Awesome

Summary

UD_Swiss_German-UZH is a tiny manually annotated treebank of 100 sentences in different Swiss German dialects and a variety of text genres.

Introduction

This repository presents work on Universal Dependency Parsing for Swiss German, which has been done as part of the Master’s Thesis Parsing Approaches for Swiss German at the Institute for Computational Linguistics at the University of Zurich.

The provided resources consist of 100 Swiss German sentences (from different sources), manually annotated with part-of-speech tags and universal dependencies.

NOAH’s Corpus of Swiss German Dialects was used as a resource for the Swiss German part-of-speech tagging process - a process preliminary to parsing.

Annotation

In general, we follow the German guidelines UD for German.

The POS annotations are generally based on the German guidelines, namely the Stuttgart-Tübingen-TagSet (STTS) and some changes according to the TIGER annotation scheme. Furthermore, dealing with Swiss German, there is the need for an additional POS tag PTKINF, not present in the STTS tagset, as well as for the "meta tag" TAG+.

• PTKINF is an infinitive particle which does not exist in Standard German but is frequently used in dialects. It comes in the form of go, cho, goge, lo to name a few, as in Si gönd go poschte. (They go shopping.) In the Standard German translation, Sie gehen einkaufen., we can see that there is no equivalent.
• TAG+ is used to handle merged words; we introduced the “+“-sign which can be added to any PoS tag. In the STTS there is one tag like this: the APPRART, used for combinations of articles and prepositions like im consisting of in + dem (in the). However, in Swiss German these kind of merges are performed with any kind of words and just merging the tags would result in a big tagset. Therefore we decided to use the "head" of the word or the first word as tag and simply add a plus to show that this word incorporates another one Hollenstein and Aepli, 2014. Like this, they can easily be found and, if needed, manually expanded. Frequent examples of such words include hemmer (haben + wir), häts (hat + es), and sinz (sind + sie), for we have, it has and they are.

The Universal Dependency POS (UPOS) tags are converted according to the mapping provided by the Universal Dependency. Additionaly:

• PTKINF are converted to PART
• the plus sign in TAG+ are disgarded

Acknowledgments

This work has been performed at the University of Zurich by Noëmi Aepli with the help of Simon Clematide.

References

• Citation:

@inproceedings{aepli2018parsing,
  title={{Parsing Approaches for Swiss German}},
  author={No\"emi Aepli and Simon Clematide},
  booktitle={{Proceedings of the 3rd Swiss Text Analytics Conference (SwissText), Winterthur, Switzerland}},
  year={2018}
}

• Download Paper
• Download Master's Thesis

Changelog

• 2019-11-15 v2.5
  • Initial release in Universal Dependencies.