Finding the way from \"a to a: Sub-character morphological inflection for the SIGMORPHON 2018 Shared Task

TL;DR

This paper presents a neural network system for morphological inflection that reduces learned edit operations by using character feature equivalence classes, evaluated across multiple languages.

Contribution

It introduces a language-agnostic neural architecture with novel techniques to improve morphological inflection, building upon prior models with new ideas and parameter configurations.

Findings

Effective across diverse languages

Reduces number of learned edit operations

Analyzes advantages and drawbacks of the approach

Abstract

In this paper we describe the system submitted by UHH to the CoNLL--SIGMORPHON 2018 Shared Task: Universal Morphological Reinflection. We propose a neural architecture based on the concepts of UZH (Makarov et al., 2017), adding new ideas and techniques to their key concept and evaluating different combinations of parameters. The resulting system is a language-agnostic network model that aims to reduce the number of learned edit operations by introducing equivalence classes over graphical features of individual characters. We try to pinpoint advantages and drawbacks of this approach by comparing different network configurations and evaluating our results over a wide range of languages.