# Sequence Labeling Parsing by Learning Across Representations

## TL;DR

This paper presents a multitask learning approach that uses sequence labeling to jointly learn constituency and dependency parsing, improving performance with minimal additional cost.

## Contribution

It introduces a unified sequence labeling framework for both parsing paradigms and demonstrates that auxiliary tasks enhance parsing accuracy.

## Key findings

- MTL models outperform single-task models in parsing accuracy
- Auxiliary losses improve constituency parsing by 1.14 F1 points
- Auxiliary losses improve dependency parsing by 0.62 UAS points

## Abstract

We use parsing as sequence labeling as a common framework to learn across constituency and dependency syntactic abstractions. To do so, we cast the problem as multitask learning (MTL). First, we show that adding a parsing paradigm as an auxiliary loss consistently improves the performance on the other paradigm. Secondly, we explore an MTL sequence labeling model that parses both representations, at almost no cost in terms of performance and speed. The results across the board show that on average MTL models with auxiliary losses for constituency parsing outperform single-task ones by 1.14 F1 points, and for dependency parsing by 0.62 UAS points.