Re: [decentralization] Re: Microsoft improves SmartTag feature (let's get back to Decentralization talks here)...
> This is the one feature I'm quite puzzled about. Installing a Smart TagYep.
> list seems to involve getting a file onto the users hard disk. Which
> means an install routine.
> There's was one rumour that the List file could be embedded or attached toa web page.
You can embed SmartTags in a page, but those tags will only work for that
> Can anyone confirm this? The whole feature would actually beThe SmartTags spec only provides for the interface to the browser. The
> considerably more powerful if it was slightly less decentralized and had
> more of an element of auto update about it.
developer is free to extend the backend any way he pleases. There are two
ways to implement SmartTags- one with an XML list of terms and actions, the
other with a DLL that searches each word dynamically. _What the dll uses to
search that term is up to the developer_. IOW, your SmartTag recognizer
could very easily search for matches in a database on another machine
(performance would be quite a factor here). Probably a better way to do it
would be to have a local database updated periodically and automatically,
and searches could be performed against that database.
The XML file method does provide for automatic updates by allowing the
developer to specify some optional tags in the bottom of the file that
specify a location to retreive updates from. From what I understand the
browser takes care of it for you.
> Conceptually this is quite like the problem of a spelling checker havingSort of. But if you're the one writing the spell checker, you can store the
> to have it's dictionary installed locally. Except that the dictionary
> data is more dynamic.
dictionary wherever you can reach it. SmartTags lets you do this...
> AI only works (somewhat) within the narrowest of subject matters, andafter considerable efforts teaching the computer. A client-based, generic
text interpreter, able to 'understand' what the text is about and make
useful decisions about what would interest a reader is totally unfeasible at
this point. If MS thinks otherwise,they simply haven't done their homework.
We don't ask so much out of our search engines, what's the use in holding an
automatic annotator to a higher, unachievable, standard?
> We don't ask so much out of our search engines, what's the use in holdingWell, there isn't, except for the fact they are called 'Smart' and the fact that
>an automatic annotator to a higher, unachievable, standard?
Microsoft claims/hopes they will provide useful information to the reader.
This morning I got the example of reading the word 'Madonna'. I imagine
seeing 4 or 5 links (not 200) to related information; but we need the client to
understand which Madonna we are talking about, otherwise the feature is
useless and laughable (and could be implemeted today, cross-brand, cross-
platform, with backwards compatibility to browsers version 4)
If the hit rate (chance of a useful annotation) is going to be similar to that of
the average search engine or spell checker getting it right, I think we should
- > > We don't ask so much out of our search engines, what's the use in holding
> >an automatic annotator to a higher, unachievable, standard?
> Well, there isn't, except for the fact they are called 'Smart' and the fact that
> Microsoft claims/hopes they will provide useful information to the reader.
So, get some smart developers.
For example, a SmartTag DLL which finds context from your Groove spaces? An example of the general "context lens" . (Hmmm. I might just have to do that!).
>from the linked article:I like the idea, but I think you underestimate the tremendous difficulty of
>>Imagine (for a second) a "context server" - a SOAP-based Web service,
>>say - which returns you useful links when you send it a word, phrase,
>>paragraph or URL. That's how I imagine SmartTags works
returning truly useful links to each indiviual user (which my original post was
As I stated before, limiting the 'subject matter' (like in a corporate context
server, which is a great idea) can make the service more useful. It will also
move much more control towards the author of the document, away from MS,
which is a good thing.
But I'm not sure Microsoft's spec will allow the author of a document to
specify their own context server (will it?). And this will most certainly not be
the case for the web of today, which has no ST-related directives at all and
seems to be totally at MS's mercy.
- Decentralisation: the most useful context is local.
A Groove- (or Jabber-)integrated SmartTags DLL is quite a serious
suggestion; the context is really local (eg. just to highlight people's
names, by looking into the local contact list, and add a "chat" action to
them), plus Groove is a smart vehicle to deliver DLLs to the desktop since
it already includes all the version management stuff you need to do that
Building a SOAP-driven "context lens" is quite doable too, by building a
SmartTags DLL which knows how to talk to your context server (and which lets
the end-user choose their own context server/s).
But, the proof will come when several of these beasts are actually deployed.
I can't wait.
- Do you know is funny about all this talk of Smart Tags? The idea was stolen
by MS - nothing unusual there. (Personally I doubt they have much really
original) There is a company that has been around for at least a year that
already has most of this technology. The way you activate it (and I don't
mean install the software), is to place the cursor on the word and hit the
"magic keys" (you assign the exact magic key sequence). They go out, pop up
a little window, giving you access to the definition, etc. It works very
----- Original Message -----
Sent: Tuesday, June 19, 2001 10:59 AM
Subject: [decentralization] Re: Microsoft improves SmartTag feature (let's
get back to Decentralization talk
> >from the linked article:
> >>Imagine (for a second) a "context server" - a SOAP-based Web service,
> >>say - which returns you useful links when you send it a word, phrase,
> >>paragraph or URL. That's how I imagine SmartTags works
> I like the idea, but I think you underestimate the tremendous difficulty
> returning truly useful links to each indiviual user (which my original
> As I stated before, limiting the 'subject matter' (like in a corporate
> server, which is a great idea) can make the service more useful. It will
> move much more control towards the author of the document, away from MS,
> which is a good thing.
> But I'm not sure Microsoft's spec will allow the author of a document to
> specify their own context server (will it?). And this will most certainly
> the case for the web of today, which has no ST-related directives at all
> seems to be totally at MS's mercy.
> To unsubscribe from this group, send an email to:
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
> document to specify their own context server (will it?). And thisThe smart tag architecture is agnostic on this for DLL-based tags.
> will most certainly not be the case for the web of today, which has
You could write a smart tag that honors some specific "context
server" information embedded in a tag. I like the idea of context
provided by site, and also (to me, even cooler) context provided
by "community", such as groove. Context provided by site is a fairly
well-solved problem (that is sort of what <a href> is for, right?).
The really fascinating application would be to have the option to
shift contexts. For example, today I want to browse the news, only
seeing opinions from people who think similar to me, but tomorrow I
want to browse the news and see the viewpoint of people who have
completely different perspective than me. Groove and such are good
ways to find people that are likely to have compatible metadata, but
finding people with opposing perspectives is trickier (for example, I
can go to slashdot for opposing perspective, but in many ways these
people's background is very similar to mine [all computer people,
mostly male, etc.]). As for *automatically* finding people with
similar and dissimilar perspectives, I believe that some simple
commodity data-mining/clustering algorithms could do a fantastic job
of this. One could probably build a working prototype in just a few
This, IMO, is something that *must* be done eventually. This
effectively solves (well, almost) the problem of someone "poisoning"
metadata. This is not a smart-tag specific issue, but applies to any
system that shares metadata (for example, song titles on Napster are
metadata). So suppose I am a performer who wishes to have my song
appear really popular. I could write (or have someone write) a
script that figures out which songs are the most popular, then create
a few thousand fake users who have these songs (all pointing to the
same copy on my HD of course). Then I just add all of my songs, so
that people who want to see what "other" Brittney Spears songs I have
will also see my band name associated with all those other "cool"
bands. I am sure others would be able to come up with much more
devious tricks, but you get the point. In fact, this is something
everyone seems to have missed in the talk of a powerful interest
controlling the metadata technology and thus controlling the
metadata -- in fact, a powerful pro-gun lobby could easily hire
people to hang out on ZDNet forums and overwhelm things with pro-gun
opinions. The point being that a powerful interest can poison the
metadata with or without having control of the metadata distribution
technology. In fact, people could complain that google has power to
shape public opinion, but the people we are *really* worried about
are the ones who create artificial crosslinks everywhere to "trick"
google into giving them higher rankings. So the solution for google
is the same as is needed for any other metadata sharing system -- a
way to detect anomalies. Simple commodity data mining algorithms can
quickly find things that don't "fit the curve", and the NRA plot is
exposed. When you cluster the metadata, the poison area simply
becomes one "cluster" sitting off by itself. In fact, the more
poison metadata gets pumped out, the more pronounced becomes the
Note I am not saying this is a panacea, and there are difficulties
doing this with certain types of metadata, but this would be a huge
step forward IMO, and the technology for it exists today.
BTW, just so nobody accuses me of trying to deliberately mislead, the
simple "list-style" smart tags allow context to be specified only by
the tag author, I believe. I am not aware of any secret smoke-filled
meetings that decided to limit it this way, and I am not aware of any
opposition to adding more options in the future.