A Spellcheck Technique for Tibetan Syllables Based on Grammatical Vector Multiplication

Automated spellcheck is a highly challenging research topic in the field of natural language processing, with broad application potential for corpus construction, text editing, speech captioning, and text recognition. Tibetan, a type of Pinyin script, is composed of 1–7 basic elements spliced horizontally and vertically. As a result, the frequency of spelling errors in Tibetan texts is very high. This article analyzes the rules of character formation in Tibetan grammar and converts complex Tibetan syllables into their corresponding structures. Prior studies have shown the positions of components in Tibetan syllables are consistent for characters with the same structure. In this study, we propose a structure-based component recognition model for Tibetan syllables (SBCRATS). In addition, four new spelling rules are developed by investigating 28 existing rules. Four Tibetan grammatical specifications are then vectorized and corresponding functions are introduced. These vectors provide a Tibetan script spell check methodology based on a vector multiplication model (MOSCTCBOGVM), which is simple and easy to implement. Algorithm performance was assessed using five real-world corpora and one corpus generated using Tibetan grammar. Five undergraduates majoring in Tibetan language marked the syllables identified by the model as being either correct or incorrect. A series of validation experiments showed the average spell check accuracy to be 99.86% at a rate of 51,050 words per second. This accuracy is comparable to the highest values reported in previous studies, but with unprecedented calculation speed.