A Unified Approach to Transliteration-based Text Input with Online Spelling Correction

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 609–618, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics A Unified Approach to Transliteration-based Text Input with Online Spelling Correction Hisami Suzuki Jianfeng Gao Microsoft Research One Microsoft Way, Redmond WA 98052 USA {hisamis,jfgao}@microsoft.com Abstract This paper presents an integrated, end-to-end approach to online spelling correction for text input. Online spelling correction refers to the spelling correction as you type, as opposed to post-editing. The online scenario is particularly important for languages that routinely use transliteration-based text input methods, such as Chinese and Japanese, because the desired target characters cannot be input at all unless they are in the list of candidates provided by an input method, and spelling errors prevent them from appearing in the list. For example, a user might type suesheng by mistake to mean xuesheng 学生 'student' in Chinese; existing input methods fail to convert this misspelled input to the desired target Chinese characters. In this paper, we propose a unified approach to the problem of spelling correction and transliteration-based character conversion using an approach inspired by the phrase- based statistical machine translation framework. At the phrase (substring) level, k most probable pinyin (Romanized Chinese) corrections are generated using a monotone decoder; at the sentence level, input pinyin strings are directly transliterated into target Chinese characters by a decoder using a log- linear model that refer to the features of both levels. A new method of automatically deriving parallel training data from user keystroke logs is also presented. Experiments on Chinese pinyin conversion show that our integrated method reduces the character error rate by 20% (from 8.9% to 7.12%) over the previous state-of-the art based on a noisy channel model. 1 Introduction This paper addresses the problem of online spelling correction, which tries to correct users' misspellings as they type, rather than post-editing them after they have already been input. This online scenario is particularly important for languages that routinely use transliteration-based text input methods, including Chinese and Japanese: in these languages, characters (called hanzi in Chinese and kanji/kana in Japanese) are typically input by typing how they are pronounced in Roman alphabet (called pinyin in Chinese, romaji in Japanese), and selecting a conversion candidate among those that are offered by an input method system, often referred to as IMEs or input method editors. One big challenge posed by spelling mistakes is that they prevent the desired candidates from appearing as conversion candidates, as in Figure 1: suesheng is likely to be a spelling error of xuesheng 学生 'student', but it is not included as one of the candidates. Figure 1: Spelling mistake prevents the desired output (学生) from appearing in the list of candidates This severely limits the utility of an IME, as spelling errors are extremely common. Speakers of a non-standard dialect and non-native speakers have a particularly hard time, because they may not know the standard pronunciation of the word to …

Content truncated to 500 words