Contents I. Introduction
Spell checking is a well-known and well-researched problem in computational linguistics and NLP applications. The Myanmar Language (Burmese) is an under-resourced and the official language of the Republic of the Union of Myanmar. To date, detecting and correcting Burmese spelling errors is one of the problems that intrigued NLP researchers from an early stage. In this technological age of computers and social media, proper spelling is important for efficient communication between people. The number of social media users in Myanmar was equivalent to 53.1% of the total population in January 2021. We analyzed and classified the Burmese spelling errors that are commonly found on social media into 10 categories of errors. The details of these ten error types are discussed in Section IV-B. According to our study, the most frequent errors are phonetic errors that are related to the wrong use of vowels and pronunciation. The second most frequent errors are typographic errors, which are caused by insertion, omission, substitution, and transposition. Consonant errors, the combination of phonetic and typographic errors, and sequence errors are the third, fourth and fifth frequent errors. Errors are caused by dialect usage, stacked words, short form word usage, and slang word usage. Some of the spelling errors caused by using both Unicode and de facto standard encoding named Zawgyi. Currently, spell checkers for the English language are well established. Burmese does not have a publicly available spell checker. Thus, the very first Burmese spell checker based on SymSpell [1] was experimented on both syllable and word segmentation for these ten error types, together with four frequency dictionaries according to maximum edit distance 1 to 6 respectively.
