Simple vs. Complex Ontology Alignment
Thing A: Simple Alignment
One-to-one mappings between classes (or properties) in two ontologies. “Class A in Ontology 1 equals Class B in Ontology 2.”
Thing B: Complex Alignment
Mapping rules that go beyond one-to-one correspondence. These capture structural transformations, conditional mappings, and relationships that require understanding how concepts compose.
Key Differences
| Dimension | Simple Alignment | Complex Alignment |
|---|---|---|
| Structure | Direct equivalence | Rules with conditions, transformations |
| Required understanding | Label/definition similarity | Structural relationships, composition |
| Automation success | LLMs achieve good results | Historically defied automation |
| Benchmark focus | Main focus of research community | Only recently addressed |
| Real-world need | Handles subset of data integration | Required for most practical scenarios |
Why This Matters
The research community has focused heavily on simple alignment, partially because it’s tractable, partially because benchmarks exist. But practical data integration almost always requires complex mappings. The gap between what’s benchmarked and what’s needed is significant.
Interestingly, LLMs have been applied to simple alignment with good results, but the community initially assumed complex alignment would remain intractable. The modular approach suggests otherwise, when given conceptually coherent modules rather than full ontologies, LLMs achieve surprisingly good complex alignment results.
When Each Applies
Simple alignment works when:
- Ontologies have clear 1:1 concept overlap
- Vocabularies differ but structures match
- Quick, automated matching is acceptable
Complex alignment is required when:
- Concepts in one ontology compose from multiple concepts in another
- Structural differences require transformation logic
- Real-world data integration scenarios
Related: 05-molecule—two-stage-modular-prompting, 05-atom—modularity-unlocks-llm-performance