However, LAN effects are not restricted to syntactic violations (Kaan & Swaab, 2003a) so, to the extent that syntactic difficulty is captured by word information, we could have observed a LAN effect in our data. In a review of the literature, Van Petten and Luka (2012) write that most ERP studies that compare higher- and lower-cloze (but semantically congruent) words find not only the N400 but also an (E)PNP in response to the lower-cloze word. Hence, there was every reason for our word surprisal measure to predict the (E)PNP as well. In fact, results by Thornhill and Van
Petten (2012) suggest that surprisal should be more predictive of the (E)PNP than of the N400: They found that presenting a low-cloze (i.e., high surprisal) see more synonym of a highly expected word elicits an (E)PNP but EX 527 ic50 no N400. Kaan and Swaab (2003b) found an anterior post-N400 positivity, much like the (E)PNP, in response to syntactic disambiguation. Be reminded from the Introduction that entropy reduction can be viewed
as the amount of ambiguity resolved by a word or PoS. Therefore, entropy reduction might predict the (E)PNP. Indeed, our exploratory analysis did reveal a potential (E)PNP effect of word entropy reduction, which closely followed findings on reading time in that the effect grew stronger with higher linguistic accuracy and larger lookahead distance. Somewhat 4��8C disappointingly, no such effect remained in the confirmatory analysis. Although this striking difference between the two data sets may well be a statistical fluke, it raises the question if there was any relevant difference between
the subject groups of the two analyses. There were no large differences in either mean age or gender (Exploratory: 29.5 years, 6 females; Confirmatory: 26.4 years, 4 females) but the groups might have differed in other properties. In any case, the possible effect of entropy reduction on (E)PNP deserves further study. The P600, which is a more posterior component than the (E)PNP, is well known to occur when there is a syntactic garden path (e.g., Kaan and Swaab, 2003b, Osterhout and Holcomb, 1992 and Osterhout et al., 1994). This has given rise to claims that it reflects a process of syntactic reanalysis that takes place when an initial sentence interpretation turns out to be incorrect (Friederici, 2002 and Osterhout et al., 1994). A garden-path effect is necessarily triggered by the appearance of a word with an unexpected syntactic category. As such, syntactic reanalysis should co-occur with high surprisal and, indeed, surprisal has been shown to account for particular garden path phenomena (Brouwer et al., 2010 and Hale, 2001). Levy (2008) proves that surprisal equals the extent to which the current input forces reallocation of the probability assigned to possible sentence interpretations.