KurdNet is the the Kurdish WordNet. WordNet has been used in numerous natural language processing tasks such as word sense disambiguation and information extraction with considerable success. Motivated by this success, many projects have been under taken to build similar lexical databases for other languages.

The current version of KurdNet is created following the Expand Approach, which is carried out semi-automatically and centred around building a Kurdish alignment for Base Concepts, i.e. the core subset of major meanings in the Princeton WordNet. Additionally, we use a bilingual dictionary and simple set theory operations to translate and align synsets and use a corpus to extract usage examples. The effectiveness of our prototype database is evaluated via measuring its impact on a Kurdish information retrieval task.

Base Concepts' Entity-Relationship schema in WordNet

The following table shows the main statistical properties of Base Concepts and its alignment in KurdNet:

Folder ckb contains OMW compatible files of KurdNet. Unlike definitions in Sorani Kurdish which are translated manually by experts, lemmata are collected semi-automatically. Therefore, it is possible to find incorrect cases in the provided lemmata. This is something that should be more focused on in the next version of the resource. According the our experiments, the semi-automatic minimal alignment (referred to as min) outperforms the maximal alignment (max). Therefore, only lemmata which are semi-automatically aligned based on the min technique are provided in the Open Multilingual WordNet version.

The current version of KurdNet only contains synsets and translations in Sorani. Since 2014 when we initiated this project, we have been looking for people interested in translating the current definitions and synsets into other variants of Kurdish, particularly Kurmanji. If you can contribute to this project, please get in touch. You can take a look at the glosses to be translated in the Translated_Glosses.tsv file.

Note that the .tsv files are exported based on the original Excel files and some of them are cleaned, e.g. ZWNJ removed. It is recommended to work on the .tsv files for future developments.

If you use this resource, please cite our paper:

@inproceedings{W14-0101,
  title = "Towards Building KurdNet, the Kurdish WordNet",
  author = "Aliabadi, Purya  and
    Ahmadi, Sina  and
    Salavati, Shahin  and
    Esmaili, Kyumars Sheykh",
  booktitle = "Proceedings of the Seventh Global Wordnet Conference",
  year = "2014",
  address = "Tartu, Estonia",
  url = "https://www.aclweb.org/anthology/W14-0101",
  pages = "1--6",
}