Brains and algorithms partially converge in natural language processing.
Brains and algorithms partially converge in natural language processing.
复制标题
DOI：
10.1038/s42003-022-03036-1
复制
发表时间：
2022-02-16
影响因子：
5.9
通讯作者：
King JR
中科院分区：
文献类型：
--
作者：
Caucheteux C;King JR
最近，被训练用于从大量文本中预测掩码词的深度学习算法被证明可以产生类似于人类大脑的激活。然而，是什么驱动了这种相似性目前仍不清楚。在这里，我们系统地比较了各种深度语言模型，以确定导致它们生成句子的大脑表征的计算原理。具体来说，我们分析了大脑对400个孤立的句子在一个大的队列中的102名受试者，每个记录了两个小时的功能性磁共振成像（fMRI）和脑磁图（MEG）。然后，我们测试这些算法中的每一个在何时何地映射到大脑反应。最后，我们估计这些模型的架构、训练和性能如何独立地解释类脑表征的生成。我们的分析揭示了两个主要发现。首先，算法和大脑之间的相似性主要取决于它们根据上下文预测单词的能力。其次，这种相似性揭示了每个皮层区域内知觉、词汇和成分表征的产生和维持。总的来说，这项研究表明，现代语言算法部分收敛于类似大脑的解决方案，从而描绘了一条有希望的道路来解开自然语言处理的基础。夏洛特·考舍特（Charlotte Caucheteux）和让-雷米·金（Jean-Rémi King）研究了在单词预测任务中训练的Transformer神经网络与人脑表征的匹配能力，这些表征是通过功能磁共振成像（fMRI）和脑磁图（MEG）测量的。他们的研究结果为Transformer语言模型的工作原理及其与大脑反应的相关性提供了进一步的见解。
Deep learning algorithms trained to predict masked words from large amount of text have recently been shown to generate activations similar to those of the human brain. However, what drives this similarity remains currently unknown. Here, we systematically compare a variety of deep language models to identify the computational principles that lead them to generate brain-like representations of sentences. Specifically, we analyze the brain responses to 400 isolated sentences in a large cohort of 102 subjects, each recorded for two hours with functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG). We then test where and when each of these algorithms maps onto the brain responses. Finally, we estimate how the architecture, training, and performance of these models independently account for the generation of brain-like representations. Our analyses reveal two main findings. First, the similarity between the algorithms and the brain primarily depends on their ability to predict words from context. Second, this similarity reveals the rise and maintenance of perceptual, lexical, and compositional representations within each cortical region. Overall, this study shows that modern language algorithms partially converge towards brain-like solutions, and thus delineates a promising path to unravel the foundations of natural language processing. Charlotte Caucheteux and Jean-Rémi King examine the ability of transformer neural networks trained on word prediction tasks to fit representations in the human brain measured with fMRI and MEG. Their results provide further insight into the workings of transformer language models and their relevance to brain responses.
登录
查看更多内容
影响因子：
3.4
作者：
Fedorenko, Evelina;Blank, Idan Asher;Mineroff, Zachary
通讯作者：
Mineroff, Zachary
影响因子：
3.5
作者：
Abraham A;Pedregosa F;Eickenberg M;Gervais P;Mueller A;Kossaifi J;Gramfort A;Thirion B;Varoquaux G
通讯作者：
Varoquaux G
影响因子：
5.7
作者：
Eickenberg, Michael;Gramfort, Alexandre;Thirion, Bertrand
通讯作者：
Thirion, Bertrand
影响因子：
48
作者：
Esteban, Oscar;Markiewicz, Christopher J.;Gorgolewski, Krzysztof J.
通讯作者：
Gorgolewski, Krzysztof J.
影响因子：
3.7
作者：
Anderson, Andrew James;Lalor, Edmund C.;Wang, Xixi
通讯作者：
Wang, Xixi