Knowledge Representation and Reasoning

KRR - Unit 10: Reasoning with Protégé

Overview

As per the course website, "This unit focuses on the architecture of a knowledge-based system and how the various concepts discussed so far fits within. It also introduces further concepts to build on the knowledge of reasoning in knowledge-based systems."

My Reflection

Overall Reflection

This unit was the first after the return from the winter break. I had only to do a quick reading for a paper on modeling a description logic vocabulary for cancer research (Hartel et al., 2005), a quick formative activity building on Protégé pizza tutorial, and to provide peer responses for the collaborative discussion on ontology languages started in the previous unit.

Reading Reflection

Modeling a Description Logic Vocabulary for Cancer Research by (Hartel et al., 2005) offers a detailed explanation of how the National Cancer Institute developed a large description logic vocabulary to support cancer research. It presents the NCI Thesaurus as both a practical resource and a methodological example, showing how Ontylog’s features such as kinds, roles, and defined concepts enable precise modeling of diseases, anatomy, drugs, genes, and molecular abnormalities. The authors also describe how the vocabulary was converted into OWL Lite to improve interoperability, outlining the mapping choices and the limitations encountered during this process.

In addition to the technical modeling, the paper offers a structured approach for collaborating with domain experts who may not think in formal logic. It introduces tools such as use cases, pseudo T Box models, hierarchy reviews, and wireframe diagrams to help experts understand and refine the ontology. The authors highlight lessons about designing intuitive kinds, grounding modeling decisions in real user needs, and focusing on meaningful semantic relationships rather than the number of concepts. Through these contributions, the paper provides both a blueprint for building large biomedical ontologies and practical guidance for maintaining them in real research environments.

Below are a few quotes from the paper that I found insightful on ontology development in general:

As relations between concepts, roles are what drive the building of an ontology in determining the structure of the graph. A concept is modeled in the semantics of Ontylog as a set of things. It is the roles as relations between these sets of things that determine how the subsumption graph is created.
However given the costs of ontology development one cannot go on doing that very long. Since no ontology is ever complete, a yardstick is needed to help decide where in the namespace to create fine granularity concept hierarchies and richly detailed inter-concept semantics on expressed user needs.
[D]evelopers of applications that utilize ontologies should be encouraged to develop suites of regression tests to validate specific data that their applications utilize.

Reference

Hartel, F.W. et al. (2005) ‘Modeling a Description Logic Vocabulary for Cancer Research’, Journal of Biomedical Informatics, 38(2), pp. 114–129. Available at: https://doi.org/10.1016/j.jbi.2004.09.001.

Artefacts

Collaborative Discussion

Peer Response 1

The first peer that I responded to argued that effective knowledge representation on the Web requires ontology languages with clear, formal semantics that both humans and software agents can interpret reliably. While RDF provides a basic structural foundation, it lacks the expressive power needed for reasoning, and alternatives such as OWL‑Lite and KIF are limited either by reduced expressiveness or poor scalability. In contrast, OWL 2 offers a strong balance of formal rigour, modelling capability, and compatibility with web standards, enabling automated reasoning and semantic interoperability across diverse data sources. Ongoing improvements in OWL 2 tooling further strengthen its suitability, leading to the conclusion that it best meets the criteria for representing shared conceptual knowledge on the World Wide Web.

My Response

Thank you for your comprehensive analysis of OWL2's suitability for expressing ontologies on the WWW. I strongly agree with your conclusion that OWL2 represents the optimal choice among the traditional ontology languages presented, particularly given its formal semantics, standardisation, and reasoning capabilities. Your point about RDF's limited expressiveness and OWL-Lite's intentional restrictions effectively demonstrates why OWL2 strikes the necessary balance between computational tractability and semantic richness (Ranatunga et al., 2025).

However, I would like to extend the discussion beyond the four languages specified in the original question to consider contemporary developments in the agentic AI landscape. While OWL2 provides essential logical reasoning capabilities (what Magana and Monti (2025) describe as a "symbolic anchor" to mitigate LLM hallucinations) I argue that JSON-LD deserves equal consideration as a complementary ontology representation format for modern web-based agents.

Recent agent communication protocols, including Agent Network Protocol (ANP) and Agent to Agent (A2A), have standardised on JSON-LD for representing agent capabilities and identities through Agent Cards (Agent Network Protocol, 2024; Kakde et al., 2025). This adoption reflects JSON-LD's unique position as semantic "glue" that combines RDF's graph-based rigour with JSON's developer accessibility, enabling agent discovery and interoperability at web scale.

Your observation about OWL2's ongoing evolution through enhanced reasoning frameworks (Bilenchi, 2025) aligns with broader trends toward neuro-symbolic architectures, where formal ontologies constrain probabilistic AI systems. In this emerging paradigm, OWL2 and JSON-LD serve complementary roles: OWL2 provides logical guardrails and reasoning, while JSON-LD enables practical agent communication and discovery on the open internet.

Reference List

Agent Network Protocol (2024) Agent Description Protocol Specification. Available at: https://agentnetworkprotocol.com/en/specs/07-anp-agent-description-protocol-specification/ (Accessed: 23 December 2025).

Kakde, A.P. et al. (2025) 'Advancing Agentic AI through Communication Protocols', International Journal of Scientific Research in Science and Technology, 12(5), pp. 299–308.

Magana, I. and Monti, M. (2025) Enhancing Large Language Models through Neuro-Symbolic Integration and Ontological Reasoning. Available at: https://arxiv.org/abs/2504.07640v1 (Accessed: 23 December 2025).

Ranatunga, S. et al. (2025) 'Use of semantic web technologies to enhance the integration and interoperability of environmental geospatial data', International Journal of Geo-Information, 14(2), p.52.

Peer Response 2

The second peer I responded to argued that while RDF lacks the semantic precision needed for reliable reasoning and OWL 2 can be overly complex and computationally demanding, OWL Lite offers a practical middle ground for web‑based software agents. It provides enough structure to support shared conceptual understanding while remaining lightweight, efficient, and easier to implement in distributed environments. By balancing essential ontology features with manageable complexity, OWL Lite is presented as the most suitable choice for scalable applications that depend on formal, consistent, and interoperable knowledge representation on the World Wide Web.

My Response

Thank you for presenting a well-reasoned argument for OWL Lite based on computational efficiency and practical scalability. Your emphasis on balancing expressiveness with implementability raises an important consideration for web-based agent systems, particularly where resource constraints exist. I appreciate your point that OWL2's advanced constructs may introduce unnecessary complexity in certain applications (W3C, 2012).

However, I would respectfully challenge the premise that OWL Lite remains the optimal choice in the contemporary landscape of agentic AI. While OWL Lite's simplicity offered advantages historically, the language has been effectively deprecated since OWL 2's release, with W3C recommending migration to OWL 2 profiles (EL, QL, or RL) that provide equivalent or superior computational tractability while offering greater standardisation and tool support (Grau et al., 2008).

More fundamentally, the emergence of large language models and neuro-symbolic architectures has shifted the computational bottleneck away from ontological reasoning complexity. Modern agents require formal ontologies not primarily for their own reasoning efficiency, but as "symbolic anchors" to constrain probabilistic AI systems and prevent hallucinations (Magana and Monti, 2025). In this context, OWL2's richer expressiveness becomes essential rather than burdensome, as it provides the logical guardrails necessary to ensure deterministic performance and traceability in LLM-driven applications.

Furthermore, contemporary agent communication protocols have adopted JSON-LD rather than traditional OWL variants for practical interoperability. Agent Network Protocol, Agent to Agent, and Agent Communications Protocol all standardise on JSON-LD for Agent Cards, which enable agent discovery and collaboration through Decentralised Identifiers (Agent Network Protocol, 2024; Kakde et al., 2025). This suggests that practical scalability in modern multi-agent systems derives from lightweight semantic formats combined with formal OWL2 reasoning where needed, rather than from restricting ontological expressiveness.

Reference List

Grau, B.C. et al. (2008) 'OWL 2: The next step for OWL', Journal of Web Semantics, 6(4), pp. 309-322.

Kakde, A.P. et al. (2025) 'Advancing Agentic AI through Communication Protocols', International Journal of Scientific Research in Science and Technology, 12(5), pp. 299–308.

W3C (2012) OWL 2 Web Ontology Language Document Overview (Second Edition). Available at: https://www.w3.org/TR/owl2-overview/

Artefacts

Formative Activity - Pizza Tutorial Update

The formative activity of this unit was to create a NonVegetarian Pizza class in the Pizza Tutorial that I started in Unit 8. I created a defined class of Pizza similar to VegeterianPizza, but that is defined as having MeatTopping or SeafoodTopping instead. Here is the screenshot of the updated class hierarchy in Protégé:

Protégé Screenshot - NonVegetarian Pizza Class