ProCreations commited on
Commit
0c0abcb
·
verified ·
1 Parent(s): a54fe75

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +62 -3
README.md CHANGED
@@ -1,3 +1,62 @@
1
- ---
2
- license: mit
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ task_categories:
4
+ - text-generation
5
+ language:
6
+ - en
7
+ tags:
8
+ - typo
9
+ pretty_name: M2M
10
+ size_categories:
11
+ - 10K<n<100K
12
+ ---
13
+
14
+ # Clear Spelling Dataset
15
+
16
+ ## Overview
17
+
18
+ The **Clear Spelling Dataset** is a carefully crafted collection of **100,000 unique English spelling mistakes and their correct forms**, intended for training high-quality typo correction and spell checking AI models. It covers various types of common mistakes observed frequently in real-world scenarios, such as:
19
+
20
+ - Keyboard adjacency typos
21
+ - Letter swaps and omissions
22
+ - Duplicate characters
23
+ - Phonetic substitution errors
24
+ - Commonly confused homophones (e.g., "their" vs. "there")
25
+
26
+ ## Dataset Format
27
+
28
+ The dataset is provided in **CSV format** with two clearly defined columns:
29
+
30
+ | Column | Description | Example |
31
+ |----------|---------------------------------------------|---------------------|
32
+ | `error` | The misspelled or incorrect word or phrase | "teh" |
33
+ | `correct`| The correct word or intended phrase | "the" |
34
+
35
+ ## Usage
36
+
37
+ This dataset is ideal for:
38
+
39
+ - Training and fine-tuning **typo correction** models
40
+ - Benchmarking **spell-checking algorithms**
41
+ - Enhancing NLP model robustness to real-world noisy input
42
+
43
+ ## Quality Assurance
44
+
45
+ - **No duplicates:** Each (error, correct) pair is unique.
46
+ - **Hand-curated seed set:** Includes hundreds of common misspellings verified against real-world usage patterns.
47
+ - **Realistic noise generation:** Uses realistic error transformations mimicking genuine human typing behavior.
48
+
49
+ ## License (MIT)
50
+
51
+ This dataset is released under the permissive **MIT License**, which allows commercial and non-commercial use, distribution, and modification. Attribution is required:
52
+
53
+
54
+ ## Citation
55
+
56
+ If you use this dataset in your research or projects, please provide attribution similar to:
57
+
58
+ ```
59
+ This [your project type] uses the Mistake to Learning dataset by ProCreations.
60
+ ```
61
+
62
+ Enjoy training your typo-correction models!