Category: Semantics

Recovered from the Wayback Machine.

Interrupting the play I’m having with the RDF Poetry Finder essays to see what others are saying, and also to add some geek stuff so people know that there really is a string at the end of this particular balloon.

Joseph Duemer (and Frank Paynter, indirectly) expressed some concerns about the image=abstraction view of poetry, suggesting instead a possible alternative:

 

I would tend to think that finding ways of clustering related images around themes or subjects might be a more better way to proceed. This whole project plunges us into the depths of cognative science & the ways in which we can model human consciousness.

This opens up a new way of looking at the core of the system, or perhaps more accurately, looking at the core of what is poetry. I am very intrigued. I am also out of element, not being a poet and only recently becoming a poet enthusiast. My hope is that Joseph, and the other poets/enthusiasts in the audience will expand on this fascinating view.

I also agree with Joseph when he wrote I salute Bb & will happily join her project, but, boy, this is a huge undertaking!. This is a huge undertaking, but not necessarily a new one, or even an impossible one. And this is where today’s bit of geekery enters the picture.

Years ago when I worked in the Acoustical and Linguistics group at Boeing, one of the projects the group was working on was the idea of a concept search engine that could be used foir more precise searching in a massive database of stored documents. We lost our funding before we could explore this concept beyond our simple prototype, which was an intelligent front end to the company’s data dictionary.

At the time, there was no Web, and there was no XML or RDF or any of the tools, technologies, and specifications available today. We take for granted the Web, XML, and even RDF, but it’s amazing how much these technologies can simplify a search of this nature. Compared to the technologies we had back then, in the late 1980’s.

I wasn’t one of the brains in the group — my strengths were in taking their efforts and finding practical uses for same. However, right from the start I could see the power of a concept based search, one in which a person can search on an idea or a thought or a need, rather than search on keywords and exact phrases:

“I need all documents focusing on the cost of stress testing compared to the payback of same.”

ALIA’s efforts were focused on finding automated ways of performing these searches, using fairly complicated heuristics and rather intimidating technologies such as neural networking. Heavy stuff. And our organization wasn’t the only one interested in this — just search on concept based search engines in Google to find several references (here, here, here, and so on).

[I was particularly intrigued by Joseph’s discussion about “clustering related images about themes or subjects”, because concept-based searching is sometimes referred to as cluster based searching (here and here and here, here, and so on).]

Of course, we now know that completely automated systems to support concept-based search aren’t really the answer. Concept-based search engines are built on the premise of a partnership between human and machine, between the synaptic and the digital. We have the technology, we have the expertise — we just need to find a way to bring it all together. That’s what we’re doing now, here in this weblog and at the Renaissance Web dialog group: ad hoc discussions bringing together the poet, the enthusiast, and the technologist to explore ideas, test out possibilities without necessarily restricting the discussion this early to specific implementation.

Still, there is one technical implementation issue that can be discussed separate from the heuristics of the search itself: how does one take data from a closed system using RDF Poetry Finder to a more global one?

For instance, we’ll implement an ontology, an architecture, and even open APIs (Application Programming Interfaces) based in different programming languages to support a semantic search engine for poetry and we’ll provide it free of charge to sites such as poets.org and any other poetry-related site that wants to use it. We’ll also give it out to webloggers for use at their own sites.

Hopefully with our help and encouragement, these sites will begin to incorporate this technology, providing semantic search capability for their own needs. In addition, though, they’ll also be generating data that can be accessed and consumed outside of the closed system.

This is doable, practical, and even has precedence in previous technologies. The annotation of the semantic markup for each poem will be based on an ontology created using the W3C’s OWL (Ontology Language), itself based on a universal model of semantic data, RDF, and serialized to a file using a universal markup, XML. Accessing a poem’s semantic markup will be no different than accesssing, and processing, my RSS 1.0 file — except that the structure of the data might be a bit more complex, and the information in the file persists, though it will change over time (as new interpretations of the work are incorporated).

Today’s technology used for today’s needs. Have no doubts this can be done. In a closed system, such as in use at poets.org.

However, how do we globalize this system? We have access to the RDF/XML files containing the semantic markup, and it’s based on a universally used ontology — anyone with an RDF parser can access the data, and use it to build a more globalized search engine. Right?

Well, the answer is: yes and no. The technology doesn’t prevent this, but the data does. Or at least, it provides an interesting challenge.

Scenario:

The poem Do not go gentle into that good night by Dylan Thomas is almost universally known. If you search in Google for this poem, you’ll find thousands of references to it, including this one at poets.org, which also includes an audio reading of the poem by Thomas, himself.

Within the closed system that is poets.org, when this Thomas poem is identified as a poetry resource, the identifier given it, based on the RDF model, is the URL of the poem, above. This makes sense within the closed system, because the poem and the web resource (the page) are one in the same within the closed system that is represented by poets.org.

However, another closed system such as Loren’s weblog, also featuring this poem, has a different identifier for it — the page where it is located.

How do we reconcile that the resource identified by “http://www.poets.org/poems/poems.cfm?45442B7C000C07040C7A” is the same as that identified by “http://www.lorenwebster.net/In_a_Dark_Time/archives/000236.html#000236”? And once we do, which ‘identifier’ do we use within the global system? we could pick one closed system over another and just use their identifiers, but which would we choose? All add their value to this poem, through a unique perspective as well interpretations associated with the work?

This is where we start bringing the purely automated solutions back into the picture — we determine if two resources are the same by performing a relatively sophisticated pattern match on the titles of the work. Chances are, accounting for spelling differences as well as language, we’ll be able to find matches at least 98% of the time. Or more.

And rather than choose one closed system over another, we merge the information and interpretations and concepts from both systems into one data store, something that can be accomplished quite easily with RDF/XML, and generate a third, unique identifier within this higher-level system.

Of course, it then is nothing more than a repeat of the same processes to merge the data from several higher-level systems into an even more globally placed system. A system such as Google, if it was so inclined.

I hesitated about bringing specific technical details into the discussion on Poetry Finder at this time, because we’re at the stage where we’re letting our imaginations roam, without hinderance by technical limitations. We’re not ready yet for the constraining environment of actual implementation issues.

Still, it’s hard to dance about semantics, when you’re not sure if there’s a floor underneath you as you twirl about. So go ahead, tap your feet — there’s something there.

Recovered from the Wayback Machine.

Third in a multi-part series focusing on RDF (Resource Description Framework) and poetry and demonstrating two-way integration between art and technology. No prior experience with either RDF or poetry is required.

“My father was a drunk. He beat my mother, he beat me. And my mother, rather than fight back, rather than protect me, as a mother should, she did nothing. Just took it, kept silent and told me to be silent. I hated them both for what they did to me. I loved them both for giving me life.”

I sometimes wonder if we’ll ever meet intelligent life from another world. If we do, will they admire our technology? Our physical appearance? Will they understand our religion and our governments?

Will they like our art, our paintings, and our music? How about our writing and our poetry? Yes, will they like our poetry? Or will they return home and tell the people, “They perform a dance, a ritual with their words. They hide both their pain and their joy, but they do so by talking about it. And the better they are at hiding, the more they’re exposed.”

“We don’t understand it. We don’t understand it, at all.”

The beginning of this post contains what could be a journal entry about a child, a little girl, raised in an unhappy home. One doesn’t have to search hard to find the story in the words—it’s told in bright light, characters reflected as shadows in harsh relief. You might pity the child, but the story’s been told before a 100 times, a thousand, before. You hear the words, but you soon forget them.

Now compare this to the words of the poem, I go back to May 1937, written by Sharon Olds, that Jonathon shared today. Each contains what could be the same story told by the same little girl. Which one is remembered, and which forgotten?

Poetry has always allowed us to share our deepest thoughts, our darkest fears, our happiest times; but it does so indirectly, through the use of imagery, the cadence of words, the subtlety of symbols. Rather than tell you “I’m lost, I’m sad, I’ve been betrayed in the worst way”, the poet writes, as William Blake wrote in The Sick Rose, about a worm, and a rose:

O Rose thou art sick.
The invisible worm.
That flies in the night
In the howling storm:

Has found out thy bed
Of crimson joy:
And his dark secret love
Does thy life destroy.

A story about a rose and a worm. Or is it a story about a rape of a young woman by a trusted friend, like a priest?

Tragic, forbidding, uncomfortable, uplifting, even whimsical, all poems share one unique characteristic that provides a challenge for the Internet — poems are impossible to discover through purely mechanical means. No matter the power of the engine, the brilliance of the designers, the intelligence of the web bots, poems can not be crunched like so many numbers, discovered based on simple pattern matches, and regular expressions.

If we want to grow beyond keyword matches, we’re going to have to do a little work. Rightfully so. Since when did we start believing that innovations on the web happen between the hours of midnight and 2AM, performed by unseen fairly like creatures clothed in gossamer threads held together by jewels, requiring no effort on our part?

Contrary to popular opinion, the web didn’t just ‘growed’ up, like Topsy. This page took work. The software to build it took work. The technology to support the software took work. You are looking at the combined effort of a thousand cubicle bound slaves, fingers permanently crooked, fish belly pale, and permanently wired from Mountain Dew. It took years to get to the point where you can search on the perfect woman in Google, and find, well, me, and this is based on a keyword search. How much more can we expect from our machines, to go looking for betrayal and find it in a rose?

“We don’t understand it. We don’t understand it, at all.”

Of course, I had you convinced of the necessity of direct human intervention back in The Beginnings of a Beautiful Friendship — why persist?

Why? Because it’s fun. And because there is no fire, and I’m holding no bucket. However, it’s now time to get down to work. And since the technician was out having fun, the poets have had a chance to speak out.

In my comments, Maria wrote, Does this mean that the force of poetry is about what we bring to it — our emotional reactions? In a word, Yes. The feelings evoked by a poem are one of the many characteristics, or facets by which we classify the poem internally. Ask a person to think of a happy poem, and you can watch their face as they go through a mental ‘search’, first doing a filter based on ‘happy’, and then a more selective search based on recency and partiality.

At their most primitive, emotions are sometimes used to create poetic ‘themes’ — love poems, happy poems, poems about fear, and so on. However, many writers, and readers, object to classifying a poem based on such bland bit buckets, and I’m one of them. The feelings evoked by poems are usually more complex than ‘happy’ or ‘sad’; and, many times, bring with them a frame of reference unique to the reader.

In the Renaissance Web discussion list, Dan Lyke made an important observation:

 

One of the problems with Shelley’s poetry example is that much of the
meaning in poetry is from the frame of reference of the reader. I
remember an NPR interview nearly two decades ago with some author of a
novel that was being critically regarded as a brilliant look at
Central American politics in the Reagan era, and the woman who wrote
it said something like “no, it’s a story about a deformed child coming
of age”.

Maybe it was, maybe it wasn’t, but there was no way that that author
was ever going to semantically mark-up that book to have the meanings
that most people had ascribed to it.

Dan’s right, there is no way that an author can capture each reader’s interpretation, and they shouldn’t even try. In fact, poets and other writers should be actively discouraged from adding their own input into a system such as the RDF Poetry Finder. Nothing ruins a poem or other writing more than the writer providing their own interpretation.

Each writing is a contract between writer and reader: the writer evokes, the reader responds. If the writer wants the reader to respond exactly in a certain way, they should use powerful drugs. And hypnosis.

Dan concurs with this, and wrote:

So one of the ways, perhaps, to view the markup process is as distinct
from the authoring process, perhaps collaboratively done by readers. A
smarter version of the whole Slashdot moderation system, for
instance. “I think this is about the plight of Nicaraguans in the
context of their struggle for agricultural autonomy”, and maybe it’s
some peer-to-peer process through which these ratings propagate so
that the meanings I search on are the meanings that friends have also
seen.

Besides, this will give some meaning to the otherwise empty lives of
post-modernists.

We attach categories and topics to our own postings, but for the RDF Poetry Finder to be effective, it must allow readers to read other interpretations, and either enhance existing ones, agree with them as is, or create all new ones.

The dynamic nature of this system is important — just as there isn’t a finite number of keywords in Google, there isn’t a finite number of concepts in poetry. Xian, at the discussion group, expressed his concern on this point:

I was more concerned that there would be a big lookup table of
symbols that would lead people to believe, for example, that the
missing bird in my No Bird but an Invisible Thing story
(here) must represent a human soul.

(I’m stretching the equation here, as in the case of my story the
bird isn’t an owl, but I wanted to try to tie it to something
concrete.)

Valid concern. I shudder to think what it would be like on a committee of people trying to come up with a universal set of concepts. Not only that, but I can’t imagine that either the writer or the reader would accept it.

The dynamic nature of the system doesn’t have to be complicated. As a quick demonstration about a possible approach, I discussed how something like this might work with Sylvia Plath’s poem, Mirror, which generates so many different interpretations:

 

Consider a concept, from my earlier message about Plath’s “Mirror”.

Metaphor (symbol) is “mirror”

Concept is “growing older”

But someone else comes along and says, it’s more than just ‘growing
older’. It’s about coming to terms with growing older. So they
pick “growing older”, and then add annotation.

“coming to terms with growing older”

Another person comes along and they think it has to do with a parent
and child and the relationship between each other. Has nothing to do
with getting older. They add their interpretation of the concept:

“infant’s interaction with mother”

And a third person comes along, says the mirror is just a mirror. A
fourth says that the mirror reflects our view of ourselves, our
ugliness, our beauty, and so on.

We end up with a hierarchy of concepts, so to speak, with like
concepts related to each other, unlike ones separate.

So Poetry Finder must be dynamic, but once we have the interpretations recorded, what then? How do we use them? We understand how to search on keywords and even key phrases, but how do we search on concepts?

We start small, starting with the ability to search on symbols, as recommended by Loren in a comment>. Doing this, we’re combining the familiar, a keyword search, with the new, a smart keyword search.

An example of ‘symbol’ we kicked around was the use of birds to represent souls. Specifically, the use of an owl in a poem to represent a person’s soul. From the discussion group, I wrote:

 

…if you have a way of mapping that soul can be symbolized by
owl, and owl can symbolize soul, then within that system you can
search for either all symbols that represent soul, pick owl, and then
focus on these results. Or you can search for ‘things symbolized by’
owl, pick soul from the results, and focus on these results.
Technically, both are ‘keywords’, because they are a single word that
serves as a reference point — but they’re very smart keywords. There
is an inference associated with the keywords, not to take them at
face value, but to look at their association, mapping, what have you
with other words.

In other words, there is a predicate, ‘symbolize’, that maps owl to soul, and soul to owl. Using a system that allowed me to make smart keywords out of symbols, I could then find a poem such as the following (kindly provided by Loren, who is now resting from his labors):

 

Edward Thomas
The Owl

Downhill I came, hungry, and yet not starved;
Cold, yet had heat within me that was proof
Against the North wind; tired, yet so that rest
Had seemed the sweetest thing under a roof.

Then at the inn I had food, fire, and rest,
Knowing how hungry, cold, and tired was I.
All of the night was quite barred out except
An owl’s cry, a most melancholy cry

Shaken out long and clear upon the hill,
No merry note, nor cause of merriment,
But one telling me plain what I escaped
And others could not, that night, as in I went.

And salted was my food, and my repose,
Salted and sobered, too, by the bird’s voice
Speaking for all who lay under the stars,
Soldiers and poor, unable to rejoice.

Of course, for this system to work, we couldn’t lock ‘owl’ into just one role, metaphor for soul. Maria writes:

 

In my repertoire of symbols, or almost languageless associations of objects with thoughts and feelings, the owl has been always a representative of wisdom, or vision, and is forever tangled up with my sense of Greece and Athena.

There’s that dynamic nature of RDF Poetry Finder. Growing by leaps and bounds.

Is this a book yet? Enough for one night.

 

Next: Interpretations — filtering the hit and run kiddies