A lot of research uses ERP, which is time-locked EEG. I am reading about semantic and syntactic anomalies. So when a semantic anomaly occurs there is supposed to be a negative peak in ERP at 400ms. I have some questions how this this component was established and how the data is collected?

Is there an assumption that these peaks happen in all individuals always at the same time? Is it only an assumption or is there evidence of invariability? Let's say the peak happens at 420 ms would you than use the amplitude from 400 ms or from 420?

Is the data collected from the whole brain and averaged out?

  • 1
    $\begingroup$ Can you link the paper? $\endgroup$
    – AliceD
    Jun 30, 2020 at 12:38

2 Answers 2


The semantic N400 and syntactic P600 are some of the most well established ERP components in the literature. They were discovered decades ago and the results have been replicated hundreds of times in other papers. There is just no question that they are present and very robust. That's not saying they are there 100% of the time in 100% of people. All we know is that it's a very easy response to get out and by now it would be surprising / worthy of investigation if people didn't show them in the standard paradigm.

There is a paper from a few years ago that looks at the 30 years of N400 research and all the papers that found the effects. It's probably a bit out of date now but it does a great summary of the research up until that point.

I think Hillyard & Kutas (1983) is the original paper.

ERP components are not always so tightly linked in time. You can module them based on task context and shift them by all sorts of ways so it's never that the peak is exactly at 400 ms post-stimulus. It can start WAY before then and end WAY after then. Broadly, we use N400 to talk about a large response that peaks "in that time window" (so, from 300-500ms). I wouldn't get too fixated on the specific milliseconds. Therefore you don't "use" the amplitude for anything (not sure I understood that part of your question).

Typically you see N400 over central-parietal electrodes. There is a bit of confusion over exactly what the cortical sources are but MEG and fMRI do support the idea it's a strong dipole with the positive component slightly in the LH and therefore EEG electrodes over central and parietal sites pick up the strongest signal. This also gets at what you use as your reference electrode because EEG potentials are differential and the patterns you see on the head are very dependent on reference / ground settings.

Typically you would group electrodes over general position and look at central-parietal ones or you could plot individual responses (both ways would still show the N400). As with anything in science, it's up to you what you show.


Just for reference, I'm an undergrad but I'm doing an N400 study currently.

As for your latter questions, since I think the former have been answered really nicely(it is kinda a harder concept, data visualization has helped me a lot)

The process I used summarized is:

  1. epoch data (i.e. take out time points you aren't interested in). I set my time window from -100 to 1200ms from the point they perceived the stimuli.
  2. artifact detection, where you find trials where someone blinked or an electrode contact was off (aka impedance)
  3. compile this into a .csv file, data is saved per electrode cite. I only wrote 4 electrodes to .csv because it is a lot of data and we are more interested in the front left hemisphere in observing the N400 because that's where the largest neural activation is generally seen.
  4. After you have clean data to work with, you can average mean amplitude in a time window. I used 250 to 650ms. So you get one mean value to run statistical tests on. Or you can make a line plot with all the timepoints you want to see the classic N400 peak (hopefully). You usually need a pretty decent sample size for the ERP to show anything meaningful.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.