Simplifying the Controls of an Interactive Movie Game :I

Transcription

1 65 CHI APRIL 1998 PAPERS : Simpifying the Contros of an Interactive Movie Game :I Jeff Johnson UI Wizards, Inc. 231 Moscow St. San Francisco, CA USA t jjohnson@uiwizards.com ; t. ABSTRACT Eight months before an interactive movie game was due to be shipped, its deveopers and funders decided that its user interface had to be radicay simpified. The author was given the task of designing a new, simper contro scheme. This paper describes the redesign process, the design issues that arose and how they were resoved, the tests that were conducted to evauate new design ideas, and concudes with an evauation of the resuting design, essons earned, and thoughts on user-interface design vs. game design. Keyaoids User interface, design, games, usabiity testing, interactive movies INTRODUCTON In Juy 1996, Advance Reaity, a game software deveoper, hired me as a consutant to sove a user-interface design probem in a game they were deveoping. I was referred to them by AnyRiver Entertainment, the game s producer. AnyRiver was responsibe for testing the game on computer-gamenthusiasts (referred to and their tesrs indicated that payers were having difficuty operating the game s contros. The Game The game is caed A Fork in the Tae, abbreviated Fork. The deveopers had shot a fu-ength feature fim viewed through the eyes of the protagonist. In contrast to a norma movie, each scene was fimed with many outcomes. The deveopers were editing the footage together to create an interactive game. Such games are known in the game software industry as fu motion video (F&IV) games. Every few seconds whie the movie is paying on an individua s computer screen, mouse-sensitive contro symbos appear over the movie action providing opportunities for the payer to affect what the protagonist does or says, e.g., eave, enter door, ask question, make comment, don magic coak, dodge punch. The software seects the sequence of cips to show based on the payer s choices and various state-variabes. Permission to make digimd copies ofa or part of this materia for persona or cassroom use is granted without fre provided thatthe copies are not made or diiiuted fir profit or conuoe&d advantage, the copyri&tnotice, the tite ofthe uubiution and its date aooear. and notice is &en that copyright is by pkmission of the ACM, I&. To copy ohwise, to repubiisb, to post on servers or to rediibute to ists, requires specific permission andor fee CHI 98 Los Angees CA USA copyright 199s 0-s /9814..%5.00 The high frequency of choices provided in Fork represents an advance over previous interactive movies, which offer payers a choice ony every severa minutes. The i?equency of choices aso makes the game much more compex (i.e., branchy) than most PMV games: five CD-ROMs are required to hod a of the footage a payer might encounter. To keep the game s compexity manageabe, at most three contro symbos are presented at a time. Another difference between Fork and most other PMV games is that it is rea-time: the movie does not stop whie a payer considers which action to choose. If a payer fais to make a choice within a few seconds, the opportunity passes and a choice is made by defaut. The game s premise is that you, the payer and movie protagonist, innocenty wak into the midst of a shoot-out. get wounded, back out, and wake up washed ashore on an isand. The goa is to figure out where you are, who the other peope on the isand are, and how to get back home. Your character (the,protagonist) is an average guy who has an active sense of humor (his voice is that of comedian Rob Schneider). He is basicay non-vioent, but wi defend himsef if attacked. Others on the isand range from friendy and hepfu to hostie and dangerous. The Design Probem The probem facing the deveopers was that the symbos for controing the movie-action were much too compicated for payers to master: they were non-intuitive, unsystematic, and too numerous. Payers coud not make meaningfu choices between contro symbos quicky enough. Simpifying the contros was regarded as critica to the success of the game. The game deveopers of course wanted the game s mysteries and puzzes to be chaenging, but they didn t want operating the game to be difficut. The deveopers had tried three times to design action contros for the game, but none of the resuting contro schemes had proved satisfactory. They were therefore open to the game producer s suggestion to hire a consutant. The Assignment My task was to devise a new contro scheme: a way of organizing and representing protagonist actions that was easier to understand than the previous schemes had been. The game deveopers first wanted to be sure that they coud work with me and that I woud design a contro scheme that

2 PMEW CHI APRIL 1998 they iked, so they structured my assignment such that they coud decide quicky if they wanted me to continue. In three days, I was to produce a document describing a new contro scheme. It didn t have to have the fina action categories or contro art, but it did have to expain, abstracty, how actions woud be categorized and approximatey how many categories woud be needed. ANALYSIS I started by paying Fork and asking questions about it. I payed it not as a norma payer woud, but rather using specia production software that aowed me to jump around, repay cips, etc. I focused on a few game scenes that the designers said contained intensive action or conversation. My goa was to understand the expressive requirements of the game. Previous Action Contros I aso examined the most recent contro scheme. Though it had been designed with care and thought, rather than in an ad hoc fashion as many game designers do, it had severa faws that contributed to its poor usabiity: 1. Too many symbos. Action-symbos were organized in a three-dimensiona set, with coor, shape, and bitmap image as independent axes of meaning. With five coors, about 10 shapes, and dozens of images, the od scheme used hundreds of symbos. Furthermore, the meanings assigned to shapes and coors were arbitrary, e.g., coor: yeow = hepfu, green = curious; shape: spiked = aggressive, rounded = passive. It was neary impossibe for payers to earn a of the symbos. 2. More semantic resoution than necessary. The od contro scheme distinguished each action-situation in the game from every other. This woud make sense if payers were in contro of the time, subject, object, and mood of their actions, but in this case the game is in fu contro of the timing and effect of actions, oniy a few of which are avaiabe at a time. Therefore, the action-resoution required is much ower. 3. Simiar representation of very different actions. In the od contro scheme, most symbos were stationary opaque pates that appeared at the bottom of the screen. Whether a dispayed symbo depicted, e.g., a physica movement or a speech act, was indicated ony by the shape of the pate. This hindered recognition of even what genera1 type of action a symbo represented. 4. Fawed impementation. The game editors (who, aong with their other editing duties, added action-symbos to movie cips) didn t understand the od scheme, so they impemented it haphazardy. Each editor tended to use a subset of symbos s/he understood or iked the appearance of. Furthermore, editors made up new symbos occasionay, not a of which fit the scheme. The resut was an ad hoc impementation of a principed (abeit fawed as described in points -3) design. Expressive Requirements Athough the od contro scheme was to be repaced, it was cear that an important feature of it shoud be retained: cickabe action-symbos that appear at choice-points in the movie. The rea-time (i.e., non-stop) nature of the game required that a avaiabe protagonist actions at choicepoints be simutaneousy visibe, ruing out a contro scheme such as that used in sower-paced games, where the cursor changes shape as it is moved around the screen to indicate what action is avaiabe there. The main re-design task was therefore to design a simper but sufficient categorization of actions, and symbos to represent the categories. Initia anaysis of the game suggested that protagonist-actions fe into six categories: navigate in a direction, ook in a direction, interact physicay with objects or peope, speak, think sienty. and memorize events or faces. Experience with iconic user interfaces [2] suggested to me that choosing among avaiabe actions woud be easier if these six casses were grouped into two superordinate action-casses -- physica movements (navigate, ook, and interact) vs. speech acts and thoughts -- with gross differences in how the two casses woud be presented. Opportunities for physica actions woud be represented by back and white, semi-transparent, animated hotspots. They woud usuay foat over the action and pan with it (e.g., enter door, bock punch), but might appear at the edges of the screen if they had no cear direct object (e.g., turn right, jump). A few such hotspots had aready been introduced into the game by editors dissatisfied with the existing contros, and seemed successfu. In contrast, opportunities for speech and thought woud be represented by cartoon speech baoons containing a symbo representing the specific speech or thought sub-category. They woud appear at the bottom edge of the screen (to appear to come from the protagonist s own mouth) in one of three fixed positions. Memorizing events and faces posed a diemma: which main symbo cass shoud it be in? Memorizing is a menta act. but has a definite target in the movie-action, ike ooking cosey at something. The game editors had aready used hotspots to represent opportunities to memorize things, and pay-testing indicated that it worked we. Foowing the principe if it isn t broken, don t fix it, I eft this aspect of the game as it was. An important goa was to reduce the number of action subcategories (and symbos). The recognizabiity and discriminabiity of graphica symbos in a set does not depend excusivey on set size: peope can discriminate fairy arge sets if the symbos in the set are different enough from each other and if the mappings from symbos to meaning are intuitive [3]. Nevertheess, other things being equa, smaer symbo sets are preferabe. Whatever 66

3 67 CH APRIL 1998 PAPERS the optima number might be, hundreds of action subcategories was ceary too many. It was aso cear, however, that the game s expressive requirements rued out having ten or fewer sub-categories per main action-cass. Somewhat arbitrariy, I set mysef the goa of having at most 30 symbos in each of the two main casses. After paying critica game-scenes, I devised preiminary categorizations of movement and speech actions that seemed to cover the expressive requirements whie not exceeding this imit (detais beow). Athough the fina contro symbos woud be drawn by graphic artists (one artist per set to assure consistent appearance within each set and cear differences between the two sets), I made rough sketches for each of the speech and movement sub-categories. Having satisfied mysef that a simpified contro scheme that met the game s expressive needs was feasibe, I presented it to the deveopers. They iked the simpified contro scheme and asked me to continue refining it and to work with the graphic artists, game editors, and programmers to get it impemented Additiona Requirements Over time, I earned that, in addition to the game s expressive requirements, there were requirements and constraints arising from Advance Reaity s schedue and resources, and from the wishes of the game designer- An important constraint on the I-e-design effort was that the ony funded usabiity testing was the pay-testing being conducted by AnyRiver. Their support for Fork was based, in part, on positive resuts from testing eary prototypes of the game on computer-gam enthusiasts. They continued pay-testing throughout the deveopment period, reporting payer enjoyment, usabiity probems, and bugs to Advance Reaity. Any other usabiity testing had to be done very quicky and at essentiay no cost. However, we sometimes needed to test specific design ideas before adding them to the game. In such cases, we had to devise quick, cheap tests. A fina set of requirements were based on the game designer s wishes and goas. Before describing them, it is important to carify my roe in the project In the computer game industry, the game designer is usuay aso the user-interface designer (supported by art directors and graphic artists). Because the roes were separate in this case, it was possibe to observe how our concerns overapped and interacted. The game designer was concerned mainy with how entertaining and aestheticay peasing the game was and how we it embodied his overa vision. I was concerned mainy with usabiity. Our different concerns often ed us to differences of opinion. As a consutant, I had to accept the game designer s judgment if he insisted. He respected my judgment and yieded to it more often than he overrued it, but, naturay, he was difficut to convince if a proposa ran counter to his vision of the game. His ideas about the game that affected the re-design incuded the foowing: Athough he reaized that the new contro-scheme had to have far fewer action sub-categories than the od scheme, the game designer tended to favor having more categories than I did. To him, each protagonist-action had a meaning, and he preferred to represent actions in terms of the intended meaning, rather than in terms of just what was necessary to distinguish them at choice points. This issue arose repeatedy during the re-design effort. The game designer had strong feeings about the appearance of the contro symbos. He wanted them to have three-dimensiona depth, i.e., not appear fat in the pane of the screen. They had to ead payers eyes into the movie action. The contros aso had to fit with the game s woodsy, od-word fantasy theme: they had to ook antique and cassy. Whie these graphic stye requirements did enhance the aesthetic appea of the game, they aso biased the symbo art towards more-detaied styes, rather than toward the abstract, minimaist styes that graphic designers agree are easier to recognize [3]. The game designer wanted to minimize the use of text throughout the game. The game software can dispay a text abe under any action symbo, but the game designer fet that abeing every contro symbo woud sow payers down and divert their attention from the movieaction, as we as hinder transating the game into other anguages. I agreed, so my initia design used text abes: 1) ony on speech symbos, not on physica action symbos, 2) mainy in eary game scenes to hep payers earn the symbos, and 3) thereafter, ony where necessary to distinguish choices (e.g., when mutipe speech acts of the same type were offered). However, as is described ater, the game producer disagreed with us on this issue. RE-DESIGN: PHYSICAL ACTIONS As described above, my initia anaysis of the requirements for controing the protagonist s physica actions yieded four action categories that were to be represented by hotspots: navigating in a direction, ooking in a direction, interacting physicay with objects or peope, and memorizing events or faces. From paying the game, it seemed that the foowing sub-categories of each of these categories were needed: Navigate: forward, 45 right, 45 eft, right, eft, backward, turn around right, turn around eft, turn right, turn eft, turn 45 right, turn 45 eft, stop/stand-sti. Look: up, down, eft, right, here. Interact: hit/knock, kick, push, bock, grab/catch, duck, dodge eft, dodge right, spin eft, spin right, break free. Memorize: this. Each action sub-category woud be represented by its own symbo. Figure 1 shows eary sketches of some of these symbos.

4 CHI APRIL 1998 Figure 1: Sketches of physica-action symbos. The tota number of sub-categories (and therefore symbos) in this initia version of the physica-action contros was 30. This was worrisome, because it seemed ikey that additiona experience with the game woud expose the need for physica actions not covered by these sub-categories. For exampe, I soon rea&d that we needed a subcategory and symbo for Jump. Fortunatey, further anaysis aso indicated that some of the initia sub-categories were not needed. For exampe, it became cear that many Look opportunities in the game coud be represented by navigate symbos such as Turn Left, and the rest coud be represented by a singe Look Here symbo paced in the reevant screen-ocation. It aso became cear that because users were choosing between avaiabe actions, not generating actions, it was not necessary to distinguish Hit from Kick: both coud be represented by a singe Strike symbo. Simiary, it was not necessary to distinguish Push from Bock, or Dodge Right and Left from Navigate Right and Left. Finay, the need for some sub-categories disappeared as the game evoved: Break Free and Turn 45 right and eft. The fina set of action sub-categories to be represented by hotspots numbered 21. It contained: Navigate: forward, right, eft, 45 right, 45 eft, backward, turn around right, turn around eft, turn right, turn eft, stop/stand-sti. Look: here. Interact: strike, push/bock eft, push/bock right, grab/catch, duck, jump, spin eft, spin right. Memorize: this. Once these sub-categories had been finaized, we instructed the game editors to use them throughout the game. It didn t matter that the symbos for the sub-categories did not yet exist, because each use of an action symbo in the game is just an index into a tabe of symbo bitmaps. We fied the tabe with pacehoder images initiay, and repaced them one by one as the symbos were finaized. Even as the physica-action subcategories were being refined, symbos to represent them were designed. My sketches served as input for a graphic artist, who produced successive iterations of the symbo set with feedback from me and the game designer. Some of the sub-categories were easy to represent graphicay. It was obvious that symbos for navigation actions shoud be arrows such as those painted on roads and road-signs. The ony difficuty in designing the symbos for these was in achieving the animated three-dimensiona perspective appearance that the game designer wanted. Other physica actions that proved easy to design symbos for were: strike (a jagged starburst borrowed from comic books), grab/catch (a target), spin (an animated atrow chasing its tai), and memorize (a sma pusating coud). Once these symbos were in the game, AnyRiver s paytesting indicated that game payers easiy understood them. In contrast, some physica action sub-categories were hard to represent graphicay. I initiay sketched Stop/Stand- Sti as a stop sign, but it proved impossibe to create a recognizabe back and white stop-sign bitmap within the size and perspective constraints. We eventuay setted on a perspective X. To represent Look Here, I initiay sketched a pair of eyes, but the game designer was righty concerned that symboic eyes woud not be recognizabe, and that reaistic eyes woud ook ike something peering out of the movie rather than a symbo. We chose a zooming rectange. Other action sub-categories that were difficut to find suitabe symbos for were: Jump, Duck, and Bock. With each of the hard-to-depict sub-categories, the game designer and I worked with the graphic artist unti we had a hotspot we considered worth trying in the game, then put it into the image tabe, and judged its effectiveness for ourseves as we as awaiting feedback from AnyRiver s pay-testing. If it didn t work, we sent the artist back to the drawing board (or more precisey, the paint program). Figure 2: Fina images for some physica-action symbos. Pay-testing yieded many compaints about eary versions of the hotspots, but with revisions, the compaints ceased. In fact, it eventuay seemed that payers were using the physica action symbos without being fuy conscious of them. When AnyRiver was preparing the game s instruction manua, they asked for a ist of the hotspots. When we sent them the ist, they were surprised that there were so many distinct images. We pointed out that, with the exception of the Stop/Stand-Sti symbo, peope don t perceive the navigation symbos as distinct symbos, but 68

5 69 CH I APRIL 1998 PAPERS : rather as the same arrow pointing in different directions. AnyRiver made use of this in the manua to simpify the expanation of the hotspots. RE-DESIGN: SPEECH AND THOUGHT The speech&ought contros were designed in parae with the physica-action contros. Opportunities to speak wem to be depicti by a cartoon speech baoon encosing a symbo representing the type of speech-act. Opportunities to think interna thoughts woud be represented by a thought baoon encosing a subset of the speech-act symbos. As with the physica-action contros, we had to deveop both the sub-categories and symbos. hjy initia anaysis of conversation-intensive game scenes suggested a need for 12 sub-categories of speech: statement, question, accept, refuse, offer hep, aggressive/ /insut, sarcastic, humorous, need hep/frustrated, fatterpraisedthank, grove/beg/pead, and reca memory. An additiona speech type was to represent keeping sient (athough see discussion beow). The sub-categories for rhought were to be a subset of those for speech. After defining the initia speech sub-categories, I began sketching symbos for them. I suggested using etters for some speech types (e.g.,?,i for Question), faces for some (e.g., a comedy mask for Humorous), and hand gestures for others (e.g., a shaking fist for Aggressive). The gamedesigner preferred a scheme having a common visua theme for a speech symbos (in addition to their being inside the speech baoon), and decided that we woud use human figures making mime-gestures. Figure 3: Initia sketches of speech figures. Based on my initia sketches (see Fig. 3) and input from the game designer, an artist (not the one who created the hotspots) drew human figures miming the current set of speech subcategories. Bach mime-figure consisted of two frames, which woud aternate rapidy on the dispay to produce a simpe animation (see Fig. 4). Initiay, the game designer and I met with the artist amost weeky over a two-month period to critique the atest mime-figures and inform the artist of changes in the speech subcategories. By this time, based on further experience with the game and pressure from the game designer (who tended to favor more specific speech sub-categories), the number of speech subcategories had expanded to 27. They were: statement, question, accept/yes, decine/no (poite), refuse/no (adamant), request (poite), demand (adamant), greet, offer/hepfu, excaim, ie/exaggerate, whineseeksympathy, accuse, gratitude/thank, fatter, humorous/witty, sarcastic/snide, aggressive/insut/defiant, grove/beg/pead, firt, manipuate, reason/persuade/expain, care/empathy, dea/negotiate, get-rea/get-a-grip, and reca memory. Figure 4: Artist drawings of animating mime-figures. Unike the hotspots, which had been drawn using painting software, the new speech symbos and the speech baoon were drawn on paper. Digitizing them and combining the mime-figures with the speech baoon woud be very expensive and time-consuming, so we wanted to wait unti we were more confident that the speech sub-categories were stabie and the symbos were adequate. But this meant that the speech contros weren t being pay-tested. Speech Symbo Participant Responses -I LeExaggerate 1. Exaggerating or ying. 2. tie. Teing a ta tae. 3. Teing a ie. 4. Bigger. 5. He tod a big ie. 6. Teing a fish story. 7. tie. 8. The Pinocchio syndrome. 9. Teing a big ie. 10. Lying - Exaggerate. Sarcastic/Snide 1. Nudging. Getting their attention. 2. Bravado, brag. 3. Let me te you. 4. Hey, big guy. 5. Don t you get it? Heh, heh. 6. D ya get it? Huh? : 7.??? 8. Anger. ibit I m the man. I I 10. Coughing. Tabe 1: Exampe responses to paper speech symbo test. The upper symbo conveyed its intended meaning; the ower didn t. To get empirica data on how we the mime-gestures conveyed their intended meanings, I conducted informa 1

6 - - _.._ -_II_-,_.., PAPERS CHI APRIL 1998 paper tests: I recruited ten vounteers who were not associated with Advance Reaity or AnyRiver and knew nothing about the game, showed them the figure-pairs, tod them that each pair was an animation of a man trying to express something sienty, and asked them to write on an answer sheet, in a few words, what the man was trying to say. I then assigned each response a subjective score between 1 (missed intended meaning) and 10 (got it). The resuts of the first such test were discouraging: a few symbos were conveying their intended meaning to most subjects, but most were not (see Tabe 1). I reported the poor test resuts to the artist and game designer, and we changed the basic miming figure, revised some of the symbos (see Fig. 5), and repeated the test with different participants. Some symbos did better in the second round of testing; others didn t. Based on the resuts of the second test, we revised some of the figures again. Figure 5: Exampes of revised mime-figures. We briefy considered adding coor to the mime-figures as a redundant indicator of mood (in contrast to the od scheme s use of coor as an independent dimension). To be hepfu, the coor scheme had to be obvious to payers, rather than something additiona to earn. Further anaysis reveaed that such a pan required finding not ony a universa mapping of emotion-eve to coor, but aso a universa mapping of our speech subcategories to emotion-eve. To check the atter, I conducted a quick informa test: I asked eight empoyees of Advance Reaity who weren t invoved in deveoping the speech symbos to rate each speech subcategory for amount of emotion on a O-5 5=high). Some subcategories received uniformy high emotion-ratings (e.g., accuse, excaim, refuse), whie others received uniformy ow ones (e.g., statement, keep sient), but most received a wide range of ratings (e.g., accept, fear, manipuate, humorous). Comments from test participants suggested that variabiity for certain items was due to some peope separating amount of emotion from Qj7e of emotion whie others did not. Since finding the required two universa mappings seemed unachievabe, we abandoned the idea of augmenting speech. symbos with coor. The speech symbos sti weren t satisfactory, and time was getting short. We decided to put some of them into seected game-scenes and use our own reactions and pay-testing to guide further design. This exposed two new issues: On-screen, the mime-figures were too sma to be easiy recognized. Enarging both the figures and the encosing speech baoon woud obscure too much movie action. Our soution was twofod. First, we enarged the figures but repositioned them in the speech baoons so that their egs were cipped off. A the information was in the upper bodies anyway. Second, we exaggerated whichever features carried the meaning (e.g., hands, mouth). This is a common practice of cartoonists and graphic artists [31. Athough the two-frame animation of the mime-figures worked we, it proved distracting and disorienting to have severa speech symbos dispayed at once, a animating in synchrony over the movie action. Simpy assigning different animation rates to different mimefigures broke up this effect, reducing the visua impact of mutipe speech symbos to acceptabe eves. Once these graphic-design issues were resoved, the main issues returned to the fore: determining the fina speech sub-categories and symbos to convey them. The deadine was ooming and we had to aow time for the editors to add the new speech subcategories throughout the game. We were worried that we had too many speech sub-categories for payers to remember: 27, not incuding the thought subcategories. We were aso concerned that many of the mimefigures didn t convey their intended speech-category we. To finaize the speech sub-categories and symbos, we had to resove severa issues about which the game designer and I had different opinions: Did we need speech sub-categories that were proving ver) dijficut to depict? These were: sarcastic/snide. reason/persuade/expain, care/empathy. dea/negotiate. whine/seek sympathy, and fatter. I had concuded that the probem was not simpy that we hadn t yet found the right mime-figure to depict these concepts, but rather that they were too abstract to represent graphicay. I convinced the game designer that we shoud drop them and bend other sub-categories to cover those situations. Did we reay need mutipe symbos for thoughts? The game designer preferred having a variety of thoughtsymbos to depict the different types of sient remarks the protagonist coud make. I argued that thoughts can t be heard by other characters and don t affect the outcome of the game (most are just opportunities to hear one of Rob Schneider s funny wisecracks), so there was no need to distinguish one from another. We ended up with ony one Thought symbo (an empty thought baoon). 70

7 71 fh APRIL 199s PAPERS.+ SJ Did we reay need a symbo for not speaking? I favored eiminating the symbo and having payers keep sient by not seecting any speech symbo. The game designer advantages of text abes outweighed the disadvantages. I went back through the game, composing and adding terse abes for each speech opportunity. wanted keeping sient to be an expicit choice because 2. They wanted the set of speech symbos reduced further. there are situations in the game where remaining sient is They were preparing the user manua and were worried a strategic move. We compromised: eiminate Keep that the ist of speech symbos woud seem daunting. Sient, use no symbo wherever keeping sient isn t The game designer and I fet that the set was as sma as strategic, and use the Thought symbo wherever keeping it coud be. A compromise was reached, based on the sient is strategic [athough this meant finding and adding recognition that in most of the game, ony a few speech a suitabe recorded protagonist thought ine). sub-categories were needed (seven), but some scenes * Did we reay need symbos that were hanfy used? Two were Lconversation games requiring more speech symbos were needed ony once or twice in the entire game: manipuate, get-rea/get-a-grip. They seemed to me to be good candidates for eimination, but the game designer wanted to keep them because he fet that the attitudes they represented were critica for the scenes in which they appeared. We kept them. We aso added a symbo for Shhh! (i.e., teing someone ese to be quiet), even though it was ony used once. With these decisions behind us, the set of speech/thought sub-categories numbered 22. We fet comfortabe with this number and with how the remaining mime-symbos were working, so we began instructing the game-editors on how to insta the new speech sub-categories throughout the game. Assigning speech sub-categories required much more care and judgment than did adding physica-action subcategories. I deveoped guideines for editors to foow. Unfortunatey, instaing the new speech contros was as difficut for editors to do correcty as the od ones had been. After the editors had made one pass through the game assigning speech symbos, I did some checking and estimated that about haf of the thousands of assignments were incorrect. I revised my guideines to be more expicit, and expained them verbay to most of the editors, to no avai. The game designer asked the editors to be more carefu, to no avai. The editors had a compex job, which incuded editing together video sequences, voice-tracks, and sound effects as we as pacing the contro symbos in space and time. Furthermore, they were working under intense timepressure. One finay tod me that she didn t have time to determine the correct speech sub-categories; she simpy made everything a Statement, and assumed that someone ese woud fix it ater. In response to this, I earned to use the editing software we enough to aow me to go through me entire game, isten to a the protagonist s ines, and assign speech subcategories. Once the speech contros were in the game, feedback from AnyRiver (based in part on their pay-testing) was swift in coming. They wanted two further changes: 1. They wanted a speech symbos to incude text abes. Athough the game designer and I had decided to use text very sparingy, AnyRiver convinced us that the expressiveness and hence more speech sub-categories. AnyRiver wrote the manua so as to ist the seven frequenty-used speech symbos separatey from the rest. Ansv/er Questrons *r make staemens Ask quesrons Express anger or make a rhrearenq srarement Figure 6: Exampes of fina speech-figures. REDESIGN: EVALUATION AND DISCUSSION One goa of the re-design effort was to reduce drasticay the number of contro symbos game payers had to distinguish and choose between without compromising the expressive choice that was needed to make the game entertaining and rewarding. We ended with 21 physica-action symbos and 22 speech/thought symbos: a tota of 43. Obviousy, this was a drastic reduction from the severa hundred symbos of the od contro scheme. The game designer aso fet that we had done so without compromising expressive choice. In critiquing the od contro-scheme, I treated the ack of compiance by game editors with it as an indictment of it. However, the editors aso had troube assigning the new speech sub-categories. It is tempting to concude that this has more to do with their heavy work-oad than with the contro scheme, but that concusion won t withstand the observation that editors did not have much troube instaing the new physica-action contros. Perhaps workoad contributes to the probem, perhaps deciding the appropriate sub-category for a speech act is as hard as is devising a good categorization of speech acts, and perhaps our speech categorization was sti not intuitive enough. We can aso ask how we the fina contro symbos depicted their intended meanings. My feeing is that the symbos for physica actions and memorizing were highy

8 72 PAPERS CHI APRIL 1998 successfu: so good that payers hardy notice them even whie using them heaviy. Whether the speech/thought contros were successfu is debatabe. In the end, we had to augment them with text abes. Pay-testing indicated that, thus augmented, they were usabe. With more time, I might have suggested revising the mime-figures yet again or trying other ways to represent speech acts. OTHER ASPECTS OF THE USER INTERFACE Athough I had been hired to redesign the action-contros, the game deveopers expanded my charter to incude other aspects of the game s user interface that needed refinement or redesign. These incuded: the cursor s various shapes, which indicate whether or not the movie is at a choice point and how much time remains to choose; the visua feedback when contro symbos are hit or missed; the score counters that appear in certain scenes; the user interface for making use of saved magica objects; and the contro pane for starting, saving, and oading games and setting game options. Pay testing indicated that these aspects of the user interface were successfu, but the processes by which they were designed are beyond the scope of this paper. LESSONS AND CONCLUDING THOUGHTS Working on Fork taught me the foowing essons: A picture may be worth a thousand words, but finding the right picture to convey a verba concept can be very hard. The best way to convey some verba concepts is verbay. If symbos in a set depict their meanings we, peope can discriminate and recognize them even if the set is arge. Users may not even reaize that the set is arge. Universay recognized mappings between coor and emotion depend on a universa emotion scae, which may not exist. Designs that map coor to emotion are risky. Game deveopers understand the need to test on rea users (better than do most software deveopers), but they coud use hep from user-interface speciaists in understanding the distinction between pay-testing and more focused usabiity testing, and the importance of the atter. User-interface designers have different concerns and skis than game designers and graphic designers. User-interface skis are as important for designing games as for productivity software, but the reverse is aso true: the skis of game designers coud be usefu in making productivity software more engaging and fun. Earier, I stated that the game designer often yieded to my judgment on user interface issues. But what is a userinterface issue, especiay in contrast to a gamedesign issue? The game s pot and diaogue and the chaenge of soving its mysteries and puzzes ceary fa under game design. Users nabiity to navigate successfuy in the gameword and keep track of how they are doing ceary fas under user-interface design. But consider the issue of deciding who the intended users are and designing a suitabe product for them. That is often cited as an important, though negected, concern of userinterface designers [4]. After I had payed Fork, I bad concerns about who it was for and whether it woud appea to them. As someone who rarey pays computer games, I fet that the interactive-movie aspects of Fork were great, i.e., it was easy to become engrossed in the mystery and wanting to sove it. However, I was put off by some embedded puzzes that were not integra to the story: points at which the movie disappears and the payer (not the protagonist) negotiates a maze or manipuates objects to find a combination. These were incuded to make Fork appeaing to hard-core gamer?. I worried that they might make it unappeaing to ight-garners and non-garners, and that we ought to think more about the question of who Fork was for rather than giby answering everyone! but even 1 didn t consider this a user-interface issue at the time. Simiary, AnyRiver, Fork s producer, prepared cover art for the package and manua that incuded sophomoric drawings of near-naked women, even though there is nothing so racy in the game. Again, this was done to appea to garners. but I fet that it woud turn off non-garners, ight-garners, parents (who often contro purchases), and femae garners. As it turns out, the game is seing rather poory. I franky don t think that this is due to poor contros: pay-testing indicated that they were adequate. I think it is due to insufficient consideration of the target market. Maybe that shoud be treated as part of user interface after a. ACKNOWLEDGMENTS The author is gratefu for the opportunity to work with the creative designers and deveopers at Advance Reaity, especiay fim director and game designer Rob Lay and ead programmer Ken Carson. Thanks aso to the graphic artists at 415 Productions and McMacken Graphics, to staff members at AnyRiver who contributed ideas to the user interface, especiay Chuck Canton and Stewart Bonn, and to severa anonymous CHI 98 reviewers. NOTES AND REFERENCES 1. It is common for three companies to be invoved in the preparation and marketing of a computer game: the designer/deveoper, the producer/funder, and the distributor. The distributor of Fork is Eectronic Arts Johnson, J., Roberts, T., Verpank, W., Smith, D.C., Irby, C., Beard, M., and Mackey, K. (1989) The Xerox Star: A Retrospective, IEEE Computer, September. 22(9), pages Muet, K. and Sano, (1995) D. Designing Visua Inte@xes, Mountain View, CA: SunSoft Press. Landauer, T.K. (1995) The Troube with Computers, ambridge, MA: MIT Press.