Abstract: 

Recent advancements in Question Answering (QA) and Machine Reading Comprehension (MRC) have been driven by deep learning and large language models, typically using synchronous document collections like Wikipedia and the Web. However, archival collections, such as historical newspapers, are underutilized despite their rich historical information. To address this gap, we introduce ChroniclingAmericaQA, a temporal QA dataset for supporting diverse QA tasks across raw OCR text, corrected OCR content, and scanned images. In the second part of the talk we overview TempDPR, an extension to Dense Passage Retrieval (DPR) approach that embeds time information for retrieving relevant documents from diachronic document collections. Together, ChroniclingAmericaQA and TempDPR advance time-sensitive QA by establishing new benchmarks for temporal retrieval and question answering.