ai-forever commited on
Commit
4038365
1 Parent(s): 3ee777f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +27 -4
README.md CHANGED
@@ -7,7 +7,9 @@ size_categories:
7
  - 1K<n<10K
8
  ---
9
 
10
- # Dataset Description
 
 
11
 
12
  Spellcheck Benchmark is a collection of datasets dedicated to spelling correction problem for Russian language.
13
 
@@ -15,8 +17,29 @@ Spellcheck Benchmark is a collection of datasets dedicated to spelling correctio
15
 
16
  **Paper** - <link>
17
 
18
- # Dataset Summary
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
 
20
- Spellcheck Benchmark includes four datasets, each of which consists of pairs of sentences in Russian language. Each pair embodies sentence, which may contain spelling errors, and its corresponding correction. Datasets were gathered from various sources and domains including social networks, internet blogs, github commits, medical anamnesis, literature, news and jurisprudence texts.
 
 
 
21
 
22
- All datasets were passed through two-stage manual labeling pipeline.
 
7
  - 1K<n<10K
8
  ---
9
 
10
+ # Spellcheck Benchmark
11
+
12
+ ## Dataset Description
13
 
14
  Spellcheck Benchmark is a collection of datasets dedicated to spelling correction problem for Russian language.
15
 
 
17
 
18
  **Paper** - <link>
19
 
20
+ ## Dataset Summary
21
+
22
+ Spellcheck Benchmark includes four datasets, each of which consists of pairs of sentences in Russian language.
23
+ Each pair embodies sentence, which may contain spelling errors, and its corresponding correction.
24
+ Datasets were gathered from various sources and domains including social networks, internet blogs, github commits, medical anamnesis, literature, news, reviews and more.
25
+
26
+ All datasets were passed through two-stage manual labeling pipeline. The correction of a sentence is defined by an agreement of at least two human annotators.
27
+ Manual labeling scheme accounts for jargonisms, collocations and common language, hence in some cases it encourages annotators not to amend a word
28
+ in favor of preserving style of a text.
29
+
30
+ ## Supported Tasks and Leaderboards
31
+
32
+ - Automatic spelling correction
33
+
34
+ ## Dataset Structure
35
+
36
+ ### Splits
37
+
38
+ The benchmark consists of the following splits that reflect corresponding dataset:
39
 
40
+ - **MultidomainGoldSet**: collected and annotated by SalutDevices Team;
41
+ - **RUSpellRU**: by Sorokin et al.(2016);
42
+ - **MedSpellCheck**: ...;
43
+ - **GithubTypos**: ...;
44
 
45
+ Depending on which