Negation causes NPI and NCI illusions in Czech

Grammaticality illusions are cases where comprehenders fleetingly consider ungrammatical sentences to be acceptable. The phenomenon been used to tap into the mechanism of online negative polarity item (NPI) licensing (Muller & Phillips, 2020). The effect has high specificity and not all environments that ordinarily license NPIs create illusions—sentential negation being the prime case that is ineffective as an illusory licensor (Orth et al., 2021). In the current study, we contrast NPIs with another class of expressions, namely negative concord items (NCIs), which are said to be licensed in the syntax unlike NPIs. We test whether sentential negation gives rise to illusions with Czech NPIs and NCIs respectively, predicting illusions for NCIs and no effect for NPIs. However, in three speeded acceptability judgement experiments, we find illusions with both types of expressions. Our data are inconsistent with the active scope (Orth et al., 2021) and scalar theories (Muller, 2022; Schwab, 2023). The cue-based retrieval (Vasishth et al., 2008) and the feature percolation approaches (Eberhard et al., 2005) are both compatible with the results. We argue in favour of feature percolation as the best explanation for both our Czech data as well as the data from other languages, appealing to cross-linguistic differences in the placement of negation within syntactic structure. While negation is high up in the syntactic structure in Czech and can therefore percolate up and cause illusions, it is low in English, not allowing for illusions to arise. This result highlights the importance of language-specific syntactic structure for online comprehension.