Advanced search in Research products
Research products
arrow_drop_down
Searching FieldsTerms
Any field
arrow_drop_down
includes
arrow_drop_down
Include:
The following results are related to Digital Humanities and Cultural Heritage. Are you interested to view more results? Visit OpenAIRE - Explore.
60 Research products, page 1 of 6

  • Digital Humanities and Cultural Heritage
  • Publications
  • Other research products
  • Preprint
  • Netherlands Organisation for Scientific Research (NWO)

10
arrow_drop_down
Date (most recent)
arrow_drop_down
  • Publication . Article . Conference object . Preprint . 2021
    Open Access
    Authors: 
    Danny Merkx; Stefan L. Frank; Mirjam Ernestus;
    Country: Netherlands
    Project: NWO | Language in Interaction (2300176475)

    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

  • Publication . Conference object . Article . Preprint . 2020 . Embargo End Date: 01 Jan 2020
    Open Access
    Authors: 
    Oskar van der Wal; Silvan de Boer; Elia Bruni; Dieuwke Hupkes;
    Publisher: arXiv
    Project: NWO | Language in Interaction (2300176475), EC | MAGIC (790369)

    In this paper, we consider the syntactic properties of languages emerged in referential games, using unsupervised grammar induction (UGI) techniques originally designed to analyse natural language. We show that the considered UGI techniques are appropriate to analyse emergent languages and we then study if the languages that emerge in a typical referential game setup exhibit syntactic structure, and to what extent this depends on the maximum message length and number of symbols that the agents are allowed to use. Our experiments demonstrate that a certain message length and vocabulary size are required for structure to emerge, but they also illustrate that more sophisticated game scenarios are required to obtain syntactic properties more akin to those observed in human language. We argue that UGI techniques should be part of the standard toolkit for analysing emergent languages and release a comprehensive library to facilitate such analysis for future researchers. Comment: Accepted at EMNLP 2020

  • Publication . Article . Preprint . 2020
    Open Access English
    Authors: 
    Lisa Beinborn; Rochelle Choenni;
    Publisher: The MIT Press
    Country: Netherlands
    Project: NWO | Language in Interaction (2300176475)

    Multilingual representations have mostly been evaluated based on their performance on specific tasks. In this article, we look beyond engineering goals and analyze the relations between languages in computational representations. We introduce a methodology for comparing languages based on their organization of semantic concepts. We propose to conduct an adapted version of representational similarity analysis of a selected set of concepts in computational multilingual representations. Using this analysis method, we can reconstruct a phylogenetic tree that closely resembles those assumed by linguistic experts. These results indicate that multilingual distributional representations which are only trained on monolingual text and bilingual dictionaries preserve relations between languages without the need for any etymological information. In addition, we propose a measure to identify semantic drift between language families. We perform experiments on word-based and sentence-based multilingual models and provide both quantitative results and qualitative examples. Analyses of semantic drift in multilingual representations can serve two purposes: they can indicate unwanted characteristics of the computational models and they provide a quantitative means to study linguistic phenomena across languages. The code is available at https://github.com/beinborn/SemanticDrift. Almost final version. Paper will appear in the Computational Linguistics Journal, Volume 46, Issue 3

  • Publication . Conference object . Preprint . Article . 2020 . Embargo End Date: 01 Jan 2020
    Open Access
    Authors: 
    Mathieu Dehouck; Mark Anderson; Carlos Gómez-Rodríguez;
    Publisher: arXiv
    Project: NWO | MicroDish BV - A Rapid Gr... (2300150313), NWO | Nutrients to modulate obe... (2300171675), EC | FASTPARSE (714150)

    We present the system submission from the FASTPARSE team for the EUD Shared Task at IWPT 2020. We engaged with the task by focusing on efficiency. For this we considered training costs and inference efficiency. Our models are a combination of distilled neural dependency parsers and a rule-based system that projects UD trees into EUD graphs. We obtained an average ELAS of 74.04 for our official submission, ranking 4th overall. Comment: To published in the proceedings of the IWPT 2020 Shared Task on Parsing into Enhanced Universal Dependencies

  • Publication . Preprint . Conference object . Article . 2020
    Open Access
    Authors: 
    Marzieh Fadaee; Christof Monz;
    Publisher: Association for Computational Linguistics
    Project: NWO | Constraint-Based Language... (2300178358), NWO | Surface Realization in St... (2300172930)

    Recent works have shown that Neural Machine Translation (NMT) models achieve impressive performance, however, questions about understanding the behavior of these models remain unanswered. We investigate the unexpected volatility of NMT models where the input is semantically and syntactically correct. We discover that with trivial modifications of source sentences, we can identify cases where \textit{unexpected changes} happen in the translation and in the worst case lead to mistranslations. This volatile behavior of translating extremely similar sentences in surprisingly different ways highlights the underlying generalization problem of current NMT models. We find that both RNN and Transformer models display volatile behavior in 26% and 19% of sentence variations, respectively. Accepted to Neural Generation and Translation Workshop (WNGT) at ACL 2020

  • Open Access English
    Authors: 
    Van Der Boon, Annique; Kuiper, Klaudia F.; Van Der Ploeg, Robin; Cramwinckel, Margot J.; Honarmand, Maryam; Sluijs, Appy; Krijgsman, Wout; Paleomagnetism; Marine palynology and palaeoceanography; Paleomagnetism; +1 more
    Country: Netherlands
    Project: EC | SPANC (771497), NWO | The evolution of Parateth... (2300163618)

    The Middle Eocene Climatic Optimum (MECO), a ∼500 kyr episode of global warming that initiated at ∼ 40.5 Ma, is postulated to be driven by a net increase in volcanic carbon input, but a direct source has not been identified. Here we show, based on new and previously published radiometric ages of volcanic rocks, that the interval spanning the MECO corresponds to a massive increase in continental arc volcanism in Iran and Azerbaijan. Ages of Eocene igneous rocks in all volcanic provinces of Iran cluster around 40 Ma, very close to the peak warming phase of the MECO. Based on the spatial extent and volume of the volcanic rocks as well as the carbonaceous lithology in which they are emplaced, we estimate the total amount of CO2 that could have been released at this time corresponds to between 1052 and 12 565 Pg carbon. This is compatible with the estimated carbon release during the MECO. Although the uncertainty in both individual ages, and the spread in the compilation of ages, is larger than the duration of the MECO, a flare-up in Neotethys subduction zone volcanism represents a plausible excess carbon source responsible for MECO warming.

  • Publication . Article . Preprint . 2020
    Open Access
    Authors: 
    M Saiful Bari; Shafiq Joty; Prathyusha Jwalapuram;
    Publisher: Association for the Advancement of Artificial Intelligence (AAAI)
    Project: NWO | Development of antimicrob... (2300190139)

    Recently, neural methods have achieved state-of-the-art (SOTA) results in Named Entity Recognition (NER) tasks for many languages without the need for manually crafted features. However, these models still require manually annotated training data, which is not available for many languages. In this paper, we propose an unsupervised cross-lingual NER model that can transfer NER knowledge from one language to another in a completely unsupervised way without relying on any bilingual dictionary or parallel data. Our model achieves this through word-level adversarial learning and augmented fine-tuning with parameter sharing and feature augmentation. Experiments on five different languages demonstrate the effectiveness of our approach, outperforming existing models by a good margin and setting a new SOTA for each language pair.

  • Publication . Article . Preprint . Other literature type . 2020
    Open Access English
    Authors: 
    G.-J. A. Brummer; G.-J. A. Brummer; B. Metcalfe; B. Metcalfe; W. Feldmeijer; W. Feldmeijer; M. A. Prins; J. van 't Hoff; J. van 't Hoff; G. M. Ganssen;
    Publisher: HAL CCSD
    Countries: Netherlands, France, France
    Project: NWO | SCAN-2: Scanning Sediment... (2300165588), ANR | L-IPSL (ANR-10-LABX-0018)

    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.

  • Publication . Article . Conference object . Preprint . 2020 . Embargo End Date: 01 Jan 2020
    Open Access
    Authors: 
    Robert Litschko; Ivan Vulić; Željko Agić; Goran Glavaš;
    Publisher: arXiv
    Country: United Kingdom
    Project: EC | LEXICAL (648909), NWO | MicroDish BV - A Rapid Gr... (2300150313), NWO | Inorganic porous hollow f... (2300175491)

    Current methods of cross-lingual parser transfer focus on predicting the best parser for a low-resource target language globally, that is, “at treebank level”. In this work, we propose and argue for a novel cross-lingual transfer paradigm: instance-level parser selection (ILPS), and present a proof-of-concept study focused on instance-level selection in the framework of delexicalized parser transfer. Our work is motivated by an empirical observation that different source parsers are the best choice for different Universal POS-sequences (i.e., UPOS sentences) in the target language. We then propose to predict the best parser at the instance level. To this end, we train a supervised regression model, based on the Transformer architecture, to predict parser accuracies for individual POS-sequences. We compare ILPS against two strong single-best parser selection baselines (SBPS): (1) a model that compares POS n-gram distributions between the source and target languages (KL) and (2) a model that selects the source based on the similarity between manually created language vectors encoding syntactic properties of languages (L2V). The results from our extensive evaluation, coupling 42 source parsers and 20 diverse low-resource test languages, show that ILPS outperforms KL and L2V on 13/20 and 14/20 test languages, respectively. Further, we show that by predicting the best parser “at treebank level” (SBPS), using the aggregation of predictions from our instance-level model, we outperform the same baselines on 17/20 and 16/20 test languages.

  • Publication . Other literature type . Article . Preprint . 2019
    Open Access English
    Authors: 
    N. Blom; N. Blom; N. Blom; A. Gokhberg; A. Fichtner;
    Countries: Switzerland, United Kingdom
    Project: EC | CSEM (714069), NWO | Full waveform inversion f... (2300172344)

    We present a seismic waveform tomography of the upper mantle beneath the central and eastern Mediterranean down to the mantle transition zone. Our methodology incorporates in a consistent manner the information from body and multimode surface waves, source effects, frequency dependence, wavefront healing, anisotropy and attenuation. This allows us to jointly image multiple parameters of the crust and upper mantle. Based on the data from ∼ 17 000 unique source–receiver pairs, gathered from 80 earthquakes, we image radially anisotropic S velocity, P velocity and density. We use a multi-scale approach in which the longest periods (100–150 s) are inverted first, broadening to a period band of 28–150 s. Thanks to a strategy that combines long-period signals and a separation of body and surface wave signals, we are able to image down to the mantle transition zone in most of the model domain. Our model shows considerable detail in especially the northern part of the domain, where data coverage is very dense, and displays a number of clear and coherent high-velocity structures across the domain that can be linked to episodes of current and past subduction. These include the Hellenic subduction zone, the Cyprus subduction zone and high-velocity anomalies beneath the Italian peninsula and the Dinarides. This model is able to explain data from new events that were not included in the inversion. Solid Earth, 11 (2) ISSN:1869-9510 ISSN:1869-9529

Advanced search in Research products
Research products
arrow_drop_down
Searching FieldsTerms
Any field
arrow_drop_down
includes
arrow_drop_down
Include:
The following results are related to Digital Humanities and Cultural Heritage. Are you interested to view more results? Visit OpenAIRE - Explore.
60 Research products, page 1 of 6
  • Publication . Article . Conference object . Preprint . 2021
    Open Access
    Authors: 
    Danny Merkx; Stefan L. Frank; Mirjam Ernestus;
    Country: Netherlands
    Project: NWO | Language in Interaction (2300176475)

    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

  • Publication . Conference object . Article . Preprint . 2020 . Embargo End Date: 01 Jan 2020
    Open Access
    Authors: 
    Oskar van der Wal; Silvan de Boer; Elia Bruni; Dieuwke Hupkes;
    Publisher: arXiv
    Project: NWO | Language in Interaction (2300176475), EC | MAGIC (790369)

    In this paper, we consider the syntactic properties of languages emerged in referential games, using unsupervised grammar induction (UGI) techniques originally designed to analyse natural language. We show that the considered UGI techniques are appropriate to analyse emergent languages and we then study if the languages that emerge in a typical referential game setup exhibit syntactic structure, and to what extent this depends on the maximum message length and number of symbols that the agents are allowed to use. Our experiments demonstrate that a certain message length and vocabulary size are required for structure to emerge, but they also illustrate that more sophisticated game scenarios are required to obtain syntactic properties more akin to those observed in human language. We argue that UGI techniques should be part of the standard toolkit for analysing emergent languages and release a comprehensive library to facilitate such analysis for future researchers. Comment: Accepted at EMNLP 2020

  • Publication . Article . Preprint . 2020
    Open Access English
    Authors: 
    Lisa Beinborn; Rochelle Choenni;
    Publisher: The MIT Press
    Country: Netherlands
    Project: NWO | Language in Interaction (2300176475)

    Multilingual representations have mostly been evaluated based on their performance on specific tasks. In this article, we look beyond engineering goals and analyze the relations between languages in computational representations. We introduce a methodology for comparing languages based on their organization of semantic concepts. We propose to conduct an adapted version of representational similarity analysis of a selected set of concepts in computational multilingual representations. Using this analysis method, we can reconstruct a phylogenetic tree that closely resembles those assumed by linguistic experts. These results indicate that multilingual distributional representations which are only trained on monolingual text and bilingual dictionaries preserve relations between languages without the need for any etymological information. In addition, we propose a measure to identify semantic drift between language families. We perform experiments on word-based and sentence-based multilingual models and provide both quantitative results and qualitative examples. Analyses of semantic drift in multilingual representations can serve two purposes: they can indicate unwanted characteristics of the computational models and they provide a quantitative means to study linguistic phenomena across languages. The code is available at https://github.com/beinborn/SemanticDrift. Almost final version. Paper will appear in the Computational Linguistics Journal, Volume 46, Issue 3

  • Publication . Conference object . Preprint . Article . 2020 . Embargo End Date: 01 Jan 2020
    Open Access
    Authors: 
    Mathieu Dehouck; Mark Anderson; Carlos Gómez-Rodríguez;
    Publisher: arXiv
    Project: NWO | MicroDish BV - A Rapid Gr... (2300150313), NWO | Nutrients to modulate obe... (2300171675), EC | FASTPARSE (714150)

    We present the system submission from the FASTPARSE team for the EUD Shared Task at IWPT 2020. We engaged with the task by focusing on efficiency. For this we considered training costs and inference efficiency. Our models are a combination of distilled neural dependency parsers and a rule-based system that projects UD trees into EUD graphs. We obtained an average ELAS of 74.04 for our official submission, ranking 4th overall. Comment: To published in the proceedings of the IWPT 2020 Shared Task on Parsing into Enhanced Universal Dependencies

  • Publication . Preprint . Conference object . Article . 2020
    Open Access
    Authors: 
    Marzieh Fadaee; Christof Monz;
    Publisher: Association for Computational Linguistics
    Project: NWO | Constraint-Based Language... (2300178358), NWO | Surface Realization in St... (2300172930)

    Recent works have shown that Neural Machine Translation (NMT) models achieve impressive performance, however, questions about understanding the behavior of these models remain unanswered. We investigate the unexpected volatility of NMT models where the input is semantically and syntactically correct. We discover that with trivial modifications of source sentences, we can identify cases where \textit{unexpected changes} happen in the translation and in the worst case lead to mistranslations. This volatile behavior of translating extremely similar sentences in surprisingly different ways highlights the underlying generalization problem of current NMT models. We find that both RNN and Transformer models display volatile behavior in 26% and 19% of sentence variations, respectively. Accepted to Neural Generation and Translation Workshop (WNGT) at ACL 2020

  • Open Access English
    Authors: 
    Van Der Boon, Annique; Kuiper, Klaudia F.; Van Der Ploeg, Robin; Cramwinckel, Margot J.; Honarmand, Maryam; Sluijs, Appy; Krijgsman, Wout; Paleomagnetism; Marine palynology and palaeoceanography; Paleomagnetism; +1 more
    Country: Netherlands
    Project: EC | SPANC (771497), NWO | The evolution of Parateth... (2300163618)

    The Middle Eocene Climatic Optimum (MECO), a ∼500 kyr episode of global warming that initiated at ∼ 40.5 Ma, is postulated to be driven by a net increase in volcanic carbon input, but a direct source has not been identified. Here we show, based on new and previously published radiometric ages of volcanic rocks, that the interval spanning the MECO corresponds to a massive increase in continental arc volcanism in Iran and Azerbaijan. Ages of Eocene igneous rocks in all volcanic provinces of Iran cluster around 40 Ma, very close to the peak warming phase of the MECO. Based on the spatial extent and volume of the volcanic rocks as well as the carbonaceous lithology in which they are emplaced, we estimate the total amount of CO2 that could have been released at this time corresponds to between 1052 and 12 565 Pg carbon. This is compatible with the estimated carbon release during the MECO. Although the uncertainty in both individual ages, and the spread in the compilation of ages, is larger than the duration of the MECO, a flare-up in Neotethys subduction zone volcanism represents a plausible excess carbon source responsible for MECO warming.

  • Publication . Article . Preprint . 2020
    Open Access
    Authors: 
    M Saiful Bari; Shafiq Joty; Prathyusha Jwalapuram;
    Publisher: Association for the Advancement of Artificial Intelligence (AAAI)
    Project: NWO | Development of antimicrob... (2300190139)

    Recently, neural methods have achieved state-of-the-art (SOTA) results in Named Entity Recognition (NER) tasks for many languages without the need for manually crafted features. However, these models still require manually annotated training data, which is not available for many languages. In this paper, we propose an unsupervised cross-lingual NER model that can transfer NER knowledge from one language to another in a completely unsupervised way without relying on any bilingual dictionary or parallel data. Our model achieves this through word-level adversarial learning and augmented fine-tuning with parameter sharing and feature augmentation. Experiments on five different languages demonstrate the effectiveness of our approach, outperforming existing models by a good margin and setting a new SOTA for each language pair.

  • Publication . Article . Preprint . Other literature type . 2020
    Open Access English
    Authors: 
    G.-J. A. Brummer; G.-J. A. Brummer; B. Metcalfe; B. Metcalfe; W. Feldmeijer; W. Feldmeijer; M. A. Prins; J. van 't Hoff; J. van 't Hoff; G. M. Ganssen;
    Publisher: HAL CCSD
    Countries: Netherlands, France, France
    Project: NWO | SCAN-2: Scanning Sediment... (2300165588), ANR | L-IPSL (ANR-10-LABX-0018)

    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.

  • Publication . Article . Conference object . Preprint . 2020 . Embargo End Date: 01 Jan 2020
    Open Access
    Authors: 
    Robert Litschko; Ivan Vulić; Željko Agić; Goran Glavaš;
    Publisher: arXiv
    Country: United Kingdom
    Project: EC | LEXICAL (648909), NWO | MicroDish BV - A Rapid Gr... (2300150313), NWO | Inorganic porous hollow f... (2300175491)

    Current methods of cross-lingual parser transfer focus on predicting the best parser for a low-resource target language globally, that is, “at treebank level”. In this work, we propose and argue for a novel cross-lingual transfer paradigm: instance-level parser selection (ILPS), and present a proof-of-concept study focused on instance-level selection in the framework of delexicalized parser transfer. Our work is motivated by an empirical observation that different source parsers are the best choice for different Universal POS-sequences (i.e., UPOS sentences) in the target language. We then propose to predict the best parser at the instance level. To this end, we train a supervised regression model, based on the Transformer architecture, to predict parser accuracies for individual POS-sequences. We compare ILPS against two strong single-best parser selection baselines (SBPS): (1) a model that compares POS n-gram distributions between the source and target languages (KL) and (2) a model that selects the source based on the similarity between manually created language vectors encoding syntactic properties of languages (L2V). The results from our extensive evaluation, coupling 42 source parsers and 20 diverse low-resource test languages, show that ILPS outperforms KL and L2V on 13/20 and 14/20 test languages, respectively. Further, we show that by predicting the best parser “at treebank level” (SBPS), using the aggregation of predictions from our instance-level model, we outperform the same baselines on 17/20 and 16/20 test languages.

  • Publication . Other literature type . Article . Preprint . 2019
    Open Access English
    Authors: 
    N. Blom; N. Blom; N. Blom; A. Gokhberg; A. Fichtner;
    Countries: Switzerland, United Kingdom
    Project: EC | CSEM (714069), NWO | Full waveform inversion f... (2300172344)

    We present a seismic waveform tomography of the upper mantle beneath the central and eastern Mediterranean down to the mantle transition zone. Our methodology incorporates in a consistent manner the information from body and multimode surface waves, source effects, frequency dependence, wavefront healing, anisotropy and attenuation. This allows us to jointly image multiple parameters of the crust and upper mantle. Based on the data from ∼ 17 000 unique source–receiver pairs, gathered from 80 earthquakes, we image radially anisotropic S velocity, P velocity and density. We use a multi-scale approach in which the longest periods (100–150 s) are inverted first, broadening to a period band of 28–150 s. Thanks to a strategy that combines long-period signals and a separation of body and surface wave signals, we are able to image down to the mantle transition zone in most of the model domain. Our model shows considerable detail in especially the northern part of the domain, where data coverage is very dense, and displays a number of clear and coherent high-velocity structures across the domain that can be linked to episodes of current and past subduction. These include the Hellenic subduction zone, the Cyprus subduction zone and high-velocity anomalies beneath the Italian peninsula and the Dinarides. This model is able to explain data from new events that were not included in the inversion. Solid Earth, 11 (2) ISSN:1869-9510 ISSN:1869-9529