Review of When Peanuts Fall in Love: N400 Evidence for the Power of Discourse PSC 132 – Language and Cognition In the article When Peanuts Fall in Love: N400 Evidence for the Power of Discourse, Nieuwland and Van Berkum addressed the effects of discourse level context and real-world knowledge on sentence level language cognition. More specifically, they addressed the how the brain comprehends fictional stories in which inanimate objects act in an animate fashion (e. g. a dancing peanut). The researchers’ aim was to examine how the brain handles stories that are internally coherent yet externally impossible.
They addressed how we can read and enjoy stories such as Alice in Wonderland, despite the many animacy violations. For example, based on real-world knowledge, our brain knows that a deck of cards cannot ‘paint the roses red’ and a girl cannot hold a conversation with a clock. And yet we can read a story about such events and comprehend it as if it were feasible. The article begins by addressing several past theories about how the brain processes these violations. The “two-step” model theorizes that local semantics at the sentence level cannot be overruled by the larger discourse of a paragraph or story (e. . Millis & Just, 1994). Therefore, animacy violations will always require more processing, shown through an N400 effect after the violation. The opposing theory is more interactive, a “single-step” model (e. g. Trueswell & Tanenhaus, 1994). This theory states that local semantic cues do not take precedence over globally supplied contextual clues; therefore animacy violations within a discourse may not be any more difficult to integrate. To address these issues, the researchers used event related potentials (ERPs) to monitor brain activity.
Specifically, they used EEGs to monitor for an N400 effect (i. e. , a negative shift in the waveforms that began around 300 ms and peak around 400 ms), in order to determine if it is possible for contextual information in a discourse to overrule a semantic and animacy violation within a sentence. The N400 amplitude has been shown to be inversely affected by the semantic compatibility of a given word and its context; the more cognitive effort is involved in integrating a word into an ongoing context, the larger the N400 amplitude that will be elicited by that word.
The researchers used the N400 to test whether sentence level violations could be overruled by a larger discourse, much like the experiment by Nirit, et al (2002), although that experiment did not use animacy violations, which are a more distinct violation of real-world knowledge rather than simply a semantic violation. In order to test their theory, the researchers used two separate experiments oriented towards different findings. In the first experiment, the researchers constructed sixty naturally spoken Dutch stories, each of which were six sentences long and involved either a woman and a man or a woman and an inanimate object (e. . , a yacht). In the inanimate condition, each sentence contained an animacy violation (the control condition contained no such violations). However in sentences one, three, and five, the sentences contained a selection restriction violation, so those sentences were used for evaluation. Thirty of these animate stories and thirty inanimate stories were mixed in with ninety filler stories. The participants were thirty-one right handed college students, all native Dutch speakers with no neurological impairment. Participants’ only task was to sit still and listen to the recording of the stories told by the same female researcher.
The participants EEGs were recorded from thirty standard scalp locations while they were listening to the stories, with ocular and muscular artifacts corrected. The ERPs were time-locked to a 300-600ms time window after the critical word. The second experiment was set up in a similar fashion. Sixty stories were constructed, each with six sentences and two entities, a woman and an inanimate object. The variable in this experiment was a sentence within the discourse in which the inanimate object is described either with a context-appropriate but animacy-violating property (e. . ‘the peanut was in love’), or with a canonical but highly context-inappropriate property (e. g. ‘the peanut was salted’). In addition to a similar mix of 60 stories and 90 filler stories seen in the last experiment, there were also 120 independent sentences mixed in. The participants were forty-two right handed college students, all native Dutch speakers with no neurological impairment, whose only task was to sit still and listen to the recording of the same female speaker reading the stories and sentences. The EEGs were recorded the same way as in the first experiment.
The experiments showed different but complementary results. Analysis of the first experiment showed that the inanimate condition in the first sentence of the story elicited a large N400, as expected. However, there was not a significant difference in the N400 of the animacy violations in the third or fifth sentences. This is significant, because it is evidence towards the interactive “single-step” model discussed above. If the “two-step” model was correct, this experiment should have shown an N400 for both the third and fifth sentences in addition to the first sentence.
However, no such effect was seen, showing that discourse can make it easier to integrate animacy violations in a larger discourse. The second experiment took the findings one step further and tested whether the animacy violation could be overruled by a larger discourse. Experiment two’s results showed that predicates which were locally fitting but contextually inappropriate elicited a stronger N400 than did predicates that were contextually appropriate but semantically inaccurate.
This showed that discourse has the power not only to make it easier to integrate animacy violations, but even to make them easier to integrate than a word that would normally be easier to integrate. This should be impossible, according to the “two-step” model. The basic question that the researchers addressed was whether a local animacy violation disrupts semantic processing when the violation is congruent within a larger discourse. According to the two-step approach it always does, regardless of context. The one-step approach says it does not, because discourse-level and local animacy constraints are processed simultaneously.
The first experiment supports the one-step approach because while the first sentence with a violation elicited a high N400, the same violation in the third and fifth sentences did not, as the two-step model predicted. These experiment shows that it is possible that an interpretive context can completely overrule the effects of a local semantic violation. There are several counter arguments which can be made against the researchers’ interpretation of the findings in the first experiment. The first argument that can be made is that the decline in the N400 is the esult of simple lexical repetition rather than any overruling of local violations. However, this argument is not credible because the lexicons “she” and “he” have extremely high frequency of repetition in the English language and yet mistakes between them are easily discernable. Therefore, this argument has no validity. The second criticism of the interpretation is that subjects became indifferent during cartoon-like stories with unrealistic content and therefore stopped paying attention to them. And yet when asked, all subjects said that they found the stories easily comprehendible and entertaining, refuting this criticism.
The second experiment was designed to refute some of these possible counter-arguments. This second experiment showed that the effects of contextual appropriateness can override the effects of animacy and real-world plausibility combined. These results cannot be explained away by simple lexical repetition because none of the variable words had been featured in the story prior to the test. As for the second argument, that participants got bored and stopped paying attention to the story, the anomaly was presented near the end of the story, so this experiment falsifies the claim that participants were simply tuning out an ‘odd’ story.
The essential information from this work is the absence of an elicited N400 after both the verb and the inanimate noun in the fifth sentence of experiment one. This suggests that discourse context information can completely neutralize local animacy violations. Experiment two expounds upon the findings of experiment one to suggest that discourse can not only neutralize the local animacy effects, but actually outweigh the combined effects of local animacy violations and real-world plausibility.
These findings were later supported by Hald, et al, (2007), which found that local discourse has the power to provide a context for new information to be more acceptable, for example a peanut to be in love. Their findings support this research because they also found that even “correct” world knowledge information can be less acceptable to the brain if the local discourse context does not support that world knowledge (Hald, et al, 2007). While the experiments and their findings addressed many of the possible arguments and is procedurally robust, there was one shortcoming in how they analyzed their data.
In the first experiment, the experimenters only analyzed half of the animacy violations, namely those in which the verb preceded the violating noun. The other half of the sentences, in which the violating noun was followed by the incongruous verb were not analyzed. This leaves the door open to questions on how the syntax of the sentence affects its comprehension and integration into the discourse as a whole. Other researchers who are interested in the same discussion about discourse and processing include Camblin, et al, who tested discourse congruence and lexical association during sentence processing.
They used a method of combining ERPs and eye-tracking procedures while participants read short stories. They were able to show evidence that discourse representations can have an earlier effect on processing than local manipulations of association, which supports the conclusions of this article. This article and the experiments done by Nieuwland and Van Berkum lend insight into how the brain integrates sentences and words into a larger discourse, and helps us to understand language integration as a whole. References . Camblin, C. , Gordon P. C, and Swaab T. Y. (2007) The Interplay of Discourse Congruence and Lexical Association During Sentence Processing: Evidence from ERPs and Eye Tracking. Journal of Memory and Language, 56, 1, 103-128 2. Hald, L. A, Steenbeck-Planting, E. G, and Hagoort, P. (2007). The Iinteraction of Discourse Context and World Knowledge in Online Sentence Comprehension. Evidence from the N400. Brain Research. Special Issue: Mysteries of meaning. 1146, 210-218 3. Millis, K. K. , and Just, M. A. (1994).
The Iinfluence of Connectives on Sentence Comprehension. Journal of Memory and Language, 33, 128–147. 4. Salmon, N. , and Pratt, H. (2002). A Comparison of Sentence- and Discourse-Level Semantic Processing: An ERP Study. Brain and Language. 83(3), 367-383 5. Trueswell, J. C. , and Tanenhaus, M. K. (1994). Toward a Lexicalist Framework of Constraint-Based Syntactic Ambiguity Resolution. In C. Clifton, Jr. , L. Frazier, & K. Rayner (Eds. ), Perspectives on Sentence Processing (pp. 155–179). Hillsdale, NJ: Erlbaum.