Towards a Neurosymbolic Reasoning System Grounded in Schematic Representations

Fran\c{c}ois Olivier; Zied Bouraoui

REVIEW 3 major objections 4 minor 31 references

Towards a Neurosymbolic Reasoning System Grounded in Schematic Representations

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Large language models can be guided to reason through spatial image schemas encoded as executable programs, reaching 91% on logical deduction and solving a zebra puzzle.

desk verdict A clean but thin proof-of-concept: the architecture is sensible and the code is real, but the headline 91% accuracy is not yet evidence that the spatial encodings are semantically faithful. read the letter →

arxiv 2509.03644 v1 pith:FX4V723V submitted 2025-09-03 cs.AI cs.CL

François Olivier , Zied Bouraoui This is my paper · ORCID

classification cs.AIcs.CL

keywords imageschemasembodiedcognitionneurosymbolicreasoninganswersetprogrammingdeclarativespatiallargelanguagemodelslogicaldeduction

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

Embodied-LM is a prototype that tries to give LLMs the kind of mental representation humans use when reading: image schemas, recurring spatial patterns drawn from sensorimotor experience. The paper's claim is that GPT-4 can be prompted to read a natural-language scenario and translate each statement into a spatial relation—older goes left on a timeline, an owner lives inside a house—and that these relations can be turned into an executable Answer Set Program. A spatial reasoner then checks the constraints and derives the answer, with witnesses showing the geometric arrangement. On the LogicalDeduction dataset the system reaches 91% accuracy, and on a three-house zebra puzzle it returns the correct owner while flagging that two configurations are compatible with the premises. The point is not only accuracy: because the generated program is inspectable, mistakes can be traced to a specific translation error, suggesting a path toward more explainable neurosymbolic reasoning.

What carries the argument

The load-bearing mechanism is the image schema used as a translation layer: a recurring spatial pattern such as PATH (a line with ordered positions) or CONTAINER (a bounded region with an inside) that maps abstract relations to geometric facts. The paper formalizes schemas in Declarative Spatial Reasoning, where objects are points, segments, or rectangles with coordinate parameters and relations are equations and inequalities over those coordinates. These spatial constraints are embedded as theory atoms in Answer Set Programs, a declarative logic-programming paradigm for constraint solving; the Clingo solver searches stable models while a Z3-based propagator checks that the logical requireme

What would settle it

Re-running the pipeline on LogicalDeduction while auditing each generated ASP program for directional consistency—counting programs in which 'newer' is encoded as right_pp in one rule and left_pp in another—would settle whether the 91% accuracy comes from the spatial representation or from the LLM's latent reasoning; if internally inconsistent programs still answer correctly, the schematic encoding is not the source of the success.

Watch

Extended reading notes

Core claim

The central discovery is that image schemas can be made computationally executable: a language model can be guided to interpret a scenario through a small set of spatial primitives (point, segment, rectangle; left, right, inside, on), and the resulting spatial program, once solved by an ASP engine with declarative spatial reasoning, produces answers that follow logically from the premises. The authors show this with the PATH schema, where age and order relations become positions on a line used with on_ps, and the CONTAINER schema, where nationality, pets, and drinks become points inside house rectangles used with in_pr. Because the same schema grounds different domains, the representation is

Load-bearing premise

The whole pipeline trusts the LLM's spatial encoding to preserve the meaning of every statement—for instance, 'older than' must consistently become the same side of a timeline—and if that encoding is inconsistent, the ASP program solves a different problem.

Editorial extensions

If this is right

A single image schema can underwrite several kinds of relation: PATH handles age, order, and position; CONTAINER handles living in, drinking, and owning, so a system trained once on the schema transfers across domains.
Because spatial predicate meanings are fixed and objective, the LLM does not need to define the semantics of its own formulas, reducing a common failure of LLM-generated logic programs.
Cautious reasoning over all stable models lets the system reject an answer option when another model invalidates it, a capability black-box LLM prompting lacks.
Generated programs are human-inspectable, so errors can be localized to specific translation steps and corrected by prompt refinement, making the reasoning pipeline auditable.
The architecture is a foundation for adding spatiotemporal and force-dynamic primitives, opening classic planning-style tasks such as Tower of Hanoi, river-crossing, and Blocks World.

Reading between the lines

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

Editorial inference: The directional-flip failure the authors report could be mitigated by adding explicit consistency rules to the prompt, such as a global 'newer means right' axiom, and measuring the flip rate would give a cheap diagnostic of schema faithfulness.
Editorial inference: If the spatial encoding proves reliable, the coordinate witnesses returned by the solver could serve as human-readable diagrammatic proofs, effectively turning the geometry into an explanation of why an answer follows.
Editorial inference: The same pipeline could be tested on consistency-checking tasks by asking whether the generated spatial program has any stable model at all, turning the system into a satisfiability-based detector of contradictory narratives.
Editorial inference: Because the system's accuracy is bounded by the LLM's translation quality, a natural extension is to sample several interpretations per scenario and let the solver discard those that cannot be geometrically realized, using space itself as a filter on LLM outputs.

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.

Desk Editor's Note

read the letter

The core of this paper is a prototype that works as described: GPT-4 generates ASP programs with spatial predicates, Clingo/DSR does the inference, and the authors report 91% on LogicalDeduction plus one solved zebra puzzle. That is a real implementation, with code and data, and the prompts in the appendices make the pipeline transparent. The soundness pattern—delegating inference to a solver and using cautious reasoning for answer selection—is the right backbone. I believe the system does what they say it does on these examples.

What is actually new is modest: this is a direct application of the authors' own AAMAS 2025 framework to an LLM-to-ASP pipeline, with the specific twist of using image-schema spatial primitives as the target formalism. That is a legitimate increment, not a breakthrough. The paper is honest about being a prototype.

The soft spots are real and proportionate to the claims. The empirical support is thin: one dataset, one puzzle, no variance, no discard rate from the three-attempt protocol, no ablation. More importantly, the 91% number does not, on its own, establish that the schematic representations are faithful. The stress-test concern is on target: the retry loop only checks syntax and satisfiability, not whether the generated program encodes the same problem as the narrative. The authors' own error analysis—"newest" mapped leftward while later constraints assume rightward—confirms exactly this fragility. The example programs in Appendix D are illustrative, not proof of systematic correctness. So the strongest claim, about LLMs being guided to interpret scenarios through schematic structures, is currently unverified. The mild circularity the reader flags also holds: the hand-written prompt examples define the encodings that are then used to demonstrate the value of those encodings.

Even so, I would not call this a flawed paper. The claims are modest, the implementation is transparent, and the authors identify the key limitation themselves. It deserves a serious referee—not a desk reject. The right outcome is probably revision with more rigorous evaluation: report variance and discards, audit a sample of generated programs for semantic fidelity, and add at least one more task domain.

Who is this for? Researchers working on neurosymbolic reasoning or LLM-to-solver pipelines, who will find it a useful data point. I would not cite it in my own work yet, but I would mention it as a proof-of-concept. Bring it to a reading group only if the group is interested in the interpretability angle rather than benchmark performance.

Referee Report

3 major / 4 minor

Summary. The paper introduces Embodied-LM, a prototype neurosymbolic system in which GPT-4 is prompted to translate natural-language reasoning scenarios into Answer Set Programming (ASP) programs that use the Declarative Spatial Reasoning (DSR) framework with spatial predicates intended to formalize image schemas such as PATH and CONTAINER. The resulting programs are solved by Clingo with a custom Z3-based propagator. The authors report 91% accuracy on the LogicalDeduction dataset, a successful solution of a three-house zebra puzzle, and claim that the system provides interpretable, executable schematic representations that support effective logical reasoning. The paper positions this as a proof of concept for grounding LLM reasoning in embodied cognitive structures.

Significance. If the central claim is established, the paper makes a useful conceptual contribution by connecting cognitive image-schema theory with declarative spatial reasoning and LLM-based program synthesis. The delegation of inference to a sound ASP solver is a sensible design that avoids LLM unsoundness in the final step, and the release of code/data is a strength. However, the empirical support is currently too thin to validate the claim that LLMs can be guided to produce faithful schematic encodings: the evaluation uses one dataset and one puzzle, with no variance or reproducibility information, and the key semantic-faithfulness assumption of the LLM-to-spatial-predicate mapping is not independently verified. The significance is therefore conditional on a more rigorous audit of the generated programs and a more complete reporting of the experimental protocol.

major comments (3)

[§4.3 and §6] The load-bearing assumption is that GPT-4's mapping from natural-language expressions to spatial predicates (e.g., 'older' to left_pp/right_pp, 'owns' to in_pr) preserves truth conditions across the entire generated program. The retry protocol in §4.3 checks only syntax and satisfiability, not semantic fidelity relative to the original narrative. A generated program can be syntactically valid and satisfiable yet encode a different problem, making the solver's output meaningless. The paper's own error analysis in §6 confirms directional inconsistency ('newest' mapped leftward while later rules assume rightward). The discard rate after the three-attempt protocol is not reported, and no sample of generated programs is audited against the ground-truth order relations of the LogicalDeduction dataset. Without such an audit, the reported 91% cannot be interpreted as evidence that the schematic
[§5.1, Figure 6] The empirical support for the claim that the approach 'maintains competitive performance' is thin: a single point estimate of 91% on LogicalDeduction with no variance, no number of runs, no seed information, and no ablations. The comparisons to other systems are taken from a different paper and may not be controlled. The zebra result is one hand-picked instance. For a conference paper, this level of evaluation may be acceptable for a proof of concept, but the abstract and §6 make broader claims about 'effective logical reasoning' and 'performance comparable to state-of-the-art models.' Please report standard deviations across multiple runs, the number of discarded scenarios, and ideally an error breakdown by reasoning type.
[Appendices B.1, C.1, and §5.1] The hand-written prompt examples in Appendices B.1 and C.1 define the exact spatial encodings that the system then uses to solve the LogicalDeduction problems. Success on the dataset therefore partly reflects these manually provided conventions rather than the LLM's autonomous discovery or detection of image-schema structures. This does not invalidate the approach, but it weakens the claim that the system 'grounds understanding' in image schemas: the grounding is substantially provided by the prompt engineer, not learned or inferred. The discussion should acknowledge this split of credit more explicitly and, ideally, test sensitivity to alternative prompt conventions.

minor comments (4)

[Throughout] The manuscript contains several typographical issues: 'F rançois', 'inconsistentcontexts', and inconsistent spacing in 'Proceedings ofMachineLearningResearch'. Please proofread carefully.
[Figure 6] The right panel (witnesses for the zebra models) is not explained in enough detail: what do the rectangles and points represent, and how should the reader verify that the displayed configurations satisfy all constraints? A legend or caption would help.
[Related Work / Figure 6] The score for SymbCoT is listed in the table but no reference is given for SymbCoT. Please cite the original source.
[§4.2] The formal semantics via Here-and-There with constraints is described in words, but no formal definition of the denotation of spatial atoms is given. A precise definition (or a pointer to Cabalar et al.) would make the integration mechanism easier to verify.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical evaluation is external and solver-based; self-citations are contextual only.

full rationale

The paper's chain is: (1) define spatial relations in DSR/ASP with fixed semantics (Table 2 and Section 4.2); (2) prompt GPT-4 to map narratives to these predicates (Section 4.3 and Appendices A–C); (3) run Clingo/Z3 to compute cautious answers; (4) compare to the LogicalDeduction ground truth and to a zebra-puzzle answer. The 91% figure is an empirical accuracy against an external dataset, not a quantity derived from the prompt's own definitions. The hand-written example rules in Appendices B and C are part of the system design, openly disclosed, and target the same problem type; they do not constitute fitted parameters being renamed as predictions, and the paper does not claim zero-shot transfer beyond the demonstrated domain. The self-citations (Olivier and Bouraoui, 2025; Olivier, 2022) introduce the theoretical framework (image schemas as a foundation) but are not used to infer or justify the experimental result; the correctness of the solver's output depends on the external Clingo/Z3 machinery and on GPT-4's generated program, and the paper's own error analysis (Section 6) reports failures of directional consistency. No load-bearing step reduces by construction to an input, and no uniqueness theorem is imported from the authors' prior work. Hence no circularity is identified.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The system introduces no fitted numerical parameters. Its burden is conceptual: it leans on image-schema theory, on the adequacy of spatial relations for the chosen tasks, and on GPT-4's reliability. These are domain assumptions rather than hidden free parameters.

assumptions (4)

domain assumption Image schemas (PATH, CONTAINER) can be operationalized as spatial relations sufficient for the selected logical reasoning tasks.
Sections 3 and 5 assume that ordering maps to left/right and ownership or containment maps to in_pr without proof that this preserves all logical consequences.
domain assumption GPT-4 can produce syntactically valid and semantically faithful ASP programs from the provided prompts.
Section 4.3 relies on the LLM for program generation; three retries only partially mitigate failures.
standard math The spatial theory propagation between Clingo and Z3 correctly implements the declarative semantics of the extended programs.
Section 4.2 assumes the soundness of the custom propagator and the Here-and-There with constraints semantics.
domain assumption The natural-language statements in the benchmarks are fully captured by the chosen spatial predicates.
The LogicalDeduction evaluation in Section 5.1 assumes the PATH encoding captures all ordering constraints, and the zebra evaluation assumes CONTAINER encoding captures all ownership and house constraints.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Towards a Neurosymbolic Reasoning System Grounded in Schematic Representations." pith.science (2026). https://pith.science/paper/FX4V723V

@misc{pith2026250903644,
  author       = {Pith},
  title        = {Pith review of: Towards a Neurosymbolic Reasoning System Grounded in Schematic Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FX4V723V}},
  note         = {Machine review of arXiv:2509.03644}
}

read the original abstract

Despite significant progress in natural language understanding, Large Language Models (LLMs) remain error-prone when performing logical reasoning, often lacking the robust mental representations that enable human-like comprehension. We introduce a prototype neurosymbolic system, Embodied-LM, that grounds understanding and logical reasoning in schematic representations based on image schemas-recurring patterns derived from sensorimotor experience that structure human cognition. Our system operationalizes the spatial foundations of these cognitive structures using declarative spatial reasoning within Answer Set Programming. Through evaluation on logical deduction problems, we demonstrate that LLMs can be guided to interpret scenarios through embodied cognitive structures, that these structures can be formalized as executable programs, and that the resulting representations support effective logical reasoning with enhanced interpretability. While our current implementation focuses on spatial primitives, it establishes the computational foundation for incorporating more complex and dynamic representations.

Figures

Figures reproduced from arXiv: 2509.03644 by the authors.

**Figure 1.** Temporal relationships naturally map to spatial positions on a timeline, illustrating how humans use spatial schemas to understand abstract concepts. By placing these individuals along the timeline according to their relative ages, this spatial arrangement immediately reveals additional temporal relationships that weren’t explicitly stated—such as Alice being older than Charles. Shimojima termed such inference a "… view at source ↗

**Figure 2.** Spatial relationships are formalized in the DSR framework through coordinate constraints. Point a is inside rectangle b when its coordinates fall strictly within the rectangle’s boundaries on both dimensions [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Architecture of Embodied-LM. Once a valid ASP program representing the context of the scenario is generated, GPT-4 is called a second time (see Appendix C) to add the question and options to the program in accordance with the context formalization. Each option becomes an additional rule, called an answer rule, whose purpose is to succeed when the corresponding answer option is true. We instruct GPT-4 to use a predic… view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Representative problem from LogicalDeduction dataset (Srivastava et al., 2023) naturally invoking a schematic representation with the PATH primitive. When presented with such an example, GPT-4 positions the different entities from the context on a line representing the…

**Figure 5.** Figure 5: Instance of a zebra puzzle. System Score SymbCoT 93.00 Our Approach 91.00 VERUS-LM 88.67 Logic-LM 87.63 GPT4-CoT 75.25 GPT4 71.33 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

**Figure 6.** Figure 6: Left: Performance comparison on the LogicalDeduction dataset. Results for other systems from (Callewaert et al., 2025). Right: Witnesses of the two possible models for the Zebra problem. 6. Results and Discussion Our experimental results achieve the primary goal of dem…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 24 canonical work pages

[1]

CLP(QS): A declarative spatial reasoning framework

Mehul Bhatt, Jae Hee Lee, and Carl Schultz. CLP(QS): A declarative spatial reasoning framework. In Max J. Egenhofer, Nicholas A. Giudice, Reinhard Moratz, and Michael F. Worboys, editors, Spatial Information Theory - 10th International Conference, COSIT 2011, Belfast, ME, USA, September 12-16, 2011. Proceedings , volume 6899 of Lecture Notes in Computer S...

work page doi:10.1007/978-3-642-23196-4 2011
[2]

A uniform treatment of aggregates and constraints in hybrid asp

Pedro Cabalar, Jorge Fandinno, Torsten Schaub, and Philipp Wanko. A uniform treatment of aggregates and constraints in hybrid asp. In Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning, volume 17, pages 193--202, 2020

work page 2020
[3]

Asp-core-2 input language format

Francesco Calimeri, Wolfgang Faber, Martin Gebser, Giovambattista Ianni, Roland Kaminski, Thomas Krennwallner, Nicola Leone, Marco Maratea, Francesco Ricca, and Torsten Schaub. Asp-core-2 input language format. Theory and Practice of Logic Programming, 20 0 (2): 0 294--309, 2020

work page 2020
[4]

Verus-lm: a versatile framework for combining llms with symbolic reasoning, 2025

Benjamin Callewaert, Simon Vandevelde, and Joost Vennekens. Verus-lm: a versatile framework for combining llms with symbolic reasoning, 2025. URL https://arxiv.org/abs/2501.14540

arXiv 2025
[5]

Formal specification of image schemata--a step towards interoperability in geographic information systems

Andrew U Frank and Martin Raubal. Formal specification of image schemata--a step towards interoperability in geographic information systems. Spatial Cognition and Computation, 1: 0 67--101, 1999

work page 1999
[6]

Theory solving made easy with clingo 5

Martin Gebser, Roland Kaminski, Benjamin Kaufmann, Max Ostrowski, Torsten Schaub, and Philipp Wanko. Theory solving made easy with clingo 5. In Technical Communications of the 32nd International Conference on Logic Programming (ICLP 2016). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2016

work page 2016
[7]

Image schemas and concept invention: cognitive, logical, and linguistic investigations

Maria M Hedblom. Image schemas and concept invention: cognitive, logical, and linguistic investigations. Springer Nature, 2020

work page 2020
[8]

The diagrammatic image schema language (disl)

Maria M Hedblom, Fabian Neuhaus, and Till Mossakowski. The diagrammatic image schema language (disl). Spatial Cognition & Computation, pages 1--38, 2024

work page 2024

Show all 31 references

[9]

Leveraging large language models to generate answer set programs

Adam Ishay, Zhun Yang, and Joohyung Lee. Leveraging large language models to generate answer set programs. In Proceedings of the 20th International Conference on Principles of Knowledge Representation and Reasoning, pages 374--383, 2023

2023
[10]

The body in the mind: The bodily basis of reason and imagination

Mark Johnson. The body in the mind: The bodily basis of reason and imagination. Chicago: University of Chicago Press, 1987

1987
[11]

How to build your own asp-based system?! Theory and Practice of Logic Programming, 23 0 (1): 0 299--361, 2023

Roland Kaminski, Javier Romero, Torsten Schaub, and Philipp Wanko. How to build your own asp-based system?! Theory and Practice of Logic Programming, 23 0 (1): 0 299--361, 2023

2023
[12]

Logic-lm++: Multi-step refinement for symbolic formulations, 2024

Shashank Kirtania, Priyanshu Gupta, and Arjun Radhakirshna. Logic-lm++: Multi-step refinement for symbolic formulations, 2024. URL https://arxiv.org/abs/2407.02514

2024 arXiv
[13]

Metaphors We Live By

George Lakoff and Mark Johnson. Metaphors We Live By. University of Chicago Press, 1980

1980
[14]

Where mathematics comes from, volume 6

George Lakoff and Rafael N \'u \ n ez. Where mathematics comes from, volume 6. New York: Basic Books, 2000

2000
[15]

Clingo2dsr-a clingo-based software system for declarative spatial reasoning

Beidi Li and Carl Schultz. Clingo2dsr-a clingo-based software system for declarative spatial reasoning. Spatial Cognition & Computation, pages 1--51, 2024

2024
[16]

Ivanova, Idan A

Kyle Mahowald, Anna A. Ivanova, Idan A. Blank, Nancy Kanwisher, Joshua B. Tenenbaum, and Evelina Fedorenko. Dissociating language and thought in large language models. Trends in Cognitive Sciences, 28 0 (6): 0 517--540, 2024. ISSN 1364-6613. doi:https://doi.org/10.1016/j.tics....

2024 doi
[17]

On defining image schemas

Jean M Mandler and Crist \'o bal Pag \'a n C \'a novas. On defining image schemas. Language and Cognition, 6 0 (4): 0 510--532, 2014

2014
[18]

Embers of autoregression: Understanding large language models through the problem they are trained to solve

R Thomas McCoy, Shunyu Yao, Dan Friedman, Matthew Hardy, and Thomas L Griffiths. Embers of autoregression: Understanding large language models through the problem they are trained to solve. arXiv preprint arXiv:2309.13638, 2023

2023 arXiv
[19]

Spatial relations in reasoning : a computational model

Fran c ois Olivier. Spatial relations in reasoning : a computational model . Theses, Universit \'e Paris sciences et lettres , March 2022. URL https://theses.hal.science/tel-03984759

2022
[20]

Grounding agent reasoning in image schemas: A neurosymbolic approach to embodied cognition

Fran c ois Olivier and Zied Bouraoui. Grounding agent reasoning in image schemas: A neurosymbolic approach to embodied cognition. In Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2025), pages 2875--2879. IFAAMAS, 2025. URL ...

2025
[21]

Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning

Liangming Pan, Alon Albalak, Xinyi Wang, and William Wang. Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 3806--3824, 2023

2023
[22]

Computational geometry: an introduction

Franco P Preparata and Michael I Shamos. Computational geometry: an introduction. Springer Science & Business Media, 2012

2012
[23]

language is spatial

Daniel C Richardson, Michael J Spivey, Shimon Edelman, and Adam J Naples. " language is spatial": Experimental evidence for image schemas of concrete and abstract verbs. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 23, 2001

2001
[24]

Answer set programming modulo ‘space-time’

Carl Schultz, Mehul Bhatt, Jakob Suchan, and Przemys aw Andrzej Wa e ga. Answer set programming modulo ‘space-time’. In International Joint Conference on Rules and Reasoning, pages 318--326. Springer, 2018

2018
[25]

Semantic Properties of Diagrams and Their Cognitive Potentials

Atsushi Shimojima. Semantic Properties of Diagrams and Their Cognitive Potentials. CSLI Publications, Stanford, California, 2015

2015
[26]

Beyond the imitation game: Quantifying and extrapolating the capabilities of language models (arxiv: 2206.04615)

A Srivastava, A Rastogi, A Rao, AAM Shoeb, A Abid, A Fisch, AR Brown, A Santoro, A Gupta, A Garriga-Alonso, et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models (arxiv: 2206.04615). arxiv, 2023

2023 arXiv
[27]

Talking about the moving image: A declarative model for image schema based embodied perception grounding and language generation

Jakob Suchan, Mehul Bhatt, and Harshita Jhavar. Talking about the moving image: A declarative model for image schema based embodied perception grounding and language generation. CoRR, abs/1508.03276, 2015. URL http://arxiv.org/abs/1508.03276

2015 arXiv
[28]

L anguage (re)modelling: T owards embodied language understanding

Ronen Tamari, Chen Shani, Tom Hope, Miriam R L Petruck, Omri Abend, and Dafna Shahaf. L anguage (re)modelling: T owards embodied language understanding. In Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault, editors, Proceedings of the 58th Annual Meeting of the As...

2020 doi
[29]

Systematic analysis of image schemas in natural language through explainable multilingual neural language processing

Lennart Wachowiak and Dagmar Gromann. Systematic analysis of image schemas in natural language through explainable multilingual neural language processing. In Nicoletta Calzolari, Chu-Ren Huang, Hansaem Kim, James Pustejovsky, Leo Wanner, Key-Sun Choi, Pum-Mo Ryu, Hsin-Hsi Che...

2022
[30]

Non-monotonic spatial reasoning with answer set programming modulo theories

Przemysław Andrzej Wałęga, Carl Schultz, and Mehul Bhatt. Non-monotonic spatial reasoning with answer set programming modulo theories. Theory and Practice of Logic Programming, 17 0 (2): 0 205–225, 2017. doi:10.1017/S1471068416000193

2017 doi
[31]

Exploring spatial schema intuitions in large language and vision models

Philipp Wicke and Lennart Wachowiak. Exploring spatial schema intuitions in large language and vision models. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar, editors, Findings of the Association for Computational Linguistics: ACL 2024, pages 6102--6117, Bangkok, Thailand, Au...

2024 doi

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

CLP(QS): A declarative spatial reasoning framework

Mehul Bhatt, Jae Hee Lee, and Carl Schultz. CLP(QS): A declarative spatial reasoning framework. In Max J. Egenhofer, Nicholas A. Giudice, Reinhard Moratz, and Michael F. Worboys, editors, Spatial Information Theory - 10th International Conference, COSIT 2011, Belfast, ME, USA, September 12-16, 2011. Proceedings , volume 6899 of Lecture Notes in Computer S...

work page doi:10.1007/978-3-642-23196-4 2011

[2] [2]

A uniform treatment of aggregates and constraints in hybrid asp

Pedro Cabalar, Jorge Fandinno, Torsten Schaub, and Philipp Wanko. A uniform treatment of aggregates and constraints in hybrid asp. In Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning, volume 17, pages 193--202, 2020

work page 2020

[3] [3]

Asp-core-2 input language format

Francesco Calimeri, Wolfgang Faber, Martin Gebser, Giovambattista Ianni, Roland Kaminski, Thomas Krennwallner, Nicola Leone, Marco Maratea, Francesco Ricca, and Torsten Schaub. Asp-core-2 input language format. Theory and Practice of Logic Programming, 20 0 (2): 0 294--309, 2020

work page 2020

[4] [4]

Verus-lm: a versatile framework for combining llms with symbolic reasoning, 2025

Benjamin Callewaert, Simon Vandevelde, and Joost Vennekens. Verus-lm: a versatile framework for combining llms with symbolic reasoning, 2025. URL https://arxiv.org/abs/2501.14540

arXiv 2025

[5] [5]

Formal specification of image schemata--a step towards interoperability in geographic information systems

Andrew U Frank and Martin Raubal. Formal specification of image schemata--a step towards interoperability in geographic information systems. Spatial Cognition and Computation, 1: 0 67--101, 1999

work page 1999

[6] [6]

Theory solving made easy with clingo 5

Martin Gebser, Roland Kaminski, Benjamin Kaufmann, Max Ostrowski, Torsten Schaub, and Philipp Wanko. Theory solving made easy with clingo 5. In Technical Communications of the 32nd International Conference on Logic Programming (ICLP 2016). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2016

work page 2016

[7] [7]

Image schemas and concept invention: cognitive, logical, and linguistic investigations

Maria M Hedblom. Image schemas and concept invention: cognitive, logical, and linguistic investigations. Springer Nature, 2020

work page 2020

[8] [8]

The diagrammatic image schema language (disl)

Maria M Hedblom, Fabian Neuhaus, and Till Mossakowski. The diagrammatic image schema language (disl). Spatial Cognition & Computation, pages 1--38, 2024

work page 2024