Showing papers by "School for Advanced Studies in the Social Sciences published in 2020"

Libri-Light: A Benchmark for ASR with Limited or No Supervision

Abstract: We introduce a new collection of spoken English audio suitable for training speech recognition systems under limited or no supervision. It is derived from open-source audio books from the LibriVox project. It contains over 60K hours of audio, which is, to our knowledge, the largest freely-available corpus of speech. The audio has been segmented using voice activity detection and is tagged with SNR, speaker ID and genre descriptions. Additionally, we provide baseline systems and evaluation metrics working under three settings: (1) the zero resource/unsupervised setting (ABX), (2) the semi- supervised setting (PER, CER) and (3) the distant supervision setting (WER). Settings (2) and (3) use limited textual resources (10 minutes to 10 hours) aligned with the speech. Setting (3) uses large amounts of unaligned text. They are evaluated on the standard LibriSpeech dev and test sets for comparison with the supervised state-of-the-art.

Living in Networks: The Dynamics of Social Relations

Abstract: How do personal networks emerge from social contexts? How do these evolve during the course of a lifetime? How are relationships established, maintained, connected, disrupted? How does the structure of a network evolve as people face transitions and events? Based on a classic text originally published in France and that has become the standard on the empirical study of social networks there, for the first time, a network analysis perspective is extended from contexts and social circles to relationships and life events through empirical studies. Following in the tradition of personal network studies, this contribution to the field of structural analysis in sociology offers both a synthesis of knowledge and original results from two immense surveys carried out in France. This volume proposes an original theory grounded in relational dynamics, offering novel perspectives on individual social relations over the course of a lifetime through the context of personal networks, access to social resources, and inequalities.

Urban Science: Integrated Theory from the First Cities to Sustainable Metropolises

Abstract: Urban science seeks to understand the fundamental processes that drive, shape and sustain cities and urbanization. It is a multi/transdisciplinary approach involving concepts, methods and research from the social, natural, engineering and computational sciences, along with the humanities. This report is intended to convey the current “state of the art” in urban science while also clearly indicating how urban science builds upon and complements (but does not replace) prior work on cities and urbanization in many other disciplines. The report does not aim at a fully comprehensive synopsis of work done under the rubric of “urban science” but it does aim to convey what makes urban science different from discipline-based examinations of cities and urbanization. It also highlights novel insights generated by the inherently multidisciplinary inquiry that urban science exemplifies. The authors of the report are all based in academic or research institutions but several of them are close to practice by virtue of collaboration with NGOs and community groups and engagement with policy. The authors also represent different academic disciplines and varied traditions of scientific inquiry. The report is meant to facilitate, and hopefully also provoke, discussion among the many stakeholders for whom a scientifically based, empirically rich, and historically deep understanding of cities and urbanization is not only intellectually compelling but also socially urgent and ethically pressing. We believe that the innovative scholarship constituting urban science can importantly provide scientific leadership to support meeting the urgent challenges of global sustainable development.

Internet users engage more with phatic posts than with health misinformation on Facebook

Abstract: Social media like Facebook are harshly criticized for the propagation of health misinformation. Yet, little research has provided in-depth analysis of real-world data to measure the extent to which Internet users engage with it. This article examines 6.5 million interactions generated by 500 posts on an emblematic case of online health misinformation: the Facebook page Sante + Mag, which generates five times more interactions than the combination of the five best-established French media outlets. Based on the literature on cultural evolution, we tested whether the presence of cognitive factors of attraction, that tap into evolved cognitive preferences, such as information related to sexuality, social relations, threat, disgust or negative emotions, could explain the success of Sante + Mag’s posts. Drawing from media studies findings, we hypothesized that their popularity could be driven by Internet users’ desire to interact with their friends and family by sharing phatic posts (i.e. statements with no practical information fulfilling a social function such as “hello” or “sister, I love you”). We found that phatic posts were the strongest predictor of interactions, followed by posts with a positive emotional valence. While 50% of the posts were related to social relations, only 28% consisted of health misinformation. Despite its cognitive appeal, health misinformation was a negative predictor of interactions. Sexual content negatively predicted interactions and other factors of attraction such as disgust, threat or negative emotions did not predict interactions. These results strengthen the idea that Facebook is first and foremost a social network used by people to foster their social relations, not to spread online misinformation. We encourage researchers working on misinformation to conduct finer-grained analysis of online content and to adopt interdisciplinary approach to study the phatic dimension of communication, together with positive content, to better understand the cultural evolution dynamics of social media.

Accuracy of the Language Environment Analysis System Segmentation and Metrics: A Systematic Review.

Abstract: Purpose The Language Environment Analysis (LENA) system provides automated measures facilitating clinical and nonclinical research and interventions on language development, but there are only a few, scattered independent reports of these measures' validity. The objectives of the current systematic review were to (a) discover studies comparing LENA output with manual annotation, namely, accuracy of talker labels, as well as involving adult word counts (AWCs), conversational turn counts (CTCs), and child vocalization counts (CVCs); (b) describe them qualitatively; (c) quantitatively integrate them to assess central tendencies; and (d) quantitatively integrate them to assess potential moderators. Method Searches on Google Scholar, PubMed, Scopus, and PsycInfo were combined with expert knowledge, and interarticle citations resulting in 238 records screened and 73 records whose full text was inspected. To be included, studies must target children under the age of 18 years and report on accuracy of LENA labels (e.g., precision and/or recall) and/or AWC, CTC, or CVC (correlations and/or error metrics). Results A total of 33 studies, in 28 articles, were discovered. A qualitative review revealed most validation studies had not been peer reviewed as such and failed to report key methodology and results. Quantitative integration of the results was possible for a broad definition of recall and precision (M = 59% and 68%, respectively; N = 12-13), for AWC (mean r = .79, N = 13), CVC (mean r = .77, N = 5), and CTC (mean r = .36, N = 6). Publication bias and moderators could not be assessed meta-analytically. Conclusion Further research and improved reporting are needed in studies evaluating LENA segmentation and quantification accuracy, with work investigating CTC being particularly urgent. Supplemental Material https://osf.io/4nhms/.

An Annotated Corpus for Sexism Detection in French Tweets

Abstract: Social media networks have become a space where users are free to relate their opinions and sentiments which may lead to a large spreading of hatred or abusive messages which have to be moderated. This paper presents the first French corpus annotated for sexism detection composed of about 12,000 tweets. In a context of offensive content mediation on social media now regulated by European laws, we think that it is important to be able to detect automatically not only sexist content but also to identify if a message with a sexist content is really sexist (i.e. addressed to a woman or describing a woman or women in general) or is a story of sexism experienced by a woman. This point is the novelty of our annotation scheme. We also propose some preliminary results for sexism detection obtained with a deep learning approach. Our experiments show encouraging results.

Hydroxychloroquine with or without azithromycin and in-hospital mortality or discharge in patients hospitalized for COVID-19 infection: a cohort study of 4,642 in-patients in France

Abstract: Objective To assess the clinical effectiveness of oral hydroxychloroquine (HCQ) with or without azithromycin (AZI) in preventing death or leading to hospital discharge. Design Retrospective cohort study. Setting An analysis of data from electronic medical records and administrative claim data from the French Assistance Publique - Hopitaux de Paris (AP-HP) data warehouse, in 39 public hospitals, Ile-de-France, France. Participants All adult inpatients with at least one PCR-documented SARS-CoV-2 RNA from a nasopharyngeal sample between February 1st, 2020 and April 6th, 2020 were eligible for analysis. The study population was restricted to patients who did not receive COVID-19 treatments assessed in ongoing trials, including antivirals and immunosuppressive drugs. End of follow-up was defined as the date of death, discharge home, day 28 after admission, whichever occurred first, or administrative censoring on May 4, 2020. Intervention Patients were further classified into 3 groups: (i) receiving HCQ alone, (ii) receiving HCQ together with AZI, and (iii) receiving neither HCQ nor AZI. Exposure to a HCQ/AZI combination was defined as a simultaneous prescription of the 2 treatments (more or less one day). Main outcome measures The primary outcome was all-cause 28-day mortality as a time-to-event endpoint under a competing risks survival analysis framework. The secondary outcome was 28-day discharge home. Augmented inverse probability of treatment weighted (AIPTW) estimates of the average treatment effect (ATE) were computed to account for confounding. Results A total of 4,642 patients (mean age: 66.1 ± 18; males: 2,738 (59%)) were included, of whom 623 (13.4%) received HCQ alone, 227 (5.9%) received HCQ plus AZI, and 3,792 (81.7%) neither drug. Patients receiving ‘HCQ alone’ or ‘HCQ plus AZI’ were more likely younger, males, current smokers and overall presented with slightly more co-morbidities (obesity, diabetes, any chronic pulmonary diseases, liver diseases), while no major difference was apparent in biological parameters. After accounting for confounding, no statistically significant difference was observed between the ‘HCQ’ and ‘Neither drug’ groups for 28-day mortality: AIPTW absolute difference in ATE was +1.24% (−5.63 to 8.12), ratio in ATE 1.05 (0.77 to 1.33). 28-day discharge rates were statistically significantly higher in the ‘HCQ’ group: AIPTW absolute difference in ATE (+11.1% [3.30 to 18.9]), ratio in ATE (1.25 [1.07 to 1.42]). As for the ‘HCQ+AZI’ vs neither drug, trends for significant differences and ratios in AIPTW ATE were found suggesting higher mortality rates in the former group (difference in ATE +9.83% [-0.51 to 20.17], ratio in ATE 1.40 [0.98 to 1.81];p=0.062). Conclusions Using a large non-selected population of inpatients hospitalized for COVID-19 infection in 39 hospitals in France and robust methodological approaches, we found no evidence for efficacy of HCQ or HCQ combined with AZI on 28-day mortality. Our results suggested a possible excess risk of mortality associated with HCQ combined with AZI, but not with HCQ alone. Significantly higher rates of discharge home were observed in patients treated by HCQ, a novel finding warranting further confirmation in replicative studies. Altogether, our findings further support the need to complete currently undergoing randomized clinical trials. WHAT THIS PAPER ADDS? What is already known on this subject -The use of Hydroxychloroquine (HCQ) or HCQ with azithromycin (AZI) has been associated with viral load reduction at 6 days in COVID-19 infected patients -No difference between HCQ and no-HCQ groups in terms of risk of death or need for mechanical ventilation was found in two large cohorts of hospitalized COVID-19 infected patients What this study adds -Using a large non-selected population of inpatients hospitalized for COVID-19 infection in 39 hospitals in France and robust methodological approaches, we found no evidence for efficacy of HCQ on 28-day mortality -Our results suggest an excess risk of mortality in patients treated by a combination of HCQ and AZI, but not with HCQ alone -Significantly higher rates of discharge home were observed in patients treated by HCQ, a novel finding warranting further confirmation in replicative studies

A parsimonious approach for spatial transmission and heterogeneity in the COVID-19 propagation

Abstract: Raw data on the number of deaths at a country level generally indicate a spatially variable distribution of COVID-19 incidence. An important issue is whether this pattern is a consequence of environmental heterogeneities, such as the climatic conditions, during the course of the outbreak. Another fundamental issue is to understand the spatial spreading of COVID-19. To address these questions, we consider four candidate epidemiological models with varying complexity in terms of initial conditions, contact rates and non-local transmissions, and we fit them to French mortality data with a mixed probabilistic-ODE approach. Using statistical criteria, we select the model with non-local transmission corresponding to a diffusion on the graph of counties that depends on the geographic proximity, with time-dependent contact rate and spatially constant parameters. This suggests that in a geographically middle size centralized country such as France, once the epidemic is established, the effect of global processes such as restriction policies and sanitary measures overwhelms the effect of local factors. Additionally, this approach reveals the latent epidemiological dynamics including the local level of immunity, and allows us to evaluate the role of non-local interactions on the future spread of the disease.

The growth equation of cities

Abstract: The science of cities seeks to understand and explain regularities observed in the world's major urban systems. Modelling the population evolution of cities is at the core of this science and of all urban studies. Quantitatively, the most fundamental problem is to understand the hierarchical organization of cities and the statistical occurrence of megacities, first thought to be described by a universal law due to Zipf, but whose validity has been challenged by recent empirical studies. A theoretical model must also be able to explain the relatively frequent rises and falls of cities and civilizations, and despite many attempts these fundamental questions have not been satisfactorily answered yet. Here we fill this gap by introducing a new kind of stochastic equation for modelling population growth in cities, which we construct from an empirical analysis of recent datasets (for Canada, France, UK and USA) that reveals how rare but large interurban migratory shocks dominate city growth. This equation predicts a complex shape for the city distribution and shows that Zipf's law does not hold in general due to finite-time effects, implying a more complex organization of cities. It also predicts the existence of multiple temporal variations in the city hierarchy, in agreement with observations. Our result underlines the importance of rare events in the evolution of complex systems and at a more practical level in urban planning.

Hydroxychloroquine Controversies: Clinical Trials, Epistemology, and the Democratization of Science.

Abstract: The claim that anti-malaria drugs, chloroquine and hydroxychloroquine, can cure COVID-19 became a focus of fierce political battles that pitted promoters of these pharmaceuticals, Presidents Bolsonaro and Trump among them, against "medical elites." At the center of these battles are different meanings of effectiveness in medicine, the complex role of randomized clinical trials (RCTs) in proving such effectiveness, the task of medical experts and the state in regulating pharmaceuticals, patients' activism, and the collective production of medical knowledge. This article follows the trajectory of chloroquine and hydroxychloroquine as anti-COVID-19 drugs, focusing on the reception of views of their main scientific promoter, the French infectious disease specialist, Didier Raoult. The surprising career of these drugs, our text proposes, is fundamentally a political event, not in the narrow sense of engaging specific political fractions, but in the much broader sense of the politics of public participation in science.

Longform recordings of everyday life: Ethics for best practices.

Abstract: Recent advances in large-scale data storage and processing offer unprecedented opportunities for behavioral scientists to collect and analyze naturalistic data, including from underrepresented groups. Audio data, particularly real-world audio recordings, are of particular interest to behavioral scientists because they provide high-fidelity access to subtle aspects of daily life and social interactions. However, these methodological advances pose novel risks to research participants and communities. In this article, we outline the benefits and challenges associated with collecting, analyzing, and sharing multi-hour audio recording data. Guided by the principles of autonomy, privacy, beneficence, and justice, we propose a set of ethical guidelines for the use of longform audio recordings in behavioral research. This article is also accompanied by an Open Science Framework Ethics Repository that includes informed consent resources such as frequent participant concerns and sample consent forms.

The Welfare Conventions Approach: A Comparative Perspective on Social Impact Bonds

Abstract: The Social Impact Bond (SIB) is a new funding mechanism for welfare programs. It is supposed to create savings for the public sector from which private returns can be deducted. Presented as a purel...

He said “who’s gonna take care of your children when you are at ACL?”: Reported Sexist Acts are Not Sexist

Abstract: In a context of offensive content mediation on social media now regulated by European laws, it is important not only to be able to automatically detect sexist content but also to identify if a message with a sexist content is really sexist or is a story of sexism experienced by a woman. We propose: (1) a new characterization of sexist content inspired by speech acts theory and discourse analysis studies, (2) the first French dataset annotated for sexism detection, and (3) a set of deep learning experiments trained on top of a combination of several tweet’s vectorial representations (word embeddings, linguistic features, and various generalization strategies). Our results are encouraging and constitute a first step towards offensive content moderation.

Tubes and bubbles topological confinement of YouTube recommendations.

Abstract: The role of recommendation algorithms in online user confinement is at the heart of a fast-growing literature. Recent empirical studies generally suggest that filter bubbles may principally be observed in the case of explicit recommendation (based on user-declared preferences) rather than implicit recommendation (based on user activity). We focus on YouTube which has become a major online content provider but where confinement has until now been little-studied in a systematic manner. We aim to contribute to the above literature by showing whether recommendation on YouTube exhibits phenomena typical of filter bubbles, tending to lower the diversity of consumed content. Starting from a diverse number of seed videos, we first describe the properties of the sets of suggested videos in order to design a sound exploration protocol able to capture latent recommendation graphs recursively induced by these suggestions. These graphs form the background of potential user navigations along non-personalized recommendations. From there, be it in topological, topical or temporal terms, we show that the landscape of what we call mean-field YouTube recommendations is often prone to confinement dynamics. Moreover, the most confined recommendation graphs i.e., potential bubbles, seem to be organized around sets of videos that garner the highest audience and thus plausibly viewing time.

Speech Technology for Unwritten Languages

Abstract: Speech technology plays an important role in our everyday life. Among others, speech is used for human-computer interaction, for instance for information retrieval and on-line shopping. In the case of an unwritten language, however, speech technology is unfortunately difficult to create, because it cannot be created by the standard combination of pre-trained speech-to-text and text-to-speech subsystems. The research presented in this article takes the first steps towards speech technology for unwritten languages. Specifically, the aim of this work was 1) to learn speech-to-meaning representations without using text as an intermediate representation, and 2) to test the sufficiency of the learned representations to regenerate speech or translated text, or to retrieve images that depict the meaning of an utterance in an unwritten language. The results suggest that building systems that go directly from speech-to-meaning and from meaning-to-speech, bypassing the need for text, is possible.

Catastrophic climate change, population ethics and intergenerational equity

Abstract: Climate change threatens irreversible and dangerous impacts, possibly leading to extinction. The most relevant trade-off then may not be between present and future consumption but between present consumption and the mere existence of future generations. To investigate this trade-off, we build an integrated assessment model that explicitly accounts for the risk of extinction of future generations. Using the class of number-dampened utilitarian social welfare functions, we compare different climate policies that change the probability of catastrophic outcomes yielding an early extinction. We analyse the role of inequality aversion and population ethics. Low inequality aversion and a preference for large populations favour the most ambitious climate policy, although there are cases where the effect of inequality aversion on the preferred policy is reversed. This is due to the fact that a higher inequality aversion both decreases the welfare loss of reducing consumption of the current generation and also decreases the welfare gain of reducing the future risk of extinction.

Technological Fix or Divisible Object of Collective Concern? Histories of Conflict over the Geological Disposal of Nuclear Waste in Sweden and France

Abstract: Science and technology studies (STS) has cultivated a positive vision of technological controversies. By raising new issues to address, controversies are seen as generating more thorough and exhaus...

Dual language input and the impact of language separation on early lexical development

Abstract: We examined properties of the input and the environment that characterize bilingual exposure in 11-month-old infants with a regular exposure to French and an additional language, and their possible effects on receptive vocabulary size. Using a diary method, we found that a majority of the families roughly followed a one-parent-one-language approach. Yet, the two languages co-occurred to various extents within the same half-hour both within and across speakers. We used exploratory correlation analyses to examine potential effects of the dual input on the size of infants' vocabularies. The results revealed some evidence for an impact of language separation by speakers.

A three-level classification of French tweets in ecological crises

Abstract: The possibilities that emerge from micro-blogging generated content for crisis-related situations make automatic crisis management using natural language processing techniques a hot research topic. Our aim here is to contribute to this line of research focusing for the first time on French tweets related to ecological crises in order to support the French Civil Security and Crisis Management Department to provide immediate feedback on the expectations of the populations involved in the crisis. We propose a new dataset manually annotated according to three dimensions: relatedness, urgency and intentions to act. We then experiment with binary classification (useful vs. non useful), three-class (non useful vs. urgent vs. non urgent) and multiclass classification (i.e., intention to act categories) relying on traditional feature-based machine learning using both state of the art and new features. We also explore several deep learning models trained with pre-trained word embeddings as well as contextual embeddings. We then investigate three transfer learning strategies to adapt these models to the crisis domain. We finally experiment with multi-input architectures by incorporating different metadata extra-features to the network. Our deep models, evaluated in random sampling, out-of-event and out-of-type configurations, show very good performances outperforming several competitive baselines. Our results define the first contribution to the field of crisis management in French social media.

The effect of three basic task features on the sensitivity of acceptability judgment tasks

Abstract: Sprouse and Almeida (2017) provide a first systematic investigation of the sensitivity of four acceptability judgment tasks. In this project, we build on these results by decomposing those tasks into three constituent task features (single versus joint presentation, number of response options, and use of response labels), and explore the consequences of those task features on the sensitivity of acceptability judgment experiments. We present 6 additional experiments (for a total of 10) designed to probe the effect of those task features on sensitivity, both independently and in combination. Our results suggest three notable conclusions: (i) there is a clear advantage to joint presentation of theoretically-related sentence types, regardless of the type of response scale used in the experiment; (ii) tasks involving a continuous slider (which have an infinite number of response options, and few labels) offer good sensitivity, while relying solely on spatial reasoning rather than numeric reasoning; and (iii) there are a number of subtle interactions among the three task features that may warrant further investigation. We discuss the potential benefits and concerns of each of these features in detail, along with the relevance of these findings for deciding how to investigate both simple and higher-order acceptability contrasts.

Crowdsourcing Latin American Spanish for Low-Resource Text-to-Speech

Abstract: In this paper we present a multidialectal corpus approach for building a text-to-speech voice for a new dialect in a language with existing resources, focusing on various South American dialects of Spanish. We first present public speech datasets for Argentinian, Chilean, Colombian, Peruvian, Puerto Rican and Venezuelan Spanish specifically constructed with text-to-speech applications in mind using crowd-sourcing. We then compare the monodialectal voices built with minimal data to a multidialectal model built by pooling all the resources from all dialects. Our results show that the multidialectal model outperforms the monodialectal baseline models. We also experiment with a “zero-resource” dialect scenario where we build a multidialectal voice for a dialect while holding out target dialect recordings from the training data.

The Politics of Land Acquisition in Haryana: Managing Dominant Caste Interests in the Name of Development

Abstract: Anti-Special Economic Zone (SEZ) mobilisation in Haryana failed to generate a mass movement. This is despite the political strength of farmers and their deep resentment of the government’s policy t...

Obtaining World Fame from the Periphery

Abstract: Although book translations are overwhelmingly made from English and a small number of other central languages, translations occasionally also flow in the opposite direction, i.e. from peripheral co...

Subjective disorientation as a metacognitive feeling

Abstract: There is a large body of literature on disorientation, ranging from behavioral studies to the analysis of search and rescue operations. However, the subjective side of disorientation remains insuff...

Egalitarian redistribution in the era of hyper-globalization

Abstract: Two traditional theorems of welfare economics posit a trade-off between a government redistribution targets and efficiency. We propose a third ‘claim’ of welfare economics, stating that in closed e...

‘Picking up the neighbours’ waste’: migration of Bangladeshi villagers to India metropolises

Abstract: From a small area of the South West of Bangladesh, thousands of landless farmers cross the border with their family in order to work in the informal sector, mostly in Delhi and Bangalore: a...

Neoliberal exception to liberal democracy? Entrepreneurial territorial governance in India

Abstract: Focusing on the industrial area local authority (IALA), a governance regime widely applied in the south Indian states of Andhra Pradesh and Telangana, this paper examines pathways to illiberal gove...