PaperlayerView exact revision on arXiv

Source-linked AI summary

Cognitive Schemas, Laws and Tasks

Antal Jakovác, András Telcs

arXiv:2610.09495v1cs.AI

TL;DR

The paper asks how explicit representations can support reusable cognitive schemas in knowledge-based problem solving. It develops a task–schema framework that derives schemas from representational structure and distinguishes ill-posed tasks from solver failures; it also formalizes when representational reduction preserves task-relevant solutions.

  • Problem

    The paper asks how explicit representations can support reusable cognitive schemas without introducing them as unrelated primitives.

    Source: §1 · p. 2

  • Method

    It organizes schemas by their representational prerequisites and connects them to tasks specified by context, unknowns, and constraints.

    Source: §1 · p. 2

  • Results

    The reduced task’s solutions correspond one-to-one to equivalence classes of original-task solutions under the stated conditions.

    Source: §5.2 · p. 20

  • Takeaways & Limitations

    The framework derives schemas from contexts, relations, representative objects, and admissible transformations, and distinguishes inconsistency, underdetermination, and contextual insufficiency.

    Source: §8 · p. 22

    Interaction between this structural layer and language-model architectures remains a research direction.

    Source: §8 · p. 22

Abstract

from arXiv · show

This paper asks how explicit representations can support reusable cognitive schemas in knowledge-based problem solving. We develop a structural framework in which schemas are organized by the information and relations required for their use, rather than introduced as unrelated primitives. The framework also distinguishes context-dependent relations from more stable structures that can be reused across different representations. Tasks are described through the information available, the unknowns to be determined, and the constraints that admissible solutions must satisfy. This makes it possible to separate limitations of the representation from limitations of the solving procedure. In particular, we distinguish inconsistency, underdetermination, and contextual insufficiency, where the current representation lacks distinctions or relations required by the external task meaning. We also show formally when a reduction of representation preserves the task-relevant solution structure. The resulting task--schema interface offers a structured way to describe representational conditions relevant to problem solving. It supports the reuse and stabilization of derived knowledge while remaining independent of the particular mechanism used to generate candidate solutions. This may provide a useful component for future solver architectures that combine structured knowledge, verification, and learned proposal mechanisms.

1 Introduction

The paper derives reusable cognitive schemas from the structures already available in a representational framework, then organizes them in a dependency hierarchy linked to task formulation. This interface distinguishes representational ill-posedness from failures of a solving procedure.

1 Introduction

  • Schemas such as equivalence, order, dependence, implication, and number-like structure are derived patterns of contexts, objects, and admissible transformations, not independent faculties or algorithmic modules.Their applicability follows from the relations among those structures.Source: §1 · p. 2
  • A dependency-structured hierarchy specifies the representational conditions for richer schemas, while a task specifies its context, given elements, unknowns, and constraints.Together these determine admissible instantiations and which schemas can legitimately be applied.Source: §1 · p. 2
  • Task formulation distinguishes inconsistency, underdetermination, and contextual insufficiency as representational failures that prevent a unique admissible solution.This separates limitations of the representation from limitations of the solving procedure.Source: §1 · p. 2
  • Contexts are task-relative structures, schemas specify conditions for structural applicability, and laws capture relations stable across relevant contexts.Task formulation determines which of these structures can participate in problem solving.Source: §1 · p. 2

2 Related Work

The paper situates its framework among work on structured knowledge, relational representation, abstraction, learning, and task specification. Its distinctive focus is organizing these structures around task-relative contexts that determine which schemas apply.

2 Related Work

  • Schema theory, AI frames and plans, and cognitive architectures organize knowledge and action; this paper isolates the representational conditions that make schemas applicable to tasks.Source: §2 · p. 3
  • Rough sets, Formal Concept Analysis, pattern structures, granular computing, and conceptual graphs formalize distinctions and concepts, but their structures play different roles from this framework’s task-relative contexts and Concept Graph.Source: §2 · p. 3
  • Analogy and abstraction study relational correspondence and structural transformation, while program induction and rule learning delimit candidate structures through available primitives and composition [24] [26].The framework likewise treats a schema’s applicability as dependent on task-provided distinctions and relations.Source: §2 · p. 3
  • Relational structure and invariance enter machine learning through representations, architectures, or objectives [28] [32]; this framework locates invariance recognition in task-relative contexts rather than model classes or loss functions.Source: §2 · pp. 3–4
  • Formal methods and program synthesis specify admissible solutions before search, while causal models specify variables and relations; this framework also treats representational context as part of task specification.This supports distinguishing task ill-posedness from search, learning, or inference failures.Source: §2 · p. 4
  • LLM research includes assessing implicit knowledge, enriching explicit knowledge structures, and translating questions into structured queries.Source: §2 · p. 4
  • Unlike knowledge graphs and meta-knowledge stores used in solver architectures, the Knowledge Space is intended to hold task-relative concepts, schemas, and stabilized laws reusable across contexts.The paper presents a representational layer for organizing interpretation, proposal, verification, and controlled knowledge extension.Source: §2 · p. 4
  • The framework’s distinctive feature is its organization around task-relative contexts that determine available distinctions, schema applicability, and which stable laws can support problem solving.Existing work addresses rich but distinct parts of this representational problem.Source: §2 · pp. 3–4

3 Cognitive Schemas and Structural Derivation

The section defines schemas as reusable structural mappings and specifies how their typed inputs, outputs, prerequisites, and local constraints govern application. It also describes how derived structures can be stabilized as conceptual objects and reused in further schema construction.

Schemas, Application Frames, and Applicability

  • Schemas are reusable structural mappings, with primitive schemas in an initial basis and derived schemas using structures made available by prior schemas or stabilized concepts.The dependency-compatible presentation introduces prerequisites before the schemas that use them, without requiring a single linear chain.Source: Schemas, Application Frames, and… · pp. 5–6
  • A schema specifies typed input and output placeholders, structural prerequisites, and a structural mapping; its mapping states what output structure becomes available when compatible inputs satisfy those prerequisites.The mapping is structural rather than algorithmic.Source: Schemas, Application Frames, and… · p. 5
  • An application frame selects the active context, available objects, output placeholders, and local constraints relevant to a prospective schema application.An admissible instantiation must satisfy type compatibility, structural prerequisites, and local constraints; applicability requires at least one such instantiation.Source: Schemas, Application Frames, and… · pp. 5–6
  • Direct dependency means one schema’s output or stabilized concept can satisfy an input or prerequisite of another, but dependency alone does not make the latter applicable.Remaining typed inputs, structural prerequisites, and local constraints must also be satisfied.Source: Schemas, Application Frames, and… · p. 6
  • A stable schema may be lifted into the Concept Graph as a conceptual object, distinct from the structural mapping that is instantiated and applied.The concept can be referred to, compared, generalized, specialized, or used as another schema’s input.Source: Schemas, Application Frames, and… · p. 6
  • Stabilized schemas can become inputs to further schema construction, allowing an evolving Knowledge Space while distinguishing schemas, their conceptual representations, context-relative instantiations, and executable realizations.The paper uses the Knowledge Space for persistent organization and reuse, not a particular graph-search, retrieval, or navigation mechanism.Source: Schemas, Application Frames, and… · p. 7

Equivalence

The section derives equivalence from context granularity, then defines order, successor, and structural dependence over represented structures. These relations remain tied to the context and constraints in which they are specified.

Equivalence

  • A context partition induces equivalence among elements in the same cell; refining or coarsening the context changes that relation, with singleton cells yielding identity.Source: Equivalence · p. 8

Order and Comparison

  • Order is a strict partial or total relation over context cells or representatives, enabling comparison, precedence, containment, or hierarchy without numerical structure.The order operates on equivalence classes rather than raw states.Source: Order and Comparison · p. 8

Successor

  • Successor refines an order by identifying minimal progression, and applies to ordered structures without assuming physical or metric time.Temporal precedence is one instance; process states, refinement chains, or representation stages can also provide the progression.Source: Successor · p. 8

Structural Dependence

  • Structural dependence holds when changing one component’s admissible realization changes another’s admissible possibilities under the frame’s constraints.It is frame-relative and directional, but may hold in both directions; combining it with successor structure and an explicit precedence requirement yields the restricted causal schema.Source: Structural Dependence · p. 9

General and Conditional Implication

Implication is separated from structural dependence, and the section develops derived succession and number-like schemas from prerequisite structures. These constructions distinguish restricted causal precedence from general causality and abstract number-like objects from the elements realizing finite successions.

General and Conditional Implication

  • General implication holds over the complete context domain, whereas conditional implication holds only on a specified subset.Both differ from structural dependence, which concerns variation under application-frame constraints rather than preservation of truth over a domain.Source: General and Conditional Implication · p. 10

Structural Causal Precedence

  • Structural causal precedence combines directed structural dependence with successor order, defining one restricted causal relation rather than causality in general.The dependent pattern must precede the pattern that depends on it along the available successor structure.Source: Structural Causal Precedence · pp. 10–11

Discrete Succession

  • Discrete succession consists of finite context-cell sequences generated by repeated successor steps, without assuming numerical magnitude.Longer successions arise through repeated admissible successor extension.Source: Discrete Succession · p. 11

Number-Like Structure

  • Order supplies successor, which enables finite chains whose equivalence classes yield number-like objects based on succession structure rather than element identity.This illustrates abstraction and possible reuse of a stabilized schema, not an axiomatization of the natural numbers.Source: Number-Like Structure · p. 12

Higher-Order Structural Correspondence

Higher-order structural correspondence uses previously available relations or schema concepts to identify common organization across configurations. It preserves relational structure among typed objects without requiring the objects or relations to be identical.

Higher-Order Structural Correspondence

  • Higher-order schemas can use previously available relations or schema concepts to identify common organization across configurations, operating on relational structures formed from objects.The construction may combine several previously available structures.Source: Higher-Order Structural Correspondence · pp. 12–13
  • A structural correspondence maps typed objects and, when participating relations differ, relates those relations while satisfying the application frame's type and structural prerequisites.Preservation is determined by relational structure, not object names or displayed order.Source: Higher-Order Structural Correspondence · p. 13
  • The correspondence preserves organization among typed objects and relations even when the two configurations do not contain identical objects or relations.Source: Higher-Order Structural Correspondence · p. 13
  • Compositional derivation uses one schema's output as another's input or prerequisite, whereas higher-order construction may combine structures without a single predecessor.Table 1 summarizes dependencies rather than a single linear construction sequence.Source: Higher-Order Structural Correspondence · pp. 13–14, Table 1 · p. 14

4 Relations and Laws

Relations are evaluated relative to contexts, while laws are stable structures represented by compatible families of context-relative realizations. The framework distinguishes refinement by pullback from coarsening by descent.

Relations and Laws

  • A context-relative relation is evaluated against the distinctions represented by its context, so granularity and organization can affect it and it is not by itself a law.Source: §4 · p. 14
  • Admissible context transformations preserve indistinguishability: objects indistinguishable in the source context map to objects indistinguishable in the target.Source: §4 · p. 15
  • Refinement supports canonical pullback of a coarser relation, while coarsening requires descent, possible only when the finer relation is constant on projection fibers.When descent holds, the coarser relation is unique.Source: §4 · p. 16
  • A law is a compatible family of context-relative realizations, not any single relation, and the context family specifies its domain of applicability.Source: §4 · pp. 16–17
  • Compatibility is stable under composition, so a law's realization is unchanged whether a compatible context change occurs directly or through intermediate admissible contexts.A stable law may be lifted into the Concept Graph, where its context-relative relations remain realizations of that law.Source: §4 · p. 17

5 Tasks, Contexts, and Schema Applicability

Tasks are represented relative to contexts, with givens, unknowns, and constraints defining admissible instantiations and available schemas. Examples show how representation changes can suppress irrelevant distinctions, while reduction preserves task-relevant solutions under stated conditions.

5 Tasks, Contexts, and Schema Applicability

  • A task is specified relative to a context, with givens, unknowns, and constraints determining admissible instantiations and which schemas may apply.The external presentation is assumed to have been converted into this task representation.Source: §5 · p. 17

5.1 Illustrative Examples

  • The examples illustrate applying an appropriate transformation once relevant invariances or structural requirements are available; they do not address how that transformation is discovered.The paper grounds general claims about schema applicability and cross-context stability in its formal constructions, not these examples.Source: §5.1 · p. 17

Example 1: Permutation-Invariant Structure

  • Quotienting sequences by permutation makes invariant outputs depend only on the permutation class, here whether a symbol repeats.Permutation-equivalent sequences are grouped into one context cell, suppressing order distinctions irrelevant to the task.Source: Example 1: Permutation-Invariant… · p. 18

Example 2: Abstraction and Generalization by Removing Irrelevant Coordinates

  • After relevant coordinates are identified, projection removes distinctions on the rest, yielding simpler laws that support generalization across related task instances; finding coordinates is outside the example.The task states that a relevant subset exists but does not specify its members.Source: Example 2: Abstraction and… · p. 19

Example 3: Relational Composition under Contextual Constraints

  • With adjacency left without transitivity or symmetry assumptions, the original query is underdetermined; asking about paths instead introduces a new relation and reformulates the task.The context treats adjacency as local, so composition does not by itself resolve the original adjacency question.Source: Example 3: Relational Composition… · pp. 19–20

5.2 Solution preservation under representational reduction.

  • When constraints are invariant under an admissible reduction, quotient-task solutions correspond one-to-one with original solution classes, so differences only in suppressed distinctions can cease to count as alternatives.The reduction preserves task-relevant solution structure under the stated invariance condition.Source: §5.2 · p. 20

6 Ill-Posedness of Tasks

The section distinguishes task inconsistency, underdetermination, and contextual insufficiency, which differ in how the task representation fails. It also describes stable concepts and laws as reusable across contexts while context-specific realizations remain local.

6 Ill-Posedness of Tasks

  • Ill-posedness distinguishes internal inconsistency and underdetermination, properties of the formal task, from contextual insufficiency, which depends on adequacy relative to external task meaning.Contextual insufficiency cannot be diagnosed from the formal task tuple alone.Source: §6 · pp. 20–21
  • A task is inconsistent when no admissible instantiation satisfies its constraints; contradictory constraints, incompatible givens, or schema-context mismatch can produce this case.Source: §6 · p. 21
  • A task is underdetermined when multiple mutually incompatible assignments satisfy its constraints; additional constraints, data, context specificity, or an evaluation criterion may resolve it.Source: §6 · p. 21
  • A task is contextually insufficient when its context cannot express distinctions, relations, schemas, or laws required by the external task meaning.Examples include needing distance with only adjacency represented, or temporal order without successor structure.Source: §6 · p. 21

Objects Lifted into the Concept Graph

  • Stable concepts, schema concepts, laws, and recurring diagnostics may be lifted into the Concept Graph, while context-specific realizations stay local and procedures reside separately in the Procedure Graph.Compatible context families can provide reusable domains of applicability for laws and schemas.Source: Objects Lifted into the Concept Graph · pp. 21–22

7 Discussion

The discussion connects task structure and schema applicability to the distinctions available in a context, separating representational difficulty from procedural difficulty. It presents cross-context stability as the basis for reusable laws and a structural layer that can be analyzed independently of discovery or selection mechanisms.

7 Discussion

  • Inconsistency means no admissible instantiation satisfies constraints, underdetermination leaves several nonequivalent instantiations, and contextual insufficiency lacks distinctions or relations required by the intended solution.The examples show quotienting removing irrelevant order, projection suppressing irrelevant coordinates, and insufficient relational structure leaving a query unresolved.Source: §7 · p. 22
  • A relation may depend on a particular context, whereas a law has compatible realizations across contexts under admissible transformations and can be reused as a Concept Graph object.Source: §7 · p. 22
  • The structural layer analyzes quotienting, projection, context transformation, and schema applicability independently of the mechanism that discovers or selects them.This separation provides a basis for architectures with distinct proposal-generation, verification, and persistent-knowledge components.Source: §7 · p. 22

8 Conclusion

The conclusion presents schemas as structures derived from contexts, relations, representative objects, and admissible transformations, and links them to task formulation and ill-posedness. It also identifies future directions for combining explicit structural knowledge with language-model architectures.

8 Conclusion

  • Equivalence, order, successor, dependence, implication, discrete succession, and number-like structure are derived from representational structure rather than introduced as independent primitives.Source: §8 · p. 22
  • Task specification connects schemas to problem formulation by determining available distinctions and relations for instantiating unknowns under constraints.The conclusion distinguishes inconsistency, underdetermination, and contextual insufficiency, where representation must change before the solution can be expressed.Source: §8 · p. 22
  • Relations may be context-specific, while laws have compatible realizations across contexts under admissible transformations and can be reused in the Concept Graph.Source: §8 · p. 22
  • Future work could pair explicit Knowledge Space structure with an initially minimally trained model or with a pretrained LLM proposing candidates while verification and knowledge stabilization remain external.Both directions aim to separate proposal generation from verified persistent knowledge and controlled structural learning.Source: §8 · pp. 22–23

9 Declarations

The declarations state that data and material availability is not applicable, the authors report no competing interests or funding, and all authors contributed equally.

9.1 Availability of Data and Material

  • The authors state that data and material availability is not applicable.Source: §9.1 · p. 23

9.2 Competing Interests

  • The authors declare that they have no competing interests.Source: §9.2 · p. 23

9.3 Funding

9.4 Authors’ Contributions

  • The authors contributed equally to all aspects of the work.Source: §9.4 · p. 23
Loading 2610.09495v1…