Re: alternative translations

[cbmvax!uunet!MATH.UCLA.EDU!pucc.PRINCETON.EDU!jimc]

Bob Chassell <bob@GNU.AI.MIT.EDU> writes about alternative translations:
>     le nanmu cu jukpa le sovda le zu'o                febvi
>        man      cook     egg      activity-abstractor boil
>        x1       gismu    x2                           x3
> is translated as:                             [in the style he suggested]
>     x1: cooker, I have in mind, the man
>         gismu: is in a relationship with
>     x2: food-prepared, I have in mind, an egg or eggs
>     x3: method-of-food-preparation, I have in mind,
>              activity abstraction related to boiling
>
> This form of translation appears to work for gismu that take two or
> more places, but fails for one place gismu.

Bob has expressed in English a formalism that I often use (and that some
people have complained gives them headaches).  But when correctly
viewed, the same formalism works equally, though with an Excedrin-
strength headache, on one-argument predicates.

My goal is to "interpret the sentence" independent of the language
used, rather than to "translate it into English".  Of course most
people on this list are English native speakers so there's an obvious
relation between the goals.

The formalism is this:
1.  A predicate word is a symbol for a relation.
2.  A relation (in this context) is a true/false valued function of N
        arguments, the places.  If "the rat eats the cheese", given
        particular referents for the two sumti, either it does or it
        doesn't.  Fuzzy logic is accomodated easily.
3.  Such a function can equivalently be represented as a set of pairs
        (for citka-eat) of thus-related referents.  One of those pairs
        will be (the rat, the cheese), whereas (me, the cheese) will be
        missing since I didn't get to eat the cheese.  Predicates with
        N places will have place occupant lists with N members.  Fuzzy
        logic is not easy with this representation.

Then for 1-place predicates, the relation definition set will consist of
monads (sets of one item each)!  Or the truth function will have only one
argument.

It is clear that with this kind of definition of relation, it is very
important that each relation have a specific number of places, possibly
infinite though I hope not; an "indefinite" collection of BAI cases is
very hard to handle.

                -- jimc