Abstract
Text generation can target different types of languages and take different types of knowledge as input. In this presentation, I will show how neural language models can be adapted to generate text from semantic representation graphs, knowledge graphs and multiple documents. The neural architectures presented will also illustrate how to generate texts from a single source, either in twenty-one languages of the European Union, or in so-called poorly endowed languages such as Breton, Welsh and Irish. Finally, work on the generation of Wikipedia biographies from multiple documents will shed light on the impact of data bias on the quality of the texts generated. The work presented was carried out as part of the IA xNLG Chair (Multilingual and Multisource Text Generation), co-funded by the ANR, Meta and the Grand-Est region.
Claire Gardent
Claire Gardent is Director of Research at the Centre national de la recherche scientifique (CNRS) at LORIA, Nancy. Her research focuses on automatic language processing, with a particular emphasis on multilingual and multisource generation. She has been appointed President of the European Chapter of the Association of Computational Linguistics (EACL), Editor-in-Chief of the journals Traitement Automatique des Langues and Language and Linguistic Compass (Computational and Mathematical Section). In 2022, she was awarded the CNRS Silver Medal and selected by the Association of Computational Linguistics (ACL) as an ACL Fellow. She currently holds the IA xNLG (Multilingual and Multisource Text Generation) chair, co-funded by ANR, Meta and the Grand-Est region.
