Biomedical NLP pipelines usually assume clean input text, yet large‑scale corpora assembled via automated PDF parsing contain pervasive OCR‑like artifacts, token splits and merges, hyphenation remnants, and character‑level corruption. These issues systematically erode lexical evidence and degrade downstream classifiers.
We introduce a conservative, fully auditable spell‑correction reliability layer, conceived as a safety‑oriented preprocessing module rather than a maximal‑accuracy corrector. Under uncertainty the system abstains from editing, adhering to a medical "do‑no‑harm" philosophy. The deterministic architecture couples bounded edit‑distance candidate generation with corpus‑derived n‑gram scoring and a suite of biomedical safety gates that protect domain‑critical terminology.
In intrinsic evaluation we use a manually curated benchmark of 2,104 token‑level cases. The layer achieves 94.61% error‑fix recall on synthetic errors while incurring zero harmful edits on negative controls. Extrinsically, we test a three‑class CORD‑19 topic classifier (Prevention, Treatment, Epidemiology) on 10,000 examples under a principled four‑run protocol (Clean, Noisy, Restored, Safety). The layer recovers roughly 80.45% of the noise‑induced macro‑F1 degradation, raising macro‑F1 from 0.7654 (Noisy) to 0.7717 (Restored) and preserving near‑clean performance (Safety: 0.7721).
A supplementary case study on 103 real‑world OCR‑extracted abstracts classified with BioBERT shows that transformer encoders are relatively robust to mild noise, motivating a future grey‑box architecture that integrates bounded neural signals and UMLS lexicons without sacrificing auditability.
The system is fully deterministic, artifact‑driven, and designed with deployment and auditability in mind.
Blogger's Review: This work substantially improves classification performance on noisy biomedical text while maintaining safety, making it a promising addition to real‑world medical NLP deployments.