Language models use abstract binding IDs to represent and bind entities and attributes in context, enabling them to reason about multiple objects of the same kind.