2009/03/28

Semantic Role Labeling: An Introduction to the Special Issue

®  Sentence + designated verb: SRL identify the arguments of the verb. And label them with semantic roles.

®  Steps:

¯  1 Pruning.

¯  2 Local scoring

¯  3 Joint scoring

(4) Fixing common errors

 

Pruning

®  Filtering the set of arguments for a given predicate.

¯  Any subsequence of words in the sentence is an argument candidate.

®  Xue and Palmer (2004)

Joint scoring      

®  Good structure of labeled arguments.

®  Arguments do not overlap,

®  Core arguments do not repeat, etc.

®  Re-ranking

®  Probabilistic methods.

Conditional random fields

SRL architecture:

®  Combination of systems and input annotations.

¯  Increase robustness,

¯  Gain coverage

¯  Reduce effects of parse errors

®  One my combine:

¯  Output of independents srl basic systems

¯  Outputs from same srl s. changing input annotations or parameters.

Gaing of 2-3 F1 points.  

®  Joint labeling…

®  Dependency parsing

®  Combine parsing and srl in a single step. …

®  Characterize candidate argument
phrase type, headword, …

®  Characterize verb predicate +cntx
lemma, voice,

®  Characterize the relation.
Syntactic + semantic.
Left/right position of the constituent  with respect to the verb…

®  Recall:

®  81% argument identification

®  95% assigned correct semantic role.

®  SemEval-2007

®  Disambiguation of 50 verbs.

®  FrameNet

®  40 frames

®  F1=92% asigning semantic roles

®  F1 83% segmenting and labeling arg.

®  Complete analysis of semantic roles on unseen texts,

®  Precision 60s

®  Recall 30s

®  SRL relies on syntactic structure.

®  Output by statistical parser 90% matching.

®  Is common to use parser trees.

®  Gold-standard trees.???

®  Most of the errors are by having incorrect syntactic constituents.

®  SRL relies on syntactic structure.

®  Output by statistical parser 90% matching.

®  Is common to use parser trees.

®  Gold-standard trees.???

®  Most of the errors are by having incorrect syntactic constituents.

®  CoNLL-2005 Brown corpus annotated

®  Performance drop below 70%

®  They clamed errors are in assigning the semantic roles rather than identification of argument boundaries.

®  Spanish and catalan, CESS-ECE corpus.

®  86% disambiguation predicates

®  83% labeling arguments.

®  chinese

®  semEval-2007 25K SENTENCE

®  80% core elements.

Top performing team use machine learning techniques

2009/03/27

helping COST2102

2009/03/26

helping cost
burocracy
reading papers

2009/03/25

borderline-smote
slides ready for the reading group
PREPARING slides
helping COST 2102

2009/03/23

helping carl in cost

Semantic role labeling: an introduction to the special issue

 

1 Introduction SRL:

APLICATIONS:

Information extraction

Question ansering

Summarization

MT

NLP tasks

major conclusion of that work is that the patterns of syntactic alternation exhibit regularity that reflects an underlying semantic similarity among verbs, forming the basis for verb classes major conclusion of that work is that the patterns of syntactic alternation exhibit regularity that reflects an underlying semantic similarity among verbs, forming the basis for verb classes ?? what it means?

 

2  Semantic roles in linguistics

 

3 From Linguistic Theory to computational resources:

2002 first statistical machine learning approach to SRL

Verb lexicon

Proposition bank

Each verb has a framset listing its allowed role labeling.  

Arg0, arg1 -> to be specific with semantic roles

Abstract semantic role: agent, theme, location.

4 Approaches to automatic SRL:

Only supervised approach.

4.1 SRL Step by Step

First step: filtering or pruning.

The second step consists of a local scoring of argument candidates: argument identification and classification.

The third : joint scoring.

 ----------

FRAME SEMANTICS: 

Syntactic-oriented arguments and adjuncts

frame semantics core and peripheral.

...


posible topics :

 

2009/03/21

start to read 
semantic role labeling: an introduction to the spe3cial issue
helping carl to prepar COST2102

2009/03/20

we decided to work on semantic role labeling
by spanish book for spanish lectures + return books

2009/03/19

TCD
2 PART
machine learning

2009/03/18

ACM news

New System for Improving Decision Support Systems
Universidad Politecnica de Madrid (03/13/09) 

Universidad Politecnica de Madrid School of Computing researchers have developed a system designed to improve decision-making processes in complex situations. The system was tested on the restoration of Lake Svyatoye in Belarus, which was contaminated by the Chernobyl accident. Professors Antonio Jimenez, Alfonso Mateos, and Sixto Rios, from the Department of Artificial Intelligence's Decision Analysis and Statistics Group, aimed to account for incomplete information and any possible effects those gaps could have on decision making. Multi-Attribute Utility Theory is often used to solve decision-making problems. The theory says that after building a hierarchy of objectives and identifying a set of alternatives and each alternative's value for impact on the objectives, the decision maker's preferences are quantified. The new system uses two approaches to manage incomplete information, which occurs when the impacts of some alternatives and attributes are unknown. The first approach redistributes criteria weights with missing values or impacts throughout the objectives hierarchy and across other criteria, which means the criteria hierarchy and its assigned weights vary when each alternative is analyzed depending on the criteria with missing values. The second approach associates the citerion range, the set of possible values, as the impact for a criterion with missing values, which means the entire range of values are considered possible and equally likely.

Machine learning

A tentative table of contents:

  • Bayesian Decision Theory
    • minimum error rate classification
    • discriminant functions and decision surfaces
  • Parametric models and parameter estimation
  • Non-parametric techniques
    • K-Nearest neighbors classifier
    • Decision trees
  • Linear models
    • Perceptron
    • Logistic regression (Maxent)
    • Large margin and kernel methods
  • Generative versus discriminative modeling
  • Sequence labeling and structure prediction
    • Hidden Markov and MaxEnt Markov Models
    • Sequence perceptron
    • Conditional Random Fields
    • Learning + Inference models

2009/03/17

summer schools

2 hours searching

start

finish

duration

Price:

web

1

pisa

italy

Evaluation, Best Practices and Collaboration for Multilingual Information Access

15-Jun

19-Jun

4


 €  200.00

http://www.trebleclef.eu/summerschool_registration&grants.php

only tutation fees

GRANT possibility

2

Bordeaux

france

18-Jul

02-Aug

15


 €  225.00

http://esslli2009.labri.fr/

ACEPTED

3

belfast

Uk

artificial intelligence

23-Aug

29-Aug

6


 €  340.00

http://193.61.130.114/acai09/index.html

recommendation letter

4

cambridge

UK

Machine learning

29-Aug

10-Sep

12


 900pounds

http://mlg.eng.cam.ac.uk/mlss09/registration.htm

recommendation letter, all included

5

Petrozavodsk

russia

Information retrival

11-Sep

16-Sep

5


 €       0.00

http://romip.ru/russir2009/eng/registration.html

padua

Italy

Information retrival

31-Aug

04-Sep

4


 500

http://essir2009.dei.unipd.it/

 

Information Retrieval ESSIR2009

at the same time as in cambridge


Registration: Apr 30, 2009

School: Aug 31 - Sep 4, 2009


Introduction

The European Summer School in Information Retrieval (ESSIR) 2009 gathers in Padua, Italy lectures and satellite meetings for exchange and dissemination in the modeling, design and implementation of advanced systems for the representation, storage, search and retrieval of information.

ESSIR 2009 will be the 7th European School in Information Retrieval. For more information on the past editions, check the History of ESSIR.


Location

Padua is a pleasant historical city, home to one of the oldest and most prestigious Universities in Europe. Now little more than 400 years ago, Galileo, came here as a Professor of Mathematics to spend, in his own words, "the 18 best years of my entire age". Venice is only 15 miles away, connected by frequent buses and trains. Bologna or Verona, Florence or Milan, Rome or Turin, are conveniently reached by train in one, two, four hours, respectively.


Speakers

M. Agosti, R. Baeza-Yates, N. Fuhr, A. Gionis, P. Ingwersen, M. Lalmas, M. Melucci, J.Y. Nie, S. Robertson, S. Rüger, I. Ruthven, M. Sanderson, J. Shanahan, C.J. van Rijsbergen, H. Zaragoza.


Topics

Foundations of Information Retrieval, Indexing Techniques, Formal Models, Evaluation of Information Retrieval Systems, Users and Context, Multimedia and Multilingual Retrieval, Digital Libraries, Semi-structured Data, Machine Learning, Distributed Search, Advertising and Mining on the World Wide Web.


Chairs

2009/03/16

Get enrroled
21st European Summer School in Logic, Language and Information
Bordeaux July 20-31, 2009
study statistics

2009/03/15

HCI in TED



Human computer interaction

2009/03/14

updating blog with keywords from the last two days conferences 

Grammar Induction and Iterated Learning

 Grammar Induction and Iterated Learning.

 

Dr. Willem (Jelle) Zuidema from the Institute for Logic, Language and Computation, University of Amsterdam

 

 

 

key words:

bayesian learning

signal-> concept

 

Grammar induction:

Merging and chinking non terminals

Chinking repeats the pattern.

 

The correct secuence can reproduce any free context grammar.

 

PTSG Probabilistic Tree substitution Grammars

DOP grammar , assign all possible binary structure

 

 

 

 

 

 

Iterative learning

Cause:

Mechanism

Ontogeney -> Bayesian

Functions

Phylogeny :  evolution

 

Grammar -> language -> grammar -> language -> …

Posteriors

Likehood

Prior

Probability of the data.

2009/03/13

Unsupervised and Constrained Dirichlet Process Mixture Models for Verb 


Speaker: Andreas Vlachos Computer Laboratory Cambridge 


University Abstract:  

In this work we apply Dirichlet Process Mixture Models (DPMMs) to a learning task in natural language processing (NLP): lexical-semantic verb clustering. Furthermore, we propose a novel method of guiding the DPMM towards a particular clustering solution using pairwise constraints. The quantitative and qualitative evaluation performed highlights the benefits of both standard and constrained DPMMs compared to previously used approaches




Intelectual Propety Management

by Steve Gotz

Patents
trade marks
designs
copyright
know-how


requeriments:
novel
inventive
industrial aplication




Important to get a Lab Notebook

the pattens do not need to be perfect





2009/03/12

fixing the contract outreach with DCU

Tutorial on Grammar induction

key words:


Bayesian learning

signal '--> concept



making up a grammar:


we start with a grammar where each rule describe one of the sentence of the corpus: so there are the same number of non terminal nodes/symbols as a terminal nodes or words at the corpus.


2 operations:

Merge - > reduce the number of nodes.

Chinking - > increase the number of nodes.


Selecting operations the grammar become more and more open.



The objective is to compare trees analysis, so the accuracy is measure as how near get the tree analysis of the emerge parser compared with the tree of a stablish parser.

(??? not very clear point)


for artificial grammars it give a very good results.

For real data sets not very good.




Full information:

www.webexperiment.nl

http://turing.science.uva.nl/~jzuidema/teaching/dublin09/


First day at DCU, Grammar induction

Spainish Lecture

First spanish lecture 1.5h

2009/03/11

searching for:
summer schools: --> machine learning in cambridge 


Speaker
Talk Title
Christopher BishopIntroduction to Bayesian Inference
Zoubin GhahramaniGraphical Models
David MacKayInformation Theory
Iain MurrayMarkov Chain Monte Carlo
Bernhard SchölkopfKernel Methods
Michael LittmanReinforcement Learning
Carl Edward RasmussenGaussian Processes
John Shawe-TaylorLearning Theory
Lieven VandenbergheConvex Optimization
Andrew BlakeComputer Vision
David BleiTopic Models
Tom MinkaApproximate Inference
Yee Whye TehNonparametric Bayesian Models
Josh TenenbaumMachine Learning and Cognitive Science
Geoffrey HintonTBA
Simon GodsillParticle Filters
Phil DawidCausality
Thomas HofmannInformation Retrival
Peter OrbanzFoundations of Nonparametric Bayesian Methods
Michael JordanTBA
Speaker
Course Title
Joaquin Quiñonero CandelaGaussian Processes
Andrew FitzgibbonComputer Vision
John Winn & Tom MinkaProbabilistic Programming
Iain MurrayMarkov Chain Monte Carlo
preparing first spanish lecture.
THIMESHEET FOR RESEARCH CONTRACTS

sematic role labeling

2h meeting

sematic role labeling:
how to predict the semantic role relation:

predict the name of the relation.
predict the relation it-selve

1.- tree distance knn
matching with the k neigbours the best fit.

2.- three kernels:
analizing how looks each posible subtree as a target



2009/03/10

some  preliminar concepts:


Online Passive-Agressive Algorithms

Online Passive-Agressive Algorithms

online classification,
regrsion
uniclass problems.

singles algorithmic framework
trying to make work google book downloader:
Weka in matlab:
looking source code on internet:

knn in matlab:











in c++
Version 2.09, 15-aug-2008 (latest): Interface to XTAL regression package was added the STPRtool. XTAL implements the following regression methods:
  • Projection pursuit regression (SMART)
  • Multilayer perceptron
  • Multivariate adaptive regression splines
  • k-nearest neighbors and Constrained topological mapping
  • Constrained topological mapping
reading acm new
updating cv
fixing problems with
ACM & CNGL
Estady english 1h
pick up maps for workshop

2009/03/09

preparing spanish lecture

CoNLL-2009

sematic role labeling meeting 3h

2009/03/05

ILT3 meeting

2009/03/03

cngl

meting at dcu

2009/03/02

english class 1 hour
ILT3 Meeting 3hours.
reading foundations of statistical natural language processing