Daejeon, July 27–31
An iconic moment in Swedish television history is the 1975 broadcast from the West German embassy siege in Stockholm, a hostage stand-off initiated by the terrorist group the Red Army Faction (RAF). The dramatic climax, when a bomb explosion rocked the embassy building, has been etched into public memory through the grainy, night-time television footage of distressed news reporter Bo Holmström exclaiming: “Go live! Go live!” (“Lägg ut! Lägg ut!”). This moment, both live and discursively constructed, highlights television’s role in framing terrorism as a media event, while also marking the emergence of so-called international terrorism as an urgent national issue that has captured the Swedish public imagination since the late 1960s.
Although previous studies have shown that a modern notion of terrorism entered Swedish public discourse around 1970 (Fridlund et al. 2022; Brodén et al. 2023), the pivotal role of television in shaping its perception remains underexplored. This paper addresses this gap. Building on research that situates the modern concept of “international terrorism” as a discursive product of the global 1968 and the Cold War (Stampnitzky 2013; Zoller 2021), we develop and pilot a context-sensitive multimodal workflow for large-scale analysis of historical television, specifically tracing how Swedish public service news integrated “terrorism” as a modern media category.
Methodologically, we integrate distant viewing (Arnold and Tilton 2019) with media historical contextualization (Guldi 2023), aligning speech-to-text outputs with visual and sonic features of news broadcasts, drawing on the digitized collections of Swedish television broadcasts of the National Library of Sweden (Kungliga biblioteket; hereafter KB).
Our paper is situated within the emerging subfield of digital film studies (Heftberger 2019; Dang et al. 2024). As recently as 2019, observers noted that digital humanities methods remained largely “deaf and blind” to non-textual media (Fickers et al. 2018). Only recently have digital film studies scholars begun to use algorithms for image mining (Eriksson et al. 2022; Stelmach 2024), leaving television’s cross-modal character – where moving images, sound, and editing converge – underexplored. Over the past decade, there has been a gradual increase in scholarship engaging with speech-to-text methods for large-scale analysis and distant viewing of audiovisual data (Fickers et al. 2018, Heftberger 2019, Burghardt et al. 2020). Within the Swedish context, however, there are virtually no studies that systematically integrate broadcast audio and video as primary data (Stjernholm 2022).
Theoretically, we proceed from the theoretical assumption that aesthetic elements in television not only reflect but also structure meaning-making. Because televisual narratives extend beyond mere representation of terrorist acts and public discourse, we understand broadcast mediation as shaped both semantically (statements, interviews, commentaries, voice-over, etc.) and formally (framing, editing, sound, etc.) (Prince 2011; Tavares Furtado / Auchter 2024). Accordingly, we treat television news as a site where period-specific and distinct repertoires of audiovisual narratives about terrorism are constructed.
This pilot develops a form of large-scale multimodal analysis of television archival material, that traces (i) the lexical meanings attached to terrorism in speech, and (ii) how visuals and sound depict forms of political violence framed as terrorism, weighing these patterns against close readings and contextualization informed by television history and terrorism studies.
The material consists of Swedish public service television news from the 1970s, a period in which terrorism became an increasingly visible issue of national and international concern. From 1970, Sweden operated a two-channel public service system, with TV1 and TV2 under Swedish Radio and organized through a shared broadcasting mandate. The main news programmes were Aktuellt on TV1 and Rapport on TV2, launched in December 1969. Together, these programmes structured televised reporting and shaped public understandings of current events.
Any large-scale analysis of this period must account for the uneven character of the surviving archive. Before the introduction of legal deposit requirements for audiovisual media in the late 1970s, preservation practices at Swedish Radio were selective and inconsistent, shaped by institutional priorities, technical constraints, and storage conditions. The surviving material from the early and mid-1970s thus takes the form of a partial archival record.
For this pilot, the analysis maps preserved news material from Aktuellt and Rapport between 1970 and 1979. The corpus consists of 540 recordings, amounting to approximately 9,230 minutes of video. This represents a fraction of the original output, yet provides a sufficient basis for testing speech-to-text search and audiovisual analysis. The material remains unevenly distributed across the decade, but its internal consistency as public service news allows for tracing broad patterns in the mediation of terrorism while identifying segments for closer analysis.
The analysis begins with the generation of searchable transcripts from selected broadcast material. Rather than relying on existing catalogue metadata, speech-to-text is performed using Whisper large-v3 through the faster-whisper implementation. The model runs in Swedish transcription mode with conservative decoding settings, voice activity detection, and anti-hallucination thresholds, producing time-coded segment-level CSV files. These transcripts function as a discovery layer, enabling the identification and comparison of passages containing relevant terms.
In this pilot, the analysis traces occurrences of terror across the preserved news material and compares its distribution with other prominent topic of the period. This comparison situates terrorism-related language within a broader news agenda and allows for an assessment of its prominence in relation to archival density. The results show a marked peak in 1975, likely linked to increased attention following the West German embassy siege in Stockholm. At the same time, they indicate a more continuous presence of the topic across the decade. Terrorism-related discourse can thus be traced in the spoken content of the archive even where conventional metadata remains silent.
Selected segments are then examined through multimodal analysis, where machine-generated labels are treated as heuristic aids. In the visual domain, open-vocabulary object recognition is applied, with an initial vocabulary generated through exploratory runs with Moondream and object detection carried out with OWL-Lite. In the audio domain, Audio Spectrogram Transformer analysis is used to identify recurring sound events and acoustic patterns. These outputs are normalized and interpreted alongside transcripts and video material, with attention to recurring objects, sound types, visual settings, and sonic atmospheres.
This approach also brings forward the role of error in computational analysis. Recognition models describe historical television footage from a culturally and contextually distant position. They do not interpret a broadcast as a historically situated news event, but register objects, sounds, surfaces, and settings in a partially decontextualized form. This distance can be analytically productive, since it redirects attention toward features that often remain peripheral in conventional close viewing, such as windows, street façades, police formations, and the acoustic construction of liveness. Computational outputs thus produce a form of estrangement as they offer a displaced description allowing for familiar material to be reappraised from a new interpretive angle.
Balancing automated search with a media-historical approach that situates the broadcasts in their cultural context, we adopt a layered method that avoids narrow, purely data-driven analysis by attending to the “historical and contextual complexities of the documentary record” (Guldi 2023).
More broadly, television coverage of terrorism took shape alongside the expansion of television reporting on wars and conflicts around the world, the rise of live news, shifts in Swedish television aesthetics (including the adoption of color television), and the wider institutional and technological transformations of postwar public-service television (Weibull / Djerf-Pierre 2001; Bignell / Fickers 2008; Nykvist 2011; Djerf-Pierre / Ekström 2013; Scannell 2014; Wahlberg 2017).
Specifically, we move iteratively between distant viewing and close reading of the TV2 broadcasts to test and substantiate our results, noting where relevant camera work, voice-over narration, and editorial structure. We also, to some extent, take into account the aestheticization of different forms of militant violence (aircraft hijackings, bombings, kidnappings, etc.) and situate the analysis within the wider historical contexts of the emergence of international terrorism in the 1970s, associated with militant organizations such as the PFLP, Black September, the RAF, and the IRA (Chouliaraki 2006; Melzer 2015).
We conclude by outlining how our context-sensitive multimodal approach yields methodological insights for large-scale analysis of macro-level patterns concerning the terrorism discourse in Swedish television news and for identifying specific segments suited to close reading. Although our integrative approach is exploratory, scaling it will produce a decade-long, single-programme corpus that is internally consistent enough for in-depth analysis yet remains representative of 1970s Swedish television news conventions. Perhaps most notably, our preliminary findings from this limited case study also indicates that visual and sonic features, to some extent, prefigures the subsequent stabilisation of the spoken discourse on modern terrorism discourse.