From: Barry S. <phi...@bu...> - 2010-07-27 21:28:57
|
At 01:25 PM 7/27/2010, David Osumi-Sutherland wrote: >On 27 Jul 2010, at 17:19, Barry Smith wrote: > > > At 11:40 AM 7/27/2010, Robert Stevens wrote: > > > >>> <snip> > >> > >> > >> > >>> It is clear that many application ontologies will need to include, > >>> in > >>> addition to representations of universals (types, kinds, whatever > >>> you > >>> prefer to call them), also representations of defined classes. The > >>> (heuristic) proposal is (1) that single inheritance holds for the > >>> former, but not for the latter and (2) that is_a links should be > >>> asserted for the former and inferred from the latter. This coheres > >>> well, I think, with Alan's ideas on normalization. > >>> BS > >> > >> Are defined classes in OWL not universals (or whatever)? > > > > The idea is that there are certain repeatable features (e.g. charge > > of an electron, influenza), which reference ontologies (in the ideal > > case) should represent. We call these 'universals', because the word > > is sufficiently toxic that it is unlikely to be purloined (in the way > > of 'class') for multiple other uses. > > > > Given these repeatable features we can create terms representing many > > defined classes, e.g. the class of influenza patients in the Long > > Island Jewish North Shore Hospital System > >This makes it sound like the distinction between Universals and >defined classes is clear enough to be useful. In my experience it is >not. It is supposed to be as clear as the distinction between a type or kind of entity and a collection of instances of that type or kind. Thus AIDS is caused by the HIV virus (Here 'AIDS' refers to the disease-universal) AIDS is spreading rapidly through Asia (Here 'AIDS' refers to the collection of instances) >Imagine a single instance of a fly sensory organ - a small bristle on >the edge of a wing. > >This bristle is variously and usefully classified by scientists >according to: 'bristle' is, it seems to me to be a good candidate name for a universal; that is perhaps why you feel tempted to use the word 'instance' There is, it seems to me, only one attractive axis of classification for bristle, and that is the anatomical axis. >its function (it is both chemosensory and >mechanosensory) Here you are not classifying bristles. You are classifying functions. GO-BP has 'mechanosensory behavior' which, in the proposed idiolect, would refer to a certain process-universal; I am an advocated of creating a GO-BF (for biological functions), which would include function-universal terms like chemosensory function mechanosensory function >; the shape and structure of its cuticular >specialization (a socketed bristle with a terminal pore), by the >number of neurons whose dendrites terminate in it (5), PATO has some of these shape universal terms, doesn't it? > by its >location (the dorsal wing margin) 'dorsal wing margin' names an anatomical location universal >, by its developmental origin and >pathway. and so forth. The bristle can be classified in those various different ways because there are those various universals instances of which inhere in it, or provide its location, and so forth. > In future it may be classified according to the chemosensory >genes its neurons express... > >Which one of these classifications is the Universal? Classifications are possible because entities are similar to each other in certain respects. Universals are the respects are these respects. > As far as I can >see, different classifications are useful depending on the questions a >scientist is interested in. If Universals are simply the subjects of >scientific discourse - the things that scientists make assertions >about - then this could apply to 'mechanosensory organ', wing margin >sensillum, bristle, bristle with pore, socketed bristle... Exactly. >I haven't found it very productive to get a bunch of ontologists and >scientists together to agonise over how we can make a single >inheritance hierarchy of 'univerals' for some domain of biology. That is what the GO has been trying to do for 10 years, now. Perhaps you, too, have been speaking prose all along. > Even >if we arbitrarily pick, say 'structure' as a single 'axis of >classification', there's no reason to believe there will not be many >useful ways to classify something structurally. Exactly. These different useful ways are possible because there are those different universals, classified in different ontologies. (That is why we have these different ontologies.) >What I have found to be very practical is to list all of the ways that >scientists want to classify something, and then try to automate as >much of this classification as possible using defined classes (e.g.- >wing margin bristle EquivalentTo: bristle and part_of some wing >margin). I'd happy be to automate all classification in this way and >assert no classification at all. The choice between automated >classification using defined classes and asserted classification then >comes down to the practical question of what we can easily formalize. >The asserted classification is not necessarily more important or less >arbitrary. It's just what's left. The problem is one of coordination. If just one person is dong all of this, along the lines you suggest, then we can leave it to that one person to draw the line between asserted and inferred classification on practical grounds. But we have to rely on multiple groups. How are these multiple groups to be aligned? I hope to give an answer to this question in my response to Merrill. >Now, is the ontology I maintain an application ontology simply because >of my use of defined classes? What reason is there to believe that >any of the ontologies that are now full OBO foundry members can >reflect scientifically useful classification within their domains and >be single inheritance asserted classifications with little or no use >of defined classes? If there is no reason to believe this, are these >all actually application ontologies? And if there is reason to believe this? >You might have guessed by now that I'm pretty skeptical that the >distinctions between Universal and defined class and between >application and reference ontologies are clear enough to be useful. >I'm happy to say that we should avoid filling our ontologies with >obviously contingent and arbitrary classes and if that is all that is >meant by 'Universal' then I'm fine with it. I don't think you've formulated it very well, but I think, as concerns the idea -- which is the practical one of assuring coordination -- then you have, I believe, come pretty close. > But I still haven't >heard any convincing theoretical arguments for why we might expect >that good classification consists of single inheritance hierarchies of >'universals' or for that clearly differentiates universals from >defined classes. Practically, I can say from experience that an >approach to ontology building that places great emphasis on judging >what counts as a universal and finding ways to construct single >inheritance hierarchies with these is highly unproductive. There are multiple elements involved in building ontologies, and certainly 'judging what counts as a universal' as not an appropriate way of describing any of them. (More in the Merrill response.) BS >- David > > > > >> Is this kind of technique only to be used in application ontologies? > > > > We are attempting not to lay down the law too narrowly on this and > > many related questions until we know what works best in practice; > > however we anticipate that application ontologies will likely include > > many representations of defined classes, while reference ontologies > > will likely include few. > > > >> So, not in GO? Does multiple inheritance still need to be removed > >> from reference ontologies? > > > > Efforts along these lines for GO are underway. Experimentally > > BS > > > > > >> Robert. > >> > >> > >> > >> > >> > >>>> Barry Smith <phi...@bu...> writes: > >>>>> Here, too, we are broadly in agreement -- the GO itself is > >>>>> attempting > >>>>> to realize a modified version of your proposals here, namely to > >>>>> have > >>>>> an asserted single inheritance hierarchy and to infer the rest. > >>>> > >>>> > >>>> > >>>> ------------------------------------------------------------------- > >>> ----------- > >>>> The Palm PDK Hot Apps Program offers developers who use the > >>>> Plug-In Development Kit to bring their C/C++ apps to Palm for a > >>>> share > >>>> of $1 Million in cash or HP Products. Visit us here for more > >>>> details: > >>>> http://ad.doubleclick.net/clk;226879339;13503038;l? > >>>> http://clk.atdmt.com/CRS/go/247765532/direct/01/ > >>>> _______________________________________________ > >>>> Obo-discuss mailing list > >>>> Obo...@li... > >>>> https://lists.sourceforge.net/lists/listinfo/obo-discuss > >>> > >>> > >>> > >>> ------------------------------------------------------------------- > >> ----------- > >>> The Palm PDK Hot Apps Program offers developers who use the > >>> Plug-In Development Kit to bring their C/C++ apps to Palm for a > >>> share > >>> of $1 Million in cash or HP Products. Visit us here for more > >>> details: > >>> http://ad.doubleclick.net/clk;226879339;13503038;l? > >>> http://clk.atdmt.com/CRS/go/247765532/direct/01/ > >>> _______________________________________________ > >>> Obo-discuss mailing list > >>> Obo...@li... > >>> https://lists.sourceforge.net/lists/listinfo/obo-discuss > >> > >> > >> > ------------------------------------------------------------------------------ > >> The Palm PDK Hot Apps Program offers developers who use the > >> Plug-In Development Kit to bring their C/C++ apps to Palm for a share > >> of $1 Million in cash or HP Products. Visit us here for more details: > >> http://ad.doubleclick.net/clk;226879339;13503038;l? > >> http://clk.atdmt.com/CRS/go/247765532/direct/01/ > >> _______________________________________________ > >> Obo-discuss mailing list > >> Obo...@li... > >> https://lists.sourceforge.net/lists/listinfo/obo-discuss > > > > > > > > > ------------------------------------------------------------------------------ > > The Palm PDK Hot Apps Program offers developers who use the > > Plug-In Development Kit to bring their C/C++ apps to Palm for a share > > of $1 Million in cash or HP Products. Visit us here for more details: > > http://ad.doubleclick.net/clk;226879339;13503038;l? > > http://clk.atdmt.com/CRS/go/247765532/direct/01/ > > _______________________________________________ > > Obo-discuss mailing list > > Obo...@li... > > https://lists.sourceforge.net/lists/listinfo/obo-discuss > >David Osumi-Sutherland, PhD >Ontologist / Curator >Virtual Fly Brain / FlyBase >Department of Genetics >University of Cambridge >Downing Street >Cambridge, CB2 3EH >UK >+44 (0)1223 333 963 > > > > >------------------------------------------------------------------------------ >The Palm PDK Hot Apps Program offers developers who use the >Plug-In Development Kit to bring their C/C++ apps to Palm for a share >of $1 Million in cash or HP Products. Visit us here for more details: >http://ad.doubleclick.net/clk;226879339;13503038;l? >http://clk.atdmt.com/CRS/go/247765532/direct/01/ >_______________________________________________ >Obo-discuss mailing list >Obo...@li... >https://lists.sourceforge.net/lists/listinfo/obo-discuss |