A decadal systematic review of contextual quote extraction architectures: paradigm shifts, benchmarks, and future trajectories

Introduction Context quote extraction (CQE), which selects the most relevant verbatim span from a source document given an external context, sits at the intersection of information retrieval and retrieval-augmented generation (RAG). As LLM-based systems increasingly require grounded, attributable outputs, the ability to extract the correct span has become a practical necessity. To the authors knowledge, no prior systematic review has mapped CQEs methodological development. Method Six databases were searched following the PRISMA 2020 guidelines, supplemented by citation snowballing from five anchor studies. Of the 388 records identified, 142 were removed before screening (98 duplicates, 44 non-English), leaving 246 screened by title and abstract and 146 assessed in full text. The 57 retained studies were appraised using a five-criterion quality rubric (mean: 6.4/10; range: 4–10). Results Four method clusters were identified, tracing an arc from lexical/feature-based approaches (2012–2017) through encoder-based neural models (2016–2022) and generative LLM prompting (2022-present) to multitasking RAG pipelines (2024–present). Among the benchmarks where systems were directly evaluated, RA-MTR reports the strongest results (Precision@1:71.8%; BoW F1:61.4% end-to-end). Two structural gaps persist: the journalistic domain accounts for approximately 5% of the included studies despite representing the highest volume of real-world applications, and only two publicly available CQE-native benchmarks were identified. Discussion The field has advanced rapidly in architectural sophistication while remaining constrained by the scarcity of benchmarks, metric fragmentation, and unresolved subjectivity. Six research priorities are proposed, with news-domain CQE and hallucination-resistant extraction being the most important. This review provides what is, to the authors knowledge, the first comprehensive evidence base for directing future CQE research.