holylovenia commited on
Commit
34d1875
1 Parent(s): 7f18583

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +78 -0
README.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - coreference-resolution
5
+ language:
6
+ - ind
7
+ ---
8
+
9
+ # indocoref
10
+
11
+ Dataset contains articles from Wikipedia Bahasa Indonesia which fulfill these conditions:
12
+
13
+ - The pages contain many noun phrases, which the authors subjectively pick: (i) fictional plots, e.g., subtitles for films,
14
+
15
+ TV show episodes, and novel stories; (ii) biographies (incl. fictional characters); and (iii) historical events or important events.
16
+
17
+ - The pages contain significant variation of pronoun and named-entity. We count the number of first, second, third person pronouns,
18
+
19
+ and clitic pronouns in the document by applying string matching.We examine the number
20
+
21
+ of named-entity using the Stanford CoreNLP
22
+
23
+ NER Tagger (Manning et al., 2014) with a
24
+
25
+ model trained from the Indonesian corpus
26
+
27
+ taken from Alfina et al. (2016).
28
+
29
+ The Wikipedia texts have length of 500 to
30
+
31
+ 2000 words.
32
+
33
+ We sample 201 of pages from subset of filtered
34
+
35
+ Wikipedia pages. We hire five annotators who are
36
+
37
+ undergraduate student in Linguistics department.
38
+
39
+ They are native in Indonesian. Annotation is carried out using the Script d’Annotation des Chanes
40
+
41
+ de Rfrence (SACR), a web-based Coreference resolution annotation tool developed by Oberle (2018).
42
+
43
+ From the 201 texts, there are 16,460 mentions
44
+
45
+ tagged by the annotators
46
+
47
+ ## Dataset Usage
48
+
49
+ Run `pip install nusacrowd` before loading the dataset through HuggingFace's `load_dataset`.
50
+
51
+ ## Citation
52
+
53
+ ```
54
+ @inproceedings{artari-etal-2021-multi,
55
+ title = {A Multi-Pass Sieve Coreference Resolution for {I}ndonesian},
56
+ author = {Artari, Valentina Kania Prameswara and Mahendra, Rahmad and Jiwanggi, Meganingrum Arista and Anggraito, Adityo and Budi, Indra},
57
+ year = 2021,
58
+ month = sep,
59
+ booktitle = {Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)},
60
+ publisher = {INCOMA Ltd.},
61
+ address = {Held Online},
62
+ pages = {79--85},
63
+ url = {https://aclanthology.org/2021.ranlp-1.10},
64
+ abstract = {Coreference resolution is an NLP task to find out whether the set of referring expressions belong to the same concept in discourse. A multi-pass sieve is a deterministic coreference model that implements several layers of sieves, where each sieve takes a pair of correlated mentions from a collection of non-coherent mentions. The multi-pass sieve is based on the principle of high precision, followed by increased recall in each sieve. In this work, we examine the portability of the multi-pass sieve coreference resolution model to the Indonesian language. We conduct the experiment on 201 Wikipedia documents and the multi-pass sieve system yields 72.74{\%} of MUC F-measure and 52.18{\%} of BCUBED F-measure.}
65
+ }
66
+ ```
67
+
68
+ ## License
69
+
70
+ MIT
71
+
72
+ ## Homepage
73
+
74
+ [https://github.com/valentinakania/indocoref/](https://github.com/valentinakania/indocoref/)
75
+
76
+ ### NusaCatalogue
77
+
78
+ For easy indexing and metadata: [https://indonlp.github.io/nusa-catalogue](https://indonlp.github.io/nusa-catalogue)