Figure 1
Schematic diagram of OB1-reader. (i) OB1 sees multiple words at a time. Within the visual input, letter processing is modulated by visual acuity (shown through the letter contrast) and visuo-spatial attention (yellow oval). The focus of attention can shift relative to the eye’s fixation. (ii) Open-bigram nodes are activated by the visual input, with stronger activation of letters that suffer less from crowding because they are at a word edge, and that are close to the centers of fixation and attention. (iii) Word nodes are activated by bigram nodes coding for the open bigrams that occur in the word (e.g., ‘reading’ by ‘re’ but also ‘ra’). Word nodes are inhibited by word nodes that share the same bigrams. (iv) Upon fixating a text, OB1 generates a spatiotopic sentence representation that codes the location and approximate length of yet-to-be-identified individual words. Word nodes that reach a recognition threshold are matched to locations (‘blobs’) in the spatiotopic representation based on length. OB1 only recognizes a word when it can be mapped onto a plausible location. When a word is successfully recognized, attention moves ahead of the eyes to the most salient word (defined by its number of letters weighed by their proximity to the centers of fixation and attention). (v) Whether a saccade is initiated is stochastically determined in each 25 ms processing cycle, with successful word recognition increasing the chance of initiation. The center of the most salient word becomes the saccade target. Saliency-based saccade targeting is overruled when a word location to the left of fixation has not yet been marked as recognized. Instead, a regression to that location will be executed. The attentional window is widened after each fixation during which a word is successfully recognized, while it is narrowed after each fixation without successful recognition.
Figure 2
Average response times (top panels) and error rates (bottom panels) in flanker task. Left panels: human performance data from Snell et al., (2019) replotted; right panels: OB1 simulation results. Dots indicate individual participants’ average measure, solid lines indicate the group average. Error bars represent the 95 confidence interval.
Figure 3
ERPs for each condition at electrode sites Fz, Cz and Poz. Figure replotted using the data of Snell et al., (2019).
Figure 4
OB1 lexical activity generated by stimuli from the different conditions.
Figure 5
Correlation between OB1 lexical activity and the N400 from Snell et al. (2019). Data from Snell et al. taken at 400 ms, with stimuli binned on the basis of the generated lexical activity in OB1 for the same stimuli.
Figure 6
Behavioral data from Wen et al. (2019) sentence task and OB1 simulations. A) Human accuracy data from Wen et al., (2019) replotted. B) OB1 simulations: percentage correct and C) time that it took the model to recognize the words (simulated RT). Error bars represent the 95 confidence interval.
Figure 7
ERPs generated by normal and scrambled sentences at electrodes Fz, Cz and Pz. Figure replotted from data of Wen et al., (2019). Light gray bars show the N400 interval used by Wen et al., dark gray bars their N250 interval.
Figure 8
Simulated N400 for normal and scrambled sentences. The simulated N400 was generated by summing activity of all OB! lexical units while presenting the stimuli of Wen et al. (2019).
Figure 9
Correlation between OB1 lexical activity and N400 size in the sentence task. OB1-generated lexical activity was measured at 400 ms, N400 was computed from the EEG data from the Fz, Cz and Pz electrodes of Wen et al. (2019), using their interval. Stimuli were binned on the basis of OB1’s generated lexical activity.