You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Korean Effective Crowdsourcing of Multiple Tasks (ECMT) for Comprehensive Knowledge Extraction

Details

Annotated text from Korean Wikipedia and KBox (Korean DBpedia). Includes a crowd sourced training set and expert annotated (reviewed by four experts) test set.

The dataset was annotated by crowdworks in multiple stages.

  • Phase I: entity mention detection annotation; candidate entity mentions are selected in a text
  • Phase II: entity linking annotation; candidate mentions can be linked to a knowledge base
  • Phase III: coreference annotation; entities can be linked to pronouns, demonstrative determiners, and antecedent mentions
  • Phase IV: relation extraction annotation; relations between entities are annotated

Annotation Notes

Phase I

  • For each mention, the annotator selects a category from one of 16 options: person, study field, theory, artifact, organization, location, civilization, event, year, time, quantity, job, animal, plant, material, and term.
  • Entities can be things, concepts, ideas, or events:
๊ฐœ์ฒด๋ž€ ๋‹ค๋ฅธ ๊ฒƒ๋“ค๊ณผ ๋ถ„๋ฆฌ๋˜์–ด ์กด์žฌํ•˜๋Š” ๊ฒƒ์œผ๋กœ, ๊ฐœ์ฒด๋Š” ๋ฌผ์งˆ์  ์กด์žฌ์ผ ํ•„์š”๋Š” ์—†์œผ๋ฉฐ ๊ฐœ๋…์  ์•„์ด๋””์–ด ํ˜น์€ ์‚ฌ๊ฑด๋„ ๋  ์ˆ˜ ์žˆ๋‹ค ๊ฐœ์ฒด์˜ ๋Œ€ํ‘œ์ ์ธ ๋ฒ”์ฃผ์—๋Š” ์‚ฌ๋žŒ, ๋ฌผ์ฒด, ์กฐ์ง, ๊ธฐ๊ด€, ์žฅ์†Œ, ์‹œ๊ฐ„, ์‚ฌ๊ฑด ๋“ฑ์ด ํฌํ•จ๋œ๋‹ค
  • Compound nouns are tagged with the largest span:
๋ณตํ•ฉ๋ช…์‚ฌ์ธ ๊ฒฝ์šฐ ๊ฐ€์žฅ ๋„“์€ ๋‹จ์œ„๋กœ ํƒœ๊น…ํ•ด์ฃผ์„ธ์š” ex) [์ƒํ•˜์ด] [๋””์ฆˆ๋‹ˆ๋žœ๋“œ] -> [์ƒํ•˜์ด ๋””์ฆˆ๋‹ˆ๋žœ๋“œ]
  • Final result is created by merging annotations from two separate annotators.

Phase II

  • For each mention, a list of candidates from the knowledge base are shown. The annotator can select a candidate, not in candidate list, or not an entity.
  • Each document was annotated by a single annotator.

Phase III

  • For each mention, the annotator can select a preceding mention, no antecedent, or error. Noun phrases and pronouns are extracted using the parse information.
  • "We scaled down the coreference resolution by limiting the scope of the target mentions to a named entity, pronoun, and definite noun phrase."
  • Postfixes particles (์กฐ์‚ฌ) are not included in the antecedent:
[์ž‘์—…๋Œ€์ƒ] ์•„๋ž˜ ํ•ญ๋ชฉ์—์„œ ์กฐ์‚ฌ๋“ฑ์„ ์ œ์™ธ(๊ต์ •)ํ•ด ์ฃผ์„ธ์š”. ๊ทธ๋…€๋Š” -> ๊ทธ๋…€

Citation

@inproceedings{nam-etal-2020-effective,
    title = "Effective Crowdsourcing of Multiple Tasks for Comprehensive Knowledge Extraction",
    author = "Nam, Sangha  and
      Lee, Minho  and
      Kim, Donghwan  and
      Han, Kijong  and
      Kim, Kuntae  and
      Yoon, Sooji  and
      Kim, Eun-kyung  and
      Choi, Key-Sun",
    editor = "Calzolari, Nicoletta  and
      B{\'e}chet, Fr{\'e}d{\'e}ric  and
      Blache, Philippe  and
      Choukri, Khalid  and
      Cieri, Christopher  and
      Declerck, Thierry  and
      Goggi, Sara  and
      Isahara, Hitoshi  and
      Maegaard, Bente  and
      Mariani, Joseph  and
      Mazo, H{\'e}l{\`e}ne  and
      Moreno, Asuncion  and
      Odijk, Jan  and
      Piperidis, Stelios",
    booktitle = "Proceedings of the Twelfth Language Resources and Evaluation Conference",
    month = may,
    year = "2020",
    address = "Marseille, France",
    publisher = "European Language Resources Association",
    url = "https://aclanthology.org/2020.lrec-1.27",
    pages = "212--219",
    abstract = "Information extraction from unstructured texts plays a vital role in the field of natural language processing. Although there has been extensive research into each information extraction task (i.e., entity linking, coreference resolution, and relation extraction), data are not available for a continuous and coherent evaluation of all information extraction tasks in a comprehensive framework. Given that each task is performed and evaluated with a different dataset, analyzing the effect of the previous task on the next task with a single dataset throughout the information extraction process is impossible. This paper aims to propose a Korean information extraction initiative point and promote research in this field by presenting crowdsourcing data collected for four information extraction tasks from the same corpus and the training and evaluation results for each task of a state-of-the-art model. These machine learning data for Korean information extraction are the first of their kind, and there are plans to continuously increase the data volume. The test results will serve as an initiative result for each Korean information extraction task and are expected to serve as a comparison target for various studies on Korean information extraction using the data collected in this study.",
    language = "English",
    ISBN = "979-10-95546-34-4",
}
Downloads last month
0
Edit dataset card

Collection including coref-data/korean_ecmt_raw