8.1 Introduction

The previous chapter considered change over short periods of time involving health-related corpora that reflected present-day language use. There is a long-standing tradition within corpus linguistics of analysing historical language data (e.g., Taavitsainen et al., Reference Taavitsainen, Hiltunen, Lehto, Marttila, Pahta, Ratia, Suhr and Tyrkkö2014). In this chapter we also consider time, yet here we consider language use over longer stretches of time using historical corpora. In dealing with data which is remote in time, more groundwork often needs to be carried out in terms of locating and cleaning texts and forming relevant research questions (as already mentioned in Chapter 3). A detailed consideration of context is also important, in order to provide interpretations and explanations of findings which are informed, for example, by historiography. Accordingly, when looking at such data, we have found that it can be very helpful to include a historian as part of the research team.

In the following sections we will look at two case studies. The first explores an issue – vaccination – that is widely discussed in public discourse, both in the past and in the present day. The second case study – on the topic of sexually transmitted diseases (STDs) – is more exploratory and faced challenges in terms of identifying relevant examples in the corpus. This led to an analysis of collocates, which in turn resulted in the identification of new research questions.

8.2 Anti-vaccination Discourse in Nineteenth-Century England

In Chapter 3 we discussed the rationale and process for the construction of the Victorian Vaccination Discourse corpus (VicVaDis) – a 3.5-million-word collection of anti-vaccination material published in England between 1854 and 1906. During that period (which roughly overlaps with the reign of Queen Victoria), a series of Vaccination Acts made the vaccine against smallpox compulsory for babies at three months of age and imposed penalties on parents who did not comply, including fines and, potentially, imprisonment. Response to the Vaccination Acts included a large-scale organised anti-vaccination movement that persisted until a Vaccination Act in 1907 made conscientious objection easier, to the extent that in practice compulsion no longer applied. Contentious as they may have been, the Vaccination Acts marked the beginning of a process which led, thanks to vaccination, to the eventual eradication of smallpox in 1980.

As noted in Chapter 3, in 1853 immunisation against smallpox in England involved the insertion of material from the pustules of people infected with cowpox – a mild disease which had been found to confer protection against the much more dangerous smallpox in humans. Previous work based on document analysis had identified three main concerns about compulsory smallpox vaccination in nineteenth-century England (Durbach, Reference Durbach2005; Fajri Nuwarda et al., Reference Fajri Nuwarda, Ramzan, Weekes and Kayser2022). The first concern involved potential dangers to health. This was partly because of some inherent risks involved in vaccination practices at the time, but there were also broader beliefs that vaccination interferes with what might be seen as the purity of the human body. The second concern related to the later practice of propagating the cowpox virus on the skin of calves and sheep – the resultant vaccination was viewed as the insertion of animal products into people, which raised religious objections. The third concern involved civil liberties and the mandatory nature of vaccination; critics objected to the penalties that applied to non-compliant parents, particularly from the poorer sections of society.

The availability of the VicVaDis corpus makes it possible to systematically investigate the manifestations of these different concerns in a collection of pamphlets and other texts that were widely available at the time, as well as to identify further patterns of objections. In turn, these can be compared with what is known about vaccine-hesitant views from other historical periods – particularly the present day.

In the rest of this section, we will report the main findings of an initial analysis of the VicVaDis corpus carried out by Hardaker and co-authors (Reference 132Hardaker, Deignan, Semino, Coltman-Patel, Dance, Demjén, Sanderson and Gatherer2024), and complement these findings with further observations we have made by exploring the corpus, which can be freely accessed from the website of the Questioning Vaccination Discourse project (www.lancaster.ac.uk/vaccination-discourse/vicvadis/).

8.3 Representations around Sexually Transmitted Diseases in Early English Books Online

In this section we outline a study which provides insight into how data can help us research while challenging analysts to both revise their research questions and improve the data they use. The study also shows how a shift in the contemporary context in which work is conducted may create a relevance for a research project that was not the intended goal of the project but was, nonetheless, its outcome.

The study was part of a long-standing research programme using corpus data from the early modern period to explore marginalised identities in seventeenth-century English writing. The data in question was Early English Books Online (EEBO), a digitised collection of more than 146,000 printed works in English produced largely before 1700. This database had originally been made available as scanned images which granted researchers access to these texts without the need for expensive field trips to libraries. While these scans were of limited use to corpus linguists – they could not be searched using corpus software – they did at least allow key works to be consulted at scale, with work such as McEnery (Reference McEnery2006) using EEBO, to explore the use of ‘bad language’ in Early Modern English. The scans were often derived from poor-quality texts. Even so, optical character recognition (OCR) scanning of these texts made it possible for users to search for a word in order to determine which pages from which texts might be of interest. While the results were prone to a high degree of error, the search function was still helpful. The process of improving the utility of the data began in 1999 when a group called the Text Creation Partnership (TCP) was formed, consisting of University of Michigan Library, Bodleian Libraries at the University of Oxford, ProQuest, and the Council on Library and Information Resources. The TCP worked to produce high-fidelity, machine-readable, textual transcriptions of the documents in EEBO. The resulting data was suitable for corpus analysis. In the work reported here, EEBO version 3 was used, providing just under a billion words of data for the seventeenth century. A billion words would, in the recent past, have presented its own challenges to standard corpus search packages, exceeding by far the limits of those software packages to search the data. However, the work on EEBO reported here used CQPweb (Hardie, Reference Hardie2012), which allows the rapid search and analysis of multi-billion-word corpora.

While the EEBO data still presented challenges, these were challenges linguists always had to face when working with data from this period – for example, limited metadata and variant spellings which made it difficult to search for all the occurrences of a specific word. Nonetheless, EEBO TCP opened up new avenues of possibility for corpus linguists, and in the programme of work described here, a corpus linguist and a historian worked together to carry on in the spirit of the work of McEnery (Reference McEnery2006) to look at the representation of marginalised groups such as the poor (McEnery and Baker, Reference McEnery2019) and prostitutes (McEnery and Baker, Reference McEnery and Baker2017).

In pursuing their research on marginalised groups in the early modern period, the researchers used methods from corpus linguistics; where appropriate, these were supplemented with other methods, notably geographical information systems, to explore the locations where marginalised groups resided (Baker et al., Reference Baker, Gregory, Hartmann, McEnery, Wiegand and Mahlberg2019). The analysis blended the close reading and critical archive research of a historian with insights provided by linguists and corpus data.

The study summarised here began with a research question in a similar vein to those discussed previously – the researchers wanted to look at the representation of people with STDs in the seventeenth century (McEnery and Baker, Reference McEnery, Baker, Hiltunen and Taavitsainen2022). While EEBO addressed the issue of access to a sufficient volume of data, availability was only the first challenge. While earlier work had proved possible with the data, when the researchers sought to explore the representation of people with STDs in EEBO, they encountered a substantial problem – one of the primary terms for referring to STDs, pox, was also a very frequent swear word of the time. It was also a very frequent word – there were 9,960 examples of pox in the section of the corpus covering the seventeenth century. With so many occurrences, reading all of the concordance lines to distinguish mentions of STDs from the use of the term as a curse or an insult would have been prohibitively time-consuming. An alternative approach would be to work with a sample of the data, which would have allowed the analysts to broadly characterise the usage of the word. As a way of trying to down-sample the data in a principled way, the analysts turned to collocation (see Chapter 1).Footnote ¹ This approach proved partly helpful. Some collocates clearly related to cursing, such as rogue, which collocated with pox 30 times in the corpus and never referred to STDs, being used instead as a curse (e.g., in sentences such as ‘A pox on you for a rogue’).Footnote ²

However, the collocates also revealed a further problem, along with presenting two opportunities. The problem related to those examples of pox which clearly related to disease. As the word refers to a symptom of disease, a pustule and its consequent scarring of the skin, it is quite vague as a term. On examining the examples of pox clearly related to disease, the researchers could see that only in a subset of cases could they be sure of the nature of the disease causing the pock marks. Collocates such as venerous, occurring 12 times, always in the fixed expression venerous pox, clearly relate to what would now be called syphilis. Yet when the word collocates with other STDs, such as gonorrhoea, it is not necessarily the case that pox is indicating syphilis. In some cases it probably does as it is linked to other venereal diseases (e.g., ‘cures the greatest Pox, Gonorrhea’s, Cankers, and all Venereal Diseases’).Footnote ³ However, the word may also appear in a list of diseases and be ambiguous, as in ‘the Gonorrhea, Pox, Gout, Leprosy, and other such like Diseases perfectly cured’.Footnote ⁴ That ambiguity is made apparent when we consider other collocates which indicate various diseases collocating with pox that are also linked to pock marks, notably small producing small pox (mentioned 2,001 times in the seventeenth-century data).

Yet one type of collocate always denoted an STD and was of interest from the point of view of discourse analysis – nationality terms (e.g., French pox). Through such examples, the agency of the infection was, in part, ascribed to foreigners. This offered an opportunity, resulting in the analysts reconsidering their research question. Rather than focusing on the construction of those afflicted with the disease, they decided to focus on the agency for the introduction of the disease. This was both tractable as a research question and somewhat contested in the literature. The insight gained from that collocation allowed the exploration of a new, and fruitful, angle on STDs in discourse in seventeenth-century England.

The second opportunity that the collocation analysis provided related to the close reading that was carried out to support the interpretation of collocates. Through looking at the examples and checking the metadata to ascertain the date of the example and the name of the text in which it was written, an impression was formed that certain collocates of pox increased over time, but they also increasingly appeared in certain types of texts as the century progressed. As the texts were marked with a date of publication, it was easy to explore how a term such as French pox was dispersed over time. There were 1,181 such examples in the seventeenth century, and when the researchers looked at the relative frequency of the term in the decades across that century, it was found that the peak of mention for the term occurred in the 1650s (1.78 examples per million words), with a later peak in the 1690s (1 example per million words). While there was no reliable metadata that enabled an examination of distribution of the terms by genre of text, scattered, compelling evidence was found that suggested a link between time and genre relating to pox. For example, when exploring the texts in which pox and rogue collocated, all but two were clearly plays.

To explore the link between the construction of the disease in terms of nationality on the one hand, and genre of text on the other, it was decided to produce a genre classification of all the texts in the EEBO v.3 corpus. The process of doing that is described in McEnery and Baker (Reference McEnery, Baker, Hiltunen and Taavitsainen2022), but in brief the authors used the titles of the works in question, supported by a limited amount of close reading, to first develop a genre classification of the texts in the corpus and then assign texts to that classification. The classification itself was composed of five major genres (Literary, Religious, Administrative, Instructional, and Informational), which were further divided and subdivided to produce a total of 25 sub-genres and 80 sub-sub-genres. The classification was further developed and used in a subsequent project focused on the creation of a dictionary of Shakespeare’s language (see Culpeper et al., Reference Culpeper, Hardie and Demmen2023; Murphy, Reference Murphy2019). While painstaking, the genre classification task enabled the exploration of the link between nationality, STDs, and genre.

The analysis found that while there was a wide range of nationalities associated with the pox, French people were most frequently linked to it. This gave plausibility to claims in the literature that proximity was a driver in the naming of such diseases – English writers blamed a local rival power. Three main terms were used for this, in descending order of frequency: French pox, French disease, and morbus gallicus. By normalising the frequency of mentions to per million words, the terms were seen to peak in use in the 1650s and 1660s, fall, and then rise in frequency again in the 1690s. However, there was also an elevated mention of Naples with reference to the pox, as English writers did, on occasion, mention the possibility that it was in fact Naples that was the source of the disease and more specifically soldiers of the army of Charles VIII who had marched on and captured Naples. However, as Charles was a French king at the head of a French army, this also, indirectly, pinned the blame on France. So while the Neapolitan pox may have been linked to Naples, more broadly it was still linked to France in wider discourse.

The genre analysis of the corpus corroborated the hypothesis that the discussion of the geographical origins of STDs varied by genre and through time. The main finding of the genre analysis was that, as the century proceeded, the discussion of French pox rose substantially in one genre: medical writing. By 1690, the number of mentions of French pox (and associated variations) had grown. From the 1630s onwards, it occurred most often in medical texts, and by the 1690s, the term was barely used in any other genre. Prior to the 1630s, there was a peak of mentions of syphilis in the genre labelled as treatise (a form of formal writing systematically investigating a particular subject), in which French pox was linked to a range of topics, including anti-Catholicism and the discussion of far-off places. Peaks in this period of texts from the history genre tend to discuss the possible geographical origins of the disease.

After a series of initial difficulties working with the data, the analysts were able to both usefully refocus their initial research question and enrich the data to discover new insights into the framing of discussions of disease in terms of blame and agency. In this case, both of those converge on one party – the French. However, as the century proceeds, we see the discourse around the disease shifting markedly towards medicalisation. That medicalisation is apparent in the pattern of collocation surrounding the phrase French pox – if the top-10 collocates of this word are considered in the medical writing in the corpus, it is seen that they relate to other diseases (scurvy, dropsy, leprosy, gout, diseases), medical treatments (dose, cures, cured, cure), and the process of infection (infected). As the discussion of French pox rises in this genre, by association this medicalised framing of it also rises. Importantly, if the collocates across the whole corpus are examined, it is seen that while collocates relating to other diseases are typical of general usage, the mention of supposed cures (cinnabar, guajacum, quick silver), moral judgements (evil), and naming strategies (called, Naples) are present in general English but not medical discourse. The rise of a medicalised discourse around French pox, therefore, is also marked by a move away from a culture of blaming, moralising, and bogus cures.

At this point it appeared that the work on this data was complete. However, in the process of submitting the work for publication, an event occurred that impacted on the publication itself. The work was submitted for publication in late 2019, just before the COVID-19 pandemic struck. Early in the COVID-19 pandemic, a process began of linking the disease, and its variants, to specific geographies, just as had happened with syphilis centuries before. The initial set of reviews for the paper asked that this be acknowledged, which was done. After resubmission, the public debate around COVID-19 had moved on, and the practice of linking COVID variants to the places in which they were detected had been much reduced, to avoid stigmatising national groups. When the paper returned from its second review, the reviewer noted this and asked that the researchers remove the reference to COVID-19 from the paper. The request was discussed with the editor, and ultimately the researchers decided to leave the reference in, as the paper was precisely about how geography and disease can combine to stigmatise. COVID-19, appearing in the publication process when it did, showed just how little had changed in terms of disease-naming practices, while also being illustrative of how naming practices can be just as sensitive and harmful today as they were four centuries ago.

8.4 Conclusion

Historical data can be much more difficult to collect, particularly if one wants to ensure a fully representative corpus. Poor OCR transfers or natural spelling variation can make it especially difficult to carry out lexicographic analysis based on word frequencies in such a context. Tools like VARD (Baron and Rayson, Reference Baron and Rayson2008) can help introduce a degree of systematic standardisation into historical corpus data. There is also the possibility that artificial intelligence software will be able to further improve on poorly scanned historical data, though that promise lies in the future at the time of this writing.

As the two case studies have shown, one aspect of historical corpus research on health-related topics is that the findings can sometimes help shed new light on more contemporary topics, showing how discourses can sometimes remain remarkably entrenched or are able to resurface, given similar conditions. However, there is a danger when analysing historical corpora in assuming that the attitudes and language use of the past are the same as the present. Unlike the corpora examined in the previous chapter, we do not have direct experience in the time period that these corpus texts were written in, and thus we should not impose our own values on the authors of historical texts. Context is key – it is notable how the first case study in this chapter cites large chunks of texts in order to better make sense of the use of a keyword, whereas in the second case study it was necessary to classify the genres of texts before they could be analysed. Additionally, we often need to go beyond the texts themselves, to account more holistically for the time period that they were produced in, which is one of the reasons why it can be useful to work with a historian in this challenging but rewarding form of health linguistics.