Countries: France, France, Germany, France, France
Project: EC | MAGIC (649081)
International audience; The Cenozoic strata of the Xining Basin, NE Tibet, have provided crucial records for understanding the tectonic and paleo-environmental evolution of the region. Yet, the age for the lower part of the sedimentary stratigraphy and consequently the early tectonic evolution of the basin remain debated. Here, we present the litho- and magnetostratigraphy of various early Eocene sections throughout the Xining Basin and provide two possible age models independently constrained by the radiometric age of a carbonate bed. Our study extends the dated Eocene stratigraphy down to an unconformity at 53.0 Ma, which is coeval with increased uplift of the nearby Western Qinling Shan and the formation of flexural basins in northern Tibet related to the far-field effects of the India-Asia collision. However, the Paleogene Xining Basin lacks the characteristic features of these foreland basins such as high sedimentation rates and coarsening due to foredeep propagation, which appear only later during the Neogene. Instead, the strata show NW-SE extensional features during the Cretaceous. Here, we propose that this regime persisted until the Paleogene, coeval with Eocene grabens developing further east and related to the subduction of the Pacific Plate. Yet, the rotations and unconformities observed in the Xining Basin strata show that the basin was increasingly affected by the growing Tibetan Plateau throughout the Paleogene and Neogene while experiencing a transition from extension to transpression and/or transtension.
Static word embeddings that represent words by a single vector cannot capture the variability of word meaning in different linguistic and extralinguistic contexts. Building on prior work on contextualized and dynamic word embeddings, we introduce dynamic contextualized word embeddings that represent words as a function of both linguistic and extralinguistic context. Based on a pretrained language model (PLM), dynamic contextualized word embeddings model time and social space jointly, which makes them attractive for a range of NLP tasks involving semantic variability. We highlight potential application scenarios by means of qualitative and quantitative analyses on four English datasets. Comment: ACL 2021
AbstractThe Late Bronze Age (1700–900 BC) represents an extremely dynamic period for Mediterranean Europe. Here, we provide a comparative survey of the archaeological record of over half a millennium within the entire northern littoral of the Mediterranean, from Greece to Iberia, incorporating archaeological, archaeometric, and bioarchaeological evidence. The picture that emerges, while certainly fragmented and not displaying a unique trajectory, reveals a number of broad trends in aspects as different as social organization, trade, transcultural phenomena, and human mobility. The contribution of such trends to the processes that caused the end of the Bronze Age is also examined. Taken together, they illustrate how networks of interaction, ranging from the short to the long range, became a defining aspect of the “Middle Sea” during this time, influencing the lives of the communities that inhabited its northern shore. They also highlight the importance of research that crosses modern boundaries for gaining a better understanding of broad comparable dynamics. Funder: Alma Mater Studiorum - Università di Bologna
Methods for linking individuals across historical data sets, typically in combination with AI based transcription models, are developing rapidly. Probably the single most important identifier for linking is personal names. However, personal names are prone to enumeration and transcription errors and although modern linking methods are designed to handle such challenges, these sources of errors are critical and should be minimized. For this purpose, improved transcription methods and large-scale databases are crucial components. This paper describes and provides documentation for HANA, a newly constructed large-scale database which consists of more than 3.3 million names. The database contain more than 105 thousand unique names with a total of more than 1.1 million images of personal names, which proves useful for transfer learning to other settings. We provide three examples hereof, obtaining significantly improved transcription accuracy on both Danish and US census data. In addition, we present benchmark results for deep learning models automatically transcribing the personal names from the scanned documents. Through making more challenging large-scale databases publicly available we hope to foster more sophisticated, accurate, and robust models for handwritten text recognition.