1 of 50

2 of 50

Newspaper day 20/03/2023

3 of 50

Program

13u30-13h45: Introductions

13h45-13h55: Brecht Deseure, presentation of KBR newspaper department and CAMille project

13h55-14h05: Julie Birkholz, presentation of Digital Research Lab

14h05-14h25: Niklas Stenzel (ZOG)

14h25-14h45: Elias Degruyter (UGent)

14h45-15h05: coffee break

15u05-15u25: Vincent Ducatteeuw (UGent)

15h25-15u45: Leon Castelein (UGent)

15u45-16u05: Tess Dejaeghere (UGent)

16h05-16u25: Concluding discussion

3

WiFi network: KBR_Readers

Access code: KBR_2019

WiFi network: KBR_Events

Access code: Ghs442uF

4 of 50

Feedback

Green post-its

What can KBR do to improve access to the newspapers collections for researchers?

Yellow pos-its

Ideas for future newspapers days.

Establish a network/mailing list?

Yearly event with open cfp?

Focus on newspapers or also include periodicals?

…

4

5 of 50

KBR Newspapers department

•Largest and most diverse historical newspapers collection in Belgium

•From the 17th century until today

•Legal deposit: 1 physical issue of each Belgian newspaper deposited daily. Digital legal deposit underway

•Ca. 2000 Belgian titles of the ‘grande presse’

•Ca. 1000 Belgian titles of the ‘petite presse’

•Ca. 500 foreign newspapers from 60 countries

•Colonial newspapers

5

6 of 50

Special collections

•Fonds Gaston Mertens: over 60.000 specimens from ca. 1.650 Belgische localities (17th century - 1948)

•Fonds Philippe Vandermaelen: 2.405 specimens of foreign newspapers of the 19th century (Europa and USA)

•Fonds Jacques Delmelle: complete journalistic oeuvre of one Belgian journalist, comprising over 10.000 articles/publications

6

7 of 50

Access

•One of KBR’s most consulted and vulnerable collections

•Different supports: paper, microfilm (600 titles), digital

•127 digitized and OCR-ed titles (1814-1950) = ca. 4 million pages

•1950-1989 (= ca. 3,5 million pages) will be digitized within the following three years

•Until 1918: freely accessible, also remotely

•After 1918: only accessible on KBR computers (or remotely with special permission)

•Simple search functions

7

8 of 50

8

9 of 50

9

10 of 50

10

11 of 50

CAMille FED-tWIN (KBR-ULB)

Goals

Writing the history of Belgian journalism
Digitizing Belgian newspapers and journalistic archives

Components

CAMille platform of digitized newspapers and archives (data level access)
Database of Belgian journalists
Research projects

11

12 of 50

The CAMille platform

12

13 of 50

Database of Belgian journalists

Prosopography of Belgian journalists (1830-today)
Paper dictionary inherited from Pierre Van den Dungen, transformed into spreadsheet
Enriched via data extracted from library catalogues
Development of relational database underway

13

14 of 50

Research projects

PhD project Alexia Vidalenche
Analysing discourse about journalists and journalism. Current focus: women journalists
Reconstructing journalistic careers, e.g. Alice Bron (with Alexia Vidalenche, Florence Le Cam, Manon Libert and Sébastien de Valériola)
Reconstituting editorial teams
Press releases of the Belga press agency
Developing DH solutions for newspapers corpora: signature recognition, OCR improvement, topic modelling (with Isabelle Gribomont)

14

15 of 50

15

16 of 50

16

Digital Research Lab�

Julie M. Birkholz, Lead of KBR’s Digital Research Lab & Assistant Professor Digital Humanities,

Ghent University

17 of 50

Goals

support and facilitate data level access to KBR’s diverse, multilingual digitised and born-digital collections;�
stimulate the (re)use and research of these digital sources, data and metadata of these collections.

17

18 of 50

in practice

Supporting research on datakbrbe, BESOCIAL & BelgicaWeb;
SUpporting interns and students doing DH research on the collections;
Coordinating the future FWO Junior Project on DH in Navez’s correspondence & historical networks; & ESFRI Virtual Lab- development of a data on demand service

18

19 of 50

DATA-KBR-BE

Facilitating data-level access to KBR’s digitised and born-digital collections for digital humanities research

https://www.kbr.be/en/projects/data-kbr-be /

20 of 50

Digitised Historical Newspapers as Data

https://www.kbr.be/en/projects/digital-research-lab/

https://www.kbr.be/en/projects/data-kbr-be/

21 of 50

Interdisciplinary Research Scenarios

22 of 50

From Collections to Corpora

Vooruit: socialistisch dagblad

https://en.wikipedia.org/wiki/General_strikes_in_Belgium

Which newspapers are digitised?

Importance of historical context

23 of 50

Zentrum für Ostbelgische Geschichte

https://www.geschichte.be/quellen/

24 of 50

German Language Newspapers in Belgium

https://opac.kbr.be/LIBRARY/doc/SYRACUSE/16388591

25 of 50

Discursive Identity Construction in East Belgium

�

A Linguistic and Discourse Historical Analysis of the Patterns of Language and Communication in East Belgian Mass Media

25

26 of 50

�

Agenda

Research interest
Link to digital newspapers
Methodology
Discussion

26

27 of 50

�

Research Interest

How are East Belgian identities constructed in public media discourses?

Which contents are objects of discourse?
Which communicative strategies are used to establish a certain identity?
How are these strategies realised in terms of language use?

How is the situation of East Belgium as a border region referred to? Does the relation towards Germany or the rest of Belgium play a role?
Does the construction of identities itself become subject of discourse?

27

28 of 50

�

Link to digital newspapers

A lot of our knowledge about the society we live in is influenced by mass media (e.g. newspapers)
Digitisation makes them usable for corpus linguistics

28

29 of 50

Methodology

Discourse historical approach (DHA)

Synchronic and diachronic linguistic discourse analysis that considers historical background information and sources
Three levels of analysis

Contents
Strategies
Forms of realisation

Computer aided corpus linguistics

Combination of qualitative and quantitative methods
Search engines, KWIC, n-gram-analysis, cluster-analysis, collocation-/cooccurrence-analysis

Niklas Stenzel

Zentrum für Ostbelgische Geschichte

Kaperberg 2-4

B-4700 Eupen

www.geschichte.be

30 of 50

�

Methodology

Corpus construction

Identifying events that could probably be associated with collective identity
Search for articles within a certain time period around the event that refer to it
Step by step extension of corpora by identifying keywords through qualitative analysis
Converting to txt and tagging for better workability with corpus linguistic tools

30

31 of 50

Methodology

AntConc

Freeware corpus analysis toolkit for concordancing and text analysis

Niklas Stenzel

Zentrum für Ostbelgische Geschichte

Kaperberg 2-4

B-4700 Eupen

www.geschichte.be

32 of 50

Discussion

Niklas Stenzel

Zentrum für Ostbelgische Geschichte

Kaperberg 2-4

B-4700 Eupen

www.geschichte.be

33 of 50

Presentation Elias Degruyter

https://drive.google.com/file/d/12-QqeNw4Pr6hHJWTxtanOHrer0ridMZ-/view?usp=sharing

33

34 of 50

Presentation Vincent Ducatteeuw

34

35 of 50

Presentation Leon Castelein

35

36 of 50

Ownership of public space

Research of collective action in Ghent

Gita Deneckere: “Sire het volk mort”
Theory by Charles Tilly: changing of repertoires�

Criticisms

Relevancy of locality and territoriality
Martin Schoups: “Meesterschap over de straat”
Hypotheses

36

37 of 50

Data collection & newspapers

37

38 of 50

38

39 of 50

Spatial analysis

Katrina Navickas, Protest and the politics of space and place, 1789-1848, 180

39

40 of 50

40

41 of 50

Beyond Babylonian Confusion �a case-study based approach for multilingual NLP on historical literature

Author: Tess Dejaeghere

Supervisors: Julie Birkholz, Els Lefever, Christophe Verbruggen

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101004984

42 of 50

42

43 of 50

Personalia

Background

Applied Linguistics -Translation/Interpreting – Digital Humanities & Digital Text Analysis
PhD student @ Ghent Center for Digital Humanities:

Deliverables CLSINFRA Working Package 8

ML Named Entity Recognition (NER) pipeline
ML & lexicon-based Sentiment Analysis (SA) pipeline
Relation Extraction (REX) pipeline

Personal PhD research

Title: “Beyond Babylonian Confusion: a case-study based approach for multilingual NLP on historical literature”

So first up, some background information. I am not a historian, nor a literary scholar, but I have a background in applied linguistics (specifically translation and interpreting). After that I further specialized in Digital Humanities and, more specifically, Digital Text Analysis – which is a Master’s programme at the University of Antwerp specifically aimed at the analysis of text data through AI, statistics and NLP.

Since October last year, I’ve been working as a PhD student at the Ghent Center for Digital Humanities. My work is actually two-fold; I’m partly involved with the deliverables promised by working package 8 (coordinated by Julie Birkholz present here) as part of the CLSINFRA-project, which includes the creation of multilingual NLP pipelines for applications such as Named Entity Recognition, Sentiment Analysis and Relation Extraction. But my main occupation is my (closely related) personal PhD project on the application of NLP to literary-historical texts.

As I said before, title of my proposal starts with “beyond babylonian confusion”. This babylonian confusion in part refers to the multilinguality often playing a big part in literary-historical data processing, but it also refers to two research fields which currently struggle to find one another, because they have a hard time communicating. Namely: Natural Language Processing and the Humanities, digital and non-digital.

44 of 50

Ongoing Research

Natural Language Processing vs. (Digital) Humanities

(Digital) Humanities: analyze text(s) according to a hermeneutic model to make inferences about past and present.

Stylometry
Authorship attribution
…

Natural Language Processing: subfield of artificial intelligence and computational linguistics aimed at making natural language understandable to computers.

Machine translation
Speech recognition
Chatbots
…

45 of 50

Ongoing Research

Natural Language Processing vs. (Digital) Humanities

(Digital) Humanities: analyze text(s) according to a hermeneutic model to make inferences about past and present.

Stylometry
Authorship attribution
…

Natural Language Processing: subfield of artificial intelligence and computational linguistics aimed at making natural language understandable to computers.

Machine translation
Speech recognition
Chatbots
…

46 of 50

Ongoing Research

Natural Language Processing vs. (Digital) Humanities

Named Entity Recognition (NER)

Automatically extract and classify entities from a text.
Commercial applications in e.g. text summarization and question answering systems.

Example: “Julie works in Ghent.”

Sentiment Analysis (SA)

Automatically extract sentiment from a text.
Commercial applications in e.g. opinion mining and social media mining.

Example: “Julie was happy with her cup of tea!” 🡪 POSITIVE

PERSON

LOCATION

Now, Natural Language Processing is an umbrella term referring to a lot of techniques, and a couple of the more prominent ones are Named Entity Recognition and Sentiment Analysis, which are the two NLP-techniques that we will be focussing on in this research.

I’m sure you are all acquainted with these terms, but I’ll briefly explain them for clarity’s sake.

Named Entity Recognition is a task which can be applied to a text consisting of two parts: firstly it identifies the entities in a text, and secondly it classifies them. For example, like the example on the slide, for the sentence “Julie works in Ghent”, the tool would be able to identify Julie as an entity, and classify this entity in the category “person”. Then it also identifies the entity “Ghent”, and labels this entity as a “LOCATION”. Named entity recognition is currently widely applied in for example tekst summarization and question answering systems.

Sentiment Analysis is another task which allows users to automatically extract the sentiments from a text. This could be in just three categories/classes (Positive, neutral or negative are common ones), or it could be more fine-grained. So categories could also be for example angry, sad or even jealous. In the example: “Julie was happy with her cup of tea”, the program can recognize that the sentence can be labeled as “positive”.

47 of 50

Ongoing Research

The gaps

! Need for transparent, reproducible and durable NLP workflows which are tailored to heuristic research.

! Need for an overview of the possibilities and limitations of NLP-tools in literary-historical research settings.

! Need for practical insights regarding NLP application to data with literary-historical characteristics.

! Need for workflow communication standards.

🡪 Facilitate exchange of practices between NLP and DH.

Clearly, applying NLP in DH work currently feels like you’re standing in a big forest. And you know there’s paths you can walk on, but you have no idea how to get there and which path to walk to get out. DH-papers with NLP appplication can be very obscure and they assume a lot of technical knowledge. There are increasingly more calls for a methodology-driven agenda in DH, but a couple of gaps still need to be addressed if we want to facilitate a true exchange of practices between NLP and DH, of which both fields will reap the benefits.

Clearly, there is a need for transparent, reproducible and durable NLP workflows which are fit for heuristic research settings. How can we use these tools, not simply just to generate output, but also to transform this output in visualizations and statistics which can be interpreted by the historian or literary scholar. In short, we need to take a step back from making a workflow adequate for that one special case study, and build sound infrastructures which are widely applicable and can form bridges between these research fields.
We also need to foster a tool-critical attitude within DH. There is a need for an overview of not only the possibilities, but also the limitations of NLP-tools in DH-settings, so researchers are aware of the fact that these tools do not always render complete results. More specifically: why and how do we select adequate tools? What does the tool tell you as an end user, and where do we have to be careful to not draw big conclusions?
Thirdly, there is a need for practical insights regarding NLP application to data with literary-historical characteristics – how do you deal with your data? How can you make your data fit for NLP application?
And last but not least, once we designed a sound methodology – how can we communicate our research workflow in a clear and step-wise fashion? What technical details do we have to mention, and how can we publish and dissemminate the code that we used.

48 of 50

Ongoing Research

Approach: case-study based

Rooted in practice and flexible

OCR output evaluation
NLP tool evaluation
Tool application
Visualization
Interpretation
Dissemination

! Suggestions regarding tool selection, tool output interpretation, workflow dissemination

So how will we reach these objectives? We are currently opting for a case-study based approach.

We will design workflows based on a research question which can be answered through NER or SA or a combination of both.

These workflows could processes such as (…) which do not necessarily have to be taken in a linear fashion, nor do they all have to be present in each workflow simultaneously.

Besides working out th ecase study, we will reflect on the methodology itself – we will list what worked and what didn’t. The tools that were used will also be explained and their pitfalls will be taken into account when interpreting these results.
Once ready, the case study workflows themselves will then be very clearly communicated in a step-wise fashion through Jupyter Notebooks with commented code.

This way we surpass the methodological obscurity of papers, and transform it into something that is transparent and reproducible.

This methodology also allows us to be very flexible regarding the data we use and the type of case studies we want to implement. It is also a great way to convey workflow information in a DH project-context, because it accepts that methodologies in DH cannot be one-size-fits all, and they can never be rigid. because every DH-project is so unique and has such particular needs that it’s not possible to make turn-key methodologies. So what we want to do is work towards these case studies which can serve as transparent step-by-step example projects which can be easiliy adapted and reproduced by future researchers so we have a more open technical discourse which should eventually facilitate NLP-integration in DH.

49 of 50

THANK �YOU

Tess.Dejaeghere@ugent.be

50 of 50

Feedback

Yellow post-its

What can KBR do to improve access to the newspapers collections for researchers?

Blue post-its

Ideas for future newspapers days.

Establish a network/mailing list?

Yearly event with open cfp?

Focus on newspapers or also include periodicals?

…

50