Towards JointUD: Part-of-speech Tagging and Lemmatization using Recurrent Neural Networks

TL;DR

This paper presents a multitask neural network model that jointly performs POS tagging and lemmatization using extended LSTM architecture, demonstrating the approach's potential despite current performance gaps.

Contribution

It introduces a novel joint neural network architecture for POS tagging and lemmatization, extending LSTM models to generate character-level sequences for multiple tasks.

Findings

The model successfully performs joint POS tagging and lemmatization.

Performance is promising but still below state-of-the-art levels.

Multitask learning shows potential for NLP tasks.

Abstract

This paper describes our submission to CoNLL 2018 UD Shared Task. We have extended an LSTM-based neural network designed for sequence tagging to additionally generate character-level sequences. The network was jointly trained to produce lemmas, part-of-speech tags and morphological features. Sentence segmentation, tokenization and dependency parsing were handled by UDPipe 1.2 baseline. The results demonstrate the viability of the proposed multitask architecture, although its performance still remains far from state-of-the-art.