Martin, A. E. 1 , Nieuwland, M. S. 1 & Carreiras, M. 1, 2, 3
1 Basque Center on Cognition, Brain, and Language, Donostia-San Sebastián, Spain
2 Ikerbasque Foundation, Basque Foundation for Science, Bilbao, Spain
3 Department of Basque Philology, UPV, Bilbao, Spain
Language comprehension requires retrieval of recently-processed representations from memory. How is missing information recovered during online sentence comprehension and what factors affect its retrieval? During ellipsis (the interpretation of unpronounced material), retrieval cues at the ellipsis site may make direct contact with the antecedent in memory. Under this mechanism, processing complexity is determined by representational factors, such as the retrieval cues’ diagnosticity or match to the target item or items in memory. Tracking the online processing of ellipsis in Castilian Spanish allowed us to tap into these mechanisms as a function of the relevant representations’ diagnosticity. We recorded event-related potentials from 22 participants while they read 120 sentences containing elided noun phrases in one of 4 conditions (“Marta se compró la camiseta que estaba al lado de la falda/el vestido y Miren cogió otra/*otro igual para salir de fiesta.”). The determiner ('otra'/'otro') correctly or incorrectly cued the retrieval of the antecedent ('la camiseta'), and occurred in the context of a matching or mismatching local agreement attractor ('la falda'/'el vestido') of similar lexical frequency. We predicted that if gender morphology serves as a retrieval cue, then incorrect gender would elicit a LAN/P600 effect compared to correct gender. Moreover, if retrieval is vulnerable to interference from locally matching representations, this ERP effect would be reduced in the context of a matching local agreement attractor. No P600 effect was observed. Instead, incorrect determiners evoked a sustained, anterior negativity compared to correct ones from 300ms onwards. This negativity was reduced in the context of a matching local agreement attractor, especially at central and posterior electrodes. Consonant with results regarding referential ambiguity, we take this anterior negativity to reflect the degree of cue diagnosticity to relevant linguistic representations during retrieval and interpretation of ellipsis.