# On Multilingual Training of Neural Dependency Parsers

## TL;DR

This paper demonstrates that joint multilingual training of a neural dependency parser, which uses orthographic representations, improves parsing performance across related languages and reveals linguistic properties learned by the model.

## Contribution

It introduces a multilingual neural dependency parser trained on orthographic inputs and analyzes the linguistic features learned by the model.

## Key findings

- The parser learns to associate Latin and Cyrillic characters.
- It groups words with similar grammatical functions across languages.
- The model achieves competitive results on Universal Dependencies datasets.

## Abstract

We show that a recently proposed neural dependency parser can be improved by joint training on multiple languages from the same family. The parser is implemented as a deep neural network whose only input is orthographic representations of words. In order to successfully parse, the network has to discover how linguistically relevant concepts can be inferred from word spellings. We analyze the representations of characters and words that are learned by the network to establish which properties of languages were accounted for. In particular we show that the parser has approximately learned to associate Latin characters with their Cyrillic counterparts and that it can group Polish and Russian words that have a similar grammatical function. Finally, we evaluate the parser on selected languages from the Universal Dependencies dataset and show that it is competitive with other recently proposed state-of-the art methods, while having a simple structure.