Quantifying training challenges of dependency parsers
Lauriane Aufrant, Guillaume Wisniewski, François Yvon · HAL (Le Centre pour la Communication Scientifique Directe) · 2018
Not all dependencies are equal when training a dependency parser:some are straightforward enough to be learned with only a sample ofdata, others embed more complexity. This work introduces a series ofmetrics to quantify those differences, and thereby to expose theshortcomings of various parsing algorithms and strategies. Apartfrom a more thorough comparison of parsing systems, these new toolsalso prove useful for characterizing the information conveyed bycross-lingual parsers, in a quantitative but still interpretableway.