>Pat,
>
>We may actually be making progress! See comments below. (01)
Indeed. I sense a convergence. (For those watching from the
sidelines, this is remarkable!) (02)
>>>>Logic does not *require* reasoning; but in any case, deriving the
>>>>conclusion that <this>=<that> is a very simple kind of reasoning.
>>>>
>>>Oh, part of your logic-fixation again. Yes, you can characterize a
>>>mapping of <this>=<that> as simple logic if that makes you feel
>>>any better.
>>
>>
>>Its not a case of feeling better, but of whether or not this is a
>>logical inference; and it is. Quite a lot of OWL applications are
>>used almost exclusively to come to conclusions of this form, using
>>inverse-functional properties. Quite a lot is known about
>>techniques of (reasonably) efficient equality reasoning, which I
>>imagine would be directly applicable to your area of concern. I
>>think its you who has logic-phobia. :-)
>>
>Sorry, not guilty!
>
>The Topic Maps Reference Model does *not* specify or require any
>particular method for determining that two (or more) identifications
>are of the same subject. (If we failed in that regard I would like
>to know so we can fix it.)
>
>Let's take another specific example: The gene for the chemokine
>lymphotactin was was discovered by by three different groups, who
>promptly named it SCM1, ATAC and LTN. It is also known as LPTN and
>has an "official" name of XCL1. (03)
Quite. This sort of thing happens all the time. I am Patrick John
Hayes and Pat Hayes and P.J.Hayes and Hayes, P.J. and many others.
Its only even noted as an issue in technical areas where there might
be an expectation of a single 'official' name for a thing. (04)
>There are at least two ways that we could devise a mapping that
>would result in a "merging of those different identifications.
>
>1) A researcher reading the literature notices that ATAC identifies
>the same subject as XCL1 and enters that mapping in a topic map. (I
>am grossly over simplifying, the string "ATAC" is obviously
>inadequate for identifying this gene. A google search today
>returnds some 4,900,000,000 "hits" and not all of them are about
>this gene.
>
>2) Assuming the existence of an ontology upon which to base
>inferencing (yes, I know gene ontologies exist) and assuming that
>all these terms havebeen mappped to that (note singular) ontology,
>then you could use an inferencing engine to produce the same result. (05)
Well, you don't need the singular. There can be several ontologies
involved: one of them might be nothing more than a term-to-term
mapping. (06)
>As I said above, topic maps are not constrained to use only one or the other. (07)
Quite. I didn't intend to imply otherwise. (08)
>So, why would I ever want to use #1 and not #2? (09)
Why not allow the possibility of using both? But to do that requires
that your formalism is provided with a clear logical semantics. Which
is all that I have been urging on you: that your notation be provided
with a semantic theory, so it can clearly be related to the logic
used by inference engines. To give it such a semantics is not to
'logify' it, or rule out its use by humans (in fact, despite the dark
reputation of logic, we found that giving RDF a precise formal
semantics actually helped many real live people, including
developers, not least by providing quick ways of resolving otherwise
interminable debates.) (010)
>Well, let's look at : "The success (or not) of HUGO nomenclature,"
>Genome Biology 2006, 7:402. (http://genomebiology.com/2006/7/5/402)
>That is the source of the foregoing example. The authors found that
>there is no clear tendency for authors to adopt official symbols.
>And that 14% of genes are never referenced using the official
>symbols.
>
>That doesn't bode well for any system, such as an inferencing
>engine, that depends upon a particular nomenclature mapping for
>success. (011)
Im not sure what you mean here. Logic uses names, of course: but
inference engines don't depend on particular nomenclature mappings.
The whole point of having equality in a logical language is to allow
for the possibility of there being multiple such mappings, and being
able to state relations between them. Rather simplistically, perhaps,
in classical FOL= ; but more recent logics such as the IKL we
developed for IKRIS have much more elaborate name-mapping facilities.
In fact, the entire purpose of IKL was to provide for interoperation
between systems which use different logics and different ontological
and naming conventions. (012)
>That is *not* a dig at logic or inferencing engines, just
>recognition that the initial condition for sucess, a common ontology
>to which genes are mapped, is highly unlikely. (013)
That isn't an initial condition for success. There might be inference
paths through multiple ontologies (which may include things like
wordnet) which allow the conclusions to be drawn: and that is much
more plausible. For example (I know zilch about gene technology) if
there is some condition on a gene which is such that only one gene
can satisfy this condition, and one ontology entails that the gene
satisfying this condition is called 'ATAC' and another ontology does
the same thing, perhaps using a different inference path, but uses
the string 'XCL1', then the engine, and we, can infer that ATAC=XCL1
without being explicitly told it. In a much simpler setting, the FOAF
project uses this kind of inferencing to decide that the Pat Hayes on
one web page is the P.J.Hayes on another, based on the fact that they
have the same phone number or home page. (014)
>So, it isn't really logic-phobia but facing what I think you would
>call an "objective fact," that is that people are going to identify
>the same subject differently and that includes the ontologies that
>they will choose (or not) to use. If you have the information
>necessary to support an inferencing engine, by all means, use one in
>your topic map.
>
>Or to put it another way, if you can support an "intelligent agent,"
>then do so. But don't overlook "intelligent users" while waiting for
>"intelligent agents" to arrive on the scene. (015)
See below for comment on this. (016)
<snip> (017)
>>>But more to the point and what seems to be completely ignored in
>>>this discussion, is that most people don't use first-order
>>>semantic frameworks to make those mappings.
>>
>>
>>Ah, this is the nub of the matter. I reject this claim. True,
>>people don't formalize their conclusions to themselves in a
>>classical textbook first-order notation; but are they in fact using
>>first-order valid inferences? I think in fact they often are. Bear
>>in mind that machine inference is not done by using a
>>textbook-style display of a linear proof sequence, A following from
>>B by a logical inference rule: it is done by heuristic techniques
>>generating a semantic tableau, or by a Davis-Putnam process, or the
>>like. But it is still first-order inference. And humans reason in
>>ways that seem to be almost directly first-order, including when
>>deciding identities. Bill was wearing a yellow jacket; very few
>>people wear yellow jackets; the only person I can see wearing a
>>yellow jacket is that guy over there; that guy is probably Bill.
>>This is a first-order logical inference. Or: Joan should be here by
>>now; Joan hasnt phoned; if Joan had known she was going to be late
>>she would have phoned; so, something unexpected must have delayed
>>her. That is a first-order logical inference. And so on, and on. I
>>don't want to claim that *all * human reasoning is first-order: but
>>a surprisingly large amount of it seems to be.
>>
>Well, I was using your "first-order semantic frameworks" in the
>sense of formal application, without reference to their thinking
>processes being first-order. May well be a large amount of it is
>first-order as you say. But that elides over the issue that so far
>as I know, subject to your gentle correction, there is no
>generalized inferencing engine that comes close to those first-order
>processes when performed by a human agent. (018)
True. But so what? There are engines which can do useful reasoning
which humans can't do because their attention span is too limited
(which is why companies like Fair Isaac have spent millions of
dollars developing the world's fastest inference engines.) The point
is that machines have talents which are useful to people, and what we
ought to be trying to do is find ways for them to help us. (019)
<snip> (020)
>>Of course not. I agree this is an important issue; but topic maps
>>simply record this, as far as I can see. To record an equation is
>>easy: it can be done in almost any formalism.
>>
>Sorry, topic maps simply record .... what? (021)
That two names (used in their appropriate context) mean the same
thing. That is, an equation. (But I see now that this was a
misunderstanding.) (022)
>>>Moreover, with a topic map I can record my mapping between
>>>different identifications for the same subject, which would be a
>>>benefit to the next person who searches for that subject under any
>>>of its identifications.
>>
>>
>>If I follow what you mean here, we also invented a notation for
>>this in IKL. Its basically the use of typed literals, where the
>>'datatype' is the identification mapping. But I agree, having a
>>very general notation for this is useful.
>>
>>So let me see if I understand this. as you explain it. A topic map
>>is basically a complex name for a thing, one which records a
>>variety of 'superficial' names, each used in a different context of
>>identification to refer to the same thing. The TM records both the
>>fact of these names being coreferents and records, and links to,
>>the various contexts of identification where the superficial names
>>are used. So one might express it as a collection of <namestring,
>>identification-context> pairs. Is that right?
>>
>Yes, modulo that a topic map is a collection of such "complex names"
>for things, where the "complex name" (in the reference model we call
>them proxies) can also include any other properties that are
>associated with a thing. (023)
Oh but wait: that little extra provision is very important. It means
that (as I had previously thought) TMs are not *just* a way of
organizing names to indicate coreference. They are also a way of
making assertions about things. They are an assertional language.
Once you get into that territory, you are definitely in need of a
semantic theory, and you really ought to be taking advantage of all
the scholarship and analysis that has been done by logicians (and
philosophers of language). To try to do this from scratch, inventing
your own terminology as you go, seems to be just silly, when there is
a mature science available, with a fully worked-out mathematical
basis, a precise technical vocabulary and a mature deployed
technology. It wouldn't diminish or deprecate Topic Maps to admit
that they are a way of packaging a part of modern logic for a
particular purpose. (024)
<snip> (025)
>>>*(The SW proposal that every subject have a single unique identifier
>>
>>
>>Whoa, that isn't the SW proposal, and its not the TAG Web
>>architecture position either. Of course there isn't a single unique
>>identifier, in general: if there were, owl:sameAs would be vacuous.
>>I agree, the idea of single unique 'true name' is ridiculous. I
>>call it the EarthSea theory of reference, after the idea in the
>>Ursula LeGuin novels.
>>
>Ok, so we agree that a "true name" is ridiculous. Great!
>
>I really think our difference is one of emphasis if that. You want
>to say that users to create mappings between different
>identifications of the same subject are using "first-order"
>processes. (026)
I wouldn't even want to insist on that. People can of course use any
processes at all to come to a conclusion. What interests me more is
that when that conclusion, however it was arrived at, is *recorded*
in a way that is intended to be used by machinery - if only in a very
simple and straightforward way, such as substituting one name for
another - that the notation or encoding method that is used to record
it in be provided with a precise, mathematically described,
'logic-style' semantic theory. Not in order to impose a tyrannical
mainstream logical cultural hegemony, and not to subtly trick users
into using alien notations, and not to require that all users have a
graduate degree in logic. Rather, the point of having such a
semantics for the formalism is purely pragmatic: it provides the only
secure, non-procedural basis for interoperability between all the
different formalisms. No one formal notation is going to be the
single final form that everyone uses. But as long as all the
formalisms have a common semantic base, there is at least the
possibility of making translators between them. That is what I have
devoted the last several years to achieving: giving a variety of
formalisms a common semantic base. So far we have done it for Common
Logic, IKL, RDF, RDFS and OWL. (It looks as though OWL 1.1 will
deviate from this is some subtle ways that may not be very important;
more seriously, the long-awaited Rule Language (RIF) may be even less
semantically aligned with the RDF/CL vision. Oh well, I did my best.)
But if one steadfastly refuses to even give a formalism a semantics
at all, it is simply marks on a surface, or a bucket of bits. I'm not
meaning to imply that TMs mean nothing: they clearly do mean
something. But until that something is described in ways that we can
analyze with enough mathematical precision to be the foundation of
writing correct code, interoperation with TMs must always be a matter
of guesswork. Which is a poor basis on which to try to build a
planet-wide system of communication. (027)
>If you want to describe their activity that way, I have no objection
>as it is your description. Where I would object is telling users
>that they have to use explicit "first-order" processes to make those
>mappings. (028)
Im not wanting to tell users they *have to* use anything. What I do
ask, however, is that when they go home and leave some stuff in the
machine, that I know how to interpret it without having to call them
up and interrupt their dinner. (029)
Pat (030)
>(Noting that I certainly agree you can use an inferencing engine to
>make to same mappings, well, asssuming you can find one as robust as
>a human user. Your mileage may vary.) (031)
PS. One final remark. I get the sense - and if this is wrong then I
apologize in advance - that you often see these debates in a kind of
human vs. machine way, with TMs being on the side of the humans, and
Logic, in all its mechanical clunkiness, as being somehow
representative of the machine world. This seems to come through in
your disparaging comments about the state of the AI, your tailpiece
slogan, and your contrast between "intelligent agents" (a term which
I abhor, by the way) and "intelligent users". (032)
If this is even slightly close to a reasonable analysis, let me urge
you to not think of AI technology this way. Most practicing AI
technologists don't. For several years now, my colleague Ken Ford and
myself have been arguing a completely different agenda for AI, in
which the acronym could be re-understood as 'Amplified Intelligence'.
Ken calls this 'human-centered computing': the idea is create machine
systems which can act as "cognitive prostheses" or amplifiers of
human abilities, so that the entire system of (person + AImachine) is
capable of more than either can achieve alone. I can go an about this
idea at length: too much length for this message. But the point I
want to get across is that it is helpful to think of AI methods,
including mechanical inference, as aids to people rather than
competitors to human dominance. Forget that damn silly Turing Test
(http://dli.iiit.ac.in/ijcai/IJCAI-95-VOL%201/pdf/125.pdf), and stop
worrying about the inhumanity of the machines. Backhoes and
eyeglasses aren't human either, but they are very useful muscle- and
vision-enhancers. What we need now are mind-enhancers :-) (033)
Pat (034)
>Hope you are having a great day!
>
>Patrick
>
>--
>Patrick Durusau
>Patrick@xxxxxxxxxxx
>Chair, V1 - Text Processing: Office and Publishing Systems Interface
>Co-Editor, ISO 13250, Topic Maps -- Reference Model
>Member, Text Encoding Initiative Board of Directors, 2003-2005
>
>Topic Maps: Human, not artificial, intelligence at work! (035)
--
---------------------------------------------------------------------
IHMC (850)434 8903 or (650)494 3973 home
40 South Alcaniz St. (850)202 4416 office
Pensacola (850)202 4440 fax
FL 32502 (850)291 0667 cell
phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes (036)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (037)
|