This is an experimental reading tool for the Myanmar language, designed to make complex texts more accessible through interactive linguistic annotation. It combines word segmentation, hover pop-ups drawn from multiple dictionaries, a custom-built romanization engine, as well as a full natural language processing pipeline including POS tagging, dependency parsing, and named entity recognition in order to extract semantic and syntactic meaning from a variety of raw inputs. Entire books can be uploaded in PDF form and are parsed page by page using an in-house renderer.
Burmese is a notoriously difficult language to process computationally: it is written in an unbroken script without whitespace, making word boundary detection difficult, and suffers from limited high-quality annotated machine-learning corpora.
This project addresses those limitations through a statistics-driven, dictionary-aware segmentation process, as well as a suite of specially trained neural networks. Many of the datasets used by these modules were produced by Ye Kyaw Thu, without whose work this project would not be possible.
Core lexicons include the Myanmar Language Commission's Myanmar–English Dictionary, the Universal Burmese–English–Pali Dictionary by Hoke Sein, and Burmese/Myanmar Dictionary of Grammatical Forms by Anna J. Allott and John Okell.
Conrad Compagna holds a PhD in history from Birkbeck, University of London and is a specialist in the history of the British Empire in South and Southeast Asia as well as Southeast Asian borderlands, in particular the region between northeast India, Myanmar, and Yunnan in China.
He also has expertise in digital humanities, including applied natural language processing for text extraction and analysis, text mining, and data visualization.
