1Jesse Shera’s important body of work in social epistemology (Egan & Shera, 1952; Shera, 1961), especially as it relates to bibliographic control and organization (1951), continues to serve as a point of departure for authors in library and information science (LIS) (e.g., Budd, 2002). Social epistemology has also been applied in knowledge organization (KO) (Furner, 2018). Social epistemology and the context for production is tightly tied to some of the quintessential areas of study in KO, including the creation and use of knowledge organization systems (KOSs), especially the structures employed in classification and the terminology used in controlled vocabularies.
2The concept of data is notoriously difficult to define. Christine Borgman (2016) reminds us that data “may be facts, sources of evidence, or principles of argument that are used to assert truth or reality” (p. 17). Borgman (2012) had previously noted that the properties of data can include “samples, software, field notes, code books, instrument calibrations, archival records, or a myriad of other information objects” (p. 1072), many of which may be used for research and the creation of new knowledge. Furthermore, Borgmann et al. (2012) includes the “letters and symbols that describe an object, idea, condition, situation or other factors” (p.485), and Uhlir and Cohen included “digital manifestations of literature (including text, sound, still images, moving images, models, games, or simulations)” (Uhlier & Cohen, 2011, c.b. Borgmann et al., 2012)), as data. Research data can also be considered as an equally difficult notion to define; in reality “data can be represented in units of any size, whether pixels, photons, characters strokes, letters, words, cells in a spreadsheet, datasets, or data archives” (Borgman, 2016 : 50). Data, according to Buckland (2018), are often numeric or fragments from within one or more documents. Fragments can be recombined to provide different insights, as datasets are often extracted, reshaped, and refined to illustrate different patterns through statistical operations and abstractions. These data transformations require domain expertise, also emphasized by Borgmann (2012). Eschewing the oversimplification of research data as just datasets, which are often conceptualized as spreadsheets or other machine-readable file meant for number crunching, this paper explores the concept of data, especially research data, more broadly as it relates to organizable and transformable resources.
3There is, however, a dearth of information on the place of data of any kind in KO and specifically, how concepts such as social epistemology play out relative to the role and function of research data in the bibliographic universe. In this piece, our fundamental argument is that research data are equivalent to, and as important as, other knowledge resources subject to bibliographic control. They are an information object subject to instantiation (Smiraglia, 2008), further hastening the argument that research data are therefore not only worthy of study in KO, but must be investigated. In part data as information objects must be studied to elucidate how to leverage such objects in a document-centric paradigm of organizing the products of human thought. In other words, data is not agnostic; as a community, we need to explore the role of data in shaping research goals, outcomes, aligning to disciplinary identities, and the nature of data itself. In order to begin this effort, this paper examines the foundations of knowledge organization’s documentary roots, and in doing so, establishes the connection between information “objects” that may be more traditionally subjected to bibliographic control, and to the many current definitions and interpretations of “data”.
4Communication in general is often framed using Saussure’s (1959) semiotics, and Peirce’s (1992) semiology in the field of knowledge organization; as data is a series of symbols to communicate meaning, discussing data as a series of signs and symbols understood through the process of communication makes sense. Semiotics and semiology are both theoretical models representing the flow of communication from a sender to a receiver. Saussure’s model is a dichotomous representation of a sign passing from the signifier to the signified. Peirce used a three-part model included a representamen (representation), object, and interpretant that are used by both the sender and receiver of a message. In our case, data are the observed and encoded components of a message. Data is symbolic in nature in order to facilitate its communication. Simultaneously, data is an abstraction of an observation; it requires the sender possess the expertise to assemble a meaningful message that can be interpreted by a receiver of the message.
5The type of symbols used to represent data, as well as the data selected to be represented, depend on the intent of the message. One researcher’s finished product (e.g. a published document) may well be data for another research (e.g. a record in a natural language corpus). In such a case, the published document then becomes data. If, indeed, data was encoded as and intended to serve as data, its meaning may be altered or reinterpreted with the added lenses of the passage of time, epistemological stances of other research communities, or at the individual level of understanding. Signs bear meaning steeped in the traditions of their research communities of origin, and are mutable based on their context. Research data assembled to communicate the observations of one researcher to another is typically done with data assembled in some form of document.
6The creation of any definition requires social consensus, reflecting a shared epistemological stance. Already, “knowledge…has been deemed meaningful through some socially agreed-upon process” (Budd, 2002: 93). Like documents that support knowledge sharing, data also should be meaningful, as in that meaning can be made from the item, be it in the form of a spreadsheet, metadata about geographic location embedded in a digital photo, or evidentiary data including the primary source documents and objects studied by humanists (Posner, 2015). Defining data requires social consensus grounded in experience and domain expertise, and serves as a reflection of a particular social epistemological stance or research paradigm.
7Analyses of data and their respective relationships to documents vary greatly in both proximity and precision of the relationship between the two terms (Furner, 2016), and, we note, do not necessarily clarify their relationship. Furner advocated for an informational, as opposed to computational, interpretation of data and documents as it affords the ability to name (and thus we note, control), frequently used concepts. Furner identifies the document as primary, preceding data. Buckland (2018) identifies documents as being related to “data, facts, texts, works, information, knowledge, signs, and other documents,” (p. 1) and having tangible (“physical”), cognitive, and social aspects. Buckland notes that what is called a document is ultimately a pragmatic decision, and the contrast between data and documents are a matter of form—texts are often considered to be documents, and numerical representations are often referred to as data. Hjørland (2019) contrasted definitions of data and documents, claiming that although the two concepts are interrelated, they remain discrete. In his comparison, data are produced for a purpose (p.699), while documents exist or are created in support of a fact (drawn from Briet). Hjørland also states that there is not sufficient differentiation between data and documents by those who study and reference such information objects, referencing several sources.
8By extension, data is agreed to be meaningful through similar, if not identical, processes to other, more traditionally-addressed documents. For example, the collection, creation, assemblage, and management of research data all requires consensus. In the same way that the evidentiary aspects of a document render it a primary source document, and one that can serve as an object of study (Moulaison-Sandy & Wenzel, 2023), data can also be a primary source document with the potential to inform. As a research method should parallel epistemology (Darlaston-Jones, 2007), the defined form of data should align (or be shaped to align) to the research questions at hand. Darlaston-Jones, whose home discipline is psychology, clearly outlines the natural flow from theory to method. This same philosophical flow is emphasized in LIS pedagogy, but the articulation of this flow is rarely discussed outside of the classroom. The flow of research question, to research method, to research data is open to discussion; the form of the research data is not neatly prescribed and should be given flexibility in form and consideration as its own object that shapes research. In many cases, especially in social sciences and humanities scholarship, a research question might best be addressed through the analysis of data in the form of an artifact worthy of study and analysis (Moulaison-Sandy & Wenzel, 2023).
9In the francophone tradition, according to Michael Buckland (1998a), Paul Otlet (1934) extended the definition of “document” to include “object” “if you are informed by observation of them” (p. 216). The technical definition of document devised by the International Institute for Intellectual Cooperation in conjunction with the Union française des organismes de documentation, states that a document is “any source of information, in material form, capable of being used for reference or study as an authority” (Buckland, 1998a : 216). In this same vein, 1951 Suzanne Briet published a treatise entitled “What Is Documentation?” that has been criticized for its insistence on the “cultural dimension” (Coadic, 2007). Like her predecessors, however, Briet arrives at the notion that evidence is more than text-based documents.
10In alignment with this mid-century francophone approach to considering that nature of documents as being evidentiary, we place ourselves alongside current approaches to social sciences and humanities (including digital humanities) research. Similar to the aforementioned francophone approach, Drucker (2011) uses a humanistic lens, reconceptualizing data from the latin “given” (Drucker, 2011; Furner, 2016), to “capta,” or something that is taken or constructed. Using such a humanistic lens, any object that can be studied can be “data;” however, it must be constructed into a surrogate representation, and that representation must correspond to “facts” (Furner, 2016: 297). Data has been more than just “facts,” but was only the “givens” until the 20st century (Seadle, 2021), after which the notion of data became more computational. Traditional datasets or spreadsheets are equivalent to the collected (or curated) antelope in the zoo (Briet, 1951); as such, data deserve the same degree in knowledge organization as other resources, since data of all kinds have equal potential to be informative.
11More recently, Buckland (2018) has established that data is itself an instantiation of knowledge; instantiation being when iterations of information content exist simultaneously (e.g. all versions up to or including the “current” or “most recent” versions of any information object) (Smiraglia, 2008). Representations of data exclude the “primary source,” of the document or documentation of data, as such representations do not include appropriate metadata to make the source findable as an information object. The idea that text (or other information) can be stored in memory, either in the memory of an individual or that of an electronic device in order to enable the future copying and transmission of text or information (or text as information), was explored in detail by Wilson (1962). He further argued that such memory-stored text is itself a type of information object, and therefore resides within the scope of things that are subject to bibliographic control. Data is also such an object, and can be stored in working human, electronically stored, chiseled in marble, or otherwise recorded in memory (human or machine); as such it is an object that is subject to bibliographic control and is therefore an object of interest to knowledge organization. Next, we discuss in more detail the idea of data as an object, as well as parallels between data objects and other types of objects subject to bibliographic control.
12Data as an object that is subject to bibliographic control is an extension of information. As noted by several scholars (including Day, 2014; Furner, 2016; Buckland, 2018; Hjørland, 2019), information is itself phenomenon that serves as the base concept of what it is that we analyze, describe, and organize in knowledge organization is more often discussed than data. In fact, the delineation between data and information has been described as having a “hazy border” (Borgmann, 2007: 41), and the delineation between information and data is not well defined (Hjørland, 2019). Furner (2016) discussed several historical models of data as a type of under-addressed information phenomenon, emphasizing that “information” is primarily identified as a foundation for knowledge while “data” are neglected. His argument is that the document, and not the data, are the primary concept. Assuming the document as primary shapes a framework in which the document, the primary form of data easily allows for data to be information objects in the form of documents.
13Wilson (1962) emphasized that the distinctions between a work, text, and as we argue, data, are “less easy to make with any precision” (p. 7). Extending this reasoning to the notion of the instantiation of data is a natural fit. Instantiation is distinguished into two types; “derivations” which describe instantiations that have little to no change from the initial work, and “mutations” which describe instantiations of works that were more extensively altered (2008: 11). Works, and data, can be (and are) purpose-tailored, as in the case of Buckland (see: 1997, 1998a, 1998b), where the same work was reframed to suit the venue and audience for each instantiation of the publication, “What is a digital document?”
14In academia research data are encouraged to not only be shared, but to be reused, despite the evidence to the contrary (Pasquetto et al., 2017). Reuse of research data runs in contrast to the practice of openly repurposing textual publications, as it is often discouraged, as in the aforementioned case of Buckland 1997, 1998a, 1998b. Buckland’s works serve as an example of neatly traceable instantiation of works that self-cite in such a way as to build on the original work for a different disciplinary audience; this practice could be applied to publications reusing research data in order to clarify their provenance. Pasquetto et al. (2017) surveyed studies of research data sharing and reuse, finding in their meta-analysis that most data “sharing” was done on private servers rather than in public repositories (Jasny et al., 2011). In order to reuse research data, it must be open research data, or kept in a manner that can be searched (e.g. has appropriate searchable metadata and is linked to from a database), then accessed upon request by a researcher. The reuse of research data requires a process of curating what is needed and applying appropriate transformations, such as reorganizing data to be used by a specific tool, or selecting examples or statistical of data to include to represent high-level observations and abstractions about the data itself. Such acts of transformation mirror the phenomenon of instantiation (Smiraglia, 2001, 2004, 2008), allowing for discrete versions stemming from an original instance of a dataset to exist in a state of multiplicity through time and as recorded in information systems. Each extraction and transformation of a dataset should be appropriately documented to detail changes in provenance; however, such examples are few and far between.
15One unique aspect of research data is the context in which it functions as evidence. In this way, research data as an object has two problematic terms with which it is associated: interpretable and reusable (Borgman, 2012). Essentially, data that is “interpretable” requires that the user be expert enough that the data bears meaning, and that the data has documentation sufficient to loan context to how the data was created, how it has been or should be processed, and the provenance of the data. In that data is interpretable as an object in the bibliographic universe, experience and familiarity with similar data objects are key. “Reusable” data is such if the user has the knowledge and expertise to manipulate data objects to other forms necessary to extend the data’s use to a purpose other than the original, in addition to access to the data itself.
16As theories and models that serve as the foundations of research are referred to as frameworks, we adopt the term to apply to the foundation of any data-in-hand (much as cataloging refers to the book-in-hand). As previously noted, data are not always referred to as data, and may be better described in a way that reflects the disciplinary traditions of researchers creating such an information object. Data frameworks, as well as the more easily understood definitions of data, vary across disciplines. “Information” is more often referenced in discussions, while “data” appears to be used to describe an assembly of raw facts (Hjørland, 2019). Disciplinary fields included in this section are not all-inclusive, and a high-level overview that includes statistics, computer science, library and information science, various social sciences, and humanities.
17“Data” comes from the Latin word “given,” similar to the understanding of the English word “gift” (Drucker, 2011; Furner, 2016). Furner’s (2016) historical overview of views on data grouped data into nine types: 1) gifts, 2) metadata, 3) gifts of God, 4) geometric premises, 5) mathematic premises, 6) evidence, 7) attribute-values, 8) bits, 9) differences; and logical permutations of documents, document-sets, data, datasets, and documents of documents. Furner (2017) collected definitions and compared and and contrasted them, noting the similarities and differences between each set of “data” and “information.” He argues that the fields of statistics, computer science, and library and information science have good reason to define an ontology of data, as each field works with data in specific ways. In statistics, Yule (1911) described the “characteristics” of a state, and the explanation of such characteristics being done verbally. Despite Yule’s statement that statistics is verbalized, Furner notes the mechanism of explanation to be numerical. Furner (2017) draws attention to the distinction between “statements,” “measurements,” “observations,” and “facts.”
18In statistics, Kurtz and Edgerton (1939) defined data as “measurements, observations, or estimates; numerical statements of facts in any department of inquiry (such as economics); a collection of statistical facts or figures.” This definition reflects the beginning of the 20th century turn toward a more computational understanding of data. Metrics, statistical observations, calculations, and abstractions with numbers all become a type of data.
19Humanities data is not merely qualitative in nature, it often defies the attempt at clean cut metrics applied to research data from science-based disciplines. As mentioned previously, Drucker (2011) proposed reframing the idea of data being something that is “given,” be it through “facts,” or something more divine in its means of acquisition, and instead as “capta,” or something that is taken. Reframing data as being “taken” provides the flexibility for disciplines that do not traditionally see themselves as having data to allow anything to be data. This is particularly poignant in this 21st century era of proving value and merit through the use of metrics when these developed calculations do not reflect the identity of scholars or disciplines (Day, 2014). Metrics themselves are data in that they are a statistical abstraction, but the application of metrics-based calculations meant to demonstrate the value of a given line of research, usually through adoption and use as evidenced by citation by other sources, is not one size fits all (Day, 2011). The h-index is notoriously biased to favor hard science disciplines, and perhaps, certain social sciences. At best, the practice is unfair to humanities disciplines that may rely solely on a primary source to derive an analysis, or where the translation of an old text to a new language provides value by making the text accessible to researchers who do not speak the original language of the text.
20In summary, this paper has explored the social epistemology of data, especially research data, as an entity that can and should be subject to theories of knowledge organization. Data is not neutral, as it is in of itself a form of work. Data is often treated as agnostic by researchers when the very act of formulating a research question is the first in a set of steps that guide the selection and development of ways data is curated, refined, transformed, and interpreted. The steps used to refine data for a specific purpose to answer a question are often referred to as methodology; methodologies are weighted and purpose-designed to logically guide researchers from a research question to an answer using data. Researchers are adept at articulating theoretical frameworks on which their research relies, but often neglect to describe a data framework to further ground their research. By leaning into an existing theoretical framework, emphasizing the need to address and theory in a given area (e.g., a theory of citation), or developing a new theoretical framework, researchers use theory to help illustrate the common themes and shared understanding in their work.