• Digital Humanities and Cultural Heritage
• Netherlands Organisation for Scientific Research (NWO) close

Isoprenoid glycerol dialkyl glycerol tetraethers (isoGDGTs) are membrane lipids of Archaea. Organic biomarker proxies associated with these lipids, such as the TEX 86 paleothermometer and Branched and Isoprenoid Tetraether (BIT) index, are often used in paleoenvironmental reconstructions for the marine environment, but their general applicability in lacustrine settings is hampered by limited understanding of the biological sources and environmental drivers influencing isoGDGT production. To validate the use of isoGDGT proxies in lakes, we studied the occurrence of isoGDGTs in Lake Chala, a permanently stratified (meromictic) crater lake in equatorial East Africa. We analyzed the abundance and distribution of isoGDGTs in 17 depth profiles of suspended particulate matter (SPM) collected monthly between September 2013 and January 2015, and compared this with the abundance and composition of archaea based on 16S rRNA gene and quantitative PCR analysis. Both isoGDGTs and archaeal abundance in the SPM were exceptionally low throughout the study period. In the oxygenated part of the water column, higher fractional abundances of crenarchaeol are matched by a predominance of the ammonia-oxidizing Thaumarchaeota I.1b that are known to produce this GDGT, whereas deep anoxic water layers are characterized by high fractional abundances of GDGT-0 as well as the anaerobic heterotrophic Group C3 MCG Bathyarchaea and specific euryarchaeotal methanogens. Analysis of intact polar lipid (IPL) isoGDGTs using SPM depth profiles from three months representing distinct seasons during the study period revealed the presence of several IPLs of GDGT-0 in the anoxic lower water column, which are rarely found in natural settings. IPL GDGT-0 with a phosphatidylglycerol (PG-), monohexosephosphatidylglycerol (MH-PG-) and dihexose-phosphatidylglycerol (DH-PG-) head-group was typically only present just above the lake bottom at 90 m depth and probably reflect specific communities of anaerobic archaea. We also determined the flux and distribution of isoGDGTs in settling particles collected monthly between November 2006 and January 2015 from a sediment trap suspended at 35 m water depth to assess seasonal and inter-annual variability in surface-water isoGDGT production, and compared this with the temporal distribution of isoGDGTs in the 25,000-year long sediment record from Lake Chala. Monthly variation of isoGDGTs in the 98-month settling-particles record did not show a strong annual pattern related to seasonal water-column mixing and stratification, likely because the oxycline was regularly situated below sediment-trap depth. Episodes of high GDGT-0 concentrations relative to crenarchaeol in the settling particles can therefore be linked to periods of exceptionally shallow oxycline depth, which suppresses the thaumarchaeotal bloom. During such intervals, TEX86 based paleotemperatures are not reliable because isoGDGT input from other archaeal sources disproportionally influences TEX86 values and creates a cold-temperature bias. Additionally, the abundance of the crenarchaeol isomer relative to crenarchaeol (f[CREN]) gradually increases during such episodes of high GDGT-0/crenarchaeol ratio, suggesting increasing dominance of Group I.1b over Group I.1a Thaumarchaeota, and might prove a good marker for prolonged shallow-oxycline conditions. Most importantly, the associated near-absence of crenarchaeol during times of strong upper-water-column stratification results in high BIT-index values. We propose that this suppression mechanism may be the principal driver of BIT-index variation in the sediment record of Lake Chala, and the main source of observed congruence between the BIT index and climate-driven lake-level variation on long time scales. (C) 2021 The Authors. Published by Elsevier Ltd.

Identifying factors that make certain languages harder to model than others is essential to reach language equality in future Natural Language Processing technologies. Free-order case-marking languages, such as Russian, Latin or Tamil, have proved more challenging than fixed-order languages for the tasks of syntactic parsing and subject-verb agreement prediction. In this work, we investigate whether this class of languages is also more difficult to translate by state-of-the-art Neural Machine Translation models (NMT). Using a variety of synthetic languages and a newly introduced translation challenge set, we find that word order flexibility in the source language only leads to a very small loss of NMT quality, even though the core verb arguments become impossible to disambiguate in sentences without semantic cues. The latter issue is indeed solved by the addition of case marking. However, in medium- and low-resource settings, the overall NMT quality of fixed-order languages remains unmatched. Comment: Accepted to TACL, pre-MIT Press publication version

Natural language is compositional; the meaning of a sentence is a function of the meaning of its parts. This property allows humans to create and interpret novel sentences, generalizing robustly outside their prior experience. Neural networks have been shown to struggle with this kind of generalization, in particular performing poorly on tasks designed to assess compositional generalization (i.e. where training and testing distributions differ in ways that would be trivial for a compositional strategy to resolve). Their poor performance on these tasks may in part be due to the nature of supervised learning which assumes training and testing data to be drawn from the same distribution. We implement a meta-learning augmented version of supervised learning whose objective directly optimizes for out-of-distribution generalization. We construct pairs of tasks for meta-learning by sub-sampling existing training data. Each pair of tasks is constructed to contain relevant examples, as determined by a similarity metric, in an effort to inhibit models from memorizing their input. Experimental results on the COGS and SCAN datasets show that our similarity-driven meta-learning can improve generalization performance. ACL2021 Camera Ready; fix a small typo

Abstract In this article, we compare representations of the pilgrimage to Mecca posted on Facebook and Youtube by ‘ordinary’ pilgrims from Morocco and (semi)professional bloggers of Moroccan parentage in the Netherlands. We discuss how such posts challenge representations that circulate in the mainstream media in both countries about Islam in general and the hajj in particular. For Morocco we demonstrate that this kind of digital mediation of pilgrimage contests the ways in which the state-organized hajj is framed in Morocco’s national media. For the Netherlands, we argue that bloggers deconstruct dominant images of the Muslim ‘other’ in their self-presentations as specifically Dutch Muslim pilgrims by connecting the meanings they attribute to the pilgrimage to Mecca to universal issues.

In a seminal study, Bott & Noveck (2004) found that the computation of the scalar inference of ‘some’ implying ‘not all’ was associated with increased sentence verification times, suggesting a processing cost. Recently, van Tiel and colleagues (2019b) hypothesised that the presence of this processing cost critically depends on the polarity of the scalar word. We comprehensively evaluated this polarity hypothesis on the basis of a sentence-picture verification task in which we tested the processing of 16 types of adjectival scalar inferences. We develop a quantitative measure of adjectival polarity which combines insights from linguistics and psychology. In line with the polarity hypothesis, our measure of polarity reliably predicted the presence or absence of a processing cost (i.e., an increase in sentence verification times). We conclude that the alleged processing cost for scalar inferencing in verification tasks is not due to the process of drawing a scalar inference, but rather to the cognitive difficulty of verifying negative information.

Materials In Paintings (MIP): An interdisciplinary dataset for perception, art history, and computer vision.Download the README.txt first to help you decide what you want/need to download!In this dataset, we capture the painterly depictions of materials to enable the study of depiction and perception of materials through the artists' eye. We annotated a dataset of 19k paintings with 200k+ bounding boxes from which polygon segments were automatically extracted. Each bounding box was assigned a coarse label (e.g., fabric) and a fine-grained label (e.g., velvety, silky).Note that the data can be browed and explored on https://materialsinpaintings.tudelft.nl. If you only want to download a few paintings, using that website might be faster.

Abstract The sociological literature on social networks overwhelmingly considers the number of core social contacts. Social networks, however, reach far beyond this small number of social ties. We know little about individual variation in the size of such extended social networks. In this study, we move beyond core networks and explain individual variation in the extended social network size among youth. We use survey data of Dutch adolescents (N = 5,921) and use two state-of-the-art measurements to compute extended network sizes: network scale-up methods through Bayesian modeling and the observed number of contacts on Facebook. Among both measurements, we find that extended networks are larger among ethnic majority members, girls, and those who often engage in social foci. This highlights a crucial role for preferences and opportunity in the genesis of extended networks. Additionally, we find that differences between both network sizes (scale-up and Facebook) are smaller for girls and higher educated. We discuss the implications of these findings and suggest directions for future research.

pIn the Pliocene Model Intercomparison Project Phase 2 (PlioMIP2), coupled climate models have been used to simulate an interglacial climate during the mid-Piacenzian warm period (mPWP; 3.264 to 3.025 span classCombining double low line"inline-formula"Ma/span). Here, we compare the Atlantic Meridional Overturning Circulation (AMOC), poleward ocean heat transport and sea surface warming in the Atlantic simulated with these models. In PlioMIP2, all models simulate an intensified mid-Pliocene AMOC. However, there is no consistent response in the simulated Atlantic ocean heat transport nor in the depth of the Atlantic overturning cell. The models show a large spread in the simulated AMOC maximum, the Atlantic ocean heat transport and the surface warming in the North Atlantic. Although a few models simulate a surface warming of span classCombining double low line"inline-formula"g1/4/span 8-12 span classCombining double low line"inline-formula"g C/span in the North Atlantic, similar to the reconstruction from Pliocene Research, Interpretation and Synoptic Mapping (PRISM) version 4,span idCombining double low line"page530"/ most models appear to underestimate this warming. The large model spread and model-data discrepancies in the PlioMIP2 ensemble do not support the hypothesis that an intensification of the AMOC, together with an increase in northward ocean heat transport, is the dominant mechanism for the mid-Pliocene warm climate over the North Atlantic./p.

This study addresses the question whether visually grounded speech recognition (VGS) models learn to capture sentence semantics without access to any prior linguistic knowledge. We produce synthetic and natural spoken versions of a well known semantic textual similarity database and show that our VGS model produces embeddings that correlate well with human semantic similarity judgements. Our results show that a model trained on a small image-caption database outperforms two models trained on much larger databases, indicating that database size is not all that matters. We also investigate the importance of having multiple captions per image and find that this is indeed helpful even if the total number of images is lower, suggesting that paraphrasing is a valuable learning signal. While the general trend in the field is to create ever larger datasets to train models on, our findings indicate other characteristics of the database can just as important important. Comment: This paper has been accepted at Interspeech 2021 where it will be presented and appear in the conference proceedings in September 2021

The Aral Sea basin in Central Asia and its major rivers, the Amu Darya and Syr Darya, were the center of advanced river civilizations, and a principal hub of the Silk Roads over a period of more than 2,000 y. The region's decline has been traditionally attributed to the devastating Mongol invasion of the early-13th century CE. However, the role of changing hydroclimatic conditions on the development of these culturally influential potamic societies has not been the subject of modern geoarchaeological investigations. In this paper we report the findings of an interdisciplinary investigation of archaeological sites and associated irrigation canals of the Otrār oasis, a United Nations Educational, Scientific and Cultural Organization (UNESCO) World Heritage site located at the confluence of the Syr Darya and Arys rivers in southern Kazakhstan. This includes radiometric dating of irrigation canal abandonment and an investigation of Arys river channel dynamics. Major phases of fluvial aggradation, between the seventh and early ninth century CE and between 1350 and 1550 CE coincide with economic flourishing of the oasis, facilitated by wet climatic conditions and higher river flows that favored floodwater farming. Periods of abandonment of the irrigation network and cultural decline primarily correlate with fluvial entrenchment during periods of drought, instead of being related to destructive invasions. Therefore, it seems the great rivers of Central Asia were not just static "stage sets" for some of the turning points of world history, but in many instances, inadvertently or directly shaped the final outcomes and legacies of imperial ambitions in the region.