Improved description in Source Data section

README.md

@@ -69,9 +69,12 @@ This dataset is aimed at promoting the development of Machine Translation betwee
 The dataset is a combination of the following original datasets collected from [Opus](https://opus.nlpl.eu/):
 NLLB, MultiCCAligned, WikiMatrix, GNOME, KDE4, OpenSubtitles, GlobalVoices.
 
-All datasets are 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.
+All data was 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.
+
+The filtered datasets are then concatenated and deduplicated to form the final corpus.
 
 #### Who are the source language producers?