Optimizing Planar and 2-Planar Parsers with MaltOptimizer

Miguel Ballesteros , Carlos Gómez Rodríguez , Joakim Nivre


MaltOptimizer is a tool that is capable of finding an optimal configuration
for MaltParser models, taking into account that nowadays dependency parsers require careful
tuning in order to obtain state-of-the-art results, and this tuning is normally based on
specialized knowledge. The Planar and 2-Planar parsers are two different parsing algorithms
included in MaltParser. In the present paper, we show how these two parsers can be included in
MaltOptimizer processes comparing them with the rest of MaltParser algorithm families,
and how we can define a deep feature search and selection by using MaltOptimizer for
these two algorithms. The experiments show that by using MaltOptimizer we can improve parsing
accuracy for Planar and 2-Planar parsers by up to 8 percent absolute (labeled attachment
score) compared to default settings.

Texto completo: