How does a language-learning robot know which object a child means?

A child points past a toy rabbit toward a cup and asks what it's called. Before a robot teaches the word, it has to work out which thing the child means. That small exchange makes me more curious than another dancing demonstration. AiMOGA opened pre-orders for its Mobo home companion robot this week. Its announcement describes spatial awareness and family interaction. Aju Press's coverage, which calls the robot Mobao, describes object questions and English conversation for children. Those are launch descriptions, not evidence that children learn more with it. I'd like to see a play session with several objects close together and a child who changes their mind halfway through pointing. Can the robot indicate the object it means and accept “no, the other one” without making the child restart the conversation? “Do you mean the cup?” could be a useful part of learning together, rather than an embarrassing pause to hide. There's something lovely about learning a word for a thing you can actually pick up. I'd want the robot to leave room for the child to disagree about what they're looking at. A confident answer about the wrong toy would miss the whole reason to put the lesson in the room. What would you want to watch a language-learning robot do with a child before trusting its lessons?