https://aclanthology.org/2021.nodalida-main.20/
In 2020, the Stanza tool was trained and tested on a set of gold-standard Swedish corpora (following SUC3-style annotation) in order to create a high-quality analysis. Currently (in 2024), this is the default analysis for Swedish in Språkbanken's analysis platform Sparv.
This analysis is used with Sparv. Check out Sparv's quick start guide to get started!
To use this analysis, add the following line under export.annotations in the Sparv corpus configuration file:
- <token>:stanza.pos # Part-of-speech tags
For more info on how to use Sparv, check out the Sparv documentation.
Example output:
<token pos="PN">Det</token>
<token pos="AB">här</token>
<token pos="VB">är</token>
<token pos="DT">en</token>
<token pos="NN">korpus</token>
<token pos="MAD">.</token>
For a model trained on SUC3 and validated on a part of TalbankenSBX_dev the results are as follows:
tested on Talbanken SBX_test: exact match = 0.97; POS = 0.98; msd = 0.99
tested on SIC2: exact match = 0.92; POS = 0.93; msd = 0.96
More info: https://spraakbanken.gu.se/en/resources/flair/evaluating-pos-tagging
Stanza: Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton and Christopher D. Manning. 2020
Stanza: A Python Natural Language Processing Toolkit for Many Human Languages. In Association for Computational Linguistics (ACL) System Demonstrations. 2020
SUC3: https://spraakbanken.gu.se/en/resources/suc3
TalbankenSBX: https://spraakbanken.gu.se/en/blog/20200609-the-five-lives-of-talbanken
SIC2: https://spraakbanken.gu.se/en/resources/sic2