It is striking that in the literature about the Digital Humanities not much attention has been focused on “the digital” (e.g. [Laue 2004], [Deegan 2014, 30]. References to the digital tends to focus on the historical development of computers as artefacts and of their use within the humanities, without systematic reflections about “the digital” and the digitality of the computer, which is what characterizes its way of being digital.

One exception is Evens, who regards the binary code of 0s and 1s as the point of departure for the development of digital artefacts and culture: “The common element in all digital technologies, the unifying aspect of the cultures, arts, and media that we call digital, is the discrete, binary code.”  [Evens 2012, 7]. Thus, the binary code is a condition of possibility for technological artefacts (Barry also stresses the digital by focusing on software and drawing on work in software and critical code studies), but without defining the digital [Berry 2012, 4–6]. Evens understands the binary code of 0s and 1s as numbers that can be calculated [Evens 2012], but another possibility would be to understand 0s and 1s as letters, thus displacing the digital from being a matter of mathematics to being a matter of text. The latter approach can be found in Finnemann’s definition of the digital computer.

In his conceptualization the digital computer is characterized by three invariant traits: 1) a mechanical alphabet composed of a finite number of letters, each of which is void of meaning, 2) an algorithmic syntax, and 3) an interface which determines the semantics of the syntax [Finnemann 1999], see also [Brügger and Finnemann 2013, 68–69]. The idea that the 0s and 1s can be understood as letters and not numbers is based on the assumption that numbers are not void of meaning — 0 means “zero,” 1 means “one” — but in order to be able to combine them via the algorithmic syntax they must be void of meaning, like other kinds of letters that are combined into words. All three invariant traits are necessary as such, but they are not necessary in any specific form. Today the mechanical alphabet composed of a finite number of letters is the binary alphabet with the two letters 0 and 1, the algorithmic syntax comes in a number of forms, from machine to user level, and the interface can be any kind of device that can handle the digital alphabet and its algorithmic syntaxes, ranging from PCs and handheld devices to mainframe and networked computers.

The consequence of this approach to the digital is that on a very fundamental level the digital computer can be understood as a “writing machine,” with the letters being combined to form “texts” on all the levels of the computer, and these texts are editable at any time, down to the individual bit. In line with the point made above by Evens, it is important to include the fundamental digitality of computers in the understanding of the Digital Humanities since it constitutes the condition of possibility of all the concrete forms of the Digital Humanities.

It is worth noting that the humanities always come in the plural. Of course, the plural refers to the many humanistic disciplines, but one could also see this as an indication of the fact that “the humanities” is a very varied field, and that there is no common definition of what it is. Nevertheless, it is possible to identify a number of themes that are often addressed in definitions of the humanities.

First, the question of boundaries has to be addressed: what is in-/outside the humanities? Is anthropology a social science or part of the humanities? Should, for instance, law be considered part of the humanities or part of the social sciences [Howarth 2004], and is communication a humanistic or social science discipline [Gronbeck 2005]? And how should we perceive new disciplines such as literature and medicine [Bolton 2008], or studies in the borderland between the humanities and the cognitive sciences [Beňuš 2010]?

Second, discussions of the humanities tend to revolve around the issue of whether a single, clearly delimited object of study can be identified, as well as a set of predominant theoretical and methodological approaches to be used in all disciplines.

Third, the aim of the humanities is often regarded as a defining feature of this field, for instance whether the humanities should contribute to the advance of mankind and of human culture, or be of a more descriptive nature, or be transformed into individual, societal or industrial value.

And fourth, the humanities can be delimited based on purely administrative divisions of higher education and research, which is a more formal argument with institutional affiliation determining what the humanities are (the humanities are simply what belong to an institutional entity such as a faculty of humanities).

Although there is no precise and concurrent definition of the humanities, these four themes constitute recurring issues in most attempts to define the humanities. Thus, any such definition constitutes a specific constellation of answers to the questions raised within each of the four themes: what are the boundaries, the object(s) of study, the theories and methods, the aims, and the institutional affiliation? And these constellations of answers vary throughout history and may depend on the national setting. Therefore, it is difficult to identify “the humanities” in an unequivocal manner with a view to using this definition as a stepping stone in an understanding of the Digital Humanities. Instead, one must content oneself with the idea that the humanities constitute a fuzzy and complex field, constantly under construction, and that the humanities vary a good deal. And therefore any understanding of the Digital Humanities should start by being very specific about how the humanities that become digital are understood.

As mentioned above, one of the greatest challenges facing any attempt to define the Digital Humanities is that neither “Digital” nor “Humanities” comes with a clear-cut definition, and in particular it is worth noting that the fuzziness and complexity of the “humanities” does not disappear or dissolve just because the term is combined with “digital.” On the contrary, the many differences remain and may even multiply, and any attempt to define the Digital Humanities without having defined the humanities will rebound into discussions of what is understood by the humanities.

Therefore, it may be a better solution to start with “the digital,” for instance, based on the fundamental definition of the digital outlined above. Such an approach does not solve the complexities of the humanities, but it gives a minimal common ground for debating how the interface between the humanities and the digital may be understood.

Thus, the following understanding of the Digital Humanities is suggested as the overall framing of the following discussion of the specific nature of digital sources: Humanities (regardless of what is understood by “humanities”) which to some extent use digital computers and thereby shape the invariant traits of the computer to fit specific scholarly needs, based on how they define themselves as “humanities,” their research questions and their practices. The advantage of this definition is that on the one hand it offers a shared common ground for any detailed definition of the Digital Humanities — that is, the conceptualization of digitality — while on the other hand it is open and flexible enough to incorporate the variety of detailed understandings of the humanities that have been established in theory or by the concrete scholarly practices, and instantiations of the use of the computer. It is important to stress that the interplay between “the digital” and “the humanities” is not regarded as a one-way deterministic logic from “digital” to “humanities.” Instead, it is understood as a dialectical interplay of shaping and re-shaping between “digital” and “humanities” — the two are interdependent.

With this understanding of the Digital Humanities, let us now have a closer look at some of the concrete ways in which the invariant traits of the computer — the digital alphabet, the algoritmic syntax and the interface — are (and have been) shaped by the scholarly practice within the humanities.

On the one hand, all kinds of digital material share the same feature, namely that they are digital. But on the other hand, this digitality is always already embedded in a set of semantic, technical and academic structures, which implies that digital material is not just digital, but that it is digital in a variety of ways (a similar argument of embeddedness in already existing structures is put forward in relation to digital tools in [Collins 2011]). In the following section a general typology of digitality is suggested, based on the provenance of the digital material, and on a distinction between three main types of digital material: digitized, born-digital, and reborn-digital material.

Digitized material is analogue material that has been digitized, for instance, written documents on paper, parchment etc., or electronic media such as radio and television, and even pictures or 3D-models of artefacts. The process of digitization is any form of transformation of analogue material into digital form, be it the laborious keyboarding of written documents to punched cards, the more easily performed scanning of documents to image files, or the digital recording of sound and moving images. The main characteristic of digitized material is that its “becoming digital” is based on an original that was not digital — an original that can in many cases still be retrieved and thus function as a baseline (for reflections on digitized collections, see [Hockey 2000, 11–23], [Terras 2012].

Born-digital material is digital material that has never existed in any other form than digital. This includes all types of material on digital media such as CD-ROMs, DVDs, or the internet and the web. This type of digital material does not have an analogue original to go back to, we only have the digital original ([Berry 2012, 4] as well as [Jones 2014] and [Kirschenbaum 2013] also use the term born-digital).

Reborn-digital material is the term that is suggested to characterize born-digital material that has been collected and preserved, and that has to a large extent been changed in the process of collecting and preserving. Examples of this are emulated computer games or material in a web archive, the latter being presented in greater detail in section 4 below.

Each of the three general types can be subdivided, as the examples above illustrate, and as seen with (for instance) digitized documents such as newspapers that are different from digitized temporal media such as radio or television; just as born-digital material such as computer games on DVDs is very different from online web or apps on iPads (cf. also section 3.2).

The specific nature of each of the three types of digital material has an impact on how each of them can be used and approached by the scholar in the research process. But before going into more detail about how this unfolds in the digital world, let us have a look at a schematic representation of how the research process usually unfolded before the advent of the digital (Figure 1). This very general model does not necessarily fit all parts of the humanities, but it provides a good idea of where the digital can later be located within the research process of most disciplines. In addition, it has to be stressed that what are presented as two distinct worlds — an analogue and a digital — are in practice in today’s digital world very often intermixed in the sense that even scholars claiming to be Digital Humanists switch between analogue and digital in various phases of their research. But for the sake of argument, they are presented as two distinct scholarly environments.

The research process is boiled down to four main phases: the material of study is collected, it is analyzed and the results may be debated and later disseminated. This may be an iterative process, for instance, the analysis may mean that new material has to be procured, the debate may lead to corrections of the analysis, and the dissemination may affect all the previous steps (the iterative nature is indicated by the arrows looping back).

The key argument in this approach is that the nature of the material to be studied to a large degree determines — or at least sets up a range of possibilities for — each of the four steps. Analogue material has to be collected in an “analogue” way, which may imply that the scholar has to visit the collections or objects by moving physically to where they are; the analysis must rely on ancillary tools such as printed indexes, records on index cards, and on how things are ordered in some of the physical buildings and institutions holding the collections (e.g. in “GLAM” institutions, which stands for galleries, libraries, archives and museums). Discussions generally have to take place either face-to-face, for instance at conferences, or in pre-publications such as conference papers, with the process of dissemination taking place in print media such as books, journals or newspapers, or possibly in electronic media (radio, television).

But not only does the analogue nature of the objects of study affect each of the four phases of study. Their nature also has an impact on a more detailed level. Therefore, a systematic distinction is introduced between, on the one hand, semiotic and non-semiotic objects of study, and on the other hand material that is either handed down to, found or created by the scholar. Semiotic documents that are handed down from the past could, for instance, be objects carrying semiotic systems, based on discrete units (e.g. documents, books, newspapers with writing or still images, or film/television with writing, sound and moving images), whereas non-semiotic objects that are handed down are any kind of artefact. As to the objects of study that have been created by the scholars themselves, this type of material can also be either semiotic (for instance a taped interview, field study notes, surveys on paper, or the like), or non-semiotic, such as design mock-ups, or reconstructions of artefacts from the past (e.g. experimental archaeology). It has to be stressed that the two sets of distinctions are analytical distinctions that are made with a view of identifying the most predominant feature of the different types of objects of study. Thus, the clear distinction between semiotic and non-semiotic plays down the fact that semiotic materials are also artefacts (books are also artefacts), and that artefacts also mean something in themselves, although they do not “carry” a semiotic system, such as written letter for instance (an ancient weapon can be a sign of cultural power); and with regard to the latter distinction, material handed down from the past is never approached unbiased but is always selected by the scholar with a view to creating his or her own object of study, whereas the scholar’s own creation of the material of study here and now often relies on material that was handed down.

As mentioned above, the aim is to underline that the characteristics of each of the different types of objects of study has an impact on how they can be approached in the four phases of the research process (as indicated by the thin arrows). The analytical approach to the object of study varies depending on whether we are studying handed-down documents or artefacts, or interviews or archaeological experiments that have been created by the researcher, just as the debate and dissemination of the results may vary for the same reasons.

Now let us have a look at how this unfolds in a digital world, where the scholar is confronted with the three main types of digital material outlined in section 3.1.

The fact that the nature of the material to be studied sets up a range of possibilities for each of the four steps in the research process also applies to digital material. In contrast to analogue material, digital material can be collected at a distance if it is online, digital tools for searching, filtering, and analysing the objects of study can support the analysis, and digital material opens an array of new ways of debating (e.g. blogs, wikis, pre-print databases) and of disseminating and interacting (e.g. computer files, databases, websites, visualisations such as virtual worlds, GIS, interactive maps and time lines).

However, although these opportunities for interacting with digital material are by and large a function of the digitality of the material as such, on a more detailed level they are also a function of each of the three main digital types mentioned above. Digitized, born-digital, and reborn-digital material all have their own array of possible forms of scholarly use, although overlaps may occur: digitized material can be collected, analyzed, debated and disseminated in ways that are specific to this material’s digitality, whereas born-digital material has to be approached in other ways, and reborn-digital material in other ways again. This point will be elaborated in greater detail in section 4, where the nature of digitized and reborn-digital material is compared.

The three main types of digital material affect the research process in various ways, but there are other questions worth raising here as well: is the material semiotic or non-semiotic, and has it been handed down/found or created? With regard to digitized material, one main issue is that analogue artefacts and objects created by researchers cannot become digital as such but have to be transformed into images of some kind. And with regard to born-digital material, the non-semiotic objects are probably not entirely digital, but only come with digital components. As to reborn-digital material, each of the four subtypes of material will probably differ because they are most likely to be transformed in different ways when they are moved from the born-digital area to an archive.

The systematic approach outlined above situates the digital within the scholarly process by introducing the lens of the three main types of digital material, based on their various provenances. But this analytical lens can also be used to give a brief outline of the historical shifts in the interplay between the humanities and the digital. As a working hypothesis, three waves of the nexus between the humanities and the digital can be identified (for other types of periodisation, see for instance [Berry 2012, 3–6], [Thaller 2012, 14–17], and [Finnemann 2014, 95–100].

The first wave starts in the 1960s, when new research fields such as humanities computing, literary computing, computational linguistics, and later digital history emerge [Screibman], [Cohen and Rosenzweig 2005], [Finnemann 2014], and [McCarty 2005]. These new disciplines are initially based on the use of stand-alone mainframe computers; later, proprietary computer networks are introduced, and later again, the open internet. In terms of digital material, these traditions are in the main focused on digitized material.

The second wave starts in the mid-1990s, when two partly overlapping new research fields emerge, namely “New Media Studies” and “Internet Studies.” New Media Studies as well as Internet Studies are often not considered part of the Digital Humanities, but probably they should be, if the focus is on how the humanities — or at least a number of humanistic disciplines — have integrated the digital as an object of study. New Media Studies and Internet Studies are very broad fields of study, including a range of disciplines, but they all converge in the study of new media and of the internet (cf. [Brügger 2005, 106–107] for a brief account on the Digital Humanities and internet studies). In contrast to the first wave, the second wave has had a digital object of study from the very outset, and therefore it is focused on born-digital material.

The third wave is a further development and a subset of the second wave, since it adds a historical dimension to the study of born-digital material: historical studies of the internet within internet studies emerge in the early 2000s. One of the main objects of study is the archived web, but other types of preserved born-digital objects of study are also included; in this third wave the focus is on reborn-digital material (for an overview of web historiographies, see [Brügger 2010, 8–13]).

It is important to stress that the three waves have not replaced one another, but have continued to co-exist, which is why today we have the three main types of digital material identified above: digitized, born-digital, and reborn-digital.

Before going more into detail about the characteristics of the archived web, it is relevant to reflect on why the web is archived at all since it is a born-digital medium that is “out there” all the time. Even though we may have the impression that things can always be found on the online web, the web content changes at an unprecedented pace [Brügger 2005, 15], [Dougherty 2010, 8], [Brügger 2012a, 318]. The acknowledgement of this ever-changing nature of the online web, combined with an awareness of its growing importance for our societies and its importance as a historical source in the future, probably constitute the major impetus for cultural heritage institutions and individual scholars to transform the web into the archived web.

Web archiving refers to any form of deliberate and purposive preserving of online web material [Brügger 2011a, 25]. Two ways of web archiving can be distinguished: macro and micro web archiving [Brügger 2005, 10–11].

Macro web archiving means web archiving carried out by professional archiving institutions, such as national libraries, with the aim of preserving the cultural heritage of, for instance, a nation state, and allowing as many different kinds of research projects as possible in the future.

Micro web archiving means web archiving carried out by people who are not professional web archivers, such as individual scholars or groups of scholars in relation to, for instance, a specific research project with the aim of preserving material of relevance for the research project in question.

What follows primarily applies to macro archiving, but much of it also applies to web archiving in general and to relatively small researcher-generated web collections.

Digitization is based on the existence of an analogue original that in most cases is stable, whether it is a document on parchment, a newspaper, or a radio or television broadcast on tape.

In contrast, the web original to be archived is much more ephemeral in the strict sense of the word. As mentioned above, it is very likely to have changed or disappeared within a very short time interval, and there is thus no original to go back to (cf. [Schneider and Foot 2004, 115], [Masanès 2006, 1]).

Before starting to digitize an analogue collection, the main concern is what to digitize; whereas the how to digitize is in the main limited to questions about which software and hardware to use and how to arrange it.

The web can be archived in a variety of ways that will each result in genuinely different versions. For instance, it has to be decided where the archiving should start and stop, if specific file types should be included/excluded, if the archiving should be allowed to retrieve material on other servers, etc. Thus, the question of what to archive also applies to a web archive, but what is even more important is how to archive the web. The consequences of this are (1) that strictly speaking the archived web does not exist as such before it is archived, but is only created in the process of archiving; and (2) that if two archiving institutions decide to archive the same website at the same point in time, the results are very likely to be different, due to the different archiving settings. Thus, in contrast to digitized collections, where differences as to how to perform the digitization of the same analogue object will probably only lead to minor differences in the two collections, the how to perform the archiving in web archives often creates two unique, but different versions (it is also necessary to add that differences in assembling the archived bits and pieces in the archive may even complicate these differences).

The process of digitization is to a large extent transparent: the archiving institution has an overview of the collection to be digitized and of what is happening during the digitization process.

This is not the case with the process of web archiving. For a variety of reasons web archiving is a much more opaque process, mainly because one never knows exactly what happens “out there,” once the archiving software has been launched. Technical problems may occur, for instance the archiving process may simply stop, it may get off track, for instance, when encountering crawler traps such as calendars, or bot traps such as pages generating new links (both cases create an infinite circuit of requests to the webserver), or it may encounter file or software formats that cannot be archived (streamed video, java scripts, etc.). And a phenomenon that is specific for web archiving may add to the opacity, namely what can be termed “the dynamics of updating” (cf. [Brügger 2005, 22–23]), that is the fact that the website being archived may change during the very process of archiving, and we do not know if, when, and where this may happen — as if the pages of a newspaper were continuously being edited while the newspaper was being digitized. And we cannot rely on going back and checking the original, since it may be either gone or changed.

In general, the analogue media being digitized are usually related to either one point in time (for instance the publication date of printed media) or periods of time with clear-cut starting and stopping times (as with radio and television).

The web has a totally different “publication cycle.” Entire websites are published neither at a specific point in time nor within a clearly delimited time span, but are a continuous publication with no clear-cut starting and stopping time(s). In addition, the publication time is often not mentioned on the web by the “publisher.” The consequence of this is that the temporal subdivisions of the archived web material (in days, for instance) is added by the archiving institution, and is thus not an inherent part of the archived material — and therefore it is random and editable.

The different ways of archiving the web combined with the opacity of the process means that the result of the web archiving process cannot be considered a copy on a 1:1 scale, as can a digitized object that is close to an identical copy of an original. Instead, the archived web must be regarded as a unique version — a version of a (probably) lost original, and very likely one version among other versions, none of which can be identified as the original.

In a digitized collection there is usually only one copy of each document, be it a handwritten manuscript, a newspaper, or a radio broadcast, basically because there is only one original and there is no good reason to create more identical copies.

In contrast, in a web archive there are very often both too few and too many versions of “the same” thing. On the one hand, it is very unlikely that any website has been archived all the time, for instance every day, which means that the scholar will be missing some of the material that was initially online (in addition, elements on a website (images, video, adds, etc.) may be missing too). On the other hand, some websites (or parts of websites) may have been archived several times at very short intervals, so there may be several versions from (almost) the same point in time. Thus, it is very likely that the scholar who wants to study material in a web archive does not only have one copy of each document. Instead, the processes and strategies of archiving the web and the preservation of the web in the web archive entail that in some cases a lot of what was once online is missing; whereas in other cases there may be too much material that is partly identical, but not exactly a duplicate. Finding one’s way through this network of (possibly) overlapping versions is especially difficult because the various extra versions of “the same” may not be versions of the same. There may be partial overlaps between, for instance, a webpage on a website archived in the morning and in the afternoon on the same day — as if a newspaper existed in several versions from the same day, with some pages (or elements on pages, e.g. images) being different in each version, or some being with fewer/more pages.

Digitized collections are in the main based on an original analogue collection of which the archiving institution has a systematic overview, for instance there may be a register or catalogue.

This may also to a certain degree be the case in a web archive, namely if the web archive creates metadata for each archived website (as does, for instance, the Australian web archive Pandora, pandora.nla.gov.au). But this is not always the case, especially in web archives that use the so-called bulk archiving strategy, where vast amounts of websites are archived, either based on a list of all the domain names to archive (e.g. the Danish Netarkivet) or on following links from what has already been archived (e.g. the US-based Internet Archive). In these cases the mere number of domain names makes it impossible to manually register the websites being archived. For instance, the UK country code top level domain .uk consists of approximately 10 million domain names. In these cases the only types of register are either the list of archived domains or the technical crawl log, where information about the archiving can be found, although the first of these two does not tell much, and the latter has to be transformed into information that makes sense to a scholar, and manually added information is still not an option.

For the reasons mentioned in the sections above, a web archive may not exactly know what is in the archive, and this state of affairs is even aggravated by the absence of a systematic register (and even the making of a register of a small number of websites is a challenge because of the nature of the archived web [Brügger 2011b]).

The advent of the hyperlink in the mid-1980s opened up new ways of searching and connecting documents in digitized collections. For instance, hyperlinks can be added to digitized newspapers or radio broadcasts, making it possible to get from one entity to another.

In contrast to digitized collections, where the hyperlink is an optional add-on, the hyperlink is an inherent part of the online web: in the main “no hyperlink” is the same as saying “no web.” And for the very same reason it is not optional to include the hyperlink in a web archive. At first sight one may not consider the necessity of including hyperlinks in a web archive a problem, but on a closer inspection it becomes evident that the inclusion of the online web’s hyperlinks in the web archive may challenge the scholarly use of the archive in ways that are unseen in digitized collections, basically because in digitized collections there is only one copy of each entity, and the publication of the material has already taken place in the past, at a specific and identifiable point in time.

The necessity of including hyperlinks in a web archive is a challenge, partly because archiving takes time (e.g. it takes two months to archive the entire Danish web on .dk), and partly because all web entities are not necessarily archived in the same depth below the front page (due to deliberate choices in the archiving settings, or to unexpected problems). Therefore, the presence of hyperlinks creates two sorts of inconsistencies in the web archive as a whole. First, a temporal inconsistency, because there may not be temporal coherence between the link source and the link target, that is the webpage where the link is found and the webpage to which the link points — they may have been archived with an interval of several days, weeks or even months. Second, when comparing the hyperlinks from a subset of archived websites, for instance in a network analysis, this hyperlink network may be inconsistent as such if the websites in question were not all archived in the same depth. Some may appear to have many links whereas others do not, as if we were comparing digitized newspapers which only had a front page in some cases but which had several sections in others; or comparing radio programmes which only had the first 30 seconds in some cases but only 10 seconds every minute in others. All in all, the compulsion to include hyperlinks in a web archive tends to make the collection temporally and spatially inconsistent when focusing on the software-based relations between websites or other web entities.

The first step when setting out to create the digital object of study in either a digitized collection or a web archive is to search the collection. This is one of the tasks where the advantages of the binary digital text (cf. section 2.1) become evident.

In digitized collections searches can usually be performed in a variety of ways, depending on the preferences and technical possibilities of the collection. But in many cases searches by name and date as well as some kind of free text searches are possible, often combined with various sorts of filtering of the results.

Within a web archive, searching for the name of a website may not be possible, basically because websites usually do not come with names [Brügger 2011b], but also because very often the only way to search the archive is to search for a specific URL, which is the web domain name which can be found in the location bar on the online web. However, some web archives have recently started to support free text searches, which are a great help for the scholar. But this leap forward also comes with a number of challenges which highlight the difference compared with digitized collections.

The well-known and relatively simple act of presenting search results may be slightly more challenging, partly because it is unclear what entity to show, partly because there may be many overlapping versions of “the same” thing in terms of time and space. Imagine a search for the term “humanities” in a collection of digitized newspapers: the result page will show all the pages where the word appears, and the results may be filtered further by the name of the newspaper, the date range or other parameters. But no matter what, there is only one copy of each of the pages on which the word is found; in other words, there is no temporal depth of each page. If we search for the term “humanities” in a ten-year-old web archive, the result page may show the individual webpages on which the term was found, and we may also be able to filter by the URL of the websites to which each page belongs, by date range, etc. But the results page also has to handle the existence of several — often overlapping — versions of “the same” webpage, since in the web archive “one found page” does not necessarily equal “one copy from each point in time,” but rather, as outlined above, something like “one webpage” equals “several versions.” In other words, each webpage may come with its own temporal and spatial depth in the form of partly overlapping versions, overlapping in both time and space (size). In addition, if the web archive does not have any metadata to display, it can be very difficult for the scholar to evaluate different versions.

Thus, the messiness of the web archive is replicated in the search results in such a way that the simple fact of displaying the results of a free-text search may prove to be much more complicated in a web archive than in a digitized collection. And there is no easy technical solution to this, because the core of the problem is the nature of the archived web itself.

The next step after having searched the collection is usually to select and isolate the material that has been found for later study. This is also the case in a web archive, but before doing this one has to perform a step that is usually not necessary in digitized collections: deciding which of the many possible versions of “the same” should be included in the object of study.

As mentioned above, there are a number of reasons why a web archive may contain a large number of versions of “the same” web material; but it is important to stress that these versions are usually not identical duplicates, but genuinely different versions, and the differences may range from being very important to being insignificant. Therefore, the scholar using a web archive has to perform the tasks of identifying, evaluating and selecting the possible versions to include in the study. An example: we may want to study the BBC website (bbc.com), for instance the front page and the two sub-sections News (bbc.com/news) and Sport (bbc.com/sport). The front page may have been archived eight times per day, the front page of News and Sport three times every day, and the sub-pages at varying intervals (once every day, every second day, etc.). We then have many versions of some of the pages, fewer versions of others, and no versions of others; so the task is to decide which of these versions to choose to combine to form our object of study, even though one should be aware that because of the temporal and spatial inconsistencies of the archived web elements, one cannot reconstruct the website’s front page and the two sub-sections exactly as they looked at a specific point in time, but only as they may have looked at several points in time. Identifying, evaluating and selecting versions can be quite difficult, and it is therefore important that tools are developed to help the scholar to perform this task in such a way that his or her choices are as informed as possible. This will include tools that can provide as much information about each of the versions as possible (e.g. starting/stopping time of the archiving), along with visualization tools to make it possible to display the different versions in a manageable manner, and tools to help select — or maybe even combine — which version should be used as the object of study.

Once the scholar has chosen which versions to study, it is time to return to the step mentioned above, which involves isolating the material for later study. In a digitized collection this is in general a trivial task, mainly because each document is clearly identifiable as one (or more) file(s), and in general it is not linked to other documents (and if it is, this is usually done in a very unequivocal manner). For instance, digitized newspapers are usually kept as pdf files, whereas radio programmes are often preserved as avi files. Therefore, the relevant files can either be taken out of the collection and analyzed outside the archive, or bookmarked, put into a folder etc for further study.

Again, things work differently in a web archive, mainly because of the existence of hyperlinks as an inherent part of most web documents, since the hyperlink makes it slightly more complicated to remove the files from the archive. An example: in the study of bbc.com we have identified the versions to be studied, but if we then remove them from the archive, for instance by exporting them to our own computer, we also remove them from the communicative environment in which they were initially embedded in the archive, and suddenly things may appear either not to work anymore or to behave strangely. Links may no longer work because the link target is not part of our corpus, and this is true of the links that are immediately visible on the webpage and the more invisible links that make the webpage look as it does, for instance links to adds, images, or embedded videos located on other servers. Or links may work, but in strange ways if the material is removed from a web archive with no access to the live web and placed on a computer with online access. For instance, if a script on the front page of the archived version of bbc.com requests today’s weather from the server of the national meteorological service, this server-side information may still be displayed if the server is still online, thus displaying today’s weather and patching the website of the past with today’s web (cf. [Brügger 2008, 159]).

Thus, the best way of isolating archived web material for study is to do this within the archive, for instance by generating an index of what has been chosen, and then limiting the study to this index. However, for technical reasons this may not be straightforward. For instance, the so-called deduplication procedure has to be taken into consideration; that is the fact that the web archive continually checks if a file is already in the archive (e.g. a pdf file), and if it is, it is not archived again with a view to saving disk space. But if a scholar later delimits a corpus to a certain date range, and the file was archived before the starting date of this date range, the file will not be displayed, even if it is in the archive. There are technical solutions to this, but the phenomenon illustrates that a web archive is a rather complex collection because of the existence of hyperlinks.

Once the object of study has been delimited and isolated, the scholar can start analyzing it. It is not possible to go into much detail about the great variety of analytical tools that can be applied in analyses of archived web material, but the main challenge for all these tools is that their possible use is a function of how the web archive has been constructed as well as how the concrete construction of the object of study within the web archive has been performed, and with what result. Thus, in a very fundamental way the nature of a reborn-digital collection such as a web archive constitutes the condition of possibility of the analysis.

When faced with the web archive’s challenges with regard to its scholarly use, are there any possible solutions for the scholar to address these challenges?

Since the heterogeneous, messy and opaque nature of the web archive is an inherent part of the archiving and of the archive itself, there are no easy solutions to overcome the challenges. However, this should not leave the humanities scholar paralysed, because one of the most important tools in the traditional humanities toolbox is the skill of dealing with sources and their provenance. These skills just have to be reinterpreted and translated to fit a new digital environment. For instance, the task of comparing versions outlined above is in many ways similar to traditional philology, which means that a future “web philology” does not have to start from scratch but can be developed to fit the specific nature of the archived web by identifying and focusing on the differences between texts written on parchment or paper and texts archived on the web (for an outline of a web philology, see [Brügger 2008, 161–171].

The first steps towards a web philology would be to acknowledge the specific digitality of archived web, and to recognize that access to the digital code may even be a great help to the scholar, since this layer of invisible text beneath the visible text holds a lot of important information and metadata, among others about the provenance. Thus, what is needed is a new web philological toolbox that can help the scholar gain as much information as possible about the object of study. Such an approach will not solve the challenges, but it will help scholars to make as informed choices as possible.