A comparison of top-down effects on phonological processing induced by lexical context and by situational language context: An investigation using electrocortikography and magnetoencephalography

一个单词听起来会因说话者或背景噪音的不同而不同。然而，我们非常擅长从嘈杂和可变的声音信号中提取稳定的感知。已经确定这种能力是基于对音素的识别，音素是改变意义的最小的语音单位（辅音和元音），定位于左侧颞上回（STG）。音素识别的一个基本原则是，声学变化的输入被快速而毫不费力地映射到有限数量的音素类别。然而，源自近似输入或情境信息（如输入语言）的预测可以改变这种映射。这种预测编码的一个例子是词汇自顶向下的效果，这种效果可能需要将/woo/前面的语音标记分类为/d/，但是/boo/前面的语音标记分类为/t/，从而产生/wood/和/boot/。同样，由于不同语言之间音素的声学差异，相同的符号在英语上下文中可能被理解为/d/，但在西班牙语上下文中可能被理解为/t/。尽管描述这些机制与语音识别的神经模型有关，但目前尚不清楚是什么神经机制介导了这种对左STG音素表征的预测作用。事实上，直到最近，这些努力都受到人类神经科学中现有方法的粗糙分辨率的阻碍，缺乏足够的空间（例如脑磁图，MEG）或时间（磁共振成像）分辨率。这种情况随着人类皮质电图（ECoG）的发展而改变。ECoG是一种高时空分辨率的硬膜下受限皮质区域的神经活动记录。我建议在5项研究中采用全脑脑磁图（MEG）和后STG脑电图（ECoG），旨在探讨词汇和情景语言语境对音素识别影响的神经机制。研究1将使用脑磁图来描述音位分类中语音输入和词汇期望之间整合的时间。研究2将使用ECoG进一步识别这些词汇自上而下效应背后的时空变化。研究3和4将阐明第一语言（L1）和第二语言（L2）语音和音位序列概率的神经编码的差异。为此，将在西班牙语-英语双语者听第一和第二语言句子时记录ECoG信号。第五项脑磁图研究将比较词汇语境和语言语境对音位分类的影响。受试者将对嵌入在第一语言和第二语言的词汇或语言语境中的符号进行音位分类，从而使这两个预测来源独立。词汇语境和语言语境诱导平等感知的条件之间的关键对比将直接使它们的潜在机制相互对立。总体而言，本研究计划将为语音感知预测效应的神经模型奠定基础。

A word can sound different depending on the speaker or background noise. Yet, we are highly skilled in extracting stable percepts from noisy and variable acoustic signals. It is well established that this ability is based on the recognition of phonemes, the smallest speech units that alter meaning (consonants and vowels), localized to the left superior temporal gyrus (STG). A fundamental principle of phoneme recognition is that acoustically varying inputs are mapped fast and effortlessly to a finite number of phonemic categories. However, predictions originating in proximate inputs or in situational information, such as the input language, can alter this mapping. An example for such predictive coding are lexical top-down effects that can entail categorization of a phonetic token preceded by /woo/ as /d/, but as /t/ when preceded by /boo/, resulting in /wood/ and /boot/. Similarly, due to acoustic differences between these phonemes across languages, the same token might be perceived as /d/ in an English context, but as /t/ in a Spanish context. It is currently unknown what neural mechanisms mediate such predictive effects on phonemic representations in the left STG, although describing these mechanisms is pertinent to neural models of speech recognition. Indeed, until recently, such endeavors were hindered by the crude resolution of available methods in human neuroscience, lacking sufficient spatial (e.g. magnetic encephalography, MEG) or temporal (magnetic resonance imaging) resolution. This changed with the development of human electrocorticography (ECoG). With ECoG, neural activity is recorded subdurally from a confined cortical area with high spatiotemporal resolution. I propose to employ whole-brain MEG and ECoG of posterior STG in 5 studies, designed to investigate the neural mechanisms mediating the effects of lexical and situational language context on phoneme recognition. Study 1 will employ MEG to describe the timing of integration between acoustic input and lexical expectations in phonemic categorization. Study 2 will employ ECoG to further identify spatiotemporal changes underlying these lexical top-down effects. Studies 3 and 4 will elucidate the differences between neural encoding of first (L1) and second (L2) language phonology and phonemic sequence probabilities. For this, ECoG signal will be recorded in Spanish-English bilinguals listening to sentences in their L1 and L2. A fifth MEG study will compare effects of lexical and language contexts on phonemic categorization. Subjects will perform phonemic categorizations on tokens embedded in either lexical or only language context in L1 and L2, thus making these two sources of predictions independent. The crucial contrast between conditions in which lexical and language context will induce equal percepts will directly pit their underlying mechanisms against each other. Overall, this research program will set the foundations for neural models of predictive effects on speech perception.