Notes from Last Week

OK, fell a bit on my research blogging last week, but overall still maintaining. In summary, I finished the "Little Schemer" and now feel like I have my head around Scheme (although the Y Combinator is till a bit confusing, but I did manage to go through the last chapter and implement a giant table-based interpreted for Scheme in Scheme!). Can't wait to get my hands on the "Seasoned Schemer", and I can already see how lambdaXML will allow a level of flexibility in pipelines neverbefore imagined while relying on the good-old-friendly idioms of LISP and the lambda calculus.

Had an interesting meeting with Ewan in which he told me about the IR stuff he was doing and how his work and Bundy (!) could fit in mine possibly...Bundy apparently can clean up messy propositions, which I have lots of!

I managed to write a mapping from XPMDL to lambdaXML for Henry:

http://www.ibiblio.org/hhalpin/homepage/notes/lambda.html

Henry at the meeting revealed to me that the MTP actually decompiles to a significantly more powerful langauge that he believes can implement all the basics needed by lambdaXML version 1.0.
Which I'm not sure if I believe, but if he can explain it to me, then I'll implement it in short order.

I also wrote lengthy notes on Dretske and B.C. Smith, and am going through a giant stack of papers related to the philsophy of computation (starting with Chomskey's classic "Rules and Representations"..), but these aren't quite ready for HTML yet. However, I did manage to write down my draft thinking, available here:

http://www.ibiblio.org/hhalpin/homepage/notes/philosophyabstract.html
The stories from Senga have mysteriously not arrived, which has me worried. Will investigate tomorrow, and will dedicate the rest of next week to finishing the web-interface for the pipeline for Johanna and wrapping up draft one of the philosophy paper for Andy and Henry.

Meeting with Johanna

Quick notes. With the stories coming in next week, we should be well on the way to writing a EACL paper. Also, need to finish getting Web interface for the pipeline up, which should occupy me for most of next week. Lastly, it appears we might be able to automate the "pyramid" scheme using the DUC summarization system, since their system essentially grabs "semantic content units" in the same way that we do, just via humans and not by hands, and ranks in a similar manner.

Pipelines to Functional Programming, KRL to the SemWeb

Had another meeting with Henry today, very productive. As for the Semantic Web Science workshop in Cambridge, Henry wrote a very nice little position paper that mentioned me :)

In essence, we went over a number of topics. First, that the f(X) Henry will be presenting as a W3C Member Submission is going to be rather simple (only wrap the outside of an Infoset) and only encode W3C standards, but he thinks this is the easiest way to get the ball rolling so to speak. He's quite interested in how f(X) could solve the PHP embedding code problem (as when random non-Infoset code starts appearing in Infosets!) and so lead to better code modularity, and how it could solve the AJAX problem of scripts just ad-hoc modifying DOM trees ( by letting an XML tree expose itself per se). Since I'm more of a Java hacker than a C hacker, it makes sense for me to modify the Markup Technology pipeline to bootstrap f(X) rather than the LX Toolset or coding it all myself, which I agree. I need to develop a side-by-side comparison of MT pipeline vs. f(X).

The second is that there are a number of lessons for the Semantic Web from old KRL projects. First, that Maturana is right on many counts, but his hardline stance against representation is wrong. Maturana has a hard time explaining how "internal structural changes" that have no connection to the signal, can explain the following Brian Smith parable: "To get through a rapids on a canoe, it actually makes sense to paddle upstream, where a balance in the rapids can be obtained so one can better reach a sort of stasis and get through the rapids safely." - and if you tell this to someone who doesn't know it, that they will try this line of action. It seems that there are quite a few stories that only representation can explain easily! However, if analytic philosophy seems to be coming up against all of these problems, and the obvious other choice to build AI upon is hermeuneutics, why isn't anyone following it up? Barwise never formalized situation semantics, Robin Cooper did note that "misunderstaning doesn't just happen, it's constitutive of natural language", and Brian only got so far. So what happened? Is hermeuneutics not enough, due to its *lack* of a representational story?

Lastly, that the KRL project got really complicated and more or less collapsed (having Rumelhart run away into neural networks!) when to properly update semantic networks they finally had to incorporate both action (such as backward and forward chainining triggers and traps!), and reflection to control. So - what's the main problem for the Semantic Web? When the Semantic representations change.