Richard Hall

Richard Hall is currently a senior software engineer at Melbourne University Virtual Environments for Simulation http://www.muves.
. His research interests are computational story modelling, visualisation and haptics.

Kirsty Baird

Kirsty Baird has been working on community based film and video projects for the last five years. She has been involved with numerous communities including indigenous women, homeless young people, prisoners and children. Kirsty has recently completed two short films funded by the Federal Attorney-General's Department about young people, alcohol and safety which will screen on 35mm in cinemas over summer.

Improving Computer Game Narrative Using Polti Ratios

by Richard Hall, Kirsty Baird


Computer games are criticised sometimes as lacking in narrative. In this paper we introduce a model of stories (Polti ratios) into which computer games (meeting particular criteria) can be abstracted. Using this model, we designed a process to analyse and explore computer game narrative. We give a highly detailed example of applying this process to a computer game treatment currently under production called Street Survivor.

1. Introduction

The question of whether computer games should be analysed using stories has been intensely debated (Frasca, 2003). This discussion aside, we believe that stories will be used for such analysis due partly to their universality. Stories have acted as the primary structure of communication for millennia in many cultures (Chambers, 1984), and they easily transcend cultural boundaries (Colby, Kennedy and Milanesi, 1991). Explanations come from several academic fields. Historians claim that narrative fashions human experience into structures of meaning that are generally human rather than culture-specific (White, 1991). Philosophers have argued that stories have the highest degree of structure in the world (Bakhtin, 1981 Holquist, 1990).

In addition, psychologists have demonstrated that stories play an important role in cognition and in modelling the world (Graesser, 1995 Zwaan and Magliano, 1995). Studies have shown the superiority of stories over other information forms in various ways. Narrative texts took half the reading time of expository texts (Graesser, 1995). The recall accuracy of words organised in a story is twice as good as randomly organised words (Bower and Clark, 1969) and the same with expository text (Haberlandt and Graesser, 1985). Also, jurors structure volumes of unordered and conflicting evidence as stories (Pennington and Hastie, 1991).

However, narrative is not the only tool for computer game analysis in town there are other crucial aspects to successful computer games such as game playability (Fabricatore et. al., 2002). Nonetheless, within the scope of this paper we are going to naively assume that these aspects are independent for example, good narrative does not logically imply bad game playability. Integrated models must necessarily explore and define the conditions for which such relations hold in order to properly subsume these aspects.

So we focus on stories, but there are hundreds of ways in which stories can be modelled (see the Dictionary of Narratology Prince, 1987). We wanted the model to be described by quantitative measures so that any story, however complex, could be boiled down into single numbers. Such reduction can be justifiably accused of a corresponding loss of nutrition, for instance, how can good narrative equal a high number? Accordingly, these numbers alone will not necessarily show that one story is “betterthan another, because there are so many dimensions to stories and our measures will ignore many of these. But whatever their drawbacks, and however they are constructed, scores are undeniably easy to grasp, discuss and compare.

Since scores are easy to compare, we could possibly exhaustively analyse and compare current popular games, such as The Sims or Grand Theft Auto, or go beyond the computer game context into other media such as film or literature. But the purpose of this paper is to go beyond evaluation our project is to invent a simple and practical way to improve the narrative in a computer game design treatment. There are certainly aspects of evaluation, but the target audience of this paper is practicians rather than theoreticians.

The organisation of this paper is as follows. In Section 2 we discuss the classes of story model that have been used in computational story modelling, under the assumption that such models might be amenable to the creation of quantitative measures. The story model we chose to describe in terms of these measures has the following three properties: it can differentiate between computer games that contain stories and those that do not (Section 3) it can be used to specify the characteristics of artificial characters in computer games (Section 4), and it can be used to evaluate and reconceive stories quantitatively (Section 5). The first property is important because a proper story model should be able to distinguish between a story and a non-story. The second property is important because our distinguishing feature is precisely that stories contain characters and we should be able to say what a character is. We then move to application in Section 6 we consider a real computer game design treatment (for a game called Street Survivor) as an extended case study. We apply our story model to this game and reconceive the story within the model (which was largely adopted).

2. The Appeal of Polti’s Dramatic Situations

Computational story models appear to have two fundamental characteristics: a certain level of abstraction and a top-down or bottom-up construction (from a systems theory point of view). By abstraction we mean that components can be defined vaguely (e.g. “the impact was positive” à la plot units, Lehnert, 1982) or precisely (e.g. “Joe ordered a hamburger” à la scripts, Schank and Abelson, 1977). By top-down we mean that the model defines a sequence of components (like a template) that must be filled in (e.g. “starting equilibrium, falling event, rising event, turning point, final equilibrium” à la the Aristotelian model, Cassady, 1991). By bottom-up we mean the model defines a number of components that can be combined and used in any desired manner (e.g. “subsumption state, cause of termination event, problem-state description” à la points, Wilensky, 1982).

Within this framework, Polti’s situations (Polti, 1895 Lucille, 1977, Hall et al, 2000, see Appendix 1) are highly abstract but have varying levels of detail. Some situations are explicit regarding what happens to whom in what relationships, while in others these details are omitted. For an example of an explicit situation, consider rivalry of superior and inferior. Clearly, relations exist between characters and these relations entail norms that should not (but are) compromised. For an example of a less explicit situation consider error of judgment. In this situation character(s) A makes a mistake of some kind, but no description of the mistake is given. It requires the audience to recognise that A has made the mistake.

In terms of systems theory, Polti’s situations can be characterised largely as bottom-up. Polti does not specify how to combine or synthesise his units his project is identifying the minimum number of units (reviewing over a thousand stories). Many units do describe short causal sequences (e.g. crime pursued by vengeance) that are internally top-down but with respect to the other units are bottom-up. In terms of events occurring in a story, it appears that multiple units can be concatenated or nested as long as (most of the time) the causal sequences are completely represented. For example, pursuit can halt temporarily for various reasons.

The main advantage of these bottom-up abstract units is that they are descriptive, as opposed to prescriptive. They essentially provide a pattern language that can be either matched to existing stories or used in story authoring. And they help us substantially in our quest to create quantitative measures for stories in that these units and their components are countable, by and large. But firstly we need to show that this model can distinguish between a story and a non-story in particular, between a story and a game.

3. Is It a Game or a Story?

Some authors distinguish between the two by the role of the people engaged (Newman, 2002) and the linearity of the events (Lindley, 2005). We’re going to take a slightly different tack, and initially propose the theory that if people can explain what is going on in terms of at least one of Polti’s units then the object that they are engaged with can be labelled a story. Since these dramatic situations are extracted from stories, this means that if there is a match, then people should be able to point to other stories (or genres) which share similar events, and perhaps argue that these objects are stories by analogy.

We now relate our theory to three levels of game abstraction: definition, game genre and in terms of an instance Tetris, following previous academic work (Ryan, 2001). At the level of definition, suppose we use a 10-point definition of game components (Avedon and Sutton-Smith, 1971): 1) purpose of the game, 2) procedure for action, 3) rules governing action, 4) number of required participants, 5) roles of participants, 6) results or pay-off, 7) abilities and skills required for action, 8) interaction patterns, 9) physical setting and environmental requirements, 10) required equipment. Clearly, game component 6 (results or pay-off), can be matched to Polti’s unit obtaining, thus these categories are blurred.

At the level of game genre, suppose we consider Crawford’s taxonomy of computer games (Crawford, 1984). His two main categories, skill/action games and strategy games, are equally blurred. A number of Polti’s units are easily matched: the enigma (searcher), pursuit, obtaining, falling prey to misfortune, ambition, disaster, and daring enterprise. Thus games in these genres blur with stories.

At the level of instance, consider Tetris with respect to Polti’s situations. It’s straightforward to find at least two matches: obtaining the player is continuously losing more and more ground and the enigma the player is continuously searching for a way to get rid of all the blocks. Murray’s analogy for this “a perfect enactment of the over tasked lives of Americans in the 1990s-of the constant bombardment of tasks that demand our attention and that we must somehow fit into our overcrowded schedules and clear off our desks in order to make room for the next onslaught” (Murray, 1997), has been derided as a projection (Eskelinen, 2001). Nonetheless, this analogy matches Polti units, thus both are stories under the current theory.

Clearly our initial theory requires some work as we currently have interpenetrating categories at all levels of game abstraction. To attempt clarification we propose two simple sub-categories of story: embodied and unembodied. The distinction is that an embodied story generates characters in the mind of the audience of the story. Such a definition might be contestable and incomplete, but it does pave the way for dialogue. Embodied stories, which have characters, are trivially easy to match to Polti’s units, since these units essentially describe situations of conflict between characters. On the other hand, unembodied stories require an imaginative reconceptualisation of the events in terms of characters to achieve a match. Such reconceptualisation requires more effort to match than embodied stories and must be achieved using some sort of individual (and thus contestable) justification.

How do these unembodied and embodied sub-categories impact our three levels of abstraction? At the definitional level, component 6 (where a result or pay-off is achieved) may impact characters, producing an embodied story, or vice versa. As an unembodied story, the stretching required to reconceive events and the necessary defence of such stretching might, on occasion, be too weak to warrant even considering these game in terms of Polti’s units (or perhaps any other story model). Ultimately, if we agree that no characters exist, and agree to look through Polti’s lens, component 6 cannot be matched to obtaining. Suddenly the notions of game and story can be separated into distinct categories at this abstract level using a clearer notion of embodied stories. At the less abstract level of game genre, we argued that a number of Polti’s units were matched. Similarly, analysis in terms of stories is fine when genres have characters.

Where does this leave Tetris? We proposed two matches: obtaining and the enigma. Neither of these matches generate characters in the mind of an observer, thus Tetris cannot be classified as an embodied story via our definition. Murray’s interesting angle is from the broader philosophical viewpoint, which has been considered but excluded from our category of story. Tetris can thus be classified as an unembodied story the two characterless matches we made with Polti are not, in our opinion, as interesting as the tangential philosophical view.

In this section we have separated the categories of games and stories via analysis at three levels of game abstraction. We agree with Eskelinen that the presence of characters is key to justifying the application of story models to games but disagree that philosophical viewpoints on games equate to ‘interpretative violence(Eskelinen, 2001). We believe that such viewpoints are perceptive and worthwhile articulating but have no need to be artificially positioned in the category of narrative to claim legitimacy.

4. Computer Games with “Characters” Have Stories

In the previous section we made a distinction that relied on the concept of a character without defining what characters are. Bringsjord previously proposed that computer generated characters should have a minimum of five properties of (philosophical) personhood: language, autonomy, creativity, phenomenal consciousness, and reasoning capacity, and argued that since artificial autonomy is impossible, computer generated characters are impossible (Bringsjord, 2001). Of course, it would be great if artificial intelligence advanced to the point where all would agree that a program had achieved personhood and we hope that one day it does. However, we think Bringsjord is unreasonably raising the bar for characters to necessarily require personhood. With characters, the notion of their personhood exists only in the mind of an observer of a story. The internal mental state of movie actors may not necessarily correlate to their external state.

Our specification for characters is far less ambitious than what Bringsjord has proposed. Rather than using philosophy as a basis, we developed our requirements directly from our story model. That is, we systematically identified a set of capabilities that characters should appear to have that would allow all of Polti’s thirty-six dramatic units to be realised (Hall, 2002). These capabilities are summarised in Figure 1.

Attributes: All attributes of character A which have an effect external to themselves in the situation should appear to have their normal effect. For example, if one character can run faster than another, the former would normally win a foot race.

Perception: A should appear to normally perceive their situation perfectly. For example, if A perceives their situation incorrectly then they are much more likely to make mistakes.

Mental state: The normal mental state for A should appear to be that their aim is to satisfy their goals, and every emotion which A has towards other agents reflects a perceived relationship. For example, if their actions do not reflect their internal goals then their behaviour would be described (in pop psychology) as madness.

Pursuit: The position of A (being in a specific situation) should appear to be normally independent of the position of other objects, and A can move freely around in and between situations. For example, most As are not continuously pursuing something.

Love: As should appear to be normally emotionally neutral towards other As, unless special circumstances have caused them to exhibit the important directed emotions of love or hate. For example, if A1 is in love with another, the former will not appreciate the latter being removed from those circumstances.

Family: A should appear to not belong to a family with all other characters, and special conditions must occur for a family to exist. For example, members of the one family happily coexisting in the same situation should not be enemies.

Friend: A should appear to be normally friends with characters that they interact with, although specific interactions can cause them to become enemies. For example, if A1 is friends with A2 but enemies with A3, A1 will not appreciate A2 being friends with A3.

Power: A should appear to normally not have power over other A, although A may often be involved in power structures where they or others have power. For example, if A1 has power over A2, A1 does not expect A2 to try to gain power over A1.

Exchange: A should appear to normally not exchange objects with other actors. Transactions of objects do occur between situations, but not always according to the wishes of the A involved.

Intention: A normally tries to produce the outcomes they desire. However there are four possible reasons for action from the point of view of actors: choice, accident, necessity, or irrelevant. Irrelevant means that the source of the deterministic cause is not important.

Figure 1: Character capabilities

We believe that Bringsjord’s call for game designers “to impel gamers to believe they are interacting with virtual people” is unnecessary. Dramatically compelling entertainment, by definition, must be able to be achieved via the realisation of Polti’s dramatic situations. Computer games with characters with any of the above capabilities, who represent corresponding Polti units, are thus computer games with narrative by our account.

5. Using Polti Ratios to Evaluate and Improve Stories

In this paper we evaluate stories in terms of simple measures related to Polti’s dramatic situations. We use ratios because they are supposedly the most informative scale (Reich, 1995). We’ve already broadly defined a story as events in which people can identify at least one of Polti’s dramatic situations. Intuitively, a small number of events, E, with a large number of Polti’s units, P, correspond with the notion of an intensely dramatic story but a small number of Polti’s units with a large number of events correspond to a dull story.

Level of Drama(LoD)= P/E

So how does this relate to games generally and computer games specifically? In most games, there is potential for E

©2001 - 2008 Game Studies Copyright for articles published in this journal is retained by the journal, except for the right to republish in printed paper publications, which belongs to the authors, but with first publication rights granted to the journal. By virtue of their appearance in this open access journal, articles are free to use, with proper attribution, in educational and other non-commercial settings.