A noisy channel account of ERP differences in sentence comprehension
Veena D. Dwivedi1, Janahan Selvanayagam1, Victoria Witte2, Harmonie Chan1, Ted Gibson3; 1Brock University, 2Heidelberg University, 3MIT
We used event-related brain potentials (ERPs) in order to investigate the interpretation of sentences that contained errors of deletion vs. errors of insertion. We investigated plausible (ditransitive) such as (i) The aunt mailed the letter to her niece by post as compared to sentences that deleted the preposition to, resulting in (double object) implausible sentences such as (ii) #The aunt mailed the letter_ her niece by post. In addition, related plausible (double object) sentences such as (iii) The aunt mailed her niece the letter by post were modified with the insertion of to resulting in an implausible (ditransitive) sentence such as (iv) The aunt mailed her niece #to the letter by post. All sentences were followed by yes/no comprehension questions such as Did the niece receive something? /Did the letter receive someone? Given that noisy channel models propose that deletion errors are more likely to occur than insertions, perceivers should be more likely to revise sentences with deletion vs. insertion errors. Thus, we further explored the ERP components elicited by these sentence types, based on previous work (Dwivedi et al., 2017). Previously, both sentence types exhibited a slow-going negativity, which might have been due to sentence length. These were shortened in this study. Deletion sentences still did elicit negative-going waveforms, insertion sentences did not. Instead, these elicited a late positivity. These effects were interpreted in terms of likelihood of revision; a corrected deletion sentence taxes working memory resources, whereas less likely correction at insertion sentences results in perceived ungrammaticality.
Topic Area: LANGUAGE: Syntax