As in many European countries, Swedish parliamentary records are digitized and freely available (http://data.riksdagen.se), making the study of political discourse ripe for data-driven research. This study explores the task of devising an annotation framework for mixed-methods humanities and social science (HSS) research using parliamentary data. Its origin is the ongoing research project Terrorism in Swedish Politics (SweTerror) that has raised interest in the possibility of constructing a general framework for semi-automatic annotation of salient topics in the parliamentary discourse. The aim of the paper is to discuss our exploratory work on such a framework, drawing on a limited case study of 'terrorism' as a topic in the Swedish parliamentary debate 1968-1970.
In the following, we outline a three-category framework for semi-automatic annotation. We situate the discussion in the context of the SweTerror project and the Swedish parliamentary constitution during the period in focus. We present the framework through its three main data categories: metadata, language data, and frame data. The first two are general or generic and could to a large extent be automatically extracted, using computational resources, while the third is almost entirely dependent on manual annotation. The frame data category is designed to facilitate contextual discourse analysis of a specific topic or issue in the parliamentary debate, in our case study 'terrorism'. We present this category through some theoretical distinctions concerning the concepts of framing and terrorism, which Digital Parliamentary Data in Action (DiPaDA 2022) workshop, Uppsala, Sweden, March 15, 2022. EMAIL: mats.fridlund@gu.se (Mats Fridlund); daniel.broden@gu.se (Daniel Brodén); leif-joran.olsson@gu.se (Leif-Jöran Olsson); magnus.pettersson.angsal@gu.se (Magnus P. Ängsal) ORCID: 0000-0002-5759-0027 (Mats Fridlund); 0000-0002-5914-1516 (Daniel Brodén); 0000-0001-7107-4101 (Leif-Jöran Olsson); 0000-0001-5996-5067 (Magnus P. Ängsal) have guided its design. Then, we discuss the framework's possibilities through concrete examples concerning terrorism in the Parliamentary debate 1968-1970. We conclude by suggesting how this tentative framework may be more generally applicable for HSS studies of parliamentary discourse.
While many research initiatives within the digital humanities have been conducted with either a computational or HSS focus, the aim of the SweTerror project is to truly combine HSS and language and speech technology domain expertise. Combining digital and interpretative methodologies, the project studies the sustained role of terrorism in Swedish politics 1968-2018, and, in particular, the deliberations of the Diet in the Swedish Parliament, Riksdag [1].
The Riksdag of the Estates was in 1866 reconstituted as a bicameral legislature and later in 1971 as a unicameral legislature. Thus, the period 1968-1970 studied in this paper, was the final years of the bicameral Riksdag when it consisted of the First Chamber, with 155 Members of Parliament (MPs) and the Second Chamber, with 233 MPs. The Swedish electoral system proportionately reflects the electorate's divisions in the elected body and during this period, six parties were represented in the Parliament: the Centre Party (C); the Liberal People's Party (FP); the Moderate Coalition Party (M); Civic Unity (MBS, only 1964-1968); the Social Democratic Workers' Party of Sweden (S); and the Left Party -the Communists (VPK). Historically, Swedish politics has been dominated by the Left-Right conflict, with the political parties understood as either belonging to the Left or to the Right.
Our study concerns the parliamentary communication form 'Speeches' (anföranden), i.e. public speeches and replies by MPs in the debates of the two chambers, as represented by their digitized transcripts available at the Riksdag website riksdagen.se/sv/dokument-lagar/. Specifically, in the debates MPs rhetorically argue and present their politics, raise issues, question or support government policies, etc. An MP may participate in debates without approval from the party leadership, or by being restricted by formal rules. The most common speeches are interpellation, parliamentary business and debates on general policies. MPs who initiate a debate can speak longer, while MPs can briefly reply to any speech. On a general level these exchanges of meaning are the opportunities that MPs have at disposal to review, control and to hold the government accountable [2]. In international perspective, the Swedish debates have been described as steering at verifying facts and technical in nature, with emotional outbursts being rare [3]. The speeches are transcribed and made public shortly after the debates. However, they are not transcribed exactly as spoken but edited for readability, formality and clarity and it has not been uncommon for the stenographers to include speakers' manuscripts into the debate protocols. Focus has been to represent the speaker's intended meaning rather than their in/exact wording [4,5].
The starting point is that we 'trawl' through a corpus of parliamentary speeches for keywords signifying the topic under study (topical keywords), in our case potential terrorism-related terms, identified as containing the lexeme terror. In our case material during 1968-1970, we identified a substantial number of keyword hits (71 words used in 68 sentences in 35 speeches uttered by 36 different speakers). In the illustrations of the framework's three data categories, we show only the five first and last of the 71 hits (see Figures 1-3). Notably, in our analysis the salient topic in the debate is primarily represented and understood through its core terms ('terrorism' and other closely related keywords), rather than synonyms or homonyms ('urban guerilla', 'violent extremism', etc.), which an extended analysis should also take into account.
A first step towards creating a framework that enables a detailed analysis of parliamentary debates is to select metadata labels to annotate basic parameters related to the document context for each word occurrence; the when (date), where (Chamber), why (what debate) a specific occurrence took place as well as who (speaker, party, government party member) uttered the word was attributed to and their formal role (position) (Figure 1).
We chose to include the following labels in our metadata category, of which several of the data types may be generated semi-automatically from the available open data (in italics):
• ID: unique identifier for each speech utterance (e.g., No. 3 with ID 196801182002) where the identifier includes date (19680118), Chamber (2) -First or Second -where the debate took place, and the utterance's order (002) in the debate (sole/first, second, third, etc. terrorismrelated occurrence in the debate); • Date: the date of the speech according to ISO 8601;
• Chamber: whether speech given in first or second chamber (only relevant 1866-1971);
• Debate: title of the specific debate in which the speech occurred;
• Person: speaker's surname and last name;
• Party: speaker's party affiliation;
• Gender: speaker's gender;
• Government party (Y/N): whether the speaker is a member of a party in government;
• Position: the speaker's formal role in Parliament (ordinary MP [ledamot], group leader, party leader, Minister, Prime Minister), leaving out that party leaders, Ministers and PMs most often are also MPs. While these labels are more or less self-evident, it should be noted that the label of 'Government party' (Regeringsparti) is highly significant in that MPs from government parties tend to use the opportunity to give speeches in the chamber to a lesser extent than MPs from oppositional parties. Two main ideas lie behind the construction of the metadata category. The first is that the category's labels could be understood as 'basic' when it comes to the information they hold for analytical extractions. Second, that the majority of their informational content may be automatically generated and extracted from the Riksdag's Open Data or other parliamentary databases. A few labels ('Gender', 'Government party' or 'Position') may be generated semi-automatically through the manual determination of their content using external databases, including wikipedia.se and the online version of Swedish Biographical Dictionary (Svenskt biografiskt lexikon). It is expected that we will in the future be able to automatically determine the data in these labels through enriched curation of parliamentary data from such databases or through linked-open-data (LOD) functionalities.
The second framework category concerns language data about the associated meanings and attributions of the concept in focus, in our case 'terrorism' (Figure 2). This data of a term's linguistic determinants, suffixes, affixes, prefixes, etc., could all in principle be machine-generated.
The process of designating and producing these determinants should ideally not involve further manual interpretative choices and restrictions beyond those already built-into the corpus and the corpus tools used. Thus, the following framework is based on the premise that these functionalities are already implemented or could be implemented with a limited effort (italicized in list below). • Positioning: whether the topical keyword from the perspective of the speaker is considered appropriate or not in relation to the political actualities in focus (e.g. 'so called terrorism'); • Metaphorical: whether the keyword is used in a metaphorical rather than literal sense. Currently, the Swedish Parliament provides original data with annotations from the years 1993-2018, including title, speaker, party affiliation, reply and interested parties (riksdagen.se/sv/dokumentlagar/), and in machine-readable format on (data.riksdagen.se). Substantial sections of the Swedish parliamentary data have already been annotated and curated for analysis with language technology tools by Språkbanken Text, the text division of the research infrastructure The National Language Bank of Sweden (Nationella Språkbanken), and made accessible through the online corpus system Korp [6]. The language technological enrichment of the Korp parliamentary corpora should, in principle, to a large degree allow the data in the non-italicized labels to be extracted and used from the currently existing annotations of these corpora, or enriched with some additional annotations in the future.
To be able to explore salient meanings that the topical keyword (terrorism) represents when used by speakers in the Parliament, its situated meanings need to be annotated. Thus, the framework's third category, frame data, concerns elements more contextual than those previously described. Before discussing this category, we will consider some basic distinctions of the concepts of framing and terrorism, that have guided its design.
According to Minsky [7], a frame is essentially a cognitive phenomenon and more specifically, 'a datastructure for representing a stereotyped situation' associated with a certain set of typical activities, acts and events. When encountering a situation, an event or suchlike, one draws on elements from memory in order to understand it. Thus, memory provides a set of habitual knowledge elements that help interpret and contextualize the situation or event.
Scholars in linguistics and media and communication studies have further developed the concept of framing [8]. A key notion is that certain frames are invoked by and entangled with words and other linguistic units, and furthermore, that the frames themselves select elements and are thereby a matter of highlighting and downplaying certain cognitive aspects. An influential conceptualization states that framing means 'to select some aspects of a perceived reality and make them more salient in a communicating text, in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation for the item described.' [9]. Recent linguistics elaborations emphasize that there can be no language use without frames; any verbal act is embedded in an interpretive framework [10]. To inquire into frames, then, means asking for how frames are conceptually filled, understood as basic 'slots' within the frame.
We now demonstrate how to construct the frame for a political topic, 'terrorism', and what elements are required to render it meaningful, i.e. what slots are entailed in the 'terrorism' frame, and how they are conceptually filled in linguistic acts.
'Terrorism' is a contested concept notoriously hard to define, not least as the term is used in radically conflicting ways by different actors. Our understanding of terrorism follows Jackson's 'minimal foundationalist definition', that terrorism is 'violence or its threat intended as a symbolically communicative act in which the direct victims of the action is instrumentalized as a means to creating a psychological effect of intimidation and fear in a target audience for a political objective' [11]. Following this, terrorism includes a perceived threat of a violent act. Second, an agent -a terrorist -is required. Third, there needs to be a political target (victim or object, mål) of the terror act, and fourth, a place where the attack is spatially situated. Finally, an ideological or political motive has to be assumed, since terrorism is understood as a 'symbolically communicative act' with 'a political objective'. These five slots constitute the basic terrorism frame, although it should be stressed that they in political debates are to a significant extent filled with perceived rather than experienced meanings.
The frame data category serves as a representation of context-specific elements of the concept in focus, providing answers to the question how its 'slots' are conceptually constituted (to the extent that this can be broken down and traced in the data) (Figure 3). When it comes to the concept of terrorism, we may be able to fill the slots in its frame by answering some key questions: what actions are framed as terrorism, who are the alleged agents and their victims, what are their political motives and political as well as geographical contexts? Against the background of the ambiguity of the concept of terrorism, we have opted for structuring the answers in the following labels:
• Act: what specific kind of threatening or actual violence or tactic of terrorism is referred to;
• Agent: the individual, collective or state agent referred to as performing the terrorist act;
• Target: the target (individual, group or object) of the terrorist act;
• Place: the specific location or country where the terrorist act is situated;
• Motive: the motivations or ideology of the terrorist agent. Furthermore, we have used five complementary labels to clearly characterize the contextual conditions of the various forms of (alleged) terrorism found in the frame:
• Terrorism typology: type of terrorism distinguished between apolitical, non-violent or metaphorical terror (O) and violent political terror, with the political terror either in the form of (state) regime terrorism (R) or (non-state) rebel terrorism, which in its turn is classified according to the severity of its violence as either politically motivated less-lethal violence, threats and intimidation (H) or systematic political violence and killing (T); the latter being the primary recognized type of 'terrorism' in contemporary society; • Domestic (Y/N): whether the terror-related activities occurred in Sweden or abroad;
• Nationality: the nationality of the terrorist perpetrator (Swedish, foreign, binational, multinational or global (individual/groups); • Policy area(s): the policy area that the word 'terrorism' is connected to when used in the speech (regardless of the debate's given formal policy area); • Event/s: the conflict(s), contemporary or historical, the usage of the term relates to;
• Terrorist tropes: imaginative tropes for the discursive meaning-making about terrorism ('Soviet-backed terrorism', 'US blow-back terrorism', 'false-flag terrorism', etc.). Due to the contextual nature of the labels and data, the frame data category is dependent on 'manual' interpretation. Nevertheless, the category and its set of labels will be made available in a resource of manually annotated material for language technology training and evaluation purposes.
To demonstrate the possibilities of our framework, we will look at how the terrorism frame can be traced through inquiries into the manifestation of its elements in the parliamentary debate 1968-1970. We provide three 'keyhole views', using the data for the 5 first and the 5 last hits 1968-1970 (i.e. in total 10 out of the 71 hits of terrorism-related words). The range of these views are rather limited, but they, nevertheless, provide entry points towards a better understanding of the development of the Swedish discourse of terrorism at a particular point in time.
The concept of terrorism, despite long being used to designate various forms of political terror in a wide sense in Sweden and internationally, became the preferred label for a range of acts of political violence that we today associate it with first in the early 1970s [12,13,14]. Notably, terrorism was explicitly recognized as a domestic threat in Sweden only after the period in focus. At the same time several of the specific acts of political violence that became associated with terrorism manifested or prefigured themselves in 1968. For example, in April founding members of the Red Army Faction (RAF) fire bombed a department store in Germany, and in July the Popular Front for the Liberation of Palestine (PFLP) hijacked El Al Flight 426. Furthermore, anti-colonial wars raged in Vietnam and other Third World countries, foregrounding terror violence against civilians and fueling left-wing militancy in the West, which contributed to the transition in 1968 from an Anti-Colonial wave of terrorism to a New Left wave [15].
Consequently, our case study concerns a transitional phase in the understanding of terrorism and it thus seeks to map emerging frame elements that to some extent prefigure the general discourse in the 1970s. The data in the annotation framework indicate that terrorism was not a major political concept, but, nevertheless, sometimes used by MPs. The majority of the speakers in the Parliament that discussed the topic of political terrorism -3 out of the 7 who provided our exemplifying 10 utterances -were prominent politicians (party group leaders or party leader). Furthermore, the metadata shows that terrorism was an issue of concern primarily for parties on the right (7 politicians) and opposition (10) rather than on the left (1) or the government (0). (A cursory look shows this trend in principle stands for all the 71 hits 1968-1970).
One of the main things shown by our data, is an indication of an often mentioned but little researched semantic shift concerning the rise of the concept of terrorism in the years around 1970. Previously the word 'terror' had dominated the discourse on the types of political violence today discussed in terms of terrorism and it was primarily used in relation to violence perpetrated by state actors, i.e. what nowadays would be called state terrorism. Out of the exemplifying 5 first and 5 last hits in our material, the language data category shows an overwhelming use of the word terror (9) simplexes and compounds rather than terrorism (1). But terrorism is also starting to appear and generate additional and alternative meanings, compared to its traditional uses.
While 'terror' is used in speeches both in 1968 and 1970, 'terrorism' started being used first in 1970. Rather than being coincidental, this seems to reflect a shift in meaning that researchers have traced to the early 1970s and new configurations of militancy and guerilla warfare [14]. The noun 'terrorism' and the verb 'terrorizing' (terrorisera) are used in our material to designate sub-state terrorism and even the possibility of domestic Swedish sub-state terrorism, rather than foreign regime terrorism (a search of all speeches 1945-1970 shows 'terrorism' appearing only five times, twice in 1948 and 1960 in relation to Soviet and French 20th century state terrorism, respectively, and thrice in 1970 for sub-state terrorism in Sweden and Portugal). At the same time, the word terror merges into compounds, such as terrororganisationer (1970-11-11), related to new forms of sub-state terrorism which may be taken as a harbinger of the wider uses of terror-and terrorism-related words in the 1970s.
As stated above, there is a trend toward using the word terrorism in relation to the domestic Swedish context. This is also demonstrated by the frame data where the 'agent' and 'place' slots begin to include Swedish citizens and localities in addition to foreign elements. The majority of the data in 'Place' designate African and Asian locations, but with some notable exceptions. Actually the first occurrence in 1968 is a reference to Swedish political left-wing extremists who 'terrorize' Swedish citizens: 'It however serves no purpose whatsoever if some persons with extreme views wants to act superior and terrorize other citizens with demonstrations in non-allowed places in disorderly forms' (1968-01-18, our translation). When examining the context of this utterance, one finds that the activities that the speaker is referring to include political threats by 'wrongheaded persons', but also attacks on diplomats from other countries that may have 'dangerous consequences. These words of warning could be regarded as prescient insofar as the 1971 occupation of the Yugoslavian consulate in Gothenburg and the later killing of the Yugoslavian Ambassador in Stockholm by Croatian separatists became regarded as the first terrorist incidents in modern day Sweden, adding Stockholm and Sweden to the 'Place' slot of the terrorism frame.
In the parliamentary debates we can see from the frame category that the word 'terror' is prominently used in connection to violent anti-colonial struggles in Africa and Asia, the Vietnam war in particular. This can be illustrated by an utterance by a right party (M) MP regarding the violent struggle of African subjects to the Portuguese empire and the Portuguese authorities' explanations thereof: 'Furthermore they emphasize that they experience that the terrorism that these [Portuguese Asian and African] areas are exposed to, as being organized by outside countries, among others outside of Africa. They themselves name and point to that there is Communist interests, that the Soviet Union is an arms supplier and that in the Portuguese Guinea Cuban military is appearing in the terror organizations' (1970-11-11, our translation).
A close reading of the debate's further exchange reveals the existence of an ideological tension in the semantics of terrorism. Contrary, an MP to the left (S) uses the more positive sounding 'freedom struggle' (frihetskamp) to describe the militancy in the Portuguese colonial areas in Africa. The polarization between the words terrorism and freedom struggle, or terrorist and freedom fighter, became central to the discursive conflict over the semantics of terrorism and something we expect to explore further as it diversifies in the 1970s into other colonial and political contexts.
This pejorative use of the terrorism word and the associations made to Communism among MPs to the right can be implicitly found already in the above-quoted speech from 1968, in which an MP (C) warns about Swedish extremists who want to 'terrorize other citizens with demonstrations'. A similar criticism against the left can be found again in 1970 in a speech by a liberal (FP) MP: 'When the newold Left wants to defend violent methods it cannot be mixed up with equivocation for terrorism' (1970-10-29, our translation). A close reading of the quote paragraph however reveals that the statement is meant to be sarcastic and intending to say that the New Left actually advocates terrorism. In fact, it draws on and borrows from a larger debate in the press on the New Left's views on political violence and 'the role of violence as a means of struggle [kampmedel] in democratic societies'. Thus, the media debate seems to have fed into the debate in the Parliament during what appears as a transitional phase for the parliamentary frame of terrorism.
In general, these results point to the analytical gains of taking a context-sensitive approach that extends further than the sentence level and data generated by the semi-automatic extraction. By also performing close readings of the textual context of the debates (and beyond), we are able to more deeply understand the discursive development of the parliamentary debate about terrorism.
This paper has developed a tentative framework for how to semi-automatically annotate salient topics in the Swedish parliamentary debate. We have presented the framework's three main categories: metadata, language data and frame data, that each describe different dimensions of the concept in focus. The metadata category annotates basic parameters relevant to the documentary context of the topical keyword; the when, where, why, who of each utterance. The language data category concerns the concept's linguistic attributions and determinants, which, similar to the framework's first data category, to a significant extent could be automatically generated. The frame data category, however, is distinctly contextual and thus requires manual interpretation. This category concerns more specific contextual multi-dimensional meanings associated with the concept in focus and its underlying idea is presented through the linguistic framing concept. Furthermore, we have illustrated the framework's possibilities through concrete examples from debates touching on terrorism 1968-1970.
As our annotation framework is decidedly tentative, it should be expected that the categories could be further developed and that their labels likely will be further modified in tandem with the empirical analysis in the ongoing SweTerror project. It has, as of yet, not been tested and evaluated.
We would however suggest that this tentative annotation framework might be more widely applicable for representing parliamentary discourses on other salient topics besides terrorism. It could also be developed to include other types of parliamentary communication, e.g. government bills and motions by MPs. The metadata category provides a generic framework for parliamentary data in general, as its set of labels could be regarded as basic when it comes to the information they hold for analytical extractions. At the same time, this category could be extended to include additional labels and data for other research purposes. The same applies for the language data category. The frame category, however, is more complex in context of general applicability. While the concept of 'framing' holds general promise as a reference point for the structuring of contextually bound data, the labels included in this category are partly dependent on the topic under investigation. The labels we use should, nevertheless, to some extent be applicable to other frames as most political concepts consist of a spectrum of various 'types', have varying degrees of internationalism', are used within multiple policy areas, are connected to specific events and larger tropes, etc. Thus, if modified and revised, the labels in this category may be used to address other research purposes.
SweTerror is part of the Digitisation and Accessibility to Cultural Heritage Collections (DIGARV) program funded by the Swedish Research Council, Riksbankens Jubileumsfond and the Royal Swedish Academy of Letters, History and Antiquities (digarv.se/en/). This work is also supported by the national e-infrastructure The National Language Bank and Swe-Clarin funded partly by the Swedish Research Council (under contract no. 2017-00626).