Digital libraries are online collections of digital objects that can include text, images, audio, or videos. It has long been observed that named entities (NEs) are key to the access to digital library portals as they are contained in most user queries. Combined or subsequent to the recognition of NEs, named entity linking (NEL) connects NEs to external knowledge bases. This allows to differentiate ambiguous geographical locations or names (John Smith), and implies that the descriptions from the knowledge bases can be used for semantic enrichment. However, the NEL task is especially challenging for large quantities of documents as the diversity of NEs is increasing with the size of the collections. Additionally digitized documents are indexed through their OCRed version which may contains numerous OCR errors. This paper aims to evaluate the performance of named entity linking over digitized documents with different levels of OCR quality. It is the first investigation that we know of to analyze and correlate the impact of document degradation on the performance of NEL. We tested state-of-the-art NEL techniques over several evaluation benchmarks, and experimented with various types of OCR noise. We present the resulting study and subsequent recommendations on the adequate documents and OCR quality levels required to perform reliable named entity linking. We further provide the first evaluation benchmark for NEL over degraded documents.

This paper tackles the task of named entity recognition (NER) applied to digitized historical texts obtained from processing digital images of newspapers using optical character recognition (OCR) techniques. We argue that the main challenge for this task is that the OCR process leads to misspellings and linguistic errors in the output text. Moreover, historical variations can be present in aged documents, which can impact the performance of the NER process. We conduct a comparative evaluation on two historical datasets in German and French against previous state-of-the-art models, and we propose a model based on a hierarchical stack of Transformers to approach the NER task for historical data. Our findings show that the proposed model clearly improves the results on both historical datasets, and does not degrade the results for modern datasets.

Abstract. Cross-Language Automatic Text Summarization produces a summary in a language different from the language of the source documents. In this paper, we propose a French-to-English cross-lingual summarization framework that analyzes the information in both languages to identify the most relevant sentences. In order to generate more informative cross-lingual summaries, we introduce the use of chunks and two compression methods at the sentence and multi-sentence levels. Experimental results on the MultiLing 2011 dataset show that our framework improves the results obtained by state-of-the art approaches according to ROUGE metrics.

Multi-Sentence Compression (MSC) aims to generate a short sentence with key information from a cluster of closely related sentences. MSC enables summarization and questionanswering systems to generate outputs combining fully formed sentences from one or several documents. This paper describes a new Integer Linear Programming method for MSC using a vertex-labeled graph to select different keywords, and novel 3-grams scores to generate more informative sentences while maintaining their grammaticality. Our system is of good quality and outperforms the state-of-theart for evaluations led on news dataset. We led both automatic and manual evaluations to determine the informativeness and the grammaticality of compressions for each dataset. Additional tests, which take advantage of the fact that the length of compressions can be modulated, still improve ROUGE scores with shorter output sentences.