Inkwell: Authors and Artists

    
>ckridge, you and I seem to differ on some fundamentals. E.g., I think
the world is fundamentally ambiguous, so clarifying and settling are
often reductions of truth, not enhancements of it. Is this where we
disagree? Or do we in fact not disagree all that much after all?<

Ambiguity is a property of statements. I am not sure how much the
world is like a statement, so can't answer that. Also, if librarians
have learned one thing, it is that we can't wait to find out how the
world is to start sorting information. For one thing, we are going to
need access to the information to find out how the world is. 



>Tagging systems allow us to discover clusters of related terms, not
just synonyms. Lots of sites do this well already, although it's an
area where there's always going to be room for improvement. E.g.,
search on a tag at Technorati and it will suggest "related tags," with
varying degrees of accuracy. The tag McCain" has as related tags:
bush, politics, iraq, republicans, democrats, 2008, republican, law,
torture, and immigration. This is an impressive list, but none is a
synonym. It'd be more impressive if technorati knew what the
relationships were, but I think over time, it will get better at
this.<

Those related terms are almost certainly harvested from records
bearing the tag "McCain." That is an uninformative sort of
relationship, about as useful as Amazon telling you what books people
who bought the book you are looking at also bought. It does nothing to
solve the problem of the information you are looking for being
scattered under multiple synonymous tags, since no one is going to put
two or three synonyms for the same thing on the same record. But
scattering of information is the basic problem that systems for sorting
and finding information are put in place to solve.

>As for synonyms: the ambiguity of language is (imo) its strength, not
its weakness. Oh, at times it gets in our way, but without it,
language is mere codebook. So, yes, we're going to disagree about who
the Americans are. But with tagging, folksonomies, etc., the system can
maintain multiple meanings, and allow each of us (or, better, multiple
groups) to view it in ways that make sense to them. Folksonomies are
not mere bottom-up taxonomies that give us a single way of
categorizing, albeit one from The People. Rather, the value of a
folksonomy is (imo, as always) that it knows not only that 75% of users
think America=USA, but also that 25% don't.<

Yes, but a tagging system is not a language; it is a tagging system.
And when I am looking for information, I don't give a tinker's curse
what other people call what I am looking for. If I did, I would be
looking in a thesaurus, not an online database. I want my information.

Still, something might be done with a system that allowed users to
mark tags as synonymous, and that then presented those synonymous tags
as search options: "Other terms also used for [ ] include [ ], . . . ,
and [ ]." It would be much slower than simply wrapping all the synonyms
into the search automatically, but less subject to abuse.
  

    
ckridge, 

1. I don't think ambiguity is only a property of statements. When we
notice that a statement is ambiguous, the statement has failed,and we
ask for clarification. But, imo, the world insofar as it shows itself
to us is also ambiguous, and there is often (always?) a hidden
ambiguity implicit in language.

The world is ambiguous in that there is always more than one way to
take it and to divvy it up. Language is implicitly ambiguous because
the meanings of words almost always escape their definitions. In fact,
words often have meaning in part because of the other words we didn't
choose.

2. I certainly am not suggesting that librarians hold off on
classifying until we figure out exactly how the world is classified,
since I don't think there is a way the world is classified. We
categorize it based upon needs at the moment, within a cultural and
linguistic context.  

3. As I acknowledged, the "related links" don't yet tell you what the
relationships are. But I think they will move in that direction, using
a number of techniques, including incorporating thesauruses,
gazetteers, and perhaps various ontologies. 
  

    
You must know from your research that there was a long series of
efforts to base library catalogs and subject headings on natural kinds.
They were discarded partly because it is very hard to find out what
natural kinds might be, but mostly because it is not nature that we
keep on shelves in libraries, but books, and what people write books
about has to do with their concerns, not with the order of the
universe.

So I have no problem with arbitrary, multiple, contradictory subject
tags for the same items. I grant that there are useful differences
between the tags "United States/History/Civil War, 1861-1865", "War
between the States, 1861-1865", and "War of Northern Aggression,
1861-1865". I grant that the same song can be tagged "World Music and
Reggae/North African Music/Qwali", "Religious Music", and "Dance
Music".

However, there are two huge problems. First, scattering of information
is not a bug; it is the central problem that any information retrieval
system is set up to correct. Inconsistent tagging, with its inevitable
scattering of information, cannot be ignored.

Second, tagging is deciding what things will be called, and what
things are called matters intensely to people, because what things are
called shapes how they are thought of. People will not tolerate other
people tagging items in ways they don't like. (Try posting videos of
African Americans tagged as "Niggers" on YouTube.) As it turns out, the
least controversial names to give things are those most closely based
on natural kinds. So, even though we cannot base our organizational
systems entirely on natural kinds, we cannot ignore natural kinds
either. If we do, we will inevitably wind up in endless rhetorical wars
over what things are to be called.

An aside. This whole conversation reminds me very much of
conversations that arose when corporations needed very large office
buildings. It turned out that the cheap way to do it was to make big
concrete and glass boxes, the way they had been building factories and
warehouses. Immediately a whole class of theoreticians arose to sing
the beauty of big concrete and glass boxes. As it happens, big concrete
and glass boxes have some virtues, but the fundamental fact was that
no one was going to spend the money to build anything else. That is
basically what is going on here. No one is going to hire enough
indexers to sort out the Web, or even any single free online database.
They are going to make do with the cheapest possible solutions. These
will turn out to be better in some respects and worse in others, but
the basic fact about them is that they will be the cheapest possible
solutions.
  

    
ckridge, thanks for your patience. 

At many sites, there will be explicit or implicit rules and norms
governing acceptable tags, including using words like "nigger" or
"fag." But there will still be a huge range of tags. People _will_ and
do tolerate people tagging items in contradictory way because tagging
is _not_ about deciding what things are called. It's about how
resources will be found. 

There is, of course, inevitably more to tags than that; if I tag a
photo of LA as "smog," I am saying something about LA, although I'm not
saying what what LA should be called. The multiplicity of tags not
only makes resources easier to find, it also provides a layer of
meaning that offers possibilities beyond finding and naming.

Tagging is also not the only way we need in order to find resources
and make sense of them. Even at Flickr, the paragon of tagging, there's
lots of non-tag metadata at work, including names, descriptions,
camera info, date. I haven't heard anyone say that tags are all we
need.

As for natural kinds: Yikes! That's a can of worms. 
  

    
One problem with the multiplicity of tags, though, is that it may make
it harder to find what you want when there are so  many other things
linked to a common tag. 

You need a good boolean capability in your search, so you can say yes,
I want LA + smog but not smog + ozone that might be pulling up all
sorts of climate references when you just want a picture of LA smog.

I love the kazoo search engine for providing diagrams that help
clarify groups of links that go together.
  

    

I've got a slightly contrarian take on tagging. My mom is a retired
librarian and this may be a vestitage of teen back-talk, but my experience
is that tagging introduces new issues of audience and distribution to 
search and navigation, and that those issues are more imoprtant.

I had an argument with Randy Farmer a couple of years ago at Online 
Community Summit when Yahoo had just acquired Flickr.  As a Flickrer, I 
had gone through stages of wanting to maximize page views by maximal 
tagging, to being annoyed by people who seemed to do useless tags, 
to realizing that within my own tag cloud, terms like "me" "old pal" 
or "do over" had meaning that was not going to help yahoo identify 
an image in the context of aggregate tag searching.  I was also 
charmed by the kind of tag abuse that leads people to tag things 
"I love you" or "this is for cindy and spotsy from a fellow cat person." 

He also felt that geo coordinates placed in the tag cloud in a lovely 
user hack to tie to google maps was an abuse of tag space.  However, that 
tool might not have been brought to users without the freedom for 
toolbuilders to "abuse" the space.  We now see mapping (with Yahoo maps 
of course) built in to flickr and classified as another kind of metadata.

I put a lot of work into tagging some of my early shots with similar 
terms and with color descriptors, and then I tired of that.  So out 
there in the millions of sunset shots are a few of mine that say 
"pink" "sunset" "sky"  ...  but since I doubt anybody will ever 
need to find one of mind for those attributes, I'm tending to not 
bother with supercommon tags any more unless that is how I want to sort 
things for myself, or to create a feed for an off-flickr page to display. 
It's partly about distribution -- letting someone who might 
be interested find the image.    

Esoteric tagging can help you find people who know vocabulary or 
cultural references, too.  Tagging San Francisco earthquake related 
images 1906 is not so different than making up event tags like ocu2007.

I'm more interested in connecting, play, self-organizing and 
self-expression via tagging than in making a grand index of reality.  

I have come to believe that most of the time, reality wants to be messy.
Reality also wants to be ordered, but far less often.
  

    
Gail, I'm so infatuated with the messy that I both agree and disagree
with you, simultaneously, and in the same way:)

On the one hand, I'm sympathetic to the "selfish taggers" who say a
tagger gets the best results for herself by tagging without regard to
how others might search for the resource AND that selfish tagging
results in the most useful folksonomies as well. OTOH, there are times
when "altruistic tagging" makes sense, leading me to a big It All
Depends sort of conclusion. Overall, I'm hoping we can just plunge
ahead messily and inconsistently, and let the computers sort it out.

Of course, that too depends on the domain. Tagging flickr photos is
one thing. Tagging your own photos is another. And tagging classified
dossiers is another. How much altruism, how strongly we want to enforce
norms and vocabulary, what type of searches we'll allow all depend on
what we're trying to do with the tag set.

And -- messier still -- people are doing different things with the
same tag set, and they're mixing a tag set created for one purpose with
tag sets created for other purposes. So, in the end messiness wins,
because the sum of human ends and beginnings is overwhelmingly messy.
  

    

It strikes me that there are a number of ways current sites, such as Flickr
could improve the interaction associated with tagging.  It could also be
done via browser plugins, and that would be even better, since it would
allow more and better altruistic tagging everywhere.  And one key to getting
better value from keyword tags, is to have many people assign them to the
same subject, file, object, etc..

Today, tagging and the means by which it's enabled, remains fairly
primitive.  Essentially a blank field, where a person enters a keyword (most
often the uploader, I would guess on sites such as Flickr, even though tags
can be entered by anyone on any image).

One reason I don't always feel, well, right entering tags on other people's
photos is that they're treated and presented with such a preciousness.
They're right there.  There's not room for hundreds.  But if it were
possible to enter keyword tags in a way that might benefit the tagger later,
and maybe not end up burdening the up-front view on the page, it seems more
people might do it.  There are undoubtedly some who would do a lot of it
(look at the super-editors at work on Wikipedia).  People might be able to
generate their own authoritative keyword tag databases (and these could be
further compared, averaged, and computed upon).

But let's also say that the entry field, as soon as text was entered (or a
recognized word), invokes a nearby pop-up pane listing columns of
alternatives.  Each column may represent disambiguations for the entered
keyword, and consist of ordered or popularity-ranked alternatives.

The user wouldn't be forced to choose one, but could see alternatives, in
case that mattered to them. (perhaps they were just making best guesses
anyway, and this showed them the most common similar keywords).

Clicking on one would enter it, thus streamlining things.

The thing is, if more people could be enabled/persuaded (perhaps by
convenience, or for value to their own future searching of things they've
come across previously) to altruistically enter keyword metadata, then
accumulations would insure greater accuracy.

There are ploys out there I've seen that play on this.  What's the site that
uses the game of trying to get two people to guess the same keyword for
images that they're shown?  Things like that use multiple keyword guesses to
generate higher accuracies (over time).

Such methods can also over time generate multiple keywords or keyword groups
representing different sets or tag clusters of common assumptions or
interpretations for the same image.  From that, higher orders of context can
be discerned.
  

    
One of the major problems with tags on Flickr (and similar sites) is
that there's a single global namespace, and no concept of scope or
public vs. private tags. If there were separate namespaces you could
be as selfish as you want in the private namespace, and as altruistic
as possible in the public. For example, I tag all my Flickr photos in
which I actually appear with "mcb" [my initials], which is very useful
for me, but obviously of relatively little value to anyone else who
happens to do a global search for "mcb". And with some sort of
intermediate scope, I could tag other peoples' photos in which I
appear with "mcb" as well, without polluting the global namespace. 
  

    
David, Great discussion so far – sorry I’ve missed the first part, but
I’ve been swamped with doing two enterprise taxonomy projects – which
gives you an idea of my background and bias.  

Forgive me for some late general comments, but I can’t resist.  We met
briefly at the Asis&T conference in Vancouver last year where you gave
the keynote on Everything is Miscellaneous and I have to admit I was
struck by Lou RosenFeld’s introduction when he couldn’t resist pointing
out that your talk was organized into three sections not just randomly
put together.  It was (as your book) well structured and presented –
the opposite, in fact, of messiness.

If I put you on one side of the order-messy spectrum and say Claude
Vogel with his call for large formal and standard taxonomies on the
opposite side,  I used to be closer to his side, but have moved away in
recent years.  To oversimplify, for me it is not an either/or between
a single formal classification scheme for the universe of discourse and
a let chaos reign messy free for all, but the real question is how to
create structure that supports real findability (something that
folksonomies are not really good at) but also supports and builds on
individual differences.

In #26, you talk about the limits of taxonomies including the use of
professional classifiers deciding what something is about.  Admittedly
the world inside the firewall (Intranets etc.) and the Internet call
for different balances between classifiers and social tagging, still my
experience is that there is simply no way to get rid of dedicated
classifiers – think Wiki rather than folksonomy.  

For example, a model that I’ve used for clients is a hybrid
distributed model with individuals tagging their documents by selecting
from controlled vocabularies not just tagging out of their heads – but
with a mechanism for suggesting new tags that are not in the
controlled vocabulary.  These tags are supported with smart software
that can also suggest tags (like an Inxight).  A second tier is added
with designated subject matter experts who review tags for their areas
and flag problems and communicate with individuals and with the third
tier, professional classifiers who deal with more systematic issues.

The role of classifiers in this distributed system is first to create
the taxonomic controlled vocabularies with lots of input from
individual subject matter experts and then to also review the tags both
from a tagger perspective but also monitor how the tags are being used
to find information.  They are charged with maintaining the taxonomies
but with a lot of input from all the community.

One final general point and then I’ll switch to some specifics.  You
say in #53 that there is no possibility of ever getting everyone to
agree on the categories that we want to use.  I agree, but it is also
true that we never have a situation where we can get everyone to
disagree on what categories to use.  IOW, it is not simply a messy
situation with everyone tagging differently from everyone else, but
rather there are clusters and communities of like-minded taggers who do
see the world in largely the same terms and while it is important to
build on the variety of perspectives on categories, it is just as
important to build on the similarity of perspectives on categories. 
Something my team has done by mapping formal and informal communities
within an organization as part of an enterprise taxonomy project – to
capture the significant differences and build them into the taxonomy.

More later – but this is a great discussion so far –as a late comer, I
hope it continues for a while.
  

    
On folksonomies, I’ve done some research on this (see some
presentations on my company’s web site (www.kapsgroup.com) and I have
to admit I’m less than thrilled with their performance.  First of all,
there is very little “onomy” in folksonomy and to compare them with
taxonomies is very misleading.  To date, folksonomies are simply flat
lists of keywords with no semantic relationship between terms and all
the explanations (including some of mine) of how you might develop more
semantic structure have a quality of the old T-shirt – and between
step A and step C, a miracle occurs.  And, of course, true folksonomy
advocates (and they really do tend to be impassioned advocates) argue
that even the attempt to add structure to folksonomies is wrong-headed.
 The power of folksonomies is in their lack of structure in this view.

This is because, to me anyway, the whole focus of folksonomies is in
the ease of tagging documents and photos and here it is certainly true
that having no structure makes it easy to tag.  But I have to ask, what
is the point of adding tags to documents?  Not to make it easy to tag,
but to make it easy to find things that have been tagged – and that is
where folksonomies have some real problems.

But the argument goes, true there is no semantic structural
relationships between terms but there is a structure – it is a social
structure.  You get to see how others tag and with the use of
interfaces like tag clouds you can see popularity which is the heart of
the folksonomy structure.  

The problem I have with that is two fold.  As the size of the content
domains grow there is the issue of too many things tagged with the most
popular tags which makes it harder to find individual items in the
mass and on the other hand, you have the large mass of the “long tail”
of thousands of infrequently used tags which make finding anything
equally difficult.

However, I do find folksonomies very interesting as starting points. 
First, if you look at a variety of tag clouds on Flickr, you see that
most tags are words that are what are called basic or natural level
categories.  These are terms that are intermediate within a hierarchy
and are particularly powerful terms for ease of use as well as
discriminatory power.  An example is dog (a very common tag) – this in
intermediate between say  - mammal and retrievers.  These basic level
categories become great candidates for developing structures that do
capture semantic or formal relationships of terms.  The issue then is
how (and who) to build on this.  In the commercial world, it is
actually rather easy to set up a distributed system, as described in
the last post, of  individual taggers and professional classifiers to
develop structure.  Something I think we might see more of on Internet
sites like Flickr as well?
  

    
Oh, and <ckridge>, I knew there was a reason I liked your comments in
the Future conference - you're a librarian!  My company hires a lot of
people with librarian backgrounds, but I find we usually need some sort
of mix of library science and cognitive science to deal with the messy
as David calls it.
  

    
Google has a panel of doctors who categorize links for search results.
Search for 'migraine' and you'll see results refined by category.

Amazon has Mechanical Turk.

Both provide classification by the community.  I can see how you might
organize a speech on miscellany into three sections, even if it is about
miscellany.
  

    
I think the reason tagging has taken off for photos is that there's
such a clear distinction between data and metadata.  Photos have no
words, so we attach a few words to them, or we'll never find them
again.  

Adding metadata takes effort and there's a continuum based on how much
effort you want to put in.  You start by adding a few tags, and for
some of the better photos you might write a proper caption, and then
for the really famous photos, the caption might turn into a
collaborative Wikipedia article describing who took the photo, its
historical meaning, and so on.  As you add more words and structure
them into meaningful sentences and paragraphs, with links to other
articles, you're giving search engines more data and allow for phrase
searches, and eventually they will be able to take more advantage of
sentence structure to determine relationships between the words and
therefore the relationships between the photos.  And just the level of
effort that you put into describing a photo is itself meaningful.

Maybe there's a point along the way where you stop auto-linking words
and make them explicit links, as you do in a Wikipedia article, or else
you've got lots of irrelevant links.  The data isn't lost because you
can always do a search on that word.

And of course, there is the Photosynth way of doing it, which is to
directly relate the photos as images.  But even if we end up flying
around a Photosynth world, I think attaching words to images is still
going to be pretty important, just like labeling maps remains important
when you're using Google Maps or Google Earth.
  

    

I think it's never ever an either/or situation with these technologies.
It's all additive.  In fact, we'll eventually have a multitude of ways to
sort through and get at data/files/media, etc..

Again, with separate and independent metadata sources, you could query using
specific metadata sources.  Say, highly-reputable or highly-ranked metadata
providers/sources.  These coulde be individuals or aggregated third-party
sources.

It's not about a single solution.  It's about layers upon layers of
solutions.  Same goes for the layers of the presentation and potential
interaction with that.
  

    
Yeah, I definitely agree with that.

treamy, I never supposed that tags were supposed to make specific items more 
findable; I think it's more valid to say that tags facilitate discovery and 
association. If want to find a specific item, I'm not going to follow a trail 
of tags to get there; search is far more effectivce in getting to specifics.
  

    
Now that would be interesting. A way to enable other people looking at
your photos to note what tags *they* would have used on them.
  

    
treamy, thanks for all the comments and thought. But let me be very
clear: My book does not argue in favor of chaos and lack of order. Nor
does it argue for folksonomies to the exclusion of taxonomies. Rather,
I'm trying to make the point that the "miscellaneous" has value because
of its potential for multiple, simultaneous orders. The miscellaneous
never stays that way because we are always organizing, clustering,
categorizing, etc. But, in the digital world, the organization is of
metadata, leaving the underlying data untouched. And because frequently
those digital orderings are public, in finding/creating order in the
miscellaneous pile, we are in fact adding to its potential for
order...making it messier _and_ richer in potential.

Taxonomies have their place. Wikis do, too. So do folksonomies and
curated folksonomies. So do faceted classifications, alphabetization,
and photosynth image clustering. It all depends on what you're trying
to do and what the nature of the data is. (I am thus agreeing directly
with jleft in #115 and others of you in various posts, implicitly.)
  

    
Shaku Atre gave a speech entitled "A Database is a River" some years ago.
She talked about what it was like to live along the river.  It depends where
you live along the river.

We've got more texture in the river now and more ways to express meanings.
  

    
treamy, thanks for all the comments and thought. But let me be very
clear: My book does not argue in favor of chaos and lack of order. Nor
does it argue for folksonomies to the exclusion of taxonomies. Rather,
I'm trying to make the point that the "miscellaneous" has value because
of its potential for multiple, simultaneous orders. The miscellaneous
never stays that way because we are always organizing, clustering,
categorizing, etc. But, in the digital world, the organization is of
metadata, leaving the underlying data untouched. And because frequently
those digital orderings are public, in finding/creating order in the
miscellaneous pile, we are in fact adding to its potential for
order...making it messier _and_ richer in potential.

Taxonomies have their place. Wikis do, too. So do folksonomies and
curated folksonomies. So do faceted classifications, alphabetization,
and photosynth image clustering. It all depends on what you're trying
to do and what the nature of the data is. (I am thus agreeing directly
with jleft in #115 and others of you in various posts, implicitly.)
  

    
tide pool alert
  

    
Wow,

Thanks Jon and David for having this discussion, and to everyone who
has participated so far.

I've been struggling with trying to word an observation, and the
conversation just keeps plowing ahead, so I'm breathless as I catch the
end of the trail of posts.

There are many books of late that point to the ways in which digital
information aggregation and communication is changing our perspectives.
I've tried to keep up, and a few of those books really stand out for
me. As I read _Everything is Miscellaneous_ (I've still got about 50
pages to complete it), I'm continually reminded insights that I had
while reading Jeff Hawkins' _On Intelligence_. 

As I continue to read these books, including _Everything is
Miscellaneous_, I think of how the internet is making possible a
virtual  cortex with extended memory and linkages, similar to neurons
in the brain. This idea isn't new, but I haven't seen very much
relating this  possibility to the idea of miscellany.

1) From Hawkin's book, we see that the structure of the cortex into
layers of neurons facilitates a kind of invariance which allows us to
derive meaning. (from _On Intelligence_)

2) From a talk on artificial intelligence, and artificial vision
systems, I learned that visual information isn't simply passed from the
eye to the brain for interpretation. There is an extremely complex
feedback infrastructure, such that each pathway from the retina to
brain neuron has 10,000 pathways feeding back to the retina, affecting
the sensitivity of receptors based on possible interpretations of what
is being perceived. It seems that perception is largely based on
emergent resonance patterns rather than simple lookup.

3) The idea that our mental processes represent a myriad of
perspectives all vying for dominance of conscious attention (is
Minsky's _Society of Mind_ an example of this?).

What I come to from all this is wondering if the messiness is
intrinsic, and that we are only recently becoming aware of this
necessary and intrinsic nature. The human dominant fascination with
order and control would have us believe that we can eliminate the
miscellaneous (when I hear about "junk DNA" I'm afraid some group is
going to try to figure out how to clean it all up of all that
messiness). I can't help think the same tendency is at play when we
want to clean up the messiness of folksonomies. Yes, certainly,
folksonomies are the end-all, and we shouldn't try to eliminate
taxonomies either, but find new ways to bridge the span between the two
extremes.

In thinking about this, just imagine if neurons in your brain had to
be as discerning with meta information (scrutinizing on the way in,
rather than the way out). We'd live a very impoverished mental life.
How do we open up the floodgates without drowning in meaning at the
scale of society? There are some interesting developments technically,
but they all seem just the beginning. The photosynth demo is pretty
awe-inspiring in this regard.
  

    
check out women in art  <http://youtube.com/watch?v=nUDIoN-_Hxs>
  

    <scribbled by jleft Sat 9 Jun 07 22:28>
  

    

All of this also reminds me of Bart Kosko's excellent book, "Fuzzy
Thinking," which is about Fuzzy Logic and its applications, but is also
about the fundamental difference between the Western/Aristotlean bivalency
of black and white vs. the Eastern/Buddhistic fuzziness of the grey that's
what actually exists in nature between on/off, black/white, 1/0, etc..

The idea that "messiness" is not truthful/accurate/valuable strikes me as
essentially a Western mindset judgement.