1.

As anyone, who has worked with medieval manuscripts, will know, sometimes more than half of the words are abbreviated. For example, in a forthcoming paper on Middle English and Latin manuscripts of the
Polychronicon, we found that in the most heavily abbreviated Latin sections as many as 59 percent of the words could be abbreviated, while the number for Middle English was 21 per cent (Honkapohja and Liira, in preparation). Studies comparining Latin and Romance have met with similar results (Hasenohr, 1997; Careri et al., 2011). Nevertheless, in digital scholarship, abbreviations are typically seen as something to get rid of rather than useful data to mine.

A major reason for lack of attention given to manuscript abbreviations can be found in editorial practices inherited from printed editions. It is a standard practice for editors to expand abbreviations as “a service to the reader” (cf. Driscoll, 2009). Twentieth-century editorial theory often treats abbreviations as scribal variation as “accidentals” (see e.g. W. W. Greg, 1950), not relevant for the authorial “work” contained in the manuscripts, as much scholarship focuses either directly on the work or uncovering the work under layers of scribal copying and errors. The outcome is an editorial tradition in which silently expanding abbreviations is very much the norm.

Digital approaches for making use of abbreviations as data are available, but are often not used. TEI P5 guidelines introduced the possibility of encoding both the abbreviations and their expansions using the <choice> elements with <abbr> and <expan> (cf. Driscoll, 2006, 2009; Honkapohja, 2013). Still, many digital resources continue the practice of silently expanding abbreviations. Reasons may range from considering encoding abbreviations to be too labour intensive to basing the digital resources on printed editions which expand the abbreviations (cf. Honkapohja et al., 2009). Moreover, text retrieval systems are typically unable to recognize different forms of the same word and the problem is usually solved by normalisation (cf. Kestemont, 2015: 160). Furthermore, some research questions, including investigations into syntax or stemmatology, also require normalisation. However, while normalisation may be necessary for some research questions, it also discards large amount of potentially useful data, which makes other types of research impossible.

The fairly few scholars who do work with abbreviations have identified a number of potentially interesting lines of enquiry. Abbreviations can be used, for example, for identifying change of scribe in the text (cf. Kestemont, 2015) or in historical dialectology for identifying regional characteristics in scribal language (see e.g. Smith, 2016), or studying the effect of right-margin justification on scribal spelling (Shute, 2017), or hiding endings in multilingual business writing (Wright, 2011). Consequently, the practice of expanding abbreviations is discussed and criticised by a number of scholars (Driscoll, 2006; Kyt
ö et al., 2011; Rogos 2011, 2012; Stutzman, 2014, Lass, 2004).

Even though the problems related to the prevailing practice of silently expanding are well known, and some resources such as the
Medieval Nordic Text Archive (MENOTA) do encode them, there have been relatively few studies which would have attempted to use them as data (e.g. Camps, 2016; Honkapohja, 2018; Kestemont, 2015; Rogos, 2012; Smith, 2016; Shute, 2017), especially in comparison to fields such as stemmatology and stylometry.
My proposal for short paper presents project plan and early results for a project, called
Corpus Approaches to Manuscript Abbreviations
(CAMA), funded for September 2017- February 2020.

The current project focuses on applying methodologies developed for corpus linguistics on abbreviations in the spelling system of Early Middle English, 1150-1350. The period is of interest as it was a formative one for the writing systems of English. Linguistic situation in England changed dramatically after the Norman Conquest of 1066, which introduced a new ruling class and relegated English to a tertiary role after Latin and Anglo-Norman French. When Middle English texts become more numerous in the 13
^th century, we find a very diverse dialect landscape in which the lack of a prestigious vernacular has led to the proliferation of local varieties, with almost every text appearing to represent a separate linguistic system.

Within the Early Middle English period, my project focuses on four research questions:

(Q1) Does each scribe have an individual scribal profile of abbreviations?

(Q2) Are some abbreviation usages connected to certain geographic areas?

(Q3) How are Latin and Old English abbreviations distributed in Germanic and Romance vocabulary?

(Q4) What is the function of abbreviations in the spelling system(s) of Middle English?

The data comes from the
Linguistic Atlas of Early Middle English (LAEME), a corpus of ca. 650,000 divided into scribal samples of localised Middle English. Each text in LAEME is based on a diplomatic transcription from manuscript facsimiles, not editions, and using a mark-up system that encodes the expansions of abbreviations, but in a way which makes identifying the abbreviation easy and workable (LAEME: 3.3.1). Consequently, it can be used to compile a dataset, which can be analysed quantitatively.

The methodology is based on corpus linguistics, statistical analysis and historical dialectology. I will use corpus enquiries to compile a dataset of the findings, then subject the dataset to statistical analysis using R and tried and tested techniques such as linear regression, linear correlation, principal component analysis, chi square test and cluster analysis which have yielded results in previous studies of abbreviations and spelling variation (cf. Kestemont, 2015; Smith, 2016).

Compiling the dataset consists of three steps:

• Corpus enquiries, using the web interface and scripts of LAEME.
• Corpus enquiries for unabbreviated forms of the abbreviated words found in stage 1 in each text, in which a particular abbreviation is used. These can be localised, using the lemmas tagged in the LAEME (see 2.3.2: E).
• Compiling a dataset the results, which will include
  a) results of the corpus enquiries, i.e. the abbreviation type, the abbreviated word, non-abbreviated variant(s), frequencies,
  b) information included in the LAEME metadata, i.e. text, lemma, grammatical tag, manuscript, date, script, place, co-ordinates in the LAEME localisation grid, and
  c) additional variables needed for research questions Q1 and Q3, i.e. word origin: Germanic/Romance/Latin (12), content vs. function word (13).

The dataset will be subjected to further analysis, using:

• The inbuilt mapping function in LAEME, which allows dynamically creating feature maps, based on the distribution of any form, its lemma, or grammatical tag.
• Statistical analysis,
  • linear correlation and linear regression, using the form of the abbreviation as the dependant variable, and the results encoded in the dataset (2.3.3: 3) as independent variables, calculating which of them interact with the type of the abbreviation in a certain specimen (cf. Smith, 2016),
  • Principal component analysis common in stylometry (cf. Kestemont, 2015: 168-70).

As I am giving the presentation fairly early in the funding period, I hope to receive valuable feedback on the methodology and also to build a bridge between corpus linguistics and stylometry, creating discussion on the value and potential of scribal ‘accidentals’ as data.