arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2104.05433cs.CL
收藏

多语种语言模型预测人类阅读行为

Multilingual Language Models Predict Human Reading Behavior

ETH Zurich(苏黎世联邦理工学院);University of Zurich(苏黎世大学);University of Potsdam(波茨坦大学);Vrije Universiteit Amsterdam(阿姆斯特丹自由大学)

Nora Hollenstein, Federico Pirovano, Ce Zhang, Lena Jäger, Lisa Beinborn

摘要:

我们分析了大语言模型是否能够预测人类阅读行为的模式。我们比较了特定语言和多语言预训练Transformer模型的性能,以预测反映荷兰语,英语,德语和俄语文本上自然人类句子处理的阅读时间测量。这导致人类阅读行为的准确模型,这表明Transformer模型以与人类处理机制相当的方式隐式地编码语言中的相对重要性。我们发现,BERT和XLM模型成功地预测了一系列的眼动追踪功能。在一系列实验中,我们分析了这些模型的跨领域和跨语言能力,并展示了它们如何反映人类的句子处理。

英文摘要:

We analyze if large language models are able to predict patterns of human reading behavior. We compare the performance of language-specific and multilingual pretrained transformer models to predict reading time measures reflecting natural human sentence processing on Dutch, English, German, and Russian texts. This results in accurate models of human reading behavior, which indicates that transformer models implicitly encode relative importance in language in a way that is comparable to human processing mechanisms. We find that BERT and XLM models successfully predict a range of eye tracking features. In a series of experiments, we analyze the cross-domain and cross-language abilities of these models and show how they reflect human sentence processing.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑