Preserving Inconsistency and Perspective in Knowledge Bases

17 concepts11 papers

The Myth of a Single Ground Truth

Traditional knowledge bases and semantic web architectures operate under a foundational assumption: there is a single, objective reality to be mapped. Automated knowledge harvesting frameworks excel at extracting undisputed, crisp facts—such as "Paris is the capital of France"—and organizing them into globally consistent graphs.

However, as noted by Gerhard Weikum et al. in Ten Years of Knowledge Harvesting: Lessons and Challenges, this "centralized database" mindset breaks down when applied to subjective, culturally dependent, or historically fluid information. When a knowledge graph forces a single version of the truth, it flattens the inherent nuance of human knowledge. Luna Dong et al. argue in Machine Knowledge: Creation and Curation of Comprehensive Knowledge Bases that capturing varying perspectives remains a major open challenge because standard frameworks view conflicting claims as logical errors to be filtered out or resolved.

In domains like humanistic inquiry, cultural heritage, and social sciences, disagreement is not noise—it is the data itself. As highlighted in the Proceedings of the Second Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature, forcing a consensus eliminates the context of scholarly debates. To capture human knowledge accurately, computational systems require a paradigm shift: treating contradictions not as system failures, but as vital structural features.

Reification: Turning Interpretations into First-Class Objects

If a database contains the assertions isCapital(City, Country) and two different sources name two different cities, standard logical inference engines flag a contradiction. To prevent this collapse without erasing the disagreement, systems use a mechanism called reification—turning the act of interpretation or perspective into a first-class entity within the graph.

Instead of drawing a direct edge between a subject and an object, the system introduces an intermediate "interpretation" or "narrative" node.

[Source Material] ─── (interpreted by) ───► [Interpretation Node] ◄─── (asserted by) ─── [Context/Scholar]
                                                    │
                                             (claims value)
                                                    ▼
                                             [Moral Value/Fact]

In From Digital to Computational Humanities: The VAST Project Vision, Alfio Ferrara et al. leverage this approach to track how abstract moral values like freedom and democracy change over time. By treating interpretations as independent data objects, the system can store multiple conflicting scholarly viewpoints side-by-side without creating logical contradictions.

As the authors further demonstrate in A Computational History Approach to Interpretation and Analysis of Moral European Values: the VAST Research Project, this structure explicitly maps the relationship between a source text, a specific value, and the historical context or observer providing the analysis. This polyvocal data modeling, as described by Pasquale Lisena et al. in Semantic Methods for Events and Stories (SEMMES) 2024, allows a knowledge graph to move beyond static, objective facts to accommodate diverse narrative voices and subjective perspectives simultaneously.

Formalizing Disagreement: Lattices and Standpoints

Storing conflicting claims is only half the battle; systems must also be able to reason over them mathematically. If standard binary logic encounters a statement that is both true and false, the principle of explosion causes the entire reasoning engine to fail. To make diversity tractable, researchers have designed multi-valued and context-indexed logical frameworks.

An early foundational approach to this problem was developed by Steve Easterbrook in An Algebraic Framework for Merging Incomplete and Inconsistent Views. Instead of forcing a binary true/false reduction during model integration, Easterbrook introduced a multi-valued knowledge-ordering lattice. This lattice explicitly maps states of agreement, repudiation, and active dispute:

          Dispute (Inconsistent)
          /     \
  Affirmation   Repudiation
          \     /
          Incomplete (Unknown)

By merging conflicting views into this lattice, inconsistencies are preserved as structural features of the integrated model rather than being prematurely scrubbed.

More recently, Lucía Gómez Álvarez et al. formalized this intuition for modern ontologies in Tractable Diversity: Scalable Multiperspective Ontology Management via Standpoint EL. They introduced Standpoint EL, a description logic that allows a unified formal framework to contain conflicting viewpoints by indexing axioms to specific "standpoints." This allows the system to verify what is true according to perspective X versus according to perspective Y, maintaining efficient, polynomial-time reasoning across massive ontologies without letting localized contradictions pollute the global system.

Why Embracing Inconsistency Scales Better

Intuitively, it might seem that enforcing strict global consistency makes a system cleaner and easier to manage. However, at the scale of the global web or massive crowdsourced applications, enforcing a single truth creates an artificial bottleneck.

In Mind the Web, Valentin Zacharias et al. argue that web-scale systems must abandon the centralized database mindset. Preserving contradictions across decentralized data sources allows for the identification of diverse perspectives and user groups, providing far richer contextual information than a filtered, homogenized dataset.

This preservation strategy is highly effective in collaborative and educational environments. For example, in Construction and Application of Teaching System Based on Crowdsourcing Knowledge Graph, Jinta Weng et al. outline an educational graph that deliberately retains conflicting relationships when user consensus falls below a specific threshold. By doing so, the graph accurately reflects the subjective nature of human cognition and interdisciplinary domains, preserving the exact nuances required for complex learning.

As summarized by Gerhard Weikum et al. in Knowledge Harvesting: Achievements and Challenges, moving past the assumption of objective truth allows knowledge bases to transition from rigid fact-repositories into comprehensive models of human culture, history, and thought.

Key Foundational Work

Mind the Web (2007) — contributed 2 concept elements to this direction.
An Algebraic Framework for Merging Incomplete and Inconsistent Views (2005) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_multi_perspective_knowledge_representati_8258dae3,
  title = {multi-perspective knowledge representation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-perspective-knowledge-representation/rdr_8258dae3002049e4a57b4fb746fb96c0},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_multi_perspective_knowledge_representati_8258dae3_v1,
  title = {multi-perspective knowledge representation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-perspective-knowledge-representation/rdr_8258dae3002049e4a57b4fb746fb96c0/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}

See something wrong? Report a mistake