Methodology

In order to analyse the textual transmission of the selected works, first, a thorough research of the relevant manuscripts will be conducted, focusing on both their origin (time, place, scribe, other texts copied together) and their subsequent history (codex binding, owners, glossators etc). In terms of medieval readership and reception of the texts, other works written in each manuscript will be taken into consideration and any parallels between them and our texts will be searched for.

As for the philological aspect of the research, it is necessary to create digital editions of the texts so they can be studied by means of our experimental app. Furthermore, we will study changes of each text over time, and compare differences between the exemplars, trying to interpret such developments of the texts. The result of this will be the establishment of relative chronological order of the exemplars as well as critical evaluation of their content and its evolution.

The challenge of this otherwise standard manuscript transmission research is the fact that each case study does not explore a single text but a group of closely interrelated texts on the same theme. The borders between “texts” and “versions” are often difficult to discern.

Our application will use word embeddings to compare multiple textual witnesses of the same work, and exemplars of similar works, and analyse differences between them. The front-end of the application (which shows the results of the analysis) will be written in JavaScript and follow the principles of Progress Web Application (PWA). This kind of application combines the important properties of a web application and native desktop applications and thus can run on different devices, perform sophisticated tasks, work without the internet, and be installed.

However, for the textual analysis itself, a more powerful tool is required and that will be a backend written in Python. Python unlocks us access to a broad scale of robust frameworks and libraries, as well as a large community of developers and researchers, in order to help us solve difficult challenges. The application will use a model with pre-trained word vectors for the Latin Language to analyse Latin text. This model was trained using CBOW with position-weights, in dimension 300, with character n-grams of length 5, a window of size 5, and 10 negatives. This model was trained on various recent Latin documents and thus might prove incapable to analyse complicated medieval Latin texts. In that case, we expect to be able to train a simpler model that will help us achieve our goal and could provide a starting point for follow-up research.

The results provided by the application will be evaluated and interpreted by the investigators in creating a complex analysis of the manuscript transmission of the selected texts.