Reversing expectations during discourse comprehension

Abstract

In two event-related potential experiments, we asked whether comprehenders used the concessive connective, even so, to predict upcoming events. Participants read coherent and incoherent scenarios, with and without even so, e.g. ‘Elizabeth had a history exam on Monday. She took the test and aced/failed it. (Even so), she went home and celebrated wildly’, as they rated coherence (Experiment 1) or simply answered intermittent comprehension questions (Experiment 2). The semantic function of even so was used to reverse real-world knowledge predictions, leading to an attenuated N400 to coherent versus incoherent target words (‘celebrated’). Moreover, its pragmatic communicative function enhanced predictive processing, leading to more N400 attenuation to coherent targets in scenarios with than without even so. This benefit however, did not come for free: the detection of failed event predictions triggered a later posterior positivity and/or an anterior negativity effect, and costs of maintaining alternative likelihood relations manifest as a sustained negativity effect on sentence-final words.

Keywords:

• discourse processing
• concessive connectives
• prediction
• event structures
• N400
• P600
• late negativity
• ERP

Acknowledgement

The authors thank Abigail Swain, Eric Fields, Kristina Fanucci for their assistance in data collection, and Ellen Lau and Einat Shetreet for their helpful comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Our use of ‘alternative world model’ in this sense should not be confused with the concept of ‘possible worlds’ in formal semantics.

2. In fact, most studies of sentence and discourse processing have not matched general schema-based lexical relationships in this way and have therefore not been able to distinguish between these two possibilities (see Kuperberg et al., 2011; Otten & van Berkum, 2007; Paczynski & Kuperberg, 2012, for discussion). For example, even in the classic example of an N400 effect, ‘She liked to take her coffee with cream and sugar/dog’ (Kutas & Hillyard, 1980), the attenuation of the N400 to sugar (versus dog) could, in theory, be driven by its closer semantic relationship with the general schema of coffee drinking, rather than by a more specific expectation of the most likely thing, after cream, that someone would put inside her coffee. Those studies that have used such schema-matched stimuli reveal a mixed picture, as discussed further in Experiment 2.

3. To exclude the possibility that the smaller N400 to critical words in the even-so coherent versus the plain coherent scenarios was driven purely by their later position in the sentence (see Van Petten & Kutas, 1990, for effects of word position on the N400), we looked at the N400 on the word following the critical word. We found no difference between these two conditions between 350 and 450 ms, F(1, 26) = 0.66, p > .4, suggesting word position alone did not produce this N400 difference.

4. Note that the enhanced P600 was driven entirely by a larger P600 to the incoherent critical words in the even-so (versus plain) scenarios, and that there was no P600 difference between the coherent even-so and plain scenarios. This indicates that the enhanced N400 reduction in the coherent even-so (versus the plain) scenarios cannot be simply explained by temporal and spatial overlap between the N400 and the P600 components at the scalp surface: if the N400 to coherent critical words in the even-so (versus plain) scenarios was being artificially ‘pulled down’ by an overlapping P600, we would have seen a larger (more positive-going) P600 to these coherent even-so (versus coherent plain) critical words.

5. A potential concern is that the lack of a N400 may be due to overlap of an earlier positivity. Visual inspection of more electrodes did indeed show that, at some of the central–posterior electrode sites, there was an early positivity between approximately 200–350 ms that appeared larger to critical words in the plain incoherent scenarios than the plain coherent scenarios. We carried out statistics within this time window to contrast these two conditions, collapsing across the 5-electrode sites where this effect seemed to be largest (CP1, CP2, Pz, P3, P4), but found no significant effect, F(1, 19) = 2.2, p > .1. In addition, we also redid the N400 analysis within the 350–450 ms time window, with a new baseline between 200 and 300 ms, i.e. we re-baselined right before the N400 time window. If the early difference masked the N400 in some way, we might expect to see a N400 difference emerge between the plain coherent and plain incoherent scenarios. This is not what we found: the effect of Coherence in the plain scenarios was non-significant, F(1, 19) < .1, p > .5, but remained significant in the even-so scenarios, F(1, 19) = 6.1, p < .05. Based on these two analyses, we think it is unlikely an early divergence in the waveforms masked the N400 effect in the plain scenarios.

6. A potential concern is that the absence of a N400 effect in Experiment 2 was due to a lack of statistical power due to the smaller number of participants run in Experiment 2 than in Experiment 1 (20 vs. 27). It is also possible that, in both Experiments 1 and 2, the P600 seen to the plain scenarios was due to the presence of so many even-so incoherent scenarios in the experimental environment, which may have encouraged comprehenders to engage in additional analysis to all types of incoherent critical words. Finally, the presence of this P600 in the plain scenarios may have masked any N400 on the scalp surface due to component overlap. To address these three concerns, we carried out an additional Experiment 3 in which a new group of 16 individuals read the same plain scenarios, but with no even-so scenarios in the experimental set. They carried out the same comprehension task as used in Experiment 2. Results are reported in Supplementary material at: http://www.nmr.mgh.harvard.edu/kuperberglab/publications.htm. To summarise our findings: (1) we saw no N400 or P600 coherence effect (although, as in Experiments 1 and 2, we did see a prolonged negativity effect on the sentence-final word of the incoherent versus coherent plain scenarios, indicating that participants were engaged in comprehending the sentences) and (2) when we pooled the ERPs evoked by critical words in the plain scenarios of Experiments 2 and 3 to give a total of 36 participants (exceeding the number of participants in Experiment 1), we still did not see any sign of a N400 effect. Based on these results, it seems unlikely that statistical power alone explains the absence of N400 in Experiment 2, or that the absence of the N400 effect in Experiment 2 was simply an artefact of component overlap. Finally, the absence of a P600 effect in Experiment 3 provides some preliminary support for the idea that the large number of even-so scenarios in the wider experimental environment, leading to the P600 effect to the plain scenarios in both Experiments 1 and 2, although this is speculative and requires systematic follow-up.

7. In some of our previous discussions of the P600 component, we have described these type of near certainty event predictions as arising from a ‘semantic memory based analysis’, i.e. stemming from the interaction between context and real-world knowledge stored in long-term semantic memory. By definition, incoming words that violate strong semantic memory-based predictions will, when fully integrated, yield an event representation that is highly implausible/impossible (see Paczynski & Kuperberg, 2012, for recent discussion). What the present study shows is that a P600 can also be produced when event predictions are based on an ‘alternative world model’ and full integration of a target word outputs a discourse representation that is highly incoherent but not necessarily ‘implausible’ with respect to real-world knowledge.

Additional information

Funding

This work was funded by the National Institute of Mental Health (R01MH071635 to G.R.K.).