Previous studies (Fine et al 2013, Fine & Jaeger 2016) have shown that people get less garden-pathed by sentences like (a)over the course of the experiment - i.e. as the experiment progresses, there is a reduced relative reading time in the disambiguating region
(a) The experienced soldiers warned about the dangers conducted the midnight raid
The authors claim that this is because people change their probability distribution and increase the probability of reduced relative clauses (RRCs) in order to match the statistics of the environment. In other words people adapt to expect RRCs more frequently because they are in an environment with a larger proportion of RRCs than usual. This account (if true) raises several questions. What are people adapting to and how are they changing their probability distribution? What triggers this adaptation? Is it error-driven? In this ongoing project we are trying to replicate the syntactic adaptation effects and tweak the paradigm to increase the effect size. We then want to use computational models to make specific predictions for the above questions and test these predictions with self-paced reading data.
The increase in the use of singular 'they' as a gender neutral pronoun raises the question of whether this new usage requires any additional processing effort and if yes what causes it. Consider the following sentence:
John decided to treat themselves to sushi
One might expect this sentence to have additional processing effort (when compared to 'John decided to treat himself to sushi' or 'John and Mary decided to treat themselves to sushi'). However is this because of a number mismatch (i.e. 'John' is singular whereas 'themselves' is plural), a gender mismatch ('John' is marked for gender whereas 'themselves' is unmarked for gender) or both? We ran an ERP study exploring this question by presenting participants with antecedents that mismatched in both gender and number ('John', 'some man') and antecedents that mismatched in gender only ('the participant', 'someone'). Note, while antecedents like 'the participant' are technically unmarked for gender, it is possible that participants visualize a specific person and assign gender to them. In order to minimize the likelihood of this, we ran participants who frequently interacted with gender non-binary and transgender individuals (half of whom identified as being gender non-binary themselves). Our results suggest that the processing effort is caused by gender and not number mismatch.This work was presented as a talk at the 58th Annual Meeting of the Psychonomic Society. Here are the slides.
It was also presented as a poster at the 31st Annual CUNY Conference on Human Sentence Processing. Here is the poster.