1. Introduction

The Corpus of Early English Correspondence Extension Sampler (ceeces) is the third release from the Corpora of Early English Correspondence (ceec-400), a family of linguistic resources built for the sociolinguistic study of historical English. The ceec-400 contains over 5 million words from nearly 12,000 letters spanning 1402–1800. To date, some 2.2 million words from 1410–1681 have been released (ceecs in 1998, pceec in 2006 and pceec2 in 2022). The ceeces extends this coverage, adding over 1.1 million words dating from 1653 to 1800.

The ceeces is a selection (a ‘sampler’) from the ceec Extension (ceece), which contains more than 2.2 million words from 1653 to 1800 (see Kaislaniemi, 2018). The ceece was completed in 2012, but its publication was hindered by difficulties in obtaining permissions from copyright holders. To remedy the situation, it was decided to release those parts of the ceece which were 1) out of copyright, and 2) for which we have already received full permission from the copyright holders. These datasets were published as the ceece Sampler parts 1 and 2 (ceeces 1 and ceeces 2), respectively. Further, it was decided to complement these with 3) the same text collections taken from the Tagged ceece (tceece); these were published as the Tagged ceece Sampler (tceeces). The mutual relationships of these various corpora are illustrated in Figure 1.

View Full Size

The released datasets from the ceec-400 family of corpora

Citation: Research Data Journal for the Humanities and Social Sciences 9, 1 (2024) ; 10.1163/24523666-bja10034

Note: Those corpora that are out of copyright have been highlighted in grey.

• Download Figure
• Download figure as PowerPoint slide

2. Context

The ceec-400 was compiled for the purposes of historical sociolinguistics. The original idea was to test the extent to which hypotheses derived from present-day sociolinguistics – such as “women tend to lead language change” – would be supported by empirical evidence in historical material covering hundreds of years. In the absence of spoken data, personal letters are ideal for sociolinguistic research of historical periods: they are speech-like, they have identifiable senders and recipients, and unlike published texts, they could be written by anyone who was literate. The corpus covers a wide social spectrum spanning from housemaids to kings. As such, it is of interest not only to scholars of language history but also to e.g. social historians. Compared to most other corpora of English historical correspondence, the ceec-400 is larger (thanks to its compilation process, see below) and covers a wider section of the populace. The part-of-speech tagging enables studies at a higher level of abstraction, including stylistic trends in the evolution of the letter genre.

Examples of research conducted using the ceec-400 include Nevalainen and Raumolin-Brunberg (2003, 2017), which is the seminal work in historical sociolinguistics. Nevalainen and Raumolin-Brunberg analyse the time course and social embedding of fourteen linguistic changes in English in 1410–1681, from the replacement of subject ye by you to the decline of multiple negation. The findings include that even in the past, women indeed tend to lead most changes in language, and that social aspirers often follow the lead of their social superiors. Using the ceece, this research is extended into the eighteenth century by Nevalainen et al. (2018), who find that the female advantage holds there as well but that the pace of change seems to have been slower than in the preceding centuries, possibly retarded by the ideology of standardisation. Degaetano-Ortlieb et al. (2021) use the tceece to compare the language use of women and men at three linguistic levels: vocabulary, morphology (derivational suffixes) and grammar (part-of-speech trigrams). They find that middle- and upper-class women tend to innovate in the informal setting of family letters and that women lead changes at all three levels, contributing to the colloquialisation of the letter genre over time.

The publication of the ceeces provides the wider research community with the opportunity to conduct its own studies of eighteenth-century English using this rich dataset.

3. Corpus Compilation

The ceec-400 was compiled primarily from previously published, printed editions (see Kaislaniemi, 2023). Only editions that preserved the original manuscript spellings were chosen. Text selection was guided by the aim of socio-regional coverage: to have as good a cross-section of literate English society as possible. Social categories used as selection criteria by the compilers include gender, social rank, and region (see Raumolin-Brunberg & Nevalainen, 2007). The corpus is organised into collections, which usually contain letters from a single source edition. However, the collections do not contain all of the letters in their source editions, as the selection criteria controlled for both quality (excluding for example later copies) and quantity (20 letters per writer was considered representative; more were taken when a writer’s correspondence spanned decades, but often only a few letters were available). In the ceeces, there are on average 13 letters (c. 5,700 words) from each writer.

The ceec team scanned the chosen texts, then digitised them with ocr software, and proofread the results three times against the source edition. The texts were stored as plain text, which required the conversion of formatting into simple text encoding, for example, superscripts like “S^r” Sir are marked as S=r= (these conventions follow Kytö, 1996). More recently, the texts in the ceec-400 have been converted to xml, the previous example becoming <hi rend="sup" range="1,2">Sr</hi> (see Saario, 2020).

Because the ceec-400 was designed for sociolinguistic research, the corpus texts are accompanied by rich metadata, which contains information on the correspondents and on the letters. This metadata was gathered from all available sources and recorded into a spreadsheet by the ceec team. Some metadata is also included in the corpus texts, in the headers of each collection and each letter (for details, see Kaislaniemi, 2018, 2022; Nurmi, 1998; for more, see Nevalainen & Raumolin-Brunberg, 2017, pp. 26–52).

The decision to provide the ceece with part-of-speech tagging was based on a desire to study connections between word classes of the letter texts and the social backgrounds of the letter writers. The claws tagger of Lancaster University was chosen for the task (ucrel, [1997]). Given that claws is designed for present-day English and only accepts text-level coding in xml format, the spelling of the ceece texts had to be standardised and their format converted into xml before tagging. The spelling was standardised in two stages: first semi-automatically by the Variant Detector software (vard 2; Baron, 2011a, 2011b) and then manually by a team of people paying special attention to remaining variation recognised as problematic for the tagging, such as obsolete abbreviations and non-modern punctuation marks (Saario & Säily, 2020). The standardised-spelling ceece was then converted into xml and part-of-speech tagged by claws (Saario et al., 2021).

4. Data Description

1. –Corpus of Early English Correspondence Extension Sampler (ceeces) deposited at Zenodo
   1. –ceeces 1 – doi:www.doi.org/10.5281/zenodo.4644243
      1. –License – cc by-nc
   2. –ceeces 2 – doi:www.doi.org/10.5281/zenodo.5887100
      1. –License – cc by-nc-nd
   3. –tceeces – doi:www.doi.org/10.5281/zenodo.5887230
      1. –License – cc by-nc-nd
2. –Temporal coverage: 1653–1800

The ceeces consists of 42 collections (see Appendix), which contain 2,624 letters written by 200 writers, coming to some 1.14 million words. The corpus texts are provided in plain text and xml formats, in both original and standardised-spelling versions, with the latter also provided with part-of-speech tagging (see Table 1).

View Table

Although the earliest letter in the ceeces is from 1653, there are only two letters from before 1680. Figure 2 shows the number of letters in the ceeces over time, with the proportions of men’s and women’s letters per twenty-year period. (Gender representation in the ceeces is unequal because fewer women were literate in the first place, and thanks to gender bias their letters have been less likely to survive or to be edited and published: see Kaislaniemi, 2018, pp. 51–52). Figure 3 divides the data into social ranks by proportion of the word count per twenty-year period.

View Full Size

Number of letters in the ceeces over time, gender division (%)

Citation: Research Data Journal for the Humanities and Social Sciences 9, 1 (2024) ; 10.1163/24523666-bja10034

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Words in the ceeces over time, social ranks (%)

Citation: Research Data Journal for the Humanities and Social Sciences 9, 1 (2024) ; 10.1163/24523666-bja10034

• Download Figure
• Download figure as PowerPoint slide

The standout feature of the ceeces is its metadata, which is considerably more detailed than that accompanying the previously published sections of the ceec-400 (ceecs and pceec), making the ceeces particularly well suited to sociolinguistic research. The database of metadata for the ceeces contains information on the gender, age and social status of the writers, as well as known details about their regional origins and formal education. The database also contains information on the recipients of the letters, on the relationship between the writer and recipient, and then information on the letters themselves, such as authenticity (is the letter autograph or a copy), the year of writing and word count. (See the ceeces manual [Kaislaniemi, 2022] for more information on the social breakdown of the letter-writers, and for comparisons of the ceeces with the ceece and the ceec-400).

An additional layer of information is included in the ceec-400 corpora in text-level encoding. Part of the editorial apparatus has been made computer-readable, as the corpus retains and encodes information from the edition such as scribal emendations, insertions, hand changes, and damage to the manuscript sources. In addition to information added to the texts by the editor, the corpus texts also contain information added by the compilers. This includes the flagging of foreign words, and in particular, the addition of linguistic part-of-speech annotation.

The part-of-speech tagging is provided using two different tagsets: C5 and C7 (see ucrel, [1997]). The accuracy of the tagging has been evaluated by taking a subsample of the text and manually checking the tags assigned to it by claws. The accuracy in the full tceece using the C7 tagset is estimated to be 94.5% overall. The accuracy is 95.4% for letters from men, 92.8% for letters from women, 93.5% for letters from the 17th century, and 94.7% for letters from the 18th century. The corresponding numbers for the C5 tagset are slightly higher in each case. The tagging of the Pauper collection, which probably had the lowest accuracy (87.9%), has been manually corrected in its entirety. Precisions and recalls by particular tags and the frequencies of most common incorrect–correct tag pairs are provided in the tceece manual (Saario & Säily, 2020; see also Saario et al., 2021, for an account of its creation).

To get a concrete grip on the data, see, for instance, the closing formula of the 100th letter in the Fleming 2 collection as it appears in the ceeces 1 and in the C7 version of the tceeces (letter id FLEMIN2_100):

1. 1)CEECES 1 – plain text, original spelling:

   • So with my duty to your Self, and love and Service to all with you I re[{main{]

2. 2)TCEECES – XML, normalised spelling, part-of-speech tags:

   • So_RR with_IW my_APPGE duty_NN1 to_II yourself_PPX1 ,_, and_CC love_NN1 and_CC Service_NN1 to_II all_DB with_IW you_PPY I_PPIS1 <supplied range="2,6" orig="re[{main{]"> remain_VV0 </supplied>

3. 3)TCEECES – XML, normalised spelling, tokenised, part-of-speech tags:

   • <w id="1397.1" pos="RR">So</w>

   • <w id="1397.2" pos="IW">with</w>

   • <w id="1397.3" pos="APPGE">my</w>

   • <w id="1397.4" pos="NN1">duty</w>

   • <w id="1397.5" pos="II">to</w>

   • <w id="1397.6" pos="PPX1">yourself</w>

   • <w id="1397.7" pos=",">,</w>

   • <w id="1397.8" pos="CC">and</w>

   • <w id="1397.9" pos="NN1">love</w>

   • <w id="1397.10" pos="CC">and</w>

   • <w id="1397.11" pos="NN1">Service</w>

   • <w id="1397.12" pos="II">to</w>

   • <w id="1397.13" pos="DB">all</w>

   • <w id="1397.14" pos="IW">with</w>

   • <w id="1397.15" pos="PPY">you</w>

   • <w id="1397.16" pos="PPIS1">I</w>

   • <supplied range="2,6" orig="re[{main{]">

     • <w id="1397.17" pos="VV0">remain</w>

   • </supplied>

As a comparison of examples (1–3) makes clear, the ceeces 1 retains the original spelling with minimal annotation whereas the tceeces represents the same text with normalised spelling and heavy annotation. The underlying normalisation is the result of a long and complicated process including changes to tokenisation (your Self → yourself). Rather than trying to pack all this information in one all-encompassing format, which would hardly have been readable by any existing corpus tool, the two layers of annotation have been separated into parallel versions of the same letter (for more discussion, see Saario et al., 2021, pp. 125–127).

5. Concluding Remarks

The ceeces is a unique resource for sociohistorical research into the language of English personal letters in the long eighteenth century. Its structure makes it possible to study the language of one individual as easily as that of a certain period. Since it is openly available, it is also eminently suited for many digital humanities applications as well as for teaching.