What can we learn about integration of novel words into semantic memory from automatic semantic priming?

ABSTRACT According to the Complementary Learning Systems model of word learning, only integrated novel words can interact with familiar words during lexical selection. The pre-registered study reported here is the first to examine behavioural and electrophysiological markers of integration in a task that relies primarily on automatic semantic processing. 71 young adults learned novel names for two sets of novel concepts, one set on each of two consecutive days. On Day 2, learning was followed by a continuous primed lexical decision task with EEG recording. In the N400 window, novel names trained immediately before testing differed from both familiar and untrained novel words, and, in the time window between 500–800 ms post onset, they also differed from novel names that had undergone a 24-hour consolidation, for which a small behavioural priming effect was observed. We develop an account that attributes the observed effects to processes rooted in episodic, rather than semantic, memory.


Introduction
Learning of words is a frequent experience in our lives.While the rate of words learned may be at its highest during the school years (e.g., Anglin, 2000), vocabulary development continues throughout adulthood (Brysbaert et al., 2016;Keuleers et al., 2015), with most adults also learning new words in languages other than their first language.Yet, despite the early onset and the ubiquity of word learning, its multifaceted nature makes it no trivial task.In addition to learning a new phonological form, the concept it refers to, and the association between them, word acquisition entails integrating new words in a network of thousands of already familiar words and connections (e.g., McMurray et al., 2016).In this study, we focused on some questions pertaining to the timescale of this integration process, and we also sought to broaden understanding of behavioural and electrophysiological markers of integration when learning novel names for novel concepts.Davis and Gaskell (2009) proposed an influential account of the processes by which new words are integrated into the semantic network and achieve the same status as already familiar words.Their account is based on the Complementary Learning Systems (CLS) connectionist model of learning (Kumaran et al., 2016;McClelland, 2013;McClelland et al., 2020McClelland et al., , 1995;;O'Reilly et al., 2014).This account suggests that word learning occurs over two stages.The first stage relies on the episodic memory system that is thought to reside in the medial temporal lobe, specifically, the hippocampus.In contrast, the second stage is dependent on the semantic memory system subserved by the neocortex (e.g., Squire, 2004, but see Duff et al., 2020, for a review on the hippocampal contribution to maintenance and retrieval of semantic memories).Whereas episodic memory refers to the capacity to learn, store, and evoke detailed information about specific events together with their spatiotemporal context (Dickerson & Eichenbaum, 2010), semantic memory is viewed as a store of our knowledge about the world and the meanings of verbal symbols (i.e., words) that we employ to make use of this knowledge (see Kumar, 2021, for a review).
The CLS model is rooted in the notion that new memories usually start as episodic, and later become abstracted away from the spatiotemporal context where they were acquired, and integrated into the existing knowledge networks (e.g., Meeter & Murre, 2004).The CLS model of word learning also postulates that the lexical status of the new words (i.e., whether they are treated as familiar words or as nonwords) differs from that of already familiar words until the newly formed episodic memory traces have been consolidated, i.e., transformed into long-term memory (Davis & Gaskell, 2009).Consolidation is considered a pre-condition for integration of new words into the lexicon, and it is believed to slow down the rate of forgetting, reinforce resistance to interference from other stimuli (e.g., Dudai & Eisenberg, 2004;Wixted, 2004), facilitate access to the meanings of the new words, and enhance the extent to which their lexical representations impact lexical processing of familiar words (Davis & Gaskell, 2009;Lindsay & Gaskell, 2010, and see Leach & Samuel, 2007, for a similar two-stage model of word learning).
The CLS model of word learning was originally developed to account for learning of (meaningless) spoken word forms; however, since Davis and Gaskell's seminal paper, the predictions of this model have been tested with different types of stimuli and in different modalities, often with little regard to how the predictions of the model might change depending on what is being learned and how.Thus, an extensive body of literature has examined learning of (meaningless) spoken novel word forms (e.g., Bakker et al., 2014;Davis et al., 2009;Dumay & Gaskell, 2007, 2012;Dumay et al., 2004;Gaskell & Dumay, 2003;Kapnoula & McMurray, 2016;Kapnoula et al., 2015;Lindsay & Gaskell, 2013;Tamminen & Gaskell, 2008;Tamminen et al., 2010), while other studies have focused on learning of novel words in the written modality (e.g., Bakker et al., 2014;Bowers et al., 2005;Walker et al., 2010).Likewise, there is a great deal of research on learning of novel names for familiar concepts 1 (e.g., Breitenstein et al., 2007;Dobel et al., 2009;Elgort, 2011;Elgort & Warren, 2014;Geukes et al., 2015;Korochkina et al., 2021;McGregor, 2014;McLaughlin et al., 2004;Mestres-Missé et al., 2008, 2007;Tham et al., 2015), while a smaller literature has investigated acquisition of novel names for novel concepts (e.g., Bakker et al., 2015;Balass et al., 2010;Batterink & Neville, 2011;Borovsky et al., 2012Borovsky et al., , 2010Borovsky et al., , 2013;;Clay et al., 2007;Dumay et al., 2004;Fang & Perfetti, 2017;Liu & van Hell, 2020;Perfetti et al., 2005;Tamminen & Gaskell, 2013;van der Ven et al., 2015).It is likely that the time course of integration is dependent on whether the words-to-be-learned represent novel names for already familiar concepts (typically, when learning a new language as an adult) or refer to concepts not encountered previously (typically, when learning a new word in one's first language as an adult) (cf.Lindsay & Gaskell, 2010).In the present paper, we are interested in how integration occurs in the latter scenario, i.e., when, in addition to a novel name, a novel concept has to be acquired.In the remainder of this section, we therefore restrict our literature review to research focusing on the integration of novel names for novel concepts, with the exception of a few cases where previous research has used exclusively familiar concepts.
In studies focusing on learning of novel names for novel concepts, integration has been almost exclusively examined with a semantic priming paradigm in word recognition (but see, e.g., Clay et al., 2007, for work on word production, Borovsky et al. 2010, for use of a sentence processing task, and Dumay et al., 2018;Geukes et al., 2015, for use of emotional and semantic Stroop tasks, respectively).In a standard (paired) version of this paradigm, participants are presented with pairs of stimuli (primes and targets) and asked to judge targets for lexicality (i.e., word-nonword discrimination) or semantic relatedness to the primes.Importantly, the relationship between the primes and the targets is manipulated such that the prime is either semantically related to the target (e.g., "desk" and "bed") or not (e.g., "hat" and "bed").When targets are preceded by semantically related primes, judgements are faster: there is semantic priming (see McNamara, 2005, for an extensive review).In this condition, the prime is thought to activate its semantic representation, and this activation spreads to representations of semantically related words, including the (semantically related) target (e.g., Collins & Loftus, 1975;Posner & Snyder, 1975).This pre-activation of the target means that, when the semantically related target word is encountered, less additional activation is required, resulting in faster response times than when the target is semantically unrelated to the prime (but see de Wit & Kinoshita, 2015;Holender, 1992;Neely, 1991, for alternative accounts).Crucially, when the prime is a newly learned novel word, it is believed to be able to spread semantic activation to familiar words only if it had been integrated into the lexical-semantic network (e.g., Davis & Gaskell, 2009;Lindsay & Gaskell, 2010;Tamminen & Gaskell, 2013).
In the following sections, we provide a brief overview of the (behavioural and electrophysiological) research using the semantic priming paradigm to study integration of novel names for novel concepts.

Behavioural studies on integration of novel names for novel concepts
Although several studies have addressed integration of novel names for novel concepts, no clear picture of the timescale of integration has emerged.Dumay et al. (2004, Experiment 1) used a continuous primed lexical decision task (word-nonword discrimination for both primes and targets), in which trained novel words were primes and familiar English words were targets.They observed shorter response times for the semantic superordinates (e.g., "vegetable") of the novel words as compared to semantically unrelated category names (e.g., "vehicle") (i.e., semantic priming).In addition, form priming was found for familiar English words (e.g., "cathedral") from which the trained novel word primes were derived (e.g., "catherdruke").Because these effects were present 7 days after training but not immediately after or 24 hours after training, Dumay et al. (2004) concluded that integration of meaningful novel words may take up to one week.
This interpretation was later challenged by Tamminen and Gaskell (2013) who argued that the semantic priming effect in the Dumay et al. (2004) study could have been due to co-occurrence of the novel words (primes) and their semantic categories (targets) during training (i.e., word-to-word episodic association) rather than integration into semantic memory.Tamminen and Gaskell (2013) addressed this issue by teaching their participants two sets of words in two experimental sessions conducted either on two consecutive days (1-day consolidation opportunity group) or with a between-session interval of one week (7day consolidation opportunity group).The semantic priming effect was observed (in a paired primed lexical decision task), and the size of this effect did not differ either across the groups or across the sets, regardless of whether the primes were visible (Experiment 1) or masked (Experiment 2).However, when the data from the two experiments were combined (in attempt to increase statistical power), the interaction between semantic relatedness and time of testing was significant, with a stronger priming effect after a 7-day than after a 0-day consolidation period (the 1-day vs. 0-day consolidation period was not tested).Tamminen and Gaskell (2013) therefore inferred that novel names for novel concepts can indeed be embedded with other lexical-semantic representations relatively quickly, with the initially weak semantic priming effect growing stronger over time.Following up on this work, van der Ven et al. (2015) contrasted the 1-day and 0-day consolidation periods directly and indeed observed a priming effect of 10 ms as early as 24 hours after training (but note that this effect was found in the by-participant analysis only).
Summing up, behavioural studies with novel names for novel concepts do not seem to provide an unequivocal answer regarding the time course of integration, with van der Ven et al. (2015) arguing for a 24hour period, Dumay et al. (2004) suggesting a more protracted time course of one week (but with potential methodological issues), and Tamminen and Gaskell (2013)'s findings remaining inconclusive.

Electrophysiological studies on integration of novel names for novel concepts
Studies examining the neural signatures of integration have primarily focused on two ERP components, the N400 and the Late Positive Component (LPC).The N400 is a negative brain potential with a centro-parietal distribution and a slight right-hemisphere bias that peaks approximately 400 ms after the onset of a conceptually meaningful stimulus (e.g., a word, picture, face, sound; see Kutas & Federmeier, 2011, for a review).Its amplitude typically becomes smaller when the stimulus is congruent with the context it appears in (e.g., the preceding word or part of a sentence).In the semantic priming paradigm, a larger (i.e., more negative) N400 response is typically observed when the target word (e.g., "cat") is preceded by a semantically unrelated (e.g., "table") as opposed to a semantically related word (e.g., "dog"); this effect is usually referred to as the N400 priming effect.While the functional basis of this effect is still debated, one of the most prominent theories posits that it indexes retrieval of a word from semantic memory (e.g., Brouwer et al., 2017;Deacon et al., 2004;Delogu et al., 2019;Kutas & Federmeier, 2000, 2011;Lau et al., 2008, but see, e.g., Brown & Hagoort, 1993;Hagoort, 2008;Rabovsky et al., 2018, for other accounts).When a word is encountered (e.g., "dog"), the system makes predictions about the upcoming stimulus and pre-activates words semantically related to the stimulus (e.g., "wolf", "cat", "horse"; see, e.g., Szewczyk & Schriefers, 2018).Therefore, when the upcoming word (e.g., "table") is not among the pre-activated words, a larger (more negative) N400 is elicited than when the upcoming word is one of the pre-activated words (e.g., "cat").
The LPC is a positive-going wave with a bilateral parietal distribution, peaking at around 600 ms post-stimulus, and is believed to be a marker of episodic memory retrieval (see, e.g., Rugg, 1995, for a review) 2 .In the recognition memory literature, it has been observed that stimuli that are correctly recognised as having been encountered in a prior study list ("old" stimuli) evoke a larger (i.e., more positive) LPC response than other stimuli (correctly rejected new stimuli or incorrectly judged old stimuli; e.g., Curran, 1999;Rugg & Curran, 2007;Senkfor & Van Petten, 1998).This old/new effect has also been found when the participants are not explicitly told to make memory judgements, in, for example, repetition priming paradigms (e.g., Kazmerski & Friedman, 1997;Paller et al., 1995;Van Petten & Senkfor, 1996).In the literature on semantic processing, the LPC has been linked to explicit processes in semantic retrieval, contextual integration, and revision (e.g., Swaab et al., 1998;Van Petten et al., 1991).Consequently, more positive LPC responses for semantically and associatively related as opposed to unrelated targets (labelled the LPC priming effect) have been interpreted as indices of controlled and strategic semantic processing in word recognition (e.g., Hill et al., 2002;Hoshino & Thierry, 2012;Kandhadai & Federmeier, 2010).
Like the behavioural literature, the electrophysiological literature on integration has yielded inconsistent results.Most studies have used the primed semanticrelatedness judgement or standard (paired) lexical decision paradigms.While some studies have reported an N400 priming effect immediately after learning (Balass et al., 2010;Perfetti et al., 2005;and see Batterink & Neville, 2011;Mestres-Missé et al., 2008, 2007, for studies on learning of novel names for familiar concepts), others have failed to observe the N400 effect immediately after, 24 hours after (Bakker et al., 2015;Liu & van Hell, 2020), or even one week after exposure (Liu & van Hell, 2020).
With respect to the LPC, both the old/new LPC effect and the LPC priming effect have been reported.Where an enhanced LPC response for novel versus familiar words was observed, it was interpreted as an index of correct recognition of previously encountered stimuli (Balass et al., 2010;Perfetti et al., 2005).In studies where greater positivity in this spatiotemporal window was observed for related as opposed to unrelated targets, it was taken to index non-automatic processes during lexical retrieval and possibly incomplete integration (Bakker et al., 2015;Borovsky et al., 2013;Liu & van Hell, 2020).In addition, some studies have found a stronger LPC response for familiar words as compared to trained novel words and claimed that it reflected poor recollection of the novel words' meanings (Bakker et al., 2015;Liu & van Hell, 2020).
Importantly, the same caveats as those from the behavioural literature must be considered in the case of the electrophysiological studies.For instance, in Perfetti et al. (2005), many of the targets (familiar English words) had occurred as part of the definitions of the rare words the participants had been trained on, thus making it possible that the observed N400 effects were at least partly due to word-to-word association priming.While this confound was not present in Balass et al. (2010), both Perfetti et al. and Balass et al. used the semantic-relatedness judgement task which is thought to rely on the explicit memory system and require conscious recollection (e.g., Batterink & Neville, 2011).Indeed, when Batterink and Neville (2011) contrasted performance on the semantic-relatedness judgement task with that on the paired primed lexical decision task immediately after training, the N400 priming effect was observed only in the semantic-relatedness task.Batterink and Neville therefore suggested that the primed lexical decision task is a much more reliable measure of implicit memory and argued that explicit representations of novel word meanings can develop rapidly, whereas implicit representations may require more time to emerge.
In sum, integration of novel names for novel concepts into semantic memory has been extensively studied with both behavioural and electrophysiological methods; however, it remains unclear at what point in time and under what conditions novel words begin to interact with familiar words during lexical selection.This is partly because the design of some studies did not allow the disentanglement of word-to-word association priming, arising through co-occurrence of primes and targets during training, from priming driven by semantic relatedness.Yet, this is an important distinction, because the CLS account of word learning predicts that associative priming does not require integration; rather, it can be accommodated by episodic memory.
In addition, many studies have explicitly asked participants to consider the relationship between the primes and the targets (in the semantic-relatedness judgement task, for example; see Bakker et al., 2015;Balass et al., 2010;Batterink & Neville, 2011;Liu & van Hell, 2020;Perfetti et al., 2005), hence, restricting the extent to which priming can be unequivocally attributed to automatic lexical retrieval processes.The primed lexical decision task alleviates these concerns but, in its most common form, where only targets are judged, it is also subject to strategic and conscious processing as the participants are always aware of the fact that the words are paired (e.g., Hutchison, 2003;Lucas, 2000;McRae & Boisvert, 1998).To date, only one (behavioural) study (Tamminen & Gaskell, 2013) has used a version of this task that is less prone to strategic influences (in their case, masked primed lexical decision).Nonetheless, some authors have argued that, to minimise the influence of both prospective and retrospective strategic processes (e.g., de Mornay Davies, 1998;McNamara & Altarriba, 1988;McRae & Boisvert, 1998, but see Shelton & Martin, 1992, for the intra-lexical priming hypothesis), it is necessary to use a continuous primed lexical decision task, where lexical decision is performed on both primes and targets.
The continuous version of the primed lexical decision task is thought to be a superior paradigm for reasons to do with the differential treatment of prime and target trials in the standard and masked lexical decision tasks (e.g., de Mornay Davies, 1998).As, in these tasks, participants are not required to respond to primes, they may generate potential (related) candidates that may appear as targets.If the actual target matches one of these potential candidates, response times will be shorter (the expectancy generation strategy, see Neely, 1991).Likewise, participants may assess the semantic relationship between prime and target before making their target decision, such that their decision (on the semantically related trials) is aided by the presence of a relation between the two words (the post-lexical checking strategy).Finally, if no judgements are required for the primes, the participants may not process them fully or even ignore them, which may reduce the likelihood of detecting the (relatively small) automatic semantic priming effect (typically, 10-20 ms, see, e.g., McNamara & Altarriba, 1988).Given these considerations, we chose to use the continuous primed lexical decision task in the present study to maximise the possibility of detecting a priming effect with novel words, especially given that this effect is likely to be even smaller than that typically observed with familiar words.

Current study
We taught our participants novel names, one set on each of two consecutive days (remote and recent set, respectively), and then, using continuous primed lexical decision, studied (a) whether behavioural (faster response times for related vs. unrelated targets) and/or electrophysiological (reduced N400 and enhanced LPC amplitudes for related vs. unrelated targets) priming effects would emerge for the trained novel name targets, and (b) whether these effects would vary depending on the length of the consolidation opportunity (0 hours for the recent set vs. 24 hours for the remote set).Furthermore, since enhanced negativity for unknown pseudowords (compared to familiar words) in the N400 spatiotemporal window has been suggested to reflect differences in the lexical status (e.g., Bentin, 1987;Kutas & Federmeier, 2011), we also contrasted the ERP responses for trained novel names with ERP responses for untrained pseudowords and with those for familiar words in this spatiotemporal window.

Participants
Seventy-two 3 participants (28 male, 44 female), aged between 18 and 35 years (m = 20.94,s = 3.86), took part in the experiment.All participants reported Australian English as their first and dominant language, and none had started learning any other languages prior to the age of six.All participants reported being righthanded and having normal hearing, normal or corrected-to-normal vision, and no history of neurological or psychiatric disorders.Participants received either course credits or monetary reimbursement (70 AUD) for their participation.Prior to the start of the experiment, participants were informed about the experimental procedure and gave written consent.The experiment was granted ethical approval from the Macquarie University Human Research Ethics Committee (approval No. 3531).
One participant performed below chance on the continuous primed lexical decision task; this participant's data was discarded.The final sample thus consisted of 71 participants (27 male, 44 female; age: m = 20.94,s = 3.89).

General procedure
The experiment comprised two experimental sessions conducted on two consecutive days (Figure 1).In both sessions, in the learning phase, participants learned novel names for two sets of novel concepts (e.g., a novel bird, plant, etc.), one set in each session (henceforth referred to as the remote and the recent set, respectively).In Session 1, the learning phase was followed by a word recall task for the remote set (learned in that session) and a working memory test.In Session 2, the learning phase was followed by a word recall task for the recent set (learned in that session), a continuous primed lexical decision task, and a word recall task for both sets (always in that order).EEG was recorded throughout Session 2.
The experiment was run using Presentation® software (Version 20.2,Build 07.25.18;Neurobehavioral Systems,Inc.,Berkeley,CA,www.neurobs.com).Participants were tested individually in a quiet room at Macquarie University.The mean time elapsing between the two sessions was 25.82 hours (s = 2.80, range: [16-31.50]).All participants reported having their usual duration and quality of sleep the night between the two sessions.Session 1 lasted 1.5 hours and Session 2 took 2 hours, including breaks and the EEG set up.Data from the learning phase is reported in Korochkina et al. (2023), and that from the working memory test will be reported elsewhere.

Learning phase
Participants were required to learn novel names for 40 novel concepts having been presented with the descriptions of those concepts.

Materials
Concept descriptions.Forty (existing) concepts were selected, each from a different semantic category (e.g., one plant, one bird, one drink, etc.).From those concepts, 40 novel concepts were derived such that each novel concept represented a highly similar category coordinate.For example, if aloe was selected as an existing concept, a novel plant was generated that was similar to aloe but with partly different semantic features (e.g., ability to survive extreme drought by curling into a tight ball and uncurling when exposed to water).Each of the 40 novel concepts was described with 4 short sentences that did not include the category name of the concept (sentence length: m = 9.98 words, s = 2.58).These sentences were adapted from definitions of existing concepts extracted from online dictionaries and Wikipedia.
Twenty-three Australian native speakers (5 male, 18 female; age: m = 36.17,s = 13.47) were given these descriptions along with descriptions of 20 existing concepts (fillers) and judged whether they were familiar with the described concepts and which semantic categories the concepts were likely to belong to.Participants correctly identified the intended semantic categories of the novel concepts.For concepts that the participants believed were familiar (i.e., that the description referred to a real thing), they were asked to specify what that concept was.Those novel concepts that more than 40% of the participants reported were existing things (e.g., responded that the novel plant was an aloe), were replaced by new concepts or better descriptions of the same concepts (15% of the descriptions).
If participants were not familiar with the concept (as was expected for these novel concepts), they were asked what it reminded them of.The expected response was the semantically related existing concept that had been used to generate the novel concept.Those 20% of novel concepts for which participants showed less than 90% agreement in terms of a similar concept were replaced by new concepts.Where agreement was less than 90%, but more than 65% of the participants agreed on a similar existing concept (e.g., cactus instead of aloe), the novel concept and its description were changed to be more similar to the existing concept the participants supplied (e.g., to cactus).
The corrected descriptions were given to another 10 speakers of Australian English (2 male, 8 female; age: m = 28.20,s = 5.75), which confirmed that there was high (90%) agreement regarding the most closely semantically related existing concepts (which always belonged to the same semantic category as the novel concepts; see Table 1

for examples).
Novel names.Forty novel words were created.The novel words were generated by the first author, and we ensured that they were phonotactically and orthographically legal in Australian English.To ensure that there was some variability in syllabic structure and that the novel names were neither too easy nor too difficult to learn, half of the stimuli were monosyllabic and half disyllabic.Ten native speakers of Australian English judged that these novel words neither existed in English nor elicited strong associations with existing English words.Each novel word was audio-recorded by a male native speaker of Australian English.
Lists.Four lists were created to ensure that each concept description and each novel name appeared in both the remote and the recent set.We first split the novel concept descriptions and novel names into two sets of 20 each (Concept description sets 1 and 2; Table 1.Sentences that formed the concept descriptions of (A) a novel fish similar to a shark, and (B) a novel plant similar to a cactus.

A B
It lives in saltwater and has a prominent dorsal fin.
It has tiny round leaves and very strong roots.It is able to detect electromagnetic fields that living things produce.
It is only found in deserts.
It has a small body and an even smaller head and moves very slowly.
It can survive extreme drought by curling into a tight ball.It always swims near the surface.
It uncurls when exposed to water.
Novel name sets A and B).Each set of novel names had an equal number of monosyllabic (N = 10) and disyllabic (N = 10) novel words, and a Bayesian alternative to a ttest (as implemented in the R packages BEST and BayesFactor; see Appendix A for model output) confirmed that the two sets were matched for orthographic Levenshtein distance 20 (OLD20, see Yarkoni et al., 2008;computed

Procedure
In each session, participants learned one set of 20 novel concept descriptions paired with novel names (Figure 2).
In each trial, following a 500 ms fixation cross, the written form of a novel name (in black Arial font, size 60) appeared in the middle of the screen.Participants were asked to view it silently for 2200 ms, during which a recording of the spoken form of the novel name was played.Next, four sentences describing the novel concept associated with the novel name appeared one by one (in black Arial font, size 25), with 3000 ms between each sentence, with previous sentences remaining on the screen.After the last sentence, all four sentences stayed on the screen for another 9000 ms, together with the novel name.Following a blank screen for 1000 ms, the written form of the novel name (Arial font, size 60) was presented again for 3000 ms, this time in red and with an image of a microphone above it, and the participants were asked to read the novel name aloud.After a blank screen for 1000 ms, the next trial began.When all the novel names and concept descriptions had been presented once, the set of 20 pairs was presented three more times, each time in a different order.The presentation order was pseudo-randomised such that (a) the last novel name shown in one round was not first in the next round, and (b) none of the novel names appeared at the same position more than twice across the four rounds.Therefore, the participants were exposed to each novel concept description-novel name pair 4 times.In each session, the learning phase took approximately 45 minutes including breaks.

Recall tasks
Immediately following the learning phase, a recall task was used to enable evaluation of learning in each session.
In the recall tasks, participants were presented with the concept descriptions from the set of 20 names they had just learned (i.e., the remote set in Session 1 and the recent set in Session 2) and asked to provide the names of the novel concepts associated with the descriptions.Each trial started with a 500 ms fixation screen.The descriptions (sets of four sentences) appeared on the screen (in black Arial font, size 25) with an image of a microphone above them.The participants were required to respond orally (with a timeout set to 5000 ms) and, after the image of the microphone disappeared, type the written form of the novel name in a response box displayed under the description (with a timeout of 10000 ms).The task lasted approximately 5 minutes (including one break).By administering singleset word recall tasks immediately after a corresponding training phase we sought to ensure that both sets had an equal amount and type of exposure prior to the administration of the priming task.These tasks also gave us a measure of explicit learning (number of words recalled correctly).
Participants completed another recall task at the end of Session 2 (following the continuous primed lexical decision task, see Section 2.5 below).The procedure was identical to that described above, however, descriptions of all novel concepts (from both the remote and the recent set) were presented.The descriptions appeared in a random order (i.e., not blocked by set).The task lasted approximately 10 minutes including breaks and was used to (a) examine explicit memory for words from the recent set after a 1-day consolidation period and (b) assess whether the priming task has benefitted explicit memory for the remote set.

Continuous primed lexical decision task
The design of this task was based on McRae and Boisvert (1998, Experiment 1).Following Bakker et al. (2015) and Liu and van Hell (2020), we used trained novel names as targets and familiar words as primes.This enabled us to measure the brain responses to the trained novel words (primed by familiar English words) which would not have been possible if the trained novel words were used as primes.Moreover, we wished to test whether the outcomes of the Bakker et al. (2015) and Liu and van Hell (2020) studies could be found using a priming task that reduced the extent of strategic processing (relative to the primed semantic-relatedness judgement task they used).

Materials
Eighty word pairs, each consisting of a prime word and a target word, were created.Half of the word pairs (N = 40) comprised pairs of English words (English pairs), while the other half (N = 40) consisted of an English prime word and a novel target name from the learning phase (English-novel pairs).For both English and Englishnovel pairs, there were 20 semantically related pairs (with primes and targets belonging to the same semantic category; semantically related condition) and 20 semantically unrelated pairs (primes and targets from different semantic categories; semantically unrelated condition).The full list of stimuli used in the continuous primed lexical decision task is provided in Appendix C.
English-novel pairs.For each of the 40 trained novel names, a semantically related English word and a semantically unrelated English word were selected as primes.The semantically related primes comprised the base English words from which the novel concepts were derived (e.g., "cactus" for the novel concept shown in Table 1).The semantically unrelated pairs (N = 40) were created by pseudo-randomly re-pairing the primes and the targets to ensure that words from different but semantically close categories (e.g., animals and birds, or food and drinks) were not paired as unrelated.
English pairs.We selected one pair of familiar concepts from each of the semantic categories (N = 40) used in the English-novel pairs.We aimed to ensure that the category coordinates were as similar as possible.However, to minimise the "associative boost" observed when a semantic relationship is accompanied by a normative association (e.g., Lucas, 2000), we ensured a low associative relationship (the two words were not among the first three of each other's associates and had the associative strength of 0.1 or less as given in the Small World of Words association norms; De Deyne et al., 2018).As with the English-novel pairs, the semantically unrelated English pairs (N = 40) were created by pseudo-randomly re-pairing the primes and the targets.Fifteen Australian native speakers (2 male, 13 female; age: m = 27.87,s = 8.78) then judged the semantic relatedness of the pairs by specifying whether the pairs of words referred to the same concept.If they responded negatively, they were asked to rate how similar in meaning the words in the pairs were to each other on a 7-point scale from 1 ("very different") to 7 ("very similar").The mean rating for the related pairs was 5.26 (s = 1.68) and, for the unrelated pairs, 1.28 (s = 0.97).
Fillers.Each list included 160 filler nonwords to equalise the number of "yes" and "no" responses for lexical decision.These nonwords were created following the same criteria as for the novel names used in the learning phase (equal number of monosyllabic and disyllabic words with all words being phonotactically and orthographically legal in Australian English) and matched with the trained novel names for OLD20, number of letters, and bigram frequency (see Appendix A).The same group of 10 native speakers who judged the novel names for the learning phase confirmed that the filler nonwords neither existed in English nor elicited strong associations with existing English words.
Lists.Eight lists were created such that, for each of the four lists used in the learning phase, in the continuous primed lexical decision task, half of the participants saw the trained novel words in the semantically related and the other half in the unrelated condition.Each of the eight lists contained 80 primes, 80 targets (i.e., 80 prime-target pairs), and 160 filler nonwords (i.e., 320 items in total).In each list, there were 40 English pairs (20 semantically related, 20 unrelated) and 40 English-novel pairs (20 semantically related, 20 unrelated).Pairs containing the trained novel names from the remote and the recent sets were equally distributed across conditions, and, in each list, each prime and target only appeared once.Each semantic category (N = 40) appeared once in the English pairs and once in the English-novel pairs, counterbalanced across the semantically related and unrelated conditions (see Table 2 for an example).In each list, English targets from the semantically related and unrelated pairs were matched for OLD20, number of letters, and word frequency (obtained from the MCWord database), while the related and unrelated novel name targets were matched for OLD20, number of letters, and bigram frequency (see Appendix A).

Procedure
Both experimental (primes and targets) and filler stimuli were presented to the participants in lowercase (in black Arial font, size 60), one stimulus at a time.For each letter string, the participants were required to judge whether it corresponded to a word that they knew by pressing an appropriate button (the top left button for "yes" and the top right button for "no" on a Cedrus RB-840 button box).The participants were instructed to respond "yes" to all words that they were familiar with, regardless of whether they had learned them in the course of the experiment or had known them previously.This instruction was given to ensure that there was no confusion as to how the novel names were to be responded to, and it was also consistent with the instruction given in the learning task (the participants were told that they would be learning English names for unfamiliar objects).
Primes and targets were not paired explicitly but appeared as adjacent items, with lexical decision being required for every item.The presentation order was pseudo-randomised separately for each participant with the following constraints: (a) a maximum of four consecutive items per condition (related, unrelated, filler), (b) a maximum of four consecutive items where the novel names (targets) were from the same set (remote or recent).This design and the low proportion of semantically related pairs (25%) was used to minimise strategic processing and awareness of the experimental manipulation by the participants.
The task started with 20 practice trials.The experimental trials were presented in blocks of 20, with short breaks in between.Each block started with a filler trial.The participants were told to respond as quickly and as accurately as possible upon the presentation of the stimulus.Independent of whether and when a response was given, the stimulus stayed on the screen for 1000 ms (timeout).This was followed by a blank screen for 200 ms.The blank screen was replaced by the next stimulus, either experimental or filler.

Pre-processing of behavioural data
Recall tasks.The accuracy of the participants' vocal responses was manually checked using Praat (Boersma & Weenink, 2014), and only responses that corresponded to the trained phonological form were counted as correct.The judge was blind to the experimental condition in which the responses were collected.The accuracy of the typed responses was evaluated automatically by the Presentation software, and only responses that exactly matched the trained orthographic form were coded as correct.
Continuous primed lexical decision task.Prior to conducting any descriptive or inferential statistical analysis, we inspected the distribution of the response times (for correct responses) across participants and items.Data points that clearly stood outside the distribution (response times shorter than 250 ms or longer than 950 ms) were removed (Baayen, 2008).This resulted in less than 1 per cent data loss.
2.7.Acquisition and pre-processing of electrophysiological data EEG was recorded using a 64-channel BioSemi Acti-veTwo electrode system (Amsterdam, Netherlands) at a sampling rate of 2048 Hz.The Ag-AgCl-tipped electrodes were attached to an electrode cap using the 10/20 system.The signal was recorded relative to an additional (active) electrode, the common mode sense (CMS), which formed a feedback loop with another additional (passive) electrode, the driven right leg (DRL).These two electrodes replaced the ground electrode of more conventional systems.Further electrodes were placed on the left and right mastoid as well as at the outer canthus of the left eye.The raw EEG signal was pre-processed in MATLAB (version R2020b; Natick, Massachusetts: The MathWorks Inc), using the EEGLAB toolbox (Delorme & Makeig, 2014).Two datasets were created.
For both datasets, the continuous signal was resampled at 500 Hz and filtered with a 1 Hz high pass filter (Kaiser windowed sinc FIR filter, order = 802, beta = 4.9898, transition bandwidth = 2 Hz) and a 40 Hz low pass filter (Kaiser windowed sinc FIR filter, order = 162, beta = 4.9898, transition bandwidth = 10 Hz).Line noise removal and bad channel detection were performed using the cleanLineNoise and findNoisyChannels functions from the PREP pre-processing pipeline (Bigdely-Shamlo et al., 2015), and both datasets were re-referenced to the algebraic average of the left and right mastoid.Subsequently, one of the datasets (excluding the bad channels) was subjected to Independent Component Analysis (ICA; Chaumon et al., 2015), while the other dataset was high pass filtered at 0.1 Hz (Kaiser windowed sinc FIR filter, order = 8008, beta = 4.9898, transition bandwidth = 0.2 Hz).
The demixing matrix obtained from the ICA (run on dataset 1) was then applied to the second dataset (filtered at 0.1 Hz).ICLabel, an automated EEG independent component classifier (Pion-Tonachini et al., 2019), was used to exclude individual components corresponding to activity originating from the eyes and groups of muscle motor units as well as the electrocardiographic signals.Next, the noisy channels identified earlier were spherically interpolated.The signal was then segmented into epochs of 1.2 s, from 200 ms before the stimulus onset to 1 s after the stimulus onset.The epochs were baseline corrected using the mean of the signals between −200 ms and 0 ms relative to the stimulus onset.Finally, epochs with the amplitude below −100 mV or above 100 mV were excluded.At the end of the pre-processing, a mean of 3.73 epochs per participant (s = 9.30, range: [0-62]) were excluded (note that epochs corresponding to the primes were not analysed).In the semantic priming analysis for the English pairs (Preliminary analysis, see below), a mean of 0.32 epochs per participant per condition (related: s = 0.89, range: [0-5]; unrelated: s = 0.95, range: [0-5]) were excluded.In the semantic priming analysis for the English-novel pairs (Main analysis), a mean of 0.20 epochs per participant were removed for the recent set (s = 0.69, range: [0-5]; related: m = 0.20, s = 0.62, range: [0-4]; unrelated: m = 0.21, s = 0.75, range: [0-5]) and 0.12 epochs per participant for the remote set (s = 0.40, range: [0-3]; related: m = 0.10, s = 0.34, range: [0-2]; unrelated: m = 0.14, s = 0.46, range: [0-3]).For the filler nonwords in the lexicality analysis (Main analysis), a mean of 2.44 epochs per participant (s = 5.98, range: [0-41]) were discarded.Finally, epochs corresponding to trials with incorrect behavioural responses to the target words were removed, which, together with the epochs excluded earlier, constituted 13.7% of the data.As with the behavioural data, prior to conducting any descriptive or inferential statistical analysis, we inspected the distribution of the amplitudes across participants and items, and one additional data point (in the dataset restricted to the novel name targets, i.e., Main analysis) was removed.

Analysis
We used analytical tools from both the Bayesian and the frequentist frameworks to analyse the data.All Bayesian models (behavioural and electrophysiological, see below) were performed in R, version 4.1.1(R Core Team, 2021) using the brms package (Bürkner, 2017a(Bürkner, , 2017b)).For each model, four sampling chains were run, each with 10000 iterations.Of the 10000 iterations, the initial 2000 were warm-up iterations.Thus, there were a total of 32000 sampling iterations available for analysis.For each parameter of interest, we report the estimate of the posterior mean with its 95% credible interval (95%CrI), which denotes the range of values within which we can be 95% certain that the true value of the parameter falls (e.g., Nicenboim & Vasishth, 2016).
Note that analyses reported below 4 differ from those reported in an earlier version of this manuscript (see https://psyarxiv.com/vup25/), as we modified the priors following a reviewer's comments.While this modification does not change the results and their interpretation in any substantial way, we believe that the updated analyses facilitate comparison with previous research and ultimately provide a more comprehensive understanding of our data.For readers interested in the original analyses and in how they compare to those reported below, the code and the outcomes of the original analyses are available on OSF and their discussion can be found in the publicly available pre-print (https://psyarxiv.com/vup25/).

Behavioural data
Response times (RTs) for correct responses were analysed using Bayesian Linear Mixed Effects models (e.g., Nicenboim & Vasishth, 2016).Two analyses were preregistered: the preliminary analysis examined whether there was a semantic priming effect for the English targets, while the main analysis examined (a) whether there was a semantic priming effect for the trained novel name targets, and (b) whether this effect was modulated by the length of the consolidation period (24 hours for the remote set, 0 hours for the recent set).There were no deviations from the pre-registered analyses.
All models included the maximal random effects structure for the location parameter μ (i.e., correlated varying intercepts and slopes for subjects and items, see e.g., Barr et al., 2013).The contrasts of interest were defined using sum contrast coding such that the model coefficients corresponded to the difference between the two conditions (e.g., for the effect of semantic relatedness, targets preceded by semantically related primes were coded as −0.5 and those preceded by semantically unrelated primes as +0.5).We used the lognormal distribution to specify all model parameters except for the correlation matrix for the random effects, which was defined with an LKJ-prior (Lewandowski et al., 2009;Stan Development Team, 2017).For this prior, we chose a diffuse parameter value of 2 as it allows regularisation of the distributions of the correlation parameters such that extreme values such as +1 and −1 are less likely a priori.
The priors for the preliminary (English pairs) and main (English-novel pairs) analyses differed only in terms of the priors for the intercept.These priors were based on average response times typically reported in the literature in the (paired and continuous) primed lexical decision tasks as well as on our expectation that mean lexical decision times would be slightly faster for familiar words than for the trained novel names.Thus, for the intercept in the preliminary analysis, we assumed a normal distribution with a mean of 6.3 and a standard deviation of 0.3, reflecting a mean RT of about 570 ms (with a median of about 545 ms and SD of about 361 ms), with 95% of the values falling roughly within the 300-990 ms interval (recall that RTs below 250 ms and above 950 ms were excluded during pre-processing).In contrast, for the intercept in the main analysis, we defined a normal distribution with a mean of 6.4 and a standard deviation of 0.35, which corresponds to a distribution with 95% of the values lying between about 270 ms and 1090 ms, with a mean RT of about 640 ms, median RT of about 602 ms, and a SD of about 413 ms.
All models were fitted using four sets of priors (as part of a sensitivity analysis) that differed only in terms of the assumed range of the priming effect (see Table 3).In all sets of priors, the prior was sampled from a normal distribution centred around 0, reflecting no commitment to the direction of the effect.The standard deviation of the distribution differed across the sets: s = 0.001 in prior set 1 (i.e., 95% of the values lie within a −1 ms and to +1 ms range; no priming effect), s = 0.008 in prior set 2 (i.e., 95% of the values within a −10 ms and to +10 ms range; small priming effect), s = 0.02 in prior set 3 (i.e., 95% of the values within a −24 ms and to +24 ms range; medium-sized priming effect), and s = 0.025 in prior set 4 (i.e., 95% of the values within a −28 ms and to +28 ms range; large effect).These definitions of effect sizes (no effect, small effect, medium-sized effect, large effect 5 ) were based on average effect sizes reported in mega-studies using the standard lexical decision task (e.g., Brysbaert & Stevens, 2018).For lack of better estimates, priors for other effects of interest (in the main analysis, main effect of time of learning, remote vs. recent, and the interaction between this effect and the priming effect) were based on those for the effect of semantic relatedness.In the following section, we report the back-transformed model estimates as milliseconds for ease of interpretation.Please note that, in all figures, error bars and uncertainty intervals represent one standard error from the mean, whereby standard errors were corrected for within-subjects variability unless stated otherwise (note that, as this also applies to EEG data figures, we do not repeat this information in the following subsection).

Electrophysiological data
Electrophysiological data were analysed with Bayesian Distributional Regression models (e.g., Rigby & Stasinopoulos, 2005) and Mass Univariate analysis (e.g., Groppe et al., 2011aGroppe et al., , 2011b)).Two analyses were preregistered: the preliminary analysis examined whether there were differences in mean amplitudes between the semantically related and unrelated English targets in the N400 and in the LPC spatiotemporal windows.The main analysis examined, in the semantic priming analysis, (a) whether there were differences in mean amplitudes between the semantically related and unrelated trained novel name targets in the N400 and in the LPC spatiotemporal windows, and (b) whether this effect was modulated by the length of the consolidation period (0-hour consolidation for the recent set and 24hour consolidation for the remote set).Then, in the lexicality analysis, we examined (c) whether there were differences in mean amplitudes between the semantically unrelated English targets, the semantically unrelated trained novel name targets, and the nonwords.
In line with the pre-registered plan, we first fitted Linear Mixed Effects models with a maximal random effects structure for the location parameter μ; however, these models misfitted the by-subject standard deviations.We then fitted Distributional Regression models with a group-level random effects structure for both the location and the scale (σ) parameters, and these models provided a much better fit for our data.All models included varying intercepts for the scale parameter, while, for the location parameter, models testing for priming effects included correlated varying intercepts and slopes for subjects and items, and models testing for lexicality effects contained correlated varying intercepts and slopes for subjects and varying intercepts for items (because one item could be either an English word, or a novel name, or a nonword, but never all).In the priming analysis, sum contrast coding (−0.5, +0.5) was used to define the contrasts of interest.For the lexicality analysis, sum contrasts were specified in the pre-registration, however, this contrast coding would have been unsuitable given that each item could only be either a nonword, or an English word, or a trained novel word (from either set).We therefore employed repeated contrasts, and, given that the independent variable (word type) had 4 levels, we could N (0, 0.025) SD 1-4 N (0, 0.05) σ 1-4 N (0, 0.5) Note: Models in the preliminary analysis only included one fixed effect (semantic relatedness, related vs. unrelated), while models in the main analysis included two fixed effects (semantic relatedness and time of learning, remote vs. recent) and their interaction.
only run 3 contrasts at a time: (1) nonwords vs. English words, (2) recent novel names vs. nonwords, and (3) remote vs. recent novel names.These three comparisons were chosen for several reasons.Firstly, in the literature, the lexicality effect refers to the difference between the nonwords and the familiar, existing, words, and, by running this contrast, we wished to see whether we could replicate the lexicality effect in our study.Secondly, we wished to see whether processing of the trained novel names would be different from that of the nonwords as well as whether it would vary as a function of the length of consolidation.To explore whether consolidation resulted in more word-like processing of the trained novel names, we also performed an exploratory analysis with three more contrasts: (1) remote novel names vs. English words, (2) recent novel names vs. English words, and (3) nonwords vs. remote novel names; the results of this exploratory analysis with a brief discussion are provided in Appendix D. Another deviation from the pre-registered plan was that we fitted separate models for the frontal and parietal electrodes in the 500-800 ms time window to avoid estimation of a three-way interaction.
As with the behavioural data, four sets of priors were defined, and they differed from each other only in terms of the range of the assumed effects (see Table 4).Correlation matrices were specified using the LKJ-prior, whereas, for all other model parameters, normal distribution was used (note that we used the lognormal distribution to define the prior for the standard deviation for the prior for the intercept of sigma only to ensure that negative by-subject adjustments to the intercept did not cause sigma to become negative).In choosing the plausible ranges for the effects of interest, we relied on the fact that, across published studies, both the N400 and the LPC effects are typically rather small, ranging from 5% to 30% of the standard deviation of the EEG signal, which, for N400 and LPC averages, usually falls between 8 mV and 15 mV.As with the behavioural data, all priors were centred around 0, reflecting an assumption that the effect could be either positive or negative, while the standard deviations differed across the four sets.In prior set 1, s = 0.5, suggesting that the difference between the two conditions would be about 5% of the upper bound of the SD of the signal (i.e., a very small difference, with 95% of the values falling between −1 mV and +1 mV).Prior set 2 (s = 1) assumed an effect of about 10% of the SD of the signal, with 95% of the values falling between −2 mV and +2 mV.Finally, priors 3 (s = 2) and 4 (s = 3) assumed effects of about 20% and 30% of the SD of the signal, with 95% of the values lying between −4mV and +4mV and −6mV and +6mV, respectively.
Mass univariate analysis.This analysis was employed to explore latencies and locations other than the N400 and the LPC spatiotemporal windows as the mass univariate analysis allows testing for significant differences between the conditions of interest across the whole scalp and at every time point (see, e.g., Groppe et al., 2011aGroppe et al., , 2011b, for details).The analysis was conducted using LIMO, an open-source MATLAB toolbox (Pernet et al., 2011).In LIMO, the data is analysed using a Hierarchical General Linear model.In our study, model parameters (i.e., beta coefficients for the predictor of interest) were first estimated (using trial-based Ordinary Least Squares) separately for each participant at each time point and electrode (first-level analysis; see Bellec et al., 2012).These parameter estimates were then used to test the statistical significance of the predictors at a group level (using robust paired t-tests; secondlevel analysis).Finally, we corrected for multiple comparisons using the Threshold Free Cluster Enhancement technique (TFCE; see, e.g., Pernet et al., 2015;Smith & Nichols, 2009).This method was chosen because it avoids some of the known issues of other clusterbased correction approaches (e.g., high sensitivity to the initial cluster-forming threshold), while it provides an equally strong control of the family-wise error rate.There were two deviations from the pre-registered analysis: (1) we used the same contrast coding as in the Bayesian analysis (repeated contrasts) because the pre-registered contrasts proved to be inappropriate for the data, and, (2) while correcting for multiple comparisons, we relied on bootstrapping rather than on permutation tests because bootstrap techniques have been shown to, in most cases, perform similarly to, and, under certain conditions, even outperform permutation techniques, and can thus be considered more versatile (Pernet et al., 2011;Wilcox, 2005).

Results and discussion
In the recall task immediately after learning of the remote set, mean response accuracy, for the spoken responses, was 42% (SD = 13%) and, for the typed responses, 51% (SD = 12%).Mean accuracy in the recall task immediately after learning of the recent set Intercept N (0, 5) Effect of interest N (0, 0.5) N (0, 1) N (0, 2) N (0, 3) SD N (0, 5) Intercept for σ N (0, log (10)) SD for σ N (0, 5) was very similar, with 44% correct (SD = 12%) for the spoken and 50% correct (SD = 12%) for the typed responses.The seemingly higher accuracy for the typed responses is probably due to the fact that these were collected after the spoken responses such that the participants had more time for retrieval.In the recall task at the end of the second session which probed both sets, mean accuracy, for the recent set, was 43% (SD = 12%) for the spoken and 49% (SD = 12%) for the typed responses, and, for the remote set, 30% (SD = 13%) and 34% (SD = 13%), respectively.We take these results to indicate that the participants learned both sets to the same extent, but that some forgetting happened overnight (for the remote set).Because we did not have any research questions with respect to the recall data, we did not analyse it further; however, we elaborate on the possible implications of the relatively low recall accuracy in the General Discussion.
Response accuracy for all targets used in the continuous primed lexical decision task (English words, trained novel names, nonwords) is reported in Table 5.

Preliminary analysis
This analysisthe semantic priming analysis for the English prime-target pairsprovides a baseline against which the novel names can be compared; however, the goal of this study was to examine the integration of novel words.Therefore, to conserve space, we report the results of this preliminary analysis in Appendix E. In the paragraph below, we provide a summary of the results of this analysis, and a more detailed discussion is available in Appendix E.
We found evidence for both behavioural and N400 priming effects with the English prime-target pairs; however, only a very small effect was observed in the LPC spatiotemporal window.Our finding of a behavioural priming effect replicates some of the previous studies (e.g., de Mornay Davies, 1998;McRae & Boisvert, 1998) and shows that semantic priming can be found even if the (semantically related) primes are not associatively related to the targets.As to the N400 priming effect, to our knowledge, this is the first study to report this effect for using a continuous primed lexical decision task.We take the outcomes of our preliminary analysis to indicate that the paradigm used in our study was indeed sensitive enough to tap into automatic processes of lexico-semantic retrieval.With this in mind, we now turn to our findings for the trained novel names.

Main analysis: semantic priming for the
English-novel name prime-target pairs 4.2.1.Behavioural data Figure 3 displays the response time data (for correct responses) for the novel name targets.The models included two predictors, semantic relatedness (unrelated vs. related, coded as +0.5 and −0.5, respectively) and time of learning (recent vs. remote, coded as +0.5 and −0.5, respectively), and their interaction, with four sets of priors (see Table 6).
When the effects of interest were assumed to lie between −1 ms and 1 ms (i.e., the most constraining prior in our analysis, which corresponded to the prior distribution of N (0, 0.001)), the Bayes factor analysis did not provide evidence that the initial belief about the effects of interest (prior odds) needed revising.When effects within a range of −10 ms and 10 ms were assumed (i.e., the prior allowed for more variability, prior distribution of N (0, 0.008)), the analysis showed extreme change in evidence (BF 10 = 179.41)for an interaction, reflecting a greater difference between the unrelated and related targets in the remote (7 ms) than in the recent (1 ms) set.However, a further increase in the assumed range of the interaction effect (effects ranging between −20 ms and 20 ms and between −30 ms and 30 ms, corresponding to prior distributions of N (0, 0.02) and N (0, 0.025), respectively) resulted in agnostic Bayes factors (no evidence).With respect to the main effects, Bayes factors for the effect of semantic relatedness, across different priors, provided anecdotal evidence against a difference between the novel names preceded by the unrelated (607 ms) as opposed to the related (603 ms) primes.In contrast, there was anecdotal to moderate evidence in favour of a small effect of time of learning, with recent novel names (600 ms), on average, being responded to 11 ms faster than the remote novel names (611 ms).

Electrophysiological data
Analysis of mean amplitudes in the (pre-defined) spatiotemporal windows.For both spatiotemporal windows, fixed effects in all models included semantic relatedness and time of learning, as well as their interaction.For both windows, irrespective of the prior, Bayes factors (see Table 7) indicated anecdotal to moderate change in evidence against all main effects and interactions, suggesting that there were no differences in the mean amplitudes for the semantically related versus unrelated novel name targets trained in either set either in the N400 (Figure 4) or in the LPC (Figures 5 and 6) spatiotemporal windows.
Mass univariate analysis.The analysis showed that the recent novel names elicited more positive amplitudes than the remote novel names (main effect of set) between about 506 ms and 588 ms across a set of frontal, central, and parietal electrodes, and more negative amplitudes than the remote novel names between about 700 ms and 735 ms across a set of parietal, as well as at some frontal and central electrodes (see Figure 7).Figure 8 shows that, either for the main effect of semantic relatedness or for the interaction between time of learning and semantic relatedness, no significant t-values remained after TFCE correction.

Semantic priming analysis: results summary and discussion
In the behavioural data, we obtained (moderate) evidence that response times for the novel name targets trained in the remote set were about 11 ms slower than for the novel name targets trained in the recent set.This difference could mean that discriminating the novel names from nonwords was more difficult for the   remote names, possibly due to a longer delay between the time of training and the priming task than for the recent set.This interpretation is in line with our finding that, compared to the remote set, the novel names in the recent set elicited more positive amplitudes between about 506-581 ms and more negative amplitudes between 707-735 ms, respectively, suggesting differences in familiarity and recollection across the two sets (see General Discussion).Moreover, in the behavioural data, there was also evidence for a small interaction between time of learning and semantic relatedness, indicating that, while there were no differences in response times for the related versus the unrelated targets trained in the recent set, a small priming effect of about 7 ms was present for the remote novel names.One possible account of this finding could be that the semantically related, familiar, English words facilitated recognition of the trained novel names that had undergone a 24-hour consolidation opportunity (remote set), while no such facilitation occurred for those novel names that had no such opportunity (recent set).At the electrophysiological level, in the analysis of mean amplitudes in the N400 and the LPC pre-defined spatiotemporal windows, we obtained moderate evidence against a difference between the related and the unrelated conditions in either window, irrespective of the length of the consolidation opportunity.While Note: The graph shows significant t-values before and after TFCE correction (panel A), topographic maps of amplitudes between 510 ms and 580 ms and between 710 ms and 730 ms post onset (panel B; thick black dots correspond to the electrodes with significant t-values after TFCE), and mean amplitudes in both sets at the electrode CP4, where significant t-values were found in both time windows (panel C; the time windows between 506 ms and 581 ms, and between 707 ms and 735 ms, where the t-values were significant at this particular electrode, are shaded grey).
we acknowledge that evidence in favour of the null hypothesis can have many causes and whether it signals a true absence of an effect may warrant further investigation, we think that potential explanations for an absence of priming effects in these two spatiotemporal windows should still be considered, and we do so in the General Discussion.
Notably, while both mean amplitude and mass univariate analyses seem to suggest that there were no differences between the two sets of words in the time window typically associated with the N400 ERP component (300-500 ms post onset), the results of these analyses appear to conflict regarding latencies after 500 ms post onset (no differences between the remote and the recent names in the pre-defined LPC window but significant effects between 500-600 ms and 700-800 ms in the mass univariate analysis).This contradiction could be due to the fact that different sets of electrodes were included in the two analyses: the mass univariate analysis showed that, between 500 ms and 600 ms, the differences between the two sets of novel names were present across a set of mostly frontal and central electrodes and, between 700 ms and 735 ms, across a (small) set of mostly central and parietal electrodes.In contrast, the LPC spatiotemporal window was defined as mean amplitude across a set of frontal and parietal electrodes between 500 ms and 800 ms post onset.Therefore, it is possible that evidence against an effect in the LPC window could be the result of the amplitude averaging procedure rather than the true absence of an effect.
Summing up, with the trained novel words, a behavioural priming effect was observed after a 24-hour consolidation opportunity, however, this priming effect was smaller (7 ms) than that observed with the English words (20 ms) and was accompanied by amplitude modulations at latencies later than those typically associated with the N400 ERP component.Furthermore, on average, the remote novel names were responded to more slowly than the recent novel names.Taken together, these results could indicate that the behavioural effects observed with the trained novel words are more likely to have been subserved by processes rooted in episodic rather than semantic memory.4.3.Main analysis: lexicality effects 4.3.1.Analysis of mean amplitudes in the (predefined) N400 spatiotemporal window Three contrasts were tested in this analysis: (1) nonwords versus the English words, (2) recent novel names versus the nonwords, and (3) remote versus recent novel names.For all three contrasts, as the range of the assumed effect increased, the Bayes factor in favour of the effect of interest decreased (see Table 8).For the contrast between the recent novel names and the nonwords, Bayes factors showed extreme change in evidence in favour of an effect with all four priors, suggesting that amplitudes for recent novel names (−1.69 mV) were, on average, more negative than for nonwords (−0.70 mV).Regarding the contrast between the nonwords and the English words, the nonwords elicited more negative amplitudes (−0.70 mV) than the English words (−0.26 mV).Regarding the contrast between the two sets of novel names, Bayes factors showed anecdotal (when a 1 mV difference was assumed) to moderate (when a 6 mV difference was assumed) change in evidence against a difference between the remote (−1.23 mV) and the recent (−1.69 mV) sets of novel names (see Figure 9).Finally, our exploratory analysis (see Appendix D for a full report) showed that the recent names elicited more negative amplitudes (−1.69 mV) than the English words (−0.26 mV).

Mass univariate analysis
Nonwords vs. English targets preceded by semantically unrelated primes.The nonwords differed from the English words between about 425 ms and 598 ms and then again between 650 ms and 904 ms post onset, with the differences first appearing across a small set of central and parietal electrodes, then extending over the whole scalp and, finally, subsiding to a few central and parietal electrodes (see Figure 10).Between 425 ms and 598 ms post onset, the amplitudes were more negative for the nonwords than for the English words, and, between 650 ms and 904 ms, more positive for the nonwords than for the English words.
Recent novel name targets preceded by semantically unrelated primes vs. nonwords.The novel name targets trained in the recent set differed from the nonwords between about 74 ms and 286 ms post onset across a set of frontal and central electrodes, between 364 ms and 464 ms and between 486 ms and 698 ms post onset across sets of frontal, central, and parietal electrodes, and between 711 ms and 759 ms post onset across a set of frontal electrodes (see Figure 11).The recent novel names elicited more negative amplitudes than the nonwords in all the above-mentioned time windows except for the time window between 486 ms and 698 ms, in which the values for the recent novel names were more positive than for the nonwords.
Remote vs. recent novel name targets preceded by semantically unrelated primes.The remote and recent novel names differed between about 538 ms and 577 ms post onset across a set of frontal and central electrodes, with more negative amplitudes for the remote than the recent novel names, and between 707 ms and 758 ms post onset across a set of frontal, central, and parietal electrodes, with more positive amplitudes for the remote than the recent novel names (see Figure 12).

Lexicality analysis: results summary and discussion
In the literature on visual word recognition, it is assumed that, when a stimulus, either a word or a nonword, is encountered, it activates its lexical neighbours and their semantic representations, with this activity giving rise to negative amplitudes in the N400 spatiotemporal window (Grainger & Jacobs, 1996;Holcomb et al., 2002).As processing continues, if the stimulus is a word, lexical activity associated with it is thought to suppress the activity of its lexical neighbours, leading to a reduction in the N400 response.However, if the stimulus is a nonword, no suppression can occur, resulting in larger (more negative) N400 responses as compared to those for the familiar words (e.g., Bentin, 1987;Bentin et al., 1985;Holcomb et al., 2002).Applying this logic to word learning, shortly after training (when integration is still ongoing), trained novel names could be expected to elicit more positive responses than the nonwords but more negative responses than the familiar words.However, the more the lexical status of the trained novel words approaches that of familiar words, the less negativity in the N400 spatiotemporal window they should exhibit.
In line with previous findings, we found greater negativity in the N400 spatiotemporal window for nonwords than for English words.In addition, the mass univariate analysis also suggested that the amplitudes for the nonwords were more positive than for the English words at later latencies (approx.650-900 ms).
To foreshadow the argument that we develop in the General Discussion, this difference could be due to participants' attempts to recall whether the words they could not immediately recognise (i.e., nonwords or trained novel words that were forgotten) were presented during training.
Regarding our findings with the trained novel names, differences in the N400 spatiotemporal window were observed only between the recent novel names and the English words (see Appendix D) and between the recent novel names and the nonwords.Critically though, while the direction of the effect between the recent novel names and the English words was as expected (greater negativity for the recent novel names), greater positivity was observed for the nonwords relative to the recent novel namesan unexpected effect.Furthermore, compared to the recent novel names, the nonwords also elicited more positivity early (approx.70-290 ms) and late (approx.710-760 ms) in processing, but showed more negative responses in the time window in between (approx.490-700 ms).Interestingly, within this time window (albeit for a shorter period of time), the ERPs for the remote novel names were also more negative than those for the recent novel names, with this effect being reversed later in processing.We explore these findings further in the General Discussion.Note: The graph shows with significant t-values before and after TFCE correction (panel A), topographic maps of amplitudes between 540 ms and 750 ms post onset (panel B; thick black dots correspond to the electrodes with significant t-values after TFCE), and mean amplitudes at the electrode Fz, where the t-values were maximal (panel C; time windows 533-572 ms and 715-730 ms, where the t-values were significant at this particular electrode, are shaded grey).

General discussion
The present study investigated behavioural and electrophysiological markers of integration in learning of novel names for novel concepts.Participants learned two sets of novel names, one set on each of two consecutive days (remote and recent sets), and performed a continuous primed lexical decision task immediately after learning of the recent set.In the priming task, familiar English words were primes, and (semantically related and unrelated) familiar English (Preliminary analysis) and trained novel names (Main analysis) were targets.
To our knowledge, this is the first study to use the continuous primed lexical decision task to assess integration, and the first study to examine the N400 and the LPC components in this task (with both familiar and trained novel names).With the English targets, we found strong evidence for both behavioural and N400 semantic priming effects and weak evidence for a small LPC priming effect.Our study thus both replicates earlier (behavioural) reports in the literature and provides new insights into the neural correlates of behavioural priming.Below, we provide a possible account for our results with the trained novel names.However, we note that not all our results allow for a straightforward interpretation, and therefore our account remains preliminary and there may be other possible interpretations.
The small behavioural semantic priming effect (7 ms) for the novel names trained 24 hours previously (remote set) is in accordance with and extends the results of the previous behavioural studies: van der Ven et al. (2015) reported a 10 ms priming effect after a 24-hour consolidation opportunity, while they found no such effect immediately after training 6 .However, their study used a paired primed lexical decision task which, as we have argued earlier, does not permit the disentanglement of automatic lexical-semantic retrieval processes from controlled processes resulting from the task demands (see also Neely, 1991).Tamminen and Gaskell (2013) observed priming effects of similar magnitudes (8 ms) across different consolidation opportunities (0, 1, and 7 days) with both masked and unmasked standard (paired) primed lexical decision tasks.Tamminen and Gaskell (2013) suspected that the lack of differences between the three consolidation periods was due to low power, however, combining the data from the two tasks did not allow them to test the contrast between the 0day and the 1-day consolidation opportunities and, consequently, their results remain inconclusive with respect to the comparison examined in our experiment.
Consequently, our study extends the previous results by showing that, even when the influence of explicit and strategic processing is minimised in the continuous primed lexical decision task, recognition of trained novel names may be facilitated (behaviourally) by familiar words as early as 24 hours after exposure.It is intriguing that the behavioural priming effects observed in both our study and previous studies are rather small relative to those for familiar words, and it remains unclear whether effects of such magnitude can be considered theoretically meaningful or how they are to be interpreted.It is possible that, had the consolidation period been longer, this effect may have approached those found for the familiar words (20 ms in our study); however, Tamminen and Gaskell (2013) found an effect of a size similar to that reported here even after a 7-day consolidation period.
Likewise, one could argue that our participants have received too little exposure to the novel names as there were only four trials per word-definition pair.However, we note that the training in our study was quite intensive: during each trial, the novel name was presented twice and remained on the screen for a total of 30 s, with the participants being exposed both to the spoken and to the written form of the novel name and also repeating it aloud.In addition, the participants' accuracy on the lexical decision task was around 90%, suggesting that they were able to recognise most of the trained novel names; this accuracy rate is similar to that reported in other studies in the literature (e.g., Bakker et al., 2015;Liu & van Hell, 2020).While the relatively low recall accuracy (30-50% across the recall tasks) is potentially indicative of difficulties retrieving the correct novel name when cued by its definition, it does not necessarily imply that the participants have not learned enough to form mappings between the new concept and the novel word form, or between the novel and the familiar word forms.In fact, Tamminen and Gaskell (2013) argued that the semantic priming effect is independent of explicit recall because the priming tasks tap into more implicit semantic activation (we note, however, that this hypothesis requires further investigation).
It is also important to recognise that most of the earlier studies (e.g., Bakker et al., 2015;Batterink & Neville, 2011;Liu & van Hell, 2020) did not assess the participants' ability to recall the novel names (as opposed to, for example, their ability to recall the definitions of the novel names or their ability to provide English equivalents of the novel names), and it is unclear how well their participants would have performed if they had.Therefore, a more likely interpretation of the smaller behavioural priming effect is that it reflects ongoing integration, but that, 24 hours after exposure, the processing of the novel names is still quite different from that of the familiar words.The fact that this behavioural effect was not accompanied by a modulation of the N400 component, as was the case with the English words, is also in line with this interpretation.Indeed, those previous EEG studies where the experimental design was most similar to ours, also found no N400 priming effects either immediately after, or one day after, or even seven days after exposure (Bakker et al., 2015;Liu & van Hell, 2020, but note that these studies used the semantic-relatedness judgement task to assess integration).
While there are some studies where an N400 priming effect was observed immediately after training (Balass et al., 2010;Borovsky et al., 2012Borovsky et al., , 2013;;Perfetti et al., 2005), it is contentious whether these effects can be ascribed to integration, for several reasons.Firstly, the N400 component is known to be sensitive to task demands, with larger priming effects when instructions explicitly call for semantic analysis (e.g., Chwilla et al., 1995), and can therefore index both automatic and controlled processes of lexical retrieval (Kutas & Federmeier, 2011).Yet, both the paired primed lexical decision task and the semantic-relatedness judgment task are paradigms that, by design, involve controlled semantic processing.Secondly, in at least one of the studies reporting the immediate N400 effect (Perfetti et al., 2005), the primes had been associated with the targets during training, which would have enhanced the contribution of controlled processing to task performance.Finally, in some studies (Borovsky et al., 2012(Borovsky et al., , 2013)), the priming task was interleaved with training, which required the participants to read and make sense of sentences rather than learn the new words that appeared (in the sentence-final position) in these sentences.In a design like this, it is feasible that priming is, at least to some extent, driven by integrative discourse processing which entails activation of semantic and syntactic features of words constituting the sentences, retrieval of world knowledge, and prediction of the upcoming information, among other processes (e.g., Sparks & Rapp, 2010).Because discourse processing is known to modulate both the N400 and the LPC components (e.g., Brouwer et al., 2012;van Berkum, 2009;van Berkum et al., 1999), the presence of the immediate N400 priming effects in Borovsky et al.'s studies cannot be unequivocally attributed to automatic processes of lexical-semantic retrieval and, consequently, their insights regarding integration are uncertain.
To summarise, the results of the priming analysis in our study as well as the findings of other studies capitalising on the priming task suggest that neurocognitive processes that give rise to the behavioural priming effects either immediately after or 24 hours after learning may differ from those that drive such effects for the familiar words 7 .We thus hypothesise that behavioural priming with the remote novel words found in our study could be due to changes in these words' representations in episodic memory and ensuing differences in the words' lexical status.We elucidate this hypothesis further below and start by considering the outcomes of our study regarding the contrast between the recent (0-hour consolidation) and the remote (24hour consolidation) novel names.
Taken together, our findings in the semantic priming analysis suggested that, while there were no differences between the two sets of novel names in the pre-defined N400 spatiotemporal window, the two sets did differ from each other at later stages of processing.A similar pattern of results was obtained in the lexicality analysis, which compared ERPs for the remote and recent novel names preceded by semantically unrelated primes.More specifically, both in the priming and in the lexicality analysis, the recent novel names elicited more positive amplitudes than the remote novel names between about 500 ms and 600 ms post onset and more negative amplitudes between about 700 ms and 760 ms post onset across a set of frontal, central, and parietal electrodes.
Given the temporal and spatial distributions of these effects, one possible account of these differences could be that they index two distinct processes in recognition memory, familiarity and recollection (see Yonelinas, 2002, for a review).While familiarity is thought to reflect the experience of recognising a stimulus as familiar, recollection has been associated with the ability to retrieve the associative information about an event in which the stimulus was encountered (e.g., when and where a particular word was trained).Most models of recognition memory assume that, at retrieval, recollection and familiarity are initiated in parallel, but do not depend on each other (Yonelinas, 2002), rely on different brain regions (e.g., Diana et al., 2007;Eichenbaum et al., 2007), and have different electrophysiological correlates (e.g., Cansino & Trejo-Morales, 2008;Rugg & Curran, 2007;Yu & Rugg, 2010).Familiarity is believed to occur quicker than recollection and has been linked to modulations of the Frontal N400 (or, FN400) component, with familiar items exhibiting more positivity than new items at frontal-central sites between 400 ms and 600 ms post stimulus onset 8 , while better recollection has been associated with an increase in the LPC.
Therefore, in our study, the enhanced positivity for the recent novel names between 500-600 ms post onset could be indicative of higher familiarity for those names compared to the remote novel names.Indeed, since the priming task was administered immediately after learning of the recent set but about 24 hours after learning of the remote set, higher familiarity and better recollection seem likely for the recent compared to the remote set.One could further hypothesise that, for the remote novel names, familiarity may have been too low to enable their discrimination from the nonwords, prompting the participants to attempt to remember whether they had seen these novel names during training.These attempts to retrieve episode-specific information, whether successful or not, could have given rise to (a) more positive amplitudes for the remote novel names between 700-800 ms post onset, and (b) longer response times for the remote novel names observed in our study.This explanation is also in line with reports in the literature suggesting that modulations of the late ERP components that are like the LPC in latency but have a slightly broader distribution (as was the case in our study) may reflect low recognition confidence (e.g., Addante et al., 2012).Nevertheless, while this preliminary interpretation appears feasible, it requires further investigation in the future.
Summing up, we suggest that the differences in the behavioural priming effect observed between the recent and the remote novel names could be due to some transient changes of these words' representations in episodic memory.Considering the differences in processing of the two sets of novel words across both the priming and the lexicality analysis, we hypothesised that these differences could have emerged due to higher familiarity for, and better recollection of, the recent as opposed to the remote novel names.In our view, this interpretation is also in accordance with the remaining results of the lexicality analysis, to which we now turn.
One finding that, in our opinion, may offer an interesting insight in the mechanisms of word learning is that of more negative amplitudes for the recent novel names than for the English words and for the previously unseen nonwords, both in the pre-defined N400 window and beyond it.One possible account of this outcome is as follows.There is growing consensus in the literature that episodic and semantic memories for specific events can co-exist and interact dynamically with each other such that, at any given time point, general knowledge about an event is available together with specific details associated with the original experience of this event (e.g., Winocur & Moscovitch, 2011).It follows that an attempt to retrieve episode-specific details about an event would lead to a reinstatement of this event's representation in semantic memory, and vice versa.In the context of our study, at the time of the administration of the primed lexical decision task, the representations of the recent novel names were likely still highly active in episodic memory (because of the recent exposure).These high levels of activation may have therefore prompted a "search" for a corresponding representation in semantic memory.However, because no lexicalised entries would have been available for the novel names, no semantic memory would have been reinstated, leading to a larger (more negative) N400 response.
Interestingly, in our study, the differences between the recent novel names and the nonwords emerged very early, at about 70 ms post onset, and persisted up to almost 300 ms post onset (which was then followed by the N400 lexicality effect).According to models of psycholinguistic information processing, electrophysiological effects occurring before 250 ms post stimulus onset could index initial lexical, semantic, and syntactic processes which are crucial for unambiguous identification of language elements (see, e.g., Fodor, 1983;Friederici, 2002, for a cascaded model of information processing, and Pulvermüller et al., 2009, for a parallel model of psycholinguistic information access).The literature on visual word recognition lends support for these assumptions: retrieval of lexical-semantic information has been shown to be underway as early as at around 160 ms post word onset (Amsel et al., 2013;Hauk et al., 2012), with differences between familiar words and nonwords emerging already during the N100-P200 ERP complex (Rugg, 1983), while posterior negativity at 200 ms post onset (the so-called N200 component) has been linked with initial orthographic processing (Dehaene et al., 2001;McCandliss et al., 2003).In line with this literature, our finding could indicate that the trained novel names were already distinguished from the nonwords in the earliest stages of processing.
The electrophysiological correlates of the nonwords and the recent novel names also differed later in the processing, between about 480 ms and 700 ms post onset, with the trained novel names showing more positivity across a large set of frontal, central, and parietal electrodes.This effect was followed by enhanced negativity for the trained novel names across a few frontal electrodes between about 710 ms and 760 ms post onset.As we argued earlier in this section, one possible account of these effects could be that they reflect differences in recognition confidence and the ability to retrieve episodic information related to training.Prior to the start of the priming task, our participants were told that they would see familiar English words, novel names trained during the experiment, and words that they might not know.Consequently, having encountered a stimulus they were not familiar with (low recognition confidence), they may have tried to remember if this stimulus had been presented earlier by consciously retrieving
using the R package vwr), number of letters, and bigram frequency (as given in the MCWord database; Medler & Binder, 2008).The novel names were then paired with the descriptions to create four sets of pairs with 20 pairs per set: 1A -Description Set 1 with Novel name Set A; 1B -Description Set 1 with Novel name Set B; 2A -Description Set 2 with Novel name Set A; and 2B -Description Set 2 with Novel name Set B. Four lists were constructed such that there were all possible combinations of sets across the lists: List 1set 1A in Session 1 and set 2B in Session 2; List 2 -1B and 2A; List 3 -2A and 1B; and List 4 -2B and 1A (see Appendix B for the concept description-novel name pairs assigned to each list).The participants were randomly assigned to the lists, with nine participants per list.

Figure 2 .
Figure 2. Example of one trial in the learning phase.

Figure 3 .
Figure 3. Response times for the trained novel name targets.Note: The graph shows probability density, means, medians, and interquartile range for response times (for correct responses) for the semantically related and unrelated novel name targets trained in the recent vs. in the remote set.A small amount of random jitter was added to the raw data points to facilitate figure interpretation.

Figure 4 .
Figure 4. Centro-parietal electrodes: Mean amplitudes for the recent and remote novel name targets.Note: The N400 time window is shaded grey.

Figure 5 .
Figure 5. Frontal electrodes: Mean amplitudes for the recent and remote novel name targets.Note: The LPC time window is shaded grey.

Figure 6 .
Figure 6.Parietal electrodes: Mean amplitudes for the recent and remote novel name targets.Note: The LPC time window is shaded grey.

Figure 7 .
Figure 7. Mass univariate analysis for the contrast between the recent and the remote novel name targets.

Figure 8 .
Figure8.Mass univariate analysis with the trained novel names testing for the effect of semantic relatedness and interaction between semantic relatedness and time of learning.Note: Panel A (effect of semantic relatedness) and Panel B (interaction between time of learning and semantic relatedness) show that, for either predictor, no significant t-values remained after correction with TFCE.

Figure 9 .
Figure 9. Mean amplitude for the nonwords and for the semantically unrelated English and trained novel name targets (from both sets) at centro-parietal electrodes.Note: The N400 time window is shaded grey.

Figure 10 .
Figure10.Mass univariate analysis for the between the nonwords and the English targets.Note: The graph shows significant t-values before and after TFCE correction (panel A), topographic maps of amplitudes between 430 ms and 580 ms and between 700 ms and 880 ms post onset (panel B; thick black dots correspond to the electrodes with significant t-values after TFCE), and mean amplitudes at the electrode C5, where the t-value was maximal (panel C; time windows 442-577 ms and 663-863 ms, where the t-values were significant at this particular electrode, are shaded grey).

Figure 11 .
Figure 11.Mass univariate analysis for the contrast between the recent novel name targets and the nonwords.Note:The graph shows significant t-values before and after TFCE correction (panel A), topographic maps of amplitudes between 150 ms and 750 ms post onset (panel B; thick black dots correspond to the electrodes with significant t-values after TFCE), and mean amplitudes at the electrode FC1, where the t-values were significant in all four time windows (panel C; time windows 74-269 ms, 364-416 ms, 486-581 ms, and 737-759 ms, where the t-values were significant at this particular electrode, are shaded grey).

Figure 12 .
Figure 12.Mass univariate analysis for the contrast between the remote and recent novel name targets.

Table 2 .
Experimental stimuli in one of the lists used in the continuous primed lexical decision task.

Table 3 .
Priors used in the analysis of the behavioural data (log scale).

Table 4 .
Priors used in the Bayesian analysis of the electrophysiological data.

Table 5 .
Response accuracy for targets in the continuous primed lexical decision task.

Table 6 .
Behavioural data sensitivity analysis with the English-novel name prime-target pairs.

Table 7 .
EEG data sensitivity analysis with the English-novel name prime-target pairs.