Processing of Noun Complement Complex NPs with subject and object pro in Korean Nayoung Kwon, Jiwon Yun, John Hale and John Whitman Object relatives (ORs) are reported to cause heavier processing loads than subject relatives (SRs) [1], [2], [3], [4], [5]. This asymmetry has been explained in terms of the relation between filler and gap, either linear (Dependency Locality Theory, DLT; [6]) or structural (Structural Distance Hypothesis, SDH; [7]). Another approach accounts for the asymmetry in terms of structural uncertainty, independently of the filler-gap dependency (Entropy Reduction Hypothesis, ERH; [8])/ The ERH accounts for the relative processing difficulty of ORs in terms of the relatively high degree of local ambiguity they introduce. Interestingly, these accounts predict an asymmetry between subject and object gaps not just in relative clauses, but in noun complement complex NPs (NCCNPs) like the fact that S. This study aims to evaluate these accounts by investigating the processing of NCCNPs in Korean. A self-paced reading time (RT) study was conducted with NCCNPs containing subject/object pro (2 & 4), both with/without context (1 & 3). DLT predicts a processing disadvantage for SproNP (2) over OproNP (4) based on the longer linear distance between pro and its antecedent in SproNP. SDH predicts a processing disadvantage for OproNP (4) based on longer pro-antecedent structural distance. Importantly, both accounts predict the difficulty to occur at the pro-antecedent association position (chancellor-top, W10 in (2 & 4)). ERH also predicts a disadvantage for OproNP like SDH [9]. However, unlike SDH, ERH predicts the difficulty to occur at the embedded verb (threaten-and, W7) and the following head noun (fact- nom, W8), based on higher transient ambiguity of the derivational continuation at these positions in OproNP. On the other hand, ERH predicts no processing asymmetry at the pro-antecedent association position. Analysis of RTs showed an overall parsing advantage for conditions with context in several words positions, suggesting that supporting contextual information helps processing. Importantly, there were significant main effects pro type at the head noun (fact-nom, W8; p < .007) and the following word (was.revealed-as, W9; p < .0001). OproNP took longer to read than SproNP at these positions. No such asymmetry was found at the pro-antecedent association position (chancellor-top, W10). The findings are consistent with ERH but not with DLT and SDH. They are incompatible with DLT in terms of the direction of asymmetry and the position where the effects were observed. The findings are compatible with SDH in terms of the direction of asymmetry but not in terms of the position effects. Particularly, given that the effects were observed before the antecedent is reached, and no asymmetry was found at the pro-antecedent association position, the findings do not support models based on the relationship of filler/antecedent and gap/pro (and probably SR/OR as well) should be cast in terms of general parsing difficulty associated with probabilistic structural disambiguation, independent of long-distance dependency.