Szewczyk, J. 1 & Schriefers, H. . 2
1 Jagiellonian University, Cracow, Poland
2 Radboud University Nijmegen, Donders Institute for Brain, Cognition and Behaviour, The Netherlands
Recently several ERP studies demonstrated that the human language comprehension system anticipates specific words that are highly likely continuations of a given text. These studies used agreement relations that allow testing whether the noun’s lexical features are available before the noun is actually presented. However, it remains an open issue whether the language comprehension system can also make predictions that go beyond a specific word. We presented 120 short stories in Polish. Each story’s final sentence had a direct object that consisted of a prenominal adjective, and a masculine noun. The stories were constructed such that for half of them an animate direct object noun in the story’s final sentence was highly expected, and for the other half an inanimate DO noun. The critical DO noun was either semantically congruent with the preceding context, or incongruent (which was realized by using nouns with an animacy value opposite to the expected). In Polish, there is an interdependency between the semantic category of animacy (animate versus inanimate) and gender marking on prenominal DO adjectives. Since the prenominal adjective agreed in gender with the noun, adjective’s inflectional suffix reflected the animacy value of the noun, and thus the potential incongruity of the noun was visible already at the adjective. In order to ascertain that the prediction truly concerned a broad semantically defined class of nouns, we additionally manipulated stories’ constraint strength (defined as the best DO noun completion CP). In half of the items stories were (1) highly constraining, while in the other half (2) weakly constraining (fully crossed with Congruity). Adjectives with prediction-inconsistent suffixes elicited an N400 effect, relative to the same adjectives with prediction-consistent suffixes. Importantly, this effect was not modulated by context constraint strength. These findings are evidence that readers truly predict broad semantically defined classes of words.