(Kethu: Treebank "Constituency" Bahasa Indonesia dalam Format Penn Treebank)
We revised and converted an Indonesian constituency treebank built by Dinakaramani et al. (2015) to the Penn Treebank format, a de-facto standard in creating the constituency treebank. We named this treebank Kethu, a name of a forest in Wonosobo. Why a forest? Because the treebank contains parse trees, as the forest contains trees ;-)
Characteristics of the treebank:
- Number of sentences: 1030
- Average sentence length: 27.4 tokens/sentence
- Genre: news in formal Indonesian
-
2020-10-27 v2.0
- revised the POS tagging and syntactic annotation for compound words
- revised the tokenization, POS tagging and syntactic annotation for predicate nominalization cases
- revised the syntactic annotation for noun phrases containing numerical expression (especially QP)
- changed the POS tag for 'Rp' and the similar tokens from SYM to NN
-
2019-09-01 v1.0
- revised the bracketing format to PTB format
- revised the POS tag set and adjusted the POS tagging
- resolved missing punctuations, typo, incorrect reported speech annotation, etc.
- Ika Alfina and Jessica Naraiswari Arwidarasti
The explanation on why and how we convert that treebank can be read in our paper:
-
Jessica Naraiswari Arwidarasti, Ika Alfina, and Adila Alfa Krisnadhi. "Adjusting Indonesian Compound Words Annotation to the Penn Treebank Format". In the Proceeding of The 2019 International Conference of Asian Language Processing (IALP) 2020 in Kuala Lumpur, Malaysia, 4-6 Desember 2020. (accepted)
-
Jessica Naraiswari Arwidarasti, Ika Alfina, and Adila Alfa Krisnadhi. "Converting an Indonesian Constituency Treebank to the Penn Treebank Format". In the Proceeding of The 2019 International Conference of Asian Language Processing (IALP 2019) in Shanghai, China, 15-17 November 2019.