Tracking the acquisition and retention of novel word representations: an ERP study

ABSTRACT The event-related N400 provides a measure of lexical-semantic processing, indexing the encoding strength of newly learned words. We examined N400 modulations following associative learning of novel written words paired with novel object pictures from early to later stages of acquisition. During the initial learning session, novel picture-word pairs accompanied by semantic attributes were presented as stimuli for learning at multiple timepoints. Following learning at each of these timepoints, we examined behavioural and neural effects for recognition of familiar and novel words, as well as at a follow-up session several days later. Rapid behavioural learning was accompanied by N400 congruity effects for novel words, displaying a similar neurophysiological profile to familiar words. These behavioural and neural effects were largely sustained several days later in the follow-up session. Our findings indicate that explicit learning of novel object-word associations leads to the rapid formation and longer-term retention of new word memory traces.


Introduction
The human brain is a dynamic system capable of not only storing and retrieving existing representations of familiar words, but is also adept at encoding new representations of novel words.Learning a new word is suggested to take place over different time scales, ranging from the rapid hippocampal encoding of a novel word form to the slower neocortical consolidation of a new memory trace (Coutanche & Thompson-Schill, 2014;Davis & Gaskell, 2009;McClelland et al., 1995;Sharon et al., 2011).This slower consolidation process, which likely involves a period of offline processing (i.e.sleep; Davis & Gaskell, 2009;Havas et al., 2018;Schimke et al., 2021;Tamminen & Gaskell, 2013), is considered to be crucial for binding new word memory traces to existing semantic representations.
Capitalising on the superior temporal resolution of electroencephalography (EEG), event-related potentials (ERPs)particularly the N400 componentprovide a reliable measure for investigating the neural dynamics of word learning (for reviews see: Kutas & Federmeier, 2000, 2011).The N400 is a negative-going neural response peaking near 400 ms poststimulus onset, displaying broadly distributed topography typically maximal over centroparietal sites for familiar written words.The amplitude of the N400 is particularly sensitive to semantic incongruity (Kutas & Hillyard, 1980), with increasing negativity reflecting greater processing demands.For instance, a larger (i.e. more negative) N400 response is observed when a target word (lion) is preceded by an incongruous word (plate) relative to a congruous word (tiger).Although the specific function/s of the N400 component remain the subject of much debate (see : Hagoort, 2008;Rabovsky et al., 2018), for the most part it is agreed that the N400 provides a neural marker of information retrieval from lexical-semantic (Brouwer et al., 2017;Delogu et al., 2019;Kutas & Federmeier, 2000, 2011) and episodic memory (Elmer et al., 2022;Rodríguez-Fornells et al., 2009).
Substantial evidence indicates that the N400 is sensitive to various forms of word learning, particularly those involving explicit associations of new words (i.e.novel or unfamiliar words) with meaningful objects and/or descriptions (Angwin, Armstrong, Fisher, & Escudero, 2022;Perfetti et al., 2005;Balass et al., 2010;Kuipers et al., 2017), and the acquisition of new words from semantic sentence contexts (Batterink & Neville, 2011;Borovsky et al., 2010Borovsky et al., , 2012Borovsky et al., , 2013;;Chen et al., 2014;Liu et al., 2019;Mestres-Missé et al., 2007;Zhang et al., 2019).Research on associative word learning shows new words can be rapidly acquired when paired with semantic information, as evidenced by N400 congruity effects arising to new written words immediately after learning (e.g.Perfetti et al., 2005).Similar outcomes have been observed after less than a handful of contextual learning exposures (Frishkoff et al., 2010;Mestres-Missé et al., 2007), and even following a single exposure to a novel word within a constrained semantic sentence (Borovsky et al., 2010(Borovsky et al., , 2012(Borovsky et al., , 2013)).Such findings provide insight into the remarkable speed with which novel word memory traces can become associated with existing semantic schema.Together, this evidence suggests that novel word learning takes place rapidly, with exposure to meaningful information during word learning facilitating lexical-semantic processing.
The studies discussed so far have predominantly tested novel words immediately following learning within a single session (e.g.Perfetti et al., 2005).Yet, the acquisition of a new word is suggested to be a temporally protracted process, involving overlapping stages of rapid encoding followed by slower consolidation (Davis & Gaskell, 2009;Havas et al., 2018).Consistent with accounts of post-learning consolidation, studies on lexical-semantic integration have shown N400 congruity effects for meaningful novel words 48 h after, though not immediately after, their acquisition from semantic sentence contexts (Frishkoff et al., 2010), and their association with novel pictures and semantic attributes (Kaczer et al., 2018).Conversely, other studies using similar stimuli have not detected N400 congruity effects for novel written words, neither immediately or 24 h post learning (Bakker et al., 2015;Lei et al., 2022;Liu & van Hell, 2020), or even seven days after learning (Lei et al., 2022;Liu & van Hell, 2020).The absence of N400 effects in these instances might be due to the type and/or amount of semantic information that accompanied new words during learning (Laine & Salmelin, 2010).Indeed, too much semantic information has been suggested to increase cognitive demands and impede learning (Takashima et al., 2014;Whiting et al., 2007), while irrelevant or impoverished semantics are less likely to benefit learning (Cornelissen et al., 2004).
Behavioural findings from new word learning studies that both did and did not detect N400 congruity effects (Bakker et al., 2015;Kaczer et al., 2018;Lei et al., 2022;Liu & van Hell, 2020) demonstrated that behavioural accuracy consistently exceeded chance, indicating that knowledge of new words is predominantly retained over time.Improved retention has also been shown for new words learned with both semantic definitions and novel objects relative to learning words with semantic definitions alone (Lei et al., 2022; see also : Takashima et al., 2014).New words paired with existing object pictures were also better retained following explicit associative learning relative to contextually-driven fast mapping (Coutanche & Thompson-Schill, 2014;Merhav et al., 2015;Zaiser et al., 2022).Despite N400 results providing mixed support for lexical-semantic integration, behavioural findings suggest repeated learning of new picture-word pairs accompanied by semantic information (existing or accompanying) provides a benefit for encoding, leading to greater retention of new object-word associations over time.
A number of word learning studies have reported N400 congruity effects for recognition of novel words preceded by familiar (Bermudez-Margaretto et al., 2018;Dittinger et al., 2016;Elmer et al., 2021) or novel objects (Angwin et al., 2014;Ramos-Escobar et al., 2021).In research by Angwin et al. (2014), healthy adults learned novel written words paired with novel object pictures, accompanied by either two written adjectives (serving as semantic attributes e.g.Awful Fickle), or two unfamiliar nouns (e.g.Mansey Smeath).Immediately following learning on the final day, an N400 congruity effect arose for correctly recognised picture names learned with semantic attributes, with no such effect for novel words learned without accompanying semantics.Although recognition accuracy displayed consistent improvements over time, words learned with semantics displayed an accuracy advantage over words learned without semantics.Bermudez-Margaretto et al. (2018) reported similar outcomes, noting larger reductions in N400 responses to novel written words preceded by a familiar object as opposed to a non-semantic hashtag symbol.Together, these findings suggest that exposure to semantic information during new word learning enhances the depth of encoding of new object-word associations, facilitating the formation of stronger new word representations in memory.However, measuring N400s at the start and end of a word-learning session (Bermudez-Margaretto et al., 2018), or immediately following learning on the final day (Angwin et al., 2014), limits conclusions about the impact of semantic information on post-learning consolidation, and the longer-term retention of new object-word associations in memory.
To better understand the behavioural and neurophysiological mechanisms underlying associative word learning, research should further examine the timecourse of the N400 congruity effect following explicit encoding of novel written-word names for novel objects.Such research should not only examine behavioural and neurphysiological effects within the learning session, but should also assess these effects several days after learning to qualify the post-consolidation durability and retention of new word memory traces over time.Since learning new associations between entirely novel words and objects is considered a demanding form of word learning (Laine & Salmelin, 2010), novel picture-name pairs should be accompanied by unique semantic attributes to optimise encoding efficiency.It is crucial, however, that the semantic attributes only appear together with the novel word stimuli during learning; by excluding the semantic attributes from the post-learning recognition task, any N400 effects that arise to correctly recognised novel words cannot be attributed to the activation of pre-existing semantic schema.This approach has previously been suggested to facilitate lexical-semantic processing and the formation of stronger new word memory traces (Angwin et al., 2014;Bermudez-Margaretto et al., 2018;Kaczer et al., 2018), as indexed by N400 congruity effects arising to novel picture names.
Expanding on previous research by Angwin et al. (2014), the present ERP study further investigated the spatiotemporal dynamics of memory formation and retention of meaningful object-word associations in healthy young adults.Our primary aim was to examine changes in N400 congruity effects following repeated learning of explicit associations between novel written words and novel object pictures from early to later stages of the learning session.A secondary aim was to assess whether behavioural learning and neural effects were retained for new word-object pairs several days after learning.To achieve these aims, novel written words paired with novel object pictures accompanied by distinct semantic attributes were presented as stimuli for learning at multiple timepoints within a single learning session.At each of these timepoints, we examined behavioural learning and N400 congruity effects for recognition of familiar and novel picture names, as well as at a follow-up session several days later.
Our principle hypothesis was that N400 effects would rapidly arise for novel picture names, with greater negativity to novel written words preceded by incongruous relative to congruous novel pictures.Based on previous word learning studies (Bermudez-Margaretto et al., 2018;Elmer et al., 2022;Ramos-Escobar et al., 2021), novel words were expected to show a frontally distributed N400 modulation during the early stages of learning.As learning exposures increase, improved recognition accuracy and response speed are expected to be accompanied by a shift in N400 topography, with a right-hemisphere/centroparietal distribution emerging once representations of new word-object pairs had stabilised.To explore whether initial word learning was consolidated, picture-name recognition was assessed several days after learning, where we expected behavioural and N400 effects to be predominantly retained.

Participants
Thirty-nine healthy young adults (10 males; aged 18-35 years, M = 23.46,SD = 4.86) volunteered to participate in this study by responding to an advertisement placed on the University of Queensland Psychology Research Participation Scheme (SONA) website.All participants were native or fluent English speakers, had normal vision, no history of neurological damage or disease, and were right-handed (laterality M = 0.85, SD = 0.20; Edinburgh Handedness Inventory, Oldfield, 1971).They each provided written informed consent and were reimbursed $10.00AU per 30 min of participation.To encourage motivation and attention during learning, participants received a bonus of $1.00AU for each novel word correctly recognised in the final block of the learning session (maximum of $40.00AU).Details about the bonus were provided at the start of the learning session, with participants receiving their bonus when they returned for the follow-up session (details on the instructions provided to participants are included in the Supplementary Material).This study was approved by The University of Queensland Human Research Ethics Committee (Ethics Approval Number: 201000199) and performed in accordance with the principles stated in the Declaration of Helsinki.

Stimuli
Stimuli were obtained from Angwin et al. (2014;see also: Angwin et al., 2017;Schimke et al., 2023), comprising 40 novel object pictures (space alien drawings; Gupta et al., 2004) each paired with a novel written word and two familiar written adjectives.Briefly, the 40 novel words, which adhered to English spelling conventions, were created using the Australian Research Council Nonword Database (Marian et al., 2012; descriptive statistics for psycholinguistic variables can be found within the Supplementary Materials).The 80 familiar English adjectives (non-visual modality) were obtained from the Medical Research Council Psycholinguistic Database (Wilson, 1988).Adjectives were randomised into 40 semantic attribute pairs, which were re-paired if the adjectives had opposing meanings (e.g.Happy Gloomy).Each novel word was assigned to a semantic attribute pair to create 40 unique three-word names (e.g.Blags Happy Gloomy); and each three-word name was paired with one of the 40 alien pictures. 1In addition to these 40 novel picture-name stimulus items, 40 blackand-white line drawings of familiar objects together with corresponding written word names were selected from a standardised set (Snodgrass & Vanderwart, 1980).An independent t-test showed no significant difference in word length between novel words (M = 5.23, SD = 0.62) and familiar words (M = 5.40, SD = 1.60; p = .517).

Tasks and procedures
This study was conducted over two sessions, 3-5 days apart (M = 3.81, SD = 0.97).The first session comprised six blocks (Figure 1), each containing a novel picturename learning phase, followed by a picture-name recognition task (to assess learning of the novel written words), and a semantic attribute recognition task (to assess learning of the familiar written adjectives).The second session contained only two blocks of picturename recognition trials (to assess longer-term retention).Participants were tested individually.They were seated in a comfortable chair behind a desk with their eyes level approximately 60 cm from the centre of the computer screen.E-Prime 2.0 software (Psychology Software Tools Pittsburgh, PA, USA) was used to present all stimuli and collect behavioural data.

Word learning phase
The 40 novel object-word (i.e.alien picture-name) pairs were presented one at a time on the computer screen (Figure 1B).Each pair appeared once during the learning phase of each block, resulting in a total of six novel word learning exposures.In each learning trial, an alien picture together with its novel written-word name and two written semantic attributes was displayed for 5000 ms, followed by a blank screen for 500 ms, after which the next learning trial started automatically.Participants were instructed to remain still and covertly read and learn the alien pictures and names as best as they could.

Picture-name recognition task
Following each learning phase, a picture-name recognition task was conducted to assess participants' knowledge of the alien names (Figure 1C).In this task, half of the alien pictures remained paired with their correct novel word name from the learning phase (novel-congruous trials, e.g.Zealt-Zealt), while the remaining aliens were re-paired with an incorrect name, one that corresponded to a different alien during learning (novel-incongruous trials, e.g.Glocs-Zonts).Likewise, half of the familiar objects were paired with their correct name (familiar-congruous trials, e.g.Candle-Candle), while the remaining familiar objects were repaired with the name of a different familiar object (familiar-incongruous trials, e.g.Ant-Watch) from the same set of stimuli (Snodgrass & Vanderwart, 1980).To reduce repetition of identical stimulus items, picturename pairings were altered on a rolling basis using a quasi-Latin square (Yates, 1937; details of the picturename pairing approach can be found within the Supplementary Materials).In the follow-up session, two blocks of picture-name recognition trials were presented, following the same timing and structure as session 1, but with no preceding learning phase or subsequent semantic recognition task.Picture-name recognition trials started with a fixation cross for 500 ms, followed by a picture (familiar object or novel alien) for 1000 ms.After a 500 ms blank screen interval, a single written word (familiar object or novel alien name) was displayed on the computer screen.Participants had 2000 ms to decide whether the written word was congruous or incongruous with the preceding picture, pressing either the left or right button to enter their response, respectively (Figure 1E).Upon making a response, or after 2000 ms had lapsed, a 2000 ms blank screen appeared to signify the end of the trial.The next recognition trial started automatically.After completing the picture-name recognition task in the second session, participants performed some additional language and learning-related tasks.Details of the methods and results for these additional tasks are not reported since they were conducted as part of separate experiments.

Semantic attribute recognition task
Following the picture-name recognition task in each block of session 1, participants performed a brief semantic attribute recognition task to assess their knowledge of the familiar adjective pairs that accompanied the novel aliens and their novel word names during learning (Figure 1D).Eight alien picture-attribute pairs were presented in each block, giving 48 semantic trials in total.Each alien-attribute pair appeared at least once, with eight of these pairs selected to appear for a second time.This approach ensured that an equal number of picture-attribute trials was presented in each block.Like the picture-name recognition trials, half of the alien pictures remained paired with their correct semantic attributes from learning (semantic-congruous trials, e.g.Glocs Itchy Bleak), while the remaining aliens were re-paired with incorrect semantic attributes, corresponding to a different alien picture from the learning phase (semantic-incongruous trials, e.g.Zealt Happy Loud).The semantic attribute trials followed the same design as the picture-name trials, with participants responding via left or right button-press to indicate whether the semantic attributes were congruous or incongruous with the preceding alien picture, respectively.Brief rest breaks were provided to participants at the end of each block.

EEG data acquisition and processing
Continuous EEG was recorded from scalp-surface electrodes connected to a 129-channel HD-EEG system Hz (Electrical Geodesics, Inc; Figure 1F) with a sampling rate of 500 Hz.The vertex electrode served as the online reference and impedance was kept below 50 kΩ, which is acceptable when using high-impedance amplifiers (Ferree et al., 2001).The EEG data were acquired using Netstation 4.4.2software (Electrical Geodesics, Inc) on an Apple iMac computer running Mac OS X 10.6 (Apple, Inc).Six participants' data were excluded: three participants did not attend the second session, two had excessive artifacts in their recordings, and one participant did not follow the task instructions (pressed the same button for all responses).The final sample was n = 33.
EEG pre-processing was performed using EEGLAB v2021.0.0 (Delorme & Makeig, 2004) and ERPLAB v9.00 (Lopez-Calderon & Luck, 2014) running in MATLAB (Version 2019a, MATHWORX).Raw data were imported into EEGLAB, down-sampled to 250 Hz, and band-pass filtered at 0.1-30 Hz.Using the EEGLAB plugin clean_rawdata (Kothe & Makeig, 2013), bad channels (correlation with surrounding channels < 0.75) were rejected (M = 13.94,SD = 3.72, <20% of channels) and high variance artifactsclassified using 0.25 s sliding-window principal components analysis (PCA) on continuous datawere corrected (PCs ≤ 20 SD from clean recording calibration data; e.g.small head movements during tasks) or rejected (PCs > 20 SD from calibration data, e.g.large shoulder movements during breaks) using artifact subspace reconstruction (see also Chang et al., 2018Chang et al., , 2019)).Rejected channels were interpolated using spherical spline interpolation (to calculate an unbiased average reference) and the data were re-referenced to the average signal recorded from all electrodes.Rank-adjusted data (to account for interpolationinduced rank reduction) were decomposed using adaptive mixture independent components analysis (ICA; Hsu et al., 2018), followed by probabilistic labelling and rejection of artifactual ICs (M = 11.40,SD = 1.94; e.g.eye blinks, head movements, heartbeats; Pion-Tonachini et al., 2019).Trials were segmented into epochs (−100 to 1000 ms) and baseline corrected (−100 to 0 ms) relative to picture name (i.e.word target) onset.Trials corresponding to behavioural errors, correct responses made <200 ms post word onset, and non-responses were removed.The resulting epochs were visually inspected to confirm and exclude any remaining bad trials.Overall, <25% of trials were excluded from the analysis, with the larger proportion of these exclusions corresponding to behavioural errors/non-responses.
To ensure there was a sufficient number of trials in each condition for analysing ERPs, picture-name recognition trials were combined as follows: block 1 (first and second block of session 1), block 2 (third and fourth block of session 1), block 3 (fifth and sixth block of session 1), session 2 (first and second block of

Statistical analysis
Statistical analyses were conducted using SPSS (Version 27; IBM, 2020) and violations of sphericity were corrected using Greenhouse and Geisser (1959;ε < .75)or Huynh and Feldt (1976;ε ≥ .75)adjusted p values.For both the picture-name and semantic attribute recognition tasks, behavioural measures of recognition accuracy (mean percentage of correct responses to word targets) and reaction time (RT; mean number of milliseconds from word target onset for correct responses) were analysed separately using repeated measures analysis of variance (rmANOVA) with factors session (1, 2), block (1, 2, 3), lexicality (familiar, novel), and congruity (congruous, incongruous).Post-hoc analyses of simple effects were conducted using rmANOVAs or paired t-tests, with multiple comparisons corrected using the Holm-Bonferonni method (Holm, 1979).

Results
Results are first presented for behavioural measures of recognition accuracy and response speed, followed by ERP measures of N400 congruity effects and N400 difference wave amplitudes.All descriptive statistics for the behavioural results are reported as the mean or mean difference (M Diff ) ± 1 standard error (SE), rounded to the nearest percentage.

Picture name reaction time
To examine participants' response speed for correctly recognised picture names (Figure 2B), mean RTs to written word targets in session 1 were analysed using a 3 (block) × 2 (lexicality) × 2 (congruity) rmANOVA.A significant main effect was found for congruity To further assess participants' response speed for correctly recognised new picture names, RTs to novel word targets in blocks 1 and 3 of session 1 were compared to session 2 using separate 2 (session) × 2 (congruity) rmA-NOVAs.Only significant main and interaction effects involving session are reported.For the analysis of block 1 and session 2, no significant main or interaction effects involving session were revealed (all p ≥ .152)For block 3 and session 2, there was a significant main effect of session [F(1, 32) = 95.48,p < .001,η p 2 = .75],indicative of the overall slowdown in RTs from the end of session 1 to session 2 (M Diff = −121 ms ± 12).There was also a significant interaction of congruity × session [F(1, 32) = 4.67, p = .038,η p 2 = .13],confirming the attenuation of the response speed advantage for congruous relative to incongruous novel words from the end of session 1 (M Diff = 43 ms ± 16; t(32) = 2.62, p = .013,d = .20)to session 2 (M Diff = 12 ms ± 20; p = .559).

Electrophysiological results
To examine the time course of N400 modulations to newly learned picture names in session 1, mean  amplitudes to correctly recognised word targets were analysed using an omnibus 3 (block) × 2 (lexicality) × 2 (congruity) × 3 (laterality) × 3 (anteriority) rmANOVA.To simplify results, only significant main and interaction effects involving congruity are reported (for complete statistical results see Table 2).
This three-way interaction was further explored using 2 (congruity) × 3 (laterality) rmANOVAs conducted separately on familiar and novel words.Analysis of ERP amplitudes to familiar picture names in session 2 (Figure 3) revealed a significant main effect of congruity [F(1, 32)
To more closely examine changes in N400 magnitudes for novel words between sessions, difference waves over the midline and right hemisphere in the follow-up session were compared to the first block (right hemisphere only) and the last block (midline and right hemisphere) using separate paired t-tests (Figure 5).The N400 magnitude over the right hemisphere was significantly more negative for session 2 compared to block 1 [t(32) = 2.15, p = .040,d = .49],with no significant difference between session 2 and block 3 (p = .093).The N400 magnitude over the midline was significantly more negative for block 3 compared to session 2 [t(32) = 2.89, p = .007,d = .37].

Discussion
The present study examined spatiotemporal modulations in neurophysiological memory representations for explicit learning of new written words in healthy young adults.Specifically, we assessed how repeated exposure to novel picture-name associations learned with accompanying semantic attributes subsequently modulated the N400 component, an index of lexicalsemantic processing.Novel written-word names with novel object pictures and unique familiar adjective pairs were presented as stimuli for learning across multiple timepoints during an initial learning session.Following learning at each of these timepoints, we assessed behavioural accuracy and N400 amplitudes to congruous and incongruous familiar and novel picture name targets, as well as at a delayed follow-up session several days later.We expected that learning repetition and explicit exposure to novel picture-name stimuli would lead to improved recognition accuracy and response speed for newly learned written word targets.We predicted that N400 congruity effects would rapidly arise for these newly learned words, displaying a more frontally distributed modulation during the early stages of learning, shifting to a right-hemisphere centroparietal distribution once representations of the new words were stabilised.

Behavioural findings
Familiar words displayed superior recognition accuracy and response speed relative to the novel words during both the initial learning and delayed follow-up sessions.This outcome was not surprising given that familiar objects are often encountered in day-to-day life, while the novel objects needed to be learned and matched to their associated novel word names.The advantage of including familiar words with established lexicalsemantic representations in word learning and retrieval tasks has been well documented by previous behavioural (Grainger, 1990;Tamminen & Gaskell, 2013;Zaiser et al., 2022) and neuroimaging research (Lei scalp regions of interest in blocks 1-3 of of the initial learning session (session one) and 3-5 days later in the delayed follow-up (session two).In each difference wave plot, the black vertical line representing the y-axis is aligned with the onset of the written word target.The shaded grey rectangle represents the time window (247-457 ms) during which the N400 congruity effect was measured.The blue (congruous) and red (incongruous) shading along the traces represents ±1SE.For illustrative purposes, difference waves were low-pass filtered at 15 Hz.(E) Difference wave ERP measurement scale for amplitude (y-axis, µV) and time (x-axis, ms).(F) 129-channel EEG electrode montage with nine electrode cluster regions of interest shaded in green.(G) Series of scalp topology plots showing the topographical distribution of the N400 difference wave distributions (familiar minus novel).et al., 2022;Mestres-Missé et al., 2007, 2008;Takashima et al., 2017).
As expected, associative learning of novel picture names led to consistent improvements in recognition accuracy throughout the initial session.During the early stages of word learning, accuracy tends to be lower since synaptic connections encoding new object-word associations are initially weak (Brown et al., 2007).As learning repetition increases, memory representations of new associations are strengthened (Friedman & Johnson, 2000;Laine & Salmelin, 2010), enabling more accurate recognition of novel word targets over time.Continually reinforcing picture-name associations during learning is suggested to increasingly activate more connections within semantic networks (Craik, 2002;Tibon et al., 2017; see also : Hebb, 2005), which may account for the accuracy advantage shown to congruous picture names.Several days after learning, a small reduction in novel word accuracy was observed in the follow-up session, likely due to some decay and/ or degradation in the memory representations of new object-word associations over time (Bahrick, 1984;Gathercole, 2006;Loftus & Palmer, 1974).Despite this reduction, accuracy was greater for novel word recognition at the follow-up relative to the first block of the learning session.
Our findings indicate that participants successfully retained most of the 40 novel picture names.Remarkably, successful acquisition was evident midway through the initial session, where after just four picture-name learning exposures, recognition accuracy (75%) surpassed the chance accuracy level (50%).Earlier studies have also used picture-word recognition to assess learning of explicit associations between new words and existing object pictures (Breitenstein et al., 2007;Dittinger et al., 2016;Shtyrov et al., 2021), or new words paired with novel objects and semantic information (Angwin et al., 2014;Whiting et al., 2007;Zaiser et al., 2022).Such studies report word learning success following a relatively larger number of repeated learning exposures (e.g.≥10; Dittinger et al., 2016;Shtyrov et al., 2021), and/or over multiple learning sessions (Angwin et al., 2014;Breitenstein et al., 2007;Whiting et al., 2007).Using a similarly large set of novel word stimuli, our results also show that learning was successful, but with fewer learning exposures.Together with existing findings, as well as those from other studies of semantically-driven word learning (Lei et al., 2022;Mestres-Missé et al., 2007, 2008;Frishkoff et al., 2010), the present findings support conclusions suggesting that repeated learning with semantic exposure provides a benefit for explicit encoding of novel picture-name associations.
Regarding our results for novel word response speed, increased learning led to faster recognition of new picture names throughout the initial learning session.At the start of the learning session, where picturename pairs were repeated just twice (i.e.block 1), slower response times suggested weaker encoding.This suggests that successful recognition may have involved more effortful top-down strategic processing (Laine & Salmelin, 2010;Rodríguez-Fornells et al., 2009).Together with the similar response times for congruous and incongruous novel picture names, findings from the first block of the learning session suggest the involvement of a retrospective retrieval mechanism (Keatley & Gelder, 1992;Kahan, 2000;Smith, 2003), enabling post-lexical congruity checking of whether a novel name target corresponded to a novel picture cue.During initial learning, participants may have relied on retrospective congruity checking for recognition of weaker object-word associations, with increased recognition demands accounting for the slower response times to novel words.
As word learning progresses and memory traces strengthen, the need for retrospective checking of new picture-name associations diminishes.This shift makes way for less effortful (Laine & Salmelin, 2010) and more autonomised processing (Chein & Schneider, 2012), enabling faster recognition of novel words during later stages of learning.Faster recognition is consistent with a forward-acting anticipatory mechanism (Tzelgov & Eben-Ezra, 1992), where provided the encoding strength of a picture-name association is sufficient, exposure to the novel object cue prospectively activates the corresponding word target (Hutchison et al., 2013).That is, upon participants encountering a novel object picture, the associated novel name is pre-activated in anticipation of the upcoming written word.If a congruous name appears, less activation is required, resulting in a faster response than if an incongruous novel word appeared instead.Not only does this anticipatory preactivation account for increases in recognition speed throughout the learning session, but it may also explain the speed advantage for congruous novel words that arose during the latter blocks.Since the pre-activation of the target depends on the strength of its association with the cue, the weaker picture-name associations at the start of learning were unlikely to have been sufficiently encoded to elicit an anticipatory response.
Several days after learning at the follow-up session, a large decrease in recognition speed was observed for novel words.Specifically, a reduction in the speed advantage was shown for congruous picture names as well as overall slower response times that were similar to novel words recognised at the start of learning.Similar research by Dittinger et al. (2016) also reported a reduction in response speed for new picture names during a delayed follow-up, where the speed advantage for congruous picture names was abolished and response times appeared to show little difference from novel words during initial learning.Whiting et al. (2007) similarly showed slower responses to novel picture names recognised at both one-week and onemonth post learning, albeit response times at both follow-ups were faster than initial learning.Notably, both the current research and the study by Dittinger et al. involved training participants on new pictureword pairs in a single session, while Whiting et al. delivered training across multiple sessions.Despite the potential benefit of associative word-object learning over several sessions (Whiting et al., 2007), the results from the earlier studies as well as those reported here highlight a pronounced slowdown in post-learning recognition speed.These response time findings appear to suggest that recognition processing for novel words reverts to levels observed at the initial stages of word learning.
Following a post-learning interval of several days, new object-word memory associations are suggested to undergo a period of consolidation (Davis & Gaskell, 2009;see also: McClelland et al., 1995), during which episodic word traces decay and semantic representations are strengthened.Despite the present findings showing a substantial reduction in post-learning recognition speed, relatively well-retained behavioural accuracy for novel picture names suggests that recognition processing at the follow-up may have prioritised memory integrity over retrieval speed.The neural pathways initially used during rapid encoding of transient episodic word traces may not be conducive to retrieving consolidated representations of established object-word associations (Tamminen & Gaskell, 2013).Consequently, recognition of novel words retained in a long-term store might be processed using slower, more stable and efficient neural pathways.This explanation can account for the overall reduction in post-learning response times for novel words.However, the lack of a response speed advantage for congruous novel picture names in the follow-up session suggests a post-learning shift in recognition processing, from less effortful prospective target activation (Hutchison et al., 2013;Tzelgov & Eben-Ezra, 1992) to more-strategic retrospective congruity checking (Keatley & Gelder, 1992;Smith, 2003).Irrespective of the underlying mechanism/s driving postlearning retrieval processing, correct picture-name recognition would have been unattainable had learning of the object-word associations not been successful.
The semantic attribute recognition task primarily served as a control to ensure that participants attended to the semantic attributes accompanying novel picturename pairs during learning.Semantic recognition accuracy and response speed improved from the first to the second block, with little change from the second to the third block.These results parallel other studies (Angwin et al., 2014;Kaczer et al., 2018), suggesting that exposure to distinct, written semantic attributes (e.g.adjectivenoun or adjective-adjective pairs) may provide an initial benefit for encoding explicit associations between novel written words and novel object pictures.Overall, despite the smaller number of semantic recognition trials limiting the strength of our conclusions for semantic learning, at the very least these findings suggest that participants were attending to the semantic attributes accompanying novel words and objects during initial learning.

Neurophysiological findings
To examine neural responses for repeated learning of new object-word associations, we examined ERP amplitudes to correctly recognised congruous and incongruous familiar and novel picture names from initial encoding to later retention.Consistent with findings from previous studies (Angwin et al., 2014;Bakker et al., 2015;Kaczer et al., 2018;Lei et al., 2022), N400 congruity effects arose to familiar word targets during both the initial learning and the delayed follow-up sessions.The N400 topographies displayed broad distributions that were overall larger at centroparietal sites, with a right-hemisphere bias observed at the start of learning and again at the follow-up.These scalp-level effects are consistent with N400 morphology for familiar written words with established lexical-semantic representations (Kutas & Federmeier, 2011).
Following learning at the start of session 1, recognition of congruous and incongruous novel words elicited similar amplitudes, indicating that neural representations of new object-word associations were not sufficiently encoded.Considering each new picture-name pair was only encountered twice during the first block of the learning session, the absence of an N400 congruity effect for novel words was not particularly surprising.While some evidence suggests that a single exposure to a novel word is sufficient to elicit an N400 effect (Borovsky et al., 2010(Borovsky et al., , 2012)), such N400 effects arose following contextual learning of novel words from highly constrained semantic sentences.In contrast to our study, which involved learning novel names for novel objects with accompanying semantic attributes, using highly constrained meaningful sentences to promote new word learning likely provided a stronger semantic context, facilitating deeper encoding of the novel written words.
From midway through the learning session, N400 congruity effects were shown for correctly recognised novel written words, suggesting that three to four learning exposures (as per block 2) was sufficient for encoding of new picture-name associations.Other studies have also found N400 modulations for novel words following relatively few learning exposures (Mestres-Missé et al., 2007;Ramos-Escobar et al., 2021).However, in contrast to the broadly distributed N400s displayed here, the previous studies showed more frontal/frontocentrally distributed N400 modulations for novel words.According to Mestres-Missé et al. (2007), since encoding of new written words is initially weak, greater cognitive control is required to facilitate their retrieval.Following the learning of novel word meanings from semantic sentence contexts, Mestres-Missé et al. showed via current source density mapping that N400 congruity effects for familiar word pairs originated from a temporal region, while N400s for novel-familiar pairs were generated within a frontal source.Novel object-word associations learned without (pre-existing or accompanying) semantic information have also been shown to elicit a larger N400 over frontal regions (Ramos-Escobar et al., 2021), suggesting that effortful processing during early stages of learning (i.e.self-generated semantic cues; see: Grönholm et al., 2007) also increases cognitive demands.Accordingly, a frontally distributed N400 may reflect the involvement of a generalised frontal cognitive control mechanism (Mestres-Missé et al., 2007, 2008), engaged to support the short-term maintenance of new information during effortful word learning and retrieval (Ramos-Escobar et al., 2021;Rodríguez-Fornells et al., 2009).Since our research involved learning explicit picture-name associations accompanied by distinct semantic attributes, the absence of a frontal N400 during the early learning stages may suggest that learning was comparatively less demanding.
A more frontally distributed N400 does not always arise to newly acquired written words, particularly when representations of new object-word associations are thought to be more established (Elmer et al., 2022), such as during later stages of word learning.Similar to the right-hemisphere N400 often shown to familiar written words (Angwin et al., 2022;Hurley et al., 2009;Kutas & Federmeier, 2011), several studies have also reported larger right-hemisphere N400s for novel words (Angwin et al., 2014;Batterink & Neville, 2011;Perfetti et al., 2005).Although the present N400 congruity effects displayed broader distributions during the learning session, one interesting observation was that the N400 magnitudes showed similar negativity for familiar and novel picture names, particularly over medial/ right-hemisphere and centroparietal sites.Other studies have also reported comparable N400 magnitudes for familiar and novel words (Angwin et al., 2014;Batterink & Neville, 2011; see also Mestres-Missé et al., 2007), with such findings interpreted to suggest that newly formed picture-word associations are processed similarly to established familiar object-name representations.
Our results from the initial learning session are consistent with findings of fast neuroplastic changes during the encoding of new word-object associations (Dittinger et al., 2016;Ramos-Escobar et al., 2021), evidenced by N400 congruity effects arising to novel words after less than a handful of explicit learning exposures.These N400 effects suggest that semantic access was reliably taking place during the early stages of new word learning, supporting claims from previous studies that semantic information has a beneficial impact on the learning of novel names for novel objects (Angwin et al., 2014;Bermudez-Margaretto et al., 2018;Dittinger et al., 2016;Elmer et al., 2022).Together with our behavioural findings, the neurophysiological results from our initial learning session suggest that the acquisition of novel written words was driven by a rapid encoding memory mechanism.Given the initial learning session involved repeated exposure and subsequent recognition of explicit wordobject associations, the observed behavioural and electrocortical modulations appear to be consistent with processing that is typically attributed to the underlying episodic hippocampal system (Davis & Gaskell, 2009;Tamminen & Gaskell, 2013).
Measuring ERPs following a post-learning delay of several days allowed the present study to assess whether neural representations of new picture-word associations were retained over time.This approach provides an advantage over previous N400 studies that assessed learning within a single session (Bermudez-Margaretto et al., 2018;Dittinger et al., 2016), or over multiple sessions without a post-learning delay (i.e.>24 h; Angwin et al., 2014;Ramos-Escobar et al., 2021).Our findings for novel words recognised during the follow-up showed an N400 congruity effect that was localised over the right hemisphere.The negative magnitude of this N400 was similar to the N400 for familiar words in the follow-up, and to the N400 for novel words recognised at the end of learning.An overall right-centroparietal distribution was also shown for correctly recognised picture names across both the familiar and novel conditions.According to the present findings, a similar degree of right-hemisphere processing during learning appears to have been maintained several days later at the delayed follow-up.
Much research shows that a right-hemisphere N400 is typically elicited to familiar written words with established representations (for review see: Kutas & Federmeier, 2011).For newly acquired words, a centroparietal N400 has been reported to arise during later stages of learning (Dittinger et al., 2016;Elmer et al., 2022;Ramos-Escobar et al., 2021), taken as an indication that novel word memory representations have stabilised.Source analysis has previously revealed that a right-centroparietal N400 is generated within the left posterior middle temporal region (Kutas & Federmeier, 2011;Lau et al., 2008), an area known to be involved in lexical retrieval and the integration of meaningful information.Although our scalp-level data are unable to provide direct conclusions about underlying sources of neural generators, our findings do indicate a spatiotemporal shift in the N400 congruity effect for novel words, characterised initially by broadly distributed modulations indicative of rapid neuroplastic changes during initial learning, with the localised right-hemisphere/centroparietal modulation at the follow-up more consistent with slower neuroplastic changes for newly learned words.Such a shift has been suggested to reflect slower retention-and consolidation-related processing of previously formed object-word memory associations (Ramos-Escobar et al., 2021).

Implications and future recommendations
Research examining the time course of N400 congruity effects during initial learning and delayed follow-up sessions is important for improving our understanding of lexical-semantic brain responses for different forms of word learning.The present study demonstrates that less than a handful of explicit encoding exposures to novel picture-word pairs accompanied by distinct semantic attributes is sufficient for novel words to achieve a similar degree of neural processing to that of familiar words with established lexical-semantic representations.This finding was evident despite presenting the novel words without their semantic attributes (i.e.familiar written adjective pairs) during the picturename recognition task.By excluding this semantic information from recognition, N400 modulations could not be attributed to a directly encoded association between the novel object pictures and the familiar written adjectives.Therefore, the N400 congruity effects observed for novel words are likely best explained by the presence of the semantic attributes that accompanied the novel picture-name pairs during learning phases.Consistent with evidence from existing research (Angwin et al., 2014;Bermudez-Margaretto et al., 2018), our findings support conclusions suggesting that learning repetition together with semantic exposure enhances the depth of encoding for novel picture-word associations, facilitating lexical-semantic processing and the formation of stronger new word memory traces.
While it is likely that novel picture-name associations were well-encoded during the initial learning session, the decreases in recognition accuracy and response speed at the follow-up may be a consequence of the large number of novel words that needed to be learned.Indeed, 40 novel picture-name pairs may have been too many to consolidate without additional learning during the 3-5-day retention interval.To reduce the memory load, future studies could investigate whether word retention would improve if a smaller number of similarly well-controlled novel picture-name pairs were presented as stimuli for learning.Future research might also consider more extensively investigating how N400 congruity effects are sustained post learning, and whether this effect decays over time, by testing the recognition of novel picture names on days within the post-learning retention interval.
Further research examining the modulatory influence of distinct semantic attributes on novel word learning and subsequent N400 congruity effects is warranted.In addition to presenting novel words with novel objects and semantic attributes during learning, future studies could include a condition where novel object-word pairs are accompanied by non-semantic attributes (e.g.Angwin et al., 2014) or different forms of semantic information (e.g.picture-based semantic features).Recent developments in artificial intelligence models and text-to-image generation (e.g.DALL•E, OpenAI, 2021) may be beneficial for creating semantic stimuli for use in word learning experiments.

Conclusions
This study tracked changes in behavioural and neurophysiological responses to learning novel written words associated with novel pictures and semantic attributes in healthy young adults.Remarkably, less than a handful of novel picture-name learning exposures were sufficient for an N400 congruity effect to arise for novel written words, displaying a level of neural processing similar to that of familiar words.This suggests that explicit encoding of novel picture-name pairs with semantic attributes leads to the rapid formation of new word memory traces that are predominantly retained over time.To the best of our knowledge, this is the first study to demonstrate an N400 congruity effect for newly learned picture-name associations several days after learning.Overall, the present findings are consistent with theories proposing that learning repetition and exposure with accompanying semantic information provide benefits for learning novel object-word associations.

Figure 1 .
Figure 1.Task design and protocol.(A) The experiment was conducted over two sessions, 3-5 days apart.The learning session (session 1) comprised six blocks, each containing a novel picture-name learning phase followed by a picture-name recognition task and a semantic attribute recognition task.The follow-up session (session 2) only contained two blocks of the picture-word recognition trials.Examples of stimulus items and trial structures for the (B) novel picture-name learning phase; (C) picture-name recognition (novel and familiar conditions); and (D) semantic-attribute recognition task.(E) Illustration depicting the word target response screen during the picture-name recognition task and the left (congruous) and right (incongruous) button-press responses.(F) EEG electrode montage with 129 surface recording electrodes (shaded grey), nine regions of interest (shaded in light green), with the channels most prominently associated with these regions (shaded in dark green) named here starting from the top row from leftto-right, with the name of each region of interest cluster reported in parenthesis: F3 (left-frontal), FCz (mid-frontal), F4 (rightfrontal); C3 (left-central), Cz (mid-central), C4 (right-central); P3 (left-parietal), Pz (mid-parietal), P4 (right-parietal).

Figure 2 .
Figure2.Behavioural results for the picture-name recognition task.Line graphs depicting (A) the mean percentage of correct responses and (B) the mean response time in milliseconds for correctly recognised congruous and incongruous familiar and novel picture names throughout the initial learning session (session 1) and the delayed follow-up (session 2).Error bars represent ±1SE.The colours depicting the conditions is the same for both graphs.

Figure 3 .
Figure3.N400 congruity effects for familiar picture names.(A-D) Grand average ERP waveforms for correctly recognised congruous (blue) and incongruous (red) familiar picture names obtained over left-frontal (LF), midline-frontal (MF), right-frontal (RF), left-central (LC), midline-central (MC), right-central (RC), left-parietal (LP), midline-parietal (MP), right-parietal (RP) scalp regions of interest in blocks 1-3 of of the initial learning session (session 1) and 3-5 days later in the delayed follow-up (session 2).In each ERP plot, the black vertical line representing the y-axis is aligned with the onset of the familiar written word target displayed on the computer screen.The shaded grey rectangle represents the time window (247-457 ms) during which the N400 congruity effect was measured.The blue (congruous) and red (incongruous) shading along the traces represents ±1SE.(E) ERP measurement scale for amplitude (yaxis, µV) and time (x-axis, ms).(F) 129-channel EEG electrode montage with nine electrode cluster regions of interest shaded in green.(G) Series of scalp topology plots showing the topographical distribution of the N400 congruity effect for familiar words (incongruous minus congruous).For illustrative purposes, N400 waveforms elicited to familiar picture names were low-pass filtered at 15 Hz.

Figure 4 .
Figure 4. N400 congruity effects for novel picture names.(A-D)Grand average ERP waveforms for correctly recognised congruous (blue) and incongruous (red) novel picture names obtained over left-frontal (LF), midline-frontal (MF), right-frontal (RF), leftcentral (LC), midline-central (MC), right-central (RC), left-parietal (LP), midline-parietal (MP), right-parietal (RP) scalp regions of interest in blocks 1-3 of of the initial learning session (session one) and 3-5 days later in the delayed follow-up (session two).For each ERP plot, the black vertical line representing the y-axis is aligned with the novel written word onset.The shaded grey rectangle represents the time window (247-457 ms) during which the N400 congruity effect was measured.The blue (congruous) and red (incongruous) shading along the traces represents ±1SE.(E) 129-channel EEG electrode montage with nine electrode cluster regions of interest shaded in green.(F) Series of scalp topology plots showing the topographical distribution of the N400 congruity effect for novel words (incongruous minus congruous) for blocks 1-3 of session one and session two, as well as the difference for novel words between block 3 and session 2, and between block 1 and session 2. For illustrative purposes, N400 waveforms elicited to novel picture names were low-pass filtered at 15 Hz.3

Figure 5 .
Figure5.N400 difference wave results for familiar and novel conditions.(A-D) Grand average difference waveforms (incongruous minus congruous) for correctly recognised familiar and novel words obtained over: left-frontal (LF), midline-frontal (MF), rightfrontal (RF), left-central (LC), midline-central (MC), right-central (RC), left-parietal (LP), midline-parietal (MP), right-parietal (RP) scalp regions of interest in blocks 1-3 of of the initial learning session (session one) and 3-5 days later in the delayed follow-up (session two).In each difference wave plot, the black vertical line representing the y-axis is aligned with the onset of the written word target.The shaded grey rectangle represents the time window (247-457 ms) during which the N400 congruity effect was measured.The blue (congruous) and red (incongruous) shading along the traces represents ±1SE.For illustrative purposes, difference waves were low-pass filtered at 15 Hz.(E) Difference wave ERP measurement scale for amplitude (y-axis, µV) and time (x-axis, ms).(F) 129-channel EEG electrode montage with nine electrode cluster regions of interest shaded in green.(G) Series of scalp topology plots showing the topographical distribution of the N400 difference wave distributions (familiar minus novel).

Table 1 .
Mean accuracy and reaction times for semantic attribute recognition task.

Table 2 .
Results from omnibus repeated measures ANOVAs conducted on N400 amplitudes in session one and two.