Re: The author of a query is most likely the user

On Jul 3, 2005, at 3:02 PM, Adrian Walker wrote:

> Hi Bijan --
>
> Now that the long weekend is upon us (in the US of A), there's finally  
> time to study the online post-rules-workshop  discussion. (  
> http://lists.w3.org/Archives/Public/public-rule-workshop-discuss/ 
> 2005Jun/ )

I suppose I should join that.

(Grumble grumble...yet another list. I don't know why it bugs  
me...presumably the traffic will be the same.)

> I'd like please to take you up on what may seem like a small point,  
> that you made a while ago, and argue that it is very important.

Sure, but wouldn't this be better done on the www-rdf-rules or that  
other list? I've cced rdf-rules, so why don't we follow up there.

> At 02:03 PM 6/22/2005 -0400, you wrote:
> ...the author of the query is most likely the person evaluating the  
> query
>
> Depends, I suppose, by what you mean by "evaluating".  Yes, a  
> programmer will test the query after writing it.  But of course, in  
> the real world, almost all the people -- or machines -- that run the  
> query will _not_ be the programmer who wrote it.

However, it seems likely that it will be in the context the programmer  
knew and set up or otherwise could control or substantively influence.  
E.g., a website. Or a CMS.

The real point I was making is that it doesn't seem usual that queries  
will be aggregated and combined in arbitrary ways (though the *results*  
of such queries might be).

IIRC, this was in the context of why certain operators or semantics  
might make more sense in the query language (or in rules, e.g., default  
rules) "outside the data".

I'm not convinced of this general argument for the open world  
assumption. Just trying to articulate it here.

> So, there is scope the size of the Grand Canyon for a user to  
> misunderstand what the programmer had in mind.  And the data  
> underneath the query may have changed since the query was written, so  
> the programmer's testing may anyway have been incomplete.

Seems true today on the web and, strangely, not a big problem. I  
daresay it will stay that way for the majority of end users. Most  
"queries" will be either predefined web pages, or pre defined web pages  
with some customization (i.e., web forms). And users will trust the  
queries the way they trust any piece of software or website.

> On the plus side, we can write rules in lightweight, executable  
> English, eg as in [1], so that English explanations can give each user  
> a pretty good idea of what is going on, right down to the data.

I'd love to see user studies backing that up. Pointers, please?

> Reasoning chains over a static database quickly get too complex for a  
> programmer to manually check the relation to a business or scientific  
> requirement [2,3], so I'd argue that it is even more important for a  
> rules language to document real world concepts -- and to do so  
> executably rather than in comments -- when we move to reasoning over  
> the Semantic Web.

I don't know why it's so much harder than for complex software in  
general. Executable specifications are in the minority of formal  
specifications are in the minority of specifications, as far as a I  
know. "Executable natural language" in the minority of all that.

Wow, I'm making a popularity argument. Whoo, how the mighty Smalltalker  
has fallen :)

The point is that it seems like you could run this argument against  
XQuery and Java and SQL without modification. Even if your approach  
were superior, it doesn't seem compelling. So I remain unmoved.

That being said, I would love to see those studies. My personal bet is  
that people will just tend to trust things unless they are a developer  
who needs to debug a problem. (Hence my focus on explanation in support  
of <http://www.mindswap.org/2005/debugging/>.)

> Makes the Semantic Web more Semantic (:-).   What do you think?

I saw an article claiming that business rules tend to be specified in  
natural langaugey-esque languages. Don't know how true that is.

I think the problems of HCI wrt KR languages are hard ones and  
gathering evidence is difficult especially as they segue into  
programming language level complexity (if you have access to studies on  
the actual effectiveness of different programming language features,  
I'd love pointers there; the ones I've seen are rather inconclusive  
overall; the field is not mature).

Cheers,
Bijan Parsia.

Received on Monday, 4 July 2005 06:37:02 UTC