8 research outputs found
Preparing Big Manuscript Data for Hierarchical Clustering with Minimal HTR Training
No full textHTR (Handwritten Text Recognition) technologies have progressed enough to offer high-accuracy results in recognising handwritten documents, even on a synchronous level. Despite the state-of-the-art algorithms and software, historical documents (especially those written in Greek) remain a real-world challenge for researchers. A large number of unedited or under-edited works of Greek Literature (ancient or Byzantine, especially the latter) exist to this day due to the complexity of producing critical editions. To critically edit a literary text, scholars need to pinpoint text variations on several manuscripts, which requires fully (or at least partially) transcribed manuscripts. For a large manuscript tradition (i.e., a large number of manuscripts transmitting the same work), such a process can be a painstaking and time-consuming project. To that end, HTR algorithms that train AI models can significantly assist, even when not resulting in entirely accurate transcriptions. Deep learning models, though, require a quantum of data to be effective. This, in turn, intensifies the same problem: big (transcribed) data require heavy loads of manual transcriptions as training sets. In the absence of such transcriptions, this study experiments with training sets of various sizes to determine the minimum amount of manual transcription needed to produce usable results. HTR models are trained through the Transkribus platform on manuscripts from multiple works of a single Byzantine author, John Chrysostom. By gradually reducing the number of manually transcribed texts and by training mixed models from multiple manuscripts, economic transcriptions of large bodies of manuscripts (in the hundreds) can be achieved. Results of these experiments show that if the right combination of manuscripts is selected, and with the transfer-learning tools provided by Transkribus, the required training sets can be reduced by up to 80%. Certain peculiarities of Greek manuscripts, which lead to easy automated cleaning of resulting transcriptions, could further improve these results. The ultimate goal of these experiments is to produce a transcription with the minimum required accuracy (and therefore the minimum manual input) for text clustering. If we can accurately assess HTR learning and outcomes, we may find that less data could be enough. This case study proposes a solution for researching/editing authors and works that were popular enough to survive in hundreds (if not thousands) of manuscripts and are, therefore, unfeasible to be evaluated by humans
Explainable dating of greek papyri images
No full textGreek literary papyri, which are unique witnesses of antique literature, do not usually bear a date. They are thus currently dated based on palaeographical methods, with broad approximations which often span more than a century. We created a dataset of 242 images of papyri written in “bookhand” scripts whose date can be securely assigned, and we used it to train algorithms for the task of dating, showing its challenging nature. To address data scarcity, we extended our dataset by segmenting each image into its respective text lines. By using the line-based version of our dataset, we trained a Convolutional Neural Network, equipped with a fragmentation-based augmentation strategy, and we achieved a mean absolute error of 54 years. The results improve further when the task is cast as a multi-class classification problem, predicting the century. Using our network, we computed precise date estimations for papyri whose date is disputed or vaguely defined, employing explainability to understand dating-driving features
Explaining the Chronological Attribution of Greek Papyri Images
No full textGreek literary papyri, which are unique witnesses of antique literature, do not usually bear a date. They are thus currently dated based on palaeographical methods, with broad approximations which often span more than a century. We created a dataset of 242 images of papyri written in “bookhand” scripts whose date can be securely assigned, and we used it to train machine and deep learning algorithms for the task of dating, showing its challenging nature. To address the data scarcity problem, we extended our dataset by segmenting each image to the respective text lines. By using the line-based version of our dataset, we trained a Convolutional Neural Network, equipped with a fragmentation-based augmentation strategy, and we achieved a mean absolute error of 54 years. The results improve further when the task is cast as a multiclass classification problem, predicting the century. Using our network, we computed and provided precise date estimations for papyri whose date is disputed or vaguely defined and we undertake an explainability-based analysis to facilitate future attribution
List of manuscripts containing John Chrysostom's Homilies and the relevant manual transcriptions
No full textThis dataset consists of a list of all manuscripts (in the form of a .csv file) used as data in experiments with HTR training via Transkribus. The manuscripts are dated between the 10th-14th centuries and transmit John Chrysostom’s Homilies on St. Paul’s Epistles to Titus. Homilies 1 and 5 were exploited for the training process. In addition, 19 XML source files are provided in the TEI standards format, which contains a sample of the manual transcription used as ground truth data for training HTR models.
Specifically, the sample_dataset_chrysostomus_ad-titum.csv file includes the following columns:
Sigla: a capital letter used in critical editions to refer to a specific manuscript in an abbreviated form.
Manuscripts: the name of each manuscript, containing the library and the catalogue number assigned to it.
Folia: the folia (i.e., pages) of each manuscript used in the experiments. A different sequence of folia from the same manuscript is recorded in a separate row of this file.
Ground truth data sample [file_name]: the file name of the TEI/XML files that correspond to each manuscript.
Image files: most digital reproductions of manuscripts are under some degree of copyright protection. So, instead of the image files, in this column, one can find a link to the relevant library's digital archive (if applicable)
Transkribus: Reviewing HTR training on (Greek) manuscripts
No full textTranskribus is a fully developed GUI (graphical user interface) platform offering (among others) the possibility to train HTR models with AI. It supports auto-transcription and searching of historical documents and is oriented towards Archives, Libraries, and researchers. This review describes most of Transkribus’ features by outlining the results of a project researching HTR on Greek manuscripts. Transkribus proves to be a useful solution for painlessly implementing state-of-the-art technology on Humanities, regardless of technical expertise or resources’ limitations. Any discussion in this review for further development of the platform is only provided given some of the Greek manuscripts’ peculiarities
Pragmatic meaning perspectives into film dubbing: “Beauty and the Beast”
No full textΗ παρούσα έρευνα έχει ως σκοπό τη διερεύνηση της διαπολιτισμικής μεταφοράς στο πλαίσιο ενός μεταγλωττισμένου έργου κινουμένων σχεδίων, με γλώσσα-στόχο την Ελληνική. Παράλληλα, επιχειρεί να διερευνήσει τον τρόπο με τον οποίο το πολυτροπικό κείμενο του έργου έχει μεταφραστεί στη τελική του μορφή λαμβάνοντας υπ’ όψιν έναν αριθμό πραγματολογικών φαινομένων που έχουν παρατηρηθεί στην αρχική μορφή του κειμένου. Στην έρευνα έχουν αξιοποιηθεί διάλογοι από το κείμενο της Αμερικανικής ταινίας κινουμένων σχεδίων «Beauty and the Beast» (1991, Walt Disney Studios) καθώς και οι μεταγλωττισμένες εκδοχές αυτών στην Ελληνική γλώσσα αντίστοιχα. Η έρευνα απευθύνθηκε σε πενήντα τέσσερις δίγλωσσους στην Ελληνική και την Αγγλική συμμετέχοντες, με υπόβαθρο στη μετάφραση, έτσι ώστε να εξασφαλιστεί η εγκυρότητα της ητικής ερευνητικής προσέγγισης. Τα αποτελέσματα δείχνουν ότι οι περισσότεροι από τους διαλόγους μεταφράστηκαν με διαφορετικό τρόπο στη γλώσσα-στόχο, με αποτέλεσμα την τροποποίηση ή την παράλειψη εντελώς της γλωσσολογικής επίδρασης των πραγματολογικών φαινομένων της αρχικής μορφής του κειμένου.The aim of the study is to explore cross cultural transfer in a dubbed animated film with Greek as a target language, and examine how the target version of the multimodal text was rendered in relation to a number of pragmatic phenomena demonstrated in the source text. It used exchanges from the transcript of the American animation film Beauty and the Beast (1991, Walt Disney Studios) and the Greek dubbed version of these exchanges. The study asks fifty four Greek-English bilingual respondents with a background in translation, in order for the study to check the validity of the author's view. Results suggest that most of the exchanges were rendered differently in the target text, thus modifying or completely eliding the linguistic effect of a number of pragmatic phenomena
Colonic mucosa barrier defects in collagenous and ischemic colitis
No full textSummary. Aims. The subepithelial myofibroblasts (SEMFs) and the subepithelial band of macrophages (SEBM) are major components of the colonic mucosa barrier. Although their role in homeostasis is widely recognized, their contribution to disease states is largely unknown. Our ai m was to explore histological characteristics of SEMFs and SEBM in collagenous and ischemic colitis in order to identify specific changes in di st i nct mucosa backgrounds lacking significant inflammation. Methods. SEMFs, SEBM and lamina propria (LP) macrophages were identified immunohistochemically by al pha smoot h muscl e Act i n and Cl ust er of Differentiation 68 respectively in 38 colonic biopsies [14 collagenous colitis (CC), 14 ischemic colitis (IC), 10 normal mucosa]. Results. In CC, SEMFs were rarely detectable in the collagenous band while aSMA-negative pericryptal fibr oblast-like cells appear ed. In lower LP interconnecting SEMFs processes were formed. SEBM was preserved in areas with a collagenous layer up to 20 μm. In thicker layers, it was fragmented and gradually disappeared in parallel with engulfment of enlarged macrophages. LP macrophages were usually increased. In IC, slight SEMFs changes preceded discernible epithelial alterations. Rounding, disintegration and extinction of SEMFs constituted successive alterations coinciding with crypt shrinkage and denudation. SEBM displayed total or almost total abolishment in areas with crypt damage but also in sites with minimal changes and in adjacent normal mucosa. Conclusion. Our findings provide evidence of impairment of both mucosa barrier constituents in CC and IC. In CC, histological alterations are closely related to the collagenous layer which seems to affect SEMFs differentiation and migration as well as SEBM integrity. The early extinction of SEBM in IC is indicative of its high sensitivity to hypoxia and hypoperfusion. © The Author(s) 2024
Anastomosing hemangioma: Report of two renal cases and analysis of the literature
No full textBackground: Anastomosing hemangioma (AH) is a very rare vascular tumor mimicking angiosarcoma, predominately observed in kidney and less frequently in other organs. We present two new renal cases of AH at opposite ends of the clinical presentation spectrum, provide review of the literature and compare the epidemiological, clinical and pathological profiles of renal and non-renal cases. Case presentation: The first occurred in a 64-year-old woman presented with back pain and the second, a multifocal lesion, in a 47-year-old man with end stage renal disease (ESRD). Histology disclosed a vascular tumor with striking anastomosing pattern, minimal nuclear atypia and locally infiltrative pattern, mimicking superficially angiosarcoma. Extramedullary hematopoiesis, extensive perirenal fat entrapment and increased number of mast cells were additional features in the second lesion. Both patients are well, without disease, 25 and 14 months after diagnosis. Conclusion: Comprehensive review and analysis of the published literature show that the growing number of non-renal AHs exhibits similar epidemiologic, clinical, biologic and histologic characteristics with renal AHs and most mild differences vanish after exclusion of cases associated with ESRD. Better understanding of AH pathogenesis will contribute to optimal treatment choices. © 2017 The Author(s)
