This study is an attempt to further characterize individual differences in the judgment of underinformative scalar implicature (USI) statements (e.g. some cats are mammals) and reconcile the seemingly contradictory findings of three previous studies (Nieuwland et al., 2010; Noveck & Posada, 2003; Antoniou et al., 2016) which do not converge on whether autistic-like traits modulate differences in EEG response and truth-value judgments (TVJs) of USIs. Results show that the effect of autistic-like traits on the processing USIs may be smaller than previously thought, and that processing and TVJ of USIs might be separate processes which do not seem to correlate.

​

ERPs were measured during a TVJ task to USI statements; followed by an Autism Quotient (AQ) questionnaire measuring autistic-like traits. Preliminary results from 40 participants show that, the mean number of ‘false’ responses to 40 USI items was 2.75 (SD=4.89), and only 4 participants judged >10 USI as ‘false’. This invariance in TVJs makes it impossible to test these data for individual differences based on behavioral response. Participants were divided into low vs. high scores on the communication subscale of the AQ task, plots (fig.1) illustrate that both groups had a P600 response to USI items and an N400 to false items, with the higher AQ group (more autistic-like traits) having larger effects. Correlating the communication AQ score and mean peak amplitudes for the difference waves between the Underinformative and True conditions at 300-500ms at 8 electrodes showed a small correlation between the two (r=.23), significantly smaller than the correlation shown by Nieuwland et al. (r=-.53). These results seem to imply that having autistic-like traits may have less of an effect on how USIs are processed, and that there might not be a straightforward mapping of different ways of processing USIs to different judgments.  

​

We used an unsupervised learning algorithm (K-means clustering) to identify underlying patterns in the data. Results showed clear evidence for an optimal division to three subgroups, (fig. 3) in the N400 and P600 time windows. The algorithm output provides robust evidence for the real time neurodynamics of scalar implicature computation. The grand mean showing a P600 to USIs and N400 to false sentences clearly does not reflect individual differences in response to these stimuli. Future research needs to provide a functional interpretation for these individual level traits.