Practice-Based Ontologies: Bridging the Gap Between "Knowledge Soup" and Machine Reasoning

TL;DR

In the realms of history and archaeology, data production is moving faster than our ability to process it. However, the solution isn't just more computation—it's better representation. This paper argues that current ontologies are too rigid for the messy, interpretive nature of historical work. The authors propose "Practice-Based Ontologies," built through ethnographic study, to capture the fluid way experts actually reason, moving beyond simple data retrieval to support deep narrative storytelling.

Context: The "Knowledge Soup" Problem

Most computer science approaches to Knowledge Representation (KR) thrive in domains with explicit terminology and well-formed definitions. Archaeology is the opposite. It is what John Sowa calls "Knowledge Soup": a mixture of overgeneralizations, incomplete definitions, and conflicting data points.

The authors argue that history is in a "proto-scientific" state. Unlike the hard sciences, historical investigation is more like detective work—assembling clues to reconstruct a possible past. Existing standards like CIDOC CRM help organize museum data, but they don't help a researcher think or generate hypotheses.

Motivation: Why Current Ontologies Fail

The tension in archaeology lies between Codification and Interpretation.

• The Conflict: When an archaeologist is forced to use a digital form with fixed drop-down menus, they often find it "confining." Important nuances—the "stains on a bone" or the "feeling" of an object's manufacture—get lost because there isn't a box for them.
• The Insight: The authors realized that the practice of archaeology is a hermeneutic cycle: a back-and-forth between the data and the interpretation. Any system that doesn't allow for this evolution is fundamentally broken for the field.

Methodology: The Ethnographic Shift

To fix this, the authors didn't start with code; they started with people. They embedded themselves with a diverse team at the University of Pennsylvania, including zooarchaeologists, epigraphers, and philologists.

1. The Phenomenological Approach

Instead of viewing an archaeological object as a mere physical thing, they used Ingarden’s phenomenology, viewing it as a "mind-dependent" cultural object. They sought to understand the Verstehen (deep understanding) that an expert brings to a shard of pottery.

2. Thick Description

Using "Think-Aloud" protocols (asking researchers to narrate their thoughts as they work) and reviewing field notebooks (the "dirty" data before it gets "cleaned" for reports), the authors captured the "Black Box" of archaeological reasoning.

(Note: This diagram would represent the flow from ethnographic observation to the extraction of design requirements for dynamic ontologies.)

Key Findings: From Organization to Understanding

The pilot study yielded several "Provocative Moments":

• Heterogeneous Perspectives: Archaeology is a "collaborative enterprise." A single object (like a seal impression) is viewed differently by a philologist than by a stratigrapher.
• Categorization for Understanding: The paper makes a brilliant distinction. We currently build ontologies for organization (finding the file). We should be building them for understanding (helping the expert see patterns).
• Storytelling as Knowledge: The ultimate output of archaeology is a story. Success should be measured by how well a system helps "plug the evidence into the story."

(Note: This table would contrast the rigid, top-down structure of traditional KR with the bottom-up, inductive nature of Practice-Based Ontologies.)

Critical Analysis & Future Outlook

The core contribution of this work is the shift from Product to Practice. It acknowledges that archeological knowledge is non-monotonic—meaning new data can disprove old facts at any time.

Limitations: The "Thick Description" required to build these ontologies is incredibly labor-intensive. Translating these qualitative insights into actual software architectures (like specific RDF schemas or logic engines) remains the "grand challenge" for the next phase of their research.

The Takeaway for Developers: If you are building AI or KR systems for the humanities, stop trying to impose a "Universal Truth." Instead, build "Sandbox Ontologies" that allow experts to categorize, re-categorize, and link data in ways that reflect their evolving mental models. The future of the Semantic Web in history isn't just about linked data—it's about linked interpretations.