Rossini, Diego van der Plas, Lonneke 2026 We present a scalable, modular pipeline for automatic neologism detection that combines rule-based filtering with LLM classification. The pipeline is grounded in two complementary word-formation frameworks, grammatical and extra-grammatical morphology, which jointly define the scope of what counts as a neologism and inform a four-class classification scheme (NEOLOGISM, ENTITY, FOREIGN, NONE). While designed to be modular and transferable at the architectural level, the pipeline is instantiated on 527 million English-language Reddit posts spanning 2005–2024. From this corpus, we extract 124.6 million unique tokens and reduce them by over 99.99% to yield 1,021 neologism candidates, a set small enough for manual expert verification. Multiple LLMs independently classify each candidate via majority vote, with a final verification step, revealing substantial cross-model disagreement and highlighting the challenge of operationalizing neologism detection at scale. Manual annotation of all 1,021 candidates confirms that 599 (58.7%) are genuine lexical innovations. application/pdf https://susi.usi.ch/global/documents/336329 https://localhost:5000/ark:/12658/srd1336329 https://susi.usi.ch/documents/336329/files/LREC_2026_NeoLLM_Workshop_NeoPipe_Rossini_vanderPlas.pdf eng info:eu-repo/semantics/altIdentifier/issn/2522-2686 info:eu-repo/semantics/altIdentifier/doi/10.63317/4o6ks86o293r info:eu-repo/semantics/altIdentifier/ark/12658/srd1336329 info:eu-repo/semantics/openAccess CC BY-NC Proceedings of the Workshop Neology and Large Language Models. - 2026, p. 1-15 Neologism detection Lexical innovation Reddit Large language models Rule-based filtering Computational neology info:eu-repo/classification/udc/004 From 124 million tokens to 1,021 neologisms : a large-scale pipeline for automatic neologism detection http://purl.org/coar/resource_type/c_5794