Collaboration, i.e., an activity where two or more individuals share a goal or intention that motivates joint work (Léné et al., 2021; Wood and Gray, 1991), is one of the most prevalent topics under study in hyperscanning research. Much of the field has examined face-to-face social interaction under naturalistic conditions (Czeszumski et al., 2020; Fan et al., 2021; Réveillé et al., 2024; Schneider et al., 2021). Yet, collaboration in contemporary society increasingly occurs through digital means, including screen-based meetings, shared online workspaces, and immersive virtual environments (Deschênes, 2024; Wu et al., 2023; Yang et al., 2025). These developments raise critical questions about how environmental conditions (virtual vs. in-person) and task-specific tools or modalities (e.g., digital vs. analog objects, verbal vs. text-based communication) influence the brain, body, and behavioral signatures of collaboration (Leahy et al., 2025; Magni et al., 2025; Solomon and Theiss, 2022). Digital media introduces a virtual divide (e.g., a monitor), altering attentional demands (Chuang and Hsu, 2023; Snijdewint and Scheepers, 2023), perception of implicit social cues (Frith and Frith, 2008; Oh et al., 2018; Sharan et al., 2022), and cognitive workload (Luebstorf et al., 2023; Nurmi and Pakarinen, 2023). As a result, findings from face-to-face hyperscanning studies cannot be assumed to generalize to digitally mediated settings, where collaborators rely on virtual rather than physical co-presence (Balconi et al., 2022; Balters et al., 2023; Liu et al., 2019; Sarasso et al., 2022, 2024). Although prior work suggests that the degree of digitality modulates neural and behavioral responses (for reviews, see Balters et al., 2020; Barde et al., 2020), the literature remains fragmented by different choices in task, measurement, and analysis. These developments motivate a systematic review of how hyperscanning is currently used to investigate digital collaboration.

The term hyperscanning was introduced by (Montague et al. 2002) to describe simultaneous functional magnetic resonance imaging (fMRI) of interacting individuals playing a competitive deception task. Earlier work had already explored dual-brain recordings and reported synchronized electroencephalography (EEG) activity in the alpha band (8–13 Hz) in identical twins (Duane and Behrendt, 1965). Since then, hyperscanning has expanded rapidly, driven by advances in mobile neuroimaging (for a review on the historical development see Nam et al., 2020). Today, both EEG and functional near-infrared spectroscopy (fNIRS) are the most widely used mobile brain-imaging modalities, offering complementary strengths: EEG provides high temporal resolution on a millisecond timescale, whereas fNIRS affords greater spatial specificity for cortical surface activity, making it well-suited for mapping spatial patterns of brain activation (for review see Mehta and Parasuraman, 2013). Recent advances in mobile neuroimaging and physiological sensing, together with increasingly accessible wearable sensors, have expanded hyperscanning beyond stationary setups such as fMRI or magnetoencephalography (MEG), enabling the study of interpersonal dynamics in increasingly diverse and applied settings (Carollo and Esposito, 2024).

Real-world and digitally mediated interactions unfold across multiple sensory channels and involve tightly coupled cognitive, affective, and behavioral processes (Zamm et al., 2024). Accordingly, reviews increasingly emphasize the value of multimodal and ecologically grounded frameworks of social interaction (Hakim et al., 2023; Schneider et al., 2021). Empirical work has begun to integrate brain-based methods with bodily measures such as gaze alignment (Chuang and Hsu, 2023), electrocardiography (ECG), electrodermal activity (EDA; Numata et al., 2021), communication signals (Lu et al., 2020), and other camera-based physiological metrics (Shih et al., 2024). This integration, called embodied hyperscanning (for review see Grasso-Cladera et al., 2024), has facilitated the transition from controlled laboratory paradigms to applied settings such as classrooms (Dikker et al., 2017; Zhang et al., 2024), workplaces (Wikström et al., 2021; Wu et al., 2023), and other real-world environments (Balters et al., 2023). In line with this evolution, we use the term “hyperscanning” here to encompass approaches that incorporate at least one brain- or body-based measurement modality of physiology.

Irrespective of the signal measured, inter-brain synchrony (IBS) has been the most widely reported analysis approach for interpersonal dynamics in a variety of interactive scenarios (for review see Réveillé et al., 2024). IBS has been linked to various processes central to social interaction, including shared attention (Dikker et al., 2017; Snijdewint and Scheepers, 2023), simultaneous movements (Gumilar et al., 2021), imitation (Delaherche et al., 2015; Konvalinka et al., 2023), social closeness between partners (Reinero et al., 2021), and the nature of their engagement (e.g., cooperation vs. competition; Hayne et al., 2023). Given the pronounced sensitivity of IBS to contextual and situational factors, newly emerging digital collaboration formats warrant systematic investigation as they increasingly reflect naturalistic settings (Fan et al., 2021). Alongside synchrony-based metrics on the inter-brain level, multivariate modeling and machine learning have emerged in intra-brain research that could expand the analytical repertoire for hyperscanning, enabling improved decoding of interpersonal states and complementary use of EEG, fNIRS, and physiological signals (for reviews see Dissanayake et al., 2025; Lühmann and Müller, 2017; Pinto-Orellana et al., 2024).

Despite these methodological and analytic advances, synthesizing hyperscanning studies investigating digital social interaction remains challenging due to substantial heterogeneity. Heterogeneity is introduced by varying group size (dyads, triads, or larger collectives; Hou et al., 2022; Park et al., 2023), pre-existing social relationships between partners (Bae et al., 2024; Dikker et al., 2021), and the gender composition of paired participants (Zhang et al., 2023c). Furthermore, the degree of physical vs. virtual co-presence differs across paradigms with respect to how participants are situated (same or different rooms, physical or virtual co-presence), how information is exchanged, and to what extent interactions rely on digital vs. analog means (Balters et al., 2020). For instance, some experiments manipulate remoteness by prohibiting direct visual and/or auditory contact through a physical divide (e.g., Wu et al., 2023) or by placing participants back-to-back (e.g., Kütt et al., 2019).

Previous reviews have proposed frameworks to organize this diversity. For digitally mediated interaction, several categories have been identified by (Balters et al. 2021): (i) the type of communication (goal-directed vs. open-ended), (ii) the transfer of information (i.e., whether information exchange between participants happened via analog, digital, or mixed channels), (iii) the interaction medium (originally referred to as “interaction manipulative”) being digital or non-digital, and (iv) the interaction scenario, i.e., the spatial layout of how participants are situated relative to each other (e.g., face-to-face, side-by-side, virtually connected). Other work identified the experimental task itself as a central factor for recurring experimental archetypes (Barde et al., 2020; Wang et al., 2018). Moreover, a clear conceptual distinction can be made between collaborative interactions based on task symmetry, distinguishing between symmetrical interactions without predefined roles and asymmetrical tasks with designated roles (Wikström, 2022). Together, these dimensions outline the core design space of hyperscanning studies, albeit they are currently not integrated across modalities or applied specifically to digitally mediated collaboration.

Without structuring the body of research considering all these dimensions, theoretical synthesis and cross-study comparison remain difficult, particularly when assessing how digitalization shapes the neural, physiological, and behavioral foundations of collaboration. The current review organizes hyperscanning research along dimensions relevant to studying remote and digitally mediated collaboration. Building on this structured synthesis, we present the framework for a living literature database: an extensible, interactive platform that categorizes hyperscanning studies according to measurement modalities, analytic approaches, interaction design dimensions, and task characteristics. The current version includes studies of remote collaborative interactions using mobile hyperscanning methods, with the long-term goal of expanding the database to incorporate additional modalities, paradigms, and newly published work.

To identify eligible sources for review, a broadly defined search for hyperscanning studies, including or focusing on digital interaction scenarios, was conducted using a combination of search terms that reflect hyperscanning, collaboration, recording modalities, and digital context (see Table 1). Initially, Scopus (https://www.scopus.com), PubMed (https://pubmed.ncbi.nlm.nih.gov), and Web of Science (https://www.webofscience.com) were searched. This search was conducted on April 28, 2025. The exact search terms were adapted according to the specifications of each search engine (for details, see Supplementary Table 1). The initial search was carried out for journal articles published in English or German, searching paper titles and abstracts only. In line with the PRISMA guidelines, additional (unsystematic) data searches were conducted as well (see Figure 1). Any additional findings brought to us via search alerts or by colleagues were included if eligible, although no additional proactive search was conducted.

Following the PRISMA recommendation, we used the PICOS eligibility criteria (Population, Intervention, Comparison, Outcomes, Study design) to assess eligibility (see Table 2). For the presented review, we considered only studies with a population of at least two clinically healthy adult participants. As an eligible experimental intervention, we defined the application of one or more mobile brain or body imaging techniques for physiological data to focus on mobile and applied neuroergonomics contexts. We did not limit the search to one specific mobile method since we aimed to investigate the number of joint applications of measurements. Note that we also identified some studies using camera-based tracking, encompassing the assessment of eye contact and facial muscle movements. However, this modality was not explicitly included in the search scope. Nevertheless, a specific label was assigned in these cases, given that the modality might be of interest. For the comparison criterion, we included studies that scanned at least two participants simultaneously during an interaction. Given the search term, we expected to find mainly interaction scenarios with group-based collaboration. As outcomes, we defined any results addressing inter-subject dynamics as assessed by the measurement modalities of interest. Studies reporting no statistical results related to the physiological measures during the collaborative interactions were not included. Lastly, we included the additional study design criterion to limit the review to hyperscanning studies that introduced a collaborative scenario with a focus on a digital aspect in the study design. For example, studies meeting all initial criteria and focusing purely on face-to-face collaboration were excluded. Further, studies with tasks involving imitation rather than autonomous choice of all collaborators were not included (Bieńkiewicz et al., 2021).

For duplicate removal, the Zotero (version 6.0.36) and Citavi (version 6.15) software were used, as well as an automated matching of cosine similarity of titles and abstracts using sklearn (Pedregosa et al., 2011). Finally, the web-based review program Rayyan (Ouzzani et al., 2016) was used for duplicate removal and further screening procedures. All automated duplicate removals were manually confirmed before exclusion.

The identification procedure followed by duplicate removal resulted in 6,070 sources from the database searches and 120 sources from other methods (see Figure 1).

After collecting relevant sources based on the PRISMA criteria, we aimed to extract commonalities across studies by categorizing studies across 13 dimensions based on design choices, measurement modalities, analysis approaches, and targeted cognitive functions. Category definitions were adapted from existing hyperscanning reviews (e.g., Balters et al., 2020, 2021; Wang et al., 2018) or based on influential factors of the sample and design (e.g., how auditory and visual information is exchanged between participants during the interaction). Each included source was individually reviewed and subsequently categorized by extracting the following information: (1) measurement modalities of interest used in the study (e.g., EEG, eye-tracking); (2) the number of participants included in analyses; (3) the pairing configuration of participants (whether measurements were taken simultaneously on dyads, triads, tetrads, or larger groups; Hou et al., 2022; Park et al., 2023); the pairing setup of participants, specifically variables of the setup that are known to influence behavior and subsequently results, such as (4) gender (Zhang et al., 2023c) and (5) relationship (Bae et al., 2024; Dikker et al., 2021) of simultaneously measured individuals; (6) the type of hyperscanning paradigm employed (see Table 3; Barde et al., 2020; Wang et al., 2018); (7) the task symmetry (Wikström, 2022), where highly symmetrical tasks allow participants to assume equal roles (e.g., puzzling with shared pieces), whereas low-symmetry tasks involve distinct roles (e.g., navigator and pilot); (8) the type of communication defined as either open-ended or goal-driven, and (9) the transfer of information being either analog, digital, or mixed (Balters et al., 2020). Following the detailed interaction categories introduced by (Balters et al. 2021) for digital hyperscanning studies, the (10) interaction scenario during the experiment was assessed (i.e., how participants were situated in relation to each other during measurements) and (11) whether the experiment included verbal, physical, and/or digital interaction media in shared or separated manners (for details see Table 4; Balters et al., 2021). Given the diverse measurement modalities included in the present review, we aimed to categorize (12) analysis methods in a signal-agnostic manner, i.e., not tied to specific modalities like EEG or fNIRS (Hakim et al., 2023). Therefore, we distinguished on a higher level between the analysis approaches focused on temporal aspects, spatial distribution, and connectivity domains, as well as machine learning methods (for details, see Table 5). Finally, we attempted to highlight the (13) cognitive function of interest investigated primarily during analyses for each study (Balters et al., 2021). Note that multiple labels within one category may apply to the same study.

An overview of the 45 included studies can be found in Table 6. The earliest studies included were published in the mid-2010s (see Figure 3). The average sample size across studies was 55.8 ± 70.3 (SEM = 10.48). The sample sizes varied considerably, from small-scale studies with four (Pöysä-Tarhonen et al., 2021) to large-scale studies with 480 participants (Zhang et al., 2023b). Most studies included within-subjects designs (n = 33), some analyzed between-subjects effects (n = 10), and two studies had a mixed design (Fang et al., 2022; Yamaya et al., 2025). This observed distribution might be attributed to the fact that sufficient statistical power of a between-subjects design requires a substantially larger recruitment effort for hyperscanning studies.

In the following section, the results of the categorization of the selected studies according to the labels described in Section 2.4 and detailed in Table 6 are summarized.

To promote transparency, reproducibility, and meaningful comparison across hyperscanning studies, we strongly encourage authors to report specific methodological details that directly impact the interpretation of inter-brain connectivity and synchronization. Based on our review of the literature, we outlined a set of minimal reporting guidelines that, if followed, greatly improve the interpretability and meta-analytic usefulness of future research (see Table 7).

Firstly, researchers should thoroughly report participant metadata. This includes age, gender, familiarity among participants, and any pairing characteristics that might systematically influence neural coupling or social behavior (e.g., handedness, prior experiences). Studies that do not disclose this information limit the ability to meaningfully contextualize, compare, or replicate their results.

Secondly, it is crucial to specify the temporal setup of the neural recordings, i.e., whether data were recorded simultaneously across participants or if signal streams were aligned afterward. Although both methods are valid, they have different implications for interpreting time-locked neural coupling and inter-brain dynamics. On one side, simultaneous recordings offer higher ecological validity for social components compared to single-person recordings; on the other side, single-person recordings provide greater experimental control and easier replication that is less dependent on the sample (Fan et al., 2021). The included studies used simultaneous acquisition (as part of the comparison inclusion criterion). However, during full-text screening, a notable subset (n = 46) was excluded because they relied on separate acquisitions with later alignment, even though the abstract screening suggested a form of real-time social interaction.

Thirdly, authors should clearly state whether verbal communication was allowed between participants and, if so, describe the nature and structure of the dialogue. Was communication entirely open-ended, limited to specific prompts, or completely prohibited? The level of verbal interaction fundamentally alters the cognitive and emotional demands of the task (Lu et al., 2020). Apart from affecting social, cognitive, emotional, and sensory processes, it can also introduce signal noise through muscle activity. Studies should specify whether verbal and/or non-verbal communication was possible during the experiment, as both can influence interpersonal neural synchrony in collaborative settings (Shih et al., 2024).

Fourthly, it is essential to describe the visual and spatial arrangement of the participants during the experiment to help readers better understand the sensorimotor and perceptual context of the interaction. Specifically, were participants able to see each other's facial expressions, gestures, and body movements? How were they positioned relative to each other (e.g., face-to-face, side-by-side, separated by a screen)? We recommend including a schematic diagram or photo of the experimental setup whenever possible.

Fifthly, we recommend clearly stating whether analyses were conducted at an intra- or inter-individual level or both. To ensure a structured and clear understanding of the methods, we suggest dividing the description of methods accordingly. For studies involving multimodal data collection, we recommend separating the description of analysis methods by measurement modality and clearly specifying how the modalities were combined and analyzed together.

By following these reporting practices, researchers can ensure their hyperscanning studies are both understandable on their own and useful to the larger research community. Future meta-analyses, database-driven reviews, and cumulative science efforts will depend heavily on the consistency and completeness of this essential information.

The wide variety of how factors relevant for digital collaboration are implemented across hyperscanning studies, whether through virtual interfaces, perceptual barriers, or hybrid setups, illustrates the diversity and adaptability of current research methods. A key contribution of this review is the systematic organization of interaction scenarios and media, offering a clear overview of how researchers define digital collaboration. This categorization is an important initial step toward combining findings on remote collaboration across different modalities. Building on this foundation, future research will benefit from designs that directly compare multiple measurement approaches during various interaction phases, as such methods can help develop comprehensive theoretical models and support applying hyperscanning insights to real-world digital environments (Schneider et al., 2021).

Across the reviewed literature, distinct (neuro-)physiological modalities were associated with specific aspects of collaborative cognition. This pattern could reflect both methodological affordances and historical trends in the field of hyperscanning. FNIRS studies more often targeted social cognition and language processes compared to EEG, leveraging spatial specificity and relative robustness to motion artifacts to probe regions such as the inferior frontal gyrus or temporoparietal junction (Czeszumski et al., 2022). EEG, by contrast, was predominantly used to study executive functions and visuospatial cognition, leveraging its millisecond-level temporal resolution and sensitivity to oscillatory synchronization across distributed networks. Eye-tracking hyperscanning was used less frequently and primarily to investigate attentional coordination between collaborators, consistent with gaze as a behavioral marker of visual attention (Ferencová et al., 2021). Yet relatively few eye-tracking studies explicitly focused on visuospatial collaborative processes, despite shared digital workspaces being central to many contemporary platforms (see Supplementary Figure 5). This represents a missed opportunity to characterize how gaze alignment supports joint reference, turn-taking, and spatial negotiation in digital environments.

Multimodal approaches remain rare. The most common pairing combined EEG with eye-tracking or ECG, the latter likely reflecting existing EEG measurement infrastructures. Notably, we did not identify any hyperscanning studies that integrated EEG and fNIRS in digital interaction contexts, despite strong theoretical motivation. Such integration could jointly capture the spatial localization and temporal dynamics of inter-brain coupling (for review see Li et al., 2022), offering a more complete description of collaborative processes, especially in more applied, naturalistic contexts (Pinto-Orellana et al., 2024). The infrequent use of multimodal approaches in hyperscanning studies highlights an important area for future research, with the potential to significantly advance methodological rigor and ecological validity in this field.

Multimodal acquisition is not trivial: hardware integration, increased setup complexity and cost, paradigm design that accommodates diverse temporal and spatial constraints, and a lack of standardized pipelines for joint analysis can be challenging (Gado et al., 2023; Li et al., 2022; Pinto-Orellana et al., 2024). Nevertheless, multimodal acquisition can provide a uniquely holistic view of how neural, physiological, and behavioral dynamics systematically interact during collaborative interaction (Balconi and Angioletti, 2023; Gado et al., 2023; Li et al., 2022; Lühmann et al., 2020). Work outside collaborative digital contexts already illustrates this potential (Balconi and Angioletti, 2023; Zhao et al., 2023). For example, joint EEG-fNIRS acquisition during a motor synchronization task yielded complementary inter-brain coherence indices across modalities (Balconi and Angioletti, 2023). Hyperscanning in digital collaboration could benefit from adopting symmetric, unsupervised EEG-fNIRS fusion pipelines (for review see Codina et al., 2025) to capture hidden inter-brain coupling beyond standard functional connectivity metrics.

These examples highlight how multimodal setups can shift interpretation from isolated signals to integrated multidimensional models of collaboration. The scarcity of such approaches in digital collaboration studies indicates substantial untapped potential, especially as hardware synchronization and computational fusion methods advance (Codina et al., 2025; Dissanayake et al., 2025). This is particularly critical in digital contexts, where key interaction channels (e.g., eye contact, gestures, spatial proximity) are filtered, delayed, or absent. Different modalities exhibit varying degrees of sensitivity to diverse stimuli (Stuldreher et al., 2020), making multimodal metrics the most robust choice for studying complex, dynamic, and digitally mediated forms of collaboration.

In sum, future work should move beyond simple one-to-one modality-process mapping. In digitally mediated environments, where cognitive, perceptual, and communicative cues are systematically altered, each modality provides a complementary perspective on collaborative cognition. Designing studies that leverage these strengths in combination, while acknowledging their limitations, will be crucial for advancing the cognitive neuroscience of digital interaction.

Functional connectivity analyses (e.g., IBS measures) remain dominant as an analytical framework across hyperscanning modalities. This prevalence may reflect both conceptual accessibility and methodological convenience, especially in unimodal setups. Temporal-domain analyses were found to be common as well, especially in EEG, ECG, EDA, and eye-tracking, where high temporal resolution lends itself naturally to event-locked and synchrony-based metrics of coordination and attention aspects of collaboration. On the other hand, fNIRS studies frequently employed spatial-domain analyses to leverage its superior cortical specificity. These modality-aligned analytical preferences collectively restrict the kinds of collaborative processes that can be detected and meaningfully interpreted.

Interestingly, only four out of 45 studies incorporated measures of effective connectivity. A possible explanation could be that effective connectivity measures require explicit biophysical or statistical generative models and thus impose stricter assumptions on the data. Violations of these assumptions can produce misleading causal inferences (Friston, 2011). This makes effective measures harder to interpret and less reliable in dynamic social contexts with shared stimuli and mutual influence. Functional connectivity, on the other hand, is more robust and intuitive for the naturalistic data that increasingly characterize hyperscanning studies.

These methodological constraints have direct implications for the cognitive functions that become tractable targets of investigation. Although many collaborative tasks inherently engage multiple cognitive processes, a consistent pattern emerged: executive functions and social cognition are the primary targets of interest. This emphasis is expected, given their role in coordinating shared goals, managing joint attention, regulating turn-taking, and negotiating decisions during collaboration. More fine-grained characterization of the cognitive dynamics underlying collaboration will likely require multivariate and multimodal methods capable of disentangling overlapping components within complex tasks. Combining complementary modalities with data-driven methods may clarify how specific subprocesses, for example, shared attention, perspective taking, and goal monitoring, unfold over time and how they jointly support digitally mediated collaboration. A fine distinction such as this also holds potential for providing specific feedback on interaction processes that may be diminished or lost when interaction shifts from analog to digital (Chuang and Hsu, 2023; Pöysä-Tarhonen et al., 2021).

Given the limitations of conventional analyses for capturing these complex dynamics, machine learning represents an underutilized but promising avenue for advancing hyperscanning research. One reason for the underuse of machine learning probably lies in the field's early developmental stage: many research groups prioritize traditional statistical approaches to establish basic mechanisms of joint action before introducing more complex computational models that are difficult to interpret. Yet, the promise of machine learning goes beyond classification. Its strength lies in modeling non-linear, high-dimensional, and temporally evolving dependencies across multiple interacting processes. For instance, combining fNIRS and EEG features in intra-brain contexts has been shown to enable classification of cognitive workload via bivariate connectivity metrics, an approach readily adaptable to hyperscanning data (Cao et al., 2022). In digitally mediated environments, where communication channels are constrained or altered by technology, such dynamic, cross-modal modeling may be essential for explaining systematic differences between analog and digital collaboration.

A promising direction involves multimodal, physiology-informed machine learning that jointly integrates neural signals (EEG, fNIRS), autonomic measures (ECG, EDA), behavioral indicators (e.g., gaze, EMG), and contextual information into a shared, multidimensional representational space (Codina et al., 2025). Most current fusion strategies rely on methods such as feature concatenation or decision-level fusion, whereas more powerful approaches (e.g., joint independent component analysis) remain underused despite their ability to reveal latent, cross-domain structure (Codina et al., 2025). Embedding physiological priors, such as neurovascular coupling models or autonomic response patterns, directly into preprocessing and fusion could push hyperscanning beyond simple synchrony indices toward richer, context-aware models of collaborative interaction. Progress in this direction is tightly coupled to the availability of open-source, high-quality datasets. As highlighted in a recent overview of human synchronization datasets by (Velletaz et al. 2025), multimodal data remain scattered and underutilized. Expanding accessible, well-curated datasets will be critical for enabling advanced analytic developments and for refining our understanding of the fine-grained, interacting cognitive processes that support digitally mediated collaboration.

Several limitations of the present review should be acknowledged. First, by focusing on collaborative tasks with digital components, we excluded hyperscanning studies that are, for example, centered on imitation, purely economic exchange paradigms, or shared attention. The resulting distribution of paradigms does therefore not reflect the full landscape of hyperscanning research. Nonetheless, even within this constrained scope, we identified several more open-ended and applied designs, suggesting a broader trend toward ecologically valid applications. Second, we deliberately excluded fMRI and MEG to focus on modalities amenable to mobile and applied use. One could argue, however, that fMRI and MEG hyperscanning inherently involves digital elements by design. These modalities are therefore relevant to digital interaction. The associated living database is intended to be extended over time to incorporate such studies, enabling a more comprehensive overview. Notably, although the present review is static, the InterBrainDB platform is designed as a living, open-source resource. New studies can be added and re-labeled over time, and analyses can be replicated with different inclusion filters (e.g., adding modalities or relaxing collaboration criteria). In this way, some of the limitations of the present synthesis, particularly with respect to coverage and modality scope, can be progressively mitigated as the database grows.