The Engineering of Meanings

*The shape of Meanings* (*Painting of the slab stela of Wepemnefret by Norman de Garis Davies*).

[The book is available on Amazon and the O’Reilly digital platform.]

Ontologies & Languages

Ontologies are often siloed in two dimensions—data and knowledge—using semantic layers to align the meaning of words but forsaking the mapping of words with what they denote in worlds. Ontological Prisms address that oversight through the unfolding of knowledge epistemic dimensions. Benefits appear clearly for the alignment of meanings between thesauri and taxonomies.

Thesauri & Taxonomies

Enterprise ontologies are meant to organize what is known about business environments, objectives, and information systems. To integrate these perspectives into a shared knowledge architecture ontological prisms rely on three levels of indirection: thesauri for concepts, taxonomies for things, and domains for organization.

Within siloed domain ontologies, the meanings used by taxonomies can be unambiguously aligned to those established by thesauri. The same cannot be said if domain-specific vocabularies are involved.

For instance, thesauri may use skos:narrower connectors to detail cooking processes in terms of baking, boiling, and grilling; and ingredients in terms of meat, fish, milk, legumes, and grains. However, these connectors pertain to meanings and say nothing about the classification of actual instances.

Taxonomies, by contrast, are driven by applications targeting actual instances such as the classification of dishes according to dietary rules, whether culinary (e.g., vegetarian, vegan) or religious (e.g., kosher, halal).

Representation

With SKOS, the distinction between thesauri and taxonomies is represented through conceptual and classification schemes respectively — but this distinction is purely semantic and ignores the epistemic difference between words and the things they refer to. This is where combining OWL for ontological nodes with SKOS for associated names proves its worth.

In our example, four OWL’s SKOS annotation properties are introduced: label, concept scheme, classification scheme, and class. Labels attach to observations; concept schemes to sets of observations; classification schemes to partitioning aspects; and class to managed sets of observations.

*Combining OWL for nodes (above) with SKOS annotation property for words (below)*

SKOS annotations thus serve as a broker between thesauri and taxonomies, aligning the meaning of labels with their use labelling things.

The Fabric of Meanings

The term “semantic layer” maybe misleading as it suggests meanings organized in static flat semantic spaces. Introducing epistemic dimensions reveals a DNA-like threaded helix made of thesauri and taxonomies held by definitions.

Direct & Mediated Meanings

Names provide the building blocks of both direct and mediated communication, though to different ends. The former allows non-deterministic meanings, open to refinement through conversation; the latter requires deterministic ones, since settled representations leave no room for agreed-upon interpretations.

Thesauri and taxonomies give that difference concrete form — the former allowing for non-deterministic conversations, the latter intended to support deterministic operations.

Why it Matters

Meanings are the bread and butter of conversations — and to keep conversations alive, thesauri must make room for overlaps to be resolved by context. Taxonomies, by contrast, serve operational purposes and must ensure deterministic outcomes. Where outcomes are conversational — navigation, for instance — thesauri can still be used; otherwise, as in applications, non-overlapping and all-inclusive classifications are required.

Given sets of cooking processes and ingredients, subsets can be tentatively labelled as baking, boiling, and grilling; and meat, fish, dairy, legumes, and grains respectively. Such loose descriptions can be represented with SKOS narrower or broader connectors, and further extended or refined — adding frying and poaching, or shellfish and dairy. Consistent and comprehensive thesauri can thus be built, accurately covering the meanings associated with domain vocabularies. What works for thesauri, however, falls short when it comes to taxonomies.

Next, consider diet, first introduced as a concept scheme (thesaurus), with narrower meanings for culinary and religious diets. Based on those meanings, diet may also seem a candidate for a classification scheme (taxonomy). It is not, overlaps disqualify it. The same cannot be said of culinary and religious diets when considered separately. This is where thesaurus and taxonomy part ways: the former is concerned with meanings, the latter with function. In our example, thesauri are enough to find information about diets, culinary or religious; but not to decide which dishes are allowed and which are not. That can only be achieved with taxonomies. Because thesauri and taxonomies so often overlap in practice, the epistemic gap between words and realities tends to be ignored.

Conversational vs Operational Meanings

The threading of conversational meanings (thesauri) with operational ones (taxonomies) faces two constraints: it spans two temporalities, business and systems; and it must reconcile non-deterministic classifications with deterministic ones. While the first constraint rules out sequential processes, the second grants local primacy to thesauri. Elicitation processes should therefore combine local incremental steps between thesauri and taxonomies, set within broader iterations managed through domains, summarily:

Vocabularies associated with new facts and documents are compared to existing thesauri.
Candidate new terms are identified.
New terms are added to thesauri.
New terms are compared to operational schemas.
Taxonomies are updated.
Representations are updated.
Meanings and representations are consolidated across domains.

Thesauri, taxonomies, and domains correspond to semantic, operational, and organizational levels of indirection, respectively. They also provide the pivots of knowledge engineering processes.

The Life Cycle of Meanings

For enterprises operating in competitive environments, the primary objective of knowledge engineering is to align the continuous evolution of business meanings with phased changes in the representation of managed entities. That issue has been addressed by Greek philosophers a long time ago through epistemology which characterizes knowledge in terms of truths, beliefs, and reason. With ontological prisms, truths become observable facts, beliefs become concepts, and rationalities become categories.

Anchors & Iterations

Knowledge engineering at enterprise level must address three intertwined issues: semantic heterogeneity, overlapping organizational remits, and misaligned temporalities. Because these issues cannot be sorted out independently, scaling them generates exponential complexity. Domain-based ontologies can deal with conversational and operational semantics within bounded contexts, but they fall short at enterprise level, where misaligned time frames cut across domain-specific vocabularies, database schemas, and business entities. Ontological prisms manage this three-pronged heterogeneity through dedicated anchors for, respectively, names, aspects, and entities.

The meaning of names (~) in local vocabularies are consolidated through thesauri.
Aspects (≈) are derived from the classification schemes used by taxonomies and/or from schemas of managed representations.
Entities (#) are derived from concept schemes whose instances that can be independently identified in environments.

The meaning of names and the definition of aspects and entities are engineered through tiered iterations:

Inward iterations are anchored to thesauri: new names are introduced from scratch (labels) or in relation to existing names, aspects, or entities.
Middle iterations are anchored to taxonomies: new aspects are introduced with names or in relation to existing names, aspects, or entities.
Outward iterations are anchored to domains: new entities are introduced from scratch or in relation to existing names, aspects, or entities.

These iterations, managed through domains, are set within an homogeneous engineering temporality supposed to subsume epistemic temporalities of truths (facts), beliefs (concepts), and reason (categories).

Epistemic Temporalities

Knowledge, especially for enterprises operating in competitive environments, is continuously expanding and reshaping, and there is no reason to assume that changes across epistemic dimensions are set within a single time frame. Hence the need of epistemic temporalities governing observable events (e.g., patient’s state), deontic commitments (e.g., regulations), and managed representations (e.g., patients).

Synchronizations are achieved through event references, for instance between an attempted breach (environment temporality) must be recorded (managed temporality) as to comply with regulations (deontic temporality).