Linguistic conventions in value alignment — LessWrong