Spoors, G. (2000). Computational Semiotics and Ergodics: The “Works” and “Texts” of Gameplay. [Manuscript].

Computational Semiotics and Ergodics: The “Works” and “Texts” of Gameplay

“Computational semiotics” refers to the study of the computer as a medium, or distinct sign system. Its central concern is the extent to which the information inside a computer can be considered semiotic or can be usefully analysed in semiotic terms (Anderson, 1990; Anderson, Holmqvist and Jensen, 1993; Aarseth, 1997). Jens F. Jenson (1991) has argued that the threshold between information and semiotics in human-computer interaction is the user interface, and the same assumption informs Peter Bogh Anderson's (1991) taxonomy of signs at the interface. Anderson, who argues that Pierce's tripartite distinction between “symbols,” “indexes” and “icons” is not sensitive to the way users interact with computer-based signs (1991, p. 199), offers his own taxonomy. This accounts for the transformation of signs during interaction with such criteria as “permanence,” “transience,” “handling” and “activeness.” However, as Aarseth (1997, pp. 31-39) indicates, Anderson's taxonomy is not rationally consistent, and does not adequately account for the variety or dynamics of sign-functions.

Indeed, by presuming that sign-functions inhere in the signs themselves, Anderson's taxonomy contradicts Umberto Eco's (1976) argument that readers assign sign-functions on the basis of convention and contextual requirements. Eco usefully defines a lower threshold of semiotics, drawing from Charles Sanders Pierce's (1931) definition of a sign as “something which stands to somebody for something in some respects or capacity” (quoted by Eco, p. 15). According to this definition, the relationship between signifier and signified is not inherent in the sign itself, but assigned by an “interpretant” on the basis of convention. Consequently, bioelectrical impulses, binary bits, assembly language, C+ code, and words or images on a monitor, are merely processes until an interpretant assigns them “sign-functions.” The existence of a “human addressee” is therefore the only “guarantee” of semiosis (p. 16), though Eco adds that “to assert that stimuli are not signs does not necessarily mean that a semiotic approach ought not to be concerned with them” (p. 20). In terms of computer games, computational processes may have, acquire, or be read as having, a semiotic structure which partially determines how they signify.

Using Eco as a basis, Aarseth (1997) argues that in analysing computational media it is necessary to look beyond the interface to “where the real action is: in the mathematical reality beneath the surface, where the relations and objects of the system are being processed” (p. 39). This is not to deny Eco's argument that sign-functions do not inhere in objects; Aarseth is merely concerned with how semiotic structures are distributed within or by a medium. To this end, he distinguishes between potential sign sequences in a medium, called “textons,” and signs selected to appear at the surface of a medium, called “scriptons,” and uses this distinction as a basis for classifying different media. In some media, the scriptons are highly determined by the textons. In a book, the textons and scriptons are identical: the words on the page. However, we might say that sounds we produce while reading a book (scriptons) are highly determined (in syntactic terms, anyway) by the marks on the paper/ink (textons). In film, the images projected on a film screen (scriptons) are highly determined by the images on celluloid (textons). In the case of computers, by contrast, the data in memory (textons) is mediated by layers of code that determine how it is represented at the interface (scriptons). Aarseth's example is the internal structure of an adventure game (figure 1), in which information in the database (textons) are mediated by: the representation engine, before being output to the screen (synthesis, or presentation) as a scripton to be read by users. Users, usually in response to this output, enter new data that is analysed by the system (parser) and then implemented (simulation engine), acting on the state of game data.

Figure 1. Aarseth's schematic model of a computer game's internal structure (1997, p.104)

As is evident in this model, any technical description of the process whereby media produce “scriptons” from “textons” cannot be separated from the potential and actual ways in which a human operator intervenes in the process: the choices (input) of the player. Aarseth uses the term “ergodic,” derived from the Greek words ergon (work) and hodos (path), to refer to any text whose structure is physically altered by the activity of a “reader.” An ergodic text allows the reader to physically select from, alter, or input text, producing a different text that is read from . Aarseth describes the ergodic text as “textual machine,” comprised of the relationships between a sign, the medium, and a human operator:

Figure 2. Aarseth's model of the Textual Machine (1997, p. 21).

As Aarseth acknowledges, his notion of “text” “is closer to the philological (or observable) work than to the poststructural (or metaphysical) galaxy of signifiers” (p. 20). While readers of such non-ergodic texts as traditional film and print fiction perform the work of reading “in their heads,” the user of ergodic texts:

performs in an extranoematic sense. During the cybertextual process, the user will have effectuated a semiotic sequence, and this selective movement is a work of physical construction that the various concepts of “reading” do not account for. (p. 21)

Aarseth includes hyperfiction, adventure games, cyborg-authored texts, and MUDs as examples of ergodic literature, though he does not confine his examples to digital media, arguing that even readers of print text co-create the semiotic sequence through turning or skipping pages.

Aarseth offers seven variables as a means of classifying the mode of traversal, or ergodic distinctiveness, of texts (pp. 62-65). If we define graphical representations and the correlative data or processing which produces them as scriptons and textons, respectively, we might apply these variables to FFX as follows:

1. Dynamics: whether the number and content of textons and scriptons are fixed (“static”), as in printed texts, or vary, as with most digital texts, either in terms of the content (“intratextonic dynamics”, or IDT), and/or number (“textonic dynamics”, or TDT). In FFX , the only way characters can alter scriptons is when the game initially asks the player to enter characters' names.

2. Determinability: whether textons repeatedly determine the same scriptons, or if other factors, such as randomisation, create variation (hence “determinable” or “indeterminable”). Randomisation in FFX is limited to the appearance, number and type of wandering monsters, and the unfolding of each event in battle (the chance of hitting an opponent and the amount of damage inflicted). So long as the player follows the same actions, events will in all other respects be identical if the game is replayed.

3. Transience: whether scriptons are generated by simply the passing of time (“transient”) or the user's actions (“intransient”). In some Final Fantasy titles, there is an option to enter real-time in combat, so that we must choose an action before the opponent attacks. FFX s combat, however, is a Conditional Turn-based (CTB) system, in which time only passes when players choose option. In the navigation screens, non-significant activity—the movement of trees and butterflies, and so on—occurs, but otherwise events, including wandering monsters, require the character to move. It is only in some sequences, such as when the party is chased while escaping Guadosalam that events, in this case combat, may occur without the players input.

4. Perspective: if the text requires the user to play a strategic role as a character in the world, then it's perspective is “personal”; if the player merely “reads”, then its perspective is “impersonal”. FFX requires the user to play several characters in the world, and so would be personal, except that during cut-scenes the player merely watches (“reads”) the game, and so in these terms is impersonal.

5. Access: if all scriptons are instantly accessible (“random” access), or access depends upon the user's “position” in the text (“controlled” access). Access to narrative information in FFX is linear, while access to many options requires moving through menu-sequences. Most cut-scenes only play the first time an area is entered, but as the character traverses more of the world, acquires the Airship, and equips any item with the No Encounters Ability, it becomes easier to access much of the text more quickly. The save facility allows for repeated access to “earlier” sections, and the Theatre in Luca allows players to “buy” access to the game's songs and cut-scenes.

6. Linking: if scriptons are accessed through links that are fixed (“explicit”), links that are provisional upon certain conditions being met (“conditional”), or there are no links (“none”). In FFX , most scriptons may be repeatedly accessed through the same explicit ergodic sequence, but some links may be blocked until certain actions have been performed. For example, access to the Anima Fayth in the Baaj Temple requires collecting all the Spheres of Destruction in all the Trials of the Fayth.

7. User-functions: alongside the “interpretative” function, common to all texts, an ergodic text may be: “explorative”, when the user can choose which path to take; “configurative”, when scriptons may be chosen or created by the user; and “textonic”, if changes to textons or traversal is permanent. (p. 62-65). In FFX , there is: a strong explorative component; a minimal configurative component, in that players can alter some game options; can choose how to equip characters, which questions to ask some non-player characters; and a limited textonic component, through entering character names.

That Aarseth's typology is only briefly elaborated is most likely a consequence of his stated emphasis on “popular convention than [his] own model” in his discussion of the “genres” of hypertext, adventure games, cyberfiction, and MUDs (p. 18). Nonetheless, while he might have argued the case for the use of his terminology, as it stands there are some rational inconsistencies and redundancies. The “user-functions” of “configuration” and “textonic” overlap with the “dynamics” of the text, in that by editing or adding scriptons, the dynamics of a cybertext may change. The distinction between “determinability”, “access” and “linking” could be clarified, since: linking determines access; the conditions that need to be met for a link to be accessed may include randomisation; and accessing a link “early” in the text—for example, by entering a password on an options menu—may be the sole condition for greater or complete access. Furthermore, the “user-function” of exploration overlaps with “transiency”, which includes the possibility of the user as cause of traversal through the cybertext, and with “access” and “linking”, in that unless a digital text unfolds without player input, and links and access are determined randomly, traversal presumes a user.

The category of “perspective” is the most problematic. The term “perspective” itself does not distinguish between the data structure and its mode of representation. Indeed, the terms “personal” and “impersonal” invoke the terms discourse and histoir , which distinguish between whether a narrative foregrounds its narration as subjective, or disguises itself as objective representation, respectively (Culler, 1975; Genette, 1980). The dynamics of focalisation, which may utilise diegetic and stylistics codes from various media and genres, and the reader's perceptual and conceptual perspective(s), operate along a complex spectrum between discours and histoir (Branigan, 1994; Genette, 1980; Lanser, 1981). If this category is preserved, either its terminology must be better elaborated, or it must accommodate far more varieties of “perspective” whilst remaning distinct from the actual perspective experienced by the player, since processes of identification, empathy and projection may lead to diverse relationships to characters and situations. Indeed, aspects of “perspective” might be seen as overlapping with: transiency, where the user's “real-time” is integrated into “text-time”; “access” and “linking”, in that a player's strategic role emerges through the formation of a virtual map of the ergodic text; and the “user function” of “interpretation”.

Consequently, while Aarseth suggests that a single text must occupy one of the 576 unique media positions projected by his chosen seven variables, many of these dimensions could be reduced through terminological precision. Indeed, as Aarseth's own examples show, a single text may (initially, potentially and/or dynamically) occupy multiple positions. In FFX , the textonic user-function is so limited that the text can hardly be categorised as textonic upon that basis. If we were to categorise FFX in the way Aarseth attempts, we would have to distinguish between ergodic sub-systems in a game—the Menu, Battle, Navigation and other screens in FFX —and calculate results of their position on Aarseth's grid on the basis of an aggregate score, or provide a separate scores for each ergodic sub-system. It suffices to say that Aarseth's characteristics are useful less as a quantitative taxonomic device than as a way of qualitatively identifying ergodic distinctions in cybertexts. Indeed, as is argued below, what is of more general import is the way in which expectations about ergodic access are blocked, foregrounding a distinction between the ergodic structure and the aesthetic predisposition (and the emotional state) of the player.

This inconsistency within a single game must be extended to the inconsistency between different gaming platforms. Computers have, of course, largely realised Babbage's Analytic Engine by becoming “universal machines”—that is, any computer can, theoretically, perform the same calculations as any other computer (Campbell-Kelly and Aspray, 1996, pp. 45-56). In practice, however, the quality of playing any particular game depends upon a specific platforms configuration of: the hardware (the motherboard, the graphics and sound cards); the software (the same game, ported to a different platform, may be somewhat different from other versions); and peripherals (the monitor or TV, the keyboard, and the joystick and/or game controller). Inasmuch as hardware components may be readily switched around (upgrading a PC graphics card or installing a mod chip on Playstation console), a single piece of software may be altered or periodically extend to additional programs or databases (using hacks, construction editors or expansion packs on PC games), and the interface may change (swapping a small, staticky black and white screen for a large screen liquid TV screen) it may be that the extent of the “work” is inconstant throughout the duration of a game. While in Aarseth's generalised model of the adventure game the terms “scriptons” and “textons” refer to two parts of the same “work” (the inner workings and its “surface”), in practice the complex, modular and dynamic character of hardware and software means that it is sometimes difficult to talk about elements of a single “Work” throughout an entire gaming experience.

If one is truly concerned with a taxonomy of ergodic structure of cybertexts, one might argue that one needs to trace all the components of a particular system. However, the possibility, practicality and necessity of this in an analysis of the meaning and experience of gameplay is arguable. Of course, as Aarseth argues, the postmodern rhetoric about the “unmappability” of texts may be unwarranted hyperbole, in that while the meaning of “texts” is polysemic and has no absolute closure, the structure of many of the “works” of gameplay may be usefully mapped preparatory to an analysis of use. For example, Michael Joyce's hyperfiction Afternoon has 539 “lexies” (discrete blocks of text analogous to Aarseth's “scriptons”), and the links between them are perfectly visible when it is loaded into the hyperfiction editor Tail-Spin (1997, p. 88). If, like some auteurs, a game designer not only writes the entire program code but designs their own graphics engine, we may presume that they know the extent of their “work.” By extension, if one knows the workings of the hardware, can read machine assembly code, and/or the relevant program code, it is possible to map the information flow of almost any computer system.

Of course, the sum of designers for computer games had increasingly become more like the production crew for a film (Friedman, 1995, p. 77), in that no one knows the entire production process. The closing credits of FFX certainly indicate that, like most of the Final Fantasy titles, it is not the product of any single authorial agent, but the product of a collective. Even if a player can read program code, they are unlikely to have access to FFX s code. If we take into consideration the compound nature of any computer—especially when extended through a Network—and the limited knowledge of most users, then the game as a system of “works” may well be “beyond the capacity of the normal reading mind” (Jameson, 1985, p. 80). Yet this “loss” of the game “Work” or “Text” hardly leads to Jameson's “hysterical sublime” or “high tech paranoia,” except perhaps when a user blames Bill Gates, the Microsoft Monopoly, Capitalism or Technology as the cause of his/her personal frustration. In many instances, it is simply the case that much of the “work” is not functionally relevant from the point of view of gameplay. It is, then, unreasonable to suggest that analysts understand all the hardware and software involved in gameplay, and the issue becomes instead on how players construct a textual model of the ergodic relationship, and what role this model has in gameplay.

“Text(s)” of the Computational “Work”

Aarseth's original model of the textual-machine (figure 2) specifies a human operator, but in analysing the relationship between the “works” and “texts” of gameplay we need to distinguish between three kinds of textual-machines: textual-machines that operate themselves (figure 3a); textual-machines with a human reader (figure 3b); and textual-machines with a human user (figure 3c).

Figure 3. Movement from the production of “work” or “text” during (i) computation, (ii) reading and (iii) interaction, derived from Aarseth's model (1997, p. 21). [Note: h/w= “hardware”; s/w=”software”]

With textual-machines that operate themselves, the “work” stands by itself, and there is no “text”. In the case of textual-machines with a human reader, any attempt to understand the internal working of the computer requires that the computational processes (“works”) are accompanied by a “text” dynamically (re-) constructed by the operator. There is no actual cybernetic relationship if the sign of the machine “work” is simply re-anchored by/in the reader's “text,” as when the player is simply looking at hardware or output on the monitor. Of course, inasmuch as cybertexts are anamorphic, and usually “intransient” according to Aarseth's criteria, the production of the “text” requires some interaction, at the minimum, swapping hardware around to (say) check if one has a faulty graphics card, or the clicking of cursor or mouse keys to move through a hypertext. In the case of textual-machines with a human user, the “work” is accompanied by a “text” produced by a reader, and this “text” will (partly) determine how the user alters the “work.”

Given that computation is a logical and hierarchical system of linear processes, with nested modules, any “text” of the “work” will necessarily reflect this structure. We might, then, emphasise how computational semiotic offers “semiotic analogs to programming and interface design” (Anderson, Holmqvist and Jenson, 1993, p. 3). A useful reference point for a preliminary level of analysis is Fiske and Hartley's (1978) reworking of Barthes' model of signification:

Figure 4. Barthes' (1982, p. 100) model of orders of signification, adapted by reference to Fiske and Hartley's (1978) reference to a third order of signification.

This model, while familiar to any semiotician, is best exemplified to avoid confusion. To take an example, a verbal signifier (1) /man/ will refer to signified (2) the idea of a «man», and the sign (3) is constituted by the agreement between these two terms. 1 This first order of signification is denotative, in that it is seen as a neutral and value-free process of pointing to a referent, a particular //man//. However, the signified of one plane of expression is usually the signifier of another plane of expression. We speak of a second order of signification, of connotation , when the same signified may have different signifiers which affect how it is valued: for example, a //man// may be represented by a red icon or digitally realistic image. While connotation will draw from cultural codes, such that changing the signifier allows one to control the connotative meanings and so may be seen as a stylistic choice, connotations are often personal, referring to “associative, expressive, attitudinal or evaluative shades of meaning” (1994, p. 286). We speak of another second order of signification, of myth, when a signifier refers to a chain of signifieds widely used throughout a culture. So, the sign (3) may be taken as a signifier (I) /man/ that refers to signified (II) “hero,” or, more properly, the “myth” of the hero as, among other things, male, independent, active, brave, good. Fiske and Hartley (1978) also argue that a culture's connotations and myths

may fit together to form a coherent pattern or sense of wholeness, that is, the way they ‘make sense', is evidence of an underlying invisible, organizing principle – ideology. Barthes' identifies a similar relationship when he calls connotators (the signifiers of connotation) ‘the rhetoric of ideology' (O'Sullivan et als, 1994, p. 287)

Consequently, they suggest that it may be useful to think of a third order of signification: ideology .

This model indicates how the signified of one plane of expression is usually the signifier of another plane of expression, such that any sign depends upon a primary code and/or makes reference to a tertiary code. However, we may here merely take it as providing a terminology for talking about how signification is framed by multiple codes of varying complexity and generality. This is evident in figure 5, in which we see how a sign may be identified with reference to a register in memory, a sub-function, or module, and a main program loop.

Figure 5. Fiske and Hartley's (1978) model of first, second and third orders of signification applied to possible ways of representing frames of sign-functions inside a computer system.

At a broad level, the “computational text” as defined here is concerned with mapping the formal logic of the “computational work”. The effects of polysemy are minimised the more the “computational text” conforms to mathematical principles or modal logic, in that one moves towards abstraction that is independent of the medium. Of course in practice, there may be more than three “frames” in a computer program (modules calling modules calling modules, ad nauseum ) and there may be more than three codes invoked in a semiotic sequence (see below). Consequently, mapping orders of signification onto frames of computation may break up the continuity of computation into spatial and temporal units which the computer would not recognise, and/or may break up the “text” into more computational units than the reader/player would care to know. However, players are rarely aware of, or interested in, the formal relationship between the minimal units of information and the frames in which they are embedded, and in Eco's formulation sign-functions are ascribed to output. When players do make sense of a game like FFX , then, it will generally be in terms of structures better apprehended in terms of orders of signification that govern the complexity and generality of semiosis in general. That is, computation may be seen in terms of contextual frames associated less with programming than orders of signification, and acknowledging these helps to suggest the units, and scope, of code that are identified by players.

At the most particular level, we can hypothesise a variable called hero_health in FFX that has been declared as an integer with the parameter range 0 to 10000, and from programmer's point of view represents Tidus' health points (say, 1520). The execution of the code will mean that, at the level of computation, any reference to hero_health points to the state of a register in memory: the integer value currently linked to hero_health (1520). Inasmuch as the former sign simply stands for, or points to, another, we may see this relationship in semiotic terms as denotative (hero_health à 1520).

However, during the execution of a program, variables are often accessed by, or passed to, various functions, sometimes under different names. In games like EverQuest that code for physics, such that falling from a height causes damage, a variable like hero_health might be passed as a global variable to a function called object_collision as an integer called temporary_health. At another time, hero_health may be passed as a local variable to a function called calculate_hero_speed as an integer called health_lag. Since the same memory register or variable is called by multiple functions, the same signified has different signifiers and produces different effects, so we might speak of a second order of signification: connotation . In function object_collusion, temporary_health is reduced if the character has fallen and hit the ground, and its value is then passed back to hero_health; in function calculate_hero_speed, health_lag merely determines how a character's injuries slow down the character, so its value is not changed nor passed back to hero_health. The two “connotations” both ultimately refer to the same register, the register to which hero_health points, but provide it with a distinctive “meaning.”

In FFX , a players' attentiveness to this kind of process can be seen during character customisation. Unlike most RPGs, CRPGs, MUDs and MMORPGs, in which there is the possibility for initial character creation, the statistics in FFX are initially the same for each character, except for the name. In the Status menu, characters stats are listed in a sequence:


HP 1520/ 1520
MP 92/ 92

S.Level 0
Next 1059 AP

Overdrive Mode Warrior

Strength 23
Agility 20
Defence 19
Luck 18
Magic 9
Evasion 18
Magic Defence 9
Accuracy 18

Figure 2.3. From Tidus' Status Screen in FFX.

However, as in other RPG-genres, character statistics (stats) in FFX may be modified by expending Spheres accrued through adventuring by entering the Grid Sphere Menu. Spheres are of various types— Power, Mana, Speed, Fortune, Ability, and Key—and may be expended to improve certain types of statistics. For example, Power Spheres may be expended to improve Strength, Speed may be expended to improve Agility. When the character enters the Grid Sphere, they may look at each character using L1 or L2, selecting one with X. They may then select Move on the menu, to traverse the accessible nodes. If the cursor is located on or next to an unactivated node, the player can select Use on the menu, whereupon the available Spheres of each type will be displayed. The character may select a Sphere type, pressing X to confirm or 0 to cancel. This action, once made, is irreversible.

When the player makes such a choice, the associated statistic will increase accordingly. /Strength/, for example, refers to a current value: «23», and any player with a rudimentary notion about programming may readily identify /Strength/ as a variable (let us call it STR) whose current value is 23. A player attending to this relationship between a variable and its current value during character creation may be seen, then, as conceptualising a sign at the interface and/or computation at a level of specificity that is denotative. When a player presses /X/ to select /Use/, «23» changes to «24», and the number of relevant Spheres is reduced, from say «5» to «4». Inasmuch as logical operators such as – and + cannot exist without terms to act upon, what “Use” may be seen as signifying is the operations it sets in motion: STR = STR + 1 (23 + 1 = 24) and: Spheres = Spheres – 1 ( 5 – 1 = 4). We may see this in terms of a second mode of signification: connotation in that the significance of the operation may depend upon which component of the operation is being emphasised (the + or -). In other contexts, in which equipping an item decreases STR, we might see connotation in the sense that two different operations refer to the same signified (STR) but have different “meanings”.

Unlike the “connotation” involved when a variable is accessed by different functions, or when the player considers operations which act upon a variable, we might refer to the second order of signification: myth when referring to sets of algorithms that represent some relatively autonomous operations that appear through the code. For example, the program as a whole, or several functions, might code for a simplified simulation of physics. The “conventions” of the related algorithms may not be confined to a function but cross functions, or may be routinely called and affect a whole range of computational processes. A character in EQ may pick up an item, be attacked by a monster, and try to run but be unable to. The reason for this immobility relates to the way the game codes for gravity: each item in the game world is assigned an encumbrance value, and if a character possesses too many items their speed is reduced to the point that they cannot move. Ultimately, the value of hero_health may be seen as determined partly by encumbrance in that a heavily weighted character cannot run from opponents and so is more likely to be injured.

It may be useful to identify the work of a second order of signification: myth to refer to groups of context-independent operations that seem to cross a variety of other operations. For example, we may identify two groupings of “myths”: inventory items and quests. Items in inventories may be either “generic” or “special.” The former are merely tokens of a type and will be associated with a series of operations that are identified when one examines it, relating to the relevant inventory slot, the damage it inflicts, who can use it, and its effect. Some items may be used by any character, for example, Healing Potions, Gems, Tablets and Distillers, but other character equipment may only be used by the appropriate character. For example, an Al Bhed potion cues Poison, Silence, Petrification and gives 1000 HP to all characters, but the item Golem Shield can only be worn by Tidus, in his armour slot, and provides the abilities: Auto-Shell, Auto-Protect and Auto-Regen, Auto-Reflect. Warblade can only be used by Auron, in his weapon slot, and adds 5% to his Strength. Inasmuch as items can only be equipped by the right character, in the right slot, the operations associated with an inventory type will be seen as a series of IF . . . THEN statements. IF the character has picked up the Golem Shield, THEN they try to equip it. IF the character matches the item, THEN the character is allowed to equip it. IF the character's secondary are occupied, the items are swapped. IF the item is equipped, AND used in battle, AND a strike is made, THEN their effect will be applied. Inasmuch as there may be multiple items of the same type-, the same sequence of operations will apply to any object of the same type that the player comes across in the game. (Although there is a random variation in some item's effects, and players may customise items by adding new effects).

Quest items, unlike other inventory items, are often tokens of their own type, and some have no function other than as conditions that must be met for another operation, such as a quest, to be compled. Unlike many adventure and role-playing games, FFX does not have many redundant items or quest items—for example, items which must be delivered to certain NPCs—except those related to character abilities. There is the Withered Bouquet and Flint required to make a fire in the early sequences, Jecht's spheres, which may be collected to find out what happened on Jecht's prior attempt to defeat Sin. There is the Summoner's Soul, necessary to teach Aeon's abilities; the Aeon's Soul, necessary to increase Aeon's attributes; and the Blossom Crown and Flower Scepter, necessary to find hidden Aeons. There is also the Cloudy Mirror, whose function is to allow the player to obtain the Celestial Mirror, which is necessary to enhance Celestial Weapons. Finding the Celestial Weapon of each character is a mini-quest in itself which requires finding the relevant Sigal and Crest, and, in the case of Auron, the Rusty Sword. In some of these cases, the only use of the item is to allow the player to progress. In this sense, we can identify an operation not so much with the item as much as with the quest itself. The quest “myth” follows a generic computational sequence, which can be verbalised as: the character talks to an NPC; the NPC asks if the character is interested in the quest; the character states their interest, is told what must be done or acquired; the character goes looking for the quest object, or the NPC carrying it; the character finds the quest object, or takes it from the NPC who has it; the character gives the quest object to relevant NPC; the character is given a reward of gold, experience points, or another quest. This may, again, be seen as a sequence of IF . . . THEN operations in that characters can only get the reward if they hand over the object, and the character can only hand it over if they have found and taken the item, and they may only be able to take the item if they kill a certain NPC, and so on. This group of operations is, with some variation, common to CRPGs, and during gameplay it will reflect and/or draw attention to the formal logic of the code which is its precondition.

We might also consider how the entire program—or, perhaps, the entire system—may be seen as analogous to a third order of signification: ideology , in that it is only at this level of generalisation that certain sign-functions are seen as part of a coherent whole. For example, the fluctuating value of hero_health will become especially significant if it is recognised that function_main, the primary function which calls the other functions, keeps iterating while hero_health > 0. This may be taken as a third order of signification in that a wide variety of operations will be seen as organised in terms of this principal operation. In this context, a variety of variables and functions may be seen as ultimately referring back to hero_health and a whole range of tendencies related to this over-all “function” may become evident. From the perspective of the player, the operation WHILE HEALTH > 0 will become increasingly important when Tidus is about to die, taken as the organising operation. Of course, in most arcade games, players have multiple lives, so the basic operation may be: WHILE LIVES > 0, and players will increasingly attend to this operation when they are on their last few lives. In FFX , this might be extended to the equations WHILE PHEONIX DOWN > 0, since these can be used to revive characters, and/or while MP > 0, since some characters may have the ability to use their magic to heal characters. We might also include WHILE ETHER > 0, since these can be used to increase MP, and any other items that can prolong life, such that it is only when all of these approaches 0 that a certain operation becomes increasingly significant to the player.

That players consciousness of the scope of the cybernetic processes is here obvious. Players may experience their cybernetic relationship in terms of fairly minor operations, such as modifying states, or in terms of larger data structures, such as a character, including all his/her statistics, Abilities and equipment. Alternately, the player may attend more upon the operations which constitute a battle, or a quest, or the entire game, which may be rendered in formal terms as: WHILE SIN = ALIVE. In certain situations, this may extend to some an awareness of and/or sense of connection to: the software in toto (the game code, or the operating system); an aspect of the hardware (say, a faulty graphics card); the hardware in toto (the entire console or PC); or it may extend to a dial-in modem hook-up, a local area network (LAN), or an entire computer Network over the Net. Of course, this scope of awareness of the information flow, or cybernetic system, is cannot be confused with the complexity of the player's model of this system: a player who knows s/he is “connected to the ‘Net” while downloading FAQs or cheats but there may be a very large black box in their model of what goes on between clicking on the mouse button and seeing the effects onscreen.

Such analysis is, then, far removed from the actual computation which gives rise to the activity of the interface. However, it does suggest the kind of “text” that constitutes the players model of computation during gameplay. The issue here is that there may be an enormous mismatch between computation and the player's model of it, derived from feedback at the interface, and that this disparity between the “work” and “text” is central to gameplay.

Hypothesising and Testing the “Computational Text”

The players attentiveness to relationships between computational signs or processes is, it should be obvious, not usually simply a consequence of reading text from the machine. It is through each choice and its consequence that players acquire, test and refine their “computational text,” or model, of computation, and this developing model defines the rules which give a particular sign or semiotic sequence its meaning. Most technicians and programmers, of course, do not simply test a “computational text” against the “computation work,” they try to modify the “computational work” so that it conforms to the “computational text.” This modification may be achieved either directly, by modifying the computer's hardware, or indirectly, by using the peripherals that constitute the interface, as when a programmer writes or debugs code using a keyboard and mouse. In any case, the gap between “computational work” and “computational text” is made visible to the user whenever the game does not perform in the manner expected (or desired) and a reader/user considers alternate possibilities for the “computational work.” Furthermore, the reader/user may hypothesise multiple possibilities, so it might be argued that players synthesise a variety of fragmented “computational texts” into an ideal single “computational Text” which is only complete when the player has finished with a game. However, the model of the “computational work” is unlikely to be systematic, more a disconnected series of models; and, as noted, much of the “computational work” is likely to be irrelevant to a player.

The issue is that ergodic literature requires the user to engage in speculation if his/her choices for interaction are to be meaningful. Cybertexts with few ergodic characteristics but nonetheless encourage speculation about alternative and possible events, such as print or film detective stories (Murray, 1997, p. 40), may here be seen not as prefiguring or as related to complex cybertexts, but as reminding us that the relationship between the “work” and the “text” is part of the pleasure and/or frustration of interacting with a cybertext. For example, the pleasure of reading a detective novel is dependant upon a disparity between the reader's theory (“text”) about what happened and the actual explanation that is finally offered (the completed “work”). This disparity is continually exploited and redressed through the dissemination of clues, red herrings, dead ends, revelations, new events, and so on, which deliberately mislead the reader about “what happened” until the entirety of the “work” is complete. It is the dynamic relationship between guided speculations (the multiple “texts” cued by the “work”) and the actual outcome (the subsequent “text”) that is central to the dramatic experience of the game.

Part of the problem here is that Aarseth's concern with compensating for poststructuralist emphasis on “text” leads him under-emphasise the disparity between the computational “work”, the player's “text” of this “work”, and the “text” which is the concern of most analysts, which is an interpretation of the scriptons. He does, certainly, make several arguments about this in his analysis of the rhetorics of hypertext, adventure games, computer-generated literature and MUDs. He makes a useful distinction between “aporia” and “epiphany”:

In contrast to the aporias experienced in codex literature, where we are not able to make sense of a particular part even though we have access to the whole text, the hypertext aporia prevents us from making sense of the whole because we may not have access to a particular part. (p. 91)

Consequently, in “hyperliterature” we have a sense that we cannot make sense of the whole “text” because of absent parts of the “work.” With epiphany, by contrast, “sudden revelation . . . replaces the aporia, a seeming detail with an unexpected, salvaging effect: the link out” (p. 91). For Aarseth,

this pair of master tropes constitues the dynamic of hypertext discourse: the dialectic between searching and finding typical of games in general. The aporia-epiphany pair is thus not a narrative structure but constitutes a fundamental layer of human experience, from which narratives are spun. (p. 91-2)

In short, the “text” constructed from the “work” may be experienced as incomplete because of absent aspects of the “work”, or we may experience closure through an “epiphany” which seems to adequately unite the “text”, despite absent aspects of the work.

Players' attention to hardware and software may here be seen in terms of a qualified and transient “hacker” aesthetic (Turkle, 1995): players attend to the hardware and software when there is a problem, and utilise “fixes” and “patches” to make a “buggy” system work, and/or experiment with options through trial and error until the game works. Of course, seeing the disparity between “work” and “text” as a problem that needs to be “fixed” or “patched” is problematic, as gameplay is frequently characterised or motivated by discrepancies between the “works” and “texts” of gameplay. Turkle's discussion of the “hobbiest” aesthetic, or Friedman's discussion of “deconstruction,” for example, describe how players attempt to master the inner workings of the machine, and/or reiterate their mastery. In this sense, each “problem” is a means of testing and refining one's “text,” a means of providing challenge. As is evident from the Tech sections in magazines like PC Powerplay , PC games are valued precisely because they provide more ready access to the system than console games. PC players routinely buy and install modular components, download material from the Net, and customise their “rig.” The ability of PC programmers and gamers to modify hardware and access low-level computational processes gives them more freedom in maximising game performance and experimenting with the potential of the machine. The attention to computation, or, more broadly, the medium, often overlaps with attention to the rules of the game, that is, a form of meta-gaming.

So while computer interfaces deliberately facilitate the operator's actions (as evident in rhetoric about “intuitive” interfaces), a discrepancy between expectation of consequences and the actual consequences often gives rise to the sense of challenge which underlies the desire to play and master computer games. If a character's movement through the gameworld was as predictable as moving a cursor across the screen in a word processor, there would be no challenge or motivation. However, it does not do to simply speak, as does Friedman, as if a player is only satisfied when the player demystifies (and thereby masters) the game code, and perhaps the hardware, en masse . In practice, the “computational works” which concern the player are merely those aspects of the code that represent the rules of the game, glitches in the code that can be exploited, and game data, which determines where monsters or objects appear or cane be found. To take a few examples, some maps are designed to confuse the player, monsters are often placed in difficult to see places, and quest objects may require the solving of complex puzzles. A player may feel they have mastered a game when they familiarise themselves with tricks the programmer/writer has played on the player, but such a player has in no way demystified the entire “computational work,” only some of its data.

The use of the term “cheating” in gaming discourse is here useful in problematising generalisations about the relationship between the “computational work” and “computational text.” Cheating is generally conceived of as a conscious and illegitimate manipulation of the rules of a game to one's own advantage, and there is much evidence of this in computer games. Computer magazines, websites and industry game guides include sections for cheat codes and “oracles” who answer players' questions. A player may miss a secret passageway they are looking for, wander for hours trying to find a location they have a general idea is in the vicinity, or get lost in a maze. The player may download “cheats” to increase his character's speed and/or walkthroughs and maps of the area. Yet a computer game is not simply a set of rules that may be broken: different players will play the same game according to different aesthetics, and players will bracket out or circumvent those rules which intrude upon this aesthetic. Cheating in computer games is not simply a way of dealing with frustration, it is a way of defining one's aesthetic relationship to the game. The player may be seen as a kind of roving “fix,” moving through the gameworld, resolving discrepancies between expectation and consequence through persistent experimentation.

The player, then, is not necessarily concerned with understanding the entire “computational work,” merely those elements that are functionally relevant to play, and different players utilise different aesthetics to determine what is functionally relevant. The frustration consequent upon disparities between “computational work” and “computational text” is only a problem when it exceeds the frustration deemed acceptable according to the aesthetic by which gameplay is evaluated. In short, the player strives for a “computational text” adequate to the task of playing the game according to their chosen aesthetic. If the game does not provide an adequate “text,” the use of “cheats” is not, strictly speaking, cheating: within the context of the rules which the player has defined gameplay as an aesthetic practice, the cheat merely makes visible the rules of this practice, and therefore becomes a part of the medium. However, while gameplay will require continual refining of the “computational text,” its evaluation during gameplay will draw from codes and conventions beyond the formal logic of computation, and, therefore, involve shifts to the following categories.

“Texts” of the Game: Intertextuality in the Ergodic Process

While we can find a semiotic analogue between the hierarchical organization of computational processes and orders of signification, this leaves open the question of how is contextualised by broader cultural (con)texts. This may even be evident in the way a programmer evaluates code. For example, there are several different ways of programming a search algorithm, all of which have their respective advantages and disadvantages, but programmers may emphasise economy, speed, or clarity when they choose to implement one or the other type, depending upon their programming principles. In this respect the “computational work” is usually evaluated not simply in terms of formal logic, but its application. Similarly, a player is not likely to identify a series of operations with reference to the conventions understood by a programmer: they will define them with reference to myths and ideologies in the culture. For example, players will likely identify a character class, or weapon, or quest, not because they reflect actual data structures or computational processes but because they are identified with the fundamental units of the game genre and broader myths and/or ideologies, such as heroic, individual agency and capitalist accumulation (see 2b).

It is here useful to identify a parallel between frames of computation and frames of cultural reference, through another application of Fiske and Hartley's (1978) model:

Figure 5. Fiske and Hartley's (1978) model of first, second and third orders of signification applied to signification from the perspective of a user.

As is evident in this model, (con)texts range from quite specific codes to broad discourses in the culture. It is here necessary to acknowledge Kristeva's notion of “intertextuality”, derived from Bakhtin's (1981) theories of “dialogism”. Intertextuality, in its general sense, refers to how any utterance is made sense of against a background of “texts”; in specific uses of the term it refers to a particular reference to another text. We should note here that while we presume that an intertextual reference is already familiar to the reader, computer gameplay is delimited by a circumscribed body of “computational works” and other “works” which the player will sometimes actively seek out, such that new intertextual relationships are formed through the practice of gaming. We might refer to these by way of Genette's (1982) category of “paratexts”: “all the accessory messages and commentaries which come to surround the text and which at times become virtually indistinguishable from it” (Stam, Burgoyne and Flitterman-Lewis, 1992, p. 207). In the case of works of literature, paratexts include “titles, prefaces, postfaces, epigraphs, dedications, illustrations, and even books jackets and signed autographs” (p. 207). In the case of computer games, paratexts include information on the packaging, the hardware itself, manuals, and hardware detection software; they also extent to FAQs, cheats, walkthroughs, spoilers, history, and drawings by players. Many of these re-present the software or game data in a more familiar verbal or pictorial form, for example, tables of game specifications and lists of required or recommended hardware; sometimes it produces or refers to additional textual material which allows the player to make sense of the game. Nonetheless, since the act of picking up these ancillary “works” requires physical interaction (what we might refer to as “interwork” processes), we can confine the category of intertextuality to readers' prior knowledge, to the work readers perform, as Aarseth notes, “in their heads” (1997, p. 21).

The construction of a model of computation may be seen as reliant principally upon binary, analogue and logical codes. However, inasmuch as a player's model of computation is limited to surface rules at the interface, other types of codes are likely to gain pre-eminence in defining the player's relationship to the game. A single example suffices here to suggest some relationships between the visible aspects of the “computational work” and intertextuality in gameplay. Let us assume that a player of FFX is attending to Auron's health bar and has some idea that it represents a variable in the computer with a minimum and maximum value. The health bar will also denote the character onscreen, an independent and potent representation of a well-accoutred, superhuman, male individual who takes as his own every item he comes across and can carry, and (in time, sometimes with help) overcomes overwhelming odds. This representation refers to a broad myth of heroism in the culture, the idea of a single human with (nearly) supernatural attributes. That the hero is a man may be naturalised by reference to the ideology of patriarchy, and his economic ethos (of gaining wealth through personal appropriation by any means) may be naturalised by reference to a laissez faire ideology of liberal capitalism. Consequently, the health bar may become a transitory index and apex of social order. This cultural coding may mean that if a character's health is low, relative to its maximum, the player may read: “I am hopeless, no more heroic than in everyday life”; if, however, a character's health is high, relative to its maximum, the player may read it as saying: “Your character is a hero,” or “You are a hero” (or the player may appropriate the voice of the game to assert: “I am a hero”).

We must emphasise here that the relationship between the “computational text” and the “interpretative text”, which we would normally identify as our interpretation of the game—that is, what a player thinks FFX “means” or is “about”—is far more complex. At times the computational and interpretative texts seem to be relatively autonomous, in the sense that once we have an adequate “computational text”—once we know how to use the interface—we can focus on the “content” of the game, although our ongoing exploration of the virtual world extends our “map” of the game's data structures. At other times, the subtraction or addition of ergodic capabilities may be a deliberate consequence of developments in the narrative/game. For example, when fighting Seymour Tidus and Wakka have a Talk option; Tidus also has this option when fighting Jecht's Final Aeon. When fighting Evrae on the deck of the Airship, Tidus and Rikku both have the option of telling Cid to move the Airship closer to or away from Evrae. Sometimes the meaning of ergodic capabilities may be arbitrary. For example, when a player finds himself unable to move into an apparently unblocked area s/he might make an association to some magical barrier or force field.


A theoretical framework for analysing gameplay, then, might combine Aarseth's emphasis on the “work” with poststructuralist notions of the “text.” We can distinguish between the computational “work” from several computational “texts”, which may be relatively distinct from the “text” which interprets the significance of the game; or, rather, the total “meaning” of play may be determined by tensions between different textual experiences. The key issue emerging from such a framework is that gameplay is not necessarily characterised by progressive mastery of the total “computational work,” as is suggested by Friedman's (1995) aesthetic of “demystification.” The relationships between the “computational work,” the “computational text,” and other “texts” constructed about or around the game are continually renegotiated.


1. The exception would be a computer which has a model of its own internal processes. While this can be presumed to be merely an analogical “work” nested within another “work”, the emergence of AI may produce instances in which the distinction between “work” and “text” becomes increasingly complex.

2. Following Eco (1976), single slashes (/)refer to a signifier , guillemots ( « )refer to a signified , and double slashes (//)refer to a referent . Therefore, a signifier /xxx/ refers to a signified « xxx » whose object is //xxx//.

3. The commitment in RPGs to simulating believable human characters means that characters usually fall unconscious before dying. This is often achieved by allowing for negative health point values, with death occurring when health falls below zero minus the player's constitution statistic. Consequently, the actual operation might be WHILE HEALTH > (0 – CON). For example, if the player's constitution was 15, death would occur when the player's health reached -15.

4. Aarseth considers how “description,” “narration” and “ergodics” are incorporated in hyperliterature (pp. 93-95), and argues that in adventure games the player may be seen as an “intriguee” who co-operates in the production of a real-time drama. He also distinguishes between pre-, co- and post-processing in cyber-authored literature, and signals levels of discursive conflict in MUDs.

5. In considering how the computational “work” manipulates player expectations in this way, we might see the distinction between “work” and “text” as paralleling the narratological distinction between “story”— what is told, the basic structural transformations of plot—and “discourse”— how the story is told, including the rhetorical strategies of the narrative and the characteristics of the medium (Chatman, 1978; Culler, 1975). In Bordwell's model, we should note, the relationship is inverted: the sjuzhet is the incomplete artefact, and the fabula is a “text” the reader constructs from it.

6. However, while a computer program's “works” may have a physical structure inside the computer, there is no objective position from which a story can be told, and so the “story” is only a subjective description of the logical sequence of events as they can be assumed to have happened within the diegesis. Chatman's account, using Hjelmslev's model of substance/form and expression/content, brackets out the process of semiosis. If in Barthes' model of levels of signification, one signified may be taken up as a signifier within another sign system, we might say that the form-content of narrative in general may be taken as the substance-expression (“work”) within a new structure, with the form-content taken as the final, negotiated “text”. Obviously, different readers will infer a different “story,” a different structure of events (Bruner, 1986; Chambers, 1984), as they anticipate or imagine actual or alternative plots (Eco, 1979; Holland, 1968, 1975; Iser, 1978; Todorov, 1977).