With the advancement of large language models, the naturalness of responses in dialogue systems has improved dramatically. However, for dialogue systems to truly integrate into daily life and engage in smooth conversations with humans, they must possess the ability to "read the situation"—that is, to recognize the conversational context and the relationship between interlocutors, and to respond with appropriate timing and pacing. In this talk, we will introduce research we have conducted to date, including dialogue situation recognition for understanding various contexts in everyday conversations, intimacy recognition for interpreting relationships between speakers, and incremental response generation for producing responses with human-like rhythm and timing.