Search

During the last fifteen years, automatic text scaling has become one of the key tools of the Text as Data community in political science. Prominent text scaling algorithms, however, rely on the assumption that latent positions can be captured just by leveraging the information about word frequencies in documents under study. We challenge this traditional view and present a new, semantically aware text scaling algorithm, SemScale, which combines recent developments in the area of computational linguistics with unsupervised graph-based clustering. We conduct an extensive quantitative analysis over a collection of speeches from the European Parliament in five different languages and from two different legislative terms, and show that a scaling approach relying on semantic document representations is often better at capturing known underlying political dimensions than the established frequency-based (i.e., symbolic) scaling method. We further validate our findings through a series of experiments focused on text preprocessing and feature selection, document representation, scaling of party manifestos, and a supervised extension of our algorithm. To catalyze further research on this new branch of text scaling methods, we release a Python implementation of SemScale with all included data sets and evaluation procedures., Updated version - accepted for Transactions on Data Science (TDS)

Populist discourse—which tends to benefit anti-systemic parties—has been on the rise in the world’s democratic states. Powerful non-democratic states have both the means and the incentive to spread such discourse to democratic states. We clarify the incentives illiberal states have to produce such communication, and delineate how this type of political communication fuses traditional state-to-state propaganda with election interventions. We draw on the case of Kremlin-sponsored communication on the issue of refugees in Germany to illustrate the mechanisms through which the discourse operates in target countries. We create a corpus of over a million news stories to identify the prevalence of illiberal discourse and its timing relative to Germany’s elections. We show that the Kremlin intervened in the 2017 federal elections by promoting refugee stories over and above the rate at which German outlets did. We discuss the broader implications for the use of directed political communication as a form of election intervention.

Therecording of this talk canbe streamed on YouTube. The Research Engineering Group at The Alan Turing Institute started our RSE journey in autumn 2016as a new team in a new institute. Without the usual constraints and advantages of RSE groups based in universities, the team has had to find its path for sustainable existence in a rapidly evolving institute and field. In half a decade the team has grown from 3 to 36 members, going through a few rounds of refining our roles, the way we recruit new members, and how we support members of the team to progress into their next role - whether at the Turing or beyond. The journey has been guided by our principles: transparency in leadership and decisions; diversity of talent, people and experience; supporting individuals in their career journey; and focus on our role as expert collaborators. In this talk, we aim to share our journey, the successes, the false starts and the lessons learned.

The progressive digitization of historical archives provides new, often domain specific, textual resources that report on facts and events which have happened in the past; among these, memoirs are a very common type of primary source. In this paper, we present an approach for extracting information from Italian historical war memoirs and turning it into structured knowledge. This is based on the semantic notions of events, participants and roles. We evaluate quantitatively each of the key-steps of our approach and provide a graph-based representation of the extracted knowledge, which allows to move between a Close and a Distant Reading of the collection., Comment: 23 pages, 6 figures

Why do states intervene in elections abroad? This article argues that outsiders intervene when the main domestic contenders for office adopt policy positions that differ from the point of view of the outside power. It refers to the split between the government's and opposition's positions as policy polarization. Polarization between domestic political forces, rather than the degree of unfriendliness of the government in office, attracts two types of interventions: process (for or against democracy) and candidate (for or against the government) interventions. The study uses a novel, original data set to track local contenders’ policy positions. It shows that the new policy polarization measurement outperforms a number of available alternatives when it comes to explaining process and candidate interventions. The authors use this measurement to explain the behavior of the United States as an intervener in elections from 1945 to 2012. The United States is more likely to support the opposition, and the democratic process abroad, if a pro-US opposition is facing an anti-US government. It is more likely to support the government, and undermine the democratic process abroad, if a pro-US government is facing an anti-US opposition. The article also presents the results for all interveners, confirming the results from the US case.

Amid the Coronavirus Disease 2019 (COVID-19) pandemic, preprints in the biomedical sciences are being posted and accessed at unprecedented rates, drawing widespread attention from the general public, press, and policymakers for the first time. This phenomenon has sharpened long-standing questions about the reliability of information shared prior to journal peer review. Does the information shared in preprints typically withstand the scrutiny of peer review, or are conclusions likely to change in the version of record? We assessed preprints from bioRxiv and medRxiv that had been posted and subsequently published in a journal through April 30, 2020, representing the initial phase of the pandemic response. We utilised a combination of automatic and manual annotations to quantify how an article changed between the preprinted and published version. We found that the total number of figure panels and tables changed little between preprint and published articles. Moreover, the conclusions of 7.2% of non-COVID-19–related and 17.2% of COVID-19–related abstracts undergo a discrete change by the time of publication, but the majority of these changes do not qualitatively change the conclusions of the paper., During the COVID-19 pandemic, preprints in the biomedical sciences are being posted and accessed at unprecedented rates, but can we trust them? This study reveals that the majority of preprints published within the first 4 months of the pandemic are comparable to their published version.

Existing datasets of economic sanctions rely primarily on secondary sources and do not tend to take full advantage of government documents related to economic coercion. Such data may miss sanctions, and do not capture important details in how coercive measures are threatened, imposed and removed. The latter processes often have much to do with the domestic politics in sender countries. Understanding these processes may be necessary in order to fully account for sanctions’ effectiveness. We present a natural language processing (NLP) approach to retrieving sanctions-related government documents. We apply our method to the case of US sanctions. The United States is the world’s pre-eminent user of sanctions. Our method can be applied to other cases. We collect all sanctions events originating in the office of the US president, and all congressional sanctions, for 1988–2016. Our approach has three advantages: (1) by design, it captures all sanctions-related documents; (2) the resulting data are disaggregated by imposing branch of government; (3) the data include the original language of the measures. These features directly shed light on interbranch delegation, domestic (partisan) conflict, and policy priorities. We show that our data record more episodes than most existing sanctions’ data, and have features that other datasets lack. The availability of the original text opens up new avenues for research and analysis.

Existing datasets of economic sanctions do not tend to take full advantage of government documents related to economic coercion. They may miss some economic sanctions, and do not capture some important details in how measures are threatened, imposed and removed. The latter often have to do with the domestic politics in sender countries, and understanding them may be necessary to fully account for sanctions' effectiveness. We present a natural-language processing (NLP) approach to retrieving sanctions-related government documents for the US case. We collect all sanctions events originating in the office of the US President, and all Congressional sanctions, for 1988-2016. Our approach has three advantages: (1) by design, it captures all sanctions-related documents; (2) the resulting data is disaggregated by imposing branch; (3) the data includes the original language of the measures. These features directly shed light on inter-branch delegation, domestic (partisan) conflict, and policy priorities.

Recognizing toponyms and resolving them to their real-world referents is required for providing advanced semantic access to textual data. This process is often hindered by the high degree of variation in toponyms. Candidate selection is the task of identifying the potential entities that can be referred to by a toponym previously recognized. While it has traditionally received little attention in the research community, it has been shown that candidate selection has a significant impact on downstream tasks (i.e. entity resolution), especially in noisy or non-standard text. In this paper, we introduce a flexible deep learning method for candidate selection through toponym matching, using state-of-the-art neural network architectures. We perform an intrinsic toponym matching evaluation based on several new realistic datasets, which cover various challenging scenarios (cross-lingual and regional variations, as well as OCR errors). We report its performance on candidate selection in the context of the downstream task of toponym resolution, both on existing datasets and on a new manually-annotated resource of nineteenth-century English OCR'd text., Comment: 10 pages, 1 figure

The world continues to face a life-threatening viral pandemic. The virus underlying the COVID-19 disease, SARS-CoV-2, has caused over 98 million confirmed cases and 2.2 million deaths since January 2020. Although the most recent respiratory viral pandemic swept the globe only a decade ago, the way science operates and responds to current events has experienced a paradigm shift in the interim. The scientific community has responded rapidly to the COVID-19 pandemic, releasing over 125,000 COVID-19 related scientific articles within 10 months of the first confirmed case, of which more than 30,000 were hosted by preprint servers. We focused our analysis on bioRxiv and medRxiv, two growing preprint servers for biomedical research, investigating the attributes of COVID-19 preprints, their access and usage rates, as well as characteristics of their propagation on online platforms. Our data provides evidence for increased scientific and public engagement with preprints related to COVID-19 (COVID-19 preprints are accessed more, cited more, and shared more on various online platforms than non-COVID-19 preprints), as well as changes in the use of preprints by journalists and policymakers. We also find evidence for changes in preprinting and publishing behaviour: COVID-19 preprints are shorter and reviewed faster. Our results highlight the unprecedented role of preprints and preprint servers in the dissemination of COVID-19 science, and the impact of the pandemic on the scientific communication landscape.

This paper proposes a new approach to animacy detection, the task of determining whether an entity is represented as animate in a text. In particular, this work is focused on atypical animacy and examines the scenario in which typically inanimate objects, specifically machines, are given animate attributes. To address it, we have created the first dataset for atypical animacy detection, based on nineteenth-century sentences in English, with machines represented as either animate or inanimate. Our method builds on recent innovations in language modeling, specifically BERT contextualized word embeddings, to better capture fine-grained contextual properties of words. We present a fully unsupervised pipeline, which can be easily adapted to different contexts, and report its performance on an established animacy dataset and our newly introduced resource. We show that our method provides a substantially more accurate characterization of atypical animacy, especially when applied to highly complex forms of language use., 12 pages, 1 figures