Ontology files can use a format update

[kwalcock]

Justification and suggestions will follow in not too long.

[kwalcock]

The need to update the format was discussed a couple of months ago. Some of the justifications were

• An ontology editor was being added to Causemos and the present format does/did not work well for that task (IIRC).
• We need/want to support metatdata on non-leaf nodes. Eidos has a branch in which this is implemented, but in a way that changes the ontology format in the smallest possible way, which isn't very clean.
• The present version is somewhat difficult to work with in that the keys are not fixed. Some of them are the actual names of nodes and don't come from a small, fixed list. They might rather be the values. In this example, anything in front of a colon should come from a fixed set:

  - livestock_production_inputs:
  - OntologyNode:
     examples:
     - livestock feed
     name: livestock_feed

• The minimal indentation makes working with long files a little difficult, like when browsing the file on github.

The format was discussed at UAz and our top choice looked like this:

- node:
    name: wm
    examples:
        - crime
    definition:
    polarity: -1
    semantic type: event
    children: 
        - node:
            name: concept
            examples:
                - crime
            definition:

• Nodes are all called node and names have been moved to the name field.
• Children are explicitly listed under the children key.
• Indentation has been increased to four spaces where before it was two and sometimes even zero.
• Non-leaf nodes look just like the others except that they have no children.

[kwalcock]

At a recent meeting we discussed these points which are related to the ontology format:

• We could add examples/definitions for specific compositions
  • Some compositions have more meaning than the individual parts
  • Currently no way to structurally add to the ontology - may need separate place to put these
  • Would also need support at the level of reading/grounding

The format might account for adding information to specific compositions.

[bgyori]

I think the new proposed format should work well!

[johnhungerford]

The proposed format looks good to me. Is there any reason why nodes are structured in a list rather than a dict? Is there an imaginable case of multiple nodes in the same branch having the same name? Why not:

wm:
    examples:
        - crime
    definition:
    polarity: -1
    semantic type: event
    children: 
        concept:
            examples:
                - crime
            definition:

Just curious -- lists are fine with me.

[kwalcock]

It was my (false?) impression that the open vocabulary keys was a complication. wm has examples, definition, polarity, semantic type, and children that can be known and listed in advance. Children has a bunch of things whose names need to be collected in a very different way. It's probably not a big deal but it seems like it would complicate the code (and I'll be checking on that shortly). There are probably not naming conflicts among siblings, but I think there are cases in which the name of some great aunt is reused in some grand niece and if nodes get moved around there could be a problem. However, using the key "name" but having the values conflict would probably be just as bad. If that check is not already in the unit tests, I should add it.

[johnhungerford]

Yeah I think with the way we parse the yml file it would be simpler to deal with lists than dicts.

[kwalcock]

FWIW it looks like the definition field should be called descriptions.

There are old ontologies that use descriptions and new ones that use definition. Eidos doesn't know about definition and ignores it. Is it there just for human consumption?

[kwalcock]

For that matter, it's probably a good time to change pattern to patterns.

It turns out that in the compositional ontology, there is only pattern. In the flat ontology, there are both pattern and patterns. The plan is to change them all to the plural. I may submit an intermediate PR for the flat to change everything to the singular first,

[kwalcock]

I'm going to have to find out about this take_args that I see in the flat ontology. Can anyone tell me about it?

[kwalcock]

We have not yet used it in production, but at one time there was also a request for negative examples and there is an old prototype for it. If we do need these, I suggestion adding counterexamples and counterpatterns.