Being Comes From Not-Being: Open-Vocabulary Text-to-Motion Generation With Wordless Training