VLM-HOI: Vision Language Models for Interpretable Human-Object Interaction Analysis