Pith. sign in

REVIEW 3 major objections 5 minor 52 references

Can LLMs Generate Good Stories? Insights and Challenges from a Narrative Planning Perspective

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that GPT-4-class LLMs reliably generate causally sound stories only at small scales, while planning with character intentionality and dramatic conflict remains hard and seems to require reinforcement-trained reasoning…

desk verdict A genuinely useful, reproducible benchmark for LLM narrative planning whose conflict success rates are inflated by a validator relaxation the authors already half-admit; the qualitative conclusion survives. read the letter →

arxiv 2506.10161 v1 pith:RL24AWR3 submitted 2025-06-11 cs.CL cs.AI

Yi Wang , Max Kreminski This is my paper · ORCID
classification cs.CLcs.AI
keywords storygenerationnarrativeplanninglargelanguagemodelscausalsoundnesscharacterintentionalitydramaticconflictanswersetprogramminggame
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper treats the question of whether LLMs can write good stories as a question about solving narrative planning problems: a story is good to the extent its action sequence is causally sound, its characters act from established intentions, and it contains dramatic conflict. The authors build a benchmark from three classic narrative-planning examples—a secret agent, Aladdin, and a Western confrontation—and encode the planning criteria in Answer Set Programming so that LLM-generated plans can be checked automatically. Their central finding is that GPT-4-class LLMs reliably produce causally sound plans only at small scales; intentionality and conflict problems defeat most non-reasoning models and remain difficult even for reasoning models. If the finding holds, LLM-based story generation in games should be scoped to short causal plotlines or paired with external structure, rather than expected to manage motivated characters and conflicts on its own.

What carries the argument

The load-bearing object is the benchmark itself: three narrative-planning domains translated into Answer Set Programs and checked by an ASP solver. Three formal definitions carry the argument. Causal soundness requires every action's preconditions to hold when it executes and the narrative goal to hold at the end. Intentionality, following the frame-of-commitment formalism from the intentional-planning literature, requires every intentional action to belong to a subsequence driven by a character's stated goal. Dramatic conflict, following the conflict definition used in the Western-based task, is a threatened causal link: an action by one character undoes an effect that another character's later action depends on, with at least one of the two actions nonexecuted. The pipeline also includes an external-calibration mode in which the ASP solver simulates each LLM-chosen action and feeds the resulting world state and error messages back into the prompt, mimicking a reactive game environment. The encodings use the action language BC+ and follow a known compilation of intentional narrative planning into classical planning, with extra predicates added to record threats and conflicts.

What would settle it

Run the Western/conflict tasks with nonexecuted actions required to be causally sound, as the original conflict definition implies; if the o1 model's success rate collapses or its 'reward hacking' plans disappear, the reported capability depended on the Section III-C3 simplification rather than on genuine conflict planning.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes an uneven capability profile for LLM storytelling. On the original examples, GPT-4o, Claude-3.5, and Claude-3.7 solve the causal secret-agent task with moderate to high success, while nearly all non-reasoning models score zero on the intentional Aladdin task and near-zero on the conflict Western task; only the o1 model solves a meaningful fraction of both. Scaling experiments show o1 holds high causal success up to 16x16 grids, degrades as the Aladdin delegation network deepens, and handles up to about twelve adventurers in the Western domain before collapsing. The characteristic failure modes are skipping causal steps while preserving high-level plot structure, and having characters act on intentions that were never established in the plan—a closed-world tracking failure specific to intentionality. The authors conclude from this that causal soundness is achievable at small scale, intentional and conflictful planning is not, and symbolic narrative planners remain the practical choice for runtime game narrative.

Load-bearing premise

The argument depends on the benchmark's formal rules matching what narrative-planning researchers actually mean by causal soundness, intentionality, and conflict; if the simplifications—especially exempting nonexecuted actions from causal soundness—distort those concepts, the measured success rates stop measuring the claimed storytelling abilities.

Editorial extensions

If this is right

  • LLM story generation in games should currently be scoped to short causally sound plotlines; asking a model to sustain character intentions or staged conflict without external help will frequently fail.
  • External calibration—having a symbolic solver simulate and critique each action—can lift causal-planning success substantially (GPT-4o from 13/30 to 29/30 on the original secret-agent task), at the cost of more tokens and time.
  • Only reinforcement-trained reasoning models solve the intentional and conflict tasks at meaningful rates, so this capability gap is not closed by ordinary instruction-tuned LLMs.
  • Even successful conflict plans can be technically valid but narratively odd, such as conflicts produced by two nonexecuted actions or by characters repeatedly getting snakebitten, so passing the formal validator does not guarantee a commonsense story.
  • For runtime game use, symbolic planners still beat LLMs on cost and speed (0.06 seconds for Aladdin, 28.41 seconds for Western), so the practical near-term system is a hybrid rather than an LLM alone.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: failures on intentionality may be a sharper diagnostic of LLM reasoning than classical planning benchmarks, because they require the model to maintain a closed-world model of characters' goals rather than retrieving story-like associations.
  • Editorial inference: the Aladdin results hint that familiar story knowledge actively misleads LLMs; an obfuscated version of the same task—renaming characters, goals, and items—could test whether removing commonsense priors improves or worsens intentional planning.
  • Editorial inference: the observed reward-hacking behavior suggests that once LLMs are embedded in game loops with symbolic validators, they will optimize the validator's criteria rather than narrative quality, so the validator's simplifications become attack surfaces.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes an ASP-based benchmark for evaluating LLMs as narrative planners. It translates three classic narrative planning domains (Secret Agent, Aladdin, Western) into parameterized task sets with automatic validation of causal soundness, character intentionality, and dramatic conflict. Experiments on six LLMs show that reasoning models such as o1 succeed on small-scale causal planning, struggle with intentionality and conflict at larger scales, and that external calibration improves causal planning. The authors conclude that GPT-4-tier LLMs can generate causally sound stories at small scales, but planning with intentionality and conflict remains challenging and may require RL-trained reasoning models.

Significance. If the benchmark is accepted, this is a useful and reproducible contribution: it operationalizes narratological constraints as formal ASP validation, provides a public taskset, and reports falsifiable success rates. The external ASP validator and the absence of fitted parameters are strengths, and the comparison of several model families gives a concrete picture of current LLM capabilities. However, the conflict encoding has a known over-acceptance problem stated in Section III-C3 and demonstrated by the paper's own reward-hacking observations in Section IV-B3, which limits the interpretation of the conflict results. The intentionality benchmark also uses a restricted definition of intention goals, and the reinforcement-learning claim is not directly tested. These issues need to be resolved before the headline conclusions can be fully accepted.

major comments (3)
  1. [III-C3, IV-B3, Table I, Fig. 3(c)] The decision in Section III-C3 not to require causal soundness of nonexecuted actions makes the conflict validator over-accepting. Under Ware et al.'s definition, a nonexecuted action is an action that would fail if attempted, so it should be a genuine attempt with true preconditions at the attempted state whose effects are blocked by interference. The paper itself reports in Section IV-B3 that successful plans include agents merely 'thinking about' taking medicine without execution, thereby bypassing map navigation, and that plans exploit the setting where preconditions of nonexecuted actions are not enforced. Because the abstract's claim that conflict planning remains challenging is anchored to the Western success rates in Table I and Figure 3(c), the benchmark must be repaired or the success rates re-reported excluding such degenerate plans before the central claim can be accepted.
  2. [Abstract, IV-A] The conclusion that narrative planning 'requires LLMs trained with reinforcement learning for complex reasoning' is not supported by the experimental design. The comparison is between off-the-shelf model families (o1, o1-mini, Claude 3.5, Claude 3.7, GPT-4o, GPT-3.5) that differ in training method, scale, inference-time compute, and possibly prompting; the experiments do not isolate RL training as the causal factor. The authors should either soften the claim to 'reasoning models' or provide evidence from a controlled comparison, such as the same base model with and without RL post-training.
  3. [III-E2, Fig. 3(b)] The Aladdin benchmark restricts intentions to two goal types, dead(Ch) and in possession of (Ch, Obj), which is a substantial narrowing of the IPOCL definition in which intention can be an arbitrary expression over world state. Since the central claim about intentional planning is based on this restricted encoding, the paper should prominently state this limitation in the conclusions or demonstrate that the difficulty persists for a broader class of intention goals. Without this, the claim that 'planning with character intentionality remains challenging' may reflect the restricted task design rather than the general concept.
minor comments (5)
  1. [II-C] There is a typo: 'commmon-sense' should be 'common-sense'.
  2. [Table I] Success rates are reported out of 30 attempts without confidence intervals or significance tests; for key comparisons such as o1 versus Claude-3.7 on the Western domain, exact binomial confidence intervals would help readers assess the stability of the differences.
  3. [Fig. 3] The success rates in Figure 3 are based on 50 random instances per parameter setting but no error bars or variance information are shown; please add confidence intervals or describe how randomness was controlled across instances.
  4. [IV-B3] The paper describes 'reward hacking' behaviors but does not quantify how often they occur among successful plans; a breakdown of successful plans into legitimate and degenerate conflict structures would make the Western results interpretable.
  5. [III-D] The prompt contents are described at a high level, but full prompts are only available externally; including at least one complete prompt in an appendix would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LLM performance is evaluated against an independently specified ASP validator, not against a fitted or self-referential target.

full rationale

The paper's central claims are empirical measurements of LLM success on narrative planning benchmarks. The validation is external to the LLMs: plans are checked by an ASP encoding derived from published formalizations (Riedl & Young's intentional planning, Ware et al.'s conflict model, Haslum's compilation), not from quantities fitted to the LLM outputs. The benchmark tasks are parameterized variants of literature examples (secret agent, Aladdin, Western), and no constants are tuned to make the observed success rates come out a particular way. Self-citations appear only in related work and discussion (e.g., Kreminski & Martens [5], Garbe et al. [22], Wang et al. [31], Wang & Kreminski [43]) and are not load-bearing for any empirical result. The one modeling simplification that could affect what is measured—'nonexecuted actions are not required to be causally sound, although their intentionality must still be justified' (Sec. III-C3)—is a validity caveat, not a circularity: the ASP validator remains a fixed external oracle, and the paper explicitly reports the resulting reward-hacking behavior (Sec. IV-B3). No equation or fitted parameter is equivalent by construction to the reported success rates or to the abstract's conclusions.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters are fitted to data; the benchmark uses fixed domain parameters and standard solvers. The central results rest on domain representativeness and the faithfulness of the ASP encoding, both of which are assumptions inherited from prior literature.

assumptions (3)
  • domain assumption The Secret Agent, Aladdin, and Western domains selected from narrative planning literature are representative of the three target qualities (causal soundness, intentionality, conflict).
    The benchmark's validity depends on these canonical examples standing in for the broader concepts; the authors do not empirically justify representativeness.
  • domain assumption The translation of narrative planning problems into ASP (via BC+ and the compilation from Haslum 2012) preserves the semantics of intentionality and conflict.
    Section III-C describes the encoding; if the compilation is unfaithful, the automatic validation does not measure the intended narrative properties.
  • standard math The Clingo 5.7.1 solver correctly implements answer set semantics for the generated programs.
    Standard trust in a widely used ASP solver; the paper provides no formal proof but this is routine.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can LLMs Generate Good Stories? Insights and Challenges from a Narrative Planning Perspective." pith.science (2026). https://pith.science/paper/RL24AWR3

@misc{pith2026250610161,
  author       = {Pith},
  title        = {Pith review of: Can LLMs Generate Good Stories? Insights and Challenges from a Narrative Planning Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RL24AWR3}},
  note         = {Machine review of arXiv:2506.10161}
}
read the original abstract

Story generation has been a prominent application of Large Language Models (LLMs). However, understanding LLMs' ability to produce high-quality stories remains limited due to challenges in automatic evaluation methods and the high cost and subjectivity of manual evaluation. Computational narratology offers valuable insights into what constitutes a good story, which has been applied in the symbolic narrative planning approach to story generation. This work aims to deepen the understanding of LLMs' story generation capabilities by using them to solve narrative planning problems. We present a benchmark for evaluating LLMs on narrative planning based on literature examples, focusing on causal soundness, character intentionality, and dramatic conflict. Our experiments show that GPT-4 tier LLMs can generate causally sound stories at small scales, but planning with character intentionality and dramatic conflict remains challenging, requiring LLMs trained with reinforcement learning for complex reasoning. The results offer insights on the scale of stories that LLMs can generate while maintaining quality from different aspects. Our findings also highlight interesting problem solving behaviors and shed lights on challenges and considerations for applying LLM narrative planning in game environments.

Figures

Figures reproduced from arXiv: 2506.10161 by the authors.

Figure 1
Figure 1. Our evaluation pipeline consists of the task generation component, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Description of the three story domains we use for benchmarking. Bold text indicates intentional actions. All the fluents are binary-valued except [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. LLM Narrative Planning Performance on the Variations of the Three Story Domains [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 43 canonical work pages

  1. [1]

    GPT for games: A scoping review (2020-2023),

    D. Yang, E. Kleinman, and C. Harteveld, “GPT for games: A scoping review (2020-2023),” in IEEE Conference on Games (CoG) , 2024

    work page 2020

  2. [2]

    Large language models and games: A survey and roadmap,

    R. Gallotta, G. Todd, M. Zammit, S. Earle, A. Liapis, J. Togelius, and G. N. Yannakakis, “Large language models and games: A survey and roadmap,” IEEE Transactions on Games , 2024

    work page 2024

  3. [3]

    Language as reality: a co-creative storytelling game experience in 1001 Nights using generative AI,

    Y . Sun, Z. Li, K. Fang, C. H. Lee, and A. Asadipour, “Language as reality: a co-creative storytelling game experience in 1001 Nights using generative AI,” in Proc. AIIDE, 2023

    work page 2023

  4. [4]

    Playing with unicorns: AI Dungeon and citizen NLP,

    M. Hua and R. Raley, “Playing with unicorns: AI Dungeon and citizen NLP,” DHQ: Digital Humanities Quarterly , vol. 14, no. 4, 2020

    work page 2020

  5. [5]

    Unmet creativity support needs in computationally supported creative writing,

    M. Kreminski and C. Martens, “Unmet creativity support needs in computationally supported creative writing,” in Proceedings of the First Workshop on Intelligent and Interactive Writing Assistants , 2022

    work page 2022

  6. [6]

    A cognitive process theory of writing,

    L. Flower and J. R. Hayes, “A cognitive process theory of writing,” College Composition & Communication , 1981

    work page 1981

  7. [7]

    What makes a good story and how can we measure it? A comprehensive survey of story evaluation,

    D. Yang and Q. Jin, “What makes a good story and how can we measure it? A comprehensive survey of story evaluation,” arXiv preprint, 2024

    work page 2024

  8. [8]

    ROUGE: A package for automatic evaluation of summaries,

    C.-Y . Lin, “ROUGE: A package for automatic evaluation of summaries,” in Text summarization branches out , 2004, pp. 74–81

    2004

Show all 52 references
  1. [9]

    BERTScore: Evaluating text generation with BERT,

    T. Zhang, V . Kishore, F. Wu, K. Q. Weinberger, and Y . Artzi, “BERTScore: Evaluating text generation with BERT,” arXiv, 2019

    2019

  2. [10]

    Art or artifice? Large language models and the false promise of creativity,

    T. Chakrabarty, P. Laban, D. Agarwal, S. Muresan, and C.-S. Wu, “Art or artifice? Large language models and the false promise of creativity,” in CHI Conference on Human Factors in Computing Systems , 2024

    2024

  3. [11]

    Computational narratology,

    I. Mani, “Computational narratology,” Handbook of narratology , 2014

    2014

  4. [12]

    The story so far on narrative planning,

    R. E. C. Rivera, A. Jhala, J. Porteous, and R. M. Young, “The story so far on narrative planning,” in Proc. ICAPS, vol. 34, 2024, pp. 489–499

    2024

  5. [13]

    Narrative planning: Balancing plot and character,

    M. O. Riedl and R. M. Young, “Narrative planning: Balancing plot and character,” Journal of Artificial Intelligence Research , 2010

    2010

  6. [14]

    Controlling narrative generation with planning trajectories: the role of constraints,

    J. Porteous and M. Cavazza, “Controlling narrative generation with planning trajectories: the role of constraints,” in Proc. ICIDS, 2009

    2009

  7. [15]

    TALE-SPIN, an interactive program that writes stories,

    J. R. Meehan, “TALE-SPIN, an interactive program that writes stories,” in IJCAI, vol. 77, 1977, pp. 91–98

    1977

  8. [16]

    A computa- tional model of plan-based narrative conflict at the fabula level,

    S. G. Ware, R. M. Young, B. Harrison, and D. L. Roberts, “A computa- tional model of plan-based narrative conflict at the fabula level,” IEEE Transactions on Computational Intelligence and AI in Games , 2013

    2013

  9. [17]

    The stable model semantics for logic programming,

    M. Gelfond and V . Lifschitz, “The stable model semantics for logic programming,” in 5th Conference on Logic Programming , 1988

    1988

  10. [18]

    A belief-desire-intention model for narrative generation,

    T. Wadsley and M. Ryan, “A belief-desire-intention model for narrative generation,” in Proc. AIIDE, 2013

    2013

  11. [19]

    Headspace: incorporating action failure and character beliefs into narrative planning,

    R. Sanghrajka, R. M. Young, and B. Thorne, “Headspace: incorporating action failure and character beliefs into narrative planning,” in Proc. AIIDE, vol. 18, 2022, pp. 171–178

    2022

  12. [20]

    A computational model of narrative generation for surprise arousal,

    B.-C. Bae and R. M. Young, “A computational model of narrative generation for surprise arousal,” IEEE Transactions on Computational Intelligence and AI in Games , vol. 6, no. 2, pp. 131–143, 2013

    2013

  13. [21]

    Indexter: A computational model of the event-indexing situation model for characterizing narratives,

    R. E. Cardona-Rivera, B. A. Cassell, S. G. Ware, and R. M. Young, “Indexter: A computational model of the event-indexing situation model for characterizing narratives,” in Proceedings of the 3rd Workshop on Computational Models of Narrative , 2012, pp. 34–43

    2012

  14. [22]

    StoryAssembler: an engine for generating dynamic choice-driven nar- ratives,

    J. Garbe, M. Kreminski, B. Samuel, N. Wardrip-Fruin, and M. Mateas, “StoryAssembler: an engine for generating dynamic choice-driven nar- ratives,” in Proc. FDG, 2019

    2019

  15. [23]

    Hierarchical neural story genera- tion,

    A. Fan, M. Lewis, and Y . Dauphin, “Hierarchical neural story genera- tion,” arXiv preprint arXiv:1805.04833 , 2018

    2018 arXiv

  16. [24]

    Evaluating story generation systems using automated linguistic analyses,

    M. Roemmele, A. S. Gordon, and R. Swanson, “Evaluating story generation systems using automated linguistic analyses,” in SIGKDD 2017 Workshop on Machine Learning for Creativity , 2017, pp. 13–17

    2017

  17. [25]

    Wordcraft: story writing with large language models,

    A. Yuan, A. Coenen, E. Reif, and D. Ippolito, “Wordcraft: story writing with large language models,” in Proc. UIST, 2022, pp. 841–852

    2022

  18. [26]

    Plan- and-write: Towards better automatic storytelling,

    L. Yao, N. Peng, R. Weischedel, K. Knight, D. Zhao, and R. Yan, “Plan- and-write: Towards better automatic storytelling,” in Proc. AAAI, 2019

    2019

  19. [27]

    Generative agents: Interactive simulacra of human behavior,

    J. S. Park, J. O’Brien, C. J. Cai, M. R. Morris, P. Liang, and M. S. Bernstein, “Generative agents: Interactive simulacra of human behavior,” in Proc. UIST, 2023

    2023

  20. [28]

    Event representations for automated story generation with deep neural nets,

    L. Martin, P. Ammanabrolu, X. Wang, W. Hancock, S. Singh, B. Harri- son, and M. Riedl, “Event representations for automated story generation with deep neural nets,” in Proc. AAAI, 2018

    2018

  21. [29]

    TattleTale: storytelling with planning and large language models,

    N. Simon and C. Muise, “TattleTale: storytelling with planning and large language models,” in ICAPS Workshop on Scheduling and Planning Applications, 2022

    2022

  22. [30]

    There and back again: extracting formal domains for controllable neurosymbolic story authoring,

    J. Kelly, A. Calderwood, N. Wardrip-Fruin, and M. Mateas, “There and back again: extracting formal domains for controllable neurosymbolic story authoring,” in Proc. AIIDE, 2023

    2023

  23. [31]

    StoryVerse: Towards co-authoring dy- namic plot with LLM-based character simulation via narrative planning,

    Y . Wang, Q. Zhou, and D. Ledo, “StoryVerse: Towards co-authoring dy- namic plot with LLM-based character simulation via narrative planning,” in Proc. FDG, 2024

    2024

  24. [32]

    WhatELSE: Shaping narrative spaces at configurable level of abstraction for AI-bridged interactive storytelling,

    Z. Lu, Q. Zhou, and Y . Wang, “WhatELSE: Shaping narrative spaces at configurable level of abstraction for AI-bridged interactive storytelling,” in the CHI Conference on Human Factors in Computing Systems , 2025

    2025

  25. [33]

    Large language models as narrative planning search guides,

    R. Farrell and S. G. Ware, “Large language models as narrative planning search guides,” IEEE Transactions on Games , 2024

    2024

  26. [34]

    Are large language models capable of generating human-level narratives?

    Y . Tian, T. Huang, M. Liu, D. Jiang, A. Spangher, M. Chen, J. May, and N. Peng, “Are large language models capable of generating human-level narratives?” arXiv preprint arXiv:2407.13248 , 2024

    2024 arXiv

  27. [35]

    Experimental narratives: A comparison of human crowd- sourced storytelling and AI storytelling,

    N. Begu ˇs, “Experimental narratives: A comparison of human crowd- sourced storytelling and AI storytelling,” Humanities and Social Sci- ences Communications, vol. 11, no. 1, 2024

    2024

  28. [36]

    Of human criteria and automatic metrics: A benchmark of the evaluation of story generation,

    C. Chhun, P. Colombo, C. Clavel, and F. M. Suchanek, “Of human criteria and automatic metrics: A benchmark of the evaluation of story generation,” arXiv preprint arXiv:2208.11646 , 2022

    2022 arXiv

  29. [37]

    LLMs can plan only if we tell them,

    B. Sel, R. Jia, and M. Jin, “LLMs can plan only if we tell them,” arXiv preprint arXiv:2501.13545, 2025

    2025 arXiv

  30. [38]

    LLMs can’t plan, but can help planning in LLM-modulo frameworks,

    S. Kambhampati, K. Valmeekam, L. Guan, K. Stechly, M. Verma, S. Bhambri, L. Saldyt, and A. Murthy, “LLMs can’t plan, but can help planning in LLM-modulo frameworks,” arXiv:2402.01817, 2024

    2024 arXiv

  31. [39]

    Answer set programming for procedural content generation: A design space approach,

    A. M. Smith and M. Mateas, “Answer set programming for procedural content generation: A design space approach,” IEEE Transactions on Computational Intelligence and AI in Games , 2011

    2011

  32. [40]

    Answer set programming and plan generation,

    V . Lifschitz, “Answer set programming and plan generation,” Artificial Intelligence, vol. 138, no. 1-2, pp. 39–54, 2002

    2002

  33. [41]

    RoleModel: towards a formal model of dramatic roles for story gener- ation,

    S. Chen, A. M. Smith, A. Jhala, N. Wardrip-Fruin, and M. Mateas, “RoleModel: towards a formal model of dramatic roles for story gener- ation,” in Intelligent Narrative Technologies III Workshop , 2010

    2010

  34. [42]

    Generating explorable narrative spaces with answer set programming,

    C. Dabral and C. Martens, “Generating explorable narrative spaces with answer set programming,” in Proc. AIIDE, 2020

    2020

  35. [43]

    Guiding and diversifying LLM-based story generation via answer set programming,

    P. J. Wang and M. Kreminski, “Guiding and diversifying LLM-based story generation via answer set programming,” arXiv preprint, 2024

    2024

  36. [44]

    Action language bc+,

    J. Babb and J. Lee, “Action language bc+,” Journal of Logic and Computation, vol. 30, no. 4, pp. 899–922, 2020

    2020

  37. [45]

    Clingo = ASP + control: Preliminary report,

    M. Gebser, R. Kaminski, B. Kaufmann, and T. Schaub, “Clingo = ASP + control: Preliminary report,” arXiv preprint arXiv:1405.3694 , 2014

    2014 arXiv

  38. [46]

    Flaw selection strategies for value- directed planning,

    M. Williamson and S. Hanks, “Flaw selection strategies for value- directed planning,” in Proc. ICAPS, 1996, pp. 237–244

    1996

  39. [47]

    Narrative planning: Compilations to classical planning,

    P. Haslum, “Narrative planning: Compilations to classical planning,” Journal of Artificial Intelligence Research , vol. 44, pp. 383–395, 2012

    2012

  40. [48]

    Narrative planning: balancing plot and character,

    M. O. Riedl, “Narrative planning: balancing plot and character,” Ph.D. dissertation, North Carolina State University, 2004

    2004

  41. [49]

    Towards System 2 reasoning in LLMs: Learning how to think with meta chain-of-thought,

    V . Xiang, C. Snell, K. Gandhi, A. Albalak, A. Singh, C. Blagden, D. Phung, R. Rafailov, N. Lile, D. Mahan et al. , “Towards System 2 reasoning in LLMs: Learning how to think with meta chain-of-thought,” arXiv preprint arXiv:2501.04682 , 2025

    2025 arXiv

  42. [50]

    Glaive: a state-space narrative planner supporting intentionality and conflict,

    S. G. Ware and R. M. Young, “Glaive: a state-space narrative planner supporting intentionality and conflict,” in Proc. AIIDE, 2014

    2014

  43. [51]

    A possible worlds model of belief for state-space narrative planning,

    A. Shirvani, S. Ware, and R. Farrell, “A possible worlds model of belief for state-space narrative planning,” in Proc. AIIDE, 2017

    2017

  44. [52]

    A formalization of emotional planning for strong-story systems,

    A. Shirvani and S. Ware, “A formalization of emotional planning for strong-story systems,” in Proc. AIIDE, 2020

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.