BEGIN:VCALENDAR
VERSION:2.0
PRODID:ILLC Website
X-WR-TIMEZONE:Europe/Amsterdam
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
X-LIC-LOCATION:Europe/Amsterdam
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:/NewsandEvents/Archives/2005/newsitem/1087/23-
November-2005-Computational-Linguistics-Seminar-Je
lle-Zuidema
DTSTAMP:20051121T000000
SUMMARY:Computational Linguistics Seminar, Jelle Z
uidema
ATTENDEE;ROLE=Speaker:Jelle Zuidema
DTSTART;TZID=Europe/Amsterdam:20051123T160000
DTEND;TZID=Europe/Amsterdam:20051123T000000
LOCATION:Room P.327, Euclides building, Plantage M
uidergracht 24, Amsterdam
DESCRIPTION:Stochastic Tree Substitution Grammars
(STSGs), such as used in Data-Oriented Parsing, ha
ve great linguistic advantages, essentially mergin
g "construction grammar" with "probabilistic lingu
istics". However, from a computational linguistics
perspective, they pose a number of computational
challenges that have not yet been satisfactorily s
olved. Two fundamental and related problems are
"the problem of estimation" -- estimating the wei
ghts of an STSG from observed subtree frequencies
in a tree bank -- and "the problem of expectation"
-- calculating the expected subtree frequencies w
hen generating trees using an STSG with known weig
hts. A linguistic desideratum for estimation is th
at it converges to the maximally general STSG out
of the possibly many correct ones. I will briefly
discuss why none of the existing estimation method
s fulfills this desideratum. I will then present m
y recent work on the problem of expectation and di
scuss how its solution directly suggests an altern
ative approach to the first problem. For more i
nformation, see http://staff.science.uva.nl/~jzuid
ema/CLS/
X-ALT-DESC;FMTTYPE=text/html:\n \n
Stochastic Tree Substitution Grammars (STSGs), suc
h as used in Data-Oriented Parsing, have great lin
guistic advantages, essentially merging "cons
truction grammar" with "probabilistic li
nguistics". However, from a computational lin
guistics perspective, they pose a number of comput
ational challenges that have not yet been satisfac
torily solved.\n

\n \n Two
fundamental and related problems are "the pr
oblem of estimation" -- estimating the weight
s of an STSG from observed subtree frequencies in
a tree bank -- and "the problem of expectatio
n" -- calculating the expected subtree freque
ncies when generating trees using an STSG with kno
wn weights. A linguistic desideratum for estimatio
n is that it converges to the maximally general ST
SG out of the possibly many correct ones. I will b
riefly discuss why none of the existing estimation
methods fulfills this desideratum. I will then p
resent my recent work on the problem of expectatio
n and discuss how its solution directly suggests a
n alternative approach to the first problem.\n

\n \n \n For more informat
ion, see\n http://staff.
science.uva.nl/~jzuidema/CLS/\n

URL:/NewsandEvents/Archives/2005/newsitem/1087/23-
November-2005-Computational-Linguistics-Seminar-Je
lle-Zuidema
END:VEVENT
END:VCALENDAR