A Modest Pareto Optimisation Analysis of Dependency Parsers in 2021
This work provides an incremental analysis of parser efficiency for NLP researchers, focusing on practical trade-offs without major innovations.
The study compared three leading dependency parser systems across a diverse subset of languages to assess their accuracy-efficiency trade-offs, finding that biaffine parsing offers a balanced default choice while sequence-labelling parsing is better for inference speed.
We evaluate three leading dependency parser systems from different paradigms on a small yet diverse subset of languages in terms of their accuracy-efficiency Pareto front. As we are interested in efficiency, we evaluate core parsers without pretrained language models (as these are typically huge networks and would constitute most of the compute time) or other augmentations that can be transversally applied to any of them. Biaffine parsing emerges as a well-balanced default choice, with sequence-labelling parsing being preferable if inference speed (but not training energy cost) is the priority.