A Falta de Pan, Buenas Son Tortas: The Efficacy of Predicted UPOS Tags for Low Resource UD Parsing

TL;DR

This paper investigates how predicted UPOS tags influence dependency parsing performance in low-resource settings, showing they are helpful with limited data but less so as data size grows.

Contribution

It provides an empirical evaluation of predicted UPOS tags' effectiveness across various low-resource scenarios in universal dependency parsing.

Findings

Predicted UPOS tags improve parsing in very low-resource settings.

The positive effect of UPOS tags decreases with larger treebank sizes.

Predicted tags are especially beneficial when annotated data is scarce.

Abstract

We evaluate the efficacy of predicted UPOS tags as input features for dependency parsers in lower resource settings to evaluate how treebank size affects the impact tagging accuracy has on parsing performance. We do this for real low resource universal dependency treebanks, artificially low resource data with varying treebank sizes, and for very small treebanks with varying amounts of augmented data. We find that predicted UPOS tags are somewhat helpful for low resource treebanks, especially when fewer fully-annotated trees are available. We also find that this positive impact diminishes as the amount of data increases.