Theiling Online    Sitemap    Conlang Mailing List HQ   

Frame-based vs ontology-based vocabulary

From:Sai Emrys <sai@...>
Date:Tuesday, February 20, 2007, 11:15
This idea stems from something I've been mulling for a while now; for
me of course it's intended for my non-linear writing system idea, but
I believe it could be put to excellent use in a more normal conlang

Thanks to everyone at the last bay area conlangers' meet for helping
with this. :) It's still far from clear to me, and I honestly don't
think I fully understand what I'm trying to get at, which is why it's
hard for me to explain it or to figure out how to do it. Hopefully
this little ramble will make enough sense to some of you that you can
ramble in return and we get some ideas out of the process. ;-)

This is going to consist of more questions than answers, since I'm
still in the 'highly speculative brainstorming' phase.  And I'm going
to be a little bit loose with my vocabulary since it's 2:43am and I
haven't slept in.... er, a while. (I'm not really sure when I last
woke up...)

If you're not familiar with what frames or 'agent roles' are, I
suggest you take a look at John Quijada's talk at the LCC1, hosted on, for a very understandable overview. (What, you
haven't watched all the conference videos yet??)


All vocabularies that I know of, and even the fundamental ways of
approaching the naming of things, are ontologically based. That is,
you name *things* by placing them in categories, specifying their
attributes, or giving them an individual name.

This general approach has been rehashed lots of ways over the years,
for better and worse. It's clear that there are certain inherent
limitations to the method - the main one being that ontologies are
highly arbitrary in what they choose to elevate to a relevant property
(and how they're sorted), and where they draw the boundaries.

As an example, we name biological things by their hierarchical
membership. We name and describe anything we come across in terms of
properties - e.g. a cat, longhaired, white-and-black, Maine Coon, by
the name of Ki, owned by / owner of me. (^^)

This seems a bit odd to me at a fundamental level, because things are
hardly ever really discussed by themselves; they come up in context.
E.g. I brush my teeth with a toothbrush. Why would I mention it except
in a context? I might as well call it "a thing with which I clean my
teeth", though that is ambiguous. But that is still very definitely an
ontological approach, see - I'm naming it, giving it a category. And
then I'd further specify ontologically, like "... which has bristles"
or "sticklike" or somesuch.

And I think we're all familiar with the horribly borked kludge that is
practically all of 'element combination' vocabulary systems - of which
Austral is probably the prototype (I'm still not sure if it was meant
as a parody or not...).  E.g. what does tree + thread mean? What about
person + fire? Etc. Just... no. Please. :-P It's fine as a way to make
arbitrary but mnemonic-friendly names, yes, but it sucks as a way to
make the naming scheme transparent.

What I would like instead is a system based on frames. For example,
there is the 'commercial transaction' frame of cogsci standards fame;
it has various roles that are implied to exist within it - buyer,
seller, money, object - and perhaps various other peripheral ones,
like the ongoing relationship between two roles (client), habital
location (store, market), etc etc. These names are terms of
convenience; they probably could be generalized to something like the
'agent roles' list, like agent, patient, instrument, etc... though it
is clear that in standard languages at least it is possible for
(nearly?) any frame-role to be assigned any of a number of
agent-roles. E.g. money can buy an object just as much as a seller;
you could even talk about money 'using' the person as its instrument.
(Viz. Dawkins' "the selfish gene" for a more plausible example of this
sort of reversal.)

So instead of saying something like "I bought a game from the market"
it might be something like "[commercial transaction](buyer = I, place
= generic, money = ?, object = [game](instrument))". Except of course
much more sexy than that; that's just the analytical version.

I'm not sure whether or not you could completely get around
ontological naming as a supplement to specify which particular
potential frame-role-filler you mean when there are multiple plausible
options (eg "what kind of game did you buy?"... how could you specify
that using frames?). And you'll probably need to at least keep
arbitrary entity-naming (like "Bob").

So maybe all this system would do is replace certain (broad) classes
of verbs and nouns. I'm really not sure on this point.

So my main question is: how could you maximally use this?

Specifically, how could you construct as much as possible of your
"vocabulary" (if it still functions like one... I'm not at all certain
that it would) or "grammar" to work based on frames, i.e. based on
relationships in a socially-known and merely-pointed-to context,
rather than on ontologically-derived naming and role-specifying words?
Could names be abandoned altogether?

How many frames are there vs lexical items, in e.g. English?

How many frames are generalizable? How could you make as many as
possible to be variants of each other, and then only need to have a
symbol for one meta-frame and some way to generically specify which
variant you are referring to in terms of its argument structure? (E.g.
barter vs commercial transaction vs commerce vs social transaction vs
spatial movement of small-multiple items ... all share strong common

What would a generalized "agent roles" list look like for frame roles?

What does having both working in parallel give us? Could that added
information (pragmatic, presumably?) be somehow marked in another
(better?) way?

How much does this affect 'verbs' vs 'nouns' vs ...? I don't think I'm
clear enough on the concept, because this question is particularly
fuzzy to me.

How could it be implemented in a sexy, non-cumbersome, non-LISP-y way?
(Both as a NLWS and as a standard primarily-spoken language...)

Are there any languages that already do something similar (perhaps
even an ANADEW) that would be useful to ste^H^H^H learn from? I have
heard bits about some American Indian language's unusual way of
describing a canoe being rowed towards the shore as 'directed vector
motion' or some such, but I don't know the details, whether they have
other interesting features, whether it's applicable to what I'm
talking about here, etc.

Finally, I'd like to ask for specific recommendations for books,
papers, or other resources I should look at that answer questions
related to this or would give me inspiration / catalyzation.
Particularly I am interested in something that would give a good
in-depth overview of what kinds of semantic and pragmatic information
languages communicate in general - e.g. agentivity, causality
emphasis, etc. I don't care at all about how specific languages
implement them, and I definitely don't care about the formal
semanticists' shtick about whether sentences are "true" or the like.
The best I've read to date that made me think of stuff I hadn't before
is _Describing Morphosyntax_ (ftw!), but I'm sure that I'm not nearly
the first to want to think about this. I just don't know where to
look; most of what little I have seen doesn't really get at what I

Thanks in advance,

- Sai


Jim Henry <jimhenry1973@...>
And Rosta <and.rosta@...>