Logo DBNL
TS. Tijdschrift voor tijdschriftstudies. Jaargang 2015 (2015)

TS. Tijdschrift voor tijdschriftstudies. Jaargang 2015

(2015)– [tijdschrift] TS–rechtenstatus Auteursrechtelijk beschermd

Vorige Volgende

The Grapevine: Measuring the Influence of Dutch Newspapers on Delpher
Maarten van den Bos & Hermione Giffard

m.j.a.vandenbos@uu.nl; h.giffard@uu.nl

Abstract

A measure of which newspapers are the most influential is an important foundation for any large-scale study of newspapers. A reliable measure of significance would allow studies to more strongly weigh articles in more influential papers. In this article, we use digital methods to create a partial citation index of a range of Dutch newspapers with a national circulation published in the period from 1920 to 1939. We find that using the digital archive of the National Library of the Netherlands (available through Delpher) enables us to find citations, but the problems of digitization make it impossible to create any index approaching comprehensiveness. Because source criticism - a crucial part of studying history - is made more difficult by digitization, it merits special notice in the digital age.

Keywords

digital humanities, Delpher, newspaper, citation index, source criticism

Introduction

Source criticism is a central part of writing history. Digital archives make such analysis even more crucial because search engines imply that all of the documents that they display are equal. Thus far, in a given search, all documents are given equal value: an incidence of a key word in document A has the same weight as an occurrence in document B, regardless of how significant A or B are, even relatively. Searches return all of the documents fulfilling the search query, irrespective of their importance. This sort of equality, which we will call ‘digital equality’, is an illusion produced by the means of digital access and analysis, and so we aim to provide researchers with a tool to confront it.1

In more traditional historical research, a researcher's judgment of significance has always depended on his domain knowledge. But with the wide selection of newspapers available through search engines such as Delpher, which grants access to the vast collection of the National Library of the Netherlands, it becomes less likely that all users will be aware of the status of all the sources made available to them digitally. In order to enable researchers to better evaluate their results from digital searches, we propose to create a citation index for newspapers. A citation index measures the significance of a newspaper by counting references made to it in other newspapers. Our assumption is that references to a given publication imply its influence on other publications, so citations are a proxy for the significance of a newspaper. Citation indices have existed since the mid-twentieth century in both science and the humanities. Theoretical work on such indices suggests that they can be defensibly used as a rough indicator of quality, although they do have bias problems.2

In the introduction we will reflect on the importance of significance in digital historical research. This will be followed by an overview of the Dutch media landscape of the interwar period. Then we will explore the possibilities and difficulties of using Delpher to create our citation index for Dutch newspapers published nationally between 1920 and 1939. We take advantage of the possibility of full text searching in the large digital newspaper collection of the National Library of the Netherlands. However, while the search engine enables us to find citations, we find that the problems of digitization make it impossible to create a comprehensive index. The consequence of the complications of using digital search engines is that the result of our experiment is a partial citation index, giving relative rather than absolute information. We use other tools to try to overcome the problems that we encountered in our first searches as well as to better understand our results. Chief among these is Texcavator, a tool for large-scale text mining developed at the universities of Utrecht and Amsterdam. The use of this tool improves our results, but not to such an extent that they meet the standards for a comprehensive citation index. This raises questions with regard to the use of digital newspaper databases in historical research, which we address in our conclusion.

On the importance of significance

There are many measures of influence or significance for evaluating newspapers. Different forms of influence are best measured in different ways. We are not interested, as some other projects are, in un-acknowledged borrowing of content.3 Such work is specifically designed to identify ‘borrowing’ when the source is not credited. We have chosen instead to create a citation index, because citing a source is different from copying (parts of) it. Circulation numbers have also been used as a proxy for influence, but they

only give a general picture. A citation index is more useful for a historian who is not so much interested in how many people read a newspaper, as in who read it: in this case, if other newspapers accredit influence to one of their competitors. Citations are more than just an acknowledgement of authorship. They can serve as a guarantee of validity or reliability and also reflect the assumptions of the newspaper in which the citation appears. Citations are useful because they reveal which newspapers are recognized by other newspapers as authoritative. We are particularly interested in capturing this form of influence.

The danger of using digital sources without any measure of significance becomes apparent when we consider the reasons for including any particular newspaper in a digital archive. Inclusion is not entirely or even primarily, as users might think, due to the influence of a newspaper. In the case of Delpher, the influence of newspapers was considered by experts, but was not decisive in the final choice to include a paper in the archive.4 An example of a practical matter influencing inclusion is the quality of the ink of a newspaper. This not only determines whether or not a newspaper survives through time, but also if a mechanized scanner can read it. Of course, archives have always selected sources based on criteria that are not made clear to users (and historians are perhaps too liable to assume that the archives which they use are complete), but digital archives imply the equality of the sources they include in a way that other archives do not.5 The problem of digital equality is exacerbated by the fact that newspapers are digitized according to values that are unknown to the researcher.

The influence of these kinds of practicalities on archives is, or should be, something which researchers are naturally aware of. But there are important differences between digital and paper archives. Researchers often assume that digital texts are like traditional texts. Indeed, Delpher confirms this expectation by first presenting researchers with pdf images of a newspaper article rather than the flawed digitally recognized version of the text. There are many problems with the digitization process, which make this assumption a dangerously incorrect one. In fact, digital texts, while more accessible, are inaccurate copies of the original paper version.

Digitization conceals a substantial amount of information from researchers which might seem peripheral, but is of crucial importance to good source criticism. In a library, where at first glance all books may appear equal, books carry traces of use on closer inspection: check-out stamps, worn pages, coffee stains, or underlining. If you get a book

with uncut pages that are still uncut, then probably the book has not been read since it entered the library. Such traces tell researchers something which might only subliminally influence their work. These traces are, nevertheless, crucial. They add to the researcher's understanding of the significance of a certain work. The digital humanities seem to obviate the need to know what you are looking at, but they actually make it more necessary than ever: it is so easy to mistake obscure, previously inaccessible sources for relevant, influential ones.6 The sorts of information about significance that are gathered naturally in libraries and archives, have to be explicitly added to digital archives, which collect and exhibit no such signs of use.

For any researcher trying to answer a specific question, not all articles are equally valuable. An article which might be relevant as far as the subject matter is concerned, may still not be revealing to the specific aims of a historian. For a computer, all articles may be fundamentally equal, but some are deemed to be more relevant to a given query (generally calculated by counting keywords) than others.7 For a historian, some articles are more significant than others and some articles are more relevant. Unlike the question of relevance, significance is something that we can get a quantitative handle on. It is particularly useful to formalize this information, because such knowledge is needed before one can make truly effective use of a search engine.8

The interwar media landscape in the Netherlands

To understand the meaning of citations of Dutch newspapers in other Dutch newspapers, we need to understand the fiercely polarized media landscape of the interwar period between the 1920s and 1930s. This contrasts strongly with the nineteenth century, when liberal papers and magazines, like De Gids, Algemeen Handelsblad, and the Nieuwe Rotterdamsche Courant, dominated the Dutch newspaper market.

Only in the nineteenth century did newspapers explicitly aim to provide disinterested reporting. In its statement of principles published in 1844, the editors of the Nieuwe Rotterdamsche Courant specified that even-handedness would be their watchword, even-handedness without blandness. Dominant in the journalistic culture of the era was the idea that political power needed to undergo constant public scrutiny. Journalists were seen as carrying out this scrutiny on behalf of the public.9

At the turn of the century, two simultaneous developments significantly changed both the Dutch newspaper landscape and its journalistic culture. Firstly, a number of publications were acquired by large business conglomerates which were interested in profit before content. For example, businessman Hak Holdert bought De Telegraaf and changed the editorial policy of the newspaper from impartial and progressive into

‘neutral’, a concept consciously kept ambiguous.10 At the same time, the democratization of Dutch politics and society produced a whole range of new publications of a specific religious or political affiliation.11 After the First World War, newspapers were no longer seen as public watchdogs, but as defenders of interest groups. Reading a newspaper became part of belonging to a certain community. In 1955, more than 75 per cent of Catholic families subscribed to at least one Catholic daily; for Protestant families this was more than 65 per cent. Figures for the interwar period are alleged to have been even higher.12

This pattern of subscription did not mean that the news read by a Catholic family was different from the news read by a Protestant family. Yet the tone and point of view of different papers could be diametrically opposed. A good example of the differing interpretations of events by different communities within the Netherlands is shown by the coverage of the fall of the first Colijn-administration in November 1925. From the introduction of universal suffrage in the Netherlands in 1918, confessional parties maintained a small but steady majority in the Dutch parliament. Since the Catholic party was unwilling to form an administration with the social-democrats, they were forced to cooperate with two Protestant parties. In November 1925, this fragile coalition broke up as a result of a member of a small orthodox-protestant party asking for the Dutch diplomatic legation at the Vatican to be revoked. When the member's proposition was accepted by a majority of parliament, including one of the two protestant parties which supported the administration, all of the Catholic ministers resigned in protest.13

Every Dutch newspaper covered the political crisis, but their evaluation of the events differed significantly. Catholic newspapers emphasized the importance of the legation to the Vatican in diplomatic affairs and blamed protestant politicians for causing the crisis by refusing to ignore a proposition made by a ‘fringe element’ in their midst because they were afraid of being seen as indulgent towards the Catholic Church. Depending on their party affiliation, Protestants either applauded taking a firm stand against ‘organized Catholicism's’ attempt to seize power in national and international affairs, or denounced the proposition as dangerous for political stability.14 This case shows remarkable similarities to classic case studies concerning the coverage of the Russian Revolution of 1917, events in Germany in the 1930s, or the mutiny on a Dutch naval ship

in 1933, which also emphasize the small but significant differences between interwar newspapers of different affiliations.15

The fact that different newspapers were divided by affiliation complicates any determination of the significance of Dutch newspapers between 1920 and 1939. A Catholic daily would hardly ever quote a Protestant newspaper as an important or reliable source of information; indeed, it might sooner conceal the source of its information. Praise for other Catholic dailies was also rare in Catholic newspapers; frequent positive references to other newspapers could lead readers to change their subscriptions. It is easy, however, to find examples of Catholic dailies dismissing Protestant ones (or the reverse) for misinforming the public. They blamed other newspapers for misinforming their public, but, by doing so, in fact offered a testament to the significance of the other newspaper. A citation index takes advantage of this fact.

Importantly, citations often only acknowledge the significance of particular papers in particular realms. For example, economic news was often drawn from the Algemeen Handelsblad and citing this paper served as a guarantee of accuracy. But when a paper with a clear political position was cited, it was mostly cited as a leading representative of a particular position. In this way, newspapers both acknowledged and reinforced judgments about their competitors among their readers. Looking at how and when newspapers are cited by other newspapers gives us further insight into the assumptions that lie behind citations. The way in which newspapers refer to other newspapers gives us a key to their relationship to public discourse and offers part of the answer to the age-old question of whether newspapers shape or only reflect public debate. We find that newspapers were cited because they were the public mouthpieces of different groups in society. Judged representative, they were often cited to enable another paper to engage with the opinions of a particular group (generally in order to criticize them).

Creating a citation index using Delpher

We set out to create a citation index for Dutch newspapers using Delpher and the digital newspaper collection of the National Library of the Netherlands. Using the metadata to provide gross filters, we limited our searches to articles in Dutch newspapers with a national circulation.16 We first looked for references to the titles of each of the national newspapers published in the period between 1920 and 1939 by entering the name of a newspaper on Delpher. A quick search produced an astronomically high number of hits. Searching for ‘De Standaard’, the most prominent Protestant daily in the Netherlands in

the 1920s, returned 7,020 articles in national newspapers for the 1920s and 4,757 for the 1930s. A search for ‘De Telegraaf’ resulted in 11,181 and 13,613 articles respectively.

Close reading of the highest ranked results made us immediately aware of two problems. The first problem, which inflated the numbers, was that the titles of our selection of newspapers are commonly used Dutch words (see Figure 1). This affects some newspapers more than others. Therefore, it is difficult to take this problem into account in a broad, comparative search.



illustratie
Figure 1: From De Telegraaf 27.01.1979. This example shows that the title of a newspaper, such as De Standaard, can be a common Dutch word, ‘standaard’ [stand]. Simply searching for newspaper titles is therefore problematic. Moreover, in its metadata this text was identified as an article. This example shows how even more precise searching (as discussed below using ‘volgens’) will always produce false positives.


Our reliance on computer searches thus forced us to limit our search to newspapers with titles which did not consist of frequently used words, such as De Maasbode, if we wanted absolute results. Named Entity Recognition (NER) is not yet sophisticated enough to determine if common words are used as proper nouns. Although limiting our search to these newspapers could perhaps be justified, it is problematic in that it moves decision-making about research away from the researcher. We did not want to limit our research in this way, because it would exclude some of the most significant newspapers (De Standaard and De Telegraaf) from our citation index. Furthermore, because we have no way to measure our errors, we do not know if different newspaper titles are biased in particular ways. We were thus unable to measure, except anecdotally, if ‘Standaard’ is more frequently misrecognized by automated software than ‘Maasbode’, for example. A second problem we came across was the fact that newspapers often use their own name in their own articles. Even more problematic was the fact that the title banner of some newspapers has been digitized and categorized as an article, so the top hits tended to be those banners (see Figure 2).

In order to refine our search, we combined the titles of newspapers with typical words that implied citation: ‘volgens’ [according to], ‘zoals’, ‘aldus’ and ‘in’.17 This lead to search queries such as ‘de Telegraaf’ AND (volgens OR zoals OR aldus OR in). To avoid instances when both phrases appeared but were not adjacent, we used ‘PROX’, a Boolean operator which only returns a match if the two words are within ten words of each other. Thus, ‘De Telegraaf’ PROX zoals returns 34 hits for the 1930s (but none for the 1920s, which suggests that the newspapers from that time period are poorly recognized). One of the results is an article in national-socialist daily Het Volksdagblad denouncing De Telegraaf for not reporting on the complaints of small business owners about commercials authored by large multinational companies.18



illustratie
Figure 2: A screenshot of a search for ‘De Telegraaf’ on Delpher. The title banner of the newspaper is recognized as an article.


Our revised search query seemed promising, but the search for ‘De Telegraaf’ PROX in returned many more results than other citation words, because the Dutch word ‘in’ is a common preposition - implying that it is not solely (or even primarily) used to introduce a reference. So while this refined search is better, it still has numerous problems, making it hard to make sense of results, let alone compare them.

Missing a particular synonym for ‘volgens’ (or excluding one, such as ‘in’ for the reasons given above) which a particular newspaper always used, means missing all the citations in a given newspaper, so again our results would be biased and only relatively valid because we treated all the newspaper titles that we searched for in the same way. Unfortunately, although more specific than our first search, this search query also misses cases in which newspapers were the object of news reports themselves.

Creating a citation index using Texcavator

In order to avoid including self-referencing in our index, we wanted to specify that a citation could not occur in the title of a newspaper. In order to limit our search in this way, we turned to Texcavator, a digital tool developed to search and analyze digitized newspapers of the National Library. Texcavator is not only able to search the newspapers with more complex queries than Delpher, but can also be used to analyze and understand search results. Key among its functionalities are timelines and word clouds (as illustrated below), which allow for quick analysis.19 Because the tool uses a different search syntax than Delpher, our query string was slightly different from the one we used for Delpher, using ‘~’ rather than ‘PROX’. Elastic search, the syntax used in Texcavator, allows the user to specify the interval for proximity, so we used five words. Thus, ‘volgens’, for example, has to appear within five words of ‘Telegraaf’. The program also allows the researcher to limit a search using the metadata provided by the database, so we once again configured our search for articles in national newspapers. After we made the changes described above, our search string looked like this: ‘Telegraaf volgens’~5 - paper_dc_title: ‘Telegraaf’. This query excludes all articles that contain De Telegraaf in combination with volgens that were published in De Telegraaf itself, thus avoiding self-referencing.

Searching for citations using the words ‘volgens’, ‘zoals’, and ‘aldus’ showed that De Telegraaf was referenced 1,412 times by other newspapers between 1920 and the end of 1939. In comparison, the Catholic daily De Maasbode was referenced 388 times and the liberal Algemeen Handelsblad 1,077 times. Based on these figures, it can be concluded that De Telegraaf was the most frequently cited newspaper of the interwar period. These figures support the argument of previous press historical research that liberal Dutch newspapers such as the Algemeen Handelsblad lost their prominence during the interwar period, while ‘neutral’ papers, such as De Telegraaf, became more important. In order to further analyze our results, we used Texcavator to make a timeline of the articles referring to De Telegraaf as well as a word cloud, which visualizes the most frequently used words in these articles (see Figure 3).



illustratie
Figure 3: The timeline and word cloud generated by Texcavator using the articles that referenced De Telegraaf.


Word clouds show how digital tools can provide more information about citations than just a number (the relative significance of various newspapers is not taken into account in Figure 3). A word cloud visually represents the content of the articles yielded in response to a search query. The size of the words in the cloud reflects the number of times that a specific word appears in the returned data set. The cloud in Figure 3, generated from articles that referenced De Telegraaf, suggests that the references to that newspaper were mostly in articles about politics, since words like ‘regering’ [government/administration]) and ‘minister’ [minister] were used more frequently than other words. When we similarly generated a word cloud using the articles referencing the Catholic De Maasbode, some of the largest words referred to the Catholic community (see Figure 4). The two clouds are not directly comparable with respect to font size.



illustratie
Figure 4: The word cloud generated by Texcavator from articles citing De Maasbode.


These kinds of results have to be taken with a grain of salt. Firstly, including citations of Het Volk and De Tijd in our citation index remains difficult, even in conjunction with citation words, because ‘volk’ [people] and ‘tijd’ [time] are very common words. A search for Het Volk in the period that we are interested in yielded 4,960 hits, few of which are references to the newspaper. References to De Standaard are subject to a similar problem. This newspaper changed chief editor in 1920. The new chief editor of the newspaper, Henrikus Colijn, was probably the most influential politician of the Dutch interwar period.20 This alone might explain the 1,538 references to the title that we found using Texcavator. However, the timeline of the results for De Standaard and ‘volgens’ reveals a further problem (see Figure 5). It shows a burst of articles in 1933, the year when the Conference of London marked the end of the gold standard. Therefore, it is likely that the burst does not reflect an increase in the significance of De Standaard, but rather reveals the extensive coverage of the conference and the vigorous political debate in the Netherlands on whether or not to follow its conclusions.21 If we allow syntactic confusion between the names of leading newspapers and typical Dutch words of the interwar period to influence our research, then we greatly reduce the number of newspapers for which we can produce a measure of relative significance.



illustratie
Figure 5: Timeline generated by Texcavator showing the chronological spread of articles using the words ‘Standaard’ and ‘volgens’ but excluding uses in the newspaper De Standaard. The red line, known as a burst, represents many articles - the column is not to scale, because if it were, it would render the others invisible. It is likely a reflection of the vigorous Dutch debate about the gold standard, not of references to the newspaper.


In order to check the partial results that we were confident in, we compared our measures of citations to citations in different datasets. For example, we looked at the mentions of newspapers in the Dutch parliament, since its minutes have been digitized.22 Here, too, De Telegraaf was the title most frequently mentioned, with 688 references between 1920 and 1939. De Standaard was mentioned 438 times and De Maasbode 365 times. The Algemeen Handelsblad was mentioned 640 times, however, which seems to undermine our previous conclusion about the position of liberal newspapers in the interwar period. Still, close reading of the articles shows that ‘Telegraaf’ was often used in parliament to describe changes in Telegraph-laws rather than to cite the national newspaper, which makes the usefulness of this search questionable. Other possible comparisons, such as to the Digital Library of Dutch Literature (Digitale Bibliotheek der Nederlandse Letteren, dbnl), are frustrated by search interfaces that only allow searches for single keywords.23

Conclusion

Digital tools give researchers access to more newspapers than ever before, both major and marginal, so producing a relative measure of significance on a similarly large scale is imperative. A citation index will help to make search engines more effective, and go some way towards fixing the problem of digital equality. Digital researchers no longer pre-select the newspapers that they will study based on criteria deduced from knowledge of the period they study. Instead, a search engine selects sources for the researcher, using its notion of relevance, which may often be misguided. Digital accessibility is clearly changing which titles and articles are being used by researchers, and this is an acute problem for scholars who lack experience to judge whether or not a given quote is representative of community-wide sentiment. This kind of knowledge is of crucial

importance when using a tool such as a computer, which ambivalently presents good and bad quotes together.

Every time designers incorporate a specific definition of a value into digital tools, more interpretation and decision-making is taken away from the user. When researchers are not actively making decisions about significance, they are likely to be less critical of the search results presented to them. So while it is important to think about ways to incorporate new information into search tools, it is also essential to remember that if this kind of information is hidden from them, users can make, or are forced to make, fewer judgments about the specific needs of their research problem.

In this paper, we tried to use digital tools to create a citation index in order to establish an empirically based standard for significance. Because of the many problems that we faced using digital search tools, we were only able to find the relative significance of a few major newspapers. Our inability to answer our initial question brought to our attention many shortcomings of digital archives that remain to be overcome. As long as source criticism remains opaque for digital sources, and admittedly our suggested measure of significance only partly addresses this problem, digital archives will continue to contain too many problems to be used to provide quantitative evidence for historical arguments.

In general, both the fervent exponents and the avid opponents are likely to overestimate the capability of digital tools. Therefore, we need more theoretical reflection on the possibilities and problems of digital archives. Significance is just one measure necessary in order to effectively use the abundance offered by these archives. Our experiment has shown that it is very difficult, if not impossible, to be comprehensive using digital tools. We need to put much more effort into solving the problems, including those of digital equality, which frustrate interpretation.24 An important step towards facilitating the use of digital sources in the same way as any other sources is to really understand both the many sources made available in digital archives and the way in which they are structured and presented. Finding a reliable way to judge significance is part of this indispensible exercise.

•> maarten van den bos & hermione giffard are both postdoc researchers in the Departement of History and Art History at Utrecht University for the Asymmetrical Encounters: E-Humanity Approaches to Reference Cultures in Europe, 1815-1992. A project on digital text mining funded by HERA.

Vorige Volgende

Over dit hoofdstuk/artikel

auteurs