HUJI-KU at MRP~2020: Two Transition-based Neural Parsers

TL;DR

This paper presents HUJI-KU's submission to the 2020 MRP shared task, utilizing transition-based neural parsers with BERT embeddings, extending TUPA for new frameworks and languages, and exploring multitask learning.

Contribution

It generalizes TUPA to support new MRP frameworks and languages, and experiments with multitask learning using the HIT-SCIR parser.

Findings

Achieved 4th place in cross-framework and cross-lingual tracks.

Extended TUPA to support additional frameworks and languages.

Explored multitask learning with the HIT-SCIR parser.

Abstract

This paper describes the HUJI-KU system submission to the shared task on Cross-Framework Meaning Representation Parsing (MRP) at the 2020 Conference for Computational Language Learning (CoNLL), employing TUPA and the HIT-SCIR parser, which were, respectively, the baseline system and winning system in the 2019 MRP shared task. Both are transition-based parsers using BERT contextualized embeddings. We generalized TUPA to support the newly-added MRP frameworks and languages, and experimented with multitask learning with the HIT-SCIR parser. We reached 4th place in both the cross-framework and cross-lingual tracks.