Update the dataset

.gitattributes

@@ -53,5 +53,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
 *.jpeg filter=lfs diff=lfs merge=lfs -text
 *.webp filter=lfs diff=lfs merge=lfs -text
-ca-zh_all_2023_10_26.ca filter=lfs diff=lfs merge=lfs -text
-ca-zh_all_2023_10_26.zh filter=lfs diff=lfs merge=lfs -text
+ca-zh_all_2024_08_05.ca filter=lfs diff=lfs merge=lfs -text
+ca-zh_all_2024_08_05.zh filter=lfs diff=lfs merge=lfs -text

README.md

@@ -39,11 +39,11 @@ The sentences included in the dataset are in Catalan (CA) and Chinese (ZH).
 
 Two separate txt files are provided with the sentences sorted in the same order:
 
-- ca-zh_all_2023_10_26.ca.
+- ca-zh_all_2024_08_05.ca.
 
-- ca-zh_all_2023_10_26.zh.
+- ca-zh_all_2024_08_05.zh.
 
-The dataset is additionally provided in parquet format: ca-zh_all_2023_10_26.parquet.
+The dataset is additionally provided in parquet format: ca-zh_all_2024_08_05.parquet.
 
 The parquet file contains two columns of parallel text obtained from the two original text files. 
 Each row in the file represents a pair of parallel sentences in the two languages of the dataset.
@@ -69,23 +69,33 @@ This dataset is aimed at promoting the development of Machine Translation betwee
 
 #### Initial Data Collection and Normalization
 
-The first portion of the corpus is a combination of the following original Catalan-Chinese datasets collected from [Opus](https://opus.nlpl.eu/):
-WikiMatrix, XLENT, GNOME, OpenSubtitles.
+The first portion of the corpus is a combination of our CA-ZH Wikipedia dataset and the following original Catalan-Chinese datasets collected from [Opus](https://opus.nlpl.eu/): OpenSubtitles, WikiMatrix.
 
-Additionally, the corpus contains synthetic parallel data generated from the following original Spanish-Chinese datasets collected from [Opus](https://opus.nlpl.eu/):
-UNPC, CCMatrix, MultiParacrawl.
+Additionally, the corpus contains synthetic parallel data generated from Spanish-Chinese News Commentary v18 from [WMT](https://data.statmt.org/news-commentary/v18.1/training/) and the following original Spanish-Chinese datasets collected from [Opus](https://opus.nlpl.eu/): NLLB, UNPC, MultiUN, MultiCCAligned, WikiMatrix, Tatoeba, MultiParaCrawl, OpenSubtitles. 
+
+Lastly, synthetic parallel data has also been generated from the following original English-Chinese datasets collected from [Opus](https://opus.nlpl.eu/): NLLB, CCAligned, ParaCrawl, WikiMatrix.
 
 ### Data preparation
 
-The Chinese side of all datasets are passed through the [fastlangid](https://github.com/currentslab/fastlangid) language detector and any sentences which are not
-identified as simplified Chinese are discarded.
-The datasets are then also deduplicated and filtered to remove any sentence pairs with a cosine similarity of less than 0.75.
-This is done using sentence embeddings calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE).
-The filtered datasets are then concatenated to form the final corpus. 
+The Chinese side of all datasets were first processed using the [Hanzi Identifier](https://github.com/tsroten/hanzidentifier) to detect Traditional Chinese, which was subsequently converted to Simplified Chinese using [OpenCC](https://github.com/BYVoid/OpenCC).
+
+All data was then filtered according to two specific criteria:
+
+- Alignment: sentence level alignments were calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE) and sentence pairs with a score below 0.75 were discarded.
+
+- Language identification: the probability of being the target language was calculated using [Lingua.py](https://github.com/pemistahl/lingua-py) and sentences with a language probability score below 0.5 were discarded.
+
+Next, Spanish data was translated into Catalan using the Aina Project's [Spanish-Catalan machine translation model](https://huggingface.co/projecte-aina/aina-translator-es-ca), while English data was translated into Catalan using the Aina Project's [English-Catalan machine translation model](https://huggingface.co/projecte-aina/aina-translator-en-ca).
+
+The filtered and translated datasets are then concatenated and deduplicated to form the final corpus.
 
 #### Who are the source language producers?
 
-[Opus](https://opus.nlpl.eu/) 
+[Opus](https://opus.nlpl.eu/)
+
+[WMT](https://machinetranslate.org/wmt)
+
+[Projecte Aina](https://huggingface.co/projecte-aina)
 
 ### Annotations
 

ca-zh_all_2024_08_05.ca

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63cd2b05b43498ddc32e23151f1843576fcf8274d7f993bd7bc154dbe0ff1578
+size 12316506664

ca-zh_all_2024_08_05.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abde25c775e9429aa9e29bc113544c746dd81807332e4609f1adb0779362aa8a
+size 15187135710

ca-zh_all_2024_08_05.zh

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1688b94022b2862498a19b36248356bec2a35a8abf5aabe37507b3c86eafee3a
+size 9637810856