Table 5.
CEREBRA match results and statistical significance compared with human judgements across sentences where at least three of the four subjects agreed.
| (A) Matching CEREBRA predictions for approaches one to three and chance with human data | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Three of four participants average agreement (3 ratings) | |||||||||||
| Ratings | Human | Cerebra#1 | Cerebra#2 | Cerebra#3 | Chance | ||||||
| −1 | 618 | 478 | 484 | 463 | 8 | ||||||
| 0 | 456 | 2 | 2 | 3 | 0 | ||||||
| 1 | 892 | 608 | 599 | 587 | 886 | ||||||
| Total | 1966 | 1088 | 1085 | 1053 | 894 | ||||||
| Average | 55% | 55% | 54% | 45% | |||||||
| Statistical analyses for CEREBRA approaches and chance | |||||||||||
| Subjects | Chance | Cerebra #1 | Cerebra #2 | Cerebra #3 | P -value | P -value | P -value | ||||
| Mean | Var | Mean | Var | Mean | Var | Mean | Var | Cerebra #1 | Cerebra #2 | Cerebra #3 | |
| S1 | 894 | 6.01 | 1,082.5 | 149.0 | 1083 | 131.32 | 1,033 | 707.25 | 2.94E-41 | 2.99E-42 | 3.92E-24 |
| S2 | 894 | 7.21 | 1,076.8 | 199.0 | 1073 | 128.31 | 1,035 | 233.91 | 2.15E-38 | 1.80E-41 | 6.10E-33 |
| S3 | 894 | 11.52 | 1,089.4 | 186.6 | 1086 | 166.91 | 1,063 | 224.41 | 8.89E-40 | 2.48E-40 | 5.22E-36 |
| S4 | 894 | 7.21 | 1,086.7 | 39.0 | 1087 | 36.64 | 1,077 | 94.79 | 1.51E-51 | 5.06E-52 | 3.89E-44 |
| S5 | 895 | 12.03 | 1,099.1 | 183.8 | 1097 | 157.71 | 1,048 | 252.79 | 1.19E-40 | 1.12E-41 | 1.83E-33 |
| S6 | 894 | 4.62 | 1,088.0 | 179.5 | 1082 | 161.88 | 1,048 | 205.82 | 2.64E-40 | 1.24E-40 | 1.73E-35 |
| S7 | 895 | 7.21 | 1,097.6 | 64.1 | 1096 | 41.73 | 1,075 | 216.77 | 8.52E-49 | 8.54E-52 | 1.65E-37 |
| S8 | 894 | 2.52 | 1,079.6 | 229.6 | 1077 | 129.91 | 1,039 | 366.06 | 1.09E-37 | 5.10E-42 | 6.10E-30 |
(A) The table shows the average agreement of the 20 repetitions across all 8 fMRI subjects. CEREBRA Approaches 1 and 2 agree with human responses 55%, CEREBRA Approach 3 agrees 54%, when the chance level is 45%. (B) The table shows the means and variances of the CEREBRA change models and the chance model for each subject, and the p-values of the Student t-test, revealing that the differences are highly significant. Comparison agreement with human judgements where at least three of the four subjects agreed.