Spotting AI's Touch: Identifying LLM-Paraphrased Spans in Text