Can someone please explain the "Scaled dot-product attention" function from the article?

Attention(Q, K, V ) = softmax( (Q * TRANSPOSED(T)) / sqrt(Dk) ) V

That's where i start to shake my head.