mpsilfve / finer-data

finer-data

Introduction

The directory data contains a corpus of Finnish technology related news articles with a manually prepared named entity annotation (digitoday.2014.csv). The text material was extracted from the archives of Digitoday, a Finnish online technology news source (www.digitoday.fi). The corpus consists of 953 articles (193,742 word tokens) with six named entity classes (organization, location, person, product, event, and date). The corpus is available for research purposes and can be readily used for development of NER systems for Finnish. The corpus is described in the article

"A Finnish News Corpus for Named Entity Recognition"

@article{ruokolainen2019finnish,
  title={A finnish news corpus for named entity recognition},
  author={Ruokolainen, Teemu and Kauppinen, Pekka and Silfverberg, Miikka and Lind{\'e}n, Krister},
  journal={Language Resources and Evaluation},
  pages={1--26},
  year={2019},
  publisher={Springer}
}

Experiments

The repository also contains the Digitoday (digitoday.2015.test.csv) and Wikipedia (wikipedia.test.csv) evaluation sets employed in the experiments of the article

"A Finnish News Corpus for Named Entity Recognition"

The training and development sets formed of digitoday.2014.csv employed in the experiments can be found from files digitoday.2014.train.csv and digitoday.2014.dev.csv, respectively.

Finally, the directory experiments contains the predictions of systems FiNER, Gungor-NN, and Sohrab-NN in the paper "A Finnish News Corpus for Named Entity Recognition" on the Digitoday and Wikipedia evaluation sets.

FiNER:

urn.fi/urn:nbn:fi:lb-2018091301

Gungor-NN:

@InProceedings{C18-1177,
  author = 	"G{\"{u}}ng{\"{o}}r, Onur
		and {\"{U}}sk{\"{u}}darli, Suzan
		and G{\"{u}}ng{\"{o}}r, Tunga",
  title = 	"Improving Named Entity Recognition by Jointly Learning to Disambiguate Morphological Tags",
  booktitle = 	"Proceedings of the 27th International Conference on Computational Linguistics",
  year = 	"2018",
  publisher = 	"Association for Computational Linguistics",
  pages = 	"2082--2092",
  location = 	"Santa Fe, New Mexico, USA",
  url = 	"http://aclweb.org/anthology/C18-1177"
}

Sohrab-NN:

@InProceedings{D18-1309,
  author = 	"Sohrab, Mohammad Golam
		and Miwa, Makoto",
  title = 	"Deep Exhaustive Model for Nested Named Entity Recognition",
  booktitle = 	"Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
  year = 	"2018",
  publisher = 	"Association for Computational Linguistics",
  pages = 	"2843--2849",
  location = 	"Brussels, Belgium",
  url = 	"http://aclweb.org/anthology/D18-1309"
}

License

The Digitoday material is licensed under CC BY-ND-NC 1.0 and the Wikipedia material is licensed under CC BY-SA 3.0

finer-data's Issues

Sohrab-NN code?

Dear authors,

You have used the neural model from the paper "Deep Exhaustive Model for Nested Named Entity Recognition". How did you get the code for that model? Did you implement it from scratch?

Thanks

NE sequences starting with I- instead of B-

There are a few NE sequences that start with I- instead of B- (as most do) in the files in data/. For example, the following appears in data/digitoday.2014.train.csv

520   I-PRO   O
.     O       O
  
Lumia I-PRO   O
520:n I-PRO   O
osuus O       O
käytössä      O       O

(I'd be happy to dig out the details for the rest if there's interest in addressing this issue.)

Tokenization errors

At least the following appear in the data:

• Gam esIndustry-julkaisu
• ki rjoitettu
• myytyynYounitediin
• tal lennustilaa
• jaNokia
• televisionkatselu un
• Lumia-puheli mia
• koneveikoll e
• ostavatselaimia
• palkankorotustoiveistapomolle
• verta si
• sal asanoista
• vit sailtu
• viittausPohjois-Koreaan
• ki innostusta
• Kicks tarterissa
• odottama an
• liikent een
• le vittämiseen
• ko konaisuudessaan
• verkkos ivustolta

A quick google indicates that at least some of these were not mistakes in the original texts.

If there's interest in addressing these, I'd be happy to submit a pull request for the cases I'm aware of.