We explored the applicability of the rules-as-you-go approach to transforming AGROVOC into an ontology. We looked for examples of patterns that could be used. The results are shown in Table 6.
Table 6: Examples for the rules-as-you-go approach |
Pattern: plant <growsIn> soil type Rice RT moist soil ® rice <growsIn> moist soil Pattern: geographical entity <spatiallyIncludedIn> geographical entity Benin BT West Africa ® Benin <spatiallyIncludedIn> West Africa Pattern: geographical entity <isa> geographical entity Benin RT Francophone country ® Benin <isa> Francophone country Pattern: body part <containsSubstance> (substance | small particle) blood NT blood gas ® blood < containsSubstance > blood gas blood NT blood cell ® blood < containsSubstance > blood cell |
From this exploration it appears that the approach is promising. On the other hand, the rules-as-you-go approach is not error-free; results must be checked by an ontology editor. In some cases, it may not be possible to define a rule.
Another difficulty is illustrated by the concept of Francophone Africa. Does this term refer to the set of Francophone African countries, in which case it refers to a type of geographical entity, or does it refer to the area (not necessarily contiguous) covered by these countries, in which case it refers to a geographical entity (just as West Africa).
UF (Used For) may refer to a synonymous or quasi-synonymous term or to a narrower concept (generally includesSpecific). Since the terms on both sides of the UF most often refer to concepts that have the same entity type, it is difficult to formulate a rule. One might use an external knowledge source, such as a large dictionary, to detect synonymity and treat the other cases as includesSpecific, always subject to editor verification.
The rules-as-you-go approach implies that the reengineering effort should start with the top-most concepts so that entity types and patterns can be detected early.