Does She Wink or Does She Nod? A Challenging Benchmark for Evaluating Word Understanding of Language Models