CoChat
Sign Up
Cited by 0Artificial Intelligence in GamesHuman Motion and AnimationDigital Games and Media

Expressive Response Curves: Testing Expressive Game Feel with A*

Open access
Junius, Nic, Carstensdottir, Elin·2023·Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment
DOI 10.1609/aiide.v19i1.27524
Open this paper in CoChat — free accountOpen PDF

Abstract

Designing AI models for expressive character behavior is a considerable challenge. Such models represent a massive possibility space of individual behaviors and sequences of different character expressions. Iterating on designs of such models is complex because the possibility spaces they afford are challenging to understand in their entirety and map intuitively onto a meaningful experience for a user. Automated playtesting has primarily been focused on the physical spaces of game levels and the ability of AI players to enact personas and complete tasks within those levels. However, core principles of automated playtesting can be applied to expressive models to expose information about their expressive possibility space. We propose a new approach to automated playtesting for AI character behaviors: Expressive Response Curves (ERC). ERC allows us to map specific actions taken by a player to perform a particular expression to understand the affordances of an expressive possibility space. We present a case study applying ERC to Puppitor rulesets. We show that using this method we can compile paths through Puppitor rulesets to map them and further understand the nature of the expressive spaces afforded by the system. We argue that by using ERC, it is possible to give designers more nuanced information and guidance to create better and more expressive AI characters.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper introduces Expressive Response Curves (ERC), an automated playtesting method that uses A* to trace paths through an expressive possibility space and combine them into maps. A case study on Puppitor rulesets shows ERC can reveal responsiveness, gaps, and how cost-function choices shape the paths found.

Key findings

  • Applying ERC to Puppitor rulesets produced maps of paths from chosen starting affect vectors to each expressible affect; searches used six starting affect vectors with distinct prevailing affects, and at most five unique paths per start because the search repeats the same move if the goal affect is already expressed. Two additional trials used an alternative cost function.
  • The chosen ruleset often showed responsiveness: individual move choices frequently changed the prevailing affect before another move was considered; the only goal affect that required A* to repeat one action across multiple nodes was sadness.
  • Cost function changed the paths: with `|∆goal val|`, figure 3 had only two total paths and no path from a joy node to fear; with `||∆max val| - |∆goal val||`, figure 4 gave every expressible affect a unique set of moves. The first cost preferred subtle moves, while the second preferred more active moves.
  • Precalculation example: with joy at 0.35 and worry at 0.77, and a projected_energy + neutral update of 0.0005, joy needed at least 839 individual steps to equal worry; multiplying a step by 90 changed the update to 0.045 and joy to 0.395, cutting the minimum calculated steps to about 10.

Methods

  • Design: ERC translates an expressive possibility space into a graph, with nodes representing states such as affect vectors and edges representing actions or moves; A* is adapted with node information including prevailing affect and chosen move, domain-specific cost functions, a heuristic, and abstraction of repeated updates. Searches run from starting points to expressive goals, and paths are analyzed for length, complexity, and responsiveness by highlighting prevailing-affect changes and new moves.
  • Data/sample: a case study on Puppitor, an embedded domain-specific language for character expression, using rulesets extracted from *Tracks in Snow*. Six starting affect vectors were chosen, each with a unique prevailing affect, and A* searched for each affect from each start; two extra trials used an alternative cost function.
  • Analysis: paths were reconstructed by unrolling precalculated steps and recalculating prevailing affect at each step; individual ERCs were combined into maps and Sankey diagrams, and cost functions were compared to characterize the affordances of the expressive space.

Limitations

  • Stated: the Puppitor heuristic is inadmissible, so optimal paths are not guaranteed; the authors say they only care whether a path exists, and creating an admissible heuristic for this new domain is beyond the paper’s scope.
  • Stated: visualization is limited. Sankey diagrams flatten exact time progression and cannot accurately capture a single-step worry-to-joy transition via `(projected, tempo_down)`; current techniques allow only one starting point per diagram, and two or three starting points cause combinatorial explosion. Full mapping and user-friendly presentation of all paths is an open problem beyond the paper’s scope.
  • Stated: Puppitor was an ideal testbed because it has clearly defined states and built-in human-perceptible prevailing affects; other expressive domains may lack these features, making translation and adaptation harder.
  • Evident from the text: the case study examines small corners of a single Puppitor ruleset rather than the full expressive possibility space, and the figures are not comprehensive captures. The authors also state that ERC is not a recommendation tool for designers.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1609/aiide.v19i1.27524

Cites

29 of 29 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

7 of 20 cited sources checked · 4 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Automated playtesting research has mostly targeted generated levels, game balance, and imitation of human decision-making.

    “Automated playtesting for games has largely been focused on evaluating generated levels (Liapis et al. 2015;Powley et al. 2016;Holmgård et al. 2018) and game balance (Zook, Harrison, and Riedl 2015;Pfau et al. 2020Pfau et al. , 2022) ) and emulating human decision making (Devlin et al. 2016;Holmgård et al. 2016).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Liapis et al. 2015 source
    • cites Zook, Harrison, and Riedl 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Devlin et al. 2016 source
  • Premisecitation

    The widely used UCT selection formula often produces inhuman game-move choices.

    “the commonly used selection formula, Upper Confidence Bound for Trees (UCT), frequently leads to inhuman choices for game moves (Devlin et al. 2016).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Devlin et al. 2016 source

      The Monte Carlo Tree Search section calls UCT one of the most commonly used tree policies. The Biasing Monte Carlo Tree Search section shows that move choices by the baseline ISMCTS agent differed from human choices and became more human-like when its UCT score was biased. It does not establish that the UCT formula itself frequently causes inhuman moves across games; the measured comparison is for Spades agents. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
  • Interpretationcitation

    A long path of perceptually similar nodes, especially with distinct edges, signals an unresponsive region of the expressive possibility space.

    “A lengthy path that contains long sequences of perceptually similar nodes, especially if their edges are distinct, is an indicator of an unresponsive area of the expressive possibility space, the expressive equivalent of the 1000 bowls of oatmeal problem (Compton 2016;Rabii and Cook 2023).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Compton 2016

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Rabii and Cook 2023 source

      Section 5.3, “The Cost of Oatmeal” (page unavailable), describes large quantities of perceptually similar content as the oatmeal problem, and §4.1.2 says a large possibility space may contain very similar outputs. The paper does not describe paths of nodes, distinct edges, or those features as an indicator of an unresponsive region of expressive possibility space. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Findingown data

    Applying ERC to Puppitor rulesets lets paths through those rulesets be compiled and mapped, revealing the character of the expressive spaces the system offers.

    “We show that using this method we can compile paths through Puppitor rulesets to map them and further understand the nature of the expressive spaces afforded by the system.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Core principles of automated playtesting apply to expressive models.
  • Findingreasoning

    Every node in the Puppitor graph has at least twelve incoming edges.

    “In turn, this means every node has at least twelve incoming edges, as nodes with minimized and maximized values in their affect vectors are reachable from huge numbers of other nodes.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. 38
  • May combine two claims. 24

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Puppitor is an embedded domain-specific language that specifies a character's expressive qualities so human and AI players can jointly perform scenes in narrative experiences.

    “Puppitor is an embedded domain specific language that defines the expressive qualities of a character to allow human and AI players to collaboratively perform scenes within narrative experiences (Junius et al. 2022).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Junius et al. 2022 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The case study's rules were taken from Tracks in Snow, a visual novel built to showcase Puppitor.

    “Our chosen rules are extracted from Tracks in Snow (TiS) (June et al. 2021), a visual novel designed to showcase Puppitor (Junius et al. 2022).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites June et al. 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Junius et al. 2022 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    In Puppitor only one affect is expressed at a time, namely the affect with the largest magnitude.

    “Only one affect can be expressed at a time, corresponding to the affect with the highest magnitude, the prevailing affect, with logic to handle the possibility of multiple affects having the same value (Junius et al. 2022).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Junius et al. 2022 source

      The Affecter section describes selecting a highest-valued affect as a single ‘prevailing affect’ and gives tie-handling rules. But it says this conversion to a discrete value is available when a single value is preferable outside Puppitor; it does not establish that Puppitor can express only one affect at a time. (LLM, conf=0.85)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    During a Puppitor update, the prevailing action's value is multiplied by the prevailing modifier's value.

    “When a Puppitor update is applied to an affect vector in a game loop, the prevailing action's value is multiplied by the prevailing modifier's value (Junius et al. 2022).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Junius et al. 2022 source

  • May combine two claims.

    Findingreasoning

    Pre-multiplying a single step to be equivalent to 90 steps changes the update value to 0.045 and one step from the initial joy value to 0.395, cutting the minimum number of steps the search must calculate to approximately 10.

    “If we instead pre-multiply a single step to instead be the equivalent of 90 steps, changing the update value to be 0.045 and one step from the initial value for joy to be 0.395, the minimum number of steps the search must calculate can be cut to approximately 10 steps.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    In figure 1, there are three unique paths to expressing fear, each showing fear at a different point in time due to different moves chosen.

    “For example in figure 1, there are three unique paths to expressing fear, each showing the expression of fear appear at a different point in time thanks to the different moves chosen.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    With the two cost functions side by side, the authors characterized their preferred performance styles: |∆goal val| prefers more subtle moves and ||∆max val| -|∆goal val|| prefers slightly more active moves.

    “With the two cost functions side by side, we could then characterize the performance styles they preferred, with the first (|∆goal val|) preferring more subtle moves and the second (||∆max val| -|∆goal val||) preferring slightly more active moves.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The authors originally used only |∆goal val| as the cost function but noticed it rarely made paths containing obvious move choices when comparing its decisions with the ruleset definition.

    “We originally only used |∆goal val| as the cost function to produce ERCs, but noticed that it rarely made paths containing obvious move choices when comparing its decisions with the ruleset definition.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Comparing figures 3 and 4 shows that with the ||∆max val| -|∆goal val|| cost every expressible affect has a unique set of moves to reach it, whereas with the |∆goal val| cost there are only two paths in total.

    “This change is most obvious in comparing figures 3, which uses the |∆goal val| cost, and 4, which uses the ||∆max val| -|∆goal val|| cost, as every affect able to be expressed has a unique set of moves to reach it in the latter, whereas there are only two paths in total in the former.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    A different cost function doesn't always lead to dramatically different paths through the possibility space, but affects that can be easily expressed are much more likely to be found early in the search.

    “A different cost function doesn't always lead to dramatically different paths through the possibility space, but affects that can be easily expressed are much more likely to be found early in the search.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Juice does not necessarily raise player performance, and too much or poorly judged juicy feedback can slightly reduce its usefulness as performance feedback.

    “Interestingly, with as much focus as juice has received as part of game feel and feedback mechanisms, it does not necessarily improve player performance in games and too much or ill considered incorporation of juicy elements can slightly detract from their use as feedback about performance (Hicks et al. 2019).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Hicks et al. 2019 source

  • Classified as probably not a substantial claim.

    Premisenone given

    A* was selected over Monte Carlo Tree Search for building ERCs because its behavior is more consistent and reproducible.

    “Finally, we chose to use A* (Hart, Nilsson, and Raphael 1968) as part of the approach to creating ERCs over other popular algorithms like Monte Carlo Tree Search (Chaslot et al. 2006) due to its more consistent and reproducible behavior.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Hart, Nilsson, and Raphael 1968 source
    • cites Chaslot et al. 2006

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Expressive range analysis characterizes a generator's possibility space from the artifacts it produces.

    “Expressive range analysis (ERA) is a method of describing the possibility space of a generator, often a level generator, based on the artifacts it produces (Smith and Whitehead 2010).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Smith and Whitehead 2010 source

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Paths found by A* are more readily reproducible than those of other game-AI algorithms such as MCTS.

    “In addition, the paths A* finds are more easily reproducible than other algorithms commonly used in game AI like MCTS.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Game feel treats interaction with a game as assembled from many components, from single button presses to the way goals reshape a player's relation to tasks.

    “Game feel describes the way interacting with a game is built out of numerous pieces coming together, from the ways an individual button press manifests on screen and through speakers to the ways giving a player goals alters their relationship to individual tasks (Swink 2008).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Swink 2008

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    ERC can supply designers with richer information and guidance for building better, more expressive AI characters.

    “We argue that by using ERC, it is possible to give designers more nuanced information and guidance to create better and more expressive AI characters.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    ERCs let us visualize the expressive-character-AI analogue of Game Feel.

    “In this sense, therefore, ERCs allow us to visualize what is the equivalent of Game Feel (Swink 2008) for expressive character AI.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Swink 2008

  • Classified as probably not a substantial claim.

    Premisereasoning

    A*'s built-in path generation matters for ERC creation, since ERCs are paths through expressive possibility space.

    “Additionally, A* has built in path generating functionality which is important for the creation of ERCs, as they are paths through expressive possibility space, and their analysis to characterize the expressive possibility space's affordances for player experience as inspired by game feel.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Juice is a part of game feel concerned with cascading audiovisual feedback responding to player actions and input.

    “Additionally, juice is a subset of game feel focused on cascading audiovisual feedback in response to player actions and input (Gabler et al. 2005) and our usage of ERCs can provide insight about where juice may ideally be applied to heighten the play experience.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Gabler et al. 2005

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Because an ERC is a sequence of possible actions and timings relative to state, it can reveal moment-to-moment player actions and feedback.

    “Furthermore, as a sequence of potential actions and timings in relation to state, ERCs can surface information about moment-to-moment player actions and feedback.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Premisenone given

    Designing expressive behavior for AI characters is hard because the possibility spaces such models create are complex.

    “Designing expressive character behavior for AI performance is a difficult undertaking thanks to the complexity of the possibility spaces they create.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The least desirable state is one where the goal affect is among the highest-valued affects but is not the prevailing affect, because actions that appear to increase the goal affect's value will not do so, potentially causing the search to get stuck.

    “We consider the least desirable state to be one where the goal affect is among the affects with the highest value but is not the prevailing affect because any action taken that appears to increase the goal affect's value further will not in fact, which can cause the search to become stuck trying options which do not help it achieve its expressive goal.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Puppitor's prevailing affect system makes signaling state transitions a core feature of its interface, removing the need to synthesize a new approach to identifying human-perceivable state transitions.

    “Rather than having to derive a novel method for discovering important state transitions, Puppitor's prevailing affect system make this a core feature of its interface.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    Current widely available visualization techniques limit the authors to a single starting point per diagram.

    “While we have been able to visualize small pieces of a Puppitor ruleset, current widely available visualization techniques limit us to a single starting point per diagram.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The size and complexity of expressive possibility spaces can make searches spend huge numbers of repeated actions on tiny state changes.

    “Finally, the complexity and scale of expressive possibility spaces can lead to searches becoming bogged down taking huge numbers of repeated actions to make minute changes to the state.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The Sankey diagrams maintain some information about time progression through node column positions but flatten details about exact progression over time.

    “The Sankey diagrams do maintain some of the information about the progression of time through nodes' column positions, but they flatten the details about the exact progression over time.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Each step the search finds is fractionally small progress toward the goal, so many steps in a path will repeat the same move, wasting calculation time.

    “Each step the search finds is fractionally small progress towards the goal and as a result many of the steps in the path will be performing the exact same move over and over, wasting time on calculations.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    Puppitor features clearly defined states with a well-defined relationship to player actions, complete with values that could easily be turned into domain information for A* search.

    “It features clearly defined states with a well defined relationship to actions a player could perform, complete with values that could easily be turned into domain information for A* search.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The clarity of Puppitor's original domain made the translation process relatively straightforward, but this will not necessarily hold for other current and future expressive domains where ERCs may be useful.

    “This clarity in the original domain made for a relatively straightforward translation process, but this will not necessarily hold true for other current and future expressive domains which ERCs may be useful to.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Combined with the more intuitive view of an expressive possibility space facilitated by ERCs, the authors can gain a far better understanding of the complexity of expressive domains and the nuances of their relationship to the development of AI characters focused on performance and expression.

    “Combined with the more intuitive view of an expressive possibility space facilitated by ERCs, we can gain a far better understanding of the complexity of expressive domains and the nuances of their relationship to the development of AI characters focused on performance and expression.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The heuristic is inadmissible because there is no guarantee that the greatest-magnitude affect value remains constant, since Puppitor allows rules to update every available affect value and the rules used in this case study do so.

    “While a useful estimate, this heuristic is inadmissible as there is no guarantee of the value of the greatest magnitude remaining constant as Puppitor allows rules to update every available affect value and our rules used in this case study follow this pattern.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The creation and analysis of ERCs for Puppitor enabled informed iteration of domain knowledge for A* and provided insight into how domain knowledge changes the expressive performance of the search.

    “The creation and analysis of ERCs for Puppitor enabled informed iteration of domain knowledge for A* as well as garnered insight into the way that domain knowledge changes the expressive performance of the search.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The paths found by A* through the affect-vector graph correspond to meaningful sequences of player actions and experiences in Puppitor.

  • Classified as probably not a substantial claim.

    Premisenone given

    Puppitor has a built-in way of signaling human-perceptible changes in its state: its prevailing affects.

    “Additionally, Puppitor has a built in way of signaling changes in its state that are human perceptible, its prevailing affects.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Highlighting steps where the prevailing affect changes and where a new move is made allows analysis of a ruleset's responsiveness, both for how quickly a desired affect can be expressed and for responsiveness to new moves.

    “Highlighting both of these steps allows us to analyze the responsiveness of a given ruleset, both for how quickly a desired affect can be expressed as well as the responsiveness of performing new moves.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Computing costs during the search instead of fixing them in the graph lets A*'s behavior be steered to favor different domain features and to explore more action-sequence permutations.

    “This decision allows the behavior of A* to be directed to prioritize different features of the domain and more effectively explore different permutations of action sequences.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The combinatorial explosion of even two or three starting points in a single diagram quickly renders them unusable for analysis.

    “The combinatorial explosion of even two or three starting points in a single diagram quickly ren-ders them unusable for the purpose of analysis.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    ERC maps the specific actions a player takes to perform a given expression, in order to reveal the affordances of an expressive possibility space.

    “ERC allows us to map specific actions taken by a player to perform a particular expression to understand the affordances of an expressive possibility space.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    A* visits graph nodes breadth-first and uses a priority queue keyed on cost and heuristic values to steer the search toward a goal.

    “A* uses a breadth-first approach to visiting graph nodes and a priority queue based on cost and heuristic values to guide the search towards a specific goal (Hart, Nilsson, and Raphael 1968).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Hart, Nilsson, and Raphael 1968 source

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    For developers, having a fast way to view the effects of changes in domain knowledge and representation on AI characters allows for more deliberate iteration.

    “For developers, having a fast way to view the effects that changes in domain knowledge and representation have on AI characters allows for much more deliberate iteration.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The diagrams in the case study represent individual paths through small corners of a single Puppitor ruleset and are not fully comprehensive captures of the entire expressive possibility space.

    “These diagrams only represent individual paths to single possible ways of expressing affects and are not fully comprehensive captures of the entire expressive possibility space.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    A single affect vector by itself provides insufficient information for A* to search effectively between nodes and does not by itself reveal human-perceivable aspects.

    “An individual affect vector does not contain enough information for A* to effectively search between nodes and also does not easily expose the human perceivable parts of itself on its own.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    ERCs can help identify overly homogeneous areas of a Puppitor ruleset and affects that are complex or impossible to express from certain points in the ruleset.

    “On the player-oriented side, the use of ERCs can help identify overly homogeneous areas of a Puppitor ruleset and affects that are complex or impossible to express from certain points in a ruleset.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Comprehensively visualizing the information provided by ERCs is a challenge due to nuances in action sequences and their relationship to surfaced player feedback even in smaller domains like Puppitor.

    “Comprehensively visualizing the information provided by ERCs is a challenge due to the nuances in the sequences of actions and their relationship to the surfaced player feedback present even in smaller domains like Puppitor.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    For node representations to yield ERCs, human-perceivable state changes must be made primary components of a node.

    “The representation of nodes able to produce ERCs requires that human perceivable changes to state be made primary components of a node.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Not all domains that can benefit from ERCs would have Puppitor's built-in features.

    “Through our experience creating ERCs for Puppitor, we recognize that not all domains which can benefit from ERCs would have these features built in.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The system's affordance lightens the translation and adaptation burden by removing the need to synthesize a new approach to identifying human-perceivable state transitions.

    “The system's affordance therefore lightens the translation and adaptation burden by removing the need to synthesize a new approach to identifying human perceivable state transitions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    A* need not visit every node in a graph, and without an admissible heuristic it need not find an optimal path.

    “Unlike breadth-first search, A* is not guaranteed to visit every node in a graph, nor is it guaranteed to find an optimal path without an admissible heuristic, a heuristic function which never over-estimates the distance to the goal (Hart, Nilsson, and Raphael 1968).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Hart, Nilsson, and Raphael 1968 source

  • May combine two claims.

    Interpretationown data

    Through the case study, the authors demonstrated that insights provided by ERC maps allow more informed reasoning about the domain by designers, which can lead to new ways of guiding A* performance within an expressive possibility space.

    “Through our case study, we demonstrated that the insights provided by ERC maps allow for more informed reasoning about the domain on the part of designers, which in turn can lead to new ways of guiding A* performance within an expressive possibilty space.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The paths found by A* through the affect-vector graph correspond to meaningful sequences of player actions and experiences in Puppitor.

  • May combine two claims.

    Interpretationreasoning

    The first cost function generally leads to longer paths because larger changes cost more, so the search tends toward the smallest possible steps toward its goal.

    “The first cost function used generally leads to longer paths as changes of greater magnitude cost more, so the search will tend towards the smallest steps possible towards its goal.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Because Puppitor's possibility space is complex, having multiple edge cost functions can help map as much of the space as possible.

    “Due to the complexity of the possibility space created by a Puppitor ruleset, having access to multiple edge cost functions can aid our goal of mapping as much of the space as possible.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Puppitor provided an ideal testing platform for the creation and analysis of ERCs.

    “Puppitor provided an ideal testing platform for the creation and analysis of ERCs.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    When reconstructing paths between precalculated steps, it is crucial to recalculate the prevailing affect with each added step because Puppitor rules allow a single update to change every value in an affect vector, so reconstructed steps can have different prevailing affects than either interpolated node.

    “When reconstructing the path between steps, it is crucial to recalculate the prevailing affect with each step added.This part of the process is necessary as Puppitor rules allow a single update to change every value in an affect vector, meaning these reconstructed steps can have different prevailing affects than either node they are interpolating between.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The Sankey diagrams produced by the analysis are useful for illustrating flow between affects and their relationship to player actions, but they abstract away some important information for analyzing game feel.

    “Knowing this, we note that while the Sankey diagrams our analysis of ERCs produced prove useful in illustrating the flow between affects and their relationship to player actions, these diagrams abstract away some important information for analyzing game feel.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Puppitor rules can be written with very small update values; this is useful for continuous experiences over time but causes many repeated updates and slow traversal when searching expressive space.

    “Puppitor rules can be written with very small update values.This authoring pattern is useful at creating continuous experiences over time but when searching across expressive space, it leads to huge numbers of repeated updates and slow traversal.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    By creating ERCs using Puppitor's character rulesets, the authors gained significantly more insight into the affordances of the system as an expressive domain for players and developers.

    “By creating ERCs using Puppitor's character rulesets, we have been able to gain significantly more insight into the affordances of the system as an expressive domain for players and developers working in this area.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The paths found by A* through the affect-vector graph correspond to meaningful sequences of player actions and experiences in Puppitor.
55 of 55

  • The paper relies on this without support.

    A shortage of tools and techniques for understanding expressive possibility spaces is a genuine problem.

    “A lack of available tools and techniques to assist in understanding expressive possibility spaces reinforces this problem.”

  • The paper relies on this without support.

    Human playtesting and observation tend to leave large parts of expressive possibility spaces unexplored and underused.

    “which can leave large areas of these spaces unexplored and underutilized.”

  • The paper relies on this without support.

    Core principles of automated playtesting apply to expressive models.

    “With our goal in mind, we view core principles of automated playtesting as applicable to expressive models, helping us expose information about the expressive possibility spaces created by these models in more intuitive ways than currently exist.”

  • The paper relies on this without support.

    Six starting affect vectors with unique prevailing affects are enough to cover multiple parts of the possibility space for the low number of trials.

    “For our search process, we picked six different starting affect vectors, each with a unique prevailing affect so we would cover multiple parts of the possibility space with even a low number of trials.”

  • The paper relies on this without support.

    The responsiveness of ERCs, measured by steps between prevailing affect changes or new moves, corresponds to the relative player experience of expressing affects.

    “By evaluating a Puppitor ruleset by the responsiveness of the ERCs it produces, we can highlight the relative player experience of expressing affects from different locations in the possibility space.”

  • The paper relies on this without support.

    The paths found by A* through the affect-vector graph correspond to meaningful sequences of player actions and experiences in Puppitor.

    “Through our case study, we demonstrated that the insights provided by ERC maps allow for more informed reasoning about the domain on the part of designers, which in turn can lead to new ways of guiding A* performance within an expressive possibilty space.”
6 of 6

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Second cost formula does not favor larger changes

    Minor

    The formula assigns the same cost when both affect changes are zero as when both are equally large. It therefore favors similar-sized changes, not the greatest change in both affects as claimed.

1 of 1

Cites

Flagged first, then by citations

2 of the 31 references couldn’t be matched to a known paper.

Related

Authors
Junius, N., & Carstensdottir, E.
Affiliations
University of California, Santa Cruz (US)
ORCID
Junius, Nic
Venue
Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment
Publisher
Association for the Advancement of Artificial Intelligence (AAAI)
Published
2023-10-06
Volume
19
Issue
1
Pages
284-294
ISSN
  • 2326-909X (ISSN-L)
  • 2334-0924
DOI
10.1609/aiide.v19i1.27524
OpenAlex
W4387394698
Cited by
0
References
31
FWCI
0.00
Type
conference paper
Language
en
Open access
Locations
Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (publishedVersion)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence in Games
  • Physical Sciences › Engineering › Control and Systems Engineering › Human Motion and Animation
  • Social Sciences › Sociology and Political Science › Digital Games and Media
Keywords
game feel, A* algorithm, affordances

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-01.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-01.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review