Interview with Carole Lamoureux, Cassandre Maubert and Jeanne Fras

 

The Archival City programme team has undertaken the scanning and the encoding of a collection of administrative directories from the Préfecture de la Seine and the City of Paris spanning the period from 1883 to 1970. These directories list the “personnel and responsibilities of the departments of the Seine and the City of Paris” (1885-1911), later renamed “organisation and responsibilities of the departments” (1920-1966) and “administrative directory” (from 1970). This work makes it possible to navigate quickly and reliably through 38 directories. We spoke with Carole Lamoureux, Cassandre Maubert and Jeanne Fras, integral contributors to the transcription and encoding of these directories, about their methodologies and the outcomes they achieved.

How did this project come about?

CL: This project has been in motion for two and a half years. Loïc Vadelorge and the Grand Paris research team conceived the idea to process the directories of the prefecture of the Seine and the city of Paris in order to gain a better understanding of the administration of the prefecture and the city and to find one’s way around the archives. Researchers in fields such as urban planning, development, and transportation could identify documents, but lacked access to the comprehensive historical context provided by archive producers.

How were the chronological limits of the directories defined?

CL: Initially, the team envisaged a broader timeframe encompassing the 19th and 20th centuries. Ultimately, the program researchers determined the dates based on the directories themselves. They observed a remarkable consistency in the directories from 1883 onward.The cut-off date was a few years after the cessation of the Prefecture of the Seine (at the same time as the department of the Seine[1]) on 1 January 1968. This is why the last directory dates from 1970.

Where are these directories kept?

CL: The directories we work with are housed at the Archives de Paris, in the library section of the institution. The Archives’ staff conducted the digitization process. The Archives de Paris, which inherited a significant portion of the Seine department’s archives, has yet to generate authority records for archive producers (history of archive creators). Hence, our efforts at Archival City hold value not only for researchers but also for the Archives de Paris to benefit from our work.

How many directories does the corpus include?

CL: Our corpus comprises 38 directories, although only 37 will be accessible through the visualization tool. The 38th directory from 1968 will have limited access due to its unique nature. This directory profiles departmental directors shortly after the dissolution of the Seine prefecture. The directory provides detailed information about directors, complete with photos, birth years, and career paths. Due to the volume of data and its proximity to the present, we won’t publish all this data online. Eleven directories were excluded as they were incomplete, serving as supplements or corrections to previous editions, lacking comprehensive organizational details.  In terms of dates, it’s very disparate: we don’t know why there are directories for this or that year, perhaps there were directories that have not been found. During the First World War, it seems that no directory has been published, but this has yet to be confirmed. The publication resumed in 1920 and continued regularly until the Second World War.

Who contributed to the Directories project in terms of personnel?

CL: Elsa Camus was initially responsible for processing digitizations carried out by the Archives de Paris. During her tenure she proposed undertaking the use of XML-TEI encoding, a proposal accepted by the team. Initially, she devoted 80% of her working time to these directories and the remaining 20% to the coordination of Archival City. This ratio was reversed when I was hired in her place (80% on Archival City as a whole and 20% on the directories in particular). Consequently, several interns undertook substantial coding work, including Cassandre Maubert, Anahi Haedo and Nathan Dacier-Falque, all master’s degree students (M1). Their task consisted in the archival descriptions, but also in completing the digitisation of directories in eScriptorium[2] and their encoding. Between winter 2021 and spring 2022, there were again three interns (Léna Humbert, a Master 1 student, Sonya Bensaadi and Andréa Dorin, Master 2 students).

During the summer of 2022, a trainee joined the team for 3 months (Fanny Brière, who had a background in art history but wished to train at XML-TEI). Then, in September 2022, Cassandre was hired as a full-time ingénieure d’études. A large part of her working time had been devoted to the directories. Finally, Valentin Devillepoix and Jeanne Fras joined the team as interns from February 2023. As for the scientific team, several researchers work on the Greater Paris project: Loïc Vadelorge of course, Paul Lecat, Paul Lesieur, Emmanuel Bellanger, Charles Riondet and Cédric Feriel. Cédric Feriel and Paul Lecat, Archival City’s scientific secretary until August 2022, made a major contribution to the design of the visualisation tool. We also have close links with the Archives de Paris and the Bibliothèque de l’Hôtel de Ville.

What did your work involve?

CM: Our project had multiple phases, involving iterative refinement of strategies. When I was a trainee two years ago, I had a presentation of what was being done. Elsa had processed scans from the Archives de Paris and converted them into searchable texts. Once processed, she checked whether the OCR scans were usable and what the error rate was. The error rate was very high.  So we started by doing some tests with the eScriptorium automatic text recognition software, which is used for handwriting. With other trainees, we used this tool to try and improve the OCR scans. We were able to correct and optimise some of them, but it took a long time.

The problem was one of overall OCR processing. We faced challenges due to formatting disparities spanning 1883 to 1970. As a result, the process resulted in a very high error rate. Then, as the overall OCR scanning was not good enough, we decided to scan by directory. We needed to give eScriptorium some indication of how the document was formatted, and to do this we had to manually transcribe a certain part of the document. This manual transcription in the software was a very tedious operation to carry out….

As this operation took too long and did not produce satisfactory results, we decided to abandon the eScriptorium OCR process. We therefore employed Elsa’s OCR scans, as a foundation, and corrected them during XML-TEI encoding.

What does XML-TEI encoding involve?

JF: TEI encoding entails rendering a text machine-readable. While we comprehend a text’s hierarchy visually (because of its graphic, we see that the title is bigger or smaller), computers lack this ability. To bridge this gap, we give a certain number of indications (this corresponds to that; this is an address; this is the name of a department; this is the name of a person, their function, etc.). We introduce tags with distinct functions and attributes, allowing machines to interpret the document’s structure.

CL: We add metadata. This makes the text machine-readable and enables searches to be carried out afterwards. For example, I can search by tag type. For example, the “persName” tag allows me to find all the people mentioned in the directory. In the 1896 directory (fig 1.), for example, there are 2,728 “persName” entries. If I look at all the directories, there are 63,359 references to people (fig.2).

Fig. 1- Search for “persName” (mentions of people) in the XML-TEI encoding of the 1896 directory: 2,728 results.

Fig. 2- Search for “persName” (mentions of people) in the full corpus of XML-TEI directory encodings: 63,359 results.

How do you learn to use this language?

CL: In the process of mastering this language, there are several aspects to consider: there’s the use of the XML language[3], which one needs to know how to manipulate, whatever the tags behind it are. It’s important to note that tags should adhere to specific standards for various purposes. These standards encompass EAD (Encoded Archival Description) standards for describing archives[4], and the TEI (Text Encoding Initiative) for encoding primary sources[5]: the most commonly used tags are ‘persName’, ‘p’ for paragraphs, ‘lb’ for line breaks; or the ‘choice’ tag, for texts with an old spelling, for example.

Imagine you have a text with typos or you are using an old version of the language, such as the one used until the 19th century: TEI allows you to say “this is what is written in the text: this is the original version; and the standardised version is what is added next to it”. This will allow you to have both versions at the same time, and to search using two words (Old French and Modern French). And then, in the displays, you can display what you want, you can display one or the other, or both by playing with the web features for example…

To sum up, there are a certain number of essential metadata tags such as document title and author. Then there’s the online documentation that one gradually learns to use. It is primarily available through the TEI consortium’s website[6], which offers comprehensive guidelines listing the current elements (tags) and attributes. Within the extensive range of TEI tags, our specific encoding scheme was initially defined by Elsa. Additional elements were added during the encoding phase as required. The need for these additional elements arose due to the heterogeneous nature of the directories encountered, resulting in the inclusion of new features not covered by the initial schema.

JF: … This was particular true of the 1968 directory, which we mentioned. We had to process it with a much richer schema; for example, we had to add special tags to say that it contained several images.

Do you encode all the directories?

CL: The initial approach encompassed transcribing and encoding all directories. However, this proved to be time-consuming, prompting a shift in strategy. We adopted a targeted approach, encoding directories from specific years and intervals, particularly during significant historical events such as world wars. These ‘pillar’ directories were fully encoded, but as it took about one full month to process each directory, we had to make choices: therefore, we decided to encode what interested us most, i.e. departments’ names, with the transcription and encoding of all the names of people in the departments of interest to the project’s researchers, the details related to these departments, etc.

Importantly, the original directory text was not deleted; instead, it was preserved as a comment within the TEI file, enabling future work, including transcription corrections and the addition of desired tags. In the coming weeks, we will create an index encompassing all individuals of interest, including those from the roads department, prefects, major administration personnel, and archives department staff (we were interested to see how they worked). This index will significantly enhance navigability and research capabilities.

JF: The advantage is that anyone who is qualified will be able to access our TEI document and continue the improvement work, as it is standardised and we have a ‘header’ section (i.e. metadata) that explains how we worked, our methodology.

CL: Furthermore, we will provide data papers and methodological explanations, facilitating the continuation and improvement of this work. This collaborative and standardized approach ensures that qualified individuals can access and enhance our TEI documents effectively.

Once the encoding is complete, what can researchers do with this work?

CL: Upon completion of encoding, researchers can explore and utilize our work in several ways. For those well-versed in TEI and query language, the TEI documents themselves offer extensive research possibilities. These researchers can conduct detailed searches and analyses using the provided TEI files.

For those less familiar with TEI, a visualization tool will be available  (fig.3), along with data extractions in various formats, such as PDFs, mind maps (fig. 4), and Excel exports. The latter facilitates the translation of tags used in the TEI encoding into a more accessible format. Researchers proficient in Excel can leverage this to conduct filter-based searches.

Fig.3- Summary of a directory as it will be displayed on the visualization tool

Excerpt from the model and graphic proposal produced by Jeanne Fras

Fig.4-Example of the mental map of the préfecture of Paris in 1905

Excerpt from the model and graphic proposal produced by Jeanne Fras

I’ll show you an example of the 1911 directory. Here’s the directory summary (fig.5)

Fig. 5-Summary, 1911 directory

 

It’s the encoding we’ve done that allows it to appear. We’ve embedded the entire hierarchy: we’ve said “from here to there, we have this administration and within this administration we have this or that department”. This allows us to have a summary of this type (fold-out) and then, behind it, when you click on it, you see the original page on the left and the transcription on the right.

As far as the design of the visualization tool concerned, we tried not to deviate too much from what was done for Bologna[7], so as to have something homogeneous. There’s a timeline that lets you click on each of the directories and then, if you click on it, it hierarchizes them. Of course, that’s just the first level, because if you unfold the whole thing, it’s unreadable. We’ll stick to the first two/three levels. When you click on “consult”, you land on the fold-out directory summary. After that, there’s another way of accessing the directories (apart from the timeline and mind map), and that’s via the catalog and search page. If you do a keyword search, all the directories will be displayed on the left, and if you click on a directory, you’ll see how many times that term has been found. You can consult the page in question or the entire directory if you like.

Here is an example of a PDF extraction of the 1935 directory from the encoding: if you go to this, the whole administration is unfolded. If I click on one of the titles, it takes me directly to the digitization, which is further down in the PDF. In terms of functionality, with these PDFs we have something much more fluid for navigating images and hierarchies than we’ll have with the visualization tool.

When the visualization tool is ready, will it be possible to carry out diachronic analyses?

CL: As for diachronic analyses, the timeline in the visualization tool presents the evolution of prefecture and city organization year by year, up to a few levels deep. Comparative analyses become feasible through searches. For people who are indexed, a small page will appear listing all the directories in which they are mentioned. However, there are many directories for which we haven’t transcribed everything. This means that there will be gaps in the search at the time of the query. If someone does a search on a person in the education administration, for example, they’ll get the directories that have been fully encoded, but they won’t see the directories where the education part hasn’t been encoded.

JF: In the visualization tool, there’s an indication that the documents have been processed with OCR and that, potentially, there are still typos.

Was the visualization tool created in collaboration with the agency Limonade & Co?

CL: Limonade & Co is our service provider for the visualization tool. Collaboratively, we communicate technical specifications, schematics, and other essential elements to them, guiding the design and development process.

CM: Jeanne has played a pivotal role in designing the site, providing mock-ups and graphical proposals that Limonade has further refined.

CL: It’s a technical and graphic collaboration between Limonade and us (technical and scientific teams; Archives de Paris…).

When will the visualization tool be ready?

JF: The visualization tool is expected to be in an advanced demonstration phase by the time of the program’s final symposium on November 9 and 10. While it may not yet be finalized or available online, it will be well-progressed and usable for demonstration purposes.

 

 

[1] The department was replaced by the department of Paris (including both the city and the department), the department of the Hauts-de-Seine, of Seine-Saint-Denis and of Val de Marne.

[2] Link to the eScriptorium software (in French): https://escriptorium.paris.inria.fr.

[3] Presentation of XML (or Extensible Markup Language) on the website of the consortium W3C https://www.w3.org/XML/

[4] Documentation related to the EAD by the Library of Congress : https://www.loc.gov/ead/

[5] Documentation related to the TEI by the consortium TEI : https://tei-c.org/

[6] https://tei-c.org.

[7] https://bologna.archivalcity.org/

 

 

---

---