Effects of prediction error on episodic memory retrieval: evidence from sentence reading and word recognition

Figures & data

Figure 1. Histograms of article and noun cloze probabilities in the experiment.

Table 1. Stimuli characteristics for the three adjectives inserted after the article.

		spill1 (old)	spill2 (but)	spill3 (reliable)
Length	Predictable	5.10 (2.00)	6.20 (2.81)	10.33 (2.21)
	Unpredictable	5.05 (1.97)	6.23 (2.81)	10.18 (2.16)
	Welch’s t-test	t(77.96) = 0.31, p = .76	t(78) = −0.04, p = .97	t(78) = 0.11, p = .91
	(Available obs.)	58/96 words	72/96 words	61/96 words
Frequency	Predictable	4.21	3.24	1.15
	Unpredictable			1.09
	Welch’s t-test	–	–	t(58.9) = 0.36, p = .70

Notes: Frequency norms for all words were obtained from the SUBTLEX-DE data base. “(Available obs.)” indicates the number of word frequencies that could be obtained from the corpus for each one of the three words. For spill 1 and spill 2, the per-condition words for which frequency ratings could be obtained were perfectly matched because they were identical over predictable and unpredictable conditions.

Table 2. Examples of comprehension questions that were used in the self-paced reading experiment.

Target sentence	Comprehension question
Im Opernhaus dirigiert der Maestro mit Leidenschaft die große und imposante Flanke mit den Streichern, sowie den Chor, der ebenfalls sein Bestes gibt. In the opera house, the maestro conducts with passion the large and impressive side with the violins, as well as the choir that’s also doing its best.	Dirigiert der Maestro die Klarinetten? Does the Maestro conduct the clarinets?
Obwohl sie sich ein Taxi zum Bahnhof nimmt, verpasst Anna heute den zu früh abfahrenden Zug nach Frankfurt am Main, und muss am Bahnsteig lange warten. Even though she’s taking a cab to the train station, Anna’s missing the [too early leaving] train to Frankfurt and has to wait at the platform for a long time.	Fährt Anna nach München? Is Anna going to Munic?
Nach dem Einbruch im Nachbarhaus alarmieren die Anwohner die auch nachts bereitstehende Polizei, um den Fall schnell von der Spurensicherung aufklären zu lassen. After the break in at the neighbour’s house, the residents call the [at night available] police so that the case can be closed by the forensics.	Gab es einen Einbruch? Was there a break in?

Figure 2. Accuracy rates during word recognition depending on encoding condition (predictable vs unpredictable). Black triangles indicate average values per condition.

Figure 3. Accuracy rates during word recognition depending on encoding condition (predictable vs unpredictable) and subject-wise recognition scores (reflecting the discrimination between old and new items). Black triangles indicate average values per condition.

Figure 4. Residual (length-corrected) RTs for target words in predictable and unpredictable items, split out by high and low Pr recognition scores (note: all models were run with the scaled continuous variable). Words displayed correspond to the example item, When Paul finally got his driver’s license, he was always driving around with the old but reliable car/group of friends.

Table 3. Effect sizes (b), standard errors (SE), and t-values for models estimating the effects of condition and recognition scores on log-transformed RTs of the article, and the first and second word in the spill-over region (e.g. the_article old_{word 1} but_{word 2} reliable _{word 3} car/group).

	Article				Spill-over 1				Spill-over 2
	b	SE	t	p	b	SE	t	p	b	SE	t	p
Fixed effects
Condition	−0.02	0.01	−1.38		−0.01	0.01	−1.19		−0.01	0.01	−0.96
Pr Score	0.11	0.02	4.59	*	0.13	0.03	4.50	*	0.12	0.03	4.56	*
Condition: Pr score	−0.01	0.01	−1.33		0.005	0.01	0.55		0.004	0.01	0.40
Control predictors
Trial number, Scaled	−0.09	0.004	−21.18	*	−0.11	0.004	−25.27	*	−0.12	0.01	−25.27	*
Word length, scaled	–	–	–	–	0.02	0.01	2.33		0.02	0.01	2.36
Noun plausiblity, scaled	–	–	–	–	–	–	–	–	–	–	–	–
Noun frequency, log	–	–	–	–	–	–	–	–	–	–	–	–
Random effects	Variance	Variance	Variance
Subject	0.04	0.05	0.05
Subject | Condition	0.002	Ø	Ø
Item	0.005	0.002	0.003
Item | Condition	0.001	0.003	0.004
Item | Pr Score	Ø	0.001	0.001
Item | Condition: Pr score	Ø	Ø	Ø

Note: Ø is used for predictors that had to be removed during model fitting because of issues with convergence (Barr et al., 2013). Asterisks indicate statistical significance at the .0125 level.

Table 4. Effect sizes (b), standard errors (SE), and t-values for models estimating the effects of condition and recognition Pr scores on log-transformed RTs of the the third spill-over word, the noun, and the noun spill over region (e.g. … . reliable _{spill-over 3} car/group _noun of _{noun spill-over 1}).

	Spill-over 3				Noun				Noun spill over
	b	SE	t	p	b	SE	t	p	b	SE	t	p
Fixed effects
Condition	0.01	0.01	0.46		0.05	0.01	3.86	*	0.07	0.02	4.21	*
Pr Score	0.11	0.03	4.00	*	0.15	0.03	4.52	*	0.10	0.03	4.00	*
Condition:Pr Score	0.02	0.01	2.62	*	0.05	0.01	4.12	*	0.02	0.01	1.92
Control predictors
Trial number, Scaled	−0.11	0.005	−23.31	*	−0.13	0.01	−22.26	*	−0.11	0.01	−19.82	*
Word length, Scaled	0.02	0.01	2.62		0.02	0.01	2.67	*	0.02	0.01	1.46
Noun Plausibility, Scaled	–	–	–	–	0.01	0.02	0.24		–	–	–	–
Noun Frequency, Log	–	–	–	–	−0.01	0.02	−0.24		–	–	–	–
Random effects	Variance	Variance	Variance
Subject	0.05	0.08	0.04
Subject | Condition	Ø	Ø	Ø
Item	0.003	0.09	0.003
Item | Condition	0.003	Ø	0.008
Item | Pr Score	Ø	Ø	Ø
Item | Condition: Pr Score	Ø	Ø	Ø

Note: Ø is used for predictors that had to be removed during model fitting because of issues with convergence (Barr et al., 2013). Asterisks indicate statistical significance at the 0.0125 level for spill-over 3, and at the 0.025 level for the noun and the spill-over word after the noun.

Figure 5. Correlation plots and Pearson correlation coefficients, illustrating the relationship between residualized reading times during encoding (left panel: average RTs per subject for sentence-early words; right panel: prediction cost at word spill 3 (i.e. reliable) per subject) and recognition memory.

Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3), 255–278. https://doi.org/10.1016/j.jml.2012.11.001

 Web of Science ®Google Scholar