RE: SUO: Universal Time, other universals, and cultural contextof SUO.
Pat said " Ah, I misunderstood you. That idea is KR, true, but it was invented
in KR independently of the linguistic input. It is indeed interesting
that NL seems to utilise this kind of structure, let me say, and this
fact is an interesting datum; but even here, the exact kind of
'roles' that one gets from NL do not correspond in detail to the
'roles' (this term is linguistic, so may warp the discusion if not
scarequoted) that arise when one takes a more objectively ontological
perspective. Many properties of events have no explicit linguistic
markers, so are simply ignored by linguistic analysis: if there is no
case role or prepositions for it, a linguist will not even consider
it."
It depends on what you call a linguist. I agree that Hans Kamp looks at matters too much from a linguistic viewpoint only. But I'm pretty sure that a computational linguist working on deep semantics does take all aspects into account.
I am quite sure than no computational linguist takes all ontological matters into account. (Some cognitive psychologists, working in psycholinguistics, maybe: but even then I doubt it.) The point being that there are many ontological issues (to do with individuation, representation of time, of physical states, of geometry, and so on, and centrally ontological issues like whether or not to think of physical entities in mereological terms) which either simply do not arise in linguistics (even computational linguistics) or else do arise, but which natural language - which evolved for situated communication between human beings, not as a representational langauge for ontological use - has particular mechanisms for which are basically unsuitable for ontological use (I am thinking in particular of tense for representing time, and modality for representing rigidity, both of which have been an actively harmful distortion of good ontological practice which we have seen on these lists, which arise from a mistaken importing of NL devices and intuitions into the ontological realm.)
Ontology is more like Fodor's 'language of thought' than it is like NL. And of course the language of thought is not a natural language. If it were, NL comprehension would be trivial: just read in the NL sentences and store them.
The opposite is - unfortunatelly - true as well: many "ontologists" don't care at all about the properties of events (or entities in general) that are explicitly marked in language.
I think you will find that in these cases, the ontologists are deliberately disregarding the linguistic markers not through blindness, but because they beleive them to be actively misleading. (See my earlier email with excerpts from correspondence with Piek Vossen and Livia Polyani, for example, which is typical of many encounters I have had with computational linguists.) There is a basic and fundamental misunderstanding here, in my view: the idea that natural langauge and ontological formalisms are on some kind of continuum. If you look at the requirements they must satisfy, it is clear that they are in completely different functional spaces, and that the requirements on them are all in completely opposite directions For example:. NL needs to be informationally compact but can be highly ambiguous and contextual; NL utterances are often abbreviated, gnomic, or incomplete. NL is a language for communciation rather than representation; it does not need to support inference. It is rich with indexical and deictic markers and constructions, and intended to be understood in a surrounding context by creatures with sensory abilities which need to be sharply aware of the processing and epistemic abilities of the other creatures in their vicinity. It uses social mechanisms like humor, repetition, emotional content, irony, and so on. It mixes propositional utterances with commands, requests, questions and other speech acts; it uses narrative and tensed and rhetorical forms and God knows what else, It does not need to have a precise model theory or even a precisely defined syntax. None of this is true for ontological formalisms, which need to be at the opposite end of all these dimensions. Maybe most important, ontology langauges need to be able to represent all the stuff the is *never* said in NL because the users of NL always assume that the listener already knows it: all the unspoken knowledge of the world (including the stuff I once called 'naive physics', but also naive psychology, epistemology, sociology and so on.)
The resulting ontologies may then turn out to be unuseful for NL understanding.
Indeed. So what?. I would suggest that anyone whose primary interest in NL processing move to a different mailing list. This one is about ontology design. We also don't discuss the weather in Turkestan, for example.
Too often people claim that ontologies should be "language independent". But there is a big difference in being independent from specific languages (French, English, ...) or being independent from natural language as a medium of communication in general. The same people (I mean the language ignoring ontologists) claim that once the "pure" ontology is developed, it "just" takes a little additional effort to link it to NL applications.
I would not claim that. I am simply uninterested in NL applications, and do not regard them as centrally important. I think they will become less important as time goes by, in fact, as more people get used to working directly with ontologies on the Web, for example. I see no particular reason why ontology should pay particular attention to NL, NL work is one potential application, but it is not by any means the primary or most interesting one. Indeed it seems to me to be a very peripheral one, and becoming more peripheral all the time. I can forsee a bright future for ontology development without involving NL at all.
My experience is just the opposite.
A key point in any development is to know for what reason and by what animals the ontology will be used. If it is just for communication between machines, you can forget about the language aspects. But I guess that all machines are working for the benefit of humans, and hence, communication with humans is unavoidable.
Humans have many alternative ways to interact with machines than by the use of NL. I doubt that full NL comprehension will *ever* be a useful communication medium between humans and machines, in fact. (Limited comprehension may well be of use, as it is already.)
And if we want to get away with the poor point and click interfaces of today, language will come into play,
Language already comes into play. But not full-blown natural language.
and ontologies must be able to account for that.
You might be right about this. Only time will tell. I will be willing to get on with other, more productive, uses in the meantime, however.
Pat Hayes
---------------------------------------------------------------------
IHMC (850)434 8903 home
40 South Alcaniz St. (850)202 4416 office
Pensacola, FL 32501 (850)202 4440 fax
phayes@ai.uwf.edu http://www.coginst.uwf.edu/~phayes