SIGTYP 2020 Shared Task: Prediction of Typological Features

TL;DR

This paper presents a shared task on predicting typological features of languages to automatically enrich knowledge bases like WALS, highlighting the challenges and the effectiveness of feature correlation-based methods.

Contribution

It introduces a shared task with multiple submissions focusing on predicting typological features, demonstrating the potential and limitations of current approaches.

Findings

Feature correlation-based methods perform best.

Most systems struggle with languages having few known features.

Error analysis reveals challenges in low-resource scenarios.

Abstract

Typological knowledge bases (KBs) such as WALS (Dryer and Haspelmath, 2013) contain information about linguistic properties of the world's languages. They have been shown to be useful for downstream applications, including cross-lingual transfer learning and linguistic probing. A major drawback hampering broader adoption of typological KBs is that they are sparsely populated, in the sense that most languages only have annotations for some features, and skewed, in that few features have wide coverage. As typological features often correlate with one another, it is possible to predict them and thus automatically populate typological KBs, which is also the focus of this shared task. Overall, the task attracted 8 submissions from 5 teams, out of which the most successful methods make use of such feature correlations. However, our error analysis reveals that even the strongest submitted systems struggle with predicting feature values for languages where few features are known.