To celebrate the 10th anniversary of the Helsinki Digital Humanities Hackathon, a participant and a team leader shine a spotlight on its spirit and inner workings. To do this, we provide a glimpse into two projects that delved into two historical areas, colonialism and the Holocaust, following traces in newspapers (for the former) and oral histories (for the latter) via machine learning, LLMs, and digital methods.
The Helsinki Digital Humanities Hackathon (DHH) is a different kind of summer school. At this award-winning event, approximately 40 students in humanities and social sciences collaborate with computer scientists to develop a project from start to finish in ten days. The Hackathon participants are divided in four thematic groups led by one to four team leads that propose specific data that their group will work with and can offer content or methodological expertise; their task is to guide participants regarding data and methods, while nurturing collaboration. The true objective of this hackathon is to show students the value added in collaboration and while data approaches or innovative methods are applied, they should address real humanities questions and problems. Since its first edition in 2015, this event has inspired cultural heritage organisations and researchers to experiment with many types of data, from historical newspapers, parliamentary debates, tweets, and poems, to Eurovision songs, and this year, for the first time, oral history interviews1.
With the support of DARIAH and CLARIN, DHH has become an international event, offering bursaries to participants from abroad. DHH25 hosted 36 participants, of which 14 travelled from universities in Germany, Italy, UK, France and Slovenia, but altogether, this was a global Hackathon with participants from three continents speaking more than twenty languages.
This year the Hackathon offered four thematic areas2 but as both authors of this post have a shared interest in historical data, we delve into two projects that focused on European events that had a global impact which continues to the present day: Colonialism and the Holocaust. This is not the first time the Hackathon has tackled contested topics. Nationalisms and fake news have been explored in earlier editions demonstrating the commitment of the Hackathon organisers to show students how they can use digital methods to more deeply understand societal and historical phenomena.
In the 17th and 18th centuries, newspapers became a popular medium that reflected the everyday life of a changing society. What do the newspapers tell about these changes and their implications? If we look at large quantities of newspaper data over time, can we learn something about history that may remain otherwise hidden?
With the guidance from the team leaders (Ville Vaara, Jonas Fisher, Ke Shu), the group brainstormed topics such as social class and gender issues within the context of the colonisation. These difficult topics prompted us to contemplate the complexity and the dark side of the colonial project. We used the Burney3 and Nichols4 Collections, which include English newspapers with 1.7 million articles from 1649-1804, leveraging OCR text and page images. Through examining these newspaper collections via digital humanities approaches, we wanted to pose questions that would tackle the novelties that colonialism introduced to society in Britain, and we wanted to explore impact both in Europe and in the lands it colonised. The group addressed this with diverse incursions into the data. We will focus here on two: firstly, we inquired into the geographic organisation and distribution of goods; and secondly we examined advertisements to better understand the emerging consumer society in relation to the colonial project.
To explore the geographies of colonialism we used the full content of newspapers from which we traced locations relating to “sugar” and “tobacco” which largely appeared in maritime trade news and ship bulletins. The places referenced were mostly harbours from which the goods were shipped. The absence of names from plantations in the Southern States of North America indicates that the produce was not sold directly by plantations but were bundled and transported to the harbours, where trade most likely occurred. In addition, Caribbean and New England colonies most often appeared in ship bulletins concerning the movement of tobacco. This is not surprising as it was a well-documented trade route that formed a part of the triangular slave trade5. The view of the colonial landscape revealed in newspapers was closely associated with colonial goods and trade routes and did not reflect the realities of colonial production and politics.
Following this, to better understand the consumer society in colonial Britain, we examined in what contexts these goods were mentioned in advertisements through a sample extracted from the page scans. For this second inquiry, we generated word clouds based on n-grams (see below). One interesting finding was that references to colonialism (such as terms including “colonies” or “plantation”) were connected to sugar, but such associations are absent for other goods, such as tobacco. Overall, colonial products are often presented together, even when they do not originate from the same regions. Within the advertising narrative, they tend to merge with locally produced goods and are primarily promoted for their quality, with their colonial origins fading into the background. It was natural for consumers to see the product sources as part of British economic structure instead of colonised countries.
Close-reading the advertisements also highlighted how the language of the ads served to stimulate new consumption practices.
In addition, colocations of “tobacco” could indicate that “snuff” gradually became a new colonial luxury for the upper class by the end of the century:
These explorations inspired us to delve into the globalized colonial economy and power dynamics when trying to understand the changing social structure and consumption phenomena. Tackling the historical subject through Digital Humanities approaches has provided valuable perspectives to look at global issues at present as well, since exploitation and oppression is still relevant in the context of globalization today and should not be neglected.
Our process and findings were documented in detail in this poster. During the hackathon, interdisciplinary collaboration was both crucial and challenging, especially when we were working on abstract and difficult concepts. We took the time to share our different perspectives and find common grounds, which allowed us to land on suitable approaches rather than the fanciest technical solutions. However, the explorations and interpretations are limited based on what we find using a specific data set and methods. Nonetheless, the project succeeded in opening for participants new questions and exploring computational methodologies that could be used to analyse newspaper content.
Encounters and recorded testimonies from witnesses and survivors of the Holocaust are essential sources of education and sensibilization about one of the most traumatic events of the twentieth century. A wealth of Holocaust documentation has been collected by the US Holocaust Memorial Museum. This database aggregates testimonies from diverse sources and is the most comprehensive digital archive giving open access to full-length interviews. Thanks to previous work by Isuri Anuradha and Prof. Martin Wynne, a pre-curated multilingual subset of 100 interviews8 was provided to the group who undertook a layered exploration into the emotions captured by the interviews. This focus was decided by students unanimously in our first preparation meeting, and we, as team leaders, could not but support it. Anybody who has experience with in-depth, unstructured interviews knows how the expressive nuance of the interview that is captured in audio or video can vanish in the transcript.
The motivation for this project can be read in this blog-post by participant Ellen Yang. Today, as only few witnesses and survivors remain alive, these interviews may be our most direct link to understanding the personal experiences and traces left by the Holocaust and to learn about how people who lived through or witnessed the Holocaust coped with the trauma.
Conventionally testimonial interviews consist of close ups or middle shots and interviewers are positioned close to the camera, hence interviewees look at their interlocutor and face the camera at the same time. Here are two interviews selected for the project: On the left, Judith Zucker, suvivor of labor Camp Nováky in Slovakia who later emigrated to the US. On the right is US Major Alfred Sundquist, a physician involved in the liberation of Wöbbeling concentration camp in Germany.
The hypothesis posed by the students was to ascertain if there were matches between sentiment analysis performed on the transcripts and the automated emotion detection analysis of facial expressions. Additionally, the group sought to understand 1) if similar information could be extracted from the tone of voice or speech patterns, and 2) to verify if the emotion labels provided by computer models9, could be validated by human interpretation when watching the interviews. While the group’s outcomes remained in this technical task, the motivation was to open vistas for future projects, for example, to facilitate enrichment of transcripts with sentiment annotation, or identification of key emotional moments. For long and unstructured interviews such as these, the project found it is a worthwhile effort.
Rather than focusing on the outcomes10 (which can be read in this poster), we want to focus on the collaborative nature of the Hackathon, and the visualization above helps explain this. Transforming an oral history interview into “data” requires breaking tasks into comparative and interconnected bits so that three media can be analysed in tandem: the conversation contained in the transcript, the audio wave signal, and the sequence of headshots. The group distributed tasks in pairs to those who had experience working with each media, while the humanities researchers oversaw selecting data and verifying results after each extraction or analysis pipeline. Due to time and technical constraints, the group selected 10 less lengthy interviews (5 English and 5 Polish), which, nevertheless, contained emotional variation while maintaining gender balance and type of testimony (survivors vs witnesses).
Our daily afternoon meets allowed for reality-checks, re-routing, and essential collaboration. One day, one group announced that processing transcripts into machine readable format would take longer than anticipated because transcripts lacked speaker-ids and timestamps. In addition to the processing time, they were at loss as to how to sync transcripts to the other signals. At this point the audio team dropped everything to generate unified timestamps across signals facilitating the desired comparative approach. These moments also allowed for celebrations, for example when the computer vision model gave similar results as the sentiment analysis of the transcript of the same interview; and a few days later, the audio analysis agreed (see below).
In conclusion, we hope that next Spring a similar group of young talented researchers will assemble in Helsinki. As in previous years, the Helsinki DH Hackathon organisers and team leaders will make a careful selection of historical and contemporary data for participants, both students and team leaders, to collaboratively explore uncharted academic terrain or to advance the routes opened by previous projects such as these two.