wsd as distributed constraint optimization problem
TRANSCRIPT
WSD as Distributed Constraint Optimization Problem
Author: Siva Reddy, Abhilash Inumella
Publication: ACL 2010
Presenter: Po-Han Lin (林伯翰 )
Outline
Background WSD as a DCOP Experiment Conclusion and Future References
Outline
Background WSD COP DCOP
WSD as a DCOP Experiment Conclusion and Future References
Word Sense Disambiguation
Word Sense Disambiguation (WSD) The problem of WSD can be defined as the task
of assigning the most appropriate sense to the word within a given context.
WSD is one of the oldest problems in computational linguistics which dates back to early 1950’s.
Example: Noun orange has at least two senses:
color or fruit.
Constraint Optimization Problem A COP can be defined as a regular constraint
satisfaction problem in which constraints are weighted and the goal is to find a solution maximizing the weight of satisfied constraints.
x is a vector residing in a n-dimensional spacef(x) is a scalar valued objective function, gi(x) = ci for i = 1, …, n and hj(x) d≦ j for j = 1, …, m are constraint functions that need to be satisfied.
Distributed Constraint Optimization Problem DCOP is the distributed analogue to
constraint optimization. DCOP is a problem in which a group of
agents must distributedly choose values for a set of variables such that the cost of a set of constraints over the variables is either minimized or maximized. DCOP can be formalized as a tuple
(A, V, D, C, F)
Distributed Constraint Optimization Problem - (A, V, D, C, F)
A is a set of n agents. A = {a1, a2, . . ., an}
V is a set of n variables. V = {x1, x2, . . ., xn}
D is a set of domains each one associated to the corresponding variable. D = {D1, D2, . . ., Dn}
C is a set of constraints described by various utility functions fk. fk is defined over a subset of variables V.
It maps every possible variable assignment to a cost.
C = {fk |fk : Di × Dj × . . . × Dm → }ℜ
Domain of fk is Cfk = {Di × Dj × . . . × Dm}
fk(c) = utility associated with the constraint c, where c C∈ fk.
F is the objective function to be maximized.
zk is the weight of the corresponding utility function fk .
Outline
Background WSD as a DCOP Experiment Conclusion and Future References
WSD as a DCOP
Agents Each word is treated as an agent.
Variables We define the sense of a word as its variable.
Domains Domain of the variable.
Constraints A constraint specifies a particular configuration of the agents involved
in its definition and has a utility associated with it. Objective function
objective function is defined over all the knowledge sources (fk) as below:
F denotes the total utility associated with a solution zk is the weight given to a knowledge source
Detial about Constraints Part-of-speech (POS 詞性 )
Different POS can decide different domain of words. Ex: play has 47 senses, but only 17 senses correspond to noun category.
Morphology (構詞學 ) Word form (字形 ) Vocabulary (字詞 )
orange has color and fruit two senses, but oranges only be used in the fruit sense.
Domain information Information can be captured using a unary utility function defined for every word.
Sense Relatedness Sense relatedness between senses of two words wi, wj is captured by a function f, where f
returns sense relatedness (utility) between senses based on sense taxonomy and gloss overlaps.
Discourse Many different word, but same sense.
Collocations co-occur word. Ex: bank:
financial institution the edge of a river if in a given context bank co-occur with money, “financial institution”
Outline
Background WSD as a DCOP Experiment Conclusion and Future References
Experiment
Experiment Data SENSEVAL-2 SENSEVAL-3
Ex: <instance id="activate.v.bnc.00024693" docsrc="BNC">
<answer instance="activate.v.bnc.00024693" senseid="38201"/>
<context>
… which you step on to <head>activate</head> it . Used correctly , … .
</context> </instance>
Answer <lexelt item="activate.v"> <sense id="38201" source="ws" wn="activate
%2:36:00::" synset="activate actuate energize start stimulate" gloss="to initiate action in; make active."/>
<sense id="38202" source="ws" wn="activate%2:30:03::" synset="activate" gloss="in chemistry, to make more reactive, as by heating."/>
… </lexelt>
Experiment
Experiment Results Dcop:
This paper method Sinha07
Sinha and Mihalcea, 2007 page rank algorithm
Agirre09 Agirre and Soroa, 2009 page rank algorithm used additional knowledge
extended WordNet relations sense disambiguated gloss.
MFS most frequent sense popular back-off heuristic in WSD
system.
Outline
Background WSD as a DCOP Experiment Conclusion and Future References
Conclusion
Modelling WSD in a distributed constraint optimization framework.
Showed that this framework is powerful enough to encode information from various knowledge sources.
FutureWork
Only used relatedness based utility functions derived from WordNet.
Effect of other knowledge sources remains to be evaluated individually and in combination.
The best possible combination of weights (zk) of knowledge sources is yet to be engineered.
Outline
Background WSD as a DCOP Experiment Conclusion and Future References
References
WSD as a Distributed Constraint Optimization Problem http://dl.acm.org/citation.cfm?id=1858916
Distributed Constraint Optimization Problem http://www.doc88.com/p-23745051258.html http://en.wikipedia.org/wiki/Distributed_constraint_optimization
Constraint optimization http://en.wikipedia.org/wiki/Constraint_optimization
Word Sense Disambiguation http://wordnet.princeton.edu/ http://203.64.42.21/course/2005/tgbcl/poko/ohcl-13.htm http://www.scholarpedia.org/article/Word_sense_disambiguation
SENSEVAL http://www.senseval.org/