LowResourceEval-2019: a shared task on morphological analysis for low-resource languages

TL;DR

This paper reports on a shared task evaluating morphological analysis methods for low-resource Russian languages, highlighting machine learning approaches that outperform rule-based systems and providing datasets in a standardized format.

Contribution

It introduces a shared task for morphological analysis in low-resource languages and provides datasets in CONLL-U format for broader NLP use.

Findings

Machine learning approaches outperform rule-based methods

Participation included four teams with diverse solutions

Datasets are standardized in CONLL-U format

Abstract

The paper describes the results of the first shared task on morphological analysis for the languages of Russia, namely, Evenki, Karelian, Selkup, and Veps. For the languages in question, only small-sized corpora are available. The tasks include morphological analysis, word form generation and morpheme segmentation. Four teams participated in the shared task. Most of them use machine-learning approaches, outperforming the existing rule-based ones. The article describes the datasets prepared for the shared tasks and contains analysis of the participants' solutions. Language corpora having different formats were transformed into CONLL-U format. The universal format makes the datasets comparable to other language corpura and facilitates using them in other NLP tasks.