Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning