Resolving prepositional phrase attachment ambiguities in Spanish with a classifier

Resumen

In this paper we present a classifier that solves a certain kind
of ambiguities in syntactic structure for Spanish, namely, ambiguities as
to the point of adjunction of a prepositional phrase in the syntactic
structure of a sentence (PP attachment).
As a starting point, we used EsTxala dependency grammar for Spanish,
integrated within FreeLing, with an accuracy score of 61% on PP adjunc-
tion. Our target is to develop a specialized module for for PP attachment,so that the syntactic analyzer combines dependency grammars manualrules with statistical information infered out of a classifier.
We have evaluated different classifiers and different features to char-
acterize PP-attachment ambiguities. Our best approaches improve the
performance of EsTxala by 20 points, but are still far from the per-
formance of unsupervised methods reporting 94% accuracy. We gained
insight on the factors governing the disambiguation of PP attachment
ambiguities, which will arguably let us build lighter models that can be
easily integrated within a general-purpose analyzer as FreeLing.