Next Boston Lisp Meeting: Monday November 24th 2008 at 1800 at MIT 34-401B Gregory Marton will give a talk about the meanings of English words as programs. http://fare.livejournal.com/136992.html
Gregory will introduce a way of thinking about the meanings of English words as programs (c.f. SHRDLU), cast the problem of language learning as a search through the space of possible programs, and show first steps in learning. Gregory has created a system called Sepia that makes it relatively easy to write small lexicons that implement some semantic theory for a domain. Applications that Gregory and several colleagues have had fun with include:
* understanding dates and times and other measures: "next Wednesday" * finding and linking names of people, organizations, places * telling a robot what to do with stuff on a table: "touch the red one" * asking about spatial paths in video: "show people entering the kitchen" * a toy gossip world: "John loves Mary" "Who does not hate Mary?" * a little number theory: "18 is twice the sum of its digits"
These applications are inherently brittle, and manual construction gets harder as the application grows. The research goal is to construct new meaning programs with little human input, or with input from non-programmers. Given a problem phrase, Sepia uses standard ways to find similar words that it knows. Its current approach is to take the meanings of those words and makes small changes, looking for a combination that produces the target meaning.
Both the system as a whole and the language of semantics are implemented in the GNU/Guile flavor of Scheme.
Gregory Marton is a PhD student in computer science at MIT, and this is his dissertation topic. His first language is Hungarian, and since meeting English at age eight, he has wondered about meanings and how to learn language. He has a B.S. in Computer Science with a minor in Linguistics from the University of Maryland, College Park, 1999.
*
The Lisp Meeting will take place on Monday November 24th 2008 at 1800 (6pm) at MIT, Room 34-401B.
As the numbers indicate, this is in Building 34, on the 4th floor. This is the usual location, on 50 Vassar Street, Cambridge.
MIT map: http://whereis.mit.edu/bin/map?selection=34
Google map: http://maps.google.com/maps?q=50+Vassar+St,+Cambridge,+MA+02139,+USA
Many thanks go to Alexey Radul for arranging for the room, and to MIT for welcoming us.
* *
Buffet: ITA Software, a fine employer of Lisp hackers (disclosure: I work there), is kindly purchasing a buffet to accompany our Monthly Boston Lisp Meeting. Anyone who attends is welcome to partake. We appreciate it if you let us know you're coming, and what food taboos you have, so that we can order the right amount of food. Tell us by sending email to boston-lisp-meeting-register at common-lisp.net. We won't send any acknowledgment unless requested; importantly, we'll keep your identity and address confidential and won't communicate any such information to anyone, not even to our sponsors.
* * *
The previous Boston Lisp Meeting on October 27th had over 30 participants. Tim McNerney gave copious background on the Thinking Machines Corporation and its line of massively parallel computer as an introduction to his work on automated verification of a compiler optimization pass through abstract interpretation. Both the presentation and the following discussion were very lively.
We're always looking for more speakers. The call for speakers and all the other details are at http://fare.livejournal.com/120393.html
For more information, see our new web site boston-lisp.org. For posts related to the Boston Lisp meetings in general, follow this link: http://fare.livejournal.com/tag/boston-lisp-meeting or subscribe to our RSS feed: http://fare.livejournal.com/data/rss?tag=boston-lisp-meeting
Please forward this information to people you think would be interested. Please accept my apologies for your receiving this message multiple times. My apologies if this announce gets posted to a list where it shouldn't, or fails to get posted to a list where it should. Feedback welcome by private email reply to fare@tunes.org.
[ François-René ÐVB Rideau | Reflection&Cybernethics | http://fare.tunes.org ] When all lawful citizens are disarmed, will we have an omnipresent police state to protect us from armed criminals?