Emotion and cause analysis (ECA) plays a key role in understanding human emotional responses, which are inherently linked to specific triggers or causes. Traditional ECA methods in NLP often rely on data aligned with the test context, like news articles, but annotated dialogue data is scarce. We introduce a novel approach that utilizes existing corpora from different contexts (e.g., sentence-level and document-level texts) to predict emotions and causes in dialogues, even in the absence of annotated dialogue data. This “zero-shot” method, complemented by a “few-shot” setting, allows for validation using a limited number of annotated samples. We propose a Unified Option Generation (UOG) framework, leveraging instructive learning to address the challenges of emotion recognition and cause extraction across diverse data scenarios. This framework redefines ECA as an option generation problem, enhancing both zero- and few-shot performance in emotion and cause analysis. Experiments confirm the efficacy of this approach in dialogue contexts.