资料来源： 实验室的莎拉 I.吉伯尔和乔纳斯 · 卡普兰 — — 南加利福尼亚大学
理解语言是人类有能力的最复杂的认知任务之一。当结合个别单词到句子的形式意义，给出数量惊人的可能的选择，至关重要的大脑是能够识别单词形成连贯的组合和时出现异常，破坏了意义。广泛的研究表明，某些头皮记录的电气事件敏感到这种期望中的偏差。重要的是，这些电气签名的不协调是具体到意外的意思，，因此不同于大脑的一般响应其他类型的异常。
通过目前对句子的语义一致和不一致两端的范式使用实验考察了神经生理学相关的语义错位。最初是在 1980 年，语义错位任务提出了参与者与一系列句子此与字一致或不一致。若要测试的反应是从语义不一致和不更一般由于惊喜，一些句子包括词给出了在不同的大小。1一个句子的语义不一致终结有已经显示出特定电的事件是在头皮事件相关电位 (ERPs) 被称为可刻录。ERP 是从某一特定的感觉、 认知、 或运动事件造成的测量的大脑反应。事件相关电位测量使用脑电图 (EEG)，评估疾病患者和正常人的脑功能的无创性手段。在头皮上，称为 N400，发现特定的 ERP 组件显示更大振幅响应语义不一致的事件。N400 是大约 250 和 400 毫秒后刺激发病之间发生的 EEG 信号中的负向偏转。一般情况下，早期电位反映感官电机处理，而后来潜能像 N400 反映认知加工。
在本视频中，我们展示如何管理使用脑电图的语义错位任务。视频将覆盖的设置和管理的脑电图，和事件相关电位分析有关的语义错位控制和目标刺激。在此任务中，参与者设置了脑电图电极，然后大脑活动录得，而他们查看控制语句和语义不一致的句子。脑电图过程是相似的哈比比et al.，1和任务仿照 Kutas 和威尔森。2当 ERPs 的平均值的一致和不协调的句子，每个事件的神经相关可对比在选定的时间窗口。
During the semantic incongruity task where participants viewed congruous sentences, incongruous sentences, and sentences where the last word was presented in a larger size, there was a negative-going N400 response only for the incongruous sentences (Figure 2, blue). Sentences with a surprising element (larger last word) that was not semantically incongruous did not show an N400 response, but did show an increased P560 response (Figure 2, red). The N400 response started about 250 ms after the presentation of the last word of the sentence and peaked about 400 ms after the stimulus onset.
These results show that electrical activity in the brain, and particularly in the parietal lobe, registers when a semantically incongruous word is presented as part of a sentence. This electrical event reflects the neural processes that identify the interruption of ongoing sentence processing by a semantically inappropriate word. The N400 seems to provide useful information about the timing, classification, and interactions of cognitive processes involved in natural language processing and comprehension.
This study demonstrates some of the advantages of the ERP approach, in particular, its high temporal resolution. In this paradigm, to simulate natural reading, word stimuli are presented very briefly in succession. Because of the excellent temporal resolution of EEG, we are able to discern electrical responses to the stimuli individually.
As a marker of semantic processing, the N400 can be a useful tool in understanding the development of language from childhood to adulthood. Study of this component shows that even in 19-month-old babies, there is a semantic incongruity effect when they hear words that don't match pictures they are seeing.3 This demonstrates the very early presence of a mechanism for matching words to their proper context. However, while young adolescents show an N400 that discriminates between congruent and incongruent language, the response profile of this component is not yet as nuanced as that of adults; for example, it is not as sensitive to different degrees of incongruity.4 These studies demonstrate the sensitivity of this ERP component as an index of semantic processing.
Understanding language involves complex cognitive processes, and—given the incredible number of word choices and arrangements that can form a single sentence—the brain must be able to distinguish between coherent and incoherent combinations.
A person’s comprehension of a sentence, whether spoken—like when a mother tells her son that she’s going to the store—or written in a book, depends, in part, on what the brain anticipates the next word in the sentence to be.
For example, if someone begins to read “It was a dark and stormy…” at the beginning of a book, it is expected that “night” will be the following term.
However, occasionally unexpected words are encountered—like “…and the mad scientist was painting his laboratory the color raccoon…”—that disrupt the sentence’s meaning.
In this instance, the anomalous term is raccoon, as it refers to a type of animal, rather than an expected color, like black.
Such semantic incongruities—the senseless sentences—elicit unique electrical signals in the brain—responses known as event-related potentials, ERPs for short—that may provide insight into how the brain either retrieves the definition of, or reprocesses, the troublesome word in an attempt to comprehend the sentence.
This video explains how the technique of electroencephalography, or EEG, can be used to measure ERPs during semantic incongruity tasks, in which participants are shown sentences ending with unexpected words.
We demonstrate how to design stimuli, and collect and analyze data, specifically focusing on a unique component of ERPs, named N400 to reflect its characteristics.
In this experiment, EEG is used to measure brain activity in participants shown semantically coherent and incoherent stimuli, in order to investigate language processing and comprehension.
These stimuli consist of three kinds of sentences: congruous, incongruous, and size-deviant. Although each is composed of seven words, they differ in the nature of their last terms.
The final words in congruous sentences, like “She scratched her dog behind its ear.,” pose no problems with meaning, and appear in the same font type—and size—as those preceding it.
Importantly, these sentences serve as controls to gauge how the brain responds to coherent word combinations.
In contrast, incongruous sentences, like “She dipped her chicken finger in boots.,” possess last terms that are semantically anomalous.
Here, boots conflicts with the meaning of the rest of the words—it is expected that chicken fingers would be dipped in a condiment like mustard, not in articles of clothing. Thus, these stimuli evaluate how surprising, incoherent language is processed.
The final type of sentences are called size-deviant, and contain last words that are surprising in appearance—they are in a larger font—but not congruity.
For example, if in the sentence “He put his hand in his mitten.,” the term mitten is written in bigger letters, it still makes semantic sense.
These stimuli are critical, as they are meant to distinguish whether the brain’s response to the last word in a sentence is the result of general surprise—the shock of an inconsistent text size—or is specific to unexpected meanings.
After participants are prepared for EEG, they are told to carefully read sentences that appear on a computer screen, as questions will be asked about them later on.
In reality, no quiz is given at the end of the experiment; however, these instructions ensure that subjects will pay attention to the upcoming stimuli.
During the task, participants are sequentially shown—in the correct order—the seven words that make up a single sentence.
Each term appears individually in the center of the monitor—to reduce eye movements that could interfere with data collection—for 100 ms, and is followed by 1000 ms of blank screen.
EEG information is continuously recorded over 120 such trials, each of which consists of a unique sentence. Specifically, stimuli are shown at the same frequency—40 times—but in a random order. Then, the task is repeated a second time, so participants must read a total of 240 sentences altogether.
Afterwards, EEG data are processed to visualize average ERPs for each type of sentence—from each electrode—and scientists search for the N400 component in these waveforms.
The “N” in this term indicates that the peak is negative, and the “400” represents its latency—that it occurs roughly 400 ms after the last-word stimulus is shown to the participant.
Based on previous experience, it is expected that the amplitude of N400 will increase in response to semantically inconsistent events, and will be recorded from all scalp electrodes.
However, this response will likely be most prominent at the Pz electrode, positioned in the midline of the scalp above the parietal lobes—regions of which are known to be involved in processing and integrating written language.
Prior to beginning the experiment, recruit a participant who is a native English speaker, and explain to them the two main components of the procedure: that they will be wearing electrodes, and be shown sentences on a computer screen. Then, collect from them all of the necessary, signed consent forms.
Next, outfit the participant with scalp and face electrodes. For more details on this procedure, check out the methods described elsewhere in this collection. Once in the testing space, verify impedance values across all electrodes.
Upon confirming that the EEG traces are void of noise, instruct the participant to sit so that their eyes are approximately 75 cm away from the screen.
Emphasize that they should read and pay careful attention to the sentences that appear word-by-word on this display, as questions will be asked about their content later on.
To ensure that the participant understands the task, show them ten practice sentences, but do not collect data during this time. Afterwards, start the EEG system to commence continuous recording.
Proceed with the functional task by presenting 120 trials—consisting of 40 congruous, 40 incongruous, and 40 deviant-size sentences—in a random order. Then, repeat this process with an additional set of 120 stimuli to guarantee that enough data are collected.
Once data have been recorded for all 240 stimuli, process it as described in JoVE’s ERPs and the Oddball Task video.
To analyze the data, first plot the average waveforms for the timecourses of congruous, incongruous, and deviant-size stimuli collected from the Pz recording site. On the x-axis of this graph—representing time in ms—indicate when each word in a sentence is shown.
Afterwards, locate the N400 peaks, and for each, calculate its average amplitude—defined as the distance between the lowest point of the peak and the baseline value of 0 µV, also represented by the horizontal axis.
Then, calculate the latency of this component—how long in ms it takes for it to appear in the waveform after the last word in a sentence is shown.
For the ranges of these amplitudes and latencies, proceed to use F-tests to determine whether there is a difference between target and control stimuli.
Notice that the N400 response was only observed after participants were shown the last word of an incongruous sentence, indicating that this electrical event reflects neural processing—particularly involving the parietal lobes—that identify an interruption in sentence processing caused by an incoherent term.
Importantly, although N400 was not observed in waveforms collected using deviant-size stimuli, another unique component—P560, a positive peak with a latency of 560 ms—was.
This indicates that the brain responds differently to unexpected visual stimuli and semantically inconsistent terms, and suggests that N400 is a unique electrical signature of language incongruity.
Now that you know how semantic inconsistency can be used to elicit the N400 component in ERPs, let’s look at other ways researchers are examining this unique electrical signal to study language processing and comprehension.
Some researchers aim to determine when the ability to identify incoherent language develops, and whether this skill changes with age.
Such work has involved showing young children—outfitted with EEG caps—representations of recognizable objects, like a camera.
However, the trick is that when the child looks at this depiction, they’re told it’s something different—for example, a cat. Thus, this is a modified version of the semantic incongruity task, as the spoken word doesn’t match the meaning of the visible item.
Measurements of the brain’s electrical responses to these tasks demonstrated that children exhibit an enhanced N400-esque response to incongruous item-word pairs—one that lasts for several hundred ms—compared to congruous sets.
Importantly, this suggests that even at an early age, humans are able to identify and process semantic incongruity.
Other researchers are assessing whether ERPs can be used to better understand language deficits associated with certain personality disorders, such as schizophrenia.
Paradoxically, previous work has shown that individuals with pronounced schizophrenia-like characteristics, such as anxiety or the inability to feel pleasure, demonstrate a heightened N400 response to congruous word pairs—like animal and goat—compared to people with milder symptoms.
However, when these participants were treated with an antipsychotic drug called olanzapine, the amplitude of this congruity-caused N400 component decreased compared to individuals given a placebo, suggesting a possible therapy that could treat the disjointed speech sometimes observed in such disorders.
You’ve just watched JoVE’s video on how congruous and incongruous sentences can be used to investigate language processing. At this point, you should know how to present stimuli to participants, and collect and interpret ERP data. We hope you also now understand how the N400 component is being used to investigate other aspects of language comprehension, such as how it can be affected in behavioral disorders.
Thanks for watching!
