3 ms·

Ask HN: Why are all ontologies bad?

3 points by julienreszka 6y ago

w_t_payne 6y ago
I'm not sure that they are, but I think there is a cost/benefit problem that needs to be solved.
JPLeRouzic 6y ago
There are probably a myriad of reasons, but I can think of a few: * A good quality ontology costs a lot of money, it cannot be done automatically. That is the difference for example between the KEGG PATHWAY database and the GO ontology. The KEGG PATHWAY database is a collection of manually drawn KEGG pathway maps representing experimental knowledge on metabolism and various other functions of the cell and the organism while the GO ontology is the product of computer scripts. When students have no interesting results to show in an experiment, they do a "GO ontology enrichment" [0] and there is often funny stuff that shows up. * There is a contradiction in having a large ontology to reflect some knowledge field and having to find relevant information in this large dataset. Many relations are weak signals, which one is relevant? * An ontology reflects the limited knowledge and bias of its designers. * You cannot express complicated relations because ontologies must have acyclic relations and even multiple inheritance is usually not possible, so they end using hierarchical relations which cannot be very useful. In addition there is no high level logic relations such as something being a function of one or several other items (high order logic). [1] [0] https://en.wikipedia.org/wiki/Gene_Ontology_Term_Enrichment https://en.wikipedia.org/wiki/Gene_Ontology_Term_Enrichment [1] https://en.wikipedia.org/wiki/Web_Ontology_Language#OWL_Full https://en.wikipedia.org/wiki/Web_Ontology_Language#OWL_Full
- compressedgas 6y ago
  That last point is not true without qualification.
  - JPLeRouzic 6y ago
    I am sorry to not understand, but what do you mean?
    - compressedgas 6y ago
      The statement is not a fact about ontologies but rather about specific existent ontology languages and the trade-offs they made in their data model and logics to remain tractable. When an ontology is represented in a plain relational database without any fancy logic. All relations with finite extent are representable.
      - JPLeRouzic 6y ago
        >When an ontology is represented in a plain relational database without any fancy logic. All relations with finite extent are representable. Ah OK, I agree.
throwawaybutwhy 6y ago
The question is ill-posed. 'Have you stopped beating your wife?' It seems you are better off asking whether all ontologies are bad first. And no, I don't think ontologies are intrinsically bad. There's still hope NLP can make their creation and maintenance cost-efficient.

3 points by julienreszka 6y ago

I'm not sure that they are, but I think there is a cost/benefit problem that needs to be solved.

There are probably a myriad of reasons, but I can think of a few: * A good quality ontology costs a lot of money, it cannot be done automatically. That is the difference for example between the KEGG PATHWAY database and the GO ontology. The KEGG PATHWAY database is a collection of manually drawn KEGG pathway maps representing experimental knowledge on metabolism and various other functions of the cell and the organism while the GO ontology is the product of computer scripts. When students have no interesting results to show in an experiment, they do a "GO ontology enrichment" [0] and there is often funny stuff that shows up. * There is a contradiction in having a large ontology to reflect some knowledge field and having to find relevant information in this large dataset. Many relations are weak signals, which one is relevant? * An ontology reflects the limited knowledge and bias of its designers. * You cannot express complicated relations because ontologies must have acyclic relations and even multiple inheritance is usually not possible, so they end using hierarchical relations which cannot be very useful. In addition there is no high level logic relations such as something being a function of one or several other items (high order logic). [1] [0] https://en.wikipedia.org/wiki/Gene_Ontology_Term_Enrichment https://en.wikipedia.org/wiki/Gene_Ontology_Term_Enrichment [1] https://en.wikipedia.org/wiki/Web_Ontology_Language#OWL_Full https://en.wikipedia.org/wiki/Web_Ontology_Language#OWL_Full

compressedgas 6y ago
That last point is not true without qualification.
- JPLeRouzic 6y ago
  I am sorry to not understand, but what do you mean?
  - compressedgas 6y ago
    The statement is not a fact about ontologies but rather about specific existent ontology languages and the trade-offs they made in their data model and logics to remain tractable. When an ontology is represented in a plain relational database without any fancy logic. All relations with finite extent are representable.
    - JPLeRouzic 6y ago
      >When an ontology is represented in a plain relational database without any fancy logic. All relations with finite extent are representable. Ah OK, I agree.

throwawaybutwhy 6y ago

The question is ill-posed. 'Have you stopped beating your wife?' It seems you are better off asking whether all ontologies are bad first. And no, I don't think ontologies are intrinsically bad. There's still hope NLP can make their creation and maintenance cost-efficient.