From 124 Million Tokens to 1,021 Neologisms: A Large-Scale Pipeline for Automatic Neologism Detection
Researchers have built a production-scale pipeline that combines rule-based morphological analysis with LLM classification to detect neologisms at corpus scale. Processing 527 million Reddit posts across two decades, the system filtered 124.6 million unique tokens down to 1,021 high-confidence neologism candidates for expert validation. The work demonstrates how LLMs can serve as efficient classifiers within larger NLP workflows when paired with domain-specific linguistic frameworks, offering a replicable pattern for other large-scale language phenomena detection tasks.
Modelwire context
ExplainerThe actual contribution is architectural: the researchers show that LLMs work most efficiently not as end-to-end solvers but as high-precision classifiers downstream of domain-specific filtering. The 1,021 candidates represent not just detection but validated signal, which is a different claim than simply 'we found neologisms.'
This directly extends the pattern established in recent work on structured LLM workflows. The validation-driven chart generation pipeline (May 1) and the procedural execution diagnostic (also May 1) both identified that decomposing tasks into explicit stages with intermediate validation gates produces more reliable outputs than monolithic inference. This neologism work applies the same logic to linguistic classification: morphological rules act as a coarse filter, LLMs handle the nuanced boundary cases, and expert review validates the final set. The approach mirrors how constrained sensemaking (May 1) improved research ideation by scaffolding reasoning rather than removing it.
If this pipeline is replicated on other languages or linguistic phenomena (semantic drift, code-switching, technical jargon emergence) within the next 12 months with similar precision-to-candidate ratios, it confirms the architecture is genuinely portable. If adoption stalls or requires heavy task-specific tuning, the method is more brittle than the paper suggests.
Coverage we drew on
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.