Update README.md

README.md

@@ -149,18 +149,20 @@ Supported languages:
 ### Original Source?
 Around 11 months ago, I downloaded and preprocessed 2.7M rows of text data, but completely forgot the original source of these datasets...
 All I remember is that I downloaded datasets from everywhere I could: HuggingFace, research papers, GitHub, Kaggle, SurgeAI, and Google search. I even fetched 20K+ tweets using the Twitter API.
-Today (6/28/2023) I came across two newer HuggingFace datasets, so I remembered to credit them below.
+Recently, I came across two newer HuggingFace datasets, so I remembered to credit them below.
 
 Known datasets:
 - tomekkorbak/pile-toxicity-balanced2
 - datasets/thai_toxicity_tweet
+- inspection-ai/japanese-toxic-dataset (GitHub)
 
 <br>
 
 ### Limitations
-Some limitations include:
+Limitations include:
 - All labels were rounded to the nearest integer. If a text was classified as 46%-54% toxic, the text itself might not be noticeably toxic or neutral.
 - There were disagreements among moderators on some labels, due to ambiguity and lack of context.
 - When there're only URL(s), emojis, or anything that's unrecognizable as natural language in the "text" column, the corresponding "lang" is "unkown".
+- The validation data is not representative of the training data.
 
 Have fun modelling!