Introduction & Theory

Understanding how novices become experts has been an important concern in information-processing approaches to learning. Anderson (1999) described several stages of skill acquisition, showing how people gain expertise by first focusing on declarative knowledge and explicit rules about how to operate within a domain (the "cognitive stage," pg. 273) then transition to an "associative stage," where an expert internalizes a "successful procedure for performing [a] skill" that previously required declarative knowledge to perform (1999, pg. 274). Learning these procedures appears important for expertise, and, according to Anderson, lead to representational differences between experts and novices.

Citing Chase & Simon's study of expertise in the game of chess, Anderson stated that "experts seem to display a special enhanced memory for information about problems in their domain of expertise" (1999, pg. 292). In the study, chess experts and novices were presented pieces on a chessboard then asked to reconstruct board configurations (both from real games and random patterns) on a second board nearby. Participants were forced to turn their heads to glance at the stimulus board when reconstructing positions on a second board, as a means to capture how many (and which) pieces participants chose to place at any given time.

Chase and Simon showed that experts represented visual components of meaningful chessboards differently than novices, while not differing on the random chessboards. According to Anderson, "Chase and Simon defined as a chunk those pieces that subjects moved following one glance. They found that these chunks tended to define meaningful game relations among the pieces," such as pawns in clustered, defensive positions (1999, pg. 293-294; emphasis mine). Expertise thus manifested within the structure of mental representations; expert chess players perceived meaningful structure of the chess positions that novices did not and unsurprisingly did not differ on random patterns.

Historically, games have proven to be useful testbeds for theories of expertise, as with Chase and Simon's chess study. Do their conclusions generalize to other kinds of game environments, even those which seem to require much more than attending to the visual representation of the game's "pieces"?

In this paper, I will show that expert players of the interactive, multimodal game Guitar Hero II (GH2) organize the visual content of the game in similar ways to Chase and Simon's chess experts. I hypothesize that expert players of GH2 will chunk the visual information from real game screens in meaningful ways, while novices will not. As a control case, I suggest that experts and novices will not significantly differ in their chunking of random game screens which presumably will not have imbedded meaningful structures for the experts to identify. Even in a complex, physical game such as Guitar Hero II, I propose that expertise fits Anderson's theory.

Setting & Methods

Guitar Hero II (2006) is the second in a series of popular music simulation games. Like other rhythm-based games (e.g, Dance Dance Revolution, Karaoke Revolution), GH2 players synchronize motion along with the rhythm of a song and game elements on a television screen. It uses a controller designed to mimic the look and feel of an electric guitar with buttons simulating frets, a picking button, and motion sensitivity to activate "Starpower" (which boosts one's score through difficult spots in the game). Like a real guitar, "notes" are created when the player holds down the correct fret button while striking the picking button. Failure to pick a note at the correct time or picking an incorrect note increases the chance that the player fails the song.

GH2's songs are each playable at four difficulty levels: Easy, Medium, Hard, and Expert. These make it straightforward to operationalize expertise -- players who had previously completed the game on either the Hard or Expert difficulty level were assumed to have a degree of Guitar Hero II expertise (and were thus Experts), while players who had not completed the game on even Medium difficulty were classified as Novices.

Five participants participated in the study, three as experts (referred to as E1-E3) and two as novices (N1 and N2). Participants' personal gaming and music expertise varied quite widely -- both E1 and N2 professionally studied games and had several years of musical experience, however E2 and N1 were both students with some gaming experience but no guitar experience.

Task One

Each participant was asked to play through a song on Easy: "Mother" by Danzig, chosen because it contained a variety of song features (including solos and repeated chord patterns). If a participant successfully completed the song, he or she was instructed to next complete the same song on Medium, then Hard, then Expert. Each performance was videotaped for Task Three, and data were gathered on each participant's accuracy on each performance.

Task Two

Next, each participant was asked to complete a paper and pencil task, similar to Chase and Simon's chess expertise task. Participants were presented with paper representations of two potential Guitar Hero II screens: "meaningful" (designed for actual gameplay), and "random" (containing the same number of notes presented in an arbitrary order; see Figure 1 below).

Figure 1: A game screen was captured, translated into an abstracted version (the "meaningful pattern"), then a random pattern with the same number of notes was created (the "random pattern").

These simplified Guitar Hero II screens were presented to each participant in random order, one at a time. The participant was seated at a table with a paper printout of either the meaningful or random sheet behind him or her, and was asked to redraw the stimulus pattern onto a blank staff in front of them. Given the participant's physical position, each had to turn around to see the stimulus page, and, given the number of notes, often needed to turn several times. The order of each note's replication on the blank staff was recorded, along with each head turn toward the stimulus page.

Task Three

Finally, each participant reviewed the videotape of their last successful performance with the experimenter (on Expert for all three Experts, Medium for both Novices). Participants were prompted to do "retrospective thinkalouds," pointing out anything they noticed about how they played which they deemed of relevance. This was also an opportunity for participants to describe their own theories on how one develops expertise in the game.

Results

Task One

All Experts completed the song on Expert difficulty while both Novices completed the song on Medium but neither was able to complete it on Hard. Even though the Novices completed the song on Medium, it was clearly a difficult task for them, more difficult than Expert difficulty was for the Experts. On the last-completed song (Expert for Experts, Medium for Novices), experts were generally more accurate (92%, 92%, and 94% correct, versus 69% and 88%) while still facing a greater number of notes (394 for Medium vs. 625 for Expert).

Task Two

The redrawing task was designed to capture the structure of players' mental representations of GH2's graphical information (the "notes" on the screen). Like Chase and Simon's chess study, a chunk of recalled notes was operationalized as consecutive notes redrawn on the blank staff between head-turns toward the stimulus page.

The sizes of the first and second chunks redrawn on the page were calculated, as shown in Tables 1 and 2 below.

All of the Experts had three notes in their first Meaningful chunks, and very similar sizes for their second Meaningful chunks (4, 3 and 3). I saw no patterns between Experts and Novices on the Random patterns for either chunk, nor commonalities among the Novices' chunk sizes.

Task Three

The "retrospective thinkalouds" elicited responses regarding participants' knowledge of game mechanics, how they positioned their bodies, and how they believed they learned to play. Notably, neither of the Novices understood how the Starpower function within the game worked (N2 assumed it was the same as Karaoke Revolution, when it was not). All three Experts used Starpower extensively during play.

Additionally, both Novices declared that they felt like they were still learning how to use the controller. With regards to syncing up the position of each dot on the screen, associating it with a specific color, then the appropriate button to strike on the controller, N1 stated:

"I didn't notice the notch on the yellow one, until I had to move, then I felt it. I'd have to make some kind of map in my head that the one with the notch is yellow."

The "notch" refers to a bump on the yellow (third) key on the controller, similar to notches used to center touch-typists on QWERTY keyboards.

E1 made an interesting analogy to how one synthesizes the visual information and activity in other rhythm-based games (on-screen arrows and dancing moves in Dance Dance Revolution):

"[A] lot of the literacy of [Dance Dance Revolution] is looking at those arrows and knowing where to shift your weight and what leg to use, and it's not on the screen. And I feel like it's the same thing with Guitar Hero. There are things about your physical fingers that are implied by those ... certain shapes or patterns. You wouldn't understand that this is a skill you need to pick up just looking at [the screen]."

Discussion & Conclusion

After establishing that Experts could exhibit expert performance on this song while Novices could not, it was determined that there was no variation in the size and therefore structure of the chunks: Since all Experts re-drew the staffs in the same direction, each three notes were the same for each Expert. Additionally, looking at the subsequent chunk on the meaningful staffs, we see nearly the same degree of uniformity, with only one expert chunking four instead of three notes. We can hypothesize why this might be the case (see Figure 2, below).

Figure 2: Experts' first two chunks on the structure staff; a "pull-off" (in blue) followed by a "hammer-on" (in orange).

The first three (blue, in Figure 2) notes indicate, in the parlance of guitarists and Guitar Hero players, a "pull-off," or a series of notes which are to be played by first positioning one's fingers on several buttons, then pulling each off in quick succession. A task of the game on the Hard and Expert levels, it is no surprise that Novices did not chunk these notes in the same fashion, since they had likely not encountered it yet. Additionally, the subsequent three or four (orange, in Figure 2) notes correspond to a two-note "pull-off" then "hammer-on" sequence in the game, or a series of notes played in quick succession by forcefully moving one's left hand to subsequent buttons after only picking the first note with one's right hand. As with Chase and Simon's chess experts, experts in Guitar Hero chunked the visual information of the game in ways that correspond to meaningful structures within the game -- in this case, structures which indicate hand motions encountered at higher levels of the game.

Therefore, there is strong evidence that experts of Guitar Hero II chunk the game's visual information in meaningful manners. But, how might we account for the way this skill is applied? Upon investigation of participants' explanations of their expertise (or lack thereof), we can see that they match Anderson's first two stages of skill acquisition (cognitive, then associative).

First, N1 described "mapping" the visual presentation, mental representation, and controller actions during her gameplay. Using the contrller was difficult for her, and her inclination was to state declarative, cognitive strategies for learning how to use it. Her use of the term "map" indicated that learning to play involved a degree of rules -- be it learning how to implement Starpower within the game (a game mechanic), or simply which button corresponds with which event on screen. She recognized this as the task and framed it using explicit cognitive terminology ("mapping"), as she is still developing representations which Experts use to chunk the game's visual information associatively.

In contrast, Experts appeared to function well at an associative stage of expertise -- patterns indicating "pull-offs" and "hammer-ons" involve different physical motions than those used in the lower levels of the game and thus require quick association with actions in order to properly execute. As E1 stated above, like Dance Dance Revolution, GH2 requires experts to go past the information presented on the screen, and mentally represent patterns which imply specific (and sometimes complex) motions. The process of expert performance is to perceive structure in game environment and quickly associate those structures with actions.

In conclusion, it appears that Anderson's (1999) theory is still applicable to understanding expertise, even in complex, multimodal game environments. The chunking of visual information and conformity to Anderson's stages illustrate that, even in highly interactive games such as GH2, an information-processing approach can help to explain the mechanics of expertise.

References

Anderson, J. (1999). Cognitive psychology and its implications. New York: Worth Publishing, pp. 272-304.

Guitar Hero II. (2006). [PlayStation 2 software]. Santa Monica, CA: Activision.