Solving this problem led to my creating a general, flexible word learning system called the "TWIG" system. When the system encounters a sentence that it doesn't understand, it first tries to determine if the sentence makes sense under the assumption that the new word refers to something it knows about. For instance, if it hears "The pig is near you," and it knows what every word but "near" means, it will assume "near" refers to some relationship between the pig and itself. You can read more about this part of the system in this paper.