A Preliminary Study on Probabilistic Models for Chinese Abbreviations

Chinese abbreviations are widely used in the modern Chinese texts. They are a special form of unknown words, including many named entities. This results in difficulty for correct Chinese processing. In this study, the Chinese abbreviation problem is regarded as an error recovery problem in which the suspect root words are the “errors” to be recovered from a set of candidates. Such a problem is mapped to an HMM-based generation model for both abbreviation identification and root word recovery, and is integrated as part of a unified word segmentation model when the input extends to a complete sentence. Two major experiments are conducted to test the abbreviation models. In the first experiment, an attempt is made to guess the abbreviations of the root words. An accuracy rate of 72% is observed. In contrast, a second experiment is conducted to guess the root words from abbreviations. Some submodels could achieve as high as 51% accuracy with the simple HMM-based model. Some quantitative observations against heuristic abbreviation knowledge about Chinese are also observed.