Update README.md

README.md

@@ -46,6 +46,18 @@ The dataset includes prompts in multiple languages including:
 - Italian
 - Japanese
 
+## Noise types
+
+The dataset includes the following noise variants to evaluate model robustness under different types of input perturbations.
+
+N1 – Typographical Noise: Introduces spelling errors or minor typographical variations in the text.
+
+N2 – Lexical Noise: Replaces words with semantically similar alternatives while preserving the original meaning.
+
+N3 – Structural Noise: Modifies the sentence structure or word order without altering the underlying intent.
+
+N4 – Extraneous Noise: Adds additional irrelevant words or phrases that do not contribute to the core meaning of the prompt.
+
 ## Dataset Size
 
 - Total samples: **10,232**