A few years ago, we contacted the key cultural heritage organisations in Finland and asked what we thought was a simple question touching on central historical source material: “How many nineteenth-century letters do you have in your collections?” The answer was simple, though not quite what we expected: “There is no way of knowing.” This means that neither the organisations nor the humanities scholars who regularly use these materials have a comprehensive view of the extent or basic structural characteristics of these collections. Consequently, we cannot assess their inherent biases or distortions, nor can we evaluate the representativeness of the collections in relation to nineteenth-century “epistolary cultures”.

A reasonable follow-up question is why information about archived material is important. After all, scholars have successfully used the contents of epistolary collections to answer a wide variety of historical research questions since the establishment of national and other archival institutions and the development of historical methods based on primary sources. A lack of interest in the logics or “ethnographies” [Dirks 2002] of complete archival collections is likely due to hermeneutic and microhistorical research interests among scholars immersed in historical epistolary exchange. It is also due to the difficulty of obtaining accurate and comparable quantitative data on vast and diverse materials accumulated over decades or even centuries and catalogued in astonishingly diverse ways. However, when combined with the latest digital humanities technologies, collection-level archival metadata offers untapped potential for historical enquiry.

In this paper, we suggest that the research agenda of digital historians should be expanded to include questions about the composition of archival collections. We analyse a new digital archive constructed by combining letter metadata (e.g. information about senders, recipients, dates and quantities of letters) from databases, finding aids and archival catalogues of person and family archives held in various Finnish cultural heritage organisations. This “big metadata” [Enqvist and Pikkanen 2024] transcends organisational boundaries, enabling us to obtain quantifiable and comparable information about the collections. When we transform it into linked open data and connect it to external sources, such as biographical and other datasets and ontologies, we can enrich the data with categories not present in the archival metadata, such as the gender of letter writers and recipients. This allows us to observe relations that are simply not detectable in traditional archives of printed finding aids or organisational archival databases.

This paper argues that a fruitful intersection exists between digital humanities, computational humanities, and critical archival studies, which we refer to as “critical collection history”. While it may be possible to conduct such research using traditional methods, we aim to demonstrate that, at its most effective and ambitious, critical collection history is digital collection history. Repurposing information stored in archival finding aids and other repositories of archival metadata using computational methods provides us with an unprecedented and powerful tool with which to explore the interplay between historical source criticism and the archival politics of inclusion and exclusion (Whose materials are preserved by cultural heritage organisations and made available for scholarly and other purposes?). This approach also allows us to gain new perspectives on the interaction of historical actors within past contexts.

There is a well-established international scholarly tradition of compiling and editing carefully curated collections of correspondence. In the last 15 years, such data have been published as digital research resources and, in many cases, as linked open data. Projects such as the EU-funded large-scale Reassembling the Republic of Letters (RRL), the Circulation of Knowledge and Learned Practices in the 17th-century Dutch Republic (CKCC) and the German correspondence metadata aggregator correspSearch, mainly focus on the epistolary materials of learned men, because traditional epistolary editions have centred on this group.

Our material allows us to adopt a more comprehensive perspective. In what follows, we will address the gendered aspects of nineteenth-century epistolary collections in Finland. We are particularly interested in the traces that nineteenth-century women left in archival collections. The nineteenth century has been characterised as the golden age of letter writing [Lahtinen et al. 2011] and women have been identified as the main protagonists of this development [Monagle et al. 2023]. Does quantitative analysis of the collected epistolary metadata confirm these qualitative findings? Based on our data, did women play a prominent role in epistolary communication? Do they “own” their collections and archives? By delving into the collected metadata, can we find individuals who are not part of our current scholarly canon? Throughout the paper, we will revisit a broader question: Does the material tell us more about twentieth-century archival practices or particular developments in nineteenth-century society?

Compared to previous publications of linked open data on epistolary metadata, there are certain methodological and interpretative difficulties associated with the use of large collections of metadata gathered directly (“bottom-up”) from cultural heritage organisations. The primary challenges pertain to the commensurability of the archival collections themselves and the validation of the results. There are no comparable epistolary metadata datasets that we can use to measure the significance and/or Finnish and nineteenth-century specificity of our findings. The best approach is to use the entire dataset to provide context and calibration for its constituent parts (i.e. different organisational or thematic collections), and vice versa.

The rapid expansion of big data and digital humanities since the early 2000s has emphasised the importance of understanding the shape or contours of (digital) archives and corpora. The research framework proposed in this paper is similar to recent studies that have examined the “anatomies” of historical data collections and the processes involved in their creation, both before and during digitisation (e.g. [Tolonen et al. 2022], [Beelen et al. 2023], [Ortolja-Baird and Nyhan 2022]). It has been suggested that large digital datasets require a new kind of critical attitude; those who produce and use such data must pay particular attention to its local context: how it was acquired and prepared, its strengths and weaknesses, and whether it is fit for purpose [Loukissas 2019, 8, 195], [Ahnert et al. 2020, 54–55]. At the same time, it has been recognised that much digitised historical material can be analysed using computational methods. Indeed, quantitative analysis is particularly effective when we want to interrogate the characteristics, structures, and properties of digital archives and collections and reflect on the relationship between datasets and the phenomena they represent or model [Hyvönen et al. 2021, 53], [Guldi 2023, 29], [Bode 2020, 4–5].

Over the past few decades, researchers in archival and cultural studies have been addressing the same questions, albeit mostly in the context of analogue archives, close-reading individual collections. Since the archival turn of the 1990s and early 2000s, there has been growing interest in analysing the processes involved in creating, organising, appraising, and describing archival holdings, as well as the ways in which these processes reproduce and/or challenge existing hierarchies, dominant discourses, and ideologies ([Edquist 2021, 109]; see also [Schwartz and Cook 2002], [Taavetti 2016]; [Ketelaar 2017]; [Mikkola et al. 2019]. Critical archival studies pay attention to the processes prior to the possibility of digitisation. Like critical theory on a wider social scale, these studies are emancipatory in nature [Caswell et al. 2017, 2]. They seek to highlight injustices in current archival research and practice, offering recommendations for improvement. As with researchers working with large digital corpora, the importance of analysing the logics, classification systems and exclusions of archival collections has been emphasised [Dirks 2002]. Recently, digital techniques have emerged for studying curatorial practices and the historical and cultural context of museum curation and archival appraisal (e.g. [Salway and Baker 2020]; [van Lange et al. 2022], [Enqvist and Pikkanen 2024]). Paradigmatic quantitative studies of letter collections are the results of the aforementioned RRL project (e.g. [Hotson and Wallnig 2019]), as well as various case studies that use network analysis (e.g. [Ahnert and Ahnert 2019]).

Our study uses information from letter catalogues and databases from the Grand Duchy of Finland (1809–1917). However, as sets of correspondence often extend beyond the designated end year of 1917, we have also included these data where available. [1] The dataset is available online [2], “upcycling” these collections for reuse as FAIR data for research, as suggested by [Scheltjens 2023]. The current version of the dataset comprises around 1.2 million letters featuring approximately 114,052 unique actors (senders and recipients) from nine organisations and four digital editions (see Table 1). [3].

Letter catalogues have also been used to create datasets in various other projects. The information in these catalogues varies considerably (see [Kudella 2019] for more on modelling in the RRL project, for example). In our case, the archival metadata includes basic information about the senders and recipients (whether individuals, groups, or families). While dates are often clearly delineated in some well-curated collections, more often than not, they are given as the span of a set of correspondence (e.g., 25 letters between 1870–1895). Beyond archival organisations, we have knowledge of the archival collections in which the letters are held. In Finnish cultural heritage organisations, personal materials, including letters, are organised in separate “collections”, or “archives” as they are usually called. In the Anglophone world, such collections are often referred to as “papers” (for example, “the Cecil Papers”, referring to a collection gathered by William Cecil (1520–1598) and his son, Robert Cecil (1563–1612)). However, to avoid conceptual confusion between organisations and their collections, the term “fonds” will be used when discussing a body of records created and accumulated organically, reflecting the functions of its creator [Fonds SAA Dictionary].

The metadata are transformed into linked open data by an automatic transformation pipeline (see [Drobac et al. 2023b] for a detailed description). The resulting data model builds on international standards, such as CIDOC CRM ([Doerr 2003]), Dublin Core and ICA Records in Contexts, to promote interoperability with other datasets. The data model aims to facilitate the modelling of relevant metadata properties collected from source datasets. The most central classes in the current data model are “Letter”, “Actor”, “Place”, and “Time-Span”. Provenance (the MetadataRecord class) and archival/collection-level information (Fonds and Organisation classes) are also included.

To represent actors (senders and recipients of a letter) in different source datasets, we use an adapted version of the proxy concept from the Open Archives initiative Object Reuse and Exchange (OAI-ORE) specification. In our case, a proxy signifies a particular viewpoint on a person or group within a designated source. In the harmonisation process, proxies that are identified (through deduplication/disambiguation workflow [Drobac et al. 2023a], matching based on a trained probabilistic data linkage Splink model [4]) to represent the same person or group are linked through a shared instance of the class ProvidedActor. This class is an adaptation of the Europeana Data Model’s (Doerr et al., 2010) class ProvidedCHO (Provided Cultural Heritage Object). In Europeana, the ProvidedCHO “represents the Cultural Heritage Object that Europeana collects descriptions about”. In practice, a proxy represents a “local”, or source-specific, view of an actor (with metadata ingested from that source), whereas a ProvidedActor represents a “global” view of the actor by assembling all the information about the actor from various sources. The core classes and properties of the data model are presented in Table 2.

Our study is based on the availability of epistolary metadata as Linked Open Data (LOD). Having transformed the metadata collected from various cultural heritage organisations into a harmonised data model, we can analyse the resulting dataset as a whole, while taking into account the differences in archival practices between the organisations that provided us with data. As illustrated in Figure 1, our research can be extended from examining a single collection or the collections of a single archival organisation to exploring the combined dataset (or union catalogue), thereby capitalising on the inherent links between collections and organisations.

Figure 1 shows how many person-actors the different organisations have in common. Making the data available in a machine-readable, structured format (RDF) and in a SPARQL endpoint means that we have a multidimensional dataset that enables evolving historical research questions and allows us to query the data in rich and meaningful ways. Linking the data entities (e.g. person-actors) to external sources enriches the dataset with contextual information, providing ways to obtain details such as the biographical background of the people in the dataset. At the same time, possible links to external data sources, such as the National Biography of Finland or Wikidata, indicate that the person in question is recognised as relevant, at least from today’s perspective. The data for the studies reported in this paper were generated by formulating SPARQL queries to select subsets of the dataset.

In addition to SPARQL queries, we use the LetterSampo Finland semantic portal[5]. Based on the Sampo model [Hyvönen 2022], the portal is implemented using the Sampo-UI programming framework [Ikkala et al. 2022]. It allows users to search, browse and analyse letters, actors (people and organisations), archival organisations and collections (fonds), and places in the dataset. The interface is based on the faceted search paradigm [Tunkelang 2009] and enables users to search for letters from a specific time period, sent by a particular person and archived by a particular organisation, for example. Visualisations include the yearly distribution of letters, top correspondents, and correspondence networks. Data including letter-sending locations can be visualised on a map.

Some cultural heritage organisations have built up their collections over decades or even centuries, during which time cataloguing practices, descriptive concepts and standards have changed. In some cases, the material was initially handled by non-specialists. Combined with the creative individualism of archivists, this has resulted in a wide variety of finding aid formats, presenting one of the challenges of data modelling work and transforming letter catalogues into machine-readable data. The solution is to engage in continuous dialogue about humanistic knowledge interests and the emerging data model. A linked open data solution has proven suitable for modelling cultural heritage metadata relating to archives and historical figures. However, perhaps even more important is the way in which the modelling process forces us to consider the “things” being modelled. What are the catalogued collections of letters, and how can we conceptualise phenomena such as epistolary culture? As [Ciula et al. 2018, 345] write, models can be learnt from “at two different stages, in the creation of the model and in its application and successive manipulation”. In the case of epistolary metadata, modelling the letter collections has led us to focus on specific aspects of the collections themselves.

In addition to the properties provided by the linked data, we use network analysis to construct specific correspondence networks. These networks illustrate changes in women’s centrality over time, helping us to visualise the gendered structure of nineteenth-century archival collections. Our focus is on contacts established through correspondence rather than on the direction of interaction through letter-writing. The main reason for this is the partiality of our collections (more on this below). The intention behind these network visualisations is not to examine how the network functions, but rather to offer an alternative viewpoint on the women in our collection data through their interconnectedness. The layout used is a force-directed model, meaning that the most important nodes in terms of connectivity remain at the centre of the network. Although our primary focus is on the networks as collections, we also highlight other themes relating to nineteenth-century women’s historical correspondence as reflected in our dataset.

One way to challenge the boundaries of archival organisations and their collections is to create a virtual collection of metadata relating to women’s letters. We will proceed in three steps, firstly discussing the quantitative results in terms of the total number of women actors and women as so-called records creators, then examining the ten most active women in terms of letters, and finally exploring the position of women in specific correspondence networks over time. Our findings will be contextualised in relation to archival practices, scholarly focus, and the modernisation of nineteenth-century society. Finally, we will assess the representativeness of our dataset as a whole. To this end, the Åbo Akademi University Library and the Finnish Literature Society will serve as sample datasets and main points of comparison. These organisations have exceptional collections covering different geographical regions and representing the two national languages, Swedish and Finnish.

The current dataset comprises 114,052 actors, 24.1% of whom are female. While this figure is higher than that observed in previous European projects, it is less pronounced than expected based on research emphasising that women’s epistolary activity peaked in the nineteenth century [Monagle et al. 2023]. However, when we filter out groups and administrative units, and consider all person-actors in the dataset, the percentage of female actors rises to 29%. In the Åbo Akademi data, the proportion of female actors is 31.7%, and in the Finnish Literature Society data, it is 32.5%. Furthermore, of the 1,246,573 letters, on average 38% were sent by women (alone or together with someone), and the corresponding percentage of letters received was 36%.

In addition to examining the characteristics of the entire pool of person-actors, collecting and enriching epistolary metadata gives us access to another class of actors. In archival parlance, these are known as “records creators” or “entities of origin”. These are individuals, families, corporations or administrative units that “create, receive or accumulate a body of records, personal papers or objects” [Entity of Origin SAA Dictionary]. Their names are usually found in catalogue headings and indexes of letter collections. Archival fonds contain letters received by such entities, but often also letters sent by them. These may be copies kept by the entity or originals added to the collection at a later date. Often, there are also letters sent and received by other entities that may have had a family or other connection with the original records creator. The Finnish Literature Society provides a basic list of records creators on its website to serve as an entry point to its epistolary material. In contrast, the Åbo Akademi University Library provides online access both to its epistolary database and the database of records creators; however, it remains difficult to find and combine information about actors who are not records creators, even if they had extensive collections.

By comparing the total dataset of person-actors with that of records creators, we can see that the status of records creator is granted to a select few. Although the number of actors varies greatly between our two sample organisations, only around 2% of them are also records creators in both cases. In the case of the Åbo Akademi, we can see that the gender of records creators is relatively close to that of the overall actor data, but it is significantly higher in the Finnish Literature Society’s collections (37.5% female). By contrast, the National Library of Finland has a very ‘masculine’ catalogue of epistolary fonds: 85% of records creators are men, with only 14.6% being women. These differences can be explained by the archival policies and constituencies of these organisations. The Åbo Akademi University in Turku was founded in 1919 to serve the Swedish-speaking minority in newly independent Finland. Its collections include materials representing the old Swedish-speaking aristocracy and intelligentsia — the country’s early political and cultural elite. In contrast, the Finnish Literature Society, founded in 1831, has a very different profile. Its collections reflect the ideas of nineteenth-century Finnish-language nationalists in the fields of folklore studies and Finnish language and literature. The collections include the personal archives of literary writers, translators, literary critics, and others. Through the National Library’s connection to the University of Helsinki, professors, researchers, and other academics are well represented in the library’s archival collections as professional groups.

​​Linking to external data sources provides additional insight into the actors and records creators, as well as the general characteristics of the entire dataset. Only 19% of all actors can be enriched with information provided by links to databases and repositories of historical narratives that can be captured with digital methods. However, these individuals are responsible for 85% of all letters. This means that we have a very prominent group of individuals who have left an extensive trace of themselves in the form of letters and other personal papers, despite being few in number, and a large number of individuals with few letters and connections. Such “long tails” have been observed in other epistolary collections of a different nature [Ahnert and Ahnert 2019, 31]. When it comes to the narrower group of records creators, however, about 80% of them can be linked to external resources, suggesting a continuing scholarly interest in them. Collected lists of records creators can therefore serve as a kind of “who’s who” of nineteenth-century Finland and nineteenth-century studies. The collections of prominent records creators have also preserved the correspondence of individuals who would otherwise have left little or no archival trace.

Thus, we can see that nineteenth-century epistolary collections in general, and the group of records creators in particular, are dominated by literate elites and the middle classes. This is consistent with what we know about the development of literacy in this north-eastern corner of Europe. At the beginning of the nineteenth century, writing skills were still rare, despite the fact that half of all Finns could read. It was not until the introduction of primary schools in the 1860s that literacy and text comprehension really began to spread. Statistics from the Evangelical Lutheran Church indicate that, by 1880, 86.3% of the population could read, though only 12.4% were literate. By 1900, the literacy rate had risen to 38.6%, reaching 55.3% ten years later [Latomaa and Nuolijärvi 2002]. The educated elite and middle classes also appear to have taken care to preserve their correspondence in personal, family and manor archives. These archives were transferred to cultural heritage organisations when they began collecting private archival material in the late nineteenth and early twentieth centuries.

These organisations operate under the principle of provenance (respect des fonds), which emphasises preserving records according to their origin and within the units in which they were collected. However, a digital, virtual metadata archive provides an opportunity to challenge established practices and offer new perspectives on archival materials. For instance, we can explore the collections of notable politicians and academics, or large family holdings, and ‘liberate’ mothers, sisters, and daughters by granting them independent status as virtual records creators, thereby ensuring their immediate visibility and discoverability. One such individual is Aline Reuter (née Procopé).

Table 3 presents the ten most active women in the entire dataset, ranked by the number of letters. Interestingly, only half of these women are linked to content-rich external resources; that is to say, they have a Wikipedia page or an entry in the National Biography of Finland. For these women, we have readily available occupational information. In contrast, much less information is at our disposal for the other four women – the “dark horses” of our virtual archive. Aline Reuter’s correspondence at the Åbo Akademi comprises 8,178 letters (3,072 sent and 5,106 received). This makes it one of the largest collections of letters from a nineteenth-century woman in our current dataset. You might think that such a large collection, covering a period considered so formative for Finnish nationalism and studied by generations of historians, would be used extensively. This does not seem to be the case, however. The few mentions that shed light on her life can be found in the National Biography entries of her three famous professor sons and in the Reuter family entry (see, for example, [Autio 1997a] and [Autio 1997b]). There are also Wikipedia pages (e.g. “Edvin Titus Feodor Reuter”) and a Geni.com profile (“Aline Reuter”) which, for technical reasons, cannot be used to enrich the data. These sources provide information about her lifespan (1828–1916) and some details about her family (husband and four sons).

However, obituaries of Aline Reuter in the newspapers provide further insights into her life. She had 13 children in total, many of whom died young. She homeschooled the surviving children until they reached grammar school age, and she acquired such excellent skills in subjects such as Latin and mathematics that, when her son Odo Reuter, a professor of zoology, became blind, Aline Reuter (now in her 80s) took over his professional correspondence and proofread his papers. However, more than her obvious intellectual agility, the obituaries emphasise her “rich inner life” and remind readers that “there were no great events in her life” (see, for example, [Dagens Press 1916]). This statement alone arouses curiosity about the contents of Reuter’s extensive collection of letters, tempting one to immerse oneself in her “epistolary space” [How 2003] and reflect on its representativeness, for example with regard to the experienced, personal nationalism of nineteenth-century women [Eiranen 2021] or other significant ways of discursively shaping oneself and one’s surroundings [Monagle et al. 2023, 12].

“Mother Aline” is thus a prime example of a marginalised or forgotten person who only becomes clearly visible when we recontextualise her letters in the collected and enriched metadata. Indeed, it is only through the combined metadata that we can confirm the extraordinary quantity of surviving correspondence. By way of comparison, the most extensive set of male correspondence is that of Senator Leopold (Leo) Mechelin (1839–1914), whose entire collection of letters from various organisations totals 14,399.

We can make further observations about the gender distribution in the Finnish nineteenth-century data with the help of network analysis and simple degree metrics, which build on the above analysis of the number of contacts each person had. This enables us to visually assess the extent to which the women in our letter collections were connected. As our dataset spans over a hundred years, we decided to examine two periods in more detail: 1830–1860 and 1880–1910, which reflect the socio-economic, legal and social changes that affected women’s correspondence in the modernising Grand Duchy of Finland. From the 1860s onwards, women gradually gained access to higher education. As the country’s economy liberalised, women’s legal and economic rights were strengthened, and they participated in associations and societal movements. Hypothetically, this meant that women’s epistolary networks became more similar to those of men, and we were interested to see if we could find traces of this in the archived epistolary material.

As expected, looking at the position of women in the correspondence networks reveals that their connectivity increases as we move from the first to the second period. At the same time, the “archival status” of the most well-connected women and their links to external biographical databases undergoes a significant change. As illustrated in Figure 2, during the initial period of letter correspondence (1830–1860), men dominate the centre of the network, with most women (represented by blue nodes) situated in the middle and outer spheres. The ten most connected women are also not central figures from an archival perspective (see Appendix 2, Table 5). As they are not records creators, the bulk of their letters have been archived as part of their families’, husbands’ or sons’ fonds. Furthermore, linking to external databases has only resulted in links to genealogical databases. Professional information was only found for two women (Fredrika Runeberg and Elisabeth Blomqvist), who have also attracted academic interest (e.g. [Klinge 2006] [Konttinen 2000]). The situation is quite different for the ten most connected men of this period (see Appendix 2, Table 4): they are all records creators; their letters originate from 7–9 archival organisations; and they are well linked to external sources, including Wikipedia.

During the second period (1880–1910), women were more centrally positioned within the network, as illustrated in Figure 3. The profile of the women with the most connections in this later period is very different to that in the earlier period: nine out of ten of these women are records creators, and their letters are widely distributed across different archival organisations (see Appendix 2, Table 7). These women are well-known cultural figures – artists, writers and politicians – and are well connected to external data sources. They also have a Wikipedia entry. Their network profiles thus differ significantly from those of lesser-known figures such as “Mother Aline”, who, despite her extensive correspondence, plays a more peripheral role in the network. In both periods, the biggest difference in our metadata between men and women is that women have significantly fewer contacts, even those who worked in the public and private sectors in the late nineteenth century (such as the social activist and politician Aleksandra Gripenberg; see Table 3 and Appendix 2, Table 7 and Appendix 3, Table 9). Based on archival materials, women’s epistolary networks were narrower. However, as noted above, the ties were proportionally stronger in terms of the number of letters exchanged.

In addition, networks consisting exclusively of correspondence between women provide further insight into the organisation of archival collections, the focus of scholarly interest and the changes brought about by the modernisation of the country. The configurations of women’s correspondence networks are illustrated in Figure 4 for the periods 1830–1860 and 1880–1910. During the earlier period, women exchanged significantly less correspondence, and their networks formed several distinct but small clusters. While most of the women are the same as in the full network, the shape of the network now suggests a predominance of family networks, as previous research [Monagle et al. 2023] has suggested. However, further data analysis and qualitative research are needed to confirm this. Additionally, many of the women’s letters are uniquely archived at Åbo Akademi, likely due to the institution’s substantial collection of family archives from the Swedish-speaking population. For the second period (1880–1910), the situation is very different, as can be seen from the shape and size of the network in Figure 4. The women were highly interconnected, with significant nodes at the centre of the network. Examining the most central women reveals that, with the exception of two individuals, all of those in the “women's network” and the broader network remain the same during this period, albeit with slightly different rankings (see Appendix 3). Therefore, in the latter period, it is not possible to identify separate clusters of women’s correspondence. Instead, women’s correspondence is interwoven with the networks of the country’s artistic and cultural elite, whose correspondence is spread across all the archival organisations in our dataset.

Despite these extensive collections, archived materials only provide patchy and partial access to nineteenth-century epistolary communication. When we compare the dataset — and thus the collections of the largest Finnish cultural heritage organisations — with statistics on all letters sent in the Grand Duchy of Finland, it is clear that coverage and representativeness are low (Figure 5). The collected metadata does not reflect the growth in letter correspondence in the country, which was driven by the expansion of trade, commerce and administration, as well as the development of technologies that supported easy and affordable letter exchange, such as cheap postage and stationery, prepayment of letters and increasing delivery speeds (Pietiäinen, 1988). After the mid-1880s, when literacy rates began to rise, archival coverage decreased further still. It appears that archival organisations lacked the means or interest to preserve everyday correspondence in the Grand Duchy, and lower-class individuals and small businesses may not have had the inclination to create personal archives. Therefore, an interesting question is whether we can estimate what materials are missing. Can we see what is not there?

At the start of the data collection, we asked cultural heritage organisations to estimate the proportion of their collections that were unorganised and uncatalogued (see Appendix 1). The results of this survey clearly demonstrate a growing shortage of resources for the labour-intensive work of archival description. However, it is reasonable to assume that letters added to collections since the nineteenth century, perhaps up to the 1980s, were more thoroughly indexed due to the greater availability of human resources relative to the size of the collections. Nevertheless, the collections do not represent a random sample of nineteenth-century epistolary culture, as evidenced by the social and cultural status of most of the records creators. Therefore, while cultural heritage organisations’ collections are not representative of the majority of letters sent, especially in the late nineteenth century, they do reflect the membership of the elite and upper middle classes relatively well.

However, what has been preserved of the total body of correspondence that once existed is often coincidental, even among the literate elite. For instance, Senator Mechelin received 12,129 letters, which are kept in his family’s collection at the National Archives. However, we only have 2,270 of his letters (mainly in the collections of other individuals in five different organisations). Assuming that people mostly replied to the letters they received, this kind of disparity in the material points to definite gaps in the collections. Conversely, it is more challenging to determine whether the proportion of female actors (24% of the total, and 29% of those for whom gender can be identified) and the proportion of letters sent and received by them (38% and 36%, respectively) is indicative of missing material. Similar, large-scale datasets of epistolary metadata from other European countries are lacking, making it difficult to compare our findings and results. The closest we have are the results of the ambitious RRL projects mentioned above, which provide a rare insight into the gender distribution of epistolary material from earlier centuries. The datasets currently available in Germany and the Netherlands show an overwhelmingly male correspondence of 85%, but this is understandable given the projects’ premises.

In the era of flourishing research on LGBTQIA+ histories, exploring the role of women as archival protagonists may not seem very revolutionary. In fact, the first attempt to enable sustained research on women’s history in Finland was made in the mid-1990s with the publication of Naisia, asiakirjoja, arkistoja: Suomen naishistorian arkistolähteitä (Women, Documents, Archives: Sources for Women’s History in Finland, [Härkönen 1994]), which identified various archival collections of potential interest to the emerging field of gender history. However, as we have demonstrated, applying data science methods can render the process of critically revising archival collections much more systematic and powerful. Epistolary metadata is, of course, only one type of information that can be used to interrogate the role and visibility of different social, cultural, or ethnic communities in our social narratives. Indeed, cultural heritage metadata may offer a more affordable and realistic way of making dispersed archival and museum collections accessible and researchable than mass digitisation projects. However, metadata alone can be frustrating for scholars accustomed to interrogating vast textual archives, and a dialogue between digital and more traditional research methods is required. Nevertheless, using “pure” metadata enables us to “map missing bricks in the wall of knowledge”, for instance by categorising individuals and social groups whose biographies do not yet exist [Guldi 2024, 525] – a task we have begun with the help of Aline Reuter, among others.

Jo Guldi has recently referred to such digital approaches as “counting the silences” [Guldi 2024, 524]. “Silence” is a buzzword used by scholars to refer to missing data or issues that archives cannot or will not ‘talk about’. It is often used interchangeably with terms such as ‘gaps’, ‘bias’ and ‘occlusion’ (see also [Carter 2006, 217]. To clarify, [Guldi 2023, 30–37] distinguishes between occlusion and bias. For Guldi, a digital humanist who works with large textual corpora, the Occluded Archive is an archive that is closed to view. It consists of materials that we cannot access for various reasons. Occlusion operates at the level of political and historical factors, as well as archival organisation (e.g. which social groups are literate, which archives are preserved, and how relevant documents are hidden in separate datasets). For Guldi, bias is “dirt” or “filth” in the textual content. This powerful metaphor refers to racist, misogynistic and other expressions and narratives that carry a heavy ideological burden. Historians are usually aware of such omissions and biases, which they identify as part of their work. However, these factors are not always considered when data science methods are applied to materials.

Our data consist of metadata from the letter collections; we do not have the content of the letters in our dataset. This means that we need to introduce an additional feature of occlusion: the partly accidental and partly selective nature of the entire life cycle of historical and archived materials. Many archival theorists interpret “silences” as tangible gaps in the archival record. These gaps can arise from everyday practices such as the accidental disappearance or deliberate destruction of materials prior to and after archiving, or from appraisal and subsequent rejection in a cultural heritage organisation ([Fowler 2017]; [Edquist 2021]). The historical reasons for this kind of fundamental, irreversible non-preservation vary, and there is little evidence of such processes producing structural bias in historical sources. However, it may be possible to model some gaps using digital methods or conduct the time-consuming task of searching organisational records for traces of decision-making processes. In this paper, we have identified both gaps (in the representativeness of the data, compared with the information available on nineteenth-century postal activity and literacy) and silences (in the lack of research on certain notable women in our dataset) by examining what has been preserved in the collections of Finnish cultural heritage organisations. A persistent “silence” is often only recognised in retrospect, for example, when a paradigmatic change in theory or technology allows a new view of the archived material.

​​It is true that material separated into different repositories or organisations is a strong occluding factor. However, occlusion also occurs within individual organisations. A key factor here is how organisations prioritise the creation of metadata and categorise catalogued documents as open and accessible or closed (e.g. [Sherratt 2015]). Only collections that have been organised and described can be made available to researchers and other users (see Appendix 1). Furthermore, it has been argued that creating a fonds and assigning a records creator carries significant intellectual and hegemonic weight, yet these decisions often remain invisible [Drucker 2021, 326], [Friedrich 2021, 312–317].

Digital humanities scholars often work with materials that were originally indexed in a library context. While these bibliographic databases, catalogues, and digitised repositories are incomplete and less representative than many scholars might hope (see, for example, [Bode 2020]), they do comprise fairly homogeneous sets of materials. Furthermore, issues of archival access and its political constraints are frequently examined in the context of sensitive government records (e.g. [Risam 2019]; [Sherratt 2015]). However, it is also important to recognise these mechanisms in relation to cultural heritage materials and those held in private archival organisations. All organisations whose letter metadata are included in our dataset have their own acquisition policies, enabling them to accept or reject archival material offered to them, provided they justify their decisions. Compared to material generated in library contexts and governmental organisation records, archival organisations’ vast collections, particularly person and family collections, seem more “wild”, especially when considering material beyond the high research canon. Cataloguing categories vary, information comes in different formats, and many collections are only partially organised, with huge backlogs of unorganised material.

In this paper, we introduce the framework of Critical Collection History. Using the powerful qualities of Linked Open Data, this framework finds relevant features in a dataset comprising over a million letters and over 100,000 actors. It is motivated by discussions in the fields of critical archival studies and digital humanities. We argue that critical collection history occurs when the boundaries of cultural heritage organisations and their collections are erased and the structure of traditional paper archives is challenged and “weirded”. At the most general level, critical collection history asks whose actions and experiences are important enough to be archived and catalogued, and which individuals and groups are visible in finding aids, indexes and external research repositories. Digital technologies enable us to examine archival collections individually and collectively as large sets of historical data, not as static and inherently objective entities providing direct access to the past, but as multi-layered, dynamic and processual entities resulting from various historical, socio-economic and political processes. Similar to many contextualising approaches to digital collections (e.g. [Beelen et al. 2023]), the aim is not to “correct”, but to identify how such processes and layers manifest in the collections in question. Furthermore, we are not merely users of archival data; we are also actively involved in making the archival record more visible (or ‘louder’) in the digital space (Johnson, 2017). For instance, we make decisions about modelling the data and solving problems related to resource constraints when obtaining source data ([Drobac et al. 2023b]).

This article discusses materials that mainly comprise person and family collections held in private archival organisations. These collections are widely used by scholars and favoured as case studies in archival research [Edquist 2021], yet they are rarely considered in postcolonial and other critical approaches. As a case study, we have created a virtual archive of “epistolary women”. Our analyses suggest that the collections available to us as linked data are relatively representative of the upper echelons of nineteenth-century society. However, it is more difficult to assess whether the proportion of female person-actors (29% of all actors and 26–34% of the more filtered group of records creators in two sample organisations) accurately represents women’s epistolary activities. Nevertheless, network analysis indicates that the available epistolary metadata reflects changes in women’s societal roles brought about by modernisation. This demonstrates that even partial datasets can reveal significant historical phenomena and be profitably used by humanists, provided the gaps or ‘missingness’ are understood and acknowledged [Ryan and Ahnert 2021].

In addition, we have identified certain epistemic silences: women who do not feature in our current research canon but who have substantial archival collections that could offer valuable insights into a variety of historical topics. Overall, the results are tentative but promising. However, we must bear in mind that the current data include four online publications of letters and letter metadata from only nine cultural heritage organisations – albeit the largest and most central – and only from those collections that have been indexed at the level of individual letter exchanges. Nonetheless, despite all the gaps and omissions discussed in this paper, the collected metadata provides the most comprehensive sample available of who actually sent and received letters in nineteenth-century Finland.

This article has demonstrated the potential of linked open data and critical collection history in identifying gendered structures within epistolary metadata collections. In the future, we can continue to explore women’'s correspondence as a historical phenomenon by focusing more qualitatively on the women we identify using data-driven methods. Furthermore, we can direct our historical research towards archival collections as the outcome of specific political, cultural, and socio-economic pressures – pressures that leave their mark, turning the archives themselves into artefacts of history [Burton 2005, 6]. To achieve this, we should digitise the acquisition registers of various organisations and study their decision-making processes relating to archival appraisal where possible. While many organisational histories touch on these issues – most recently [Hakala 2023] study of the Swedish Literature Society in Finland’s archival collections – a combined history focusing on private epistolary collections as a case study remains to be written. A critical collection history is also always a history of the construction of collective identity; it can reveal whose experiences are available to us as historians, and also whose experiences have been narrativised as history at different stages in the past.

The project Constellations of Correspondence – Large and Small Networks of Epistolary Exchange in the Grand Duchy of Finland (CoCo) has launched a survey on nineteenth-century letter collections in Finnish cultural heritage organisations. To date, 57 archives, libraries and museums have responded. The survey contains 18 questions to help understand the current situation and the challenges of aggregating collection metadata. Here, we present results relevant to this paper in Figures 6-8. In summary, 86% of respondents either have or possibly have nineteenth-century letters in their archival collections. The word ‘possibly’ is explained by the fact that only 33% have catalogued all their letters. Conversely, 14% have not catalogued them at all. Regarding archival finding aids, only 36% of respondents with nineteenth-century letters have catalogues in electronic form, such as databases, Excel files, or Word documents. This makes the content machine-readable, but additional structuring and harmonising is required.

Computing resources of the CSC – IT Center for Science were used in our work.