Re: [ontolog-forum] language ambiguity (was: Axiomatic ontology)

["Adrian Walker" <adriandwalker@xxxxxxxxx>]

Hi Jakub & All --

A while ago, Pat Cassidy raised a good question about NL ambiguity and
whether it is a show stopper without long term research into full
natural language understanding.

As many folks on this list know, there's an online system [IBL] that
supports the writing and running of database applications -- as rules
in open vocabulary, executable English.  Pat's question was about
a specific application in the oil industry, but it serves to illustrate
one way of handling ambiguity.

Pat wrote:

As for how ambiguity can creep into IBL, I look at the first example:

estimated demand some-id in some-region is for some-quantity gallons of some-finished-product in some-month of some-year

in that-month an order for that-finished-product can consist in whole or part of some-base-product

in that-month the refinery some-name has committed to schedule some-amount gallons of that-base-product

we have some-method transportation from refinery that-name to region that-region
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

for demand that-id for that-finished-product refinery that-name can supply that-amount gallons of that-base-product


. . .  and then consider that, if someone creates a similar rule with the third line being:

in that-month the refinery some-name has uncommitted scheduled production some-amount gallons of that-base-product

 
There will be two rules having
similar intended meaning. , but with the difference that in the second
case it is more explicit that the refinery's output can be provided to
fill a new order, and is not already committed.  Whether the first
rule actually conflicts with the second depends on whether the one who
wrote the first rule actually intended to say that the output was not
already committed (which depends on a time point, not explicitly
mentioned in the rule).  The third line might also be:

in
that-month the refinery some-name has uncommitted scheduled production
some-amount gallons of that-base-product on date some-date

 
If someone, searching for the
second rule (looking for uncommitted production) runs a query that
fires the first rule, they may mistakenly believe that there is an
uncommitted production when all of the output may in fact be already
committed.  This is just one example, and I am sure you can find
many others,  of where different people, intending to say the same
thing, may phrase it differently enough to create problems in an
inference system.

To avoid unintentional
generation of conflicting rules and inferences, I expect that a system
will need true understanding of linguistic input.   
 
Adrian replied to Pat as follows:

You raise a good general
point illustrated with committed vs uncommitted in the rules for
the  oil industry supply chain example.

Fortunately, the angel is in the details.   The IBL provides
two mechanisms that help in such situations -- 1. context, and 2.
explanation.

1.  If you add the "uncommitted" rule as you suggest, the system responds as follows

In this rule, the line in some-month the refinery some-name has
uncommitted scheduled production some-amount gallons of
some-base-product should be (a) the conclusion of some rule, or (b)
the heading of a table, or (c) a predefined sentence.... You may like
to consider the sentence in some-month the refinery some-name has
committed to schedule some-amount gallons of some-product

2.  If you just leave the "uncommitted" rule in place, and
ask  for an oil production schedule, an explanation shows this
step:
  
estimated demand 523 in NJ is for 1000 gallons of product-y in October of 2005 
in October an order for product-y can consist in whole or part of product-x 
in October the refinery   Shell Canada  One  has committed to schedule 300 gallons of product-x 
we have truck transportation from refinery   Shell Canada One   to region NJ 
__________________________________________________________
for demand 523 for product-y refinery   Shell Canada One  can supply 300 gallons of product-x 


The context and explanation mechanisms are generic, so they also address the general point you illustrate with your example.

The other aspect of course is that anyone with appropriate permission
can edit the rules.  So, there is also a Wikipedia-like social
mechanism in place.  For example, you could now go in and change a
couple of rules so that your "uncommitted" improvement works, and shows
up in explanations.

(End of Adrian's reply to Pat)

The full example is 

    www.reengineeringllc.com/demo_agents/Oil-IndustrySupplyChain1.agent

and there's a paper about it

    www.reengineeringllc.com/Oil_Industry_Supply_Chain_by_Kowalski_and_Walker.pdf

So, the basic point is that if you bring in some appropriate technology
to watch and  comment on what people write, and if you also
exploit the 'social' Wiki  aspect of the Web, then you can to a
useful extent have your NL cake and eat it -- without choking on
ambiguity.

With apologies to folks who have seen some of this before.

                                      
-- Adrian

[IBL]  Internet Business Logic
A Wiki and SOA Endpoint for Executable Open Vocabulary English
Online at www.reengineeringllc.com    Shared use is free

Adrian Walker
Reengineering

On 2/13/08, Jakub Kotowski <jakubkotowski@xxxxxxx> wrote:

John,

John F. Sowa wrote:
> Rob,
>
> This is not an argument about words.
>

<snip>

> I'm not sure what you mean, but I'll state my opinions:
>
>   1. The vagueness and ambiguity of natural languages are inevitable
>      in any robust, flexible system of communication among any large
>      group of people with different goals and backgrounds who are
>      faced with a complex and changing environment.

I would like to add that the ambiguity or vagueness of natural languages
is probably not only inevitable but also a *wanted* and *beneficial*
property as Steven Pinker argues for example in this interesting talk:
http://www.ted.com/index.php/talks/view/id/164 (the core of the argument
starts rougly from the 9th minute).
It is the difference between stating a fact/an offer directly and
bluntly versus saying it so that there are more possible meanings one of
which may not be so offending as the blunt one (example could be trying
to bribe someone).
Do you think that this property could be beneficial for artificial
languages (like those for multi agent systems) too?

Jakub




_________________________________________________________________
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

_________________________________________________________________
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    (01)