Out of Order: How Important Is The Sequential Order of Words in a Sentence in Natural Language Understanding Tasks?

Do state-of-the-art natural language understanding models care about word\norder - one of the most important characteristics of a sequence? Not always! We\nfound 75% to 90% of the correct predictions of BERT-based classifiers, trained\non many GLUE tasks, remain constant after input words are randomly shuffled.\nDespite BERT embeddings are famously contextual, the contribution of each\nindividual word to downstream tasks is almost unchanged even after the word's\ncontext is shuffled. BERT-based models are able to exploit superficial cues\n(e.g. the sentiment of keywords in sentiment analysis; or the word-wise\nsimilarity between sequence-pair inputs in natural language inference) to make\ncorrect decisions when tokens are arranged in random orders. Encouraging\nclassifiers to capture word order information improves the performance on most\nGLUE tasks, SQuAD 2.0 and out-of-samples. Our work suggests that many GLUE\ntasks are not challenging machines to understand the meaning of a sentence.\n