arxiv: 2605.01359 · v1 · submitted 2026-05-02 · 💻 cs.AI

Recognition: unknown

Structural Ranking of the Cognitive Plausibility of Computational Models of Analogy and Metaphors with the Minimal Cognitive Grid

Alessio Donvito , Antonio Lieto

Authors on Pith no claims yet

Pith reviewed 2026-05-09 14:30 UTC · model grok-4.3

classification 💻 cs.AI

keywords cognitive plausibilityanalogymetaphorMinimal Cognitive Gridcomputational modelsAI evaluationcognitive theories

0 comments

The pith

The Minimal Cognitive Grid enables structural ranking of computational models of analogy and metaphor by their cognitive plausibility.

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

The paper uses the Minimal Cognitive Grid to systematically evaluate models including the Structure-Mapping Engine, CogSketch, METCL, and large language models. It formalizes the grid's three dimensions of Functional/Structural Ratio, Generality, and Performance Match to measure alignment with cognitive theories of analogy and metaphor. This matters because it supplies a consistent mathematical basis for comparing artificial systems against human cognitive processes instead of relying on informal judgments. A reader might use this to identify which models best capture the structural and functional aspects of human reasoning in these areas.

Core claim

Through the analysis of its three main dimensions (Functional/Structural Ratio, Generality, and Performance Match), the Minimal Cognitive Grid examines how well each system aligns with standard cognitive theories of the modeled phenomena, thus allowing for comparison of the models with respect to their cognitive plausibility, according to consistent and generalizable mathematical criteria.

What carries the argument

The Minimal Cognitive Grid, operationalized through the dimensions of Functional/Structural Ratio, Generality, and Performance Match, which quantifies alignment between computational models and cognitive theories of analogy and metaphor.

Load-bearing premise

That the three dimensions of the Minimal Cognitive Grid and their quantitative operationalization fully capture cognitive plausibility without omitting critical aspects of human analogy and metaphor processing.

What would settle it

An experiment in which models ranked highest by the Minimal Cognitive Grid fail to match human performance patterns in new analogy or metaphor tasks not used in the initial assessment.

Figures

Figures reproduced from arXiv: 2605.01359 by Alessio Donvito, Antonio Lieto.

**Figure 1.** Figure 1: Sensitivity of the FSR metric to ±30% variations in constraint weights. Panel (A): positive perturbations; Panel (B): negative perturbations. Color intensity represents the magnitude of percentage change relative to the baseline configuration. 12 view at source ↗

**Figure 2.** Figure 2: A schematic representation of the Cattell-Horn-Carroll taxonomy of human cognitive abilities. The general intelligence factor (G) branches into an array of cognitive/sensory-motor general domains, each further subdivided in a set of broad (Stratum II) and narrow (Stratum I) abilities. 14 view at source ↗

read the original abstract

In this paper, we employ the Minimal Cognitive Grid (MCG), a framework created to evaluate the cognitive plausibility of artificial systems, to offer a systematic assessment of leading computational models of analogy and metaphor, including the Structure-Mapping Engine (SME), CogSketch, METCL, and Large Language Models (LLMs). We present a formal and quantitative operationalization of the MCG framework and, through the analysis of its three main dimensions (Functional/Structural Ratio, Generality, and Performance Match), examine how well each system aligns with standard cognitive theories of the modeled phenomena, thus allowing for comparison of the models with respect to their cognitive plausibility, according to consistent and generalizable mathematical criteria.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper applies the existing MCG to produce a quantitative ranking of analogy models but supplies no actual scores, equations, or validation steps in the abstract.

read the letter

The main thing this paper does is take the Minimal Cognitive Grid and use its three dimensions to score models like SME, CogSketch, METCL, and LLMs for how well they match cognitive theories of analogy and metaphor. It turns the comparison into something that looks like consistent math rather than loose discussion, which is a step forward for anyone trying to pick models that align with human reasoning data. That systematic angle is the part worth noting, even if it builds directly on prior MCG work from the same group. The soft spots are clear from the abstract alone. No equations appear for how Functional/Structural Ratio, Generality, or Performance Match get computed, and there are no reported scores, error bars, or checks against actual human experiments. Without those, the rankings stay untestable and could just restate the framework's own definitions. The stress-test point about missing incremental acquisition and error signatures also lands, since nothing in the description shows those aspects are measured. This is aimed at people working on cognitive modeling who already know the MCG and want a shared scoring tool for plausibility checks. It could help that narrow group sort models, but only if the full paper adds the missing calculations and some external test. I would send it to peer review so referees can examine the operationalization and see whether the numbers actually track human data or just the authors' prior choices.

Referee Report

2 major / 1 minor

Summary. The paper employs the Minimal Cognitive Grid (MCG) framework to systematically assess and rank the cognitive plausibility of computational models of analogy and metaphor, including SME, CogSketch, METCL, and LLMs. It presents a formal quantitative operationalization of the MCG's three dimensions (Functional/Structural Ratio, Generality, and Performance Match) to compare how well each model aligns with standard cognitive theories via consistent mathematical criteria.

Significance. If the operationalization is shown to be non-circular and the three dimensions proven sufficient, this could supply a reproducible, generalizable method for quantifying cognitive plausibility in AI models of analogy and metaphor, strengthening links between computational systems and cognitive science.

major comments (2)

[Abstract] Abstract: the abstract describes the approach and dimensions but supplies no equations, data, validation steps, or error analysis, so it is impossible to verify whether the operationalization supports the ranking claims.
[MCG operationalization] MCG operationalization section: the claim that the three dimensions (Functional/Structural Ratio, Generality, Performance Match) plus their quantitative scoring rules constitute a sufficient and non-arbitrary proxy for alignment with 'standard cognitive theories' inherits any gaps in the underlying MCG definition; no systematic coverage of competing accounts (e.g., relational priming or developmental trajectories) is shown to confirm completeness.

minor comments (1)

The manuscript would benefit from a summary table listing the numerical scores for each model on the three dimensions to make the final rankings immediately visible.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments on our manuscript. We address each major comment point by point below, indicating where revisions will be made to improve clarity and completeness.

read point-by-point responses

Referee: [Abstract] Abstract: the abstract describes the approach and dimensions but supplies no equations, data, validation steps, or error analysis, so it is impossible to verify whether the operationalization supports the ranking claims.

Authors: The abstract is designed as a concise overview of the paper's goals and contributions. The equations defining the Functional/Structural Ratio, Generality, and Performance Match dimensions, the application to specific models (SME, CogSketch, METCL, and LLMs), the resulting rankings, and supporting analysis are all detailed in the Methods, Results, and Discussion sections. We agree that the abstract could better preview these elements. In the revised manuscript, we will add a brief statement summarizing the quantitative scoring approach and key ranking outcomes to allow readers to assess the claims more readily from the abstract. revision: yes
Referee: [MCG operationalization] MCG operationalization section: the claim that the three dimensions (Functional/Structural Ratio, Generality, Performance Match) plus their quantitative scoring rules constitute a sufficient and non-arbitrary proxy for alignment with 'standard cognitive theories' inherits any gaps in the underlying MCG definition; no systematic coverage of competing accounts (e.g., relational priming or developmental trajectories) is shown to confirm completeness.

Authors: The MCG is a pre-existing minimal framework whose three dimensions were selected to capture essential aspects of cognitive plausibility for computational models. Our contribution lies in the formal quantitative operationalization and its application here. We acknowledge that explicit discussion of alternative cognitive accounts would strengthen the argument for sufficiency. In revision, we will add a dedicated paragraph in the operationalization section that relates the dimensions to competing perspectives such as relational priming and developmental trajectories, supported by citations to the cognitive science literature, while noting the minimal nature of the grid. revision: partial

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper applies the MCG framework (with its three dimensions) as an external evaluative lens to rank existing models of analogy and metaphor. The abstract and provided text present this as an operationalization and application step rather than a self-referential derivation where outputs are forced by redefinition of inputs. No equations or steps are shown reducing a claimed result to a fitted parameter or self-citation chain by construction. The framework is cited as pre-existing, and the analysis compares models against it without the ranking itself being the definitional input. This is a standard use of an author-developed metric for comparative evaluation and does not meet the threshold for flagged circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that standard cognitive theories are correctly captured by the three MCG dimensions and that the quantitative operationalization is faithful to those theories.

axioms (1)

domain assumption Standard cognitive theories of analogy and metaphor are accurately represented by the Functional/Structural Ratio, Generality, and Performance Match dimensions.
Invoked when using these dimensions to assess alignment with cognitive theories.

pith-pipeline@v0.9.0 · 5419 in / 1189 out tokens · 20467 ms · 2026-05-09T14:30:25.812305+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

262 extracted references · 33 canonical work pages · 6 internal anchors

[1]

2019 , eprint=

GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding , author=. 2019 , eprint=

2019
[2]

1994 , publisher=

Unified theories of cognition , author=. 1994 , publisher=

1994
[3]

Journal of Experimental & Theoretical Artificial Intelligence , volume=

Psychometric artificial intelligence , author=. Journal of Experimental & Theoretical Artificial Intelligence , volume=. 2011 , publisher=

2011
[4]

Frontiers in Robotics and AI , volume=

Analyzing the explanatory power of bionic systems with the minimal cognitive grid , author=. Frontiers in Robotics and AI , volume=. 2022 , publisher=

2022
[5]

AI & SOCIETY , pages=

Artificial creativity: can there be creativity without cognition? , author=. AI & SOCIETY , pages=. 2025 , publisher=

2025
[6]

Cognitive Systems Research , pages=

The ghost of behaviorism: critical reflections on methodological limitations in the research of large language models psychology , author=. Cognitive Systems Research , pages=. 2026 , publisher=

2026
[7]

Neurocognitive Foundations of Mind , pages=

Confirmation and explanation in neuroscience: Reassessing the relationship between functional and mechanistic approaches , author=. Neurocognitive Foundations of Mind , pages=. 2026 , publisher=

2026
[8]

Behavioral and brain sciences , volume=

Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources , author=. Behavioral and brain sciences , volume=. 2020 , publisher=

2020
[9]

2020 , eprint=

SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems , author=. 2020 , eprint=

2020
[10]

Journal of Experimental & Theoretical Artificial Intelligence , volume=

A description logic framework for commonsense conceptual combination integrating typicality, probabilities and cognitive heuristics , author=. Journal of Experimental & Theoretical Artificial Intelligence , volume=. 2020 , publisher=

2020
[11]

Philosophy of science , volume=

The role of models in science , author=. Philosophy of science , volume=. 1945 , publisher=

1945
[12]

2002 , publisher=

The discovery of the artificial: Behavior, mind and machines before and beyond cybernetics , author=. 2002 , publisher=

2002
[13]

2013 , publisher=

Explaining the computational mind , author=. 2013 , publisher=

2013
[14]

Dimensions of Mind: A Symposium , editor =

Putnam, Hilary , title =. Dimensions of Mind: A Symposium , editor =
[15]

2012 , publisher=

Artificial intelligence: A modern approach;[the intelligent agent book] , author=. 2012 , publisher=

2012
[16]

Proceedings of the National Academy of Sciences , volume=

The simulation of judgment in LLMs , author=. Proceedings of the National Academy of Sciences , volume=. 2025 , publisher=

2025
[17]

Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse

Abstract Meaning Representation for Sembanking , author=. Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse. 2013

2013
[18]

Topics in Cognitive Science , volume =

Forbus, Kenneth and others , title =. Topics in Cognitive Science , volume =. doi:https://doi.org/10.1111/j.1756-8765.2011.01149.x , year =

work page doi:10.1111/j.1756-8765.2011.01149.x 2011
[19]

, title =

Speer R., Havasi C., and Lieberman H. , title =. Proceedings of the 23rd National Conference on Artificial Intelligence - Volume 1 , pages =. 2008 , isbn =

2008
[20]

Proceedings of the 2nd AAAI Conference on Collaboratively-Built Knowledge Sources and Artificial Intelligence , pages =

Havasi, Catherine and others , title =. Proceedings of the 2nd AAAI Conference on Collaboratively-Built Knowledge Sources and Artificial Intelligence , pages =. 2010 , publisher =

2010
[21]

Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence , pages =

Speer, Robyn and Chin, Joshua and Havasi, Catherine , title =. Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence , pages =. 2017 , publisher =

2017
[22]

2024 , eprint=

GSR-BENCH: A Benchmark for Grounded Spatial Reasoning Evaluation via Multimodal LLMs , author=. 2024 , eprint=

2024
[23]

OpenEQA: Embodied Question Answering in the Era of Foundation Models , booktitle =

Majumdar, Arjun and others , year =. OpenEQA: Embodied Question Answering in the Era of Foundation Models , booktitle =
[24]

2024 , eprint=

Can Large Language Models generalize analogy solving like people can? , author=. 2024 , eprint=

2024
[25]

Using Analogical Reasoning to Prompt LLMs for their Intuitions of Abstract Spatial Schemas , author=
[26]

arXiv preprint arXiv:2305.05050 , year=

ANALOGICAL--A Novel Benchmark for Long Text Analogy Evaluation in Large Language Models , author=. arXiv preprint arXiv:2305.05050 , year=

work page arXiv
[27]

Nature Human Behaviour , volume=

Emergent analogical reasoning in large language models , author=. Nature Human Behaviour , volume=. 2023 , publisher=

2023
[28]

Advances in connectionist and neural computation theory , volume=

The copycat project: A model of mental fluidity and analogy-making , author=. Advances in connectionist and neural computation theory , volume=. 1995 , publisher=

1995
[29]

2020 , eprint=

Language Models are Few-Shot Learners , author=. 2020 , eprint=

2020
[30]

J. C. Raven , title =
[31]

2024 , eprint=

Identifying Semantic Similarity for UX Items from Established Questionnaires Using ChatGPT-4 , author=. 2024 , eprint=

2024
[32]

Artificial intelligence , volume=

The structure-mapping engine: Algorithm and examples , author=. Artificial intelligence , volume=. 1989 , publisher=

1989
[33]

, author=

Episodic and semantic memory. , author=. 1972 , publisher=

1972
[34]

Journal of verbal learning and verbal behavior , volume=

Retrieval time from semantic memory , author=. Journal of verbal learning and verbal behavior , volume=. 1969 , publisher=

1969
[35]

, Xie X: A survey on evaluation of large language models

Chang, Yupeng and others , title =. 2024 , issue_date =. doi:10.1145/3641289 , journal =

work page doi:10.1145/3641289 2024
[36]

2017 , eprint=

Attention Is All You Need , author=. 2017 , eprint=

2017
[37]

2024 , eprint=

Text Clustering as Classification with LLMs , author=. 2024 , eprint=

2024
[38]

2023 , eprint=

Large Language Models Enable Few-Shot Clustering , author=. 2023 , eprint=

2023
[39]

2023 , eprint=

EvEval: A Comprehensive Evaluation of Event Semantics for Large Language Models , author=. 2023 , eprint=

2023
[40]

arXiv:1901.09069 (2019)

Word embeddings: A survey , author=. arXiv preprint arXiv:1901.09069 , year=

work page arXiv 1901
[41]

context-predicting semantic vectors , author=

Don’t count, predict! a systematic comparison of context-counting vs. context-predicting semantic vectors , author=. Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=
[42]

The 7th international student conference on advanced science and technology ICAST , volume=

Semantic cosine similarity , author=. The 7th international student conference on advanced science and technology ICAST , volume=. 2012 , organization=

2012
[43]

2016 , eprint=

The LAMBADA dataset: Word prediction requiring a broad discourse context , author=. 2016 , eprint=

2016
[44]

arXiv preprint arXiv:1909.01066 , year=

Language models as knowledge bases? , author=. arXiv preprint arXiv:1909.01066 , year=

work page arXiv 1909
[45]

2021 , publisher=

Cognitive design for artificial minds , author=. 2021 , publisher=

2021
[46]

Large Language Models in Cybersecurity: Threats, Exposure and Mitigation , pages=

Fundamental Limitations of Generative LLMs , author=. Large Language Models in Cybersecurity: Threats, Exposure and Mitigation , pages=. 2024 , publisher=

2024
[47]

The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

Vitron: A unified pixel-level vision llm for understanding, generating, segmenting, editing , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=
[48]

arXiv preprint arXiv:2401.17981 , year=

Enhancing multimodal large language models with vision detection models: An empirical study , author=. arXiv preprint arXiv:2401.17981 , year=

work page arXiv
[49]

arXiv preprint arXiv:2405.17247 , year=

An introduction to vision-language modeling , author=. arXiv preprint arXiv:2405.17247 , year=

work page arXiv
[50]

arXiv preprint arXiv:2202.10936 (2022)

A survey of vision-language pre-trained models , author=. arXiv preprint arXiv:2202.10936 , year=

work page arXiv
[51]

2024 , publisher=

Large Language Models in Cybersecurity: Threats, Exposure and Mitigation , author=. 2024 , publisher=

2024
[52]

arXiv preprint arXiv:2311.15732 , year=

GPT4Vis: what can GPT-4 do for zero-shot visual recognition? , author=. arXiv preprint arXiv:2311.15732 , year=

work page arXiv
[53]

International conference on machine learning , pages=

Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

2021
[56]

2024 , eprint=

MMBench: Is Your Multi-modal Model an All-around Player? , author=. 2024 , eprint=

2024
[57]

2024 , eprint=

GPT-4 Technical Report , author=. 2024 , eprint=

2024
[58]

2024 , eprint=

Gemini: A Family of Highly Capable Multimodal Models , author=. 2024 , eprint=

2024
[59]

2022 , eprint=

Flamingo: a Visual Language Model for Few-Shot Learning , author=. 2022 , eprint=

2022
[60]

2023 , eprint=

BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models , author=. 2023 , eprint=

2023
[61]

IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

Lvlm-ehub: A comprehensive evaluation benchmark for large vision-language models , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=
[63]

Advances in Neural Information Processing Systems , volume=

Lamm: Language-assisted multi-modal instruction-tuning dataset, framework, and benchmark , author=. Advances in Neural Information Processing Systems , volume=
[64]

National Science Review , pages=

A survey on multimodal large language models , author=. National Science Review , pages=. 2024 , publisher=

2024
[66]

Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages=

A corpus and cloze evaluation for deeper understanding of commonsense stories , author=. Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages=

2016
[67]

Thirteenth international conference on the principles of knowledge representation and reasoning , year=

The winograd schema challenge , author=. Thirteenth international conference on the principles of knowledge representation and reasoning , year=
[68]

2024 47th MIPRO ICT and Electronics Convention (MIPRO) , pages=

Hallucinations in llms: Understanding and addressing challenges , author=. 2024 47th MIPRO ICT and Electronics Convention (MIPRO) , pages=. 2024 , organization=

2024
[69]

2024 , eprint=

Do Large Language Models Latently Perform Multi-Hop Reasoning? , author=. 2024 , eprint=

2024
[70]

What is special about language? , author=

Language and other cognitive systems. What is special about language? , author=. Language learning and development , volume=. 2011 , publisher=

2011
[71]

, author=

Features of similarity. , author=. Psychological review , volume=. 1977 , publisher=

1977
[72]

, author=

Text comprehension, memory, and learning. , author=. American psychologist , volume=. 1994 , publisher=

1994
[73]

Cognitive science , volume=

Structure-mapping: A theoretical framework for analogy , author=. Cognitive science , volume=. 1983 , publisher=

1983
[74]

, author=

Structure mapping in analogy and similarity. , author=. American psychologist , volume=. 1997 , publisher=

1997
[75]

1987 , publisher=

Mechanisms of analogical learning , author=. 1987 , publisher=

1987
[76]

The Oxford handbook of thinking and reasoning , pages=

Analogy and relational reasoning , author=. The Oxford handbook of thinking and reasoning , pages=
[77]

Wiley interdisciplinary reviews: cognitive science , volume=

Computational models of analogy , author=. Wiley interdisciplinary reviews: cognitive science , volume=. 2011 , publisher=

2011
[78]

Exploring analogy in the large , author=
[79]

Proceedings of the 23rd national conference on Artificial intelligence-Volume 3 , pages=

CogSketch , author=. Proceedings of the 23rd national conference on Artificial intelligence-Volume 3 , pages=
[80]

Proceedings of the Tenth Annual Conference of the Cognitive Science Society , author=
[81]

Proceedings of the 22nd annual meeting of the cognitive science society , pages=

SEQL: Category learning as progressive abstraction using structure mapping , author=. Proceedings of the 22nd annual meeting of the cognitive science society , pages=. 2000 , organization=

2000
[82]

Design studies , volume=

A cognitively based approach to computer integration for design systems , author=. Design studies , volume=. 1996 , publisher=

1996
[83]

, author=

Modeling visual problem solving as analogical reasoning. , author=. Psychological review , volume=. 2017 , publisher=

2017
[84]

Cognitive Systems Research , volume=

Using analogical mapping to simulate time-course phenomena in perceptual similarity , author=. Cognitive Systems Research , volume=. 2009 , publisher=

2009

Showing first 80 references.