• Digital Humanities and Cultural Heritage
• Publicationsclose
• Other research productsclose
• Preprint close
• Netherlands Organisation for Scientific Research (NWO) close
• NARCIS close

Semantic representations have long been argued as potentially useful for enforcing meaning preservation and improving generalization performance of machine translation methods. In this work, we are the first to incorporate information about predicate-argument structure of source sentences (namely, semantic-role representations) into neural machine translation. We use Graph Convolutional Networks (GCNs) to inject a semantic bias into sentence encoders and achieve improvements in BLEU scores over the linguistic-agnostic and syntax-aware versions on the English--German language pair.

Density heterogeneities are the source of mass transport in the Earth. However, the 3-D density structure remains poorly constrained because travel times of seismic waves are only weakly sensitive to density. Inspired by recent developments in seismic waveform tomography, we investigate whether the visibility of 3-D density heterogeneities may be improved by inverting not only travel times of specific seismic phases but complete seismograms. As a first step in this direction, we perform numerical experiments to estimate the effect of 3-D crustal density heterogeneities on regional seismic wave propagation. While a finite number of numerical experiments may not capture the full range of possible scenarios, our results still indicate that realistic crustal density variations may lead to travel-time shifts of up to ∼ 1s and amplitude variations of several tens of percent over propagation distances of ∼ 1000km. Both amplitude and travel-time variations increase with increasing epicentral distance and increasing medium complexity, i.e. decreasing correlation length of the heterogeneities. They are practically negligible when the correlation length of the heterogeneities is much larger than the wavelength. However, when the correlation length approaches the wavelength, density-induced waveform perturbations become prominent. Recent regional-scale full-waveform inversions that resolve structure at the scale of a wavelength already reach this regime. Our numerical experiments suggest that waveform perturbations induced by realistic crustal density variations can be observed in high-quality regional seismic data. While density-induced travel-time differences will often be small, amplitude variations exceeding ±10% are comparable to those induced by 3-D velocity structure and attenuation. While these results certainly encourage more research on the development of 3-D density tomography, they also suggest that current full-waveform inversions that use amplitude information may be biased due to the neglect of 3-D variations in density. Solid Earth, 7 (6) ISSN:1869-9510 ISSN:1869-9529

Cosmic particles hitting Earth's moon produce radio emission via the Askaryan effect. If the resulting radio ns-pulse can be detected by radio telescopes, this technique potentially increases the available collective area for ZeV scale particles by several orders of magnitude compared to current experiments. The LOw Frequency ARray (LOFAR) is the largest radio telescope operating in the optimum frequency regime for this technique. In this contribution, we report on the status of the implementation of the lunar detection mode at LOFAR. Proceedings of the 26th Extended European Cosmic Ray Symposium (ECRS), Barnaul/Belokurikha, 2018

Changeover from a glacial to an interglacial climate is considered as transitional between two stable modes. Palaeoceanographic reconstructions using the polar foraminifera Neogloboquadrina pachyderma highlight the retreat of the Polar Front during the last deglaciation in terms of both its decreasing abundance and stable oxygen isotope values (δ18O) in sediment cores. While conventional isotope analysis of pooled N. pachyderma and G. bulloides shells shows a warming trend concurrent with the retreating ice, new single-shell measurements reveal that this trend is composed of two isotopically different populations that are morphologically indistinguishable. Using modern time series as analogues for interpreting downcore data, glacial productivity in the mid-North Atlantic appears limited to a single maximum in late summer, followed by the melting of drifting icebergs and winter sea ice. Despite collapsing ice sheets and global warming during the deglaciation, a second “warm” population of N. pachyderma appears in a bimodal seasonal succession, separated by the subpolar G. bulloides. This represents a shift in the timing of the main plankton bloom from late to early summer in a “deglacial” intermediate mode that persisted from the glacial maximum until the start of the Holocene. When seawater temperatures exceeded the threshold values, first the “cold” (glacial) then the “warm” (deglacial) populations of N. pachyderma disappeared, whilst G. bulloides with a greater tolerance to higher temperatures persisted throughout the Holocene to the present day in the midlatitude North Atlantic. Single-specimen δ18O of polar N. pachyderma reveals a steeper rate of ocean warming during the last deglaciation than appears from conventional pooled δ18O average values.

From the paper "Formality Conjecture" (Ascona 1996): "I am aware of only one such a class, it corresponds to simplest good graph, the complete graph with $4$ vertices $($and $6$ edges$)$. This class gives a remarkable vector field on the space of bi-vector fields on $\mathbb{R}^{d}$. The evolution with respect to the time $t$ is described by the following non-linear partial differential equation: ..., where $\alpha=\sum_{i,j}\alpha_{ij} {\partial}/{\partial x_{i}}\wedge {\partial}/{\partial x_{j}}$ is a bi-vector field on $\mathbb{R}^d$. It follows from general properties of cohomology that $1)$ this evolution preserves the class of $($real-analytic$)$ Poisson structures}, ... In fact, I cheated a little bit. In the formula for the vector field on the space of bivector fields which one get from the tetrahedron graph, an additional term is present. ... It is possible to prove formally that if $\alpha$ is a Poisson bracket, i.e. if $[{\alpha,\alpha}]=0\in T^2(\mathbb{R}^d)$, then the additional term shown above vanishes." By using twelve Poisson structures with high-degree polynomial coefficients as explicit counterexamples, we show that both the above claims are false: neither does the first flow preserve the property of bi-vectors to be Poisson nor does the second flow vanish identically at Poisson bi-vectors. The counterexamples at hand suggest a correction to the formula for the "exotic" flow on the space of Poisson bi-vectors; in fact, this flow is encoded by the balanced sum involving both the Kontsevich tetrahedral graphs (that give rise to the flows mentioned above). We reveal that it is only the balance $1:6$ for which the flow does preserve the space of Poisson bi-vectors. Comment: Talks given in parallel by A.B. at GADEIS VIII workshop (12--16 June 2016, Larnaca, Cyprus) and A.K. at ISQS'24 conference (13--19 June 2016, CVUT Prague, Czech Republic), 10 pages, 2 figures, 4 tables

The quality of a Neural Machine Translation system depends substantially on the availability of sizable parallel corpora. For low-resource language pairs this is not the case, resulting in poor translation quality. Inspired by work in computer vision, we propose a novel data augmentation approach that targets low-frequency words by generating new sentence pairs containing rare words in new, synthetically created contexts. Experimental results on simulated low-resource settings show that our method improves translation quality by up to 2.9 BLEU points over the baseline and up to 3.2 BLEU over back-translation. Comment: 5 pages, 2 figures, Accepted at ACL 2017

Recent studies have shown that the Early Eocene Climatic Optimum (EECO) was preceded by a series of short-lived global warming events, known as hyperthermals. Here we present high-resolution benthic stable carbon and oxygen isotope records from ODP Sites 1262 and 1263 (Walvis Ridge, SE Atlantic) between ~ 54 and ~ 52 million years ago, tightly constraining the character, timing, and magnitude of six prominent hyperthermal events. These events, which include Eocene Thermal Maximum (ETM) 2 and 3, are studied in relation to orbital forcing and long-term trends. Our findings reveal an almost linear relationship between δ13C and δ18O for all these hyperthermals, indicating that the eccentricity-paced covariance between deep-sea temperature changes and extreme perturbations in the exogenic carbon pool persisted during these events towards the onset of the EECO, in accordance with previous observations for the Paleocene Eocene Thermal Maximum (PETM) and ETM2. The covariance of δ13C and δ18O during H2 and I2, which are the second pulses of the "paired" hyperthermal events ETM2-H2 and I1-I2, deviates with respect to the other events. We hypothesize that this could relate to a relatively higher contribution of an isotopically heavier source of carbon, such as peat or permafrost, and/or to climate feedbacks/local changes in circulation. Finally, the δ18O records of the two sites show a systematic offset with on average 0.2 ‰ heavier values for the shallower Site 1263, which we link to a slightly heavier isotopic composition of the intermediate water mass reaching the northeastern flank of the Walvis Ridge compared to that of the deeper northwestern water mass at Site 1262.

We present our first efforts towards building a single multilingual automatic speech recognition (ASR) system that can process code-switching (CS) speech in five languages spoken within the same population. This contrasts with related prior work which focuses on the recognition of CS speech in bilingual scenarios. Recently, we have compiled a small five-language corpus of South African soap opera speech which contains examples of CS between 5 languages occurring in various contexts such as using English as the matrix language and switching to other indigenous languages. The ASR system presented in this work is trained on 4 corpora containing English-isiZulu, English-isiXhosa, English-Setswana and English-Sesotho CS speech. The interpolation of multiple language models trained on these language pairs enables the ASR system to hypothesize mixed word sequences from these 5 languages. We evaluate various state-of-the-art acoustic models trained on this 5-lingual training data and report ASR accuracy and language recognition performance on the development and test sets of the South African multilingual soap opera corpus. Acccepted for publication at Interspeech 2018

Abstract. Recent studies based on trace gas mixing ratios in ice cores and charcoal data indicate that biomass burning emissions over the past millennium exceeded contemporary emissions by up to a factor of 4 for certain time periods. This is surprising because various sources of biomass burning are linked with population density, which has increased over the past centuries. We have analysed how emissions from several landscape biomass burning sources could have fluctuated to yield emissions that are in correspondence with recent results based on ice core mixing ratios of carbon monoxide (CO) and its isotopic signature measured at South Pole station (SPO). Based on estimates of contemporary landscape fire emissions and the TM5 chemical transport model driven by present-day atmospheric transport and OH concentrations, we found that CO mixing ratios at SPO are more sensitive to emissions from South America and Australia than from Africa, and are relatively insensitive to emissions from the Northern Hemisphere. We then explored how various landscape biomass burning sources may have varied over the past centuries and what the resulting emissions and corresponding CO mixing ratio at SPO would be, using population density variations to reconstruct sources driven by humans (e.g., fuelwood burning) and a new model to relate savanna emissions to changes in fire return times. We found that to match the observed ice core CO data, all savannas in the Southern Hemisphere had to burn annually, or bi-annually in combination with deforestation and slash and burn agriculture exceeding current levels, despite much lower population densities and lack of machinery to aid the deforestation process. While possible, these scenarios are unlikely and in conflict with current literature. However, we do show the large potential for increased emissions from savannas in a pre-industrial world. This is mainly because in the past, fuel beds were probably less fragmented compared to the current situation; satellite data indicates that the majority of savannas have not burned in the past 10 yr, even in Africa, which is considered "the burning continent". Although we have not considered increased charcoal burning or changes in OH concentrations as potential causes for the elevated CO concentrations found at SPO, it is unlikely they can explain the large increase found in the CO concentrations in ice core data. Confirmation of the CO ice core data would therefore call for radical new thinking about causes of variable global fire rates over recent centuries.

The LOFAR radio telescope is able to measure the radio emission from cosmic ray induced air showers with hundreds of individual antennas. This allows for precision testing of the emission mechanisms for the radio signal as well as determination of the depth of shower maximum $X_{\max}$, the shower observable most sensitive to the mass of the primary cosmic ray, to better than 20 g/cm$^2$. With a densely instrumented circular area of roughly 320 m$^2$, LOFAR is targeting for cosmic ray astrophysics in the energy range $10^{16}$ - $10^{18}$ eV. In this contribution we give an overview of the status, recent results, and future plans of cosmic ray detection with the LOFAR radio telescope. Comment: Proceedings of the 26th Extended European Cosmic Ray Symposium (ECRS), Barnaul/Belokurikha, 2018