Multi-LEX: A database of multi-word frequencies for French and English.

Written word frequency is a key variable used in many psycholinguistic studies and is central in explaining visual word recognition. Indeed, methodological advances on single-word frequency estimates have helped to uncover novel language-related cognitive processes, fostering new ideas and studies. In an attempt to support and promote research on a related emerging topic, visual multi-word recognition, we extracted from the exhaustive Google Ngram datasets a selection of millions of multi-word sequences and computed their associated frequency estimate. Such sequences are presented with part-of-speech information for each individual word. An online behavioral investigation making use of the French 4-gram lexicon in a grammatical decision task was carried out. The results show an item-level frequency effect of word sequences. Moreover, the proposed datasets were found useful during the stimulus selection phase, allowing more precise control of the multi-word characteristics.

MEGALEX: A megastudy of visual and auditory word recognition.

Using the megastudy approach, we report a new database (MEGALEX) of visual and auditory lexical decision times and accuracy rates for tens of thousands of words. We collected visual lexical decision data for 28,466 French words and the same number of pseudowords, and auditory lexical decision data for 17,876 French words and the same number of pseudowords (synthesized tokens were used for the auditory modality). This constitutes the first large-scale database for auditory lexical decision, and the first database to enable a direct comparison of word recognition in different modalities. Different regression analyses were conducted to illustrate potential ways to exploit this megastudy database. First, we compared the proportions of variance accounted for by five word frequency measures. Second, we conducted item-level regression analyses to examine the relative importance of the lexical variables influencing performance in the different modalities (visual and auditory). Finally, we compared the similarities and differences between the two modalities. All data are freely available on our website ( https://sedufau.shinyapps.io/megalex/ ) and are searchable at www.lexique.org , inside the Open Lexique search engine.

A Thousand Words Are Worth a Picture: Snapshots of Printed-Word Processing in an Event-Related Potential Megastudy.

In the experiment reported here, approximately 1,000 words were presented to 75 participants in a go/no-go lexical decision task while event-related potentials (ERPs) were recorded. Partial correlations were computed for variables selected to reflect orthographic, lexical, and semantic processing, as well as for a novel measure of the visual complexity of written words. Correlations were based on the item-level ERPs at each electrode site and time slice while a false-discovery-rate correction was applied. Early effects of visual complexity were seen around 50 ms after word onset, followed by the earliest sustained orthographic effects around 100 to 150 ms, with the bulk of orthographic and lexical influences arising after 200 ms. Effects of a semantic variable (concreteness) emerged later, at around 300 ms. The overall time course of these ERP effects is in line with hierarchical, cascaded, interactive accounts of word recognition, in which fast feed-forward influences are consolidated by top-down feedback via recurrent processing loops.

Sequential modulations of poorer-strategy effects during strategy execution: an event-related potential study in arithmetic.

When participants accomplish cognitive tasks, they obtain poorer performance if asked to execute a poorer strategy than a better strategy on a given problem. These poorer-strategy effects are smaller following execution of a poorer strategy relative to following a better strategy. To investigate ERP correlates of sequential modulations of poorer-strategy effects, we asked participants (n=20) to accomplish a computational estimation task (i.e., provide approximate products to two-digit multiplication problems like 38×74). For each problem, they were cued to execute a better versus a poorer strategy. We found event-related potentials signatures of sequential modulations of poorer-strategy effects in two crucial windows (i.e., between 200 and 550 ms and between 850 and 1250 ms) associated with executive control mechanisms and allowing conflict monitoring between the better and the cued strategy. These results have important implications on theories of strategies as they suggest that sequential modulations of poorer-strategy effects involve earlier as well as later mechanisms of cognitive control during strategy execution.

Sentence superiority in the reading brain.

When a sequence of written words is briefly presented and participants are asked to identify just one word at a post-cued location, then word identification accuracy is higher when the word is presented in a grammatically correct sequence compared with an ungrammatical sequence. This sentence superiority effect has been reported in several behavioral studies and two EEG investigations. Taken together, the results of these studies support the hypothesis that the sentence superiority effect is primarily driven by rapid access to a sentence-level representation via partial word identification processes that operate in parallel over several words. Here we used MEG to examine the neural structures involved in this early stage of written sentence processing, and to further specify the timing of the different processes involved. Source activities over time showed grammatical vs. ungrammatical differences first in the left inferior frontal gyrus (IFG: 321-406 ms), then the left anterior temporal lobe (ATL: 466-531 ms), and finally in both left IFG (549-602 ms) and left posterior superior temporal gyrus (pSTG: 553-622 ms). We interpret the early IFG activity as reflecting the rapid bottom-up activation of sentence-level representations, including syntax, enabled by partly parallel word processing. Subsequent activity in ATL and pSTG is thought to reflect the constraints imposed by such sentence-level representations on on-going word-based semantic activation (ATL), and the subsequent development of a more detailed sentence-level representation (pSTG). These results provide further support for a cascaded interactive-activation account of sentence reading.

Face swapping in seizure videos for patient deidentification.

OBJECTIVE: This study aimed to test different AI-based face-swapping models applied to videos of epileptic seizures, with the goal of protecting patient privacy while retaining clinically useful seizure semiology. We hypothesized that specific models would show differences in semiologic fidelity compared to the original clinical videos. METHODS: Three open-source models, SimSwap, MobileFaceSwap and GHOST were adopted for face-swapping. For every model, an AI generated male and female image were used to replace the original faces. One representative seizure per patient from three patients with epilepsy was chosen (3 seizure videos x 3 AI models x 2 M/F swap) and remade to 18 transformed video clips. To evaluate the performance of the three models, we used both objective (AI-based) and subjective (expert clinician) evaluation. The objective assessment included four metrics for facial appearance and four metrics for facial expression changes. Four experienced epileptologists reviewed the clips and scoring according to deidentification and preservation of semiology. Kruskal-Wallis H test was used for statistical analysis among the models. RESULTS: In the reproduced videos, the swapped face cannot be recognized as the original face, with no significant difference in scores of deidentification either by objective or subjective assessment. Regarding semiology preservation, no significant differences between models were observed in the objective evaluations. The subjective evaluations revealed that the GHOST model outperformed the other two models (p=0.028). CONCLUSION: This is the first study evaluating AI face swapping models in epileptic seizure video clips. Optimization of AI face-swapping models could enhance the accessibility of seizure videos for education and research while protecting patient privacy and maintaining semiology.

The time course of strategy sequential difficulty effects: an ERP study in arithmetic.

Uittenhove and Lemaire (Exp Psychol 59(5):295-301, 2012) found that we are slower when executing a strategy following a difficult strategy than when executing the same strategy following an easier strategy (i.e., strategy sequential difficulty effects). Uittenhove and Lemaire suggested that difficult strategies temporarily reduce available executive capacities, interfering with the next strategy execution. In this study, we used ERP to determine the time course of these effects. In a computational estimation task, we found greater cerebral activities during strategy execution following a more difficult compared to an easier strategy. Interestingly, greater cerebral activities were most apparent immediately after the encoding of the problem and not during encoding or in later stages of processing. This suggests that strategy sequential difficulty effects interfere most with the retrieval of procedures in contrast to execution of these procedures. We discuss implications of these findings for further understanding of execution of cognitive strategies.

BDNF as potential biomarker of epilepsy severity and psychiatric comorbidity: pitfalls in the clinical population.

BACKGROUND: Several studies implicate brain-derived neurotrophic factor (BDNF) in the pathophysiology of epilepsy. In particular, preclinical data suggest that lower serum BDNF is a biomarker of epilepsy severity and psychiatric comorbidities. We tested this prediction in clinical epilepsy cohorts. METHODS: Patients with epilepsy were recruited from 4 epilepsy centers in France and serum BDNF was quantified. Clinical characteristics including epilepsy duration, classification, localization, etiology, seizure frequency and drug resistance were documented. Presence of individual anti-seizure medications (ASM) was noted. Screening for depression and anxiety symptoms was carried out in all patients using the NDDI-E and the GAD-7 scales. In patients with positive screening for anxiety and/or depression, detailed psychiatric testing was performed including the Mini International Neuropsychiatric Interview (MINI), STAI-Y, Holmes Rahe Stressful Events Scale and Beck Depression Interview. Descriptive analysis was applied. Spearman's test and Pearson's co-efficient were used to assess the association between BDNF level and continuous variables. For discrete variables, comparison of means (Student's t-test, Mann-Whitney u-test) was used to compare mean BDNF serum level between groups. Multivariate analysis was performed using a regression model. RESULTS: No significant correlation was found between serum BDNF level and clinical features of epilepsy or measures of depression. The main group-level finding was that presence of any ASM at was associated with increased BDNF; this effect was particularly significant for valproate and perampanel. CONCLUSION: Presence of ASM affects serum BDNF levels in patients with epilepsy. Future studies exploring BDNF as a possible biomarker of epilepsy severity and/or psychiatric comorbidity must control for ASM effects.

The time course of visual letter perception.

We describe a novel method for tracking the time course of visual identification processes, here applied to the specific case of letter perception. We combine a new behavioral measure of letter identification times with single-letter ERP recordings. Letter identification processes are considered to take place in those time windows in which the behavioral measure and ERPs are correlated. A first significant correlation was found at occipital electrode sites around 100 msec poststimulus onset that most likely reflects the contribution of low-level feature processing to letter identification. It was followed by a significant correlation at fronto-central sites around 170 msec, which we take to reflect letter-specific identification processes, including retrieval of a phonological code corresponding to the letter name. Finally, significant correlations were obtained around 220 msec at occipital electrode sites that may well be due to the kind of recurrent processing that has been revealed recently by TMS studies. Overall, these results suggest that visual identification processes are likely to be composed of a first (and probably preconscious) burst of visual information processing followed by a second reentrant processing on visual areas that could be critical for the conscious identification of the visual target.

On the relations between letter, word, and sentence-level processing during reading.

Much prior research on reading has focused on a specific level of processing, with this often being letters, words, or sentences. Here, for the first time in adult readers, we provide a combined investigation of these three key component processes of reading comprehension. We did so by testing the same group of participants in three tasks thought to reflect processing at each of these levels: alphabetic decision, lexical decision, and grammatical decision. Participants also performed a non-reading classification task, with an aim to partial-out common binary decision processes from the correlations across the three main tasks. We examined the pairwise partial correlations for response times (RTs) in the three reading tasks. The results revealed strong significant correlations across adjacent levels of processing (i.e., letter-word; word-sentence) and a non-significant correlation between non-adjacent levels (letter-sentence). The results provide an important new benchmark for evaluating computational models that describe how letters, words, and sentences contribute to reading comprehension.

General-purpose monitoring during speech production.

The concept of "monitoring" refers to our ability to control our actions on-line. Monitoring involved in speech production is often described in psycholinguistic models as an inherent part of the language system. We probed the specificity of speech monitoring in two psycholinguistic experiments where electroencephalographic activities were recorded. Our focus was on a component previously reported in nonlinguistic manual tasks and interpreted as a marker of monitoring processes. The error negativity (Ne, or error-related negativity), thought to originate in medial frontal areas, peaks shortly after erroneous responses. A component of seemingly comparable properties has been reported, after errors, in tasks requiring access to linguistic knowledge (e.g., speech production), compatible with a generic error-detection process. However, in contrast to its original name, advanced processing methods later revealed that this component is also present after correct responses in visuomotor tasks. Here, we reported the observation of the same negativity after correct responses across output modalities (manual and vocal responses). This indicates that, in language production too, the Ne reflects on-line response monitoring rather than error detection specifically. Furthermore, the temporal properties of the Ne suggest that this monitoring mechanism is engaged before any auditory feedback. The convergence of our findings with those obtained with nonlinguistic tasks suggests that at least part of the monitoring involved in speech production is subtended by a general-purpose mechanism.

Letter and word identification in the fovea and parafovea.

We investigated the extent to which accuracy in word identification in foveal and parafoveal vision is determined by variations in the visibility of the component letters of words. To do so we measured word identification accuracy in displays of three three-letter words, one on fixation and the others to the left and right of the central word. We also measured accuracy in identifying the component letters of these words when presented at the same location in a context of three three-letter nonword sequences. In the word identification block, accuracy was highest for central targets and significantly greater for words to the right compared with words to the left. In the letter identification block, we found an extended W-shaped function across all nine letters, with greatest accuracy for the three central letters and for the first and last letter in the complete sequence. Further analyses revealed significant correlations between average letter identification per nonword position and word identification at the corresponding position. We conclude that letters are processed in parallel across a sequence of three three-letter words, hence enabling parallel word identification when letter identification accuracy is high enough.

An ERP investigation of transposed-word effects in same-different matching.

Can several words be read in parallel, and if so, how is information about word order encoded under such circumstances? Here we focused on the bottom-up mechanisms involved in word-order encoding under the hypothesis of parallel word processing. We recorded EEG while participants performed a visual same-different matching task with sequences of five words (reference sequence followed by a target sequence each presented for 400 ms). The reference sequence could be grammatically correct or an ungrammatical scrambling of the same words (e.g., he wants these green apples/green wants these he apples). Target sequences for 'different' responses were created by either transposing two words in the reference (e.g., he these wants green apples/green these wants he apples), or by changing two words (e.g., he talks their green apples/green talks their he apples). Different responses were harder to make in the transposition condition, and this transposed-word effect started to emerge around 250 ms post-target onset. The transposed-word effect was first seen on an early onsetting N400 component, with reduced amplitudes (i.e., less negative ERPs) in the transposed condition relative to a two-word replacement condition. A later transposed-word effect was seen on a more temporally widespread positive-going component. Converging behavioral and EEG results showed no effects of reference grammaticality on 'different' responses nor an interaction with transposed-word effects. Our results point to the noisy, bottom-up association of word identities to spatiotopic locations as one means of encoding word order information, and one key source of transposed-word effects.

Rapid processing of letters, digits and symbols: what purely visual-attentional deficit in developmental dyslexia?

Visual-attentional theories of dyslexia predict deficits for dyslexic children not only for the perception of letter strings but also for non-alphanumeric symbol strings. This prediction was tested in a two-alternative forced-choice paradigm with letters, digits, and symbols. Children with dyslexia showed significant deficits for letter and digit strings but not for symbol strings. This finding is difficult to explain for visual-attentional theories of dyslexia which postulate identical deficits for letters, digits and symbols. Moreover, dyslexics showed normal W-shaped serial position functions for letter and digit strings, which suggests that their deficit is not due to an abnormally small attentional window. Finally, the size of the deficit was identical for letters and digits, which suggests that poor letter perception is not just a consequence of the lack of reading. Together then, our results show that symbols that map onto phonological codes are impaired (i.e. letters and digits), whereas symbols that do not map onto phonological codes are not impaired. This dissociation suggests that impaired symbol-sound mapping rather than impaired visual-attentional processing is the key to understanding dyslexia.

Parafoveal-on-foveal repetition effects in sentence reading: A co-registered eye-tracking and electroencephalogram study.

When reading, can the next word in the sentence (word n + 1) influence how you read the word you are currently looking at (word n)? Serial models of sentence reading state that this generally should not be the case, whereas parallel models predict that this should be the case. Here we focus on perhaps the simplest and the strongest Parafoveal-on-Foveal (PoF) manipulation: word n + 1 is either the same as word n or a different word. Participants read sentences for comprehension and when their eyes left word n, the repeated or unrelated word at position n + 1 was swapped for a word that provided a syntactically correct continuation of the sentence. We recorded electroencephalogram and eye-movements, and time-locked the analysis of fixation-related potentials (FRPs) to fixation of word n. We found robust PoF repetition effects on gaze durations on word n, and also on the initial landing position on word n. Most important is that we also observed significant effects in FRPs, reaching significance at 260 ms post-fixation of word n. Repetition of the target word n at position n + 1 caused a widely distributed reduced negativity in the FRPs. Given the timing of this effect, we argue that it is driven by orthographic processing of word n + 1, while readers were still looking at word n, plus the spatial integration of orthographic information extracted from these two words in parallel.

Letter perception: from pixels to pandemonium.

In 1959, Oliver Selfridge proposed a model of letter perception, the Pandemonium model, in which the central hypothesis was that letters are identified via their component features. Although a consensus developed around this general approach over the years, key evidence in its favor remained lacking. Recent research has started to provide important evidence in favor of feature-based letter perception, describing the nature of the features, and the time-course of processes involved in mapping features onto abstract letter identities. There is now hope that future 'pandemonium-like' models will be able to account for the rich empirical database on letter identification that has accumulated over the past 50 years, hence solving one key component of the reading process.

Testing computational models of letter perception with item-level event-related potentials.

In the present study, online measures of letter identification were used to test computational models of letter perception. Event-related potentials (ERPs) were recorded to letters and pseudoletters revealing a transition from feature analysis to letter identification in the 100-200-ms time window. Measures indexing this transition were then computed at the level of individual letters. Simulations with several versions of an interactive-activation model of letter perception were fitted with these item-level ERP measures. The results are in favour of a model of letter perception with feedforward excitatory connections from the feature to the letter levels, lateral inhibition at the letter level, and excitatory feedback from the letter to the feature levels.

Older and younger adults' strategies in approximate quantification.

In two experiments, participants were asked to provide a quick and rough estimate of the number of items in collections of 4-79 items. In Experiment 1 verbal strategy reports and performance on each item were collected, and in Experiment 2 performance and eye movements were collected, while young and older participants were tested in strategy-instructed conditions. Results showed that: (a) participants used six different estimation strategies, (b) overall, young and older participants used the same set of strategies, but varied in how often they used each strategy, (c) older adults' strategy repertoire was smaller than young adults' (i.e., inter-individual differences in strategy repertoire), (d) strategy use, participants' performance, and eye movements varied as a function of numerosities and configurations of items, (e) in both the age groups, each strategy was associated with distinctive performance measures and eye movement patterns. These findings show that different processes are available for approximate quantification in both young and older adults and that aging is associated with strategic variations.