I think there's a fallacy of a zero-sum game here. "If we don't get
absolutely right the first time, we've failed completely". This is a
process that is going to require ongoing development. Also,
everyone's catalog is limited. Even your very large union catalogs
aren't terribly large in Google-ian terms. But they are also edge
cases. Since majority of catalogs are less than a million titles (at
least, I think this must be true), it shouldn't be that difficult to
generate a system that knows how to find things in it. It can also
'learn' from the actions and mistakes of others.
I guess I'm just not all that concerned with worrying about 'edge
cases' until I see that users aren't getting what they want via the
'mainstream method'.
I think one of the reasons we're in the sorry state of automation that
we're currently in is this need for 'perfection across all cases' by
version 1.0. And we never get either.
-Ross.
On 6/12/06, Bernhard Eversberg <ev_at_biblio.tu-bs.de> wrote:
> Ross Singer schrieb:
> >
> > Well, right. Learning opportunities should be limited to "his name is
> > actually spelled Thomas Friedman, not Thomas Freidmann" -- because this
> > would be an issue outside of the catalog -- as opposed to "when
> > searching for authors please Friedman, Thomas". That, quite frankly, is
> > the kind of lesson that can be done away with since it's learning how to
> > use a poorly designed technology.
> >
> You are mentioning only one of the simplest aspects here.
> What must be learnt is, I think, that as a catalog or search engine
> user you're dealing with a machine that handles character strings,
> not meaning, not questions, not intentions, and what it brings up are no
> answers. While sometimes it will display something that matches your
> intentions, it can utterly fail on other occasions. This can become
> apparent much sooner when searching in non-English languages with all
> their beauty of inflections. Technology might do more here, but only
> if we had better metadata - for example, a language indicator with
> every title.
> But Google goes a long way without any metadata? Well, look a bit closer
> at what it does and doesn't do, what it can and what it cannot possibly
> do. Will be helpful.
>
> What, OTOH, *should* catalog users learn in order to be successcul
> users? Nothing at all? "You need not think!" above the input slot - is
> that the ultimate ideal?
>
> Asking a machine *is* vastly different from asking a human, there's
> no way around that, and particulary, if what you toss into the slot
> is not actually a question but a mere two or three character strings.
> What is it that people now do learn when using Google, and is that
> enough, can it be enough, ought it be enough, for catalogs? Is it
> enough, coming to think of it, for Google itself? If not, what is
> it, and should libraries bother?
>
> Is it possible these days, btw, to say, ok, my aim is to play the
> Mondschein sonata, but don't bother me with learning scales first
> and chords! But well, that's got nothing to do with catalogs, they
> can have none of that complexity.
>
> Regards, B.E.
>
Received on Mon Jun 12 2006 - 08:09:52 EDT