(Re)construing Meaning in NLP