When does word order matter and when doesn't it?
This addresses the problem of understanding model robustness for researchers in NLP, but it is incremental as it builds on existing redundancy theories.
The paper investigates when language models are insensitive to word order changes in NLU tasks, attributing it to linguistic redundancy, and finds that insensitivity varies by task, with models showing high consistency in tasks like SST-2 but near-random consistency in tasks like RTE when word order is less informative.
Language models (LMs) may appear insensitive to word order changes in natural language understanding (NLU) tasks. In this paper, we propose that linguistic redundancy can explain this phenomenon, whereby word order and other linguistic cues such as case markers provide overlapping and thus redundant information. Our hypothesis is that models exhibit insensitivity to word order when the order provides redundant information, and the degree of insensitivity varies across tasks. We quantify how informative word order is using mutual information (MI) between unscrambled and scrambled sentences. Our results show the effect that the less informative word order is, the more consistent the model's predictions are between unscrambled and scrambled sentences. We also find that the effect varies across tasks: for some tasks, like SST-2, LMs' prediction is almost always consistent with the original one even if the Pointwise-MI (PMI) changes, while for others, like RTE, the consistency is near random when the PMI gets lower, i.e., word order is really important.