Why don't functional terms have directed'ness / valence in SUMO?

[twitchyliquid64]

(Forgive me if this is justified in some of the literature, I am learning and have a lot of trouble knowing the right words to search)

When I think about the vast majority of relationships I would want to represent in an ontology, they seem to have a kind of direction/valence. For instance:

Head -[partOf]> Horse
Terra -[within]> MilkyWay

But in SUMO and everything else, the arguments of some relationship have meaning only in the context of that type of relation/function. Everything (i.e. prolog) seems to follow this pattern as well.

Why is this? I imagine theres solid principles as to why we've landed on this, I just don't know what to search.

[apease]

English has a natural order of subject-verb-object (with many exceptions). Other human languages have a different order see https://en.wikipedia.org/wiki/Object%E2%80%93subject%E2%80%93verb_word_order . But logic doesn't have an implicit linguistic order of arguments. I could have the relation 'part' and the first argument could be the part and the second the whole, or vice versa, it just depends on how I state the formulas that define the relation.

[twitchyliquid64]

I understand thats how we write it is a linguistic property, but isn't there still an underlying direction regardless of language?

• Composition relationships have a valence in that there is a progression from greater part to sub part
• Categorization relationships have a valence in that there is a progression from less-specific category to more-specific category
• Functions have a valence in the transformation of input(s) to output(s)

Phrased as a proposition: To me it seems like a lot of functional terms have a valence/direction that exists independent of language, and I'm wondering if thats a dumb thought or this is something thats discussed/contested & I dont know what to google

Thanks again, and sorry for my noob questions! (If theres a better place to direct these lmk)

[apease]

There are interesting questions and fun to study with a linguistic corpus analysis. One counter example I think of "I derive 5 from the sum of 2 and 3." That linguistically orders from outputs to inputs. We have not only "The wheel is a part of the car." (smaller first, then larger object), but also "The building has a bathroom." (bigger first). You could select all the SpatialRelation instances, get their format and termFormat expressions and see how each of the linguistic terms operates in a corpus like the Corpus of Contemporary American English. But I think just from intuition I've found a couple of counterexamples above, if I'm understanding your point correctly.

[twitchyliquid64]

Ah, interesting. I guess for every predicate with a direction, theres an opposite predicate that describes the same transform in the inverse. Which would then lend value to defining the semantics of each argument of a relation on a per-relation basis, as SUMO does.

That makes sense.

It still feels more natural to me to describe relations in terms of a progression, i.e.

(Farmer -[instance]> ?F)

instead of

(instance ?F Farmer)

but its sounding like a me thing rather than a property of relations.

[apease]

You'll see inverses in description logics, but not in SUMO, because in FOL (or beyond) they aren't needed. (part ?X MyCar) queries for the parts of a car, (part MyWheel ?X) queries about what things my wheel is a part of. The relationship of being part of something is not a property of just the containing thing or the included thing.