture ata general level: since the representation of meta-data is crucial to thefunctioning of the system, we provide the details of each module inSections 2.2.1-2.2.3. The deployment of this architecture in the appli-cation “Carletto the spider”, together with a punctual description ofthe underlying interaction strategy, is provided in Section 3.

The loop executed by the visit server is the following:

1. Sensing. The Interaction Manager gathers the input from the user.In the DramaTour methodology, the changes in the user’s locationare actively monitored by the system to identify the appropri-ate content to be delivered and the depth of presentation, whichdepends on the duration of the permanence in a given location.

Some asynchronous inputs generated by the actions allowed by theinterface, like pausing and restarting the presentation, do not affectthe interaction strategy (they are only relevant for the client-serverprotocol), but require an immediate reaction. So, cause the systemto exit the loop to execute hard-wired behaviors.

2. Decision. User’s behavior may raise some exceptions that deserveimmediate consideration by the system. For example, the user mayunexpectedly switch off the device, requiring the system to say ahurried goodbye and exit, or she/he may step into an area that isnot covered by the localization system, requiring the system to helpthe visitor regain a monitored location. If an exception has beenraised, the Interaction Manager schedules an appropriate reactionby consulting a list of exception-reaction pairs (again handled atthe meta-data level).

If no exception has been raised, the Interaction Manager determinesthe next communicative function. The selection of the communica-tive function depends on the interaction script and on the user’sinput: the interaction script defines the next communicative func-tion given the interaction history and the user input. In case thenext function is the informative one, the interaction script alsoestablishes the semantic dimension according to which content mustbe selected.

− If the selected function is the informative one, the selection ofthe next behavior unit passes onto the Presentation Manager.The Presentation Manager selects an informative unit whosetopic is appropriate to the current location and to the currentsemantic dimension. If more than one behavior units are ap-propriate, the Presentation Manager randomly selects one ofthem.

− If the selected function belongs to any other types of commu-nicative functions, the Interaction Manager randomly choosesone behavior unit that realizes the selected communicativefunctions.

3. After a behavior unit has been selected, the Delivery Managerconcatenates it with the last played behavior unit and the next

8 Damiano et al.

META DATA

DELIVERY

MANAGER

behavior units

interaction

script

interaction

history

transition units

interactional

MANAGER

user input

editing

INTERACTION

MANAGER

transition unit + behavior unit

Figure 3. The system architecture according to the DramaTour methodology. Solidlines represent control flow, dashed lines represent data flow.

behavior unit, possibly interposing a transition unit (see below) toguarantee the visual continuity or, if not possible, to minimize thediscrepancies between the two units (cross-fading is applied in allother cases).

4. Execution. Finally, the selected unit is sent to the user device;when the commit to play it is confirmed, the interaction historystorage is updated with the pair 〈user’s input, selected behaviorunit〉, and this pair will contribute to the future decisions.

The Interaction Manager informs its decision process with the in-teraction script, the overall strategy that regulates the behavior of thecharacter along the whole presentation. The interaction script has theform of a hierarchical plan (Erol et al., 1994), in which tasks decomposeinto subtasks, and the same task can be realized through a set of alter-natives depending on a condition. The hierarchical plan representing apresentation strategy takes the form of a tree: the root represents thewhole presentation, the leaves represent the communicative functionsprovided by the methodology, directly realized by the behavior units,and the intermediate levels represent the prototypical sequences of thepresentation, seen at different levels of abstraction. The theoretical ad-

vancements achieved in recent years within the paradigm of hierarchicalplanning (D. Nau, H. Muoz-Avila, Y. Cao, A. Lotem and Mitchell,2001), together with the availability of practical tools, has promotedits exploitation in interactive applications (Cavazza et al., 2002; Hoanget al., 2005). From the authoring point of view, hierarchical planningprovides a natural paradigm to conceptualize the organization of thepresentation task into subtasks; interactivity can been accounted forby means of conditional tasks, although this choice makes the systemless flexible with respect to genuine reactive planning. For example, thepresentation may be organized into an opening section, in which socialprotocols are accounted for, a body consisting of purely informativecontent, and a closing section. By doing so, the author can design theinteraction as a set of macro-constituents, which can be combined andre-ordered at an abstract level, without taking the realization detailsinto account.

In order to account for the variability of the user’s behavior andto make the presentation adaptive, abstract tasks specify into a set ofalternatives, conditioned to the behavior of the user. The latter consistsof current input and specific patterns or occurrence frequencies in theinteraction history (user modeling approach). For example, the type ofinformation to be delivered may be tied to the user’s current locationor to the changes observed in the user’s location (in other words, to theuser’s path in the visit site).

Conditional operators can be exploited to implement sophisticatepresentation schemata. By using counters, it is possible to accountfor the ratio among communicative function types. For example, theproportion between social and informative communicative functionsmay be predetermined by the software engineer in observance of thedramatic writing rules. Temporal deadlines can contribute to align therealization of the desired ratio with the advancement of the presen-tation, so that the alternation between two or more communicativefunction types evolves along the presentation according to the notionof dramatic arc. Finally, concerning informative aspects, conditionaloperators can be exploited to constrain the quantity of informationdelivered by the system, or to set some qualitative aspects of theinformation provided, like the degree of specificity.

2.2. Data Framework

The units that factorize the behavior of the virtual character form theknowledge base of the system. They contain multimedia content (anaudiovisual clip with character animation and sound) and are taggedwith the information that the system uses to generate the interactional

10 Damiano et al.

Meta-data sets Type Values

Tuple of ontological references {<O1,r1>,...,<On,rn>}

TOPICTuple of flags

f {given, new, null} {<O1,f1>,...,<On,fn>}

Presentational

Social

DirectiveCOMMUNICATIVE Communicative function

Interactional

Phatic

Shot type

ObjectsEDITING Audio-visual features

Figure 4. The sets of metadata in the DramaTour methodology. Each behavior unitof the system is tagged with meta-data belonging to all the three sets.

and presentational behavior of the character. Meta-data are dividedinto three sets: communicative, i.e. the communicative function ac-complished by the unit, topic, i.e. the description of the informativecontent of the unit, editing, i.e. the information needed for assemblingthe audiovisual clip with the adjacent ones to obtain fluency. A softwaretool to support the authoring process has been implemented with aninterface to the database of behavior units. For each behavior unit, theauthor can choose the appropriate tagging for all the sets of metadata(see upper right corner of Figure 3).

2.2.1. Communicative MetadataEach behavior unit accomplishes a communicative function. It is upto the author to make sure that at least one unit matches each of thecommunicative functions advocated by the interaction script. More-over, each unit realizes only one communicative function. The sys-tem assumes that behavior units are self-contained, i.e. each of themaccomplishes by itself some communicative function. Communicativefunctions can be informative or interactional.

− The informative function, i.e., the task of providing the user withuseful and relevant information during the visit, is the primarytask of the system. The execution of this function is assigned to thepresentation manager, which handles the selection and the organi-zation of the conveyed content. As described below, the behaviorunits that realize the informative function are labeled according totheir topic, to allow for an order of exposition that complies withrules of topic management in discourse, and is at the same timecompliant with the changes determined by the user’s behavior.

− The interactional function is subdivided into social, directive andphatic functions, respectively.

• Since the virtual character must qualify itself as a socialagent, in order to gain believability and obtain the user en-gagement, the character must perform some basic social be-haviors, like greeting the user or introducing itself.

• The directive function includes all the actions that the char-acter performs in the attempt to “direct” the user behaviorin some way, like signalling conditions that may require theuser to perform some action (for example, executing actionsrequired for the continuation of the visit). In order to in-duce the issue of personification and autonomy of the virtualcharacter, the directive function, which has no priority oversocial aspects, has priority over the informative function: bydefinition, the occasions in which the character tries to directthe behavior of the user are those strictly required either forthe accomplishment of the drama paradigm or for the devicemaintenance.

• The phatic function (Jackobson, 1990) is activated when allother functions are unapplicable. Its purpose is to signal theuser the character’s presence and will to receive input. For ex-ample, it may be activated when the character has requestedthe user to perform an action that is necessary for the visitcontinuation.

2.2.2. Topic MetadataThe topic section of the meta-data contains the description of the infor-mative content of the units. The informative content is classified withrespect to an ontological representation of the domain that constitutesthe object of the presentation. The presentation manager relies on thisontological representation to select the content to be conveyed to theuser and structure it in a coherent way. The general strategy follows thediscourse-oriented focussing rules stated by Grosz and Sidner (Groszand Sidner, 1986), here adapted to a presentation task. To supportthis approach, domain description is encoded in a hierarchical form,in which the decomposition of the presentation task is mapped ontotopic-subtopic relations. So, the methodology only admits taxonomicrelations (is-a) and mereological relations (part-of). For example, incase of a location-based application, the implementation of a location-aware adaptive presentation requires the realization of an ontologydescribing the topology of the location, with the informative behavior

12 Damiano et al.

units tagged according to their position in this ontology. Also, locationsand subareas of locations are useful for organizing the presentationof the informative content according to the topological knowledge ofthe unit, by taking into account the items contained therein and thelocation-specific historical facts.

The focusing rules incorporated in the Presentation Manager are thefollowing:

1. Maintain focus on the current topic. For example, by describing thedomain according to mereological ontology, the current focus maybe a piece of furniture - located somewhere in an historical location.Following a biographic description of the domain, the current focusmay be an artist whose works are exhibited in a museum.

2. Move focus to a sub-topic of the current topic. With reference tothe previous example, move to a subpart, a detail of the piece offurniture, or move to a certain period of the life of the artist.

3. Move focus to another sub-topic of the current topic. Again, thenew focus may be a different detail of the previously focused detailof the piece of furniture, or a later period of the artist’s life (in thiscase subtopics are visited chronologically).

Following sub-topic relations in an ontology according to the fo-cussing heuristics corresponds to structuring the presentation along acertain dimension of meaning. Since the ontology is hierarchical, thefocussing heuristics determine a depth-first visit of the ontology.

In principle, several ontologies may be predisposed by the authorto structure the same domain, corresponding to different presentationstyles. For example, the facts about an historical site may be “de-scribed” according to a topological dimension or “narrated” followinga chronological dimension. In order to enforce the dramatization prin-ciple incorporated in the system design, the DramaTour methodologyrequires the author to encode domain knowledge according to at leasttwo different ontologies. This requirement serves the function of estab-lishing a dialectics between presentation styles, thus realizing a conflictin the way the character accomplishes its presentation task.

In order to avoid abrupt transitions over presentation styles, thedomain ontologies to which the presentation refers should not be com-pletely unrelated. For this reason, the topic meta-data of a behaviorunit is expressed by encoding its position on all the available ontolo-gies1.

1 If the topic of a behavior unit is not present in one of the domain ontologies,the value element for that ontology refers to the root of the ontology, meaning thatit may equivalently subsume any specific topic in that ontology.

Figure 5. An excerpt of the ontology of historical facts defined for “Carletto thespider”, as it appears in a screen shot of the authoring utility. In this example,concepts correspond to the names of the historical characters of the Savoy family. Inthe authoring utility, behavior units are tagged with one item of the topic meta-dataontology (see the checked square), then inserted by the utility in the representationformat in Figure 4.

The system design exploits the topic description to apply the fo-cusing heuristics on all the available ontologies: at each moment, onecurrent ontology drives the presentation, determining the active pre-sentation style. When a set of behavior units match the active topicon the current ontology, they are ordered according to a preferenceorder specified by the author off line. In this way, when a style must beabandoned because of conflict-based dramatic presentation, the currentontology changes, but the transition to the new one is smooth in mostcases.

Since the exact presentation order cannot be predicted in advance,each informative behavior unit is further tagged according to a given/newdistinction of the discourse referents mentioned therein (Hajicova et al.,1998). Each unit is tagged with a tuple of referents marked as givenor new according to how they are instantiated in the unit discoursestructure: a referent introduced by the unit is marked “new”; a referentassumed as already introduced in some unit is marked “given”. Thesystem maintains an history of referents introduced. When a unit to

14 Damiano et al.

be delivered contains a “given” referent, this must already be in thereferents’ history; vice versa an item already in the referents’ historycannot be introduced again by a unit in which it is marked as “new”.The methodology admits that a null value is employed in the given/newdistinction: units containing items that are not marked can be selectedat any time without restrictions.

2.2.3. Editing MetadataEditing features connect behavior units, possibly interposing an audio-visual segment (called a transition Unit), with the aim of obtainingvisual fluency (Biral et al., 2003). The system incorporates a set ofediting rules, that implement a number of editing techniques and spatialcontinuity. Transition units are short units (1–5 second long) with onlyacting content that matches the end of some behavior unit and thestart of another behavior unit. Transition units are selected by thedelivery manager from a repository to satisfy the editing constraintsover behavior units.

The editing rules are derived from the standard practice of continu-ity editing. Continuity editing is “a dominant editing style throughoutwestern film history”, used to “ensure narrative continuity” by meansof a smooth flow from shot to shot in order to convey the illusion ofspatial and temporal continuity (Bordwell and Thompson, 2006; Reisz,1968). A number of parameters can be used to describe continuityediting techniques: for example, the visual elements (e.g. framing, mise-en-scene, props coherence) are kept roughly continuous between shots,while the cut-on-the-action rule takes advantage of the fact that editing,when motivated by narrative reasons as in the case of a dynamic scene,passes unnoticed by the viewer. Spatial continuity is guaranteed by the180 axis rule that ensures consistent screen direction, i.e. in the caseof a character walking left to right, her/his path constitutes the axisof action and only shots filmed on one side of the axis are acceptable,since crossing the axis would result in the character walking right toleft.

The PROPS coherence (a prop is any movable articles or objectsused in a movie set) accounts for the satisfaction of the continuitybetween two frames, that includes cinematographic qualities and thewhole of mise-en-scene (lighting, setting, costume, behavior of figuresand presence or absence of props). For example, if a character wearsa hat in one unit, it must wear the same hat in the next unit, unlesswe interpose a transition unit where the character removes the hat.The DISTANCE parameter is concerned with the type of shot, i.e.distance of framing depending on the camera position. Six main shottypes are considered: extreme close-up, close up, medium shot, three-

quarter shot, long shot and extreme long shot. Generally speaking, itis regarded as viable and even cinematographically appreciable cuttingfrom one shot-type to a different one. However, a straight cut betweena long shot and a close-up, given the abysmal magnification order, canbe disturbing for the viewer and is thus to be avoided or at least it isnormally avoided in the editing practice. It is equally important to avoidcutting between shots that are too similar, a thing that is perceived asupsetting by viewers. Details of the automatic editing process we usein the DramaTour methodology can be found in (Biral et al., 2003).

3. Carletto the spider

The application “Carletto the spider”, based on the DramaTour method-ology, has been developed for an historical residence in Torino, Italy.The virtual character is an anthropomorphized spider, Carletto, whichappears on the screens of portable devices that visitors carry with themduring the visit. The metaphor is that Carletto lives hidden on thewalls and ceilings of the apartment and connects to the visitors’ devicesthrough a webcam.

The presentation given by Carletto was written by a script writer,with the help of an expert in the historical and artistic aspects of thelocation. By assuming an average duration of the visit estimated in25 minutes, behavior units were produced for an overall duration of50 minutes. Due to random aspects in the presentation design, andto subjective variability in user behavior, this duration was needed toensure partially different, personalized presentations for all users.

The tour takes place in the former royal apartments, situated on thefirst floor of the palace, and includes only a subset of five rooms: theGuardroom, the Room of the Valets, the Reception Room, the DiningRoom, and the Room of the Tapestries. Visitors are free to stroll inthe apartment; however, the first three rooms are pipelined, so theyencouraged a sequential visit (see the plan of the apartments in Figure7). Rooms included in the visit are identified by large signs with thepicture of Carletto posited in a visible corner of the room.

3.1. character Design

The creation of a believable character, who will accompany the visitorin first person through the presentation, is crucial to the effectivenessof the presentation. Within the DramaTour framework, the characterfunctions as a mediator between high culture and the visitor, actingmore as a quality-improving interface rather than a simple contents

16 Damiano et al.

supplier. The goal of character design is to merge dramatization andcultural mediation to show an intention of providing information, andto display changes during the interaction. The character has to avoidto look like a know-it-all, or to provide an overwhelming mass of infor-mation. The use of a comic attitude is particularly appropriate to thetask of cultural mediation, since it reduces the distance between thesubject of the presentation and the visitor. At the same time, a comicattitude is well accepted by the majority of the users, and contributesto the generation of positive emotions during the interaction.

To implement the first person live action paradigm mentioned above,the character must show a space-time consistency (i.e. it must berelated to the current space and time, avoiding the “ghost-from-the-past” effect). The author has to design both his/her explicit-externalmotivations (the character must show a will to tell a story, a senseof duty that forces him/her to be there with the visitor) and his/herimplicit-inner motivation (the character must show a personality con-sistent with her/his wills and duties) In other words, the character hasto have an official reason to be there and, in the same time, has to havea personality that helps him/her to carry out the duty properly (e.g.if she/he is a museum guide, better to create a personality interested -in whatever sense - in the venue, in the history and happy to show it).

According to the paradigm of drama, the character has to showsome conflicts. Although literature state different kinds of conflicts(McKee, 1997), the DramaTour methodology suggests to avoid anydeep inner conflict, in favor of an external or interpersonal conflict,easier to design and better matched with the comic attitude mentionedabove. Conflict is here intended as a tool to help the user to build abound with the character, hence the author has to avoid to design anyconflict between the user and the character. Conflict must be explicit,and must recur across the whole presentation in different forms in orderto be recognized as such by the user.

The character of “Carletto the spider” has been designed by a dramaexpert, and selected among a pool of three competitors by a focusgroup formed by a museum curator, a producer, a 3D artist and aHCI expert. The competitors of Carletto were all anthropomorphiccharacters (a female archivist and a male historian of the art); thechoice of Carletto was mainly due to its zoomorphic identity, that wasassumed to overcome some of the interpersonal and social boundariesthat separate visitors from human experts.

Carletto is a teenage spider whose family has inhabited the palacefrom ages. Carletto not only knows the history of the palace in detail,but knows a lot of funny anecdotes about the people who have livedthere through the centuries, and is striving to tell them to the visitors.

The conflict between the role of an “audio guide”, who exposes factsorderly and plainly according to the topology of the location, and thedesire to recount all the trivia and the anecdotes he knows from anhistorical perspective - most of which see him or his family personallyinvolved - meet the methodology guideline of centering the presentationon an internal conflict of the character to gain the attentional andthe emotional engagement of the users. Following the author guidelineaccording to which the character itself must be carefully dramatizedin the behavior units, Carletto engages in a continuous fight with thejanitors, who would like to kick him out of the place.

3.2. Interaction Design and Adaptation

Although “Carletto the spider” has been endowed by the author witha personality of its own, its communicative strategy takes into accountthe relevant metaphor of museum guide. As long as the user’s behavioris coherent with such a metaphor, Carletto keeps the control of theinteraction, politely directing the user’s attention to relevant aspectsof the environment and communicating with her/him in the dramatizedstyle that characterizes it.

The initial design of Carletto’s interaction with the visitor consistedof a simple directive strategy, in which the user was not assignedany control, and feedback consisted of Carletto’s audiovisual behav-ior. During the implementation, a simple “Wizard of Oz” study wascarried out to evaluate the cognitive load imposed by a so engaginginteraction. This experiment allowed tuning the duration of behaviorunits, the density and timing of the content exposition. Finally, beforethe opening, a set of pilot studies were run with real users simulatingthe visit in real setting. 22 subjects were involved in this preliminarytests, and from their visit experience and feedback the interface waspartially re-designed: the “play/resume” and the “stop” buttons wereadded to give to the user more control on the system; the name ofthe current room was displayed both on the screen and in the roomsto provide more feedback about localization; the quantity of contentsdelivered per room was decreased.

The visit is structured along a topological presentation style, i.e.,the palace is seen as a set of rooms; this presentation style is basedon a specific ontology representing the topology of the location. Inorder to obtain an alternation between presentation styles, a differentontology has been predisposed, where information is structured as a setof temporally-ordered historical facts and characters. While in the topo-logical ontology the domain is modeled according to inclusion relationsbetween areas and sub-areas (the apartment is a set of rooms, which are

18 Damiano et al.

a set of walls, etc.), the historical ontology is structured according toa relation of inclusion over historical periods. The alternation betweendifferent presentation styles is a means to model the internal conflictof Carletto between the role of a museum guide (see previous section),who strictly follows a topological order, and the role of inhabitant ofthe place, who knows its history and is personally involved in it.

The system starts by enforcing the greetings by Carletto, i.e., itplays audiovisual clips where Carletto welcomes the visitors and intro-duces himself. Then, Carletto starts talking about the room in whichthe visitor is currently situated; occasionally, if the user moves fromthe entrance room while the introductive clips are played, the firstdescribed room may not be the entrance room.

Each time the user enters a room, Carletto starts (or resumes) thepresentation of the objects (furniture, artworks) in the room, accordingto an order that reflects the topology of the room. When a certainamount (80%) of the room sub-topics have been illustrated Carlettoswitches to an anecdotic presentation style for a while (based on anontology of historical facts), then gently starts inviting the user to anew room (directive). If the user does not move from the current room,Carletto activates the phatic function, by playing funny games andgazing to the user from time to time.

The idea of delivering the 80% of the whole material about eachitem along a certain ontological dimension pervades the whole archi-tecture. For each subtopic of a room, the system delivers 80% of thematerial about the subtopic, then performs a digression along the his-torical ontology. For each return to a room, the system delivers 50% ofthe remaining material to be delivered: in this way, it sees to it thatsomething relevant is left to be said about the room.

In case of missing localization, Carletto switches to the historicalpresentation style, then, after an informative behavior unit, he gentlystarts inviting the user to regain the connection by moving. Again, ifthe user does not change location, Carletto finally activates the phaticfunction. The system exits when the visitor has walked through morethan half of the rooms of the visits and has regained the exit door.

The interaction script described so far is represented as a hierarchi-cal, conditional plan in Figure 6. For simplicity, the parameters of taskoperators are not specified; conditions on the selection of alternativesare not represented either. Dashed arcs represent abstract tasks whichspecify into a set of alternatives, while solid lines represent complextasks representing sequences of simpler subtasks. The leaves of the plantree consist of primitive actions – corresponding to the communicativefunction types prescribed by the DramaTour methodology.

Make-presentation

Manage-presentationManage-exceptions

Consult-

Interaction-

History

Describe-room Social-activity

Detect-room Select-CF

Manage-deadlines

Select-Ontology

Select -

Directive

Select-

PhaticSelect -Historical

Select -

Informative

Missing-Network

Location-Unknown

Update-Interaction-

History

Update-

Manage-Interaction

Update-deadlines

Update-

Figure 6. The hierarchical plan representing the interaction script of “Carlettothe spider” (in boldface). Dashed lines represent alternative ways of accomplish-ing a task, solid lines represent sequences of subtasks through which a task isaccomplished.

For the sake of completeness, the figure also represents, at the higherlevels of abstraction of the plan, the loop of the interaction manager de-scribed in Section 2.1 (see the sequence of the manage-exceptions andmanage-presentation abstract tasks, and the subsequence consist-ing of the consult-interaction-history, manage-interaction andupdate-interaction-history tasks, entirely devoted to the mainte-nance of the interaction history).

The main distinction between social and informative behavior is rep-resented by the choice between the describe-room and the social-activityoperators. The first is directly realized by performing final or initialgreetings depending on the interaction history, while the latter is a com-plex operator that encompasses the task of detecting the current room(detect-room) and the presentation of the room itself (describe-room).

Once the current room has been detected, the threshold measur-ing the amount of information delivered about the current room istested, and the appropriate behavior is selected (manage-deadlines).As described above, this threshold, set to 80% the first time the visitorenters the room, varies to 50% the subsequent times. If the amount

20 Damiano et al.

of delivered content is above the threshold, the directive function isactivated (selective-directive). A second threshold regulates thetransition to the phatic function (select-phatic). If no threshold hasbeen reached, the informative function is selected; the most relevantaspect of the selection consists of the selection between the historicaland topological line of presentation (select-ontology). Again, thedecision depends on the interaction history: the topological ontologyis the default (select-topological), while the historical ontology isselected after the content threshold on the topological dimension hasbeen reached (select-historical). Finally, after the execution of themanage-interaction operator, the interaction history is update with thelast-selected behavior unit (new communicative function type, topic,user input - if any - are appended to the interaction history).

3.3. Cross-media Implementation

The current implementation is based on common hardware availableon the consumer market and mostly on open-source software. Theserver, that follows the specifications described in Section 2.1, is imple-mented in Java (http://java.sun.com), while the data base system ismySQL (http://www.mysql.com/). The web-based authoring interfacehas been developed in PHP (http://www.php.net/).

The input to the visit server is given by a client message, togetherwith the interaction history. When the server receives a message, itfirst decides what sequence of units to present next, then sets up thesequence for the delivery process and sends it to the client; finally, itstarts waiting from new input from the client. When the client contactsthe server for the first time, the server releases an ID to identify theinteraction, and creates a new interaction history, initially empty. No-tice that, since client and server communicate synchronously, any userinput provided to the system during the interval between each exchangeof client request and server response will be handled only when thesubsequent exchange is initiated by the client. This delay, determinedby the aim of reducing the complexity of the client-server interaction,makes the performance of the character more stable and guarantees therespect of the dramatic continuity designed by the author.

The PDA client is written in Java and runs on an ASUS A636 PDA(PocketPC series). The video clips implementing the behavior units areencoded in Macromedia Shockwave Flash (http://www.macromedia.com).The client pilots the media player by sending text-based commands tolocalhost. The PDA also supports a localization client, that providesthe user’s current location, i.e., it identifies the room is which the useris currently situated. Built-in controls of the PDA allow the user to

stop, pause and resume the presentation; so, they do not have anyconsequences in terms of the client server schema, with the exceptionof the stop command, that causes the interaction to be copied to thesystem log and destroyed on the server.

Since the DramaTour methodology is media-independent, beside thePDA-based version of the virtual tour, a web-based guided virtual tourof the same location has been developed by using the same visit server.By mapping the topological organization of the guided tour onto theelements of the web interface, the same server has been exploited forthe two media. An important advantage of this approach is that theuser who performs the two modalities of visit experiences the sameapproach, so that the web-based virtual tour can be either a preparationfor the actual tour or a prosecution of it. In fact, the two applications,beside sharing the same server, also share the same data. From theauthor’s point of view, not only the same methodology is enforced byboth applications - with an advantage for scriptwriting and meta-dataediting - but, since data are reused, the amount of work required bysetting up a version of the tour for a different media than the PDA issignificantly reduced.

The web interface simulates the tour in the virtual space, by propos-ing to the user a virtual representation of each room accompanied bythe sequence of clips in which Carletto provides information about theroom. The visitor, in the virtual tour as well as in the actual visit,is free to decide the order in which she/he wants to visit the rooms.However, it is important to stress that the presentation delivered to theuser may vary even if the user follows the same order of visit, thanksto the non-deterministic aspects incorporated in the architecture of thesystem.

Setting up the web-based version of the guided tour consists mainlyin the tasks of designing and implementing the web interface and thevirtual reality equivalent of the visit.

The layout of the interface is divided into two columns, as shownin Figure 7. The left column contains control menus, while the rightcolumn contains the clip player (at the top of the column) and therepresentation of the currently selected room (at the bottom of the col-umn). The right column occupies the two-thirds of the browser window;the clips and the room representation, in fact, constitute the primaryinformation delivered to the user through this interface.

In order to enforce the parallelism between the real and the virtualvisit, the clip player represented in the web interface is a replica ofthe PDA device on which the real visit take place, and the room isrepresented by a PMVR player aimed at improving the user’s immer-sion in the virtual reconstruction of the real location. The left column,

22 Damiano et al.

Figure 7. The DramaTour website

which hosts the interface controls and the related feedback, contains thenumbered list of the rooms (referred to by their names) and a map ofthe site, in which rooms are numbered consistently with the list. Bothrepresentation can be clicked, and their effect of the system’s behavioris equivalent. The map allows the user to figure out the natural orderof visit of the rooms by showing the constraints in the physical space,an order which is further enforced by the room numbers.

The navigation flow is very simple: at the beginning of the visit,the current room is, by default, the first room. The user can click onthe map to move to a different room, and is left free to move fromany room to any other room, since the DramaTour methodology doesnot make any commitment to a sequential access to the set of roomsif not required by a specific location. In order to provide the user with

feedback about the current room, the room name becomes highlightedin the room list. Differently from the guided tour in the physical space,the user location is always known, so the clip player itself does notreport any information about the current location.

The client/server schema on the web is implemented in the followingway: the web client, i.e., the web browser, only supports the clip player(Shockwave Flash Player 8) and the PMVR player. In order to accountfor users with different connection bandwidths, the latter may varyaccording to the user’s choice between QuickTime VR and Flash. Thevisit client is a PHP program that runs on the http server. PHP receivesthe user’s current location from the client and communicates with thevisit server via a socket.

4. The visit and the Evaluation Settings

The system evaluation lasted three days during the opening, and con-sisted of the qualitative and quantitative analysis of a survey ques-tionnaire given to the users the end of the visit, and of ethnographicobservations. The visit was free of charge and the visitors were wel-comed by hostesses which gave them a headphone-equipped PDA, withwritten instructions advising that the system was a prototype understudy, and some suggestions about the use of the guide (e.g., “Youare free to stroll in the rooms”, “Carletto is able to adapt its de-scription depending on your position”, “You can stop and resume thepresentation by pressing the button illustrated below”, etc.). Althoughinstructions were likely to influence the behavior of the users somehow,thus interfering with the evaluation, this risk was traded off against theprobability that they would take Carletto for a standard, non adaptiveaudio-guide, and limit their behavior accordingly, thus reducing therelevance of the evaluation in a relevant way.

After the visit, every user was asked to fill in a questionnaire of17 questions. Three topic areas were proposed by the questionnaire:socio-demographic data, the familiarity with technological equipments,an evaluation of the visit. The questionnaire contained single andmultiple choices, Likert scale2, and a free comment section.

The museum hostesses asked visitors to fill in the questionnaire,when they gave back the device. The questionnaires were then au-tonomously filled by the users to avoid any possible interference; theywere anonymous, voluntary, and introduced by a written presentation

2 A Likert scale is a type of survey question where users are asked to evaluatethe level at which they agree or disagree with a given sentence (Sirkin, 2005).

24 Damiano et al.

explaining the general research aim (collecting real and anonymoususers data for a human-machine interaction research).

The goals of the evaluation were to investigate, in real context set-tings, the user acceptance, the effectiveness of the dramatized character-based approach (also in comparison with conventional systems), and togather information for future system redesign. The questionnaire wasconceived with the help of an expert in social science of media, andpartially modified after the pilot test described above.

4.1. Quantitative Evaluation

All the 380 visitors compiled the questionnaire. However, we just con-sidered 249 questionnaires since some were incomplete, and a numberof users could not complete the visit because of technical problems. 3

Socio-demographic data. The visitors were 87 males and 161 fe-males, with an age distribution ranging from younger than 15 years oldto older than 64 years. In particular, the complete age distribution is:20% ≥64, 21% 55-64, 15% 45-54, 10% 35-44, 22% 25-34, 10% 15-24,2% ≤15.

Their level of education was high: 42% graduate, 44% high school,12% junior high school, 2% primary.Technological equipments. Concerning their technological equip-ments, see Figure 8, the 90,36% declared to have a cellular phone, the67% a computer, the 80% an Internet connection (home or work), andonly the 4,2% a palm. This distribution seems to reflect the commonlyassumed diffusion of technology in Italy, as reported by the NationalInstitute of Statistics (ISTAT).Evaluation about the visit As far as the visit experience is concerned(see Figure 9), users were in general quite satisfied (notice that thepossible scores were ranging in a Likert scale from 1 to 5): the averagescore for the visit evaluation is 3.51 with a standard deviation (SD) of1.4; the average score for the content quality is 3.65 with a SD of 1.3;the average score for the content quantity is 3.54 with a SD of 1.4; theaverage score concerning the visual aspect of Carletto is 3.63 with a SDof 1.4; the average score concerning the easiness of use is 3.8 with aSD of 1.4.

3 During the visit, some users suffered from technical problems mainly caused bythese sources: the maintenance of the devices, and the wireless network. The firstsource was due to the fact that 20 visits were scheduled per half an hour, with theavailability of 50 palm devices, and the palm battery sometimes took longer thanhalf an hour to recharge, and some batteries were faulty. So, some user experiencedincomplete or frequently interrupted visits, without the possibility of changing thedevice. The second source of problems concerned the wireless network of the building,which was usually used for other purposes (data exchange, Internet access).

ellite

ital T

Serie1

Figure 8. Technological equipments.

0,000,501,001,502,002,503,003,504,004,505,00

Total e

valuat

antity

sometimes

visit interrupted

Figure 9. The visit experience in relation to technical problems.

The visit experience exhibits a good acceptance of the proposedapproach for the general public, but a remarkable variability, withlarge standard deviation values. We hypothesized that a possible reasoncould be the technical problems some user suffered during the visit.Figure 9 shows the average values for the above dimensions in relationto the technical problems experienced by the users. The strong vari-ability of the results demonstrates that technical problems had a highimpact on users’ judgements, which are higher for visitors that had lessproblematic tours.

26 Damiano et al.

In order to find significant relationship between the variables understudy4, we measured Pearson correlation, since scores showed a normaldistribution (see (G. Keppel and Tokunaga, 1998) for details).

The visit experience has no significant correlation with sex, age andtechnological equipments, but the results of contents quantity show asignificant negative correlation with education (r=-0.145, significant atthe 0.05 level, 2-tailed)), and also between easiness of use and education(r=-0.139, significant at the 0.05 level, 2-tailed). This highlights thefact that visitors with high education level were less satisfied aboutthe visit experience, and they seem to be more critical and demanding.In particular, as the analysis of the free comments shows (see Sec.4.3), they would like to receive a larger amount of contents and toexperience a different modality of interaction (more specifically, a morecontrollable interface).

We also calculated correlations only considering the judgments ofvisitors experiencing less problematic tours. In this case, only educationand easiness of use show a significant negative correlation (r=-0.192,significant at the 0.05 level, 2-tailed). Thus, visitors experiencing lessproblematic tours, and with a high education level, are critical aboutthe modality of interaction proposed by Carletto.

The character-based storytelling approach was well accepted by vis-itors, since content quality receives a high score as well as Carletto’svisual aspect (not an obvious result for a spider).

Connected to storytelling is also how visitors perceived the drama-tized presentation delivered by the virtual characters. Therefore, wedesigned a multi-choice question to ask visitors which personality theywould associate to Carletto. The aim was to discover what visitorsthought about Carletto’s presentation, depending on the personalitythey associate to him. Every personality has distinctive features as-sociated, thus the analysis of the visitors’ choices is an indirect wayof discovering their impression about the key features of the overallexperience, as well as their positive or negative judgements. Figure 10shows the results. The mostly associated personality is the “talkingcricket” (23%), the character of Collodi’s “Pinocchio”. The talkingcricket is a little lively animal, which typically speaks too much, butit is also the metaphor of “the voice of the visitor’s conscience”, whichsuggests visitor what to explore next. Thus, most visitors perceivedCarletto as a funny character, sometimes annoying, but lovely, and

4 Age, level of education, technological equipments of the users, the visit evalu-ation, the content quantity, the content quality, the visual aspect of Carletto, theeasiness of use of the system, the personalities attributed to Carletto, the apprecia-tion for human and audio guides, the visitors capability to orient themselves and tofind items, the request for more interactivity, the ergonomics of the device

Carletto's personalities

Curious

Talking cricket

Guiding

TravelCompany

Annoying

NoPersonality

Figure 10. The personalities of “Carletto the spider”.

above all a friendly and un-conformist guide, which is normally prefer-able. Both the “travel mate” (20%) and the “guide” (18%) have re-ceived many preferences. This means that the visitors perceived thecharacter in a friendly way, and they are inclined to follow its advices, asdemonstrated by the ethnographic study described below. These opin-ions are always associated to positive judgements; we found significantcorrelations between:

− the visit evaluation and the guide character (r=0.189, significantat the 0.01 level, 2-tailed); when the appreciation for Carlettoincreases, visitors tend to identify him as a guide;

− the visit evaluation and the travel mate (r=0.220, significant at the0.01 level, 2-tailed); when the appreciation for Carletto increases,visitors are likely at identifying him as a travel mate, in a positiveand friendly way;

− the age and the travel mate (r=-0.158, significant at the 0.05 level,2-tailed). When the age level of visitors decreases, they are moreinclined at identifying him as a mate, in a friendly perspective.

Fewer ratings were associated to the comic character (12%), theprying one (11%), the annoying one (7%), and the sage (4%). Theywere always associated to negative judgements. We found significantcorrelations between:

28 Damiano et al.

− the visit evaluation and the annoying character (r=-0.219, sig-nificant at the 0.01 level, 2-tailed). Quite obviously, visitors whodislike Carletto perceive him as annoying;

− age and the prying character (r=-0.223, significant at the 0.01level, 2-tailed). Visitors are inclined to judge Carletto like a pry-ing character, as long as their age decreases, thus young visitorfind him “offensively curious or inquisitive”, in a negative way. Asshown by the free comments answers, they did not appreciate itsgossip about the Savoy royal family;

− education and the sage character (r=-0.284, significant at the 0.01level, 2-tailed). When the education level increases, visitors tendto not consider Carletto as a sage one, as well as they are notsatisfied about the contents he conveys (see results about the visitexperience).

visit e

uality

easine

human guide

audio guide

Figure 11. Human and audio-guide.

In the chart of Figure 11), we compare the evaluation of the virtualguide two more established approaches, namely, the human guide andthe audioguide. The visitors under study generally appreciate human(average rate 3.8, range: 1-5) and audio guides (average rate 3.5, range:1-5). We found a very significant positive correlation between thosewho like audio-guides and the visit evaluation(r=0.486, significant atthe 0.01 level, 2-tailed); the appreciation of audio-guides has a verysignificant negative correlation with age (r=-0.191, significant at the0.01 level, 2-tailed). Hence, visitors who appreciate audio-guide havealso a positive attitude toward Carletto. Those visitors do not like to

have a guided fruition in a social dimension, and they are likely to belonely in experiencing their visit. Moreover, as the age level decreases,their preference toward such devices tends to increase.

Could you find the items referred to by Carletto?

sometimes

Figure 12. The users’ orientation.

Figure 13. Modalities of interaction.

The remaining questions were aimed at evaluating the modality ofinteraction. Most visitors (83%) were able to orient themselves andto find the items Carletto was talking about (Figure 12); while mostof them would appreciate a more active interaction (Figure 13) withthe possibility of directly asking questions to Carletto (41%: yes, 39%:sometimes). By analyzing these variables we found the following corre-lations:

− the ability to orienteer in the location and to find items has a verysignificant positive correlation with the visit experience (r=0.332,significant at the 0.01 level, 2-tailed), audio-guides (r=0.214, sig-nificant at the 0.01 level, 2-tailed) and the use of PC (r=0.166,significant at the 0.01 level, 2-tailed). Thus a general good expe-rience of the visit is characterized by a satisfying fruition of thefocusing proposed by Carletto, and also by the appreciation ofaudio-guide and the use of PC.

30 Damiano et al.

− the request for more interactivity has a significant negative corre-lation with the judgement about the content quantity (r=-0.151,significant at the 0.05 level, 2-tailed) and a positive significantcorrelation with the use of PC (r=0.142, significant at the 0.05level, 2-tailed). Visitors who seem to be less satisfied with thevisit would like to have not only more contents but also moreinteractivity. These visitors frequently use the PC.

Finally, concerning specifically the ergonomics of the device (seeFigure 13), the 82% of the visitors found the device easy to handle,with significant positive correlation with the use of the mobile phone(r=0.152, significant at the 0.05 level, 2-tailed). Therefore, keeping thepalm from its shorter side was not problematic for visitors. However,after the ethnographic observation, see Sec. 4.2, we concluded that asafer solution is to put a cord on it, in order to let the users hang upthe device to the neck.

4.2. Ethnographic Observations

As noticed by (Dourish, 2001), since statistical analysis can be mis-leading and too narrow, we carried out a qualitative study relyingon insights and on the users’ observed behavior and reactions. Morespecifically, we carried out ethnographic observations and a qualitativeanalysis of the open answers collected by questionnaires (see Section4.3). Concerning ethnographic observations, we observed the behaviorsby strolling in the site amidst visitors like hostesses that moved alongrooms to take care of them (so observed visitors did not feel understudy). Often visitors asked us help and questions. The most interestingobservations that emerged are:

− the use of Carletto makes the visitors less inclined to distract theirattention from the device to observe the location and the objectaround them;

− at the beginning, visitors tend to be lonelier, isolated, and withlimitations to their ability to listen and to speak. After a while,some of them regained the group, and discussed among them aboutthe presentation;

− some visitors were completely directed by Carletto, waiting for itsadvices about what to see and where to go;

− sometimes older visitors tried to speak to Carletto, directly ad-dressing him with spontaneous comments;

− some very young visitors thought the Carletto was hidden some-where in the ceilings.

These findings enrich the results described in the previous section.An interesting observation is that some visitors do not like to be lonelyin the visit, thus the virtual guide should be designed to be used frommore than one user, while keeping the individual fruition. The broadappreciation for audio-guide, shown by the visitors under study, alsodemonstrate that there are visitors preferring a more individual expe-rience, even if visitors often visit museum in group5. The presentationmust not to capture the visitors’ attention in an all-involving way,but it has merely to guide visitors in the fruition: visitors need tobe encouraged at looking around, and not focus on the device. Finally,the last two points somehow underline that some users immerse them-selves in the dramatization process, since they tend to assign a placeto the virtual character in the real world. This can be interpreted asan example of the Laurel’s interface as mimesis (Laurel, 1993).

4.3. Qualitative Evaluation

The qualitative analysis of visitors’ free comments was inspired bythe Grounded Theory, a well defined research methodology where thetheory is derived from data, in an inductive process of data analy-sis (Strauss and Corbin, 1990). According to the Grounded Theorymethodology, collected data may be qualitative or quantitative or acombination of both types, since an interplay between qualitative andquantitative methods is advocated. For this reason we found very usefulto integrate the free comments study with the analysis of quantitativedata and of the ethnographic observations described before, in order todiscover possible interconnections and make them emerging during thestages of theory.

In particular, three phases of data analysis are involved in the theory:open coding, which is the analytical process through which concepts areidentified starting from the analysis of the collected material, and theproperties and dimensions of these concepts are discovered; axial codingthat is the process of relating categories to their subcategories, termed“axial” because coding occurs around the axis of one category, linkingcategories at the level of properties and dimensions; selective coding,namely the process of integrating and refining the theory.

The questionnaire contained a free comment section for the ques-tion “Some observation?”. Among all the 249 visitors, 114 filled in

5 Only the 13% of visitors visit a museum alone Il Pubblico di Mostre e Musei inPiemonte: http://www.fitzcarraldo.it/

32 Damiano et al.

the question. We started by examining the open answers, which werebroken down into discrete parts. Then we closely examined data andcompared them for similarities and differences, and we started to accu-mulate concepts. At the same time we started the inductive process ofthe investigation and definition of main categories, subcategories andvariables involved in the phenomenon under study. The main categoriesidentified in the study were positive evaluations without suggestions(e.g.: “cool”, “I love it”), negative evaluations without suggestions (e.g.:“ridiculous”, “I hate it”), reported technical problems, suggestions. Af-ter having analyzed the material related to the main categories, weconcentrated our analysis on the last one (suggestions), which was themost interesting for the purposes of our evaluation, namely assessingthe acceptance of the proposed approach and collecting feedback forfuture re-design. The following subcategories of suggestions emergedfrom the inductive analysis, (presented with some example of theircharacterizing properties):

− Information: the need for more information, the possibility togo into more depth about the topic presented, to receive moreinformation, etc.;

− Adaptation: the adaptation of the guide to the age or the needs ofthe visitors, Carletto is for children, Carletto is a child, the needof a more serious guide, etc.,

− Interactivity : the request of more interactivity, the possibility tochoose the quantity of delivered information, the possibility to stepback, the possibility to manually indicate the location, etc.;

− Context : the need of contextual information, the presentation hasto show the items Carletto is talking about, the presentation hasto highlight contextual information, etc.;

Then we put together these findings with quantitative data and theanalysis of ethnographic observations, and we started relating cate-gories to their subcategories along with their properties and dimensionsto form more precise and complete explanations until an acceptablelevel of saturation was reached. We also tried to investigate conditions(e.g., when does a user need more interactivity? which kind of usersneed more interactivity), action and interactions (e.g, in which situationusers experience the lack of interactivity) and their consequences. Someinteresting findings helped at refining the core theory of the selectivecoding: users/visitors have highlighted the need of some kind of a morepersonalized interaction, because they would like to be able to choose

the modality and the type of presentation, the localization, the qualityand the quantity of delivered information, the character itself, its per-sonality, etc. Therefore, as discussed below, our future re-design willbe aimed at taking into account these user needs by adding adaptationfeatures and user modeling to the system.

4.4. Summary of Results

The visitors under study seem to reflect the general distribution ofvisitor in Piedmont6, the Italian region where Torino is located: themostly represented age range is comprised between 45 and 64 yearsold; the majority of visitors are women; about the 40% of visitorshas a degree, and a similar percentage has a high school diploma.Thus we could affirm that the sample of the survey seems to be quiterepresentative of the museum population in our territory, under a socio-demographic point of view. However, it has been hard to find somedistinctive socio-demographic features on the basis of which trying tomodel the visitor.

Significative correlations show that: visitors with high educationlevel are more critical about the experience with Carletto, in particularwith the content quantity, and the modality of interaction; visitors withhigh education level do not think that Carletto is a sage one; youngvisitors who appreciate Carletto are likely to identify Carletto as atravel mate; young visitors who do not appreciate Carletto do not likehis extreme curiosity; young visitors are likely to prefer a device-drivenmuseum visit; visitors who are likely to use the PC have a good visitexperience, in term of orientation; but also the visitors who demandmore interactivity are likely to use the PC.

Even the visitors who wrote productive suggestions (e.g., many ofthem have underlined the need of a more serious guide presentinginformation in a more detached and objective way) did not exhibitdistinctive features which let us to model users. As it can be noticed,it is very difficult to create different stereotypical models of visitorsaccording to these correlations. However, in order to transform Car-letto in a user-adaptive system, strategies of adaptation merely basedon socio-demographic features seems to be quite inappropriate. To bemore accurate in these findings, we have also run a K-means clusteranalysis (Huang, 1998) to identify relatively homogeneous groups ofcases based on sex, age, education, technological equipments, and thevisit experience values. Only two clusters emerged:

6 “Il Pubblico di Mostre e Musei in Piemonte”, study from the Fitzcarraldofoundation, a local independent research center, http://www.fitzcarraldo.it/

34 Damiano et al.

− Cluster 1 is characterized by the following features: 30 years oldfemales, high school education level, medium use of web; theypossess a PC and a cell phone, they really liked the visit (theaverage visit experience value is 4) and sometimes they would liketo ask something to Carletto; they really like human and audiomuseum guides;

− Cluster 2 is characterized by the following features: 61 yearsold females, high school education level, infrequent use of web;they possess a PC and a cell phone, they really like the visit (themedium visit experience value is 4) and sometimes they wouldlike to ask something to Carletto; they really like human museumguides and like audio-guide;

These two clusters only partially covers the population under study,and in general the population of museum visitors. So probably, as otherstudies have already highlighted (Zancanaro et al., 2007), museum vis-itors who share similar visit habits can hardly be identified by similarsocio and psyco-demographic features. On the basis of our analysis,we can hypothesize some coarse classifications, useful at introducing amore adaptive interaction and of more adaptive delivering of contents.Excluding visitors who like Carletto as it is, possible distinctive visitorsfeatures are: visitors who like to be guided, and visitors who prefer tofreely move in the museum; visitors who prefer to be in a group, andvisitors who prefer to be alone; visitors who want a more controllableinterface; visitors who want the possibility of deciding the quantity andquality of information delivered; visitors who would like to have a moreserious and detached character.

5. Comparison with Related Work and Discussion

The DramaTour methodology of location-based information presen-tation and the application “Carletto the spider” for character-basedguidance to an historical site share many design issues with a num-ber of projects active in the applied research literature on informa-tion presentation and cultural heritage exploitation. We can single outthree aspects: multimodality, character-based mediation, and narra-tive/storytelling forms.

Thanks to increasing hardware and software resources, high-qualitygraphics and sound have become available for the development of newmultimodal applications. Multiplicity of modalities and of interactionparadigms is particularly wanted in mobile devices, to cope with unpre-dictable environmental conditions, and the limitations of the computa-

tional resources (memory, battery, processing power, unreliable wirelessconnection, small screen size).

Multimodality presents a variety of possible interactions in bothinput and output: speech recognition and generation, pen pointinginput, head and eye tracking, 3D gesture, schemata and browsing arebeing integrated by a number of approaches (see, e.g., (Callaway et al.,2005; Solon et al., 2004)). The use of different modalities should realizean adequate balance of complementarity and redundancy to decreasethe cognitive load of the users, and evaluations seem to confirm thedesirability by the users to be able to switch automatically or manuallythrough several modes (Oviatt, 2002). In the case of mobile devices,much attention has been devoted to peculiar aspects of mobility, likethe integration of explicit user input (e.g. speech and pen pointing)with implicit contextual knowledge (e.g. location) in the interactiondesign (Yue et al., 2005), evaluation of mode recognition in exertionconditions (Kumar et al., 2004), or tools that support multimodalityover different platforms (Simon et al., 2005). In the specific applicationsfor cultural heritage exploitation, it is worth noting the efforts towardthe improvement of the cohesion of the social experience (Aoki et al.,2002), a very difficult task when dealing with “personal” devices.

Our approach provides multimodality in input, with the implicit andthe explicit input forms, and in output, with sound, graphical visual-izations introduced by Carletto, and textual feedback on the locationdetection. The system evaluation has occurred in overload conditions,since the site infrastructure was in use for other tasks, and the mainte-nance of 50 PDA’s (including battery charging and wireless card beingstuck at times) concerned an organization beyond the mere researchissues. Finally the system supported different platforms for delivery(PDA and WWW), always keeping the same aspects of location-basedcharacter-enacted presentation. Limitations of the system on the multi-modal aspects, as revealed by the user’s responses, are the non existentsocial experience, that for many people who visit a site in group isfundamental and the possibility of making explicit content requeststhat interrupt the presentation flow.

Characters improve the naturalness of the interaction with the user,making the system appear more responsive and cooperative. Muchresearch has addressed the capability of engaging in a natural face-to-face dialogue in the framework of Embodied Conversational Agents(Cassel et al., 2000). The methodology for building ECAs ranges fromsynthesizing behavior from an abstract specification of the character toassembling predefined units (Bernsen and Dybkjaer, 2004; Kopp et al.,2005). Improved sound and graphic processing has provided virtualguides with realistic facial expressions and lip-synch speech that display

36 Damiano et al.

emotions for pedagogical aims (Leon and Fisher, 2006). On a differentand complementary perspective are the CG animated characters, nowa-days ubiquitous in the film and game industries (Maestri, 1996; Clark,2002). Virtual characters are carefully modeled and rendered, theymove, talk and speak in realistic manner, and are endowed with full-fledged, sophisticate personalities; all these factors contribute to cre-ate Coleridge’s “suspension of disbelief”, that characterizes fictionalnarration and favors identification and immersion by the user.

Carletto is posited at the junction of the two worlds of ECAs andCG characters. His behavior, cast in units, results from his presentationcapacities as well as his animated acting. The animation task has beensolved (being a research project at a tiny budget with respect to gameand movies) by identifying a number of elementary expressions thatwere composed to form the behavior units. Also, Carletto mixes anencompassing presentational strategy, encoded in the interaction script,and the predefined units that are to be edited during the presentation.The anthropomorphic design helps people in establishing a relationshipwith the virtual character, and enhances the presentation in most cases.Users’ assignment of a personality to Carletto supports such a claim.However, in the DramaTour methodology the existence of a characteris strictly tied to the existence of a dramatic scene.

Interactive storytelling aims at the real-time generation of an en-gaging plot, acted by artificial characters, generally for entertainmentand learning purposes. Story theorists like Bruner (Bruner, 1991) claimthat stories provide a framework for making sense of events and theirmeaning (Mateas and Sengers, 1999). Also, stories can convey largeamounts of information in a compact format, which is easily assimi-lated by the user (Gershon and Page, 2001). Interactive storytellinginvolves the creation of a plot and its factorization into elementaryunits, the implementation of a planning system that reacts to the user’sinputs in real time in order to assemble a story from such elementaryunits. The organization of content in the exposition usually relies onsimple templates derived from narrative theories like those of Propp(Propp, 1968) and Greimas (Greimas, 1977). Interactive drama aimsat displaying a story through the live action of characters at the presenttime (Mateas and Stern, 2003). Drama is distinctive feature of the theDramaTour methodology: enacting the story improves the receptionof the conveyed information, and the assignment of a personality toCarletto – as reported by the users of the evaluation – testifies theidentification of the virtual character.

Besides these methodological issues, parallels with other approachesalso exist on the domain of cultural heritage exploitation, where mostapplications are geared to recommend items during a museum visit

(see e.g. (Pechenizkiy and Calders, 2007)). Considering the broad divi-sion between content-based recommendations, when the user is recom-mended on the basis of her/his own past preferences, and collaborativerecommendations, when the user is recommended on the basis of otherpeople’s past preferences (people with similar tastes) (Adomaviciusand Tuzhilin, 2005), DramaTour only takes into account the contentalready delivered to the user, with the goal of not repeating itself inthe presentation. Although we notice that the drama-based approachfeatures a number of peculiarities, it is possible to update the systemby incorporating notions of personalization and adaptation. Here wesketch some proposals for future re-design.

− In order to deliver a presentation in a style that is more appropriatefor a user, the system can propose several characters (e.g., a comicand a more serious one, at least), and let the users choose, at thebeginning of the visit, the most appropriate guide. Every guideshould introduce her/himself to the visitor in a video clip, andshowing her/his character and her/his visit style. The guide per-forms the presentation in a dramatized way, reflecting immediatelythe peculiar features of her/his behavior. So, the user does not haveto specify a list of attributes she/he would like for her/him visitexperience, but can choose the guide that best matches her/hisvisit style. Of course, this choice leads to almost doubling the effortof developing the content material, since the author needs to recastthe same information in another presentation style and, thus, inanother drama.

− Each guide can show an adaptive behavior depending on the visitstyle exhibited by the user. When a visitor stays longer in a room,the system selects clips containing more detailed contents to bedelivered. These clips show information in a more detached way,e.g. without dramatization. This would satisfy one requirementthat emerged from the data analysis, i.e. that visitors who wantmore contents prefer a more serious and detached modality in thefruition of contents. On the contrary, when the visitor spends lesstime in a room, and does not look for details, the clips deliveredcan be more general and presented in a well dramatized manner.

− A number of users reported they would like to ask Carletto forsomething specific, so to improve interactivity. In order to allowvisitor to raise questions but at the same time to leave the controlto Carletto and simplify the input recognition (free vocal messagescould be difficult to recognize), we propose that, when speakingof some topic, Carletto shows a poster that concerns some specific

38 Damiano et al.

item in the room; this poster remains active for 30 seconds; if theuser clicks on it, Carletto directs his presentation towards thatitem.

6. Conclusions

In this paper we have presented the DramaTour methodology for dra-matized presentations in a location-based character-enacted manner.The application “Carletto the spider”, based on the DramaTour method-ology, has been developed for the visit guidance in an historical site,implemented for both mobile devices on site and via WWW browsingremotely. The paper has addressed the knowledge encoding provided bya “procedural” scriptwriter and the interaction strategy compliant withthe dramatic features. Personalization of the presentation is based onthe implicit user input, due to the duration of the visit in each location,and explicit input, via button-clicking for a number of actions (Kobsaet al., 2001). We have also carried on an evaluation of the system ina open-to-the-public installation with almost 400 visitors, and we haveproposed a number of extensions for increasing the adaptation andpersonalization issues as they emerged from the evaluation.

References

Adomavicius, G. and A. Tuzhilin: 2005, ‘Toward the Next Generation of Recom-mender Systems: A Survey of the State-of-the-Art and Possible Extensions’.IEEE Trans. Knowl. Data Eng. 17(6), 734–749.

Aoki, P. M., R. E. Grinter, A. Hurst, M. H. Szymanski, J. D. Thornton, and A.Woodruff: 2002, ‘Sotto Voce:Exploring the Interplay of Conversation and MobileAudio Spaces’. In: ACM SIGCHI 2002. Minneapolis, Minnesota, USA.

Aristotle: 1920, On the Art of Poetry. I. Bywater (transl.) 1920. Oxford: Clarendon.Austin, J. L.: 1962, How to Do Things with Words. Cambridge, Mass: Harvard

University Press.Bernsen, N. O. and L. Dybkjaer: 2004, ‘Domain-Oriented Conversation with H.C.

Andersen’. In: Proceedings of ADS 2004. pp. 142–153.Biral, F., V. Lombardo, R. Damiano, and A. Pizzo: 2003, ‘Cyrano goes to Hollywood:

a drama-based metaphor for information presentation’. In: AIMS 2003. pp.17–24.

Bordwell, D. and K. Thompson: 2006, Film art : an introduction. Boston: McGrawHill.

Bratman, M.: 1987, Intention, Plans, and Practical Reason. Cambridge (MA):Harvard University Press.

Bruner, J.: 1991, ‘The narrative construction of reality’. Critical Inquiry 18(1),1–21.

Callaway, C., E. Not, A. Novello, C. Rocchi, O. Stock, and M. Zancanaro:2005, ‘Automatic cinematography and multilingual NLG for generating videodocumentaries.’. Artificial Intelligence 165, 57–89.

Cassel, J., T. Bickmore, L. Campbell, H. Vilhjamlmsson, and H. Yan: 2000, ‘Hu-man Conversation as a System Framework: Designing Embodied ConverationalAgents’. In: J. Cassell, J. Sullivan, S. Prevost, and E. Churchill (eds.): EmbodiedConversational Agents. The MIT Press, Cambridge, Massachussets, pp. 29–63.

Cavazza, M., F. Charles, and S. Mead: 2002, ‘Interacting with Virtual Characters inInteractive Storytelling’. In: Proc. of the First Int. Joint Conf. on AutonomousAgents and Multiagent Systems.

Clark, K.: 2002, Inspired 3D Character Animation. Premier Press.Cohen, P. R. and H. J. Levesque: 1990, ‘Intention is Choice with Commitment’.

Artificial Intelligence 42, 213–261.D. Nau, H. Muoz-Avila, Y. Cao, A. Lotem and S. Mitchell: 2001, ‘Total-Order

Planning with Partially Ordered Subtasks’. In: Proc. IJCAI-01. pp. 425–430.Damiano, R., V. Lombardo, F. Nunnari, and A. Pizzo: 2006, ‘Dramatization meets

information presentation’. In: Proceedings of ECAI 2006. Riva del Garda, Italy.Dourish, P.: 2001, Where the Action Is: The Foundations of Embodied Interaction.

Cambridge: MIT Press.Erol, K., D. Nau, and J. Hendler: 1994, ‘HTN planning: Complexity and expressiv-

ity’. In: AAAI-94 Proceedings.Esslin, M.: 1988 (1987), The Field of Drama. London: Methuen.Freytag, G.: 1883, Die Technik des Dramas. Leipzig: Hirzel.G. Keppel, W. H. S. and H. Tokunaga: 1998, Introduction to Design and Analysis,

A Student’s Handbook. W H Freeman and Co.Gershon, N. and W. Page: 2001, ‘What Storytelling Can Do for Information

Visualization’. Communications of the ACM 44(8), 31–37.Greimas, A.: 1977, ‘Elements of a Narrative Grammar’. Diacritics pp. 23–40.Grice, H. P.: 1975, ‘Logic and conversation’. In: P. Cole and J. L. Morgan (eds.):

Syntax and Semantics, Vol. 3, Speech Acts. New York: Academic Press, pp.41–58.

Grosz, B. J. and C. L. Sidner: 1986, ‘Attention, Intentions, and the Structure ofDiscourse’. Computational Linguistics 12, 175–204.

Hajicova, E., B. H. Partee, and P. Sgall: 1998, Topic-Focus Articulation, TripartiteStructures, and Semantic Content. Dordrecht: Kluwer.

Hoang, H., S. Lee-Urban, and H. Muoz-Avila: 2005, ‘Hierarchical Plan Representa-tions for Encoding Strategic Game AI’. In: AIIDE05.

Huang, Z.: 1998, ‘Extensions to the K-means Algorithm for Clustering LargeDatasets with Categorical Values’. In: Data Mining and Knowledge Discovery,Vol. 2. pp. 283–304.

Jackobson, R.: 1990, ‘The Speech Event and the Functions of Language’. OnLanguage.

Kobsa, A., J. Koenemann, and W. Pohl: 2001, ‘Personalized Hypermedia Presenta-tion Techniques for Improving Online Customer Relationships.’. The KnowledgeEngineering Review 16(2), 111–155.

Kopp, S., L. Gesellensetter, N. Krmer, and I. Wachsmuth: 2005, ‘A ConversationalAgent as Museum Guide - Design and Evaluation of a Real-World Application’.In: 5th International Working Conference on Intelligent Virtual Agents (IVA’05).

Kumar, S., P. R. Cohen, and R. Coulston: 2004, ‘Multimodal Interaction underExerted Conditions in a Natural Field Setting’. In: ACM ICMI’04. State College,Pennsylvania, USA, pp. 227–234.

40 Damiano et al.

Laurel, B.: 1993, Computer as Theater. Reading, MA: Addison Wesley Longman.Leon, J. and M. Fisher: 2006, ‘The Use of Virtual Characters to Generate Teach-

able Moments’. In: J. Trant and D. Bearman (eds.): Museums and the Web2006: Proceedings, http://www.archimuse.com/mw2006/papers/leon/leon.html.Toronto.

Maestri, G.: 1996, Digital Character Animation. New Riders.Mateas, M.: 2002, ‘Interactive Drama, Art and Artificial Intelligence’. Phd.

dissertation, Carnegie Mellon University.Mateas, M. and P. Sengers: 1999, ‘on Narrative Intelligence’.Mateas, M. and A. Stern: 2003, ‘Integrating plot, character and natural language

processing in the interactive drama Faade’. In: 1st International Conference onTechnologies for Interactive Digital Storytelling and Entertainment TIDSE 03.

McKee, R.: 1997, Story. New York: Harper Collins.Murray, J.: 1998, Hamlet on the Holodeck. The Future of Narrative in Cyberspace.

Cambridge, MA: The MIT Press.Oviatt, S.: 2002, ‘Multimodal Interfaces’. Handbook of Human-Computer Interac-

tion.Pechenizkiy, M. and T. Calders: 2007, ‘A Framework for Guiding the Museum Tour

Personalization’. In: International Workshop on Personalization Enhanced Ac-cess to Cultural Heritage in conjunction with 11th International User ModellingConference 2007. Corfu, Greece, p. 12.

Picard, R.: 1997, Affective Computing. MIT Press.Propp, V.: 1968, Morphology of the Folktale. University of Texas Press.Reisz, K.: 1968, The technique of film editing. London: Focal Press.Searle, J.: 1969, Speech Acts: an Essay in the Philosophy of Language. Cambridge

(UK): Cambridge University Press.Seger, L.: 1990, Creating Unforgettable Characters. New York: Henry Holt and

Company.Simon, R., F. Wegscheider, and K. Tolar: 2005, ‘Tool-Supported Single Authoring for

Device Independence and Multimodality’. In: ACM Mobile HCI ’05 Conference.Salzburg, Austria, pp. 91–98.

Sirkin, M.: 2005, Statistics for the Social Sciences. Sage Publications.Solon, A., P. McKevitt, and K. Curran: 2004, ‘Mobile MultiModal Presentation’.

In: ACM MM’04. New York (NY), USA.Strauss, A. and J. Corbin: 1990, Basics of qualitative research: grounded theory

procedures and techniques. Newbury Park, Calif.: Sage Publications.Yue, W., S. Mu, H. Wang, and G. Wang: 2005, ‘TGH: A Case Study of Design-

ing Natural Interaction for Mobile Guide Systems’. In: ACM Mobile HCI ’05Conference. Salzburg, Austria, pp. 199–206.

Zancanaro, M., T. Kuflik, Z. Boger, D. Goren-Bar, and D. Goldwasser: 2007, ‘Ana-lyzing Museum Visitors’ Behavior Patterns’. In: To appear in the proceedings ofUM 2007.

A stroll with Carletto: adaptation in drama-based tours with virtual characters

Documents

Transcript of A stroll with Carletto: adaptation in drama-based tours with virtual characters

Νέα Υόρκη - Exotic Tours

R ' - Université de Tours

Final Brochure - Trinetra Tours

Your DMC in France - Excelys Tours

Absurd Drama -Martin Esslin

UK/US Drama Bibliography

Activities/Drama Exercises - Amazon AWS

Roman Drama Imagines Poetarum

LJETOVANJA - Niba Tours

Characters of Phylum-Porifera

HALLELUJAH CHORUS - St. Martin of Tours Parish

Milfytoon Drama Walkthrough - F95zone

CHARACTERS BULDING - LPM IAIN Lhokseumawe

Travel Information for France - ExperiencePlus! Bicycle Tours

Main characters

tours for schools Groups2019/20 - NST Ireland

british drama - e-Gyanagar

radio drama

Swinburne Drama Workshop 2014

2017 ESCORTED TOURS FOR THE DISCERNING TRAVELLER