de_STTS2_folk

de_STTS2_folk tagger

This is a spaCy language model trained to use the Stuttgart-Tübingen Tagset version 2.0, which was designed to tag transcripts of conversational speech in German. The model may be useful for tagging ASR transcripts such as those collected in the CoGS corpus.

The model was trained using the tag annotations from the FOLK corpus at https://agd.ids-mannheim.de/folk-gold.shtml, employing an 80/20 training/test split. Tokens in the training data for the model were converted to lower case prior to traning to match the format used for automatic speech recognition transcripts on YouTube, as of early 2023.

Usage example:

!pip install https://huggingface.co/stcoats/de_STTS2_folk/resolve/main/de_STTS2_folk-any-py3-none-any.whl
import spacy
import de_STTS2_folk
nlp = de_STTS2_folk.load()
doc = nlp("ach so meinst du wir sollen es jetzt tun")
for token in doc:
    print(token.text, token.tag_)

References

Coats, Steven. (2023). A new corpus of geolocated ASR transcripts from Germany. <i>Language Resources and Evaluation</i>. https://doi.org/10.1007/s10579-023-09686-9

Westpfahl, Swantje and Thomas Schmidt. (2016): FOLK-Gold – A GOLD standard for Part-of-Speech-Tagging of Spoken German. In: <i>Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), Portorož, Slovenia</i>

---

tags:

• spacy
• token-classification language:
• de model-index:
• name: de_STTS2_folk results:
  • task: name: TAG type: token-classification metrics:
    • name: TAG (XPOS) Accuracy type: accuracy value: 0.9191333537

---

Label Scheme

<details> <summary>View label scheme (62 labels for 1 components)</summary> </details>

Accuracy