https://belski.me/tags/nlp/Recent content in NLP on Vadzim Belski — AI Research & EngineeringHugoenFri, 01 May 2026 00:00:00 +0000https://belski.me/projects/creations/nemo-text-processing/Wed, 01 Jan 2025 00:00:00 +0000https://belski.me/projects/creations/nemo-text-processing/<p>Extended fork of NVIDIA&rsquo;s NeMo text normalization framework, adding a specialized <strong>Emirati Arabic (ar_ae)</strong> module for production TTS frontends.</p> <h2 id="whats-new">What&rsquo;s new</h2> <p>A complete WFST-based normalization pipeline for Gulf Arabic dialect, handling:</p> <ul> <li><strong>Numbers</strong> — cardinal, ordinal, Arabic-Indic numerals</li> <li><strong>Dates</strong> — Hijri and Gregorian calendar formats</li> <li><strong>Currencies</strong> — 20+ Arab world currencies (AED, SAR, KWD, QAR, BHD, OMR&hellip;)</li> <li><strong>Fractions</strong> — Arabic fraction expressions</li> <li><strong>Phone numbers</strong> — UAE and GCC formats</li> <li><strong>Electronic entities</strong> — URLs, emails in mixed Arabic/Latin text</li> </ul> <h2 id="coverage">Coverage</h2> <p>Supports 16+ languages from the base NeMo package, with Emirati Arabic as a first-class addition covering the linguistic specifics of the Gulf dialect — distinct from Modern Standard Arabic normalization.</p>https://belski.me/blog/transformer_architectures_classical_modern_nlp_healthit_finance/Fri, 01 May 2026 00:00:00 +0000https://belski.me/blog/transformer_architectures_classical_modern_nlp_healthit_finance/<p>A few months ago I was sitting in a client call with a hospital CTO who had just approved a budget to "deploy an LLM for everything." ICD-10 coding, PHI de-identification, discharge summaries, triage notes — all of it through one big decoder model. I asked him one question: "What's your inference budget per request?" He went quiet. That silence told me everything I needed to know about where this was going.</p>