Corpus linguistics has long relied on manual annotation, but as corpora expand at an unprecedented rate, this becomes a serious methodological bottleneck. The authors of a new arXiv paper address this by presenting a scalable pipeline for LLM-assisted corpus annotation, aiming to reduce the human effort required while still producing reliable linguistic data.
The pipeline is applied to the English 'consider' construction, a case study that involves tracking variation and change across large text collections. By using LLMs to assist in the annotation process, the authors hope to make it feasible to study such phenomena at a scale that would be impractical with manual methods alone.
The paper is a replacement of an earlier version, indicating ongoing refinement of the pipeline. While the abstract does not provide specific performance metrics or comparison with manual annotation, the core contribution is the pipeline itself as a response to the scalability challenge in corpus linguistics.