Large language models possess remarkable capacity for processing language, but it remains unclear whether these models can further generate creative content. The present study aims to investigate the creative thinking of large language models through a cognitive perspective. We utilize the divergent association task (DAT), an objective measurement of creativity that asks models to generate unrelated words and calculates the semantic distance between them. We compare the results across different models and decoding strategies. Our findings indicate that: (1) When using the greedy search strategy, GPT-4 outperforms 96% of humans, while GPT-3.5-turbo exceeds the average human level. (2) Stochastic sampling and temperature scaling are effective to obtain higher DAT scores for models except GPT-4, but face a trade-off between creativity and stability. These results imply that advanced large language models have divergent semantic associations, which is a fundamental process underlying creativity.

Natural language inference (NLI) is a task to infer the relationship between a premise and a hypothesis (e.g., entailment, neutral, or contradiction), and transformer-based models perform well on current NLI datasets such as MNLI and SNLI. Nevertheless, given the linguistic complexity of the large-scale datasets, it remains controversial whether these models can truly infer the relationship between sentences or they simply guess the answer via shallow heuristics. Here, we introduce a controlled evaluation set called Simple Pair to test the basic sentence inference ability of NLI models using sentences with syntactically simple structures. Three popular transformer-based models, i.e., BERT, RoBERTa, and DeBERTa, are employed. We find that these models fine-tuned on MNLI or SNLI perform very poorly on Simple Pair (< 35.4% accuracy). Further analyses reveal event coreference and compositional binding problems in these models. To improve the model performance, we augment the training set, i.e., MNLI or SNLI, with a few examples constructed based on Simple Pair ( 1% of the size of the original SNLI/MNLI training sets). Models fine-tuned on the augmented training set maintain high performance on MNLI/SNLI and perform very well on Simple Pair (~100% accuracy). Furthermore, the positive performance of the augmented training models can transfer to more complex examples constructed based on sentences from MNLI and SNLI. Taken together, the current work shows that (1) models achieving high accuracy on mainstream large-scale datasets still lack the capacity to draw accurate inferences on simple sentences, and (2) augmenting mainstream datasets with a small number of target simple sentences can effectively improve model performance.

Pre-trained language models have achieved human-level performance on many Machine Reading Comprehension (MRC) tasks, but it remains unclear whether these models truly understand language or answer questions by exploiting statistical biases in datasets. Here, we demonstrate a simple yet effective method to attack MRC models and reveal the statistical biases in these models. We apply the method to the RACE dataset, for which the answer to each MRC question is selected from 4 options. It is found that several pre-trained language models, including BERT, ALBERT, and RoBERTa, show consistent preference to some options, even when these options are irrelevant to the question. When interfered by these irrelevant options, the performance of MRC models can be reduced from human-level performance to the chance-level performance. Human readers, however, are not clearly affected by these irrelevant options. Finally, we propose an augmented training method that can greatly reduce models’ statistical biases.

本文实验发现,段落顺序会影响人类阅读理解效果;而打乱段落或句子顺序,对BERT、ALBERT和RoBERTa三种人工神经网络模型的阅读理解答题几乎没有影响。打乱词序后,人的阅读理解水平低于三个模型,但人和模型的答题情况高于随机水平,这说明人比人工神经网络对词序更敏感,但人与模型可以在单词乱序的情况下答题。综上,人与人工神经网络在正常阅读的情况下回答阅读理解问题的正确率相当,但两者对篇章结构及语序的依赖程度不同。