MMDT NER

Overview & Abstract

MMDT NER is a foundational language technology project that transforms unstructured Burmese text into structured, usable data by identifying people, organizations, locations, dates, times, and numerical expressions. Built on a 2.14-million-token annotated corpus, it combines transformer-based modeling and responsible evaluation to advance context-aware AI for Burmese—an underrepresented, low-resource language—and support research, journalism, digital archives, humanitarian initiatives, and other public-interest applications.

Products · Low Resource Languages View Product ↗

More in Low Resource Languages

mmdt-tokenizer

"mmdt-tokenizer" is an open-source Python toolkit for Myanmar-language text segmentation. It uses a transparent, grammar-informed pipeline to protect URLs, emails, dates, phone numbers, and numeric expressions before segmenting text into syllable- and word-level units. Curated linguistic lexicons and structural rules recognize postpositions, conjunctions, classifiers, verbal particles, and sentence-final particles, while merging related predicate constructions into meaningful tokens. Designed for low-resource settings, the system is lightweight, auditable, and extensible without model retraining, providing essential language infrastructure for Myanmar NLP research, corpus development, search, information extraction, and other language technologies.

← Back to Low Resource Languages All R&D Projects →