Lists Home |
Date Index |
the more i think about these issues, the more i'm convinced that, like
people, these systems can only work if there is a training feedback loop.
it takes a minimum of 8-10 years to train a human being to a level where
they can cope with modern society and communicate effectively on a wide
range of issues. and that's with the most complex device in the known
universe as the computing engine.
somehow i think it's hubris that we think we can lay down a few
relatively simple rules and get machines to understand enoug to realise
some of the ws-* ambitions. i can only assume that many of the well
meaning people behind these initiatives have somehow forgotten most of
the maths and cognitive science they learned (i know it's harsh, no need
to flame me)
eg i'm using festival in an automated dj system that drives my private
music collection. all day it announces songs and artists, and then plays
the songs. trivial? well sometimes it does the right thing by 1978 (or
similar) as a date - pronounces it as such, sometimes it pronounces it
as a number. it always gets confused by 1928-1944 (or similar). and you
should hear it's attempts at tijuana :)
now this can be fixed by identifying the specific tokens it has trouble
with and training it to recognise them and then have a special and
correct pronunciation (like learning english irregular verbs etc). at
the moment i have to train it. one day the computers should be able to
train each other.
all the ws-* stuff and all the owl stuff will need similar feedback
loops for learning. they will need to cope with generalities and
specifics. they will need access to computer equivalents of google and
they will need a way to evaluate the results.
can't remember how many nodes and connections in the human brain (up
there with stars in the universe i think) but i don't think any of the
existing stuff will come into it's own until we have computing devices
of similar complexity.
meanwhile we'll keep using biological computers to do the hard stuff
maybe i view the problem as more complex than it is, but i don't think
we have devices, let alone supporting technology, of sufficient
complexity to even begin to tackle these problems.
Chiusano Joseph wrote:
>I should be clearer: I was actually thinking of a different type of
>querying than ad hoc. I see ad hoc querying (as you know) as a type of
>query in which the queryer (is that a word?) is presented with a user
>interface (such as a portal) that may present drop-down lists of various
>category values (based on a set of taxonomies) from which they select,
>or use some type of textual query, etc. In such cases, there is full
>knowledge of the database schema by the entity that is requesting the
>information (e.g. the application code or ad hoc query tool behind the
>My example was one in which the request for information (the additional
>needed information) was made by an entity (the Hotel Reservation Web
>Service) that had no knowledge whatsoever of the "source" database
>schema - it just knows that it is missing information that it requires.
>So it seems to me that we may need some mechanism for an entity to
>communicate a request for information it needs to another entity in a
>standard mechanism, so that there is complete understanding between the
>2 entities. Of course, the Travel Agent Web Service should be fully
>aware of the source database schema - so as long as it fully understands
>the request from the Hotel Reservation Web Service, it should be simple
>for the Travel Agent Web Service to query the source database for the
>needed information information. Now that I think about it more, a
>semantically aware database may not be necessary in this case - as long
>as there is sufficient semantic understanding between the requestor
>(Hotel Reservation Web Service) and requestee (Travel Agent Web
>However, there may be cases in which such a database may be valuable -
>let's suppose that there is a sufficient level of trust between a
>requestor and a semantically aware database, but an insufficient level
>of semantic understanding (as if the Hotel Reservation Web Service made
>a direct request to the source database in the example I've given). This
>is a case where such a database may be valuable, so that the requestor -
>which has no knowledge of the database schema - can request information
>from the database and receive exactly what it was looking for.
>In closing, it looks like this could be a single mechanism to me - but
>applied to different types of entities (Web Services, databases, etc.).
>Booz Allen Hamilton
>Strategy and Technology Consultants to the World
>"Bullard, Claude L (Len)" wrote:
>>Unless one is say, creating ad hoc SQL queries and sending them
>>to a non-collocated database, why would deep schema information
>>be necessary? I don't think many private holders of data will
>>be too eager to expose all of the local schema information but
>>they would be willing to negotiate services that equal reports.
>>It is completely possible to expose the schema or to expose
>>an interface schema that middleware then transforms into
>>the local schema. As you say, Thomas, mappings. BTW, why would I
>>need OWL to implement what is essentially a business object?
>>I don't see the problem. What am I missing?
>>Joe, you may be asking for ad hoc querying. If so, it is
>>doable but not often done for external resources. Crystal
>>Reports and ODBC make it possible. The trick, as you note,
>>is that given a flat set of descriptions without the integrity
>>information, the security objects, etc., it is hard to build
>>a good query and easy to screw up a database if the ODBC
>>connection is badly configured. For that reason, the service
>>should provide the right codes and relationships for its
>>set of queries available by the role of the authorized and
>>authenticated querying agency. NCIC for example, has a
>>set of standard queries.
>>From: Thomas B. Passin [mailto:firstname.lastname@example.org]
>>Chiusano Joseph wrote:
>>>I sent this e-mail over 24 hours ago to the W3C Semantic Web Services
>>>Interest Group, and I did not receive any pushback (a gentle way of
>>>saying I did not receive a reply ;). So I'm wondering if that is good or
>>>Here is a scenario:
>>>At this point, we need the following to happen:
>>>(1) The Hotel Reservation Web Service must relay to the Travel Agent Web
>>>Service the information that is missing, and
>>>(2) The Travel Agent Web Service needs to obtain that missing
>>>information from the Travel Agent relational database
>>>It is #2 above that I perceive as a current gap - i.e. unless the Travel
>>>Agent relational database is sufficiently "semantically aware" (i.e.
>>>perhaps it implements an OWL ontology whose classes and properties are
>>>mapped to the database tables/fields respectively), there is no
>>>efficient and accurate way that the required information can be obtained
>>>from the Travel Agent relational database.
>>This seems to be no different from any other data integration problem.
>>In general, it is impossible to automatically map from one database
>>schema to another, because most databases do not/cannot contain enough
>>explicit schema and ontology information to do so. In many if not most
>>cases, people will have to create the mapping, or at least to adjust an
>>automatically-obtained mapping. If not a map, then a wrapper to make
>>the database respond like, say, and rdf database.
>>This is one reason I have never believed in the practicality of fully
>>automated service composition (or fully automated web service discovery,
>>for that matter, and for similar reasons). But if you say that the
>>agents are only to operate within known domains, the mappings or
>>wrappers could be prepared in advance, just as is done today.
tel;cell:+61 411 287 530