The N400 and the P600 (2 hours)

↵ Back to module homepage

Look at the pairs of sentences below. What is the difference between (1a) and (1b), and what is the difference between (2a) and (2b)?

1. • I drink my coffee with cream and sugar.
   • I drink my coffee with cream and dog.
     
2. • All the students were in the classroom.
   • All the students was in the classroom.

Hopefully you noticed that the difference between the first pair of sentences is one of plausibility (sugar is a word that plausibly fits in that sentence, whereas dog is not), whereas the difference between the second pair of sentences is one of grammaticality (were agrees with the plural subject of the sentence, whereas was does not.)

It turns out that these phenomena are associated with different ERP patterns as well. When people read or hear a word that is implausible or unexpected in the context, like dog in I drink my coffee with cream and dog, this tends to elicit a more negative ERP (compared to a word that is plausible and expected in the context, like sugar) about 400 milliseconds after hearing the word, as shown in the figure below. This effect (specifically, the difference between the ERP elicited by the less plausible word and the ERP elicited by the more plausible word), is called the "N400" ("N" for negative, "400" for 400 ms, the latency at which this effect tends to be strongest).

On the other hand, reading a word that is not grammatical in the context, like was in All the students was in the classroom tends to elicit a more positive ERP (compared to a word that is grammatical in the context, like were) about 600 milliseconds after hearing the word, as shown in the figure below. This effect is called, accordingly, the "P600".

Based on these differences we've seen, you should be able to try to predict what kind of ERP response will be elicited for various sentences. Let's give it a try.

First we'll look at some sentences from Alemán-Bañón and colleagues (2017). Before looking at these sentences, you need to know a tiny bit of background about Spanish grammar: in Spanish, adjectives have to agree with nouns in terms of their number. This is a feature English doesn't have. In English we have this with demonstratives (e.g., a plural noun like dogs needs a plural demonstrative like these or those, not a singular one like this or that; that's why these/those dogs is grammatical whereas *this/that dogs is not), but adjectives don't care about singular or plural; i.e. we use the same adjective black to say either black dog(singular) or black dogs (plural). In Spanish, however, if a noun is plural, any adjective describing it also must be plural; it would be like saying blacks dogs. Now, let's look at some of the sentences Alemán-Bañón and colleagues used in their ERP experiment:

What do you expect the ERPs elicited by these sentences (specifically, by barato in each sentence) to look like? After you've thought about it, click below to reveal the answer.

Alemán-Bañón et al. (2017) results

This figure is a bit more complicated than the ones we've seen so far, because it's showing the ERP waves at many different places over the head instead of just at one place. However, if you look at the middle of the head (the "C" row) or back of the head (the "P" row), you will see that the the light gray line (the ungrammatical adjective) elicits a more positive ERP than the dark black line (the grammatical adjective), particularly around 600 ms after seeing the word. (Note that in this figure, as in many ERP figures, the scale is upside-down; negative is plotted upwards and positive downards. This is shown in the upper left corner of the graph.)

This is a classic P600 effect.

Now let's look at another pair of sentences, this time from Xu et al. (2013):

Again, think about what you expect the ERPs elicited by "her" in these two sentences to show, and then click below to reveal the answer.

Xu et al. (2013) results

This figure, like the previous one, shows negative plotted upwards and positive downwards. The red line represents the ERP elicited by "her" in the condition where it doesn't match the plural antecedent, and the black line represents the ERP elicited by "her" when it does match the singular antecedent. The red line, for the mismatch condition, is lower (more positive) than the black line for the match condition. This difference seems to be biggest 400-800 milliseconds after seeing the word.

This is also a classic P600 effect.

Now let's consider another example, this time from Ye et al. (2006). To understand these, you need to know that Chinese has a function word 把 (glossed BA), which introduces an object in an SOV structure. The default sentence structure in Chinese is SVO (subject-object-verb), but SOV (subject-verb-object) is possible with this BA. In other words, a Chinese sentence can look like "I ate the apple", or "I BA the apple ate"; these mean roughly the same thing.

(Note that the verb 裁, while translated as "cut", is only used for cutting stuff like fabric, e.g. with scissors. The sort of cutting involved in, e.g., chopping down trees, uses a different verb.) Again, think about what you expect the ERPs elicited by "her" in these two sentences to show, and then click below to reveal the answer.

Ye et al. (2006) results

Here we have a messier situation. The thin gray line represents the ERP for "cut" in the correct context, and the thick dotted black line represents the ERP for "cut" in the context it doesn't fit (descring chopping a tree rather than cutting cloth). Based on what you read at the beginning of this activity, you might have predicted an N400 for this comparison. But the results are not so clear-cut. The ERP for "cut" when it doesn't fit the context is indeed more negative than the ERP for "cut" when it does fit the context, but this only happens for about the first 200 milliseconds after hearing the word. This might be an early N400 (although the authors of the paper were hesitant to claim that). But later, from at least 400-1000 milliseconds after hearing the word, the ERP for the implausible "cut" is actually more positive than the ERP for the plausible "cut".

What it actually looks like here is that there is both an N400 effect and a P600 effect.

If you're following along carefully, you may have noticed that these results challenge what you read at the beginning of the activity. Based on what you read at the beginning of the activity, you may have assumed that grammar errors elicit P600 effects, whereas violations of plausibility or expectations elicit N400 effects. But that doesn't seem to be the full story, because here we see one manipulation (an implausible/unexpected verb) which seems to elicit both an N400 and a P600. There are two different conclusions we might make from this. We could conclude that Chinese works differently than English somehow, i.e. that meaning and plausibility are fundamental parts of "grammar". (That is in fact the claim that Ye and colleagues [2006] make, although I don't really agree with it.) Or we could conclude that our understanding so far of the N400 and P600, based on what I said at the beginning of this activity, is wrong, or at least is not the full story. (In the last activity of this module you will have an opportunity to read a paper by Bornkessel-Schlesewsky and colleagues which integrates both of these conclusions; it basically says that the N400 and P600 do indeed represent something more nuanced than just "grammar" versus "plausibility", and that languages differ in which kind of ERP effects they have based on the different ways that languages mark certain kinds of information.)

Now let's try one last example, from Kim and Osterhout (2005). What do you think the ERPs here will look like?

• The hearty meal was devoured by the kids.
• The hearty meal was devouring the kids.

Kim and Osterhout (2005) results

Here, like in the Spanish study, we have ERPs for several places across the head. The important pattern is easiest to see at Pz, which is in the middle back of the head (the bottom center of this figure). Here, devouring elicits a more positive ERP than devoured, especially around 600 milliseconds; in other words, this looks like a P600 effect.

This last result may look surprising. The hearty meal was devouring the kids is a perfectly grammatical sentence; it has no grammatical error. The problem with it is plausibility (we know that meals don't usually devour kids). Based on what you've seen so far, you might then expect that it would elicit an N400, since we said above that plausibility violations tend to elicit N400s. Why did this elicit P600 instead?

This shows that there is a fundamental problem with assuming that certain linguistic phenomena (like grammar, syntax, semantics, plausibility, morphology, or whatever) are associated with ERP components. People often like to say that the N400 is elicited by "semantic" violations and the P600 by "syntactic" violations, but that is clearly not the full story; here we have a sentence with no syntactic problem and only a plausibility problem, and yet it elicits a P600 rather than an N400. This experiment, and others reporting a similar effect (which at the time was sometimes called the "semantic P600"), was very influential for just this reason; it highlighted that we need to think about ERPs in a more nuanced way.

Instead of paying attention to what kind of violation there actually is in the sentence, let's instead pay attention to what the language processing system in our mind might do when faced like a sentence like this. To make that more concrete: think about how you might try to correct this sentence as you are reading it, i.e., what you might identify as the "problem" with this sentence.

There are actually two things your language processing system might do. It might assume, as I assumed in the discussion above, that there is a problem with the plausibility of this sentence. In other words, your language system might trust that the syntactic structure of this sentence is right, and that the problem is that the sentence has a weird meaning (it's weird for a meal to devour kids). Thus, the way to "correct" the sentence in this situation would be to put in a different verb (e.g., maybe something like "The hearty meal was tempting the kids.").

On the other hand, though, your language processing system might instead assume there is a problem with the syntax of the sentence. In other words, your language system might trust that the general meaning of the sentence is normal and plausible, and instead there's just a grammatical error, sort of a typo. It might assume that the only error in the sentence is accidental use of passive voice (verb-ed by) rather than active voice (verb-ing). In that case, the way to "correct" the sentence would be to keep "devour", but just fix the suffix (change it to "devouring" instead of "devoured by").

If your language system chooses to trust the syntax and interpret this sentence as having a weird meaning we might expect it to throw an N400 at us (under the assumption that implausible/unexpected words elicit an N400). On the other hand, if your language system chooses to trust the meaning and instead interpret the sentence as having a grammatical error, we might expect it to throw a P600 at us (under the assumption that grammatical errors elicit a P600). What we see from the above results is that the latter possibility seems to be the true one: even though the sentence doesn't have a grammar error, the language processing system acts as if it does, because it trusts the meaning over the grammar that it sees.

There are many other nuances to this situation; there are other languages, and other situations, in which there are N400s (or both P600s and N400s) for these kinds of sentences. There is still debate over what these patterns mean for our understanding of how language works and our understanding of how ERPs work. So, the story I have just told might not be the right story. But regardless of what the right story ends up being, I think this example is very instructive for teaching us that we shouldn't just think about ERPs as products of things like "syntax" or "semantics", but that we have to think more carefully about what actual processes the mind is going through as it attempts to understand language.

The kind of sentence from the Kim & Osterhout paper (e.g. The hearty meal was devouring the kids) has what is often called a semantic reversal anomaly, because it involves "reversing" who usually does what to whom (usually kids eat meals, but this sentence is saying the meal eats the kids). To finish this task, find one other experiment examining what kind of ERP is elicited by a semantic reversal anomaly, and summarize it in 250 words or less. Some papers include multiple experiments; you can just summarize one experiment. To find experiments, you can search online (using e.g., Google Scholar), or you can start from some known papers on this phenomenon and search outwards (by checking related papers that these papers cite, or by checking more recent papers that cite this paper; in Google Scholar you can look up a given paper and then see all the papers that cite it). As a starting point, I recommend either of the below papers by Wing Yee Chow, which are both interesting and easy to read:

1. Chow, W. Y., Lau, E., Wang, S., & Phillips, C. (2018). Wait a second! Delayed impact of argument roles on on-line verb prediction. Language, Cognition and Neuroscience, 33, 803-828.
2. Chow, W. Y. & Phillips, C. (2013). No semantic illusion in the "Semantic P600" phenomenon: ERP evidence from Mandarin Chinese. Brain Research, 1506, 76-93.

When you have finished this activity, continue to the next section of the module: "How ERP components should and should not be understood".

by Stephen Politzer-Ahles. Last modified on 2021-05-11. CC-BY-4.0.