Global Environmental Change 69 (2021) 102308 Contents lists available at ScienceDirect Global Environmental Change journal homepage: www.elsevier.com/locate/gloenvcha Framing the frontier – Tracing issues related to soybean expansion in transnational public spheres Finn Mempel a, *, Esteve Corbera a, b a Institute of Environmental Science and Technology, Universitat Autònoma de Barcelona, Campus UAB, 08193 Bellaterra, Barcelona, Spain b Institució Catalana de Recerca i Estudis Avançats (ICREA), Pg. Lluís Companys 23, Barcelona, Spain A R T I C L E I N F O A B S T R A C T Keywords: Rapid soybean expansion in South America has been linked to numerous socio-environmental problems, Soy including deforestation in sensitive biomes. As a major importing region of soybeans, wider public awareness has Environmental governance also put pressure on the European Union. Different governance initiatives involving various groups of stake­ Environmental communication holders have sought to address these issues. However, what is identified as a relevant problem, as a region of Public sphere theory Text mining interest or which actors are mentioned in this context are all matters of claims-making processes between Topic model different groups and mediated through various channels of communication. This study uses a text-mining approach to trace the construction of socio-ecological problems related to soybean expansion and the actors and regions linked with these issues in public discourse. The focus lies on print media from the European Union, but several additional sources are included to investigate the similarities and differences between various communication channels and regions. These include newspaper articles from producing countries and interna­ tional news agencies, scientific abstracts, corporate statements, and reports from advocacy groups gathered from the mid-1990s to 2020. The results show that European mass media have shifted their focus from consumer labeling, health, and concerns over genetically modified organisms towards more distant or abstract phenomena, such as deforestation and climate change. This has been accompanied with a broader view on different stake­ holders, but also with a strong regional focus on the Amazon biome. There has also been much less attention on direct concerns for communities in producing regions, such as land conflicts or disputes over intellectual property rights. We conclude that while European public spheres appear to become more receptive to issues related to impacts in sourcing regions, there remains a narrow focus on specific problems and regions, which reflects a fundamental asymmetry in different stakeholders’ ability to shape transnational deliberations and resulting governance processes. 1. Introduction structural adjustment policies, deregulation and a general embrace of biotechnology and large agribusiness (Leguizamón, 2020; Neiman and Since the early 20th century, soybeans have evolved from a regional Blanco, 2020). While applauded as an economic success story by some, food crop into one of the world’s most important agricultural com­ this expansion has also been criticized for its negative socio- modities (Du Bois, 2018). Within the past 50 years, soybean production environmental impacts, as the soybean frontier has expanded into has increased almost tenfold and soy now ranks among the four leading highly sensitive and biodiverse biomes. Studies have pointed to patterns crops worldwide in terms of overall area harvested (FAO, 2018). The of landholding concentration, rural displacement, deforestation, soil crop’s high protein and fat contents make it valuable for a number of degradation, food insecurity and health hazards (Goldsmith, 2017; different end-uses, but the main driver has been an increasing global Leguizamón, 2014; McKay and Colque, 2016; Pengue, 2009). demand for livestock feed, particularly in Europe and China (Oliveira As media outlets and NGOs from the Global North have increasingly and Schneider, 2016). publicized these issues and called on governments, corporations, and Within the last decades, the expansion of soy production has mainly consumers to act, political pressure has mounted. New governance taken place in South America, where it thrived within a context of mechanisms have been put in place, including many examples of what * Corresponding author at: Institute of Environmental Science and Technology (ICTA), Universitat Autònoma de Barcelona, Campus UAB, 08193 Bellaterra, Spain. E-mail address: finn.mempel@uab.cat (F. Mempel). https://doi.org/10.1016/j.gloenvcha.2021.102308 Received 6 November 2020; Received in revised form 13 May 2021; Accepted 25 May 2021 Available online 3 June 2021 0959-3780/© 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 has been termed private food law (van der Meulen, 2011), such as cer­ issues affecting their lives. tification schemes, private standards or multi-stakeholder round-tables. Beyond its application as an analytical framework, public sphere These often single out individual agricultural commodities and associ­ theory also has a strong normative foundation in relation to the func­ ated social or environmental problems. Further, several international tioning of deliberative democracy. Evolving public spheres can be agreements have targeted the issues associated with soybeans among evaluated according to their inclusiveness to all members and minority other forest-risk commodities (FRCs) and the European Union (EU) is groups within a given polity (legitimacy) and according to their effec­ currently assessing policy options to address tropical deforestation tiveness at translating public opinion into political debate and binding linked to its imports (Bager et al., 2020; European Parliament, 2020). legislation (efficacy) (Fraser and Nash, 2014). These new governance processes are neither entirely embedded in While traditionally applied at the level of individual nation-states, political entities, nor truly global (Lenschow et al., 2016), they are flow- Fraser (2009) has pointed out the need to use these criteria for delib­ based rather than territorial (Sikor et al., 2013) and often involve erative processes on a transnational level, since economic globalization different institutional layers and actors. While the effectiveness of these has subjected people across individual nation-states to the same gover­ measures has often been addressed by research (e.g., Garrett et al., 2016; nance institutions and rendered them susceptible to the outcomes of the Lambin et al., 2018), the role of communication and discursive ex­ same value chains and decisions taken by the same transnationally changes in constructing and framing the problems provoking these operating actors. measures has received much less attention (Persson and Mertz, 2019). In this article, we adopt a conceptualization of issue-oriented, net­ Elgert (2012) has demonstrated how certain actors and coalitions worked, transnationalizing public spheres, which are porous to have framed issues related to soybean expansion rather narrowly to communicative flows across polities (Fraser and Nash, 2014). As in the promote a particular way of addressing them with specific governance later revision of Habermas’ original theory (Habermas, 1996), we instruments, such as certification schemes. Our contribution takes a consider mass media as crucial to the functioning of modern public broader perspective to trace the evolution of how different socio- spheres. This central role of mass media in deciding on what are being ecological problems in relation to the surge of global soybean produc­ considered key issues in public spheres is also assumed in media effect tion and trade have circulated in public discourse since the late 1990s. models such as agenda setting or priming (Scheufele and Tewksbury, To do so, we mobilize public sphere theory as a conceptual framework 2007). and use a text mining approach to analyze a large text corpus comprised Following Benson (2009) in his mapping of public spheres using of different sources. Our focus remains on the EU journalistic field as the Bourdieu’s field theory, we understand the journalistic field to be at the center of public discourse within one of the primary soy importing center of a given public sphere, interacting with and mediating between regions. other fields, including the academic, the political, the economic and the In this article we ask: What differences exist between these fields in advocacy fields. It is important to note that the assumed centrality of terms of the prevalence of certain topics? Which topics prevail at mass media should not be understood as a normative assessment of their different time periods? How are these topics linked to each other and to function in public discourse. In fact, many traditions in mass commu­ the mentions of certain actors and geographic regions? By addressing nication theory, such as mass society theory or the critical cultural trend these questions, we hope to provide insight on how the concerns of have understood mass media as promoters of hegemonic elitist world­ different groups in relation to soybean expansion have circulated views (Baran and Davis, 2011). through public discourse, particularly in importing regions such as the EU, and then shaped the ways in which transnational governance ini­ 2.2. Text mining and topic modeling tiatives, such as certification schemes, zero-deforestation commitments, or multi-stakeholder roundtables, have targeted particular problems in Empirical studies in the field of environmental communication have sourcing regions. Further, we hope to demonstrate the utility of applying traditionally relied on the same approaches to text analysis (or other public sphere theory and text mining approaches when studying pro­ forms of communication) as the social sciences more broadly. These cesses of land use change and environmental governance. differ substantially in their mode of analysis (quantitative, qualitative, In the following sections we locate our research within the field of or mixed methods) and their level of analysis (e.g., textual, contextual, environmental communication and introduce public sphere theory as or sociological), but generally all involve the acquisition, selection, our framework of analysis. We provide an overview of text mining and careful reading and manual coding of texts, sometimes using several topic modeling as the toolbox applied in our methods. We then continue independent coders. to provide a detailed overview of our methods and present our findings. The widespread use of social media, the ongoing digitalization of archival material, the creation of specialized online databases for news, 2. Environmental communication, public spheres, and text academic articles and press wires have provided an abundance of mining available text material about virtually any topic and from a variety of sources. While this provides unprecedented access to texts and other 2.1. Environmental communication and public sphere theory forms of communication, it also represents a major challenge for tradi­ tional approaches to text analysis. Manual coding of large volumes of Studies on the discursive processes shaping public perception of texts is only feasible by applying selective filters and sampling methods environmental problems can be found in the field of environmental or by hiring many additional coders, resulting in high project costs communication (Pezzullo and Cox, 2018), which emerged in the 1980s (Grimmer and Stewart, 2013). However, the increasing processing ca­ (Cox and Depoe, 2015). The field has long studied environmental pacity available to researchers even on personal computers has enabled problems as not merely materially produced, but rather as socially and the use of novel methods and applications leveraging computer algo­ discursively constructed through processes of claims-making and rithms, often referred to as text mining (Ignatow and Mihalcea, 2017). contestation, mediated through different public arenas or forums, such These had been developed in the field of computer science and are now as mass media (Hansen, 2015b). successively being added to the toolbox of the social sciences, alongside A useful theoretical underpinning when studying communicative traditional methods of text analysis. processes and their role in shaping governance mechanisms regulating More than just a way of dealing with the challenges of big data, the global commodities is public sphere theory (Habermas, 1989). The use of algorithms for language processing and supervised or unsuper­ concept of a public sphere is intimately tied to the theory of deliberative vised text classification also provides a means to counter the regularly democracy and describes a net of communicative processes, which spans raised issue of human bias in traditional text analysis. However, math­ throughout society and through which its members rationally debate the ematical models for language and text used in these methods are by 2 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 nature simplified representations. Text mining should therefore not be (mixed membership model) and every topic as a weighted mix of words. seen as replacing but rather as complementing traditional content It is common practice to perform pre-processing and filtering steps analysis (Grimmer and Stewart, 2013). Further, even unsupervised on a text corpus prior to the application of a topic model. Tokenization methods still are susceptible to bias, both, though choices in the selec­ refers to the process of breaking down text documents into smaller units tion of parameters or interpretation of output, and the reductionist (tokens), which become the basis for further analysis. In the case of topic approach to language ingrained in their algorithms. modeling, tokens are typically words and compound multi-word ex­ Originating with the groundbreaking work of Blei et al. (2003), topic pressions. Common processing steps include trimming, lemmatization, models are a rather popular tool among text mining methods. They are a named entity recognition (NER) and co-reference resolution. These are statistical framework developed in computer science research and used further explained in the methods section below. to identify the underlying (“latent”) topics in a text corpus (Wesslen, Numerous studies have applied topic modeling and other text mining 2018). Topic models are unsupervised machine learning algorithms, approaches to analyze communicative processes linked to environ­ which do not depend on training datasets, but provide classifications mental issues, most notably climate change. Topic modeling has been according to patterns identified within the data itself. Topic models applied to show how media attention to climate change has generally typically use a document-term matrix (DTM) as input, which lists rela­ increased but at significantly different rates across countries (Schmidt tive term frequencies for all documents. This representation of docu­ et al., 2013). It has helped to demonstrate how specific national fram­ ments ignores the original order of words and is therefore commonly ings of climate change reflect different countries’ specific local contexts, referred to as a bag-of-words (BoW) model. despite it being a global phenomenon (Vu et al., 2019). Further, Bous­ Topic models treat individual texts as being generated by drawing salis and Coan (2016) show the persistence of climate change denial the distributions of topics in each document as well as the distribution of among a group of conservative think tanks and Bohr (2020) demon­ words in each topic from an underlying probability distribution (Gün­ strates the importance of outlet bias as a contextual factor in U.S. media ther and Quandt, 2016). By reverse engineering the generation of a coverage on climate change. given text corpus according to this model, topic models provide the user with a representation of every document as a weighted mix of topics Fig. 1. Flow diagram illustrating major components of data acquisition, processing and analysis. 3 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 3. Methods taken from Schumacher et al. (2016). We downloaded journal abstracts from scopus.com. Search strings and filtering criteria are listed in the The entire process of data acquisition, processing, topic-modeling, supplementary material. For the year 2020, data collection stopped on and analysis is illustrated as a flow diagram in Fig. 1 and described in May 1st for all sources included. detail below. In total, the text corpus consists of 32,540 documents, which are distributed according to field and region as illustrated in Fig. 2. The 3.1. Data acquisition – Building the corpus journalistic field represents the center of our analysis and is represented with the largest number of documents (21,831). The other fields mainly Following our conceptualization of networked public spheres, we serve for comparative purposes and contribute fewer documents. Within generated a text corpus by integrating several different types of docu­ the journalistic field, print media from the EU constitute the largest ments, grouped into a nested categorization as illustrated in Table 1. All fraction (6232 Documents). More details are provided in the supple­ sources were selected according to availability during our study period mentary material. (1997–2020). For the journalistic field we aimed to select news media of record with relatively high circulations and preferably national editions. 3.2. Data processing Corporate actors were selected to represent different stakeholders along the soy value chain. For the advocacy field, we aimed to include both, Before computing the topic model, several processing steps were large international NGOs, which have directly collaborated with performed. The first step was to translate all non-English documents roundtables and other business initiatives, and movements or organi­ from their source language into English. We used the ModernMT zations working closer with local communities and peasants. machine-translation API (ModernMT, 2020), which provides a context- News articles were sourced from Factiva (Dow Jones and Company, aware translation algorithm. While machine-translation is never free of 2020) and Lexis Nexis (Lexis Nexis, 2020). We collected press releases errors, here we mostly rely on the translation of important individual via Web Scraping from the websites of several corporate actors, NGOs, key terms. Grammatical structure is irrelevant for the topic modeling social movements, and the EU Commission. EU political speeches were procedure, which does not consider the order of words, but only their Table 1 List of sources, nested according to field and region. Field Region Country Language Name Type Journalistic Field European Union U.K. English The Times News articles The Guardian The Independent Germany German Sueddeutsche Zeitung Der Spiegel taz France French Le Monde Le Figaro Libération Spain Spanish El Mundo ABC La Vanguardia Italy Italian Corriere della Sera La Stampa La Repubblica Latin America Brazil Portuguese O Globo O Estado de S.P. Folha de S.P. Argentina Spanish La Nación Clarín La voz del Interior North America USA English The Wall Street Journal The New York Times The Washington Post Transnational USA English Associated Press U.K. Reuters France Agence France Presse Germany DPA Spain EFE U.K. Press Association China Xinhua Agency Business Field North America USA English Monsanto Press Releases ADM European Union Belgium FEFAC Netherlands ForFarmer Group Latin America Brazil Portuguese Aprosoja Brasil Advocacy Field Transnational Switzerland English World Wildlife Fund (WWF) Press Releases USA World Resource Institute (WRI) Netherlands Transnational Institute (TNI) Spain GRAIN Simbabwe Via Campesina Netherlands Friends of the Earth (FoE) Political Field European Union English EU Speech Corpus Political Speeches Belgium Bulletin Quotdien Europe Print media articles Belgium EU Commission Press Releases Academic Field Transnational – English – Journal article abstracts 4 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 2. Distribution of documents according to region and field. frequency. However, this machine-translation approach still introduces always be evaluated manually. There are also various metrics designed an unquantifiable degree of error and may be highly problematic for to automatically find an appropriate range for K. After running the STM more nuanced text-mining approaches (e.g., sentiment analysis or with several different setups and comparing the output and a set of test metaphor detection). statistics (see supplementary material), we settled on a model with 80 Subsequently, we applied several filters to the corpus to remove topics, which provided sufficient detail while not overclustering. irrelevant documents and duplicates (see supplementary material). We Researchers have pointed to the need to validate any use of auto­ then used the natural language processing (NLP) library SpaCy (Hon­ mated text analysis (Grimmer and Stewart, 2013). We establish semantic nibal and Johnson, 2015) and the neuralcoref package (Clark and validity by labeling topics, taking into consideration highly associated Manning, 2016) to resolve all documents for co-references, replacing all words and documents dominated by each topic and by mapping topics mentions referring to the same real-world entity with the entity’s name. via hierarchical clustering. Following (Quinn et al., 2010), we introduce This is helpful, since an entity may be referred to many times in a text, a measure of construct validity by comparing the evolution of preva­ but only named once. lence of selected topics over time with external events, which would be We also identified and extracted real-world entities using SpaCy’s expected to lead to public debate of the given topics. Further, following Named Entity Recognition (NER) function. These were removed and the approach introduced by (Chang et al., 2009), we introduce a su­ collected separately for further analysis. The rationale behind this is to pervised element, comparing topic model outputs to the judgement of generate latent topics independent of any mentions of organizations or human coders (see supplementary material). locations. This way, the co-occurrence of topics and these entities can be analyzed separately. Finally, only nouns, verbs, adjectives, and adverbs were included as 3.4. Data analysis input tokens for the topic model. The tokens were lemmatized (trans­ formed to their noninflected dictionary form) with the lemmatizer Presented with the model output, we labeled topics according to a provided by SpaCy’s English language model. manual inspection of the terms most prevalent for each and the 20 documents with the highest relative share of a given topic. Subse­ quently, we selected the most relevant topics for our analysis and 3.3. Topic modeling merged similar topics by summing their prevalence. We then performed a network analysis based on the co-occurrence In our study, we used the Structural Topic Model (STM), based on the between the individual topics and between the extracted entities and work by Roberts et al. (2014). STM is a probabilistic model, in which topics. For this purpose, we computed cosine similarities between the both the terms associated with topics and topic proportions for each prevalence vectors of all issues and entities and used the similarity document are approximated as latent variables. We implemented the measures as edge weights for the network. Additionally, measures of STM model using the open-source stm package available for the statis­ closeness centrality for all topics are found in the supplementary tical computing environment R (Roberts et al., 2019). When converting material. our text corpus into a DTM to feed into the topic model, we performed a Lastly, for one key topic (“Deforestation”) we selected all documents, collocation analysis to represent multi-word expressions (see supple­ in which this topic was the most prevalent one. For each year in our mentary material). study period we then calculated a measure of “keyness” (Bondi and An advantage that STM provides over other topic modeling ap­ Scott, 2010), i.e. comparing a target group (documents from the given proaches is the possibility to include document metadata as covariates year) to a reference group (documents from all prior years). This for the topic model. Here, we included field, region, and year of publi­ approach provides key terms for each year, which are more frequently cation as covariates. When computing the topic model, one important used than in previous years. We performed this analysis separately for consideration is the selection of an appropriate number of topics (the documents from the EU journalistic field, Brazilian journalistic field, and model parameter K). The selection of this parameter will depend on the transnational news agencies. We then identified one year, in which this underlying research aim and the merit of the resulting topics should topic appears to be dominated by similar key terms across these three 5 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 sources. We further identified months which showed peaks in preva­ Table 2 lence of the given topic for each source. To analyze the specific content Selected topics and topic groups. of documents for the selected year and for those months, we calculated Topic Description Number Prevalence in co-occurrence networks for each, including both, tokens and named of topics Corpus (%) entities. Diets & Health These topics deal with 4 4.34 different relationships 4. Results between food items’ nutrients and health, such as concerns over different types 4.1. Topic model output and validation of fat. Pest Management These topics deal with the 3 3.21 The entire output from our topic model, including overall topic management of pests prevalence and top contributing terms for all 80 topics are provided in affecting soybeans, such as the aphid Aphis glycines the supplementary material. Labeling of topics provided initial support Matsumura, pod borers or for semantic validity, as most topics were easily categorized. The results whiteflies. from our manual labeling and grouping of the latent topics identified by Deforestation This issue is about the loss of 2 2.97 the STM are illustrated in Table 2, including short descriptions of the forest area in different selected topics or topic groups, the number of constituent topics, and biomes. Trade Disputes This topic deals with 1 2.91 topic prevalence over the entire corpus. Of the initial 80 topics, 29 were protectionist measures, such selected and grouped into 20 topics, reflecting relevant issues. The as tariffs, imposed between remaining 51 topics were excluded from the analysis. The selected topics different countries and the represent about 40% of the overall topic prevalence across the entire consequences for trade patterns (e.g. in soybeans), corpus. different economic sectors Hierarchical clustering of word weight vectors resulted in mean­ and international relations. ingful clusters around certain topic categories, even though in a few Biofuels These topics deals with the 2 2.7 cases related topics form separate clusters, likely due to differences in use of soybeans among other vocabulary between fields (see supplementary material). The corre­ plants as feedstock for biofuels and the implications spondence found between the judgement of human coders and our topic for direct and indirect model is generally encouraging. Mean model precision (comparing emissions from combustion assignment of words to topics) was found to be 88.75 percent and the engines. mean topic log odds (comparing assignment of topics to documents) was Harvest & Weather This topic deals with 1 2.61 weather patterns, such as calculated at -0.52, both at high significance levels (P < 0.001). When droughts and floods and how considering only selected topics, model precision increased to 93.1 they affect yields in various percent (see supplementary material). With respect to construct validity, agricultural areas producing Section 4.6. establishes solid correspondence between selected topics soybeans. and key real-life events. Land Use Change & This topic deals with land 1 2.23 Environmental use change and associated Impact environmental impact. 4.2. Dominant topics vary across fields Soil Management This topic deals with the 1 2.19 management of soil The upper graph in Fig. 3 illustrates topic prevalence for all fields structural properties and nutrients through included in our analysis. Topic prevalence varies quite significantly agricultural inputs or between fields. In fact, the most dominant topic is different for each cropping practices and the field. The academic abstracts included show relatively high prevalence respective impact on values for the topics Pest Management (10.68%), Soil Management soybean yields. (7.84%), Land Use Change (7.05%) and Biofuels (7.04%). The docu­ Meat & Animal These topics deal with 2 2.09 Feed concerns over different ments included from the advocacy field are dominated by issues around impacts of meat the Global Food System (19.40%) and also show the largest relative consumption and the search shares of prevalence for Deforestation (6.06%), issues around Seeds & for different protein meals Patents (6.47%), Land Conflicts (5.65%) and Climate Change & GG (e.g. oilseed cakes or fish- meal) for animal feed and Emissions (3.56%). The Business field shows low prevalence for most of the respective trade-offs our selected topics, apart from issues dealing with Harvests & Weather when substituting between (8.90%), as well as the concerns of Producers & Landowners (3.26%). them. Further, there are The journalistic field shows a more even distribution of prevalence accounts of consumers’ and between the selected topics compared to the other fields. It presents the businesses’ attempts to substitute meat products highest relative prevalence for Diets & Health (4.96%) and compared to with plant based the other fields, it also shows relatively high levels of prevalence for alternatives. Trade Disputes (4.25%) and Consumers & Food Labeling (1.77%). The Climate Change & These topics deal with 2 1.85 political field (EU) is dominated by concerns about GMOs (20.91%) and GG Emissions greenhouse gas emissions and the threat of climate to a lesser extend Trade Disputes (4.35%) and Diets & Health (3.07%). change. Economic Crisis This topic deals with 1 1.69 4.3. Topic prevalence varies across print media economic and financial crises, inflation, debt and The graph at the bottom of Fig. 3 provides the same information, unemployment. Producers & This topic deals with the 1 1.46 restricted to the journalistic field and grouped by regions from which Landowners economic and social print media were collected. The Argentinian print media has a rather (continued on next page) unique profile of topic prevalence, being dominated mostly by topics related to Harvests & Weather (15.42%), Producers & Landowners 6 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Table 2 (continued ) Diets & Health (9.30%) and Climate Change (2.84%) among all regions. Topic Description Number Prevalence in For international news agencies Trade Disputes (8.68%) dominates and of topics Corpus (%) the US print media have the highest prevalence for Consumers & Food environment for farmers and Labeling (1.45%) among all regions. landowners. Consumers & Food This topic deals with 1 1.4 4.4. Zooming into the EU print media Labeling consumers’ concern concerning various food Fig. 4 illustrates the evolution of the co-occurrence network of our items and their production, as well as the debate on food selected topics in the EU journalistic Field through five distinct time labeling. intervals between 1997 and 2020, as well as for the entire study period. GMOs This topic deals with 1 1.29 Further, Fig. 5 shows the evolution of prevalence for selected topics by biotechnology applications year. One can observe that EU print media articles in the late 1990s, in agriculture and the various concerns about which mentioned soybeans were dominated by topics dealing with Diets genetically modified & Health, GMOs, Consumers & Food labeling and to a lesser degree, organisms, such as roundup- Meat & Animal Feed. ready soybeans. These topics occur in relative isolation and there are only a few Water Resource This topic deals with water 1 1.11 strong co-occurrence values between these and other topics (e.g., be­ Management resources, hydrological alterations and issues related tween Pesticide & Health and Seeds & Patents). Over time, these topics to water availability, show a declining trend; though Diets & Health remains among the most distribution and quality. prevalent topics, GMOs is among the least prevalent in the last time Global Food System This topic deals with the 1 1.06 interval. Since the early and mid-2000 s, Deforestation and Climate challenges of global agriculture and food Change & GG Emission gain prevalence, with their highest shown values systems, particularily in for the period between 2007 and 2011 but remain relevant until 2020. feeding the world’s Between 2017 and 2020, Trade Disputes becomes the most prevalent population, overcoming topic. hunger and malnutrition and preventing food price hikes. Land Conflicts This topic deals with 1 1.03 4.5. A focus on the Amazon and more diverse stakeholders conflicts over land as a result of agribusiness expansion. Regarding the extracted named entities in the EU journalistic field, Main themes include the Fig. 6 illustrates the evolution of mentioned entities, filtered to include concern over livelihoods and rights of peasants and four of the biomes, in which soybean expansion has taken place within indigenous populations and the last decades. The graph is organized as a bipartide network to show the struggles of social how the mentions of biomes are linked with topics (using cosine simi­ movements in defending larities between the prevalence in all documents). The Amazon biome is these rights. by far the most mentioned throughout the study period and also shows Pesticides & Health This topic deals with the use 1 0.95 of pesticides and associated the strongest associations with most topics, particularly Deforestation. health concerns, such as Other associations become progressively relevant since the early 2000s, cancer and infertility. namely Land Conflicts, Land Use Change & Impacts and Climate Change. Seeds & Patents This topic deals with 1 0.78 Mentions of the Cerrado and Atlantic forest occur since the early farmers’ use of patented GMO seed and the legal 2000s and in the case of the Gran Chaco since 2007. These biomes are disputes over intellectual mainly associated with Deforestation, Wildlife & Biodiversity, Land Use property rights when saving Change and Impacts, Land Conflicts (in the case of the Gran Chaco and and reproducing seeds. Cerrado) and Water Resource Management (in the case of the Atlantic Wildlife & This topic deals with threats 1 0.68 Forest). However, both the number of mentions and the degree of as­ Biodiversity to wildlife, endangered species, biodiversity loss and sociation with our topics is less significant than for the Amazon biome the struggle of throughout our study period. conservationists. For other fields, the picture changes. For instance, in the advocacy Other (excluded) A range of topics excluded 51 61.43 field (graph provided in the supplementary material), the Amazon from the analysis. These are mainly topics dealing with biome is only as dominant as in the EU journalistic field for the time trade statistics, commodity interval 2007–2011. At other times (2017–2020), mentions of the Cer­ market updates, but also rado are comparably prevalent. Further, for Land Conflicts the associa­ general “nonsense” topics tion is strongest with the Cerrado biome. In the EU political field only reflecting the use of specific the Amazon biome is referred to at all. vocabulary not related to any particular issue. Fig. 7 illustrates the evolution of entities representing organizations or institutions mentioned in the EU journalistic field and their associa­ tions with our 20 topics as a bipartide network graph. For each time (9.51%) and Economic Crisis (4.50%). The selected newspapers from interval, the 20 most prevalent entities are selected. In order to enhance Argentina also show a relatively high prevalence of issues related to readability, only edges within the upper 12.5% quantile in terms of their Seeds & Patents (1.83%), compared to the other regions. weights were plotted. We observe that for the entire study period, the EU The included Brazilian print media sources show a high relative is the most prevalent entity mentioned in this category, followed by prevalence of issues relating to Deforestation (9.82%), Economic Crisis several companies, NGOs, government institutions and intergovern­ (4.62%) and Land Conflicts (3.47%) compared to the other regions. For mental organizations. When looking at the evolution over time, there are EU print media, international news agencies and the US print media, three clearly dominating entities in the late 1990s: the EU, the biotech topic prevalence shows more comparable patterns: relatively large company Monsanto and the NGO Greenpeace. All these show associa­ prevalence values for topics relating to Trade Disputes and Diets & tions with GMOs and Consumers & Food Labeling, two of the dominant Health. EU print media show the highest relative prevalence shares for topics of that time interval. 7 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 3. Topic prevalence by field (upper graph) and by region for journalistic field (lower graph). We further want to point to the following observations: as the focus 4.6. Topic prevalence corresponds to key real-life events on GMOs and Consumers shifts towards Deforestation and Climate Change, the prevalence of entities from the biotech industry (e.g., When cross-checking the evolution of topic prevalence in EU print Monsanto and Novartis) fades and gives way to grain traders (e.g., media with real-life events, we found strong links between peaks of topic Cargill), food processing industry (e.g., Unilever), retailers (e.g., Tesco) prevalence and certain key events or developments (Fig. 8). and food chains (e.g., McDonald’s). Further, mentions of NGOs are also Starting with the topic Meat & Animal Feed, the strong increase in mainly associated with Deforestation and Climate Change since the mid- prevalence in the years 2000 and 2001 occurs at a time when the EU 2000 s. In the most recent time interval (2017–2020), the association Commission debated and subsequently decided on a full ban on meat between the topic Deforestation and mentions of the EU is included and bone meal (MBM). Feeding MBM to animals had been accompanied within the 12.5% edge weight quantile for the first time, simultaneously by ethical concerns and linked to the outbreaks of Bovine spongiform with the appearance of the Mercosur trade bloc among the 20 top encephalopathy (BSE) since the late 1980s. The ban was implemented in entities. 2001 and led to an increase in demand for other protein sources in an­ imal feed, such as soybean cake. This interpretation is further confirmed by the strong links between the topic and the entities EU and the Eu­ ropean Commission for 1997–2001 (Fig. 7). 8 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 4. Evolution of topic co-occurrence in EU print media. Circle size represents topic prevalence and line width indicates edge weight. Only edges with weights above the mean were included here. Deforestation shows marked peaks in prevalence between 2005 and The topic Global Food System peaks in 2008, at the height of the 2009, as well as again in 2019. The first peak coincides with broader 2007/2008 food crisis, when food prices skyrocketed worldwide public concern about high deforestation rates in the Amazon region, (consider also the strong links to the entities UN and Worldbank). It leading to the publication of the Greenpeace report “Eating up the further shows strong links with the topic Biofuels, which also peaks in Amazon” (2006) and the signing of the Amazon Soy Moratorium (2006) that period. Biofuels were often seen as partly responsible for food price by different stakeholders (see also strong links to Greenpeace in Fig. 7). hikes at the time. In 2019, broad media coverage followed the fires in the Amazon Lastly, the topic concerning Trade Disputes shows a dramatic peak in rainforest. 2018, the year in which the Trump administration unilaterally imposed 9 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 5. Evolution of prevalence for selected topics in EU print media. tariffs on different products from China. This was followed by similar occurrence networks for the selected months and the entire year 2006 tariffs imposed by China on US products, including soybeans (see also for all three sources. While the Greenpeace report and the moratorium the strong links to entities such as the White House, the WTO and dominate the coverage in EU print media and news agencies for the companies such as Huawei). entire year, in Brazilian print media this is only the case for the months of July and (to some degree) April. Another event that appears to play a major role in the coverage of news agencies that year is the blocking of 4.7. A closer look at media attention toward deforestation Cargill’s port facility in Santarem by a Greenpeace ship. When examining monthly prevalence values for the topic Defores­ 5. Discussion tation for media outlets from the EU and Brazil, as well as transnational news agencies (see supplementary material), we observe that prior to Our findings provide indications regarding the evolving debates on 2003, prevalence levels are consistently higher for Brazilian news media soybean expansion and associated impacts within interlinked trans­ and peaks in prevalence occur quite independently between the three nationalizing public spheres, as well as the legitimacy or lack thereof sources. While prevalence remains highest for Brazilian outlets during these debates confer to governance interventions at different scales. most of the study period, the levels converge between the different The EU print media included in our analysis shows a clear shift from sources and peaks co-occur more frequently. topics directly related to consumer concerns over the safety of GMOs, The outputs from our keyness analysis show that for most years in food labeling and other issues of direct relevance to EU citizens towards our study period frequently used terms compared to prior years differ more distant or abstract phenomena, such as deforestation in producing between outlets. However, there are several years with strikingly similar countries and climate change. This is accompanied by mentions of a features between sources. In 2003, both Brazilian and EU media outlets broader set of actors in the production network. appear to report on then newly elected President Lula and Minister of While these topics show a general upward trend, they still seem to be the Environment, Marina Silva. Both sources show peaks for the month quite event-focused, revealing distinct peaks during times of contro­ of June, when Silva convened an important meeting on deforestation versial or disaster-like events. This shift in focus from biosafety of GMOs between scientists and civil society. The Brazilian outlets show another and consumer health towards deforestation and climate change can also peak for the month of July, when Lula issued a presidential decree, be observed in the documents collected for the advocacy field. However, laying out policy instruments to combat deforestation. Another common this field also emphasizes various issues that do not attain the same feature is the frequent mentioning of President Jair Bolsonaro in the prominence in European print media, such as those dealing with land years 2018 and 2019. conflicts or seeds and intellectual property rights. Further, problems The closest resemblance, however, can be observed for the year concerning the general structure of the global food system are dominant 2006, in which key terms include the trader Cargill, the NGO Green­ and central (highly connected to other topics) in the advocacy field, peace and the fast-food chain Mc Donald’s in all sources. Peaks in while they never play such a dominant role in the EU journalistic field. prevalence are identifiable for the months April, when the Greenpeace There is a clear regional focus on the Amazon biome, particularly report “Eating up the Amazon” was released, and for July, when the concerning topics related to deforestation and climate change in the Amazon Soy Moratorium came into force. Fig. 9 illustrates the co- 10 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 6. Topic-biome bipartide network graphs for EU print media, illustrating the evolution of associations between topics and mentioned biomes over time. Circle sizes indicate relative topic/entity prevalence and lines indicate edge weights. Only edges with weights above the median were included. 11 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 7. Topic-organization bipartide network graphs for EU print media, illustrating the evolution of associations between topics and mentioned organizations over time. Circle sizes indicate relative topic/entity prevalence and lines indicate edge weights. Only edges with weights within the upper 12.5% quantile were included. European journalistic field, which is more pronounced and constant seeds and intellectual property are a major concern in Argentina, likely than in the advocacy and academic fields. Brazilian and Argentinian due to the yearlong legal disputes over farmers’ rights to reproduce print media put more emphasis on topics dealing with economic impacts GMO seeds and the enforcement of intellectual property rights. and producers’ concerns. Brazilian print media pay more attention to We further observe two simultaneous dynamics over the period land conflicts than their European counterparts, while issues around 2017–2020: The Mercosur region appears among the most mentioned 12 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 8. Evolution of topic prevalence for selected topics in EU print media with links to key events. entities and the association between the EU and the topic “Deforesta­ concerns not only of the citizens of a given polity, but also of those tion” becomes significant for the first time (within the highest 12.5% affected by the production networks, whose final products are consumed edge weights). This suggests increasing attention towards the re­ by those citizens. This could introduce these issues into the centers of sponsibility of the EU as a major importer of soybeans and other FRCs decision-making and thereby provide legitimacy to the governance with the debate around the EU-Mercosur trade agreement. outcomes in the sense of an all-affected or “all-subjected” principle as The event driven nature of environmental journalism as a form of suggested by Fraser (2009). This could also be understood as a move contemporary news reporting has been pointed out before (Hansen, towards what O’Brien et al. (2009) see as a necessary new form of social 2011; Pezzullo & Cox, 2018). This snapshot-like focus on certain issues contract with a “larger conceptualization of ‘we’”, an expansion of moral (e.g. deforestation during the 2019 Amazon rainforest fires) does not communities, or evidence of what Eakin et al. (2014) call “feeling provide a broader historical context for audiences (e.g. continuous loss (empathy) at a distance”. of native vegetation in various biomes since the 1990s). Hansen (2011) However, while deforestation linked to soybean expansion had also points to the “authority orientation” of news coverage, which will already been significant in the Cerrado biome throughout the 1990s lead to a dominance of scientists’ and politicians’ key concerns (e.g., the (Beuchle et al., 2015), it only becomes a major concern in EU print global implications of greenhouse gas emissions from land-use change) media when the agricultural frontier moves further north and high over those of directly affected communities (e.g., land conflicts or annual deforestation rates in the Amazon biome are linked to soybean pesticide-induced intoxication). expansion in the early 2000 s. The strong focus on the Amazon biome Our longitudinal analysis of issues related to soybean expansion and the relatively low prevalence of topics related to issues directly confirms findings of other empirical studies. These are often grounded in affecting local populations (e.g., land conflicts, pesticide induced health theories of issue attention cycles (Downs, 1973), which aim to describe problems and disputes over farmers’ rights to reproduce GMO seeds) consecutive stages of attention to issues over time. Researchers have reflects a fundamental concern with the loss of an emblematic ecosystem traced media coverage on specific issues, but also the evolution of and the global threat of climate change rather than with the lives of coverage on entire domains (e.g., the environment) (Hansen, 2015a). In people on the other side of the value chain. Therefore, in can be argued this regard, our findings related to the shifting focus over time for the EU that EU print media have been more sensitive to the fate and state of journalistic field show close resemblance to Djerf-Pierre’s (2013) study ecosystems, rather than the wellbeing and struggles of people managing of news coverage relating to different environmental issues in Swedish and living within such distant Nature. television. This is particularly true for her findings concerning a focus on Thus, it is unclear, whether increasing attention towards deforesta­ “the environmental impact of food production (particularly GMOs, ‘mad tion really is an outcome of increasing porosity between public spheres. cow disease’ and the mistreatment of animals in industrialized meat EU news media appear to pay attention to major developments in Bra­ production) in the late 1990s and early 2000s; and finally, the heavy zilian political events (e.g., the stances toward deforestation adopted by focus on climate change in the late 2000s” (Djerf-Pierre, 2013, pp. the Lula administration or later by Jair Bolsonaro). However, in our 501–502). The peak in attention on climate change around 2006/2007 study period the journalistic field shows closest resemblance between has also been observed elsewhere and related to the release of the regions when a major NGO publishes an alarming report, stages a documentary film An Inconvenient Truth (2006) and the findings of the sensational protest, and pushes for a moratorium between private 2007 report issued by the Intergovernmental Panel on Climate Change stakeholders. More attention towards deforestation may therefore rather (IPCC) (Pezzullo & Cox, 2018). be the result of highly organized campaigning by increasingly powerful With regards to the significance of our findings for public sphere conservation NGOs, mainly based in the Global North. theory, the shift from topics more relevant to individuals’ personal life It is important to note that this influence is not only limited to the and in their role as consumers (GMOs, health, labeling) towards topics of level of attention an issue receives. Rather, large environmental NGOs environmental impacts at distant places (deforestation) or on a global also shape the framing of these issues and, following Dauvergne (2016), scale (climate change) over the study period, may lead to the conclusion often tend to pursue an “environmentalism of the rich”, emphasizing that public spheres become more porous and receptive towards issues acts of eco-consumerism and corporate social responsibility. Porosity from outside their own constituency. They appear to circulate the may then work in the opposite direction: governance becomes a matter 13 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Fig. 9. Co-Occurrence networks for the topic “Deforestation” and selected regions within journalistic field. Green nodes represent countries or regions, red nodes organizations or companies and yellow nodes named individuals. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) of decisions taken by corporate actors, mediated by major NGOs and exists (Rausch et al., 2019) and recently, the carbon footprint of soybean accountable mainly to final consumers, rather than an outcome of in­ exports from the Cerrado biome has been calculated to exceed those of clusive transnational public deliberation. other biomes in Brazil, largely due to emissions from land use change This casts a shadow on the promises of deliberative democracy, as (Escobar et al., 2020). Calls by experts for a Cerrado soy moratorium the amplification of voices from professional public relations de­ (Soterroni et al., 2019) have been met with great reluctance by the partments and the gatekeeping function of major media outlets do not industry. allow for a level playing field between different stakeholders. Journal­ The negotiations over the EU-Mercosur trade agreement appear to ism’s gatekeeping function may in fact prevent rather than stimulate an have temporarily pushed the question of the European Union’s re­ open public debate on fundamental asymmetries in the global food sponsibility for environmental impacts associated with its imports into system. The increasing reliance on wire services from Western based media discourse. Policy options to address this issue are currently being news agencies, due to the high expense of foreign correspondents, may assessed by the European Commission (Bager et al., 2020; European further aggravate this concern, as these tend to reflect the interests of the Parliament, 2020) and the European Parliament has recently voted for Global North (Boyd-Barrett, 2000; Hafez, 1999; Johnston & Forde, an Amendment to the EU-Mercosur trade agreement, signaling reluc­ 2011; Lewis et al., 2008). tance to ratify it, mostly over environmental impacts (Boffey, 2020). Moreover, the dominance of a single biome in media attention to­ However, the concerns voiced by European politicians primarily seem to wards soybean expansion may help explain why there has never been a echo the focus on Brazil’s handling of commodity expansion in the similar effort of private food law to prevent the further loss of native Amazon biome, particularly after the large media attention to the 2019 vegetation in the Cerrado as there has been for the Amazon. Evidence for Amazon rainforest fires. high rates of deforestation related to soybean farming in the Cerrado It remains questionable whether the general increase of attention to 14 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 abstract or distant impacts of imported agricultural commodities, such debate on asymmetries in the global food system. Finally, in the light of as soybeans, will lead to a widespread fundamental debate on the im­ the current restructuring of networked and individualized news pro­ plications of the current global food system for all groups affected by it. duction and consumption, we propose to include an analysis of social The findings of this study point instead to a pattern of recurring focus on networks and the sharing of and interactions with news content in future individual issues, mainly linked to limited geographical regions and studies. concerns. Also with regards to the political field, the outcomes of recent renegotiations on the EU’s agricultural subsidies have been read by Funding critics as mainly a continuation of the present state of large-scale, input- intensive agriculture and animal husbandry, largely dependent on the This work was supported by the Marie Skłodowska-Curie actions import of animal feed (Cwienk, 2020). (MSCA) grant agreement No 765408 from the European Commission: Further, theorists of communication studies have pointed to the COUPLED ‘Operationalising Telecouplings for Solving Sustainability individualization of media content consumption in the age of social Challenges for Land Use’. media and even proposed the end of agenda-setting (McCombs, 2005). Bennett & Iyengar (2008) suggest that the fragmentation of audiences leads to selective consumption of information, which reinforces the in­ Declaration of Competing Interest dividual’s prior views and concerns. For our findings, this could imply that some of the less prevalent issues may only reach limited audiences, The authors declare that they have no known competing financial which are largely already informed about these. However, others have interests or personal relationships that could have appeared to influence pointed to a more complex “inter-media agenda setting process” the work reported in this paper. (Anderson, 2014). It therefore remains open how the more interactive, networked configuration between traditional media and individualized Acknowledgements content generation will affect the construction of environmental issues in public spheres and what this means in terms of legitimacy and efficacy The authors wish to express their gratitude to Anna Rebekah Breding regarding processes of environmental governance. for providing help with language editing, to Ivan Savin for his feedback and the two reviewers for constructive comments and suggestions. We 6. Conclusion would also like to thank the community of open-source package de­ velopers for R and Python, particularly Dr. Chung-hong Chan for his In this contribution we have mobilized a text mining approach to support with the oolong library. trace the evolution of socio-ecological issues constructed around the expansion of soybean production and trade in the last two decades. Our Appendix A. Supplementary data focus has been on the journalistic field within the EU in relation to other fields and regions. Through this approach we have analyzed the func­ Supplementary data to this article can be found online at https://doi. tioning of contemporary, transnationalizing public spheres and their org/10.1016/j.gloenvcha.2021.102308. legitimacy and efficacy with regards to environmental governance. We have shown that in the EU journalistic field, a high prevalence of References issues around GMOs, health, and consumer labeling in the late 1990s has given way to more abstract and distant phenomena, such as deforesta­ Anderson, A.G. (Ed.), 2014. Media, Environment and the Network Society. Palgrave tion in producing regions and climate change since the mid-2000s. This Macmillan UK, London. Bager, S., Persson, M., Reis, T., 2020. Reducing commodity-driven tropical deforestation: has been accompanied with a broader perspective of different stake­ Political feasibility and ‘theories of change’ for EU policy options. SSRN Electron. J. holders, but also with a strong regional focus on the Amazon biome. Less https://doi.org/10.2139/ssrn.3624073. attention has been directed at immediate concerns for local commu­ Baran, S., Davis, D., 2011. Mass Communication Theory: Foundations, Ferment, and Future. Nelson Education. nities, such as land conflicts or disputes over intellectual property rights, Bennett, W.L., Iyengar, S., 2008. A new era of minimal effects? The changing foundations which are more prevalent in the advocacy field or the journalistic fields of political communication. J. Commun. 58 (4), 707–731. in Brazil and Argentina. These findings are broadly aligned with other Benson, R., 2009. Shaping the public sphere: Habermas and beyond. Am. Sociol. 40 (3), 175–197. https://doi.org/10.1007/s12108-009-9071-4. empirical studies in the field of environmental communication. Beuchle, R., Grecchi, R.C., Shimabukuro, Y.E., Seliger, R., Eva, H.D., Sano, E., Achard, F., Our findings suggest that there is some porosity between different 2015. Land cover changes in the Brazilian Cerrado and Caatinga biomes from 1990 public spheres and that reporting increasingly considers more distant to 2010 based on a systematic remote sensing sampling approach. Appl. Geogr. 58, and abstract impacts. However, the overarching concerns in EU news 116–127. https://doi.org/10.1016/j.apgeog.2015.01.017. Blei, D.M., Ng, A.Y., Jordan, M.I., 2003. Latent Dirichlet allocation. J. Mach. Learn. Res. media seem to relate to the loss of emblematic ecosystems and the 3 (4–5), 993–1022. https://doi.org/10.1016/b978-0-12-411519-4.00006-9. prospects of global warming as disaster-like events rather than to the Boffey, D. 2020, October 20. EU seeks Amazon protections pledge from Bolsonaro in struggles of local communities. This may help explain the relatively push to ratify trade deal. The Guardian. https://www.theguardian.com/world/20 20/oct/20/eu-seeks-amazon-rainforest-protections-pledge-from-bolsonaro-in-push narrow focus on current governance mechanisms and the reluctance of -to-ratify-trade-deal. corporate actors to expand these to other regions. Bohr, J. 2020. Reporting on climate change: A computational analysis of U.S. newspapers We also show that, even as the level of attention toward a given issue, and sources of bias, 1997–2017. Global Environmental Change, 61(August 2019), 102038. https://doi.org/10.1016/j.gloenvcha.2020.102038. such as deforestation, converges between different sources, the focus of Bondi, M., Scott, M., 2010. Keyness in Texts. John Benjamins Publishing. reporting can still differ quite considerably. Major events and the public Boussalis, C., Coan, T.G., 2016. Text-mining the signals of climate change doubt. Global relations efforts of professionalized actors may temporarily align this Environ. Change 36, 89–100. https://doi.org/10.1016/j.gloenvcha.2015.12.001. Boyd-Barrett, O. 2000. Constructing the global, constructing the local: News agencies re- focus. However, the themes thereby introduced can dominate reporting present the world. The Global Dynamics of News: Studies in International News in Western media far beyond their immediate aftermath. The asymmetry Coverage of News Agendas, 299–322. in different stakeholders’ ability to direct attention and construct Chang, J., Boyd-Graber, J., Gerrish, S., Wang, C., Blei, D.M., 2009. Reading tea leaves: How humans interpret topic models. Advances in Neural Information Processing dominant themes casts doubt on the promises of deliberative democracy Systems 22 - Proceedings of the 2009 Conference. in transnationalizing public spheres. Clark, K., Manning, C.D., 2016. Deep reinforcement learning for mention-ranking Even as the debate on the Mercosur trade agreement has provided a coreference models. In: EMNLP 2016 - Conference on Empirical Methods in Natural new spotlight on the European Union as an importing region of animal Language Processing, pp. 2256–2262. Cox, R., Depoe, S., 2015. Emergence and growth of the “field” of environmental feed, this limited focus may lead to addressing specific regional out­ communication. In: Hansen, A., Cox, R. (Eds.), The Routledge Handbook of comes through technical solutions and hinder a more fundamental Environmental Communication. Routledge, pp. 13–25. 15 F. Mempel and E. Corbera Global Environmental Change 69 (2021) 102308 Cwienk, J. 2020, October 21. EU agriculture policy: What are the bones of contention? Leguizamón, A., 2020. Seeds of Power - Environmental Injustice and Genetically DW.Com. https://www.dw.com/en/eu-agriculture-policy-what-are-the-bones-of-co Modified Soybeans in Argentina. Duke University Press. ntention/a-55352567. Lenschow, A., Newig, J., Challies, E., 2016. Globalization’s limits to the environmental Dauvergne, P. (Ed.), 2016. Environmentalism of the Rich Environmentalism of the Rich. state? Integrating telecoupling into global environmental governance. Environ. The MIT Press. Politics 25 (1), 136–159. https://doi.org/10.1080/09644016.2015.1074384. Djerf-Pierre, M., 2013. Green metacycles of attention: Reassessing the attention cycles of Lewis, J., Williams, A., Franklin, B., 2008. A compromised fourth estate?: UK news environmental news reporting 1961–2010. Public Understanding of Science 22 (4), journalism, public relations and news sources. J. Stud. 9 (1), 1–20. https://doi.org/ 495–512. https://doi.org/10.1177/0963662511426819. 10.1080/14616700701767974. Dow Jones & Company. 2020. Factiva. http://global.factiva.com. Lexis Nexis. (2020). Lexis Uni. http://www.lexisnexis.com. Downs, A. 1973. Up and down with ecology. The “Issue Attention” Cycle. McCombs, M., 2005. A look at agenda-setting: Past, present and future. J. Stud. 6 (4), Du Bois, C.M., 2018. The Story of Soy. Reaktion Books. 543–557. Eakin, H., DeFries, R., Kerr, S., Lambin, E.F., Liu, J., Marcotullio, P.J., Messerli, P., McKay, B., Colque, G., 2016. Bolivia’s soy complex: the development of ‘productive Reenberg, A., Rueda, X., Swaffield, S.R., Wicke, B., Zimmerer, K., 2014. In: exclusion’. J. Peasant Stud. 43 (2), 583–610. https://doi.org/10.1080/ Rethinking Global Land Use in an Urban Era. The MIT Press, pp. 141–161. https:// 03066150.2015.1053875. doi.org/10.7551/mitpress/9780262026901.003.0008. ModernMT. 2020. ModernMT REST API. https://www.modernmt.com/api/. Elgert, L., 2012. Certified discourse? The politics of developing soy certification Neiman, M., Blanco, M., 2020. Beyond the Pampas: Global capital and uneven standards. Geoforum 43 (2), 295–304. https://doi.org/10.1016/j. development in Argentine soybean expansion. J. Agrarian Change 20 (4), 538–561. geoforum.2011.08.008. https://doi.org/10.1111/joac.v20.410.1111/joac.12370. Escobar, N., Tizado, E. J., zu Ermgassen, E. K. H. J., Löfgren, P., Börner, J., Godar, J. O’Brien, K., Hayward, B., Berkes, F., 2009. Rethinking social contracts: Building 2020. Spatially-explicit footprints of agricultural commodities: Mapping carbon resilience in a changing climate. Ecol. Soc. 14 (2) https://doi.org/10.5751/ES- emissions embodied in Brazil’s soy exports. Global Environ. Change, 62(July 2019), 03027-140212. 102067. https://doi.org/10.1016/j.gloenvcha.2020.102067. Oliveira, G. de L.T., Schneider, M. 2016. The politics of flexing soybeans: China, Brazil European Parliament. 2020. Legislation with binding measures needed to stop EU-driven and global agroindustrial restructuring. J. Peasant Stud., 43(1), 167–194. https global deforestation. https://www.europarl.europa.eu/news/en/press-room/ ://doi.org/10.1080/03066150.2014.993625. 20201016IPR89560/. Pengue, W.A., 2009. Cuestiones económico-ambientales en las transformaciones FAO. (2018). FAOSTAT. http://www.fao.org/faostat/en/. agrícolas en las Pampas. Problemas Del Desarrollo 40 (157), 137–161. Fraser, N., 2009. Scales of Justice. Columbia University Press. Persson, Joel, Mertz, Ole, 2019. In: Telecoupling: Exploring Land-Use Change in a Fraser, N., Nash, K., 2014. Transnationalizing the Public Sphere. Polity. Globalised World. Springer International Publishing, Cham, pp. 313–336. https:// Garrett, R.D., Carlson, K.M., Rueda, X., Noojipady, P., 2016. Assessing the potential doi.org/10.1007/978-3-030-11105-2_17. additionality of certification by the round table on responsible soybeans and the Pezzullo, P.C., Cox, R., 2018. Environmental Communication and the Public Sphere, (5th roundtable on sustainable palm oil. Environ. Res. Lett. 11 (4), 1–17. https://doi.org/ ed.). SAGE Publications Inc. 10.1088/1748-9326/11/4/045003. Quinn, K.M., Monroe, B.L., Colaresi, M., Crespin, M.H., Radev, D.R., 2010. How to Goldsmith, P. 2017. The Faustian Bargain of Tropical Soybean Production. Tropical analyze political attention with minimal assumptions and costs. Am. J. Polit. Sci. 54 Conserv. Sci., 10(July), 194008291772389. https://doi.org/10.1177/194008291 (1), 209–228. https://doi.org/10.1111/j.1540-5907.2009.00427.x. 7723892. Rausch, Lisa L., Gibbs, Holly K., Schelly, Ian, Brandão, Amintas, Morton, Douglas C., Grimmer, J., Stewart, B.M., 2013. Text as data: The promise and pitfalls of automatic Filho, Arnaldo Carneiro, Strassburg, Bernardo, Walker, Nathalie, content analysis methods for political texts. Polit. Anal. 21 (3), 267–297. https://doi. Noojipady, Praveen, Barreto, Paulo, Meyer, Daniel, 2019. Soy expansion in Brazil’s org/10.1093/pan/mps028. Cerrado. Conserv. Lett. 12 (6) https://doi.org/10.1111/conl.v12.610.1111/ Günther, E., Quandt, T., 2016. Word counts and topic models: Automated text analysis conl.12671. methods for digital journalism research. Digital J. 4 (1), 75–88. https://doi.org/ Roberts, M.E., Stewart, B.M., Tingley, D., 2019. Stm: An R package for structural topic 10.1080/21670811.2015.1093270. models. J. Stat. Softw. 91 (2). https://doi.org/10.18637/jss.v091.i02. Habermas, J., 1989. In: The Structural Transformation of the Public Sphere. The MIT Roberts, Margaret E., Stewart, Brandon M., Tingley, Dustin, Lucas, Christopher, Leder- Press. https://doi.org/10.1360/zd-2013-43-6-1064. Luis, Jetson, Gadarian, Shana Kushner, Albertson, Bethany, Rand, David G., 2014. Habermas, J., 1996. Between facts and norms: Contributions to a discourse theory of law Structural topic models for open-ended survey responses. Am. J. Polit. Sci. 58 (4), and democracy. Polity. 1064–1082. https://doi.org/10.1111/ajps.2014.58.issue-410.1111/ajps.12103. Hafez, K., 1999. International news coverage and the problems of media globalization. In Scheufele, D.A., Tewksbury, D., 2007. Framing, agenda setting, and priming: The search of a ‘new global-local nexus’. Innov. Eur. J. Soc. Sci. Res. 12 (1), 47–62. evolution of three media effects models. J. Commun. 57, 9–20. https://doi.org/ https://doi.org/10.1080/13511610.1999.9968587. 10.1111/j.0021-9916.2007.00326.x. Hansen, A., 2011. Communication, media and environment: Towards reconnecting Schmidt, A., Ivanova, A., Schäfer, M.S., 2013. Media attention for climate change around research on the production, content and social implications of environmental the world: A comparative analysis of newspaper coverage in 27 countries. Global communication. Int. Commun. Gazette 73 (1), 7–25. https://doi.org/10.1177/ Environ. Change 23 (5), 1233–1248. https://doi.org/10.1016/j. 1748048510386739. gloenvcha.2013.07.020. Hansen, A., 2015a. News coverage of the environment: a longitudinal perspective. In: Schumacher, G., Schoonvelde, M., Goyal, T., de Vries, E., Traber, D., Dahiya, T., de The Routledge Handbook of Environmental Communication. Routledge, Vries, E., 2016. EUSpeech: A new dataset of EU elite speeches. In: Proceedings of the pp. 209–220. International Conference on the Advances in Computational Analysis of Political Hansen, A., 2015b. Communication, media and the social construction of the Text, pp. 75–80. https://doi.org/10.7910/DVN/XPCVEI. environment. In: The Routledge Handbook of Environmental Communication. Sikor, T., Auld, G., Bebbington, A.J., Benjaminsen, T.A., Gentry, B.S., Hunsberger, C., Routledge, pp. 46–58. https://doi.org/10.4324/9781315887586-12. Izac, A.M., Margulis, M.E., Plieninger, T., Schroeder, H., Upton, C., 2013. Global Honnibal, M., Johnson, M., 2015. An improved non-monotonic transition system for land governance: From territory to flow? Curr. Opin. Environ. Sustain. 5 (5), dependency parsing. Proceedings of the 2015 Conference on Empirical Methods in 522–527. https://doi.org/10.1016/j.cosust.2013.06.006. Natural Language Processing. Soterroni, A.C., Ramos, F.M., Mosnier, A., Fargione, J., Andrade, P.R., Baumgarten, L., Ignatow, G., Mihalcea, R. 2017. An Introduction to Text Mining: Research Design, Data Pirker, J., Obersteiner, M., Kraxner, F., Câmara, G., Carvalho, A.X.Y.Y., Polasky, S., Collection, and Analysis. SAGE Publications. 2019. Expanding the soy moratorium to Brazil’s Cerrado. Sci. Adv. 5 (7), eaav7336. Johnston, J., Forde, S., 2011. The silent partner: News agencies and 21st century news. https://doi.org/10.1126/sciadv.aav7336. Int. J. Commun. 5, 20. van der Meulen, B.M.J. 2011. Private food law - Governing food chains through contract Lambin, E.F., Gibbs, H.K., Heilmayr, R., Carlson, K.M., Fleck, L.C., Garrett, R.D., le Polain law, self-regulation, private standards, audits and certification schemes (B. M. J. van de Waroux, Y., McDermott, C.L., McLaughlin, D., Newton, P., Nolte, C., Pacheco, P., der Meulen (ed.); Vol. 6). Wageningen Academic Publishers. https://doi.org/10.392 Rausch, L.L., Streck, C., Thorlakson, T., Walker, N.F., 2018. The role of supply-chain 0/978-90-8686-730-1. initiatives in reducing deforestation. Nat. Clim. Change 8 (2), 109–116. https://doi. Vu, H.T., Liu, Y., Tran, D.V., 2019. Nationalizing a global phenomenon: A study of how org/10.1038/s41558-017-0061-1. the press in 45 countries and territories portrays climate change. Global Environ. Leguizamón, A., 2014. Modifying Argentina: GM soy and socio-environmental change. Change 58 (June), 101942. https://doi.org/10.1016/j.gloenvcha.2019.101942. Geoforum 53, 149–160. https://doi.org/10.1016/j.geoforum.2013.04.001. Wesslen, R. 2018. Computer-Assisted Text Analysis for Social Science: Topic Models and Beyond. http://arxiv.org/abs/1803.11045. 16