For the systematic review of ERP research on language processing, data were sourced from the Science Citation Index-Expanded (SCIE), the Social Science Citation Index (SSCI), and the Arts & Humanities Citation Index (A&HCI) within the Web of Science (WOS) Core Collection by Clarivate Analytics (2021). The Web of Science repository offers access to premier publications along with their citation data across the natural and social sciences, encompassing a time frame since 1975. Consequently, utilizing the core collection of this database ensures the inclusion of high-quality materials essential for this comprehensive analysis.

In accordance with the delineation and scope of the research domain, a systematic literature retrieval strategy was formulated as follows: Initially, keywords “language” and “event-related potential” were selected as primary retrieval terms, applied within the specified temporal boundary from January 1, 2002, to December 31, 2022. Subsequently, articles and reviews were exclusively considered in the document type, while the language criterion was restricted to English. 3797 publications were retrieved from the database. Following the online retrieval of literature, a meticulous manual screening process was executed based on predefined criteria to ensure the relevance and alignment with the subject matter. Two exclusion criteria are set to complete the screening: (1) Publications unrelated to ERP research on language processing. (2) Publications without abstracts. Records in either of abovementioned conditions were systematically excluded from the dataset. Consequently, a total of 3772 publications meeting the stringent inclusion criteria were identified for subsequent analysis. These selected publications were compiled and exported in a plain text file format, encompassing complete records along with cited references. The step-by-step protocol for data collection is presented in Figure 1.

In this study, a mapping-knowledge-domain analysis within ERP research on language processing was conducted employing Citespace 6.1.R6. Citespace is an information visualization software based on Java programming language. It enables the visualization of the knowledge structure and the identification of potential trends by employing co-citation analysis and pathfinder network scaling algorithms (Chen, 2016). Specifically, Citespace generates two distinct visual representations: one focused on co-citation analysis encompassing references, cited authors, and journals, while the other concentrates on co-occurrence analysis involving keywords, clusters, and citation bursts.

Preceding the formal analysis, duplicate records were eliminated using Citespace. According to the records in the current research, all 3772 publications were valid for the formal analysis. Then, several methods of mapping-knowledge-domain analysis were employed to address the pertinent research inquiries. To address questions (1), a comprehensive synthesis of the retrieved articles was presented, encompassing details such as geographic origins and publication timelines. For question (2), a co-citation analysis of references, cited authors, and journals was undertaken to identify the most frequently cited researchers and their contributions. To elucidate question (3), a co-occurrence analysis of keywords and clusters was executed to visually represent the primary themes encapsulated in the relevant studies. Finally, in addressing question (4), burst detection analysis of keywords was utilized to discern emerging trends across distinct temporal segments.

To explore the temporal distribution of literature pertaining to ERP research on language processing, we analyzed the publication statistics across various years using data obtained from WoS. Figure 2 illustrates the chronological progression of ERP research on language processing from 2002 to 2022.

Figures 3, 4 illustrate the top 10 productive countries/regions and organizations between 2002 and 2022 in the research field. Through the examination of these statistics, a broad comprehension of the geographical distribution of ERP research on language processing can be obtained.

Co-citation analysis stands as a fundamental technique within bibliometric quantitative studies, designed to uncover the specialized domains within a scientific field (Morris and Van der Veer, 2008). This method often involves the identification of specialized clusters based on the co-citation of individual entities. For instance, Author Co-Citation Analysis (ACA) and Document Co-citation Analysis (DCA) are commonly utilized to delineate prominent researchers and their works, thus elucidating the knowledge network within specific research domains (Chen, 2006; Zhao and Strotmann, 2008; Chen et al., 2010). In this present study, we employed co-citation analysis to unveil the most frequently cited authors, references, and journals. Our analysis applied a 1-year time slice and selected the top 20 entities within each slice as parameters in Citespace. Figures 5–7 portray the outcomes of this analysis.

Figure 5 illustrates the leading authorities in ERP research on language processing. Based on the analysis conducted using Citespace, the top 10 scholars and their corresponding citation counts are as follows: Kutas (1832 citations), Friederici (1033 citations), Hagoort (958 citations), Oldfield (814 citations), Osterhout (785 citations), Näätänen (738 citations), Van Petten (700 citations), Holcomb (601 citations), Federmeier (597 citations), and Kuperberg (548 citations). These esteemed scholars have played instrumental roles in shaping the landscape of ERP research on language processing, contributing significantly to the foundational framework of this field.

Figure 6 delineates the most frequently cited journals in ERP research on language processing, signifying their concentrated focus and high caliber publications within this discipline. According to the statistical data derived from Citespace spanning the period from 2002 to 2022, the top 10 journals and their corresponding citation counts are as follows: Journal of Cognitive Neuroscience (3019 citations), Psychophysiology (2726 citations), Brain and Language (2650 citations), Neuropsychologia (2587 citations), Neuroimage (2179 citations), Science (2061 citations), Trends in Cognitive Sciences (2036 citations), Brain Research (2007 citations), Cognitive Brain Research (1886 citations), and Journal of Memory and Language (1872 citations).

Figure 7 visualizes the most referenced contributions within the domain of ERP research on language processing. These findings align with the outcomes obtained through co-citation analyses of referenced authors and journals. Eminent neurolinguistic scholars such as Friederici, Kutas, and Van Petten are prominently featured in these significant works. Additionally, these works are predominantly disseminated through influential journals like Trends in Cognitive Sciences, Brain Research, Science, among others.

Keywords hold an essential role in demonstrating the primary theme and scope of a study. When two keywords frequently co-occur, it suggests a potential connection between them. It is commonly understood that the frequency of co-occurrence reflects the strength of their association. Within Citespace, the Betweenness Centrality stands as a metric to gauge this association strength. A higher centrality value assigned to a keyword signifies its relative significance within the network (Chen, 2004; Han et al., 2022; Peng and Hu, 2022). Therefore, this investigation undertakes a co-occurrence analysis to explore the centrality and frequency patterns of keywords in the domain of ERP research on language processing, aiming to identify prevalent focal points. Figure 8 visually presents the highlighted keywords and their interconnections, while Table 1 delineates the top terms alongside their respective frequencies and centrality values.

Referring to Figure 9 and Table 1, the predominant keywords in ERP research on language processing encompass event-related potential, language, language comprehension, speech perception, time course, brain, word, working memory, mismatch negativity, and N400.

Based on the keyword co-occurrence analysis, this study performed a cluster analysis to categorize the terms derived from all 3772 publications. LLR (Log Likelihood Ratio) is a text analysis algorithm used in tasks like classification. It measures the likelihood of a term occurring in one class compared to others. High LLR values indicate terms with strong discriminatory power, aiding in feature selection for accurate classification. The generated clusters, labeled using LLR, resulted in three distinct clusters. Figures 8, 10 present both synchronic and diachronic perspectives of these clusters, while Table 2 encapsulates essential information regarding each cluster.

Over the past two decades, ERP research on language processing has witnessed significant development, marked by the publication of numerous studies. Our analysis using Citespace revealed that sentence comprehension, mismatch negativity, and reading comprehension have surfaced as focal points. In particular, discussion surrounding speech perception, working memory, and temporal dynamics has been prevalent. This field appears interdisciplinary, amalgamating theories and methodologies from diverse subjects. Hence, it can be inferred that ERP research on language processing presently holds a pioneering role.

To delve deeper into the emerging trends within ERP research on language processing, our study conducted burst detection analysis on 3772 publications’ keywords. Burst terms represent keywords that undergo a remarkable surge within a specific timeframe, indicating ongoing developmental trends (Yuan and Sun, 2023). Figure 11 illustrates the top 25 burst terms:

Figure 11 presents the evolving trends in ERP research on language processing over the preceding two decades through the examination of burst terms observed in distinct time periods. In this figure, “year” represents the emerging time of the burst term, shown by deep blue column. “Begin” and “end” in red column demonstrates the burst period of the term. “Strength” refers to a measure that quantifies the intensity and significance of the sudden increase in frequency of the term over its burst period. Citespace utilizes an algorithm developed by Jon Kleinberg for burst detection (Kleinberg, 2002).

The spatial-temporal distribution of the research pattern is clearly revealed by the visualization data. Figure 2 depicts a consistent upward trend in international publications focusing on ERP research on language processing. Initially emerging as a novel methodology, there existed a relatively limited number of studies employing ERPs in language processing research. Notably, between 2002 and 2004, the yearly publication count remained below 100. However, the field witnessed a subsequent surge in research endeavors. Particularly noteworthy are significant increments observed during three distinct periods: 2006 to 2007, 2010 to 2012, and 2017 to 2021. Remarkably, in 2021, the publication count reached its zenith, registering 244 publications within a single year. Despite minor declines in publication counts observed in 2008, 2010, and 2017, each downturn was promptly followed by obvious upswings. These observations underscore the emergence and burgeoning interest in ERP research on language processing in recent years. Notably, it is posited that significant discoveries concerning ERP components and advancements in research methodologies have propelled the proliferation of studies in this field (Kutas and Federmeier, 2011; Lopez-Calderon and Luck, 2014; Bates et al., 2015). Consequently, the number of publications experiences substantial growth during specific temporal periods.

From the insights gleaned in Figures 3, 4, a dominant presence is observed in ERP research on language processing, particularly attributed to the United States. Over a span of two decades, US-based researchers have contributed significantly, accounting for approximately 35% of the total studies in this field, amassing a total of 1308 publications. Outstanding academic institutions such as the Max Planck Institute for Human Cognitive and Brain Sciences (with 203 articles), the University of California San Diego (with 127 articles), and the Radboud University Nijmegen (with 120 articles) have significantly bolstered the advancement of ERP research on language processing. Geographically, the epicenters of ERP research on language processing are chiefly concentrated in Europe and North America. Nevertheless, an encouraging trend is observed in Asia, particularly in China (with 430 articles), evidencing substantial growth and promising advancements in catching up with the forefront of ERP research.

Results from co-citation analysis of references (Figure 7 and Table 3), co-occurrence analysis of keywords (Figure 9 and Table 1) and cluster interpretation (Figures 8, 10 and Table 2) have unraveled several key points in the ERP research on language processing from 2002 to 2022. Discussion will concern three main topics, namely the ERP components, methodologies and techniques, as well as research scope.

Several ERP components have been instrumental in exploring the neurocognitive mechanisms of language processing. The N400, illustrated by Kutas and Federmeier (2000) as an index of semantic processing, is a significant electrophysiological signature in language processing. Studies highlight its role in sentence processing and neural mechanisms of language comprehension, suggesting a dynamic and modality-specific semantic processing network predominantly in the left hemisphere (Koelsch et al., 2004; Frank et al., 2015). Kutas and Federmeier (2011) extended this by demonstrating that the N400 is evoked by multimodal stimuli, including written, spoken, and sign languages, even non-linguistic stimuli like drawings and videos. This broadened the exploration of the N400 into perception, attention, and memory, reflecting how the brain utilizes top-down and bottom-up information to understand the world. Lau et al. (2008) anatomically deconstructed the N400 component, proposing a neurocognitive model for semantic processing that includes core regions like the left posterior temporal cortex (lPTC), left anterior temporal cortex (lATC), angular gyrus (AG), and left inferior frontal gyrus (lIFG). This model suggests that lexical semantic information is stored in the lPTG and higher-level semantic processes involve the lATG and AG, with the lIFG implicated in selecting and retrieving specific lexical representations.

The P600, another crucial ERP component, normally associated with syntactic processing, was interpreted by Brouwer et al. (2012) through the Retrieval-Integration Account. They found that semantically anomalous sentences evoked a P600 rather than the expected N400, suggesting that the P600 reflects the integration of lexical information within contexts and plays a role in constructing or updating mental representations.

Mismatch negativity (MMN) is closely related to sound processing and signifies the activation of memory networks for language sounds and spoken words (Pulvermüller and Shtyrov, 2006). Shtyrov and Pulvermüller (2002) observed distinct MMN responses to word stimuli compared to pseudowords, supporting immediate word processing upon identification. Jacobsen et al. (2004) demonstrated that MMN is evoked by low-frequency vowels, indicating that the auditory processing network can pre-attentively extract vowel formant structures. MMN also indicates syntactic processing. Hahne et al. (2001) found that combined physical and syntactic violations evoked larger MMNs than either alone, suggesting parallel and independent early-stage processing. Pulvermüller et al. (2007) evidenced the autonomy of syntactic processing, while Hasting et al. (2007) revealed functionally distinct neural mechanisms underlying subject-verb agreement and word class information. What’s more, studies on MMN encompass multimodal materials (Chen et al., 2022). Ishida and Nittono (2022) explored anomalies in music, finding that syntactic anomalies evoke early right anterior negativity (ERAN) and acoustic anomalies trigger MMN. Ding et al. (2022) showed that neutral and fear emotions induce greater MMN compared to positive emotions, revealing a processing bias toward negative emotions in body gestures.

The Left Anterior Negativity (LAN) and Early Left Anterior Negativity (ELAN) are components associated with syntax comprehension, typically observed over the left anterior scalp regions. ELAN, appearing within 100–200 ms after stimulus onset, is involved in early syntactic processing, while LAN, occurring around 300–500 ms, reflects sensitivity to syntactic violations or grammatical errors (Palolahti et al., 2005; Steinhauer and Drury, 2012).

Methodologies in ERP research are crucial for study quality. Luck (2014) provides a comprehensive guide on ERPs in cognitive neuroscience, covering theoretical foundations, experimental paradigms, and data analysis techniques. Lopez-Calderon and Luck (2014) introduced ERPLAB, an open-source toolbox for ERP data analysis compatible with MATLAB, enhancing the ease of offline analysis of ERP data. Bates et al. (2015) detailed the application of linear mixed-effects models using the lme4 package in R, offering a more refined model for predicting underlying patterns compared to traditional ANOVA analysis.

ERP studies on language processing often use visual stimuli, emphasizing reading comprehension. Ditman et al. (2007) proposed the self-paced reading design in ERP research for a more precise examination of reading processing. Metzner et al. (2017) integrated ERPs and eye-movement techniques, suggesting alternative strategies in sentence comprehension, such as reanalysis (P600) or tolerance of insufficient interpretations (N400).

Novel methods like time-frequency analysis and microstate analysis are gaining attention. Time-frequency analysis decomposes ERPs into frequency components over time, offering insights into cognitive functions (Herrmann et al., 2014; Morales and Bowers, 2022). Microstate analysis studies dynamic brain activity associated with cognitive processes, segmenting the ERPs signal into temporally stable topographic states, representing synchronized neural activity (Ott and Jäncke, 2013; Croce et al., 2021).

The ERP technique has found diverse applications in various linguistic research domains. Van Petten and Luka (2012) synthesized ERP studies on prediction during language comprehension, noting the N400 as an indicator of benefits from semantic context and positive components like P300 and P600 addressing prediction costs. Friederici (2002) emphasized the neural underpinnings of auditory sentence processing, positing that comprehension is organized within a bilateral temporo-frontal network. Kuperberg (2007) reviewed studies on the P600, suggesting that language comprehension involves competing neural processing streams: a semantic stream reliant on long-term memory and a combinatorial stream for morphosyntactic rules. This clash triggers reanalysis, evident in the P600. Chow and Phillips (2013) investigated the “Semantic P600” phenomenon in Chinese, finding that well-formed role-reversed sentences evoked P600 activation, highlighting the dependence of online semantic processing on surface syntax. Deng et al. (2016) revealed early syntactic processing (ELAN) and semantic-syntactic integration (N400) in Chinese verbs. Stowe et al. (2018) emphasized the need for updated experimental designs to explain late components in sentence wrap-up. Recent studies highlight pragmatic knowledge’s role in sentence comprehension. Kuperberg et al. (2007) proposed an animacy-based P600 effect influencing thematic-semantic relations. Van Berkum et al. (2007) posited active referential processing during sentence comprehension. Casado et al. (2020) revealed that motor sequencing affects early syntactic processing. Figurative language studies by Ferretti et al. (2021) and Shen et al. (2022) focused on proverbs and metaphors in sentence processing. Studies on reading comprehension and development, such as Stites and Laszlo (2017), linked ERP components with children’s reading development, finding that N250 predicted phonological awareness. Tabullo et al. (2020) found that higher reading skills were associated with reduced N400 during expected word processing. Troyer et al. (2022) explored hemispheric processing asymmetries, highlighting the left hemisphere’s engagement with specific semantic information and the right hemisphere’s involvement in broader semantic connections.

Referring to the outcomes from burst detection analysis (Figure 11), several findings of developing trends are discovered. Besides, future directions may be predicted based on the statistics.

The first finding denotes a transition toward examining larger linguistic units (Nieuwland et al., 2019). Throughout the early 21st Century, ERPs investigations in language predominantly centred on lexical processing, evident from burst terms like “lexical decision” (2003–2005) and “word” (2003–2005). Initial studies were primarily focused on unraveling the neurocognitive mechanisms involved in processing individual words, pivotal in understanding language structures. During this phase, some studies isolated word stimuli for examination (Barber et al., 2004; Carreiras et al., 2005), while others incorporated these stimuli into phrases or sentences (Allen et al., 2003; Weber-Fox et al., 2003; Schiller, 2006). Nonetheless, the primary emphasis remained on the lexical level. The proliferation of studies concentrating on lexical processing established a robust foundation for ERP research on language processing. Consequently, subsequent investigations broadened their scope to encompass varied linguistic materials, highlighted by the emergence of the burst term “sentence comprehension” (2020–2022). Beyond lexical-semantic attributes, sentence processing encompasses the integration of diverse linguistic information. Recent studies have frequently delved into neural mechanisms associated with syntactic processing (Jackson et al., 2020; Hao et al., 2021) and pragmatic processing (Allegretta et al., 2021; Jouen et al., 2021; Schoknecht et al., 2022).

The second key finding underscores the growing integration and synthesis observed in ERP research on language processing. Initially, most studies were centred around specific ERP components, aiming to explain their distinct functions. For instance, early investigations proposed the N400 as an indicator of semantic violation (Quiroz-G, 2003) and the P600 as a marker of syntactic violation (Kuperberg et al., 2003). This trend is evident from burst terms like “semantic processing” (2002–2004) and “evoked potential” (2005–2009). In more recent years, the emergence of new burst terms such as “component” (2015–2022), “N400” (2019–2022), and “integration” (2019–2022) has spurred a re-examination of ERP components. A substantial body of studies now argues that the same component may exhibit diverse effects or functions across different experimental contexts (Kutas and Federmeier, 2011; Niharika and Rao, 2020). Moreover, the appearance of the burst term “eye movement” (2018–2019) suggests a trend toward integrating the ERP technique with other methodologies in linguistic research to acquire more robust evidence (Kessler et al., 2021).

The third significant finding highlights the increasing attention given to individual participant characteristics in ERP research on language processing. Burst terms such as “individual difference” (2017–2022) and “children” (2015–2018) signify this emerging trend. Investigations into second language acquisition have begun to identify individual variations by considering between-subject factors such as second language ability (Bice and Kroll, 2021; Liang and Chen, 2022; Grey, 2023). Furthermore, discussions regarding individual factors like gender or age have gained traction in recent studies (Cheimariou et al., 2019; Choisdealbha et al., 2022; Padrón et al., 2023).

In general, ERP research on language processing is poised to advance along several key trajectories in the future. Firstly, there will be a continued trend toward investigating larger linguistic units, so as to find out human brain’s processing mechanisms of complex language structures. Secondly, there will be a synthesis and re-examination of ERP components, exploring their multifaceted nature and roles across different linguistic contexts. Thirdly, there will be an increased focus on individual participant characteristics, to better understand variability in language processing outcomes. Lastly, there will be a growing integration and update of ERP techniques with other methodologies, to provide a more comprehensive understanding of language processing mechanisms. Overall, these trends are expected to aid the development of more nuanced models of language processing.

This study is the first mapping-knowledge-domain analysis on ERP research on language processing, which pioneers the systematic review of the topic. 3772 relevant publications from the WoS core collection are exclusively extracted for precise visualization analysis.

The database in the current study is still far from inclusive, with only WoS core collection included. Publications from other databases are not considered properly, for example, PubMed and Scopus. Besides, merely publications written in English have been selected, which might lead to a bias in the analysis. Finally, the research scope in the current study is limited in language comprehension and perception. However, in language-related ERP study, language production also plays an essential role. Further research should discuss the neurocognitive mechanisms of language production in detail.