Update README.md

README.md

@@ -7,7 +7,9 @@ size_categories:
 - 1K<n<10K
 ---
 
-# Dataset Description
+# Spellcheck Benchmark
+
+## Dataset Description
 
 Spellcheck Benchmark is a collection of datasets dedicated to spelling correction problem for Russian language. 
 
@@ -15,8 +17,29 @@ Spellcheck Benchmark is a collection of datasets dedicated to spelling correctio
 
 **Paper** - <link>
 
-# Dataset Summary
+## Dataset Summary
+
+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, reviews and more. 
+
+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. 
+Manual labeling scheme accounts for jargonisms, collocations and common language, hence in some cases it encourages annotators not to amend a word
+in favor of preserving style of a text. 
+
+## Supported Tasks and Leaderboards
+
+- Automatic spelling correction
+
+## Dataset Structure
+
+### Splits
+
+The benchmark consists of the following splits that reflect corresponding dataset:
 
-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. 
+- **MultidomainGoldSet**: collected and annotated by SalutDevices Team;
+- **RUSpellRU**: by Sorokin et al.(2016);
+- **MedSpellCheck**: ...;
+- **GithubTypos**: ...;
 
-All datasets were passed through two-stage manual labeling pipeline.
+Depending on which