Re: soundness for RIF

Sandro Hawke wrote:
> I was asked to clarify my use of the term "sound" in the context of RIF.
>
> This came up because I proposed a requirement "Sound reasoning with
> unknown dialects", or more fully, "RIF Core must allow sound reasoning
> with unknown dialects." [1]
>
> I think this is one of the main requirements which will constraint the
> extensibility design.  The application context I'm imagining is use case
> 8, a query-answering process using deduction rules (database views)
> coming from a variety of sources.  My requirement here is that I be able
> to incorporate rulesets which include unknown extensions, and to know
> that even in the worst case I will not get wrong answers.
>
> In other word, I might incorporate a ruleset that includes this rule:
>
>     phoneNumberOfAssistant(Boss,Number) :-
>         assistant(Boss,Assistant),
>         phoneNumber(Assistant,Number).
>
> and also 
>
>     assistant(Boss,Assistant),
>     phoneNumber(Assistant,Number) :-
>        phoneNumberOfAssistant(Boss,Number).
>
> The first rule is a normal Horn clause.  The second is not.  I'm not
> sure what it is, really.  :-) And I don't want my program to have to
> know either.  I want it to be able to safely include the ruleset
> containing both these rules and make inferences which will be sound,
> even though it has no idea what the second rule is supposed to mean.  It
> will certainly be incomplete because of not knowing what the second rule
> means, if it wasn't incomplete already.
>   

Regardless of what Francois thinks "sound" means, this argument of yours 
above seems to make assumptions about monotonicty and negation.  What if 
e.g. the rule you understand expresses a default and the rule you don't 
understand expresses an exception?  You may draw a conclusion that is 
incorrect, i.e. unsound wrt the semantics of the unknown dialect.

-Chris

> I believe this use of the word "sound" is normal and correct for logic.
> I'm asking that the conclusions of an inference procedure follow
> logically from its premise.  The twist here is that the premise is
> written in RIF and may have parts which are opaque extensions as far as
> the inference procedure is concerned.
>
> Please note also that I understand many people invision using RIF only
> through translators, but I don't think that affects this discussion at
> all.  For my purposes, the Translator+InferenceEngine can be thought of
> as an InferenceEngine which operates on RIF.
>
> Also, I understand that I'm only thinking about deduction rules here,
> but I think this requirement has an extension which covers production
> rules as well, I'm just not as clear about how to state it.
>
> There are negation/close-world situations in which this requirement may
> not be attainable.   I'm concerned about that.   Perhaps these cases can
> be somehow identified, so that users can be warner or automated
> processes can avoid the unsound-in-this-situation constructs.
>
> Does that make sense?   
>
> Is this a requirement you would support?
>
>     -- Sandro
>
> (I believe this e-mail addresses action items 6 and 7)
>
> [1] http://www.w3.org/2005/rules/wg/wiki/Sound_reasoning_with_unknown_dialects
>
>
>
>
>
>   


-- 
Dr. Christopher A. Welty                    IBM Watson Research Center
+1.914.784.7055                             19 Skyline Dr.
cawelty@frontiernet.net                     Hawthorne, NY 10532
http://www.research.ibm.com/people/w/welty

Received on Thursday, 11 May 2006 00:58:47 UTC