Resolving Positional Ambiguity in Dialogues by Vision-Language Models for Robot Navigation.