OCR, Classification& Machine Translation (OCCAM)

OCR, Classification & Machine Translation (OCCAM) Joachim Van den Bogaert, Arne Defauw, Frederic Everaert, Koen Van Winckel, Alina Kramchaninova, Anna Bardadym, Tom Vanallemeersch CrossLang Kerkstraat 106 9050 Gentbrugge Belgium {first.lastname}@croslang.com Pavel Smrž, Michal Hradiš Brno University of Technology Božetěchova 2 612 00 Brno Czech Republic smrz@fit.vut.cz, ihradis@fit.vutbr.cz Abstract The OCCAM project (Optical Character recognition, ClassificAtion & Machine translation), which runs from 2019 to 2021, and is carried out by CrossLang and Brno University of Technology, aims at integrating the CEF (Connecting Europe Facility) Automated Translation service with image classification, translation memories, optical character recognition, and machine translation. It will support the automated translation of scanned busi- ness documents (a document format that, currently, cannot be processed by the CEF eTranslation service) and will also lead to a tool useful for the digital humanities do- main. 1 Introduction The European Commission’s Business Registers Interconnection System (BRIS) facilitates the access to information on EU companies and ensures that all EU business registers can communicate to each other electronically, in relation to cross-border mergers and foreign branches.1 Its main task is to synchronize the information that is present within Members States’ business registers. The CEF has planned an integration of BRIS with the CEF eTranslation2 Digital Service Infrastructure (DSI), to make draft translations of company information available via the European 1https://ec.europa.eu/cefdig- ital/wiki/pages/viewpage.ac- tion?pageId=46992657 2https://ec.europa.eu/cefdigital/wiki/dis- play/CEFDIGITAL/eTranslation e-Justice Portal,3 but a large volume of scanned documents would remain untranslated, because it consists of raw images that are not machine- readable. A similar problem occurs in the digital human- ities domain: while there are plenty of optical character recognition (OCR) frameworks availa- ble (both open source and proprietary), the need for OCR and translation within the digital human- ities domain remains pressing. The European Newspaper Survey Report,4 as conducted during the Europeana Newspapers project, revealed that access to twentieth-century content remains prob- lematic, and only few libraries use OCR when scanning documents. At the same time, there is a growing interest in gaining multilingual access to cultural heritage resources. 2 Proposed solution for BRIS Existing content within the member state data- bases will be leveraged to recognise, classify and translate legacy and new content. The presence of database links to scanned documents, and the tem- plate-like nature of administrative documents will be exploited to optimize OCR and translation. Since a pipelined (cascaded) implementation (i.e. an OCR step followed by a machine translation (MT) step) has the inherent risk of error accumu- lation, OCCAM proposes a more informed classi- fication-based approach, as outlined in Figure 1, to: 3 https://e-justice.europa.eu/ 4 http://www.europeana-newspapers.eu/wp-con- tent/uploads/2012/04/D4.1-Europeana-newspa- pers-survey-report.pdf • recognise document types and link them to a corresponding data model (consisting of tem- plate text, data fields, data entries and free text); • identify entities within documents and link them to corresponding entries in member state databases (e.g. by using OCR to recognise VAT numbers and retrieve the corresponding data from a national business register); • retrieve translations from translation memo- riess associated with the data model; • use class-adapted OCR and …

Content truncated to 500 words