Text summarization is the task of shortening a larger body of text into a concise version while retaining its essential meaning and key information. While summarization has been significantly explored in English and other high-resource languages, Czech text summarization, particularly for historical documents, remains underexplored due to linguistic complexities and a scarcity of annotated datasets. Large language models such as Mistral and mT5 have demonstrated excellent results on many natural language processing tasks and languages. Therefore, we employ these models for Czech summarization, resulting in two key contributions: (1) achieving new state-of-the-art results on the modern Czech summarization dataset SumeCzech using these advanced models, and (2) introducing a novel dataset called Posel od Čerchova for summarization of historical Czech documents with baseline results. Together, these contributions provide a great potential for advancing Czech text summarization and open new avenues for research in Czech historical text processing.

Abstract Historical document analysis plays a crucial role in understanding and preserving our past. However, this task is often hindered by challenges such as limited annotated training data and the diverse nature of historical handwritten documents. In this paper, we explore the potential of self-supervised learning (SSL) in historical document analysis, with a particular focus on historical handwritten document segmentation, to overcome the need for extensive annotated data while enhancing efficiency and robustness. We present an overview of SSL methods suitable for historical document analysis and discuss their potential applications and benefits. Furthermore, we present an approach for SSL in the document domain, considering various setups, augmentations, and resolutions. We also provide experimental results that demonstrate its feasibility and effectiveness. Our findings indicate that most document segmentation tasks can be effectively addressed using SSL features, highlighting the potential of SSL to advance historical document analysis and pave the way for more efficient and robust document processing workflows.

Dialogue act (DA) recognition is usually realized from a speech signal that is transcribed and segmented into text. However, only a little work in DA recognition from images exists. Therefore, this paper concentrates on this modality and presents a novel DA recognition approach for image documents, namely comic books. To the best of our knowledge, this is the first study investigating dialogue acts from comic books and represents the first steps to building a model for comic book understanding. The proposed method is composed of the following steps: speech balloon segmentation, optical character recognition (OCR), and DA recognition itself. We use YOLOv8 for balloon segmentation, Google Vision for OCR, and Transformer-based models for DA classification. The experiments are performed on a newly created dataset comprising 1,438 annotated comic panels. It contains bounding boxes, transcriptions, and dialogue act annotation. We have achieved nearly 98% average precision for speech balloon segmentation and exceeded the accuracy of 70% for the DA recognition task. We also present an analysis of dialogue structure in the comics domain and compare it with the standard DA datasets, representing another contribution of this paper.

International audience

With the author(s)’ decision to order Open Choice, the copyright of the article changed on 3rd December 2020 to [The Authors] [2020] and the article is forthwith distributed under the terms of copyright.

Abstract As the number of digitized historical documents has increased rapidly during the last a few decades, it is necessary to provide efficient methods of information retrieval and knowledge extraction to make the data accessible. Such methods are dependent on optical character recognition (OCR) which converts the document images into textual representations. Nowadays, OCR methods are often not adapted to the historical domain; moreover, they usually need a significant amount of annotated documents. Therefore, this paper introduces a set of methods that allows performing an OCR on historical document images using only a small amount of real, manually annotated training data. The presented complete OCR system includes two main tasks: page layout analysis including text block and line segmentation and OCR. Our segmentation methods are based on fully convolutional networks, and the OCR approach utilizes recurrent neural networks. Both approaches are state of the art in the relevant fields. We have created a novel real dataset for OCR from Porta fontium portal. This corpus is freely available for research, and all proposed methods are evaluated on these data. We show that both the segmentation and OCR tasks are feasible with only a few annotated real data samples. The experiments aim at determining the best way how to achieve good performance with the given small set of data. We also demonstrate that obtained scores are comparable or even better than the scores of several state-of-the-art systems. To sum up, this paper shows a way how to create an efficient OCR system for historical documents with a need for only a little annotated training data.

Current optical character recognition (OCR) systems commonly make use of recurrent neural networks (RNN) that process whole text lines. Such systems avoid the task of character segmentation necessary for character-based approaches. A disadvantage of this approach is a need of a large amount of annotated data. This can be solved by sing generated synthetic data instead of costly manually annotated ones. Unfortunately, such data is often not suitable for historical documents particularly for quality reasons. This work presents a hybrid approach for generating annotated data for OCR at a low cost. We first collect a small dataset of isolated characters from historical document images. Then, we generate historical looking text lines from the generated characters. Another contribution lies in the design and implementation of an OCR system based on a convolutional-LSTM network. We first pre-train this system on hybrid data. Afterwards, the network is fine-tuned with real printed text lines. We demonstrate that this training strategy is efficient for obtaining state-of-the-art results. We also show that the score of the proposed system is comparable or even better in comparison to several state-of-the-art systems.

This paper deals with multi-lingual dialogue act (DA) recognition. The proposed approaches are based on deep neural networks and use word2vec embeddings for word representation. Two multi-lingual models are proposed for this task. The first approach uses one general model trained on the embeddings from all available languages. The second method trains the model on a single pivot language and a linear transformation method is used to project other languages onto the pivot language. The popular convolutional neural network and LSTM architectures with different set-ups are used as classifiers. To the best of our knowledge this is the first attempt at multi-lingual DA recognition using neural networks. The multi-lingual models are validated experimentally on two languages from the Verbmobil corpus.