Executive resources shape the impact of language predictability across the adult lifespan
- Department of Psychology, University of Lübeck, Germany
- Center of Brain, Behavior and Metabolism, University of Lübeck, Germany
- Research Group Cognition and Plasticity, Max Planck Institute for Human Cognitive and Brain Sciences, Germany
- Donders Institute for Brain, Cognition and Behaviour, Radboud University, Netherlands
- Wilhelm Wundt Institute for Psychology, Leipzig University, Germany
Figures
Figure 1
Visualisation of hypotheses.
We expected main effects on reading time of (a) cognitive load and (b) surprisal, as well as (c) an interaction of surprisal and cognitive load. Additionally, (d) we explored how these effects are modulated by age.
Figure 2 with 1 supplement
Experimental design and quantification of predictability as word surprisal using a large language model (GPT-2).
(a) Participants were asked to perform a self-paced reading task (Reading Only) which was complemented in some blocks by a secondary n-back task on the font colour of the words (Reading + 1-back, and Reading + 2-back). The order of the blocks was pseudo-randomised, with Reading Only always being the first condition to be presented, followed by the two dual-task conditions, and another main block for each of the three conditions. Both dual-task paradigms (Reading + 1-back and Reading + 2-back) were first introduced in short single-task training sets. (b) We generated one surprisal score for each word in the reading material by using context chunks of two words as prompts for next-word predictions in GPT-2. The resulting probability for the actual next word in the text (here: ‘mail’, marked in teal) was then transformed into a surprisal score, which reflected how predictable the respective word was given the context. Additionally, based on the distribution of probabilities for all possible continuations, we computed an entropy score, which reflects the uncertainty in predicting the next word. Please note that the example sentence used here has been translated to English for better comprehensibility, while the original text materials were in German.
Figure 2—figure supplement 1
Comparison of age distribution between samples.
Figure 3 with 2 supplements
Estimated marginal effects of predictors age, cognitive load, and surprisal on task performance and reading time.
Main effects of cognitive load and age on accuracy in the comprehension question task (a) and on n-back task performance (d-primes; b). Please note that we do not show d-primes for the Reading Only task as there was no n-back task in this condition. Reading time increased with increasing age and word surprisal (c, left: results from linear mixed model, LMM, right: results from generalised additive model, GAM – for an explanation see section Modelling potential non-linear contributions). In (panel d), we show the two-way interaction of cognitive load and surprisal (left) and cognitive load and age (middle). In both cases, effects were strongest in the Reading Only condition (see bar plot insets). Additionally, we show how age modulates the effect of surprisal on reading time (c, right). For raw and predicted individual trajectories, please see Figure 3—figure supplements 1 and 2 in the Supplementary Material. Estimated marginal effects were adjusted for ‘Reading Only’ as the reference level. N = 175.
Figure 3—figure supplement 1
Task performance and reading time by age and cognitive load condition.
Task performance (d-primes) in conditions with an n-back task (a). Accuracy in the comprehension question performance task (b). Reading time by age and condition (c). Reading time by condition (d). Solid line: M, shaded area: 95% CI, point: mean reading time for one participant in the respective condition. N = 175.
Figure 3—figure supplement 2
Individual predicted reading time.
Reading time by age (a) and condition (b). N = 175.
Figure 4 with 3 supplements
Results of the simple slopes analysis and exemplary marginal effects plots for three different ages.
In the Johnson–Neyman plot (Johnson and Neyman, 1936) on the left side of panel (a), we show the effect of surprisal on reading time across the whole age range separated by cognitive load condition: Reading Only (top; blue), 1-back Dual Task (middle; yellow), and 2-back Dual Task (bottom; red). The stronger the surprisal effect for a certain age, the higher the value on the y-axis. Grey areas indicate age ranges for which we did not find an effect of surprisal on reading time in the respective condition, whereas blue areas indicate a significant surprisal effect (see inset on the right for a visualisation of a non-significant effect in a younger participant and a significant effect in an older participant). In panel (b), we show the predicted surprisal effect in each cognitive load for an average young (average age −1 SD), middle-aged (average age) and older participant (average age +1 SD). The bar plots illustrate the predicted effects of surprisal on reading time (Estimates ± 95% CI) across the three cognitive load conditions for those three average participants. N = 175.
Figure 4—figure supplement 1
Comparison of factor smooths for different levels of cognitive load from the three-way interaction of age, surprisal, and cognitive load.
The difference smooths show slightly stronger effects of high surprisal in young than older adults for the 1-back relative to the 2-back condition, and stronger effects of high surprisal in older adults for the Reading only relative to the n-back condition. N = 175.
Figure 4—figure supplement 2
Comparison of the results of the LMM and GAM control analyses (Estimates ± 95% CI).
Panel c illustrates the interaction between cognitive load and surprisal for a representative younger and older participant, estimated using the LMM (left) and the GAM (right). For a complementary visualisation of the three-way interaction between age, cognitive load, and surprisal, see Figure 4. For a visualisation of the main effects of age, surprisal, and cognitive load on reading time, please see Figure 3c. N = 175.
Figure 4—figure supplement 3
Estimates ± 95% CI for the three-way interaction of age, entropy, and cognitive load in the full sample (N = 175).
Figure 5 with 1 supplement
Results of the internal online replication in comparison with the results of the online sample of the original study.
Estimates ±CI for the main effects of age, surprisal, and cognitive load as well as the two-way interaction of surprisal and cognitive load are visualised. RO: Reading only. Full results are provided in Appendix 1—table 3. For a comparison of age distributions in the original online and lab sample and the online replication sample, please see Figure 2—figure supplement 1. Please note that effects are grouped by their magnitude.
Figure 5—figure supplement 1
Estimates ± 95% CI for the three-way interaction of age, surprisal, and cognitive load in the replication sample (N = 96).
Figure 6
Estimates ± 95% CI for the cumulative effect of surprisal on reading time.
To illustrate the cumulative effect of surprisal on reading time over the course of a text, we predicted reading times for an average younger (27 years, M − 1 SD) and average older (63 years, M + 1 SD) participant in the easy Reading Only condition (blue) and the most challenging condition 2-back (Dual Task; red) and computed the cumulative sum for a short example sentence. Panel a illustrates how reading time gradually increases in total over the course of the sentence, with all predictors being held constant at their average, except for the predictors age, cognitive load, and word length. In panel b, we again show cumulative reading times, this time isolating the effect of surprisal. Please note that surprisal values are zero for the first two words, as our GPT-2 model estimates surprisal based on the two preceding words, which are unavailable at the beginning of the sentence. The example sentence used in both panels is the German translation of the opening line of Anna Karenina, ‘Happy families are all alike, every unhappy family is unhappy in its own way’ (Karenina, 1878). N = 175.
Tables
Table 1
Main results for model for reading time (N = 175).
| Predictors | Estimate | Std. error | CI | t | df | p | ||
|---|---|---|---|---|---|---|---|---|
| Main effects | Surprisal | 0.001707 | 0.000151 | 0.001411 to 0.002002 | 11.320677 | 2361.37 | 1.368 × 10–28 | * |
| Age | 0.009113 | 0.000991 | 0.007158 to 0.011068 | 9.199100 | 178.46 | 1.751 × 10–16 | * | |
| Cognitive load [1-back vs. Reading Only] | 0.473800 | 0.013916 | 0.446336 to 0.501264 | 34.046321 | 176.18 | 8.399 × 10–79 | * | |
| Cognitive load [2-back vs. Reading Only] | 0.791540 | 0.026090 | 0.740046 to 0.843034 | 30.338989 | 173.76 | 7.320 × 10–71 | * | |
| Two-way interactions | Surprisal × age | 0.000035 | 0.000004 | 0.000027 to 0.000042 | 9.287151 | 287,771.27 | 3.481 × 10–20 | * |
| Surprisal × cognitive load [1-back vs. Reading Only] | –0.001093 | 0.000161 | –0.001409 to –0.000776 | –6.771521 | 287,959.11 | 2.043 × 10–11 | * | |
| Surprisal × cognitive load [2-back vs. Reading Only] | –0.001255 | 0.000163 | –0.001575 to –0.000935 | –7.681261 | 288,294.96 | 2.709 × 10–14 | * | |
| Age × cognitive load [1-back vs. Reading Only] | –0.002798 | 0.000776 | –0.004330 to –0.001267 | –3.606479 | 171.99 | 5.135 × 10–4 | * | |
| Age × cognitive load [2-back vs. Reading Only] | –0.002458 | 0.001454 | –0.005329 to 0.000412 | –1.690400 | 170.79 | 9.681 × 10–2 | ||
| Three-way interactions | Surprisal × age × cognitive load [1-back vs. Reading Only] | –0.000111 | 0.000009 | –0.000129 to –0.000094 | –12.266076 | 287,807.34 | 3.748 × 10–34 | * |
| Surprisal × age × cognitive load [2-back vs. Reading Only] | –0.000078 | 0.000009 | –0.000096 to –0.000060 | –8.483676 | 287,771.65 | 4.384 × 10–17 | * | |
| Model fit | Intra-class correlation (ICC) | 0.46 | ||||||
| Marginal R2/conditional R2 | 0.643/0.807 | |||||||
-
All continuous predictors were centred. Degrees of freedom for p-values, standard errors and confidence intervals (CIs) were computed using Satterthwaite’s approximation. All p-values reported here are FDR-corrected and were computed using ANOVAs with type III sum of squares. Results that are significant on an alpha-level of 0.05 are marked with a star.
Appendix 1—table 1
Results from models for task performance measures (N = 175).
| LMM for d-primes | GLMM for comprehension question accuracy | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Estimate | Std. error | t | df | p | OR | Std. error | z | p | |||
| Mean d-prime single-tasks | 0.469 | 0.0549 | 8.550 | 166.51 | 2.294 × 10–14 | * | |||||
| Mean comprehension question performance | 0.011 | 0.0036 | 3.053 | 171.653 | 3.943 × 10–3 | * | |||||
| De-meaned comprehension question performance | –0.001 | 0.0012 | –0.475 | 372.688 | 6.350 × 10–1 | ||||||
| Block number | –0.006 | 0.0085 | –0.676 | 349.333 | 5.618 × 10–1 | ||||||
| Recording location [online] | –0.506 | 0.0902 | –5.609 | 163.182 | 1.920 × 10–7 | * | 0.980 | 0.1876 | –0.106 | 9.155x10–1 | |
| Age | –0.005 | 0.0027 | –2.057 | 164.038 | 5.003 × 10–2 | 0.986 | 0.0053 | –2.676 | 1.304x10–2 | * | |
| Cognitive load [1-back vs. Reading Only] | 0.253 | 0.0390 | –8.928 | 1.011x10–18 | * | ||||||
| Cognitive load [2-back vs. Reading Only] | 0.156 | 0.0234 | –12.403 | 1.753x10–34 | * | ||||||
| Cognitive load. [2-back vs. 1-back] | –1.636 | 0.0626 | –26.120 | 173.125 | 2.672 x 10–61 | * | |||||
| Age * cognitive load [1-back vs. Reading Only] | 0.990 | 0.0081 | –1.183 | 3.313x10–1 | |||||||
| Age * cognitive load [2-back vs. Reading Only] | 1.003 | 0.0080 | 0.395 | 8.081x10–1 | |||||||
| Age * cognitive load [2-back vs. 1-back] | –0.014 | 0.0035 | –3.931 | 169.766 | 2.210 × 10–4 | * | |||||
| Model fit | Conditional/marginal R2 | ICC | Conditional/marginal R2 | ICC | |||||||
| 0.822/0.634 | 0.512 | 0.304/0.146 | 0.185 | ||||||||
-
Note. All continuous predictors were centred. Degrees of freedom for p-values, standard errors and confidence intervals (CI) were computed using Satterthwaite’s approximation (LMM for d-primes) and Wald’s approximation (GLMM for comprehension question accuracy). All p-values reported here are FDR-corrected and were computed using ANOVAs with type III sums of squares. Results that are significant on an alpha-level of 0.05 are marked with a star. OR = Odds Ratio.
Appendix 1—table 2
Results from the model for reading times for full original sample (N = 175).
| LMM for full original sample (N = 175) | ||||||||
|---|---|---|---|---|---|---|---|---|
| Predictors | Estimate | Std. error | CI | t | df | p | ||
| Main effects | Reading time of previous trial (log-transformed) | 0.110235 | 0.001409 | 0.107472 to 0.112997 | 78.210829 | 287,711.33 | <1.33 × 10–322 | * |
| d-prime | –0.006293 | 0.001767 | –0.009755 to –0.002831 | –3.562277 | 224,716.70 | 4.903×10–4 | * | |
| Mean d-prime single-tasks | 0.084636 | 0.019585 | 0.045974 to 0.123298 | 4.321418 | 169.86 | 3.717×10–5 | * | |
| Mean comprehension question performance | 0.002841 | 0.001276 | 0.000321 to 0.005360 | 2.225728 | 169.26 | 3.282×10–2 | * | |
| De-meaned comprehension question performance | 0.000053 | 0.000038 | –0.000023 to 0.000128 | 1.374475 | 257,871.34 | 1.693×10–1 | ||
| Word frequency | 0.643570 | 0.362057 | –0.067231 to 1.354371 | 1.777538 | 727.78 | 8.280×10–2 | ||
| Word length | 0.007839 | 0.000407 | 0.007039 to 0.008638 | 19.240951 | 1400.80 | 7.407×10–73 | * | |
| Word entropy | 0.001410 | 0.000785 | –0.000129 to 0.002949 | 1.796533 | 7594.27 | 8.280×10–2 | ||
| n-back reaction [reaction vs. no reaction] | 0.317464 | 0.001799 | 0.313937 to 0.320990 | 176.444743 | 287,862.05 | <1.33 × 10–322 | * | |
| Block number | –0.006495 | 0.000164 | –0.006816 to –0.006173 | –39.637620 | 287,253.02 | <1.33 × 10–322 | * | |
| Trial number | –0.000456 | 0.000007 | –0.000470 to –0.000442 | –64.085235 | 18,673.68 | <1.33 × 10–322 | * | |
| Recording location [online vs. lab] | –0.219339 | 0.032284 | –0.283072 to –0.155606 | –6.793975 | 168.69 | 2.686×10–10 | * | |
| Surprisal | 0.001707 | 0.000151 | 0.001411 to 0.002002 | 11.320677 | 2361.37 | 1.368×10–28 | * | |
| Age | 0.009113 | 0.000991 | 0.007158 to 0.011068 | 9.199100 | 178.46 | 1.751×10–16 | * | |
| Cognitive load [1-back vs. Reading Only] | 0.473800 | 0.013916 | 0.446336 to 0.501264 | 34.046321 | 176.18 | 8.399×10–79 | * | |
| Cognitive load [2-back vs. Reading Only] | 0.791540 | 0.026090 | 0.740046 to 0.843034 | 30.338989 | 173.76 | 7.320×10–71 | * | |
| Two-way interactions | Surprisal x age | 0.000035 | 0.000004 | 0.000027 to 0.000042 | 9.287151 | 287,771.27 | 3.481×10–20 | * |
| Surprisal x cognitive load [1-back vs. Reading Only] | –0.001093 | 0.000161 | –0.001409 to –0.000776 | –6.771521 | 287,959.11 | 2.043×10–11 | * | |
| Surprisal x cognitive load [2-back vs. Reading Only] | –0.001255 | 0.000163 | –0.001575 to –0.000935 | –7.681261 | 288,294.96 | 2.709×10–14 | * | |
| Age x cognitive load [1-back vs. Reading Only] | –0.002798 | 0.000776 | –0.004330 to –0.001267 | –3.606479 | 171.99 | 5.135×10–4 | * | |
| Age x cognitive load [2-back vs. Reading Only] | –0.002458 | 0.001454 | –0.005329 to 0.000412 | –1.690400 | 170.79 | 9.681×10–2 | ||
| Three-way interactions | Surprisal x age x cognitive load [1-back vs. Reading Only] | –0.000111 | 0.000009 | –0.000129 to –0.000094 | –12.266076 | 287,807.34 | 3.748×10–34 | * |
| Surprisal x age x cognitive load [2-back vs. Reading Only] | –0.000078 | 0.000009 | –0.000096 to –0.000060 | –8.483676 | 287,771.65 | 4.384×10–17 | * | |
| Model fit | Intra-class correlation (ICC) | 0.46 | ||||||
| Marginal R2/conditional R2 | 0.643/0.807 | |||||||
-
Note. All continuous predictors were centred. Degrees of freedom for p-values, standard errors and confidence intervals (CI) were computed using Satterthwaite’s approximation. All p-values reported here are FDR-corrected and were computed using ANOVAs with type III sum of squares. Results that are significant on an alpha-level of 0.05 are marked with a star.
Appendix 1—table 3
Results from models for reading times for original online sample and online replication sample (N = 80 and N = 96, respectively).
| LMM for online original sample (N = 80) | LMM for online replication sample (N = 96) | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Predictors | Estimate | Std. error | CI | t | df | p | Estimate | Std. Error | CI | t | df | p | |||
| Main effects | Reading time of previous trial (log-transformed) | 0.076157 | 0.002060 | 0.072120 to 0.080193 | 36.978112 | 133,340.940 | 5.049x10–297 | * | 0.149035 | 0.001818 | 0.145472 to 0.152597 | 81.985 | 161,495.845 | <3.442 × 10–281 | * |
| d-prime | 0.058086 | 0.003309 | 0.051601 to 0.064571 | 17.555499 | 58,135.052 | 2.440x10–68 | * | –0.010767 | 0.002175 | –0.015029 to –0.006504 | –4.950607 | 139,351.725 | 8.888x10–7 | * | |
| Mean d-prime single-tasks | 0.093533 | 0.027543 | 0.038674 to 0.148391 | 3.395825 | 75.930447 | 1.403x10–3 | * | 0.111742 | 0.020097 | 0.071832 to 0.151652 | 5.560231 | 92.614 | 3.324x10–7 | * | |
| Mean comprehension question performance | 0.003366 | 0.001733 | –0.000085 to 0.006816 | 1.942677 | 76.169 | 6.690x10–2 | 0.003459 | 0.001659 | 0.000163 to 0.006754 | 2.084567 | 91.938 | 4.487x10–2 | * | ||
| De-meaned comprehension question performance | –0.000481 | 0.000059 | –0.000597 to –0.000365 | –8.123202 | 118,882.498 | 8.250x10–16 | * | –0.000715 | 0.000047 | –0.000807 to –0.000624 | –15.286171 | 152,148.746 | 2.661x10–52 | * | |
| Word frequency | 0.255248 | 0.300091 | –0.335305 to 0.845800 | 0.850567 | 299.741 | 4.190x10–1 | 0.277510 | 0.253935 | –0.222530 to 0.777550 | 1.092838 | 258.972 | 2.917x10–1 | |||
| Word length | 0.006329 | 0.000414 | 0.005517 to 0.007141 | 15.297105 | 1292.834 | 2.874x10–48 | * | 0.006341 | 0.000362 | 0.005631 to 0.007052 | 17.512071 | 1287.953 | 2.794x10–61 | * | |
| Word entropy | –0.000656 | 0.000933 | –0.002485 to 0.001173 | –0.702998 | 3838.651 | 4.821x10–1 | 0.000337 | 0.000826 | –0.001283 to 0.001957 | 0.407433 | 3601.355 | 6.837x10–1 | |||
| n-back reaction [reaction vs. no reaction] | 0.366351 | 0.002603 | 0.361250 to 0.371452 | 140.757916 | 133,345.032 | <5.049 × 10–297 | * | 0.335449 | 0.002293 | 0.330955 to 0.339943 | 146.303524 | 161,785.798 | <3.442 × 10–281 | * | |
| Block number | –0.008486 | 0.000247 | –0.008970 to –0.008002 | –34.367684 | 131,940.893 | 4.797x10–257 | * | –0.008042 | 0.000224 | –0.008481 to –0.007604 | –35.946825 | 159,855.434 | 3.442x10–281 | * | |
| Trial number | –0.000407 | 0.000009 | –0.000425 to –0.000390 | –45.720722 | 8566.066 | <5.049 × 10–297 | * | –0.000386 | 0.000008 | –0.000402 to –0.000371 | –48.601047 | 8175.168 | <3.442 × 10–281 | * | |
| Surprisal | 0.001145 | 0.000162 | 0.000826 to 0.001463 | 7.046510 | 1889.625 | 4.190x10–12 | * | 0.001375 | 0.000144 | 0.001093 to 0.001656 | 9.578258 | 1886.753 | 5.358x10–21 | * | |
| Age | 0.005382 | 0.001535 | 0.002326 to 0.008438 | 3.507190 | 76.056 | 1.057x10–3 | * | 0.008953 | 0.001318 | 0.006336 to 0.011570 | 6.795181 | 91.979 | 1.460x10–9 | * | |
| Cognitive load [1-back vs. Reading Only] | 0.468989 | 0.018722 | 0.431749 to 0.506229 | 25.050620 | 82.526 | 5.650x10–40 | * | 0.507557 | 0.022113 | 0.463669 to 0.551446 | 22.953170 | 96.841 | 1.343x10–40 | * | |
| Cognitive load [2-back vs. Reading Only] | 0.824086 | 0.034653 | 0.755102 to 0.893071 | 23.781266 | 78.282 | 2.909x10–37 | * | 0.722423 | 0.031793 | 0.659316 to 0.785531 | 22.722673 | 96.183 | 3.806x10–40 | * | |
| Two-way interactions | Surprisal x cognitive load [1-back vs. Reading Only] | 0.000786 | 0.000223 | 0.000349 to 0.001223 | 3.522647 | 133,382.112 | 6.411x10–4 | * | 0.001499 | 0.000203 | 0.001101 to 0.001897 | 7.376731 | 161,262.305 | 2.667x10–13 | * |
| Surprisal x cognitive load [2-back vs. Reading Only] | 0.000375 | 0.000225 | –0.000067 to 0.000816 | 1.661967 | 133,507.349 | 1.086x10–1 | 0.001365 | 0.000203 | 0.000967 to 0.001763 | 6.721136 | 161,923.055 | 2.714x10–11 | * | ||
| Model fit | Intra-class correlation (ICC) | 0.47 | 0.50 | ||||||||||||
| Marginal R2/conditional R2 | 0.587/0.781 | 0.615/0.809 | |||||||||||||
-
Note. All continuous predictors were centred. Degrees of freedom for p-values, standard errors and confidence intervals (CI) were computed using Satterthwaite’s approximation. All p-values reported here are FDR-corrected and were computed using ANOVAs with type III sum of squares. Results that are significant on an alpha-level of 0.05 are marked with a star.
Appendix 1—table 4
Results from models for control analysis (1-back vs. 2-back) of reading times for full original sample (N = 175).
| Control analysis 2-back vs. 1-back: LMM for full original sample (N = 175) | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Predictors | Estimate | Std. error | CI | t | df | p | |||
| Main effects | Reading time of previous trial (log-transformed) | 0.062181 | 0.001755 | 0.058740 to 0.065622 | 35.420630 | 188,475.92 | 3.296×10–273 | * | |
| d-prime | –0.006667 | 0.001898 | –0.010388 to –0.002947 | –3.512595 | 168,020.99 | 7.398×10–4 | * | ||
| Mean d-prime single-tasks | 0.094749 | 0.020772 | 0.053746 to 0.135752 | 4.561318 | 171.05 | 1.756×10–5 | * | ||
| Mean comprehension question performance | 0.003146 | 0.001352 | 0.000478 to 0.005815 | 2.327238 | 170.38 | 3.018×10–2 | * | ||
| De-meaned comprehension question performance | 0.000038 | 0.000047 | –0.000053 to 0.00013 | 0.821932 | 150,546.36 | 4.837×10–1 | |||
| Word frequency | –0.128276 | 0.319365 | –0.756126 to 0.499574 | –0.401659 | 398.70 | 6.882×10–1 | |||
| Word length | 0.005220 | 0.000414 | 0.004408 to 0.006031 | 12.616091 | 1351.55 | 4.020×10–34 | * | ||
| Word entropy | 0.001711 | 0.000914 | –0.000081 to 0.003502 | 1.871897 | 4521.92 | 8.171×10–2 | |||
| n-back reaction [reaction vs. no reaction] | 0.316007 | 0.001917 | 0.312250 to 0.319764 | 164.836212 | 188,099.24 | <1.33 × 10–322 | * | ||
| Block number | –0.011363 | 0.000295 | –0.011941 to –0.01079 | –38.535522 | 185,049.49 | 1.33×10–322 | * | ||
| Trial number | –0.000450 | 0.000009 | –0.000467 to –0.000433 | –52.053811 | 10,059.10 | <1.33 × 10–322 | * | ||
| Recording location [online vs. lab] | –0.226062 | 0.034131 | –0.293438 to –0.158686 | –6.623323 | 169.71 | 8.894×10–10 | * | ||
| Surprisal | 0.001848 | 0.000161 | 0.001532 to 0.002164 | 11.467327 | 2010.37 | 3.901×10–29 | * | ||
| Age | 0.008762 | 0.001101 | 0.006591 to 0.010934 | 7.961804 | 184.31 | 3.751×10–13 | * | ||
| Cognitive load [2-back vs. 1-back] | 0.338896 | 0.020892 | 0.297669 to 0.380123 | 16.221045 | 178.98 | 1.838×10–36 | * | ||
| Two-way interactions | Surprisal x age | 0.000003 | 0.000005 | –0.000007 to 0.000012 | 0.545912 | 187,940.82 | 6.159×10–1 | ||
| Surprisal x cognitive load [2-back vs. 1-back] | –0.000148 | 0.000173 | –0.000486 to 0.000191 | –0.855146 | 187,910.28 | 4.837×10–1 | |||
| Age x cognitive load [2-back vs. 1-back] | 0.000689 | 0.001156 | –0.001593 to 0.00297 | 0.595977 | 172.03 | 6.133×10–1 | |||
| Three-way interaction | Surprisal x age x cognitive load [2-back vs. 1-back] | 0.000033 | 0.000010 | 0.000014 to 0.000052 | 3.372931 | 188,203.53 | 1.144×10–3 | * | |
| Model fit | Intra-class correlation (ICC) | 0.44 | |||||||
| Marginal R2/conditional R2 | 0.442/0.690 | ||||||||
-
Note. p-values were computed using Wald's approximation as implemented in the package mgcv. Results that are significant on an alpha-level of 0.05 are marked with a star. Edf: Effective degrees of freedom.
Appendix 1—table 5
Results from GAM for control analysis of reading times for full original sample (N = 175).
| Control analysis: GAM for full original sample (N = 175) | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Predictors | Estimate | Std. Error | t | F | EDF | p | |||
| Main effects | Reading time of previous trial (log-transformed) | 204.591 | 33.664 | <2 × 10–16 | * | ||||
| d-prime | 38.774 | 28.042 | <2 × 10–16 | * | |||||
| Mean d-prime single-tasks | 24.344 | 1.892 | <2 × 10–16 | * | |||||
| Mean comprehension question performance | 5.348 | 2.305 | 3.23×10–3 | * | |||||
| De-meaned comprehension question performance | 39.408 | 7.477 | <2 × 10–16 | * | |||||
| Word frequency | 6.837 | 7.571 | <2 × 10–16 | * | |||||
| Word length | 73.076 | 4.038 | <2 × 10–16 | * | |||||
| Word entropy | 4.027 | 3.704 | 2.44×10–3 | * | |||||
| Surprisal | 9.547 | 4.107 | <2 × 10–16 | * | |||||
| Age | 51.783 | 3.028 | <2 × 10–16 | * | |||||
| n-back reaction [reaction vs. no reaction] | 0.3167 | 0.00179 | 177.06 | <2 × 10–16 | * | ||||
| Block number | –0.0061 | 0.00017 | –36.89 | <2 × 10–16 | * | ||||
| Trial number | –0.0004 | 0.00001 | –64.47 | <2 × 10–16 | * | ||||
| Recording location (online vs. lab) | –0.2514 | 0.02602 | –9.66 | <2 × 10–16 | * | ||||
| Cognitive load [1-back vs. Reading Only] | 0.4318 | 0.02514 | 17.17 | <2 × 10–16 | * | ||||
| Cognitive load [2-back vs. Reading Only] | 0.7819 | 0.02526 | 30.95 | <2 × 10–16 | * | ||||
| Two-way interactions | Surprisal x cognitive load | 13.962 | 13.849 | <2 × 10–16 | * | ||||
| Three-way interactions | Srprisal x age x cognitive load [Reading Only] | 23.946 | 10.248 | <2 × 10–16 | * | ||||
| Surprisal x age x cognitive load [1-back] | 2.874 | 2.017 | 3.616×10–2 | * | |||||
| Surprisal x age x cognitive load [2-back] | 2.392 | 4.877 | 2.375×10–2 | * | |||||
| Random effects | Cognitive load | ID | 255.250 | 508.610 | <2 × 10–16 | * | ||||
| Text Nr. | 14,053.220 | 7.840 | <2 × 10–16 | * | |||||
| Word | 1.870 | 804.600 | <2 × 10–16 | * | |||||
| Colour | 12.530 | 2.770 | <2 × 10–16 | * | |||||
| Model fit | R2 | 815 | |||||||
-
Note. All continuous predictors were centred. Degrees of freedom for p-values, standard errors and confidence intervals (CI) were computed using Satterthwaite’s approximation. All p-values reported here are FDR-corrected and were computed using ANOVAs with type III sum of squares. Results that are significant on an alpha-level of 0.05 are marked with a star.
Appendix 1—table 6
Results from the model for reading times for full original sample (N = 175) for the effects of entropy, cognitive load, and age on reading time.
| LMM for full original sample (N = 175) | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Predictors | Estimate | Std. error | CI | t | df | p | |||
| Main effects | Reading time of previous trial (log-transformed) | 0.110066 | 0.001410 | 0.107302 to 0.112830 | 78.048894 | 287,686.846 | <8.542 × 10–79 | * | |
| d-prime | –0.006248 | 0.001767 | –0.009712 to –0.002784 | –3.535537 | 224,708.066 | 6.105x10–4 | * | ||
| Mean d-prime single-tasks | 0.084628 | 0.019587 | 0.045963 to 0.123292 | 4.320683 | 169.899 | 4.225x10–5 | * | ||
| Mean comprehension question performance | 0.002842 | 0.001276 | 0.000322 to 0.005362 | 2.226655 | 169.295 | 3.275x10–2 | * | ||
| De-meaned comprehension question performance | 0.000052 | 0.000039 | –0.000023 to 0.000127 | 1.350491 | 257,990.129 | 1.769x10–1 | |||
| Word frequency | 0.608299 | 0.360822 | –0.100080 to 1.316678 | 1.685873 | 725.615 | 9.929x10–2 | |||
| Word length | 0.007812 | 0.000406 | 0.007015 to 0.008609 | 19.220944 | 1402.270 | 9.864x10–73 | * | ||
| Surprisal | 0.001682 | 0.000150 | 0.001387 to 0.001977 | 11.177024 | 2359.191 | 7.143x10–28 | * | ||
| n-back reaction [reaction vs. no reaction] | 0.317360 | 0.001800 | 0.313832 to 0.320888 | 176.311387 | 287,866.290 | <8.542 × 10–79 | * | ||
| Block number | –0.006481 | 0.000164 | –0.006803 to –0.006160 | –39.539875 | 287,255.082 | <8.542 × 10–79 | * | ||
| Trial number | –0.000456 | 0.000007 | –0.000470 to –0.000442 | –64.065173 | 18,557.632 | <8.542 × 10–79 | * | ||
| Recording location [online vs. lab] | –0.219370 | 0.032287 | –0.283107 to –0.155632 | –6.794462 | 168.720 | 3.895x10–10 | * | ||
| Entropy | 0.001412 | 0.000785 | –0.000126 to 0.002950 | 1.800222 | 7570.659 | 8.213x10–2 | |||
| Age | 0.009110 | 0.000991 | 0.007155 to 0.011065 | 9.195551 | 178.495 | 2.325x10–16 | * | ||
| Cognitive load [1-back vs. Reading Only] | 0.473980 | 0.013924 | 0.446501 to 0.501458 | 34.041427 | 176.188 | 8.542x10–79 | * | ||
| Cognitive load [2-back vs. Reading Only] | 0.791850 | 0.026098 | 0.740340 to 0.843361 | 30.341059 | 173.750 | 7.294x10–71 | * | ||
| Two-way interactions | Entropy x age | 0.000090 | 0.000030 | 0.000032 to 0.000148 | 3.030333 | 287,391.572 | 3.257x10–3 | * | |
| Entropy x cognitive load [1-back vs. Reading Only] | 0.006638 | 0.001273 | 0.004142 to 0.009133 | 5.213908 | 287,500.035 | 3.416x10–7 | * | ||
| Entropy x cognitive load [2-back vs. Reading Only] | 0.006490 | 0.001291 | 0.003959 to 0.009021 | 5.025891 | 287,757.942 | 8.595x10–7 | * | ||
| Age x cognitive load [1-back vs. Reading Only] | –0.002785 | 0.000776 | –0.004317 to –0.001253 | –3.587447 | 171.994 | 6.142x10–4 | * | ||
| Age x cognitive load [2-back vs. Reading Only] | –0.002441 | 0.001455 | –0.005313 to 0.000430 | –1.678106 | 170.772 | 9.929x10–2 | |||
| Three-way interactions | Entropy x age x cognitive load [1-back vs. Reading Only] | –0.000399 | 0.000072 | –0.000540 to –0.000258 | –5.546582 | 287,440.357 | 5.831x10–8 | * | |
| Entropy x age x cognitive load [2-back vs. Reading Only] | –0.000188 | 0.000073 | –0.000331 to –0.000045 | –2.577310 | 287,488.317 | 1.258x10–2 | * | ||
| Model fit | Intra-class correlation (ICC) | 0.46 | |||||||
| Marginal R2/conditional R2 | 0.643/0.807 | ||||||||
-
Note. All continuous predictors were centred. Degrees of freedom for p-values, standard errors and confidence intervals (CI) were computed using Satterthwaite’s approximation. All p-values reported here are FDR-corrected and were computed using ANOVAs with type III sum of squares. Results that are significant on an alpha-level of 0.05 are marked with a star.
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Executive resources shape the impact of language predictability across the adult lifespan
eLife 14:RP108176.
https://doi.org/10.7554/eLife.108176.3
