GeoName: a system for back-transliterating pinyin place names

GeoName: a system for back-transliterating pinyin place names Kui Lam Kwok Computer Science Dept., CUNY Queens College, 65-30 Kissena Blvd. Flushing, NY, 11367 kwok@ir.cs.qc.edu Qiang Deng Computer Science Dept., CUNY Queens College, 65-30 Kissena Blvd. Flushing, NY, 11367 peterqc@yahoo.com Abstract To be unambiguous about a Chinese geographic name represented in English text as Pinyin, one needs to recover the name in Chinese characters. We present our approach to this back-transliteration problem based on processes such as bilingual geographic name lookup, name suggestion using place name character and pair frequencies, and confirmation via a collection of monolingual names or the WWW. Evaluation shows that about 48% to 72% of the correct names can be recovered as the top candidate, and 82% to 86% within top ten, depending on the processes employed. 1 Introduction Names referring to entities can be ambiguous because different entities may have been given the same name. When one encounters foreign place names within English texts, further complication arises because the English alphabet may not represent the native writing uniquely or adequately, and transliteration has to be employed. This is true for Chinese place names. In this information age, documents on Chinese events such as news stories, commentaries, reviews, analysis, can originate from various sources and languages other than Chinese. Authors may reference Chinese place names but not necessary accompany it with the actual Chinese characters. It is therefore useful to build an automatic algorithm to decode such a place name in English and map it to the original Chinese character representation. Chinese language is written as a contiguous string of ideographs (characters) without white space. Geographic names of most cities, provinces, mountains, etc are two to four characters long. Border regions have longer place names. Unlike person names, there is not a preferred closed set for family name characters. Any of the over 6K GB-encoded character is theoretically admissible as part of a place name. When one refers to them in English text, one needs to represent them using English alphabets – a process of romanization. Two main systems exist for this process: Pinyin, official in Mainland China, and Wade-Giles convention, popular in Taiwan (see e.g. http://www.romanization.com). Their objective is to spell out the pronunciation of the Chinese characters with alphabets. Unfortunately, although written Chinese is by and large uniform (except for a few hundred characters that have simplified vs. traditional forms), spoken Chinese can vary from region to region with different dialects. The Pinyin system was introduced by the PRC government in the 1950’s. It attempts to standardize the representation according to the official Beijing Potunghua dialect (Northern China Mandarin) for the whole country. The Wade-Giles system is an older convention designed by authors of the same names in the late 19 th and early 20 th century, and is popular in Taiwan and some parts of South-East Asia. There are also other haphazard romanization conventions in different regions where Chinese is used. For example, Hong Kong has its own British colonial history and Southern (GuangDong) …

Content truncated to 500 words