add dataset description

README.md

@@ -12,6 +12,14 @@ size_categories:
 ---
 
 # NLU Few-shot Benchmark - English and German
+This is a few shot training dataset from the domain of human-robot interaction.
+It contains texts in German and English language with 64 different utterances (classes).
+Each utterance (class) has exactly 20 samples in the training set.
+This leads to a total of 1280 different training samples.
+
+With this publication, we are building on our
+[deutsche-telekom/NLU-Evaluation-Data-en-de](https://huggingface.co/datasets/deutsche-telekom/NLU-Evaluation-Data-en-de)
+data set.
 
 ## Processing Steps
 - drop `NaN` values