Tag Archives: metadata

Of concepts and kings: curating a collection using EEBO-TCP

In Spring 2018, MA student Sophie Whittle dedicated 100 hours of hard graft to filling blanks in the Terms metadata. This work follows on from Winnie Smith’s work last year, identifying gaps in the Text Creation Partnership’s metadata. Sophie reflects on her experience and offers some tentative analysis focused on texts from the English Civil War and Commonwealth era:

As an MA linguistics student, I was elated to see an opportunity to apply for Linguistic DNA as part of the School of English’s work placement module. I am looking to apply for a PhD in the future, and I saw the project as a chance to undertake independent research. After completing a corpus-based dissertation about the semantic-syntactic development of the verb promise, I was looking forward to exploring historical texts in a different way—through the use of EEBO-TCP and LDNA computational methods.

After undertaking three years of linguistic study, I realised I had not delved into much historical work as my interests were very much theoretical at the time. Working on the placement has allowed me to regain historical interests which I had left back at GCSE and A-Level. During the summer prior to Masters study, I was in conversation about how King Charles I, who had intended to dine with the Governor of Hull (my home city) was stopped at Beverley Gate by Parliament. As the Governor had expressed allegiance to the Parliamentarians, he was named a traitor, but Charles was forced to return to York. This defining moment, not only for Hull but for Civil War history, piqued my interest. I was able to reflect on the conversation I’d had over summer whilst contributing to the inputting of empty metadata terms for LDNA and viewing documents of news and conflict from the Civil War. I wanted to understand the general public’s attitudes to the different sides of the Civil War in England by using LDNA’s conceptual modelling methods.

Towards the end of my placement, I wrote a proposal for a Civil War and Commonwealth collection to be included on the final LDNA interface. The large amount of texts from this period (a search via ProQuest’s EEBO interface (eebo.chadwyck.com) brings back 35,008 records) are of interest to researchers from linguistics, literature, history, theology, etc. Modelling concepts from and across the period, by determining frequently co-occurring words as pairs or trios, makes it clear that there is a wealth of information to research. My proposal employed the following term definitions:

Title contains: king
Terms contains: civil war, or commonwealth
Date range: 1642 to 1660

By defining these parameters, the idea was that a user could analyse attitudes towards different sides in the Civil War, from both Royalist and Parliamentarian perspectives.

While designing my collection, I also wanted to see if I could mimic LDNA’s use of ‘windows’. I came across a four-text sample about the preserving of peace during the Civil War, with each text from different viewpoints. For instance, one of the texts discusses the New Model Army, a dissenting faction of the Parliamentarian side, as the ‘obstructors’ to peace in the kingdom (A25836). The army was largely independent and held radical Puritan views during the Civil War. Alternatively, a different text suggests peace was only possible if King Charles I prospered, rejecting Parliamentarianism (A25857). I used the node words ‘King’ and ‘Parliament’ as a starting point and counted ten words either side the node words (W20). I then calculated the frequency of the words with the most tokens, and came up with the following table (the frequency is displayed as a percentage):

Table showing cooccurrences with the nouns king and parliament across sampled texts.

Sampled cooccurrences with ‘king’ and ‘parliament’ for a window of 20 words.

The percentages are not particularly high (due to the size of the window, perhaps increasing the size of the window might help solve this issue). However, by identifying the ten most frequently co-occurring lemmas with the node words, a number of interesting results appear. For instance, peace co-occurs more frequently alongside King than Parliament, suggesting something about the bias of the texts. Further close-reading might indicate why peace was continuously associated with Royalism (or not, which emphasises the importance of close-reading!).

Additionally, the reason for vote co-occurring alongside Parliament might seem obvious at first. Yet analysing this within its context provides a different story. Most of the co-occurrences appear in the text slandering the New Model Army. In this text, the author believes that Parliament are voting in response to pressures from the NMA, and are therefore void of their privileges as a democratic union. A list of evidence is provided by the author to explain how Parliament have been revoked of their privileges (during the ‘tumult of the Apprentices’, when apprentices were freed from their masters and asked to join the NMA in ‘a state of confusion’). In the author’s view, the NMA forced Parliament to go against their morals by undoing their previous work.

I also gained access to trio output data from Susan and her work on Newsbooks, to see if I could find something similar. The top concepts for ‘King’ and ‘Parliament’ are ‘king – lord – parliament’ and ‘parliament – state – council’ respectively, which are expected from the genre of texts. By looking at slightly less frequent trios, there are some more intriguing items. The following trios complement the data I had found manually from the four-text sample: ‘king – people – liberty’ with a PMI of 3.09 and ‘parliament – present – authority’ with a PMI of 4.09. (Both PMIs show that the observed trios occur more often than expected by chance.) There is so much data to explore here, highlighting that LDNA should be an excellent resource for conducting quantitative study. As shown, it is important to analyse the findings within their contexts to specify how concepts are cemented in history. LDNA promote the combination of distant (using statistical methods) and close reading. It was interesting to imagine the final interface with the addition of literary analysis.

Working with the TCP metadata has allowed me to explore concepts within the Civil War and Commonwealth period and finalise the work placement by writing a collection brief. As a linguistics student, it has been a real challenge to identify texts based on their literary genre. This pushed me out of my comfort zone. Being able to use my semantic skills to pull apart the meaning behind the conceptual findings has helped too. I am very grateful to have been given the opportunity to use my existing skills and gain new ones!

Featured image: Hull City Skyline. From an original photograph by John Bannon. Used under license CC 3.0.

Scientific Prose in EEBO-TCP

Among the collections provided by Linguistic DNA is one called SuperScience, a broad set of early modern science-writing originally gathered together by Dr Alan Hogarth of the Visualizing English Print project. Here he outlines some other ways to explore those texts.

In September 2017, LDNA researcher Iona Hine presented some work with TCP metadata at DRHA’s dataAche conference. In this guest post, DRHA co-panellist Alan Hogarth (pictured) examines the fruits of his own labour with EEBO-TCP. Alan was responsible for assembling sub-collections of scientific writings as part of the Visualizing English Print group in Strathclyde.

Alan writes:

Using EEBO-TCP’s metadata to search for early modern scientific texts can be a complex activity. The TCP Terms Listings include references from the extremely broad (e.g. ‘Science – Early Works’), to the very specific (e.g. ‘Turpentine – Therapeutic use’). So unless you have a clear idea of what you’re looking for, some texts are likely to escape your notice. This problem stems from the generic diversity and disciplinary fluidity that characterises early modern science. So how do we classify texts that resist classification?

With the Visualizing English Print Project, I attempted to do this very thing. Extending those generic boundaries as far as I could, I created a corpus of just under 2000 scientific texts, drawn from the period 1482-1710.

In this post, I’m going to use the science corpus to examine texts from the first group to identify as a community of like-minded scientists: the Royal Society. What were the most common linguistic features of Royal Society prose?

There are 233 Royal Society texts in the corpus, 11.8% of the whole (see Figure 1).

Figure 1 (left) Percentage of Royal Society Texts vs Rest of Science Corpus, 1482-1710. Figure 2 (right) Percentage of Royal Society Texts vs Rest of Science Corpus, 1660-1700.

Whether seen as a whole or in comparison to other texts from the period 1660 (when the Royal Society formed) to 1700 (Figure 2), Royal Society publications make up a small fraction of the total number of scientific texts. Fellows of the Royal Society wrote about subjects that spanned existing and emerging scientific fields. Yet these relatively small figures are partly explained by the scope of the corpus, which includes multiple disciplines, ranging from medicine to astrology.

The Language of the Royal Society

As well as promoting a new experimental method, key members of the Society (e.g. Robert Boyle, Thomas Sprat) were known for advocating a ‘plain’ style with which to recount experimental activities. On this basis, we might expect Royal Society texts to exhibit a certain coherence, when compared with non-society affiliated scientific texts. Figure 3 (below) shows the entirety of the science corpus (visualised using Principal Components Analysis), with Royal Society texts highlighted in red:

Fig. 3 PCA Scatterplot showing Science corpus, with RS texts highlighted.

Each text has been tagged using Docuscope software. Docuscope counts frequencies-per-text of the rhetorical categories shown on the loading plot (right). These categories contain words, phrases and punctuation. The dots in the lefthand scatterplot represent individual texts in the corpus. The scatterplot highlighting reveals that the highest density of Royal Society texts occurs in the lower right quadrant. From the categories in that quadrant we can tell that this kind of writing is broadly discursive. These categories include:

‘Comparison’, which tags comparative and superlative adjectives, such as ‘more than’ or ‘the largest’, as well as a variety of terms and phrases for difference;
‘Deny Disclaim’ (which tags terms such as ‘no’, ‘not’ and ‘none’); and
‘Private Thinking’, which captures words and expressions that deal with the processes of thought: ‘think’, ‘intellect’, ‘imagine’ etc.

But the graph also shows a distribution of Royal Society texts in the upper and lower left quadrants. The upper left quadrant has a high proportion of texts, such as medical recipe books. The lower left quadrant contains texts on mathematics and physics. Broadly speaking, discursive types of writing score high on Principal Component 1 (the x-axis), whilst procedural writing scores low. Perhaps unsurprisingly for a group with wide-ranging interests, Royal Society texts feature both procedural and discursive styles across Principal Component 1.

In order to get a closer look at the kinds of language used by members of the Royal Society, we can compare works by individual authors. Figure 4 shows the works of three figures: Margaret Cavendish, Duchess of Newcastle; Thomas Burnet; and the Society’s founding figure Robert Boyle.

Fig. 4 Table of texts under analysis by Boyle, Burnet and Cavendish, with publication date.

Margaret Cavendish’s philosophical outlook was at odds with that of the Royal Society. She was a vocal opponent of the experimental method and in her Observations upon Experimental Philosophy (1666), famously attacked new developments in microscopy, pioneered by Robert Hooke and Henry Power. In 1667 she became the first woman to attend a meeting of The Royal Society, during which she was shown a number of experiments. This demonstration did nothing to lessen her scepticism, or alter her commitment to rationalist principles. Burnet’s fame in the late seventeenth century mostly derived from his publication of The Sacred Theory of the Earth (1684), an account of the origins of the Earth after the flood (Genesis 6:9-9:17). His interactions with The Royal Society extended to correspondence with Isaac Newton in the winter of 1680-81. But the influence of The Sacred Theory was such that many Fellows of The Royal Society engaged with the book’s themes, including Edmund Halley and William Whiston.

Boyle, Cavendish and Burnet were selected because of their relative closeness within the lower right quadrant of the first scatterplot above (Fig.3). Each author’s texts can be classified as discursive prose (see Fig. 4). When compared against each other (Figure 5, below), the texts group by author, which means there may be stylistic differences at play, variations in subject matter, or both.

Fig. 5 PCA scatterplot comparing texts by Boyle, Burnet and Cavendish.

Boyle’s Language

The following graph (Figure 6) provides an indication of some of the linguistic variables that separate these texts into author groupings:

Fig. 6 Means of ‘Curiosity’, ‘Self Disclosure’ and ‘Intensity’ across works of Boyle, Burnet and Cavendish.

The graph measures the mean of individual linguistic categories, and shows the relative frequencies with which they are used by each author. So, Boyle’s texts use the language of ‘Curiosity’, ‘Intensity’ and ‘Self Disclosure’, more than those of Burnet and Cavendish.

To see what these linguistic features actually look like, here’s an example from Boyle’s Certain Physiological Essays (1669):

Extract from Boyle’s Certain Physiological Essays (1669), highlighting features of Intensity (blue), Self-disclosure (red) and Curiosity (green).

The linguistic category ‘Curiosity’ has captured Boyle’s methodological approach. Experimentation is at the heart of this method and Boyle refers to specific experiments throughout the essays. But ‘Curiosity’ also tags the language of experimental processes that precede discovery: ‘show how’, ‘when it is filtrated’. Boyle’s use of ‘Self Disclosure’ (‘I have diverse times’) assures the reader of his personal experience of conducting experiments. The frequent intensifiers (‘strong’, ‘very’, ‘so many’) create an impression of the thoroughness of descriptive detail, the kind of detail required if one were intending to replicate the experiment. Boyle professed that replication was a key aim of committing his experiments to print.

Burnet’s Language

This graph highlights some of the linguistic features that are most characteristic of Burnet – ‘Immediacy’, ‘Space Relation’ and ‘Intensity’:

Fig. 7 Means of ‘Immediacy, ‘Space Relation and ‘Intensity’ across works of Boyle, Burnet and Cavendish.

As you can see from Figure 7 (above), Boyle’s texts actually exhibit a little more Intensity than Burnet’s. I’ve included it here because Burnet uses intensifiers as a means of conveying the enthusiasm of his observations in a way that differs from Boyle.

Take this example from The Theory of the Earth (1697):

Extract from Burnet’s The Sacred Theory of the Earth, highlighting Space-Relation (green), Intensity (blue) and Immediacy (red).

Intensifiers, such as ‘greatest’, ‘most’ and ‘boundless’ contribute to this proto-romantic description of the mountains and seas. Given the geographic subject matter, it is unsurprising that Burnet’s use of ‘Space Relation’ eclipses that of Boyle and Cavendish: Docuscope has tagged phrases that indicate spaces (e.g. ‘the Earth’ ‘the tops of mountains’), but also an individual’s perception of these spaces (e.g. ‘behold’). ‘Immediacy’ adds to this immersive stylistic effect, which is achieved through a combination of geographical description and invitations to the reader to imagine him or herself within the landscape that Burnet describes: ‘let us now go’, ‘these things’, ‘these Mountains’.

Cavendish’s Language

With Cavendish, subject matter and style are different again. The following graph provides an impression of what separates her work from that of Boyle and Burnet:

Fig 8. Means of ‘Deny Disclaim, ‘Motions’, ‘Reason Forward’ and ‘Resistance’ across works of Boyle, Burnet and Cavendish.

The linguistic category ‘Motions’ captures words and phrases relating to various kinds of movement. The remaining categories (‘Deny Disclaim’, ‘Resistance’ and ‘Reason Forward’) register a pattern of argumentation in Cavendish’s texts.

We can see how these operate in this example from Observations upon Experimental Philosophy (1666), as Cavendish questions the epistemological value of the telescope and microscope:

Extract from Cavendish’s Observations upon Experimental Philosophy, highlighting Resistance (purple), Deny-Disclaim (blue), Reason-Forward (green) and Motions (red).

The ‘Motions’ that Cavendish describes (‘figurative motions’, ‘mix’, ‘sow’) are cited as examples of the essential uselessness of the experimental method: Microscopes may discern the motions of insects or reflected light in ‘Anatomical dust’, but they cannot provide answers to the larger questions, the causes of things. Cavendish’s rhetorical method supports her philosophical position: to reason, rather than to trust the senses. To this end, her pattern of reasoning is causal and tends to begin with refutations: ‘not found’, ‘no great benefit’, ‘neither’ and ‘nor’. Docuscope tags these as ‘Deny Disclaim’. Cavendish then moves onto expressions of contrast: ‘but are again dissolved’, ‘fallacies, rather than discoveries’ (tagged as ‘Resistance’). She ends with more assertive phrases that close the logical sequence of her argument: ‘so that a Gardener…’. These are tagged as ‘Reason Forward’.

The following example from the same text illustrates the circularity of this linguistic pattern:

Another extract from Observations upon Experimental Philosophy, highlighting Resistance (purple), Deny-Disclaim (blue) and Reason-Forward (green).

To conclude

This brief foray into the linguistic features of scientific prose has only scratched the surface of the differing stylistic choices that determine how a scientific author might write about a particular subject. Factors such as generic diversity mean that quantitative analysis of the whole science corpus can only give us a partial picture of how scientific language may have developed over time. This is before we take into account the incompleteness of the corpus, and of EEBO-TCP. But by looking at the relationship between specific texts and authors from a particular period, we can get a better impression of linguistic traits.

For example, Boyle, Burnet and Cavendish share an interest in the physical world. All of their texts discuss sensory objects and their motions. But their approaches to natural philosophy diverge. What’s interesting is how each author’s language use appears to support their respective methodological approaches to scientific inquiry. Where Boyle and Burnet foreground very different conceptions of experience as the way towards knowledge, Cavendish asserts that truth can only be reached through a process of rational thought, a process that can be detected in her rhetorical practice.

Header image: Dr Alan Hogarth presents at DRHA 2017 dataAche.

A distant look at EEBO-TCP’s “Controversial Literature”

Iona writes: To explore the contents of EEBO-TCP in a distant fashion (and give context to Linguistic DNA data), I have continued to experiment with the Text Creation Partnership’s metadata. Some of this work has been documented in conference papers presented at SHARP and DRHA. Having spent time examining the metadata for items (said to have been) printed in the 1550s and in the years 1610 and 1611 (to be documented in a subsequent blog post), I have now moved on to experiment with how categories observed in those slices manifest across the broader dataset.

One prominent category provided by the TCP Terms listings (a part of the TCP metadata that derives from the same source as much of the ESTC, and is in debt to prior cataloguing efforts) is “Controversial Literature”. About 5.5% of EEBO items in the GitHub TCP catalogue carry this label. Using time as a primary dimension (being the simplest metadata entity to map), it is possible to see how the distribution of “Controversial Literature” in the 1550s compares with other decades.

The following chart plots the raw quantity of EEBO-TCP items labelled “Controversial Literature” in each decade:

“Controversial Literature”: raw number of items in TCP by decade

For the first half-century of EEBO-TCP (1470-1519), no items carry the label. This is not entirely surprising: The quantity of surviving print items transcribed for this early period is much slighter. The next chart plots the raw quantity of all items in EEBO-TCP by decade as an aid to comparison:

Chart showing gradual growth, then increase on a different scale.

All EEBO-TCP: items plotted by decade of printing.

The sparsity of the early period is visible. The drop in the final column (representing the decade beginning 1700) in both charts is a by-product of decade apportionment given the criteria that define EEBO: 817 items carry the date 1700, very few post-date that. Looking at the second chart, the quantity of items printed grows steadily until the 1630s (a slight dip). From the 1640s, the trend breaks down as overall quantities increase significantly: more than 76% of EEBO-TCP items post-date 1640.

It is possible to gain some insight by comparing these two charts. For example, a dip in the 1630s is common to both TCP-whole and Controversial-part. We are not limited to the raw perspective: The Controversial values can be normalised by the total number of EEBO-TCP items from the given decade (dividing raw counts), to show how the share of Controversial literature (v. Items not-labelled-Controversial Literature) varies by decade.

Each bar represents 100% of EEBO-TCP items for the given decade

This mode of representation may make Controversial Literature seem like a marginal category when compared with the mass of other items. Yet in four decades (1540s, 1550s, 1580s and 1600s) this category makes at least 10 per cent or more of the total. If we retain the normalisation but view the information (i.e. percentage of EEBO-TCP items labelled as Controversial Literature by decade) in the plain chart format, the earlier perception of growth is effectively challenged:

Despite overall variation, the trend is downward.

On the wane: “Controversial Literature” as percentage of EEBO-TCP by decade.

From the high point of the 1540s, the market share of “Controversial Literature” appears to decline (if we allow ourselves to conceive of TCP as a market).

Of course, decades are an imposition. This is made evident when we consider the two-year period 1610 to 1611. The following chart counts EEBO-TCP items across 2-year periods, with a trendline tracking the moving average across 5 such periods (i.e. spanning a ten-year period).

“Controversial Literature”: Raw counts of items in two-year periods, with decade moving average.

The overall pattern remains one of exponential growth (a fitted curve reaches R-squared value of .916). For the two-year period beginning 1610, the item count is visibly below the relevant average, and slightly shy of the growth curve.

Here’s a snapshot from the relevant data:

Data table showing 274 items for 1610-11.

Total quantity of EEBO-TCP items for specified 2-year period(s).

The next chart plots “Controversial Literature” data in the same way, albeit omitting the periods from 1522 to 1527 (because their zero values prevent exponential curve plotting). Here, the period 1610-1611 stands proud of the growth curve, matching closely the moving average. Considered in terms of the (raw) quantity of “Controversial Literature”, this period has more in common with the 4 preceding ones (1602-1609) than those that follow.

“Controversial Literature” items by 2-year period, with decade moving average and trendline.

This is the equivalent snapshot from the raw data:

Total quantity of “Controversial Literature” items for specified 2-year period(s).

Finally, beginning in 1528-9 (when “Controversial Literature” becomes continuous) and ending in 1698-9 (to avoid a misleading dip caused by the relative absence of items in 1701), here is the distribution of Controversial literature as a component of TCP across the two-year periods.

“Controversial Literature” as a component of EEBO-TCP, 1528-1699: Each bar represents 100% of the EEBO-TCP items for the specified 2-year period.

In this view, we can see firstly that the peak occurs in 1548-9, but also that (due to the combination of a peak in items so-labelled and a relative dearth of items) in 1610-1611 so-termed “Controversial Literature” has an atypically high stake. Its 15.3% is the second highest point across the 172-year span.

The final table offers a snapshot from the data for comparison:

How raw values convert to percentages: “Controversial Literature” and other EEBO-TCP items for specified 2-year period(s).

And if all this leaves you wondering what constitutes “Controversial Literature” (and why it matters for us), that question will form the core of my next LDNA blogpost.

Documenting categories in EEBO-TCP data

As part of a work placement with Linguistic DNA, University of Sheffield MA student Winnie Smith has been examining the metadata that accompanies the Text Creation Partnership transcriptions of Early English Books Online (EEBO-TCP). Released as a CSV file (here), it combines the various codes that are used to identify different formats of EEBO (and related products). It also includes basic information like author, title, and date of publication. The focus of Winnie’s work described here are the “terms” or “subject headings”, which represent earlier attempts at cataloguing the items in the collection. What can these tell us about what’s in our main dataset?

Winnie writes:

One of the main reasons I’d applied to do a placement with LDNA was the opportunity to work with historical text, so I was very happy be given the chance to analyse EEBO-TCP metadata. Time disappears browsing EEBO (I never knew an agnus dei could be a little wax figurine as well as a mass movement), but I soon realised that analysing it in any depth is a pretty warts and all experience.

The initial target was to select a narrow date range, say 1610-1612, and try to identify broad genre and / or format categories for EEBO documents in that period (e.g. sermons, broadsides, or biographical works). The data sent back from this probe would then help LDNA in its mission to explore the “universe of printed discourse”.

However, getting the data turned out to be more complicated than it first appeared. I decided early that trying to determine semantic categories manually was both unsystematic and impractical. For 1610 alone, for example, there were 49 overtly Christian terms. How should they be grouped together? What about the classifications that might not apply to that year, but arise in other years?

In addition, the fact that the category labels in the metadata originated in the English Short Title Catalogue (ESTC), and had been provided in different ways by different people, meant that there were considerable differences in how texts were catalogued, leading to uncertainty about how they should be classified. There is also the potential for category clashes between modern labels and early modern concepts, for example of non-Christian religions. For example, did thematic entries which mentioned “Jews” refer to Jewish religious texts, learned works, or anti-Semitic material which within an Early Modern frame might claim to be religious and / or scholarly?

Terms from below

I therefore switched to looking for semantic clusters which might emerge directly from the terms. This approach had definite benefits. By manually splitting and sorting the single “Terms” column (using semicolons as the delimiter), I was able to see categories I would have missed if I had continued to rely on manual semantic classification, e.g. “broadsides”:

Screengrab of an alphabetised table, featuring entries beginning with "Broadside".

Figure 1: “Broadsides” emerging as a category in alphabetically sorted TCP metadata.

I also noticed that there were different kinds of categories present in the terms: far from all groupings being topic-based (e.g. repentance), there were several kinds of classification—in the case of broadsides, by printed format.

However, alphabetical manual sorting was time-consuming, and not guaranteed to unite similar content. For example, the entry

Conspiracies — Sermons — Early works to 1800. (A07558)

appeared separately from

Sermons, English — 17th century (A14860).

I decided to try and address that problem by further splitting, though this was complicated by different cataloguers punctuating in different ways. This made it difficult to split off information that wanted separating without separating information that didn’t. For example, all records use a double hyphen (–) to separate off separate units within a single classification. However, in the following record, a double hyphen appears within a single unit (James I) as well as between separate units, e.g. James I | King of England:

James — I, — King of England, 1566-1625. — Triplici nodo, triplex cuneus — Early works to 1800 (A20944)

After some pre-processing in Excel to try and make the punctuation and contents of the terms column more consistent (removing full stops; replacing double hyphens in front of regnal numbers; filtering out blanks or replacing them with dummy text), I loaded it into Rstudio, full of analytical idealism.

This was where the fun began.

The metadata contains widely differing numbers of term entries for different records. I had gravely underestimated the difficulties this would pose in R, which requires rows and columns to be of equal length. The issue was only compounded by the fact that the required number of new term columns (n) was also unknown. I eventually got round this by finding n first (56, as it turns out), then making all rows have n columns. This isn’t a particularly elegant coding solution, particularly because it uses multiple for-loops (which I know R frowns on) and these were slow for a dataset of over 53,000 rows. There are also issues about essentially counting a construct, X, as a proxy for term entries, where

X = something between one semicolon and another.

Observations

Caveats aside, this procedure enabled some interesting observations. The frequency of individual terms, for example, has a very long tail: 16997 items (63% of the total), occur only once. Sorting the terms alphabetically shows that there is still a great deal of noise, at least at the lower end:

Terms including a long string split by a hyphen (row 3), and the full form of 's-Hertogenbosch with its initial apostrophe.

Figure 2: A snapshot of alphabetically sorted individual “terms” and their frequencies.

This snapshot from the output table (Figure 2) shows that some rows, e.g. 3, have been split incorrectly, despite pre-processing to try and deal with different numbers of hyphens. Other potentially tricky cases are correct (e.g. row 5). For some rows (e.g. 14–15) hidden characters clearly affected the result, and it might be worth adding extra processing steps. (I didn’t, for example, remove brackets or capitalisation.)

However, the process worked and was worth it. Even accepting that treating X as a distinct unit is an approximation for catalogue terms, and that the results are imperfect, being able to see the full number and range of cataloguing terms used in the EEBO-TCP metadata (294 378 total term quasi-tokens representing 26 858 unique quasi-types) frees them up for further independent manipulation.

Snapshot includes: Early works to 1800, History, Great Britain, 17th century, England

Figure 3: Top 10 cataloguing terms in EEBO-TCP metadata (by frequency).

It was also instructive to see which expressions (with X ≈ terms) were most frequent in the dataset. Figure 3 (right) shows the top 10, while Figure 4 (below) visualises the relative distribution of the top 25.

Bar chart showing the quantity of records covered by the 25 most frequent terms.

Figure 4: The 25 most frequent cataloguing terms in EEBO-TCP metadata.

As the bar chart shows, by far the most frequent catalogue term is wholly uninformative: all works in EEBO-TCP are pre-1800, so “early literature to …” only confirms the selection criteria which apply to the dataset.

Perhaps more interesting is that the 6th most frequent item is “not supplied”; this shows how many records lack any kind of cataloguing information: “not supplied” was a dummy entry I inserted to remove blanks in pre-processing.

EDIT (7/7/17): We now realise the 2473 works from ECCO-TCP were inadvertently included in the term processing, inflating the reported blank count (“not supplied”). The true count for EEBO-TCP is 3102 items without terms, or ~6% of the total. Only 23 ECCO items have subject headings, with 2 containing the phrase “Early works to 1800”.
Keen-eyed readers may also have spotted “England” twice in the top terms, an artefact of stray spaces (following term-splitting). Removing this irregularity, we find 20,074 unique term-units; extracting commas and parentheses leaves 19864 distinct units—the various biblical figures of Figure 2 (lines 14-16) now represented by one common “biblicalfigure”.

This [nevertheless] shows that messiness extends to the upper as well as lower frequency limits of the dataset, though perhaps in a different sense: at the higher extreme, the issue appears to be less one of dealing with potentially large numbers of small inaccuracies, and more one of high-frequency results which are not all equally useful for semantic analysis.

However, one of the things I have learned on this placement is that no result is unwelcome. Unsurprisingly, EEBO-TCP was never designed for the type of digital text analysis LDNA conducts, meaning LDNA-type processes aren’t always intuitive to apply, but this does not detract from the benefits of working with EEBO, even (especially if) the results remind you to proceed with caution.

On a more personal level, the EEBO-TCP metadata is not the easiest thing to handle as a beginner (I did vainly wish it better, quite a lot), but extremely valuable for precisely that reason. Although I didn’t get much further than writing (cracking) the code to extract individual terms from the master dataset and beginning some summary statistics, I learned a vast amount about being pragmatic and patient with your data (and your programming environment), documenting your working, and the pain of struggling to make sense of messy, non-textbook data is in fact an inevitable part of the process I most enjoy: working with historical text which was never written for a machine to breeze through.

As I finish my placement, I hope my results will allow LDNA to continue analysing EEBO-TCP in more depth, and I’m looking forward to continuing the different kinds of pleasure that come from text analysis (from wonder to perseverance) into a PhD.

Note: For those who might wish to replicate Winnie’s work (noting the disclaimer about inelegant code), here is closer documentation of that process:

I used the following procedure:

a) find n

1) looping over each row of the dataset and calculating the number of term entries for each record
2) adding each result to a vector of term lengths (see code extract 1, below)
3) finding the largest number in this vector (see code extract 2, below)

b) create equal numbers of split terms for each record

4) looping back over the dataset, splitting each row, and making up any difference between the number of terms for that record and the maximum number of terms by adding missing values (NAs) so that each row was of equal length.

c) combine all the newly created term columns into a single column, removing the missing values.

Code extract 1 (above). Code extract 2 (below).