Table of Contents

Unsupervised methods for head assignments, EACL 2009

Federico Sangati and Willem Zuidema

Presented by: Zdeněk Žabokrtský
Report by: Eduard Bejček and Lasha Abzianidze

Introduction

What do we dislike about the paper

this is the case as a) this is the point of greedy technique to be theoretically able to result any possible head annotations and b) only in this case there is threat (mentioned in article) of having exponential number of elementary trees for a sentence of length <latex>n</latex> but it is less than <latex>n^2</latex>, namely <latex>n\log(n)</latex>, as for each lexical anchor we can have at most <latex>log(n)</latex> different elementary trees. -Lasha-
While extracting the elementary trees from a head annotated sentence we are first extracting spines and than they yield elementary trees (note that in this process elementary trees and spines are in one-to-one correspondence). Hence the authors means these spines in respective spines. On figures 3 and 2 there are elementary trees and their respective spines depicted. -Lasha-

What do we like about the paper