Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -70,15 +70,10 @@ The dataset contains a single split: `train`.
|
|
70 |
|
71 |
### Source Data
|
72 |
|
73 |
-
The dataset is a combination of the following authentic datasets:
|
74 |
-
|
75 |
-
| Dataset | Sentences |
|
76 |
-
|:-------|-------:|
|
77 |
-
|
78 |
The Catalan-Chinese data collected from the web was a combination of the following datasets:
|
79 |
|
80 |
| Dataset | Sentences before cleaning |
|
81 |
-
|
82 |
| WikiMatrix | 90.643 |
|
83 |
| XLENT | 535.803 |
|
84 |
| GNOME | 78|
|
@@ -90,7 +85,7 @@ The Catalan-Chinese data collected from the web was a combination of the followi
|
|
90 |
The 6.658.607 sentence pairs of synthetic parallel data were created from the following Spanish-Chinese datasets:
|
91 |
|
92 |
| Dataset | Sentences before cleaning |
|
93 |
-
|
94 |
| UNPC |17.599.223|
|
95 |
| CCMatrix | 24.051.233 |
|
96 |
| MultiParacrawl| 3410087|
|
@@ -98,10 +93,11 @@ The 6.658.607 sentence pairs of synthetic parallel data were created from the fo
|
|
98 |
|
99 |
### Data preparation
|
100 |
|
101 |
-
The Chinese side of all datasets are passed through the [fastlangid](https://github.com/currentslab/fastlangid) language detector and any sentences which are not
|
102 |
-
as simplified Chinese are discarded.
|
|
|
103 |
This is done using sentence embeddings calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE).
|
104 |
-
The filtered datasets are then concatenated to form a final corpus of 6.833.114 parallel sentences.
|
105 |
|
106 |
### Personal and Sensitive Information
|
107 |
|
|
|
70 |
|
71 |
### Source Data
|
72 |
|
|
|
|
|
|
|
|
|
|
|
73 |
The Catalan-Chinese data collected from the web was a combination of the following datasets:
|
74 |
|
75 |
| Dataset | Sentences before cleaning |
|
76 |
+
|:------------------|---------------:|
|
77 |
| WikiMatrix | 90.643 |
|
78 |
| XLENT | 535.803 |
|
79 |
| GNOME | 78|
|
|
|
85 |
The 6.658.607 sentence pairs of synthetic parallel data were created from the following Spanish-Chinese datasets:
|
86 |
|
87 |
| Dataset | Sentences before cleaning |
|
88 |
+
|:------------------|---------------:|
|
89 |
| UNPC |17.599.223|
|
90 |
| CCMatrix | 24.051.233 |
|
91 |
| MultiParacrawl| 3410087|
|
|
|
93 |
|
94 |
### Data preparation
|
95 |
|
96 |
+
The Chinese side of all datasets are passed through the [fastlangid](https://github.com/currentslab/fastlangid) language detector and any sentences which are not
|
97 |
+
identified as simplified Chinese are discarded.
|
98 |
+
The datasets are then also deduplicated and filtered to remove any sentence pairs with a cosine similarity of less than 0.75.
|
99 |
This is done using sentence embeddings calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE).
|
100 |
+
The filtered datasets are then concatenated to form a final corpus of **6.833.114** parallel sentences.
|
101 |
|
102 |
### Personal and Sensitive Information
|
103 |
|