Abstract Although seemingly easy, speech production is a complex action involving several processes and brain regions and is often studied through the use of picture naming paradigms. Several brain regions have been identified as being involved in speech production and lexical retrieval, including several areas of the left lateral, ventral, and medial temporal cortex, and of the left lateral and medial prefrontal cortex (PFC). However, little is known about how these regions interact to allow us to produce and retrieve words so efficiently as we speak. In this study, we used graph signal processing, and in particular graph learning, to analyze intracranial electroencephalography data recorded directly at the cortical surface (i.e., electrocorticography, ECoG) in the blocked cyclic picture naming task (recently reported in Ries et al., 2017). In this task, pictures are named several times in semantically homogeneous (HOM) versus heterogeneous (HET) blocks. Lexical retrieval, and in particular our ability to select words from competing alternatives, is hampered in HOM vs. HET blocks with increasing repetitions of the pictures, an effect referred to as the semantic interference effect. The method we used can infer the connectivity between different brain regions from the recorded ECoG signals through an optimization process that considers the entire connectivity map instead of pair-wise methods such as more commonly used correlation measures. Our results show that distant left frontal and temporal brain regions as well as PFC regions are functionally connected in picture naming, and that connection strength is variable between pairs of brain regions and between participants. Semantic interference on naming latencies was not present in all participants and varied in terms of when naming latencies became slower in HOM versus HET blocks. Connection weights were generally sensitive to semantic context in the participants showing a semantic interference effect on behavior but in highly variable ways across participants and in different region pairs. In sum, our results are consistent with the hypotheses that left frontal and temporal regions are functionally connected during picture naming and that intra-frontal connections seem particularly important in this paradigm manipulating lexical retrieval difficulty. Further research is needed to explore the cortical interactions supporting lexical retrieval.

中文翻译：

---

支持语音产生和词汇检索的皮层交互模式：个体层面的图形信号处理方法

摘要 虽然看似简单，但语音生成是一个复杂的动作，涉及多个过程和大脑区域，通常通过使用图片命名范式进行研究。几个大脑区域已被确定参与语音生成和词汇检索，包括左侧、腹侧和内侧颞叶皮层的几个区域，以及左侧和内侧前额叶皮层 (PFC) 的几个区域。然而，我们对这些区域如何相互作用以允许我们在说话时如此高效地生成和检索单词知之甚少。在这项研究中，我们使用图形信号处理，特别是图形学习，来分析在阻塞循环图片命名任务中直接记录在皮质表面的颅内脑电图数据（即，脑电图，ECoG）（最近在 Ries 等人报道， 2017）。在此任务中，图片在语义同质 (HOM) 与异质 (HET) 块中被多次命名。词汇检索，特别是我们从竞争替代品中选择单词的能力，在 HOM 与 HET 块中随着图片重复次数的增加而受到阻碍，这种效果被称为语义干扰效应。我们使用的方法可以通过优化过程从记录的 ECoG 信号推断不同大脑区域之间的连接，该过程考虑整个连接图而不是成对方法，例如更常用的相关度量。我们的研究结果表明，远处的左侧额叶和颞叶大脑区域以及 PFC 区域在图片命名中在功能上是相连的，并且连接强度在成对的大脑区域之间和参与者之间是可变的。命名延迟的语义干扰并非在所有参与者中都存在，并且在 HOM 与 HET 块中命名延迟变得更慢时各不相同。连接权重通常对参与者中的语义上下文敏感，显示出对行为的语义干扰影响，但在参与者之间和不同区域对中以高度可变的方式。总而言之，我们的结果与左额叶和颞叶区域在图片命名过程中在功能上连接的假设一致，并且在这种操纵词汇检索困难的范式中，额叶内连接似乎特别重要。需要进一步的研究来探索支持词汇检索的皮层相互作用。