Multiagent Narrative Experience Management as Story Graph Pruning

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

Intervention undermines the player's model of the environment's rules and reduces agency.

“Intervention subverts the player's mental model of the environment's rules and decreases agency, the player's feeling that they can take meaningful action to affect the story (Wardrip-Fruin et al. 2009).”

• cites Wardrip-Fruin et al. 2009 sourcePartly supported

  The Discussion says that incorrectly signaled expectations of a game's computational model can be detrimental to agency, and Agency and Interfaces describes increased realism making players less able to act in terms the software supports. The paper does not establish that intervention generally subverts a player's understanding of the rules or decreases agency; its Agency and Improvisation section describes designed interruptions that keep play moving. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

May combine two claims.

After cycle pruning, the story graph contains 23,159,543 nodes (a 23% reduction) and 56,783,502 edges (a 25% reduction).

“After cycle pruning, the graph has 23,159,543 nodes (down 23%) and 56,783,502 edges (down 25%).”

May combine two claims.

The random story graph has 21,115,022 nodes and 60,492,852 edges, which is roughly comparable in size to the pruned graph.

“It has 21,115,022 nodes and 60,492,852 edges, roughly comparable in size to our pruned graph.”

May combine two claims.

After dead end pruning, the story graph contains 20,365,187 nodes (a 5% reduction) and 49,669,351 edges (a 2% reduction).

“After dead end pruning, the graph has 20,365,187 nodes (down 5%) and 49,669,351 edges (down 2%).”

May combine two claims.

The pruning techniques produce a high-agency experience with believable NPC behavior, as supported by playtest results.

“We also claim these pruning techniques result in a high agency experience with believable NPC behavior, and we present the results from a playtest of our game in support.”

May combine two claims.

The final graph has no dead ends, which proves that the story is always finishable because every non-terminal state can reach a terminal state.

“That the story is always finishable is proved by the absence of dead ends in the final graph (i.e. from every non-terminal state these exists a path to a terminal state).”

May combine two claims.

After shorter plan pruning, the graph contained 93,608,267 nodes and 248,440,557 edges, each reduced by 76%.

“After this pruning, the graph has 93,608,267 nodes (down 76%) and 248,440,557 edges (down 76%).”

May combine two claims.

Both groups of 10 participants tended to agree with all four statements, even those who played the random story graph, though the sample was too small for a between-subjects analysis.

“Two groups of 10 participants was not a large enough sample for a between subjects analysis, but both groups tended to agree with all four statements, even those who played the random story graph.”

May combine two claims.

After goal priority pruning, the story graph contains 30,149,245 nodes (a 42% reduction) and 76,006,520 edges (a 45% reduction).

“After goal priority pruning, the graph has 30,149,245 nodes (down 42%) and 76,006,520 edges (down 45%).”

May combine two claims.

The goal priority prune does not prevent all cycles, so cycles of 3 or more edges are detected and broken.

“The above prune does not prevent all cycles, so we detect cycles of 3 or more edges and break them.”

Classified as probably not a substantial claim.

The pruning criteria described in the paper worked well in this domain.

“We found the criteria described below to work well in this domain, and we attempt to justify them by explaining our motivations and presenting illustrative anecdotes, but we do not claim they are best for all domains.”

May combine two claims.

After unique ending pruning, the story graph contains 52,262,059 nodes (a 10% reduction) and 138,072,434 edges (a 7% reduction).

“After unique ending pruning, the graph has 52,262,059 nodes (down 10%) and 138,072,434 edges (down 7%).”

May combine two claims.

After arbitrary pruning, the story graph contains 20,365,197 nodes (a 12% reduction) and 49,669,363 edges (a 13% reduction).

“After this final pruning, the graph has 20,365,197 nodes (down 12%) and 49,669,363 edges (down 13%).”

May combine two claims.

After lazy NPC pruning, the graph contained 58,191,971 nodes and 148,928,950 edges, reduced by 38% and 40% respectively.

“After this pruning, the graph has 58,191,971 nodes (down 38%) and 148,928,950 edges (down 40%).”

Classified as probably not a substantial claim.

The p-values in Table 1 were adjusted using Benjamini and Hochberg's (1995) correction for multiple hypothesis testing.

“The p-values in Table 1 are given after applying Benjamini and Hochberg's (1995) correction for multiple hypothesis testing.”

• cites Benjamini and Hochberg's (1995) sourceNot checked — no full text of the source

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

Pruning a complete story graph will be intractable for most domains, but this work was instructive because it allowed consideration of the long-term consequences of every experience manager decision.

“Pruning a complete story graph will be intractable for most domains, but this work was instructive because it allowed us to consider the long-term consequences of every experience manager decision.”

May combine two claims.

The described situations, such as abandoned plans and cycles, are symptoms of assuming the story graph is Markovian.

“Situations like this are a symptom of assuming the story graph is Markovian.”

• Also stated in: § Goal Priority Pruning · plain

  “Cycles like this are also a symptom of the Markov assumption.”

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

The insights from this work can probably be applied as heuristics to larger graphs that must be generated on demand.

“We believe these insights can be applied, probably as heuristics, to larger graphs which must be generated on demand.”

Classified as probably not a substantial claim.

Participants were randomly assigned to two groups, one playing the random version first and the other playing the intelligent version first.

“Participants were randomly divided into two groups, with one playing the random version first and the other playing the intelligent version first.”

Classified as probably not a substantial claim.

It is important to prune longer plans before applying unique ending pruning.

“It is important to prune longer plans before unique ending pruning.”

May combine two claims.

The method ensures believable NPC actions, a reachable ending, and that the experience manager never has to block a player action.

“We ensure NPCs act believably and that the story can always reach an ending while also ensuring the experience manager never needs to prevent the player from taking an action.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

It is possible and perhaps even desirable for the player to limit endings through their actions.

“it is possible and perhaps even desirable for the player to limit endings with their actions.”

May combine two claims.

Only NPC edges are ever removed, never player edges, which avoids the need to intervene by ensuring the narrative never reaches a state where intervention might be necessary.

“Note that we only ever remove NPC edges, never player edges; in other words, we avoid the need to intervene by ensuring the narrative never reaches a state where intervention might be necessary.”

May combine two claims.

The unique ending pruning is a tie-breaking prune, so it will never remove the last edge for an NPC.

“This is a tie-breaking prune, meaning it will never prune the last edge for an NPC.”

Classified as probably not a substantial claim.

Having the complete story graph available when making pruning decisions lets the experience manager anticipate both immediate and long-term effects of those decisions on the set of possible stories.

“Considering the entire graph allows us to foresee the long-term consequences of every pruning decision on the space of possible stories.”

• Also stated in: § Introduction · plain

  “By having the entire graph in memory when making a pruning decision, we can fully anticipate all the short-and long-term consequences of a decision on the space of possible stories.”

Classified as probably not a substantial claim.

The Markov assumption limits story graph systems because different histories reaching the same state can imply different future actions.

“The Markov assumption is a limitation for story graph systems, because different actions leading to the same state can suggest very different future actions (Farrell and Ware 2016).”

• cites Farrell and Ware 2016 sourceNot checked — no full text of the source

May combine two claims.

The tutorial ensures that participants have explored the world before playing, which allows them to focus on the narrative.

“The tutorial ensures that participants have explored the world before playing the game, allowing them to focus on the narrative.”

May combine two claims.

Because the method never removes player action edges, the experience manager can always permit whatever action the player chooses.

“By never pruning player actions, we ensure the experience manager can accommodate any choice.”

• Also stated in: § Mediation-Based Systems · plain

  “By considering the entire story graph, we ensure that we never prune player action edges-i.e.we always accommodate and never intervene.”

• Also stated in: § Experience Management · plain

  “It is our goal never to prune an edge that requires only the consent of the player, meaning the experience manager will always be able to accommodate any player action.”

May combine two claims.

The human tendency to narrativize events may be so strong that most people cannot see actions as random, only as easier or harder to explain, which makes a randomly generated story graph a stronger baseline than it might seem.

“Perhaps the human tendency to narrativize events (Bruner 1991) is so strong that most people cannot see actions as random, only as easier or harder to explain, and thus a randomly generated story graph is a stronger baseline than it might seem.”

• cites Bruner 1991 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

When a player limits what endings are available, it can be a notable moment of agency.

“When a player limits what endings are available, it can be a notable moment of agency.”

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

Unique ending pruning targets NPC actions because the experience manager tries to avoid limiting the player.

“Unique ending pruning targets NPC actions, because the experience manager tries to avoid limiting the player”

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

The presence of reachable dead ends, which existed in the original graph and remain after pruning, shows that proactive mediation considering long-term consequences is needed.

“The existence of reachable dead ends (which were present in the original graph and persist after all the above pruning) demonstrates the need for proactive mediation which considers long-term consequences.”

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

Story graph pruning will be impractical in larger domains.

“While story graph pruning will be intractable in larger domains, this is an opportunity to discover insights for making decisions when story graphs are too large to generate fully.”