Question Generation and Answering for exploring Digital Humanities collections

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4561–4568 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4561 Question Generation and Answering for exploring Digital Humanities collections Fr´ed´eric B´echet1 Elie Antoine1 Jeremy Auguste1 G´eraldine Damnati2 (1) Aix-Marseille Universit´e, CNRS, LIS (2) Orange Innovation, DATA&AI, Lannion{first.last}@lis-lab.fr {first.last}@orange.com Abstract This paper introduces the question answering paradigm as a way to explore digitized archive collections for Social Science studies. In particular, we are interested in evaluating largely studied question generation and question answering approaches on a new type of documents, as a step forward beyond traditional benchmark evaluations. Question generation can be used as a way to provide enhanced training material for Machine Reading Question Answering algorithms but also has its own purpose in this paradigm, where relevant questions can be used as a way to create explainable links between documents. To this end, generating large amounts of question is not the only motivation, but we need to include qualitative and semantic control to the generation process. In the framework the French ANR project ARCHIVAL, we propose a new approach for question generation, relying on a BART Transformer based generative model, for which input data are enriched by semantic constraints. Question generation and answering are evaluated on several French corpora, and the whole approach is validated on a new corpus of digitized archive collection of a French Social Science journal. Keywords: Question Generation, Machine reading Question Answering, Digital Humanities 1. Introduction Recent advances in representation learning of text have achieved remarkable results on benchmark Nat- ural Language Understanding (NLU) tasks as shown in the recent General Language Understanding Evalua- tion (GLUE) benchmarks (Wang et al., 2018), reach- ing even so-called human performance on several tasks (Wang et al., 2019) such as linguistic acceptabil- ity, question answering or semantic similarity. How- ever these impressive results are obtained on corpora specifically prepared for these benchmark evaluations; moreover these understanding tasks, although always related to a linguistic competency, can be considered as rather artificial as they are tailored to fit the need of system benchmark evaluation and can be quite far from a real application. In order to study how the current boost in performance in NLU models on benchmark data translates to real- life settings, the applicative framework considered here is the exploration of digitized collections by profes- sional users that are used to analyze archives in order to perform Social Science research. We chose to focus on the question/answering paradigm, as asking questions and looking for answers is at the same time a natural way for researchers to explore archives and also the task that received the most attention in recent language understanding studies, especially since the release of large training data such as SQuAD (Rajpurkar et al., 2016) 1. In this paper we will present first the self-management corpus, a collection of a French journal ranging over 20 years from 1966 to 1986, which has been cho- sen as our archival material in …

Content truncated to 500 words