To: re...@starbourne.com Subject: Fwd: [humanmarkup-comment] Out of the blue... From: jo...@enteract.com (Jorn Barger) Date: Mon, 8 Jul 2002 17:24:25 -0600 X-Rcpt-To: <re...@starbourne.com> X-DPOP: DPOP Version 2.4a Status: U

------- Begin Forwarded Message -------

Subject: [humanmarkup-comment] Out of the blue... From: Jorn Barger <jo...@enteract.com> To: <rbro...@humanmarkup.org> Date: Sat, 29 Jun 2002 08:22:14 -0600

I'm severely 'allergic' to mailinglists so I'm posting this indirectly, but I was invited to join (by Ranjeeth, I think) back on Day One, and I'm impressed you're still fighting the good fight...

I spent most of the 1970s sifting thru vocabulary-lists of emotion- words (etc) trying to come up with anything like a solid set of primitives. I believe I _did_ eventually make some progress there, but the final direction was so unexpected and counter-intuitive that before I sketch out my primitives, I want to briefly describe what I'm doing now, which really doesn't rely on the primitives at all.

Ranjeeth saw my comp.ai.* postings a year ago, about my 'Internet Timelines Project' which starts from the assumption that the most 'primitive' psychological categories must be the very same ones you find you need first, if you happen to be trying to sketch out human history on the largest possible scale. (I used the example of HG Wells' one-volume world history, but boiled down much further via hypertext, so that the top webpage will give the most abstract possible short-outline.)

So (eg) the growth and migration of human populations, and the conquest of one population by another, must be among the 'deepest' primitives according to this principle. And those same primitives that apply for human populations can be 'downsized' for human individuals as (eg) person-A-bears-child-B (population growth), or person-B-moves-to-place-B (migration), or person-A-submits-to-person-B (conquest).

So in the medium-long term, I'm working now at filling in the gaps in my own historical knowledge, so as to be able to sketch this largescale human history. (Insofar as I'm managing to proceed chronologically at all, I'm up to ancient Athens after a year of work.)

But where I think I'm already consistently managing to add real value to the Web on a daily basis, _without_ requiring the startingplace of pre-defined primitives, is in building webpages that inventory and organise big chunks of the Web, for many different topics (mainly historical timelines).

And my main point, in this posting, is that there's a whole different set of categories required here, based on ***the psychology of web-research***, that I think might be a very useful addition to HML's arsenal.

The simplest example is a categorisation of information-types that webpages may offer: etext, timeline, FAQ, biography, map, review, essay, photo, links-list, encyclopedia-article, dictionary-definition, online-purchase-offer, Internet-Movie-Database entry, mailinglist homepage, bbs, archive, etc.

I can't emphasize strongly enough that these specific types *emerged* by daily experiment, as _useful_ in the actual pages I was building. The ontologically-inclined will immediately want to subsume the list as a branch of their pre-existing ontologies, and fill in 'gaps' based on theoretical principles... but this is precisely what I want to warn *against*.

Sure, maps and photos are both 'images', and reviews and essays (etc) are all etexts, but those abstractions just _dilute_ the typology's value. And the only way you'll appreciate the danger of such 'dilution' is if you put a ton of effort into building useful webpages. But that's what semantic markup is supposed to be about, isn't it?

(An example of the dangers of dilution: I've come to believe that link-lists are doomed to extinction as an independent category, because they're much more useful when they're woven intimately into the fabric of any and every other page-type category. But this design-idea could never occur to someone who's looking thru the 'glasses' of an ontology that reifies link-lists as an independent category.)

So as a generalisation, I'm suggesting that _every link_ on the web is necessarily claiming to offer some category or categories of human value, and that the HML project might reduce its infinite-regress problem, for psychological-ontologies, by focusing on these specific, _web-pragmatic_ values.

What's more, categories like 'etext' and 'timeline' imply content- structures of their own, that can be implemented with varying degrees of success. Etexts have authors, so an etext-page may (or may not) choose to link to an author bio, and/or to a link-list of other etexts by the same author. Any timeline will usually mention people and places that can be similarly linked to secondary/offsite pages.

But this theory of web-pragmatics can't be nailed down just by standing on the sidelines and analysing-- you have to keep building pages and listening to visitors' feedback and looking for ways to optimise their value.

To start with some examples close-to-home, I dropped in today via a circuitous route that started with a Google-search for the phrase 'xml websites'. That search indirectly led to OASIS and I noticed HumanMarkup prominently linked from the OASIS homepage, and then clicked on the archive-link from the 'committee' page, and browsed the latest month of postings...

'Mailing-list posting' is definitely a useful web-pragmatic category. There are lots of possible ways to structure a mailinglist archive, and it's not immediately obvious which are the most useful. Eg, Google Groups lets you click on a person's name and see a reverse-chronological index of all their postings, which I often find useful.

The eList software is apparently configured to supply an up-link to 'Elist Home' but this is the archives page, and there's no obvious up-link back to the committee-page. (I used the sidebar, but only because I happened to have arrived that way-- someone arriving via a Google search would have trouble spotting it.)

The committee-page mentions the mailinglists and tells how to subscribe, but doesn't spell out whether you have to subscribe to post (I'm assuming you do). These are all _pragmatic_ categories that could be usefully spelled out in detail, and eventually embedded in the design of generalised software apps.

I looked on the committee-page for the startdate of the project, and didn't find it-- there are several short 'timelines' on that page that could be merged, I think. When you do this, you start to notice the gaps (like starting-date) and you start to fill in historical connections (eg mover arrives, or shaker departs).

To connect this back to the HML ontology of emotions, one bit of web-pragmatics I often miss is the concept of 'critique'-- I can go to Yahoo and find a hundred pages about XML, but suppose I want just pages that are skeptical about XML? That 'critical attitude' is by no means a _complex_ emotion, but it would be an incredibly useful one if people could cite it somehow in semantic markup. (Eg, mailinglist archives could tag which postings are critical of what topic-areas.)

Or more broadly, if someone left the mailinglist in a huff, can your committee-timeline accurately annotate that emotion? ('Accurate' here becomes a very delicate question, and I've been arguing forever that the utility of an emotions-ontology can only be demonstrated by the 'literary-historical' quality of the psychological narratives it can represent.)

Which brings me back (sort of) to the question of emotional primitives.

Back in the 70s, I gradually came to the conclusion that lists of emotion-words really weren't adequate, because each word by itself was just an ambiguous fragment of the psychological reality.

So my counter-proposal is that when you list an emotion-word like 'critical' you recognise that it's associated with what I call the "usual 'criticism' story". Different cultures may have very different ideas of what's usual, but one likely prototypical 'criticism' story in US culture is a boss (or spouse, or parent) who denigrates something you do. So the usual-critical-story might be:

person A tries to perform X person B insults person A's performance as inadequate person A feels bad person A looks for ways to improve performance etc

Any bare emotion-word can be usefully fleshed out with a story-skeleton like this. But when you start to do this, you begin to realise why the simple words by themselves never composed themselves into a coherent ontology-- they overlap and branch in far too many ways.

So my simplification-approach leverages off the most-simplified internal composition of the stories themselves-- how many persons must be involved, at minimum? how many places? how many things? what basic relationships exist between these elements? how do those relationships change in the course of the story?

I've proposed a data-structure for classifying these that requires self-similar hierarchies within hierarchies... but my Timelines project simplifies it considerably by reducing (eg) 'migrates' to:

date1= person1= place1= place2= relationship=moves

I should end with a batch of urls, I guess:

timeline project: http://www.robotwisdom.com/web/biography.html recent examples: http://www.robotwisdom.com/science/classical/ ontology-theory: http://www.robotwisdom.com/ai/ web-design theory: http://www.robotwisdom.com/web/