From 124 million tokens to 1,021 neologisms

Rossini, Diego; van der Plas, Lonneke

doi:10.63317/4o6ks86o293r

Back

Rossini_vanderPlas_2026_LREC_NeoLLM_4o6ks86o293r

Conference paper (in proceedings)

From 124 million tokens to 1,021 neologisms : a large-scale pipeline for automatic neologism detection

Rossini, Diego ORCID Istituto di argomentazione, linguistica e semiotica (IALS), Facoltà di comunicazione, cultura e società, Università della Svizzera italiana, Svizzera
van der Plas, Lonneke ORCID Istituto di argomentazione, linguistica e semiotica (IALS), Facoltà di comunicazione, cultura e società, Università della Svizzera italiana, Svizzera

2026

Published in:

Proceedings of the Workshop Neology and Large Language Models. - 2026, p. 1-15

English 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.

Collections