In their study, the authors found that the choice of pre-trained embeddings significantly affects parser performance, with XLM-R emerging as a robust option across various languages. They also observed that adding LSTM layers did not provide additional benefits when using transformer-based embeddings. Based on these insights, they proposed a simplified parser architecture that achieved state-of-the-art results in 10 out of 12 languages studied.
The code for the STEPS parser is available in the GitHub repository:
https://github.com/bewakes/steps-parser.
The repository contains our implementation of companion code for STEPS, the modular Universal Dependencies parser described in the paper:
Stefan Grünewald, Annemarie Friedrich, and Jonas Kuhn (2020): Applying Occam’s Razor to Transformer-Based Dependency Parsing: What Works, What Doesn’t, and What is Really Necessary. arXiv:2010.12699
The code allows users to reproduce and extend the results reported in the study. Please cite the above paper when using the code, and direct any questions or feedback regarding the parser to Stefan Grünewald.
Disclaimer: This software is a research prototype, solely developed for and published as part of the publication cited above. It will neither be maintained nor monitored in any way. For more detailed information, you can refer to the full paper.
Project Github
URL: https://github.com/bewakes/steps-parser