Learning to Pronounce Chinese Without a Pronunciation Dictionary

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 5687–5693, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 5687 Learning to Pronounce Chinese Without a Pronunciation Dictionary Christopher Chu, Scot Fang, and Kevin Knight DiDi Labs 4640 Admiralty Way Marina del Rey, CA 90292 {chrischu,scotfang,kevinknight}@didiglobal.com Abstract We demonstrate a program that learns to pro- nounce Chinese text in Mandarin, without a pronunciation dictionary. From non-parallel streams of Chinese characters and Chinese pinyin syllables, it establishes a many-to-many mapping between characters and pronuncia- tions. Using unsupervised methods, the pro- gram effectively deciphers writing into speech. Its token-level character-to-syllable accuracy is 89%, which significantly exceeds the 22% accuracy of prior work. 1 Unsupervised Text-to-Pronunciation Many papers address the construction of auto- matic grapheme-to-phoneme systems using rules or supervised learning, e.g. (Berndt et al., 1987; Zhang et al., 2002; Xu et al., 2004; Bisani and Ney, 2008; Peters et al., 2017). The task of unsupervised grapheme-to- phoneme conversion is introduced by Knight and Yamada (1999). Given two non-parallel streams: • A corpus of written language (characters). • A corpus of spoken language (sounds). the goal is to build: • A mapping table between the character do- main and the sound domain. • A proposed pronunciation of the written char- acter sequences. Motivated by archaeological decipherment, Knight and Yamada (1999) view character se- quences as “enciphered” phoneme sequences. Their evaluation compares the proposed pro- nunciations with actual pronunciations. With a noisy-channel expectation-maximization method, they obtain 96% phoneme accuracy on Spanish, 99% on Japanese kana, but only 22% syllable accuracy on Mandarin Chinese. In this paper, we re-visit the task of decipher- ing Chinese text into standard Mandarin pronun- ciations (Figure 1). We obtain an improved 89% Figure 1: Learning to pronounce Chinese without a dic- tionary. syllable accuracy. We further explore exposing the internals of characters and syllables to the an- alyzer, as Chinese characters sharing written com- ponents often sound similar. We find it compelling that pronunciation dic- tionaries are largely redundant with non-parallel text and speech corpora, even for writing sys- tems as complex as Chinese. We also expect re- sults may be of use in dealing with novel ways to write Chinese, such as N¨ushu script (Zhang et al., 2016), with acoustic modeling of other Chinese languages and dialects, and with novel ways to phonetically encode and decode Chinese in online censorship applications (Zhang et al., 2014). 2 Chinese Writing The most-popular modern Chinese writing system renders each spoken syllable token with a single character token (hanzi). There are over 400 sylla- ble types in Mandarin1 and several thousand char- acter types. The mapping is many-to-many: • Almost every syllable type can be written 1In this paper, we use standard pinyin syllable representa- tion, and we refer strictly to Mandarin pronunciation. 5688 with different characters (eg, zhong → {中, 重, 肿, ...}). The choice depends on context. For example, the word zhongguo (“China”) is written 中 中 中国, but zhongyao (“important”) is written 重 重 重要. • …

Content truncated to 500 words