Are there factive predicates? An empirical investigationSupplementary Material
Related article: http://muse.jhu.g.sjuku.top/article/864635
[Download PDF file
This PDF provides additional information about the following: (A) the clausal complements of the twenty predicates in experiments 1, 2, and 3; (B) the control stimuli in experiments 1, 2, and 3; (C) data exclusion criteria and details; (D) model details for experiments 1, 2, and 3; (E) comparisons of gradient and categorical ratings; and (F) mixture models applied to data from experiment 1a.
The relevant figures from the main article are repeated here in color, with their captions.
Mean certainty ratings by predicate in experiment 1a. Error bars indicate 95% bootstrapped confidence intervals. Violin plots indicate the kernel probability density of the individual participants’ ratings.
Proportion of ‘yes’ ratings by predicate in experiment 1b. Error bars indicate 95% bootstrapped confidence intervals. Vertically and horizontally jittered light gray dots indicate individual participants’ responses of ‘yes’ (coded as 1) and ‘no’ (coded as 0).
Mean certainty ratings by predicate, with number of discourses in parentheses, for 982 discourses in the CommitmentBank (de Marneffe et al. 2019). Error bars indicate 95% bootstrapped confidence intervals. Predicates included in our experiments 1a,b shown in bold.
Mean projection ratings by predicate, with 95% bootstrapped confidence intervals, for the seventy-eight predicates in the VerbVeridicality data set (Ross & Pavlick 2019), with labels for the fifteen predicates featured in our experiments.
Mean projection ratings by predicate, with 95% bootstrapped confidence intervals, for the 517 predicates in the MegaVeridicality data set (White & Rawlins 2018, White et al. 2018), with labels for the nineteen predicates featured in our experiments.
Mean inference ratings by predicate in experiment 2a, including the nonentailing and entailing controls. Error bars indicate bootstrapped 95% confidence intervals. Violin plots indicate the kernel probability density of the individual participants’ ratings.
Proportion of ‘yes’ ratings by predicate in experiment 2b. Error bars indicate 95% bootstrapped confidence intervals. Vertically and horizontally jittered light gray dots indicate individual participants’ responses of ‘yes’ (coded as 1) and ‘no’ (coded as 0).
Mean contradictoriness ratings by predicate in experiment 3a, including the noncontradictory and contradictory controls. Error bars indicate bootstrapped 95% confidence intervals. Violin plots indicate the kernel probability density of the individual participants’ ratings.
Proportion of ‘yes’ ratings by predicate in experiment 3b. Error bars indicate 95% bootstrapped confidence intervals. Vertically and horizontally jittered light gray dots indicate individual participants’ responses of ‘yes’ (coded as 1) and ‘no’ (coded as 0).
Mean veridicality ratings by predicate, with 95% bootstrapped confidence intervals, for the seventy-eight predicates in the VerbVeridicality data set (Ross & Pavlick 2019), with labels for the fifteen predicates featured in our experiments.
Mean veridicality ratings by predicate, with 95% bootstrapped confidence intervals, for the 517 predicates in the MegaVeridicality data set (White & Rawlins 2018, White et al. 2018), with labels for the nineteen predicates featured in our experiments.
Different simulated ratings distributions with overlaid optimal number of Gaussian components: One Gaussian component, mean = 0.4.
Different simulated ratings distributions with overlaid optimal number of Gaussian components: One Gaussian component, mean = 0.85.
Different simulated ratings distributions with overlaid optimal number of Gaussian components: Two Gaussian components, mean = 0.6.
Different simulated ratings distributions with overlaid optimal number of Gaussian components: Two Gaussian components, mean = 0.7.
Different simulated ratings distributions with overlaid optimal number of Gaussian components: Three Gaussian components.
Different simulated ratings distributions with overlaid optimal number of Gaussian components: Four Gaussian components.
