Contrast-Then-Approximate: Analyzing Keyword Leakage of Generative Language Models | IEEE Journals & Magazine | IEEE Xplore