Current suggested pipeline to count verbs

• Spelling correction with LLMS -> POS tagging with stanza -> Count words tagged with VERB

Thoughts

• Difference between spelling and grammatical mistakes, but choosing the correct word form depends on grammar
• Alternatives for spelling correction: https://github.com/wolfgarbe/SymSpell
• To evaulate performance of pipeline, some manually labeled data is necessary
  • manual POS tagging could be similar to: https://labelstud.io/templates/named_entity or https://labelstud.io/templates/relation_extraction
  • word spelling corrected could be pre-labeled by using a dictionary to find unknown words (however does not work for word form/grammatical mistakes)
• Everything depends on spelling correction

Setup

1. Install uv
2. Install dependencies:

   uv sync
   

3. Optionally download spacy dependencies:

    uv run spacy download de_core_news_lg
   

4. For dspy LLM functionality, add .env

   OR_KEY=sk-or-v1-key