| { |
| "paper_id": "O08-2010", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:02:29.102905Z" |
| }, |
| "title": "Generating Patterns for Extracting Chinese-Korean Named Entity Translations from the Web", |
| "authors": [ |
| { |
| "first": "Chih-Hao", |
| "middle": [], |
| "last": "Yeh", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Ze University", |
| "location": { |
| "country": "Taiwan" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Wei-Chi", |
| "middle": [], |
| "last": "Tsai", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Ze University", |
| "location": { |
| "country": "Taiwan" |
| } |
| }, |
| "email": "thtsai@saturn.cse.yzu.edu.tw" |
| }, |
| { |
| "first": "Yu-Chun", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Taiwan Univeristy", |
| "location": { |
| "country": "Taiwan" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Richard", |
| "middle": [], |
| "last": "Tzong", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Han", |
| "middle": [], |
| "last": "Tsai", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Yuan Ze University", |
| "location": { |
| "country": "Taiwan" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "One of the main difficulties in Chinese-Korean cross-language information retrieval is to translate named entities (NE) in queries. Unlike common words, most NE's are not found in bilingual dictionaries. This paper presents a pattern-based method of finding NE translations online. The most important feature of our system is that patterns are generated and weighed automatically, saving considerable human effort. Our experimental data consists of 160 Chinese-Korean NE pairs selected from Wikipedia in five domains. Our approach can achieve a very high MAP of 0.84, which demonstrates our system's practicability.", |
| "pdf_parse": { |
| "paper_id": "O08-2010", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "One of the main difficulties in Chinese-Korean cross-language information retrieval is to translate named entities (NE) in queries. Unlike common words, most NE's are not found in bilingual dictionaries. This paper presents a pattern-based method of finding NE translations online. The most important feature of our system is that patterns are generated and weighed automatically, saving considerable human effort. Our experimental data consists of 160 Chinese-Korean NE pairs selected from Wikipedia in five domains. Our approach can achieve a very high MAP of 0.84, which demonstrates our system's practicability.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "In recent years, South Korean's entertainment industry has established itself as one of the most important emerging markets on the planet. In 2006, South Korean programs on Chinese government TV networks accounted for more than all other foreign programs combined [1] . South Korean actors such as Lee Young-ae (\uc774\uc601\uc560, \u674e\u82f1\u611b), Bae Yong Joon (\ubc30\uc6a9\uc900, \u88f4\u52c7\u4fca), Rain, and Song Hye Gyo (\uc1a1\ud61c\uad50, \u5b8b\u6167\u55ac) became very popular superstars in the great China area, making text and multimedia information related to them turned hot. Such information is firstly written or tagged in Korean. Unfortunately, most users in this area cannot directly specify queries in Korean. Therefore, it is necessary to translate queries from Chinese to Korean for Chinese-Korean (C-K) information retrieval systems. The main challenge involves translating named entities (such as names of shows, movies and albums) because they are usually the main concepts of queries.", |
| "cite_spans": [ |
| { |
| "start": 264, |
| "end": 267, |
| "text": "[1]", |
| "ref_id": "BIBREF0" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "Named entity (NE) translation is a challenging task because, although there are many online bilingual dictionaries, they usually lack domain specific words or NEs. Furthermore, new NEs are emerged everyday, but bilingual dictionaries cannot update their contents frequently. Therefore, it is necessary to construct a named entity translation (NET) system. In [2] , the authors romanized Chinese NEs and selected their English transliterations from English NEs extracted from the Web by comparing their phonetic similarities with Chinese NEs. Yaser Al-Onaizan [3] transliterated an NE in Arabic into several candidates in English and ranked the candidates by comparing their counts in several English corpora. Chinese-Korean NET is much more difficult than NET considered in previous works because a Chinese NE may not have similar pronunciation to its Korean translation.", |
| "cite_spans": [ |
| { |
| "start": 359, |
| "end": 362, |
| "text": "[2]", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 559, |
| "end": 562, |
| "text": "[3]", |
| "ref_id": "BIBREF2" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In this paper, we propose an effective pattern-based NET method which can achieve very high accuracy. All patterns are automatically generated and weighed, saving considerable human effort.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "To translate an NE originated from Chinese into Korean, we begin by considering the two C-K NET approaches. The older is based on the Sino-Korean pronunciation and the newer on the Mandarin. For example,\"\u81fa\u7063\" (Taiwan) used to be transliterated solely as \"\ub300\ub9cc\" (Daeman). However, during the 1990s, transliteration based on Mandarin pronunciation became more popular. Presently, the most common transliteration for \"\u81fa\u7063\" is \"\ud0c0\uc774\uc644\" (Ta-i-wan), though the Sino-Korean-based \"\ub300\ub9cc\" is still widely used. For Chinese personal names, both ways are used. For example, the name of Chinese actor Jackie Chan (\"\u6210\u9f8d\" Cheng-long) is variously transliterated as \"\uc131\ub8e1\"Seong-ryong (Sino-Korean) and \"\uccad\ub8fd\" Cheong-rung (Mandarin). Translating Chinese NEs by either method is a major challenge because each Chinese character may correspond to several different Hangul characters or character sequences that have similar pronunciations. This results in thousands of possible combinations of Hangul characters (e.g., \"\u5f35\u97f6\u6db5' Zhang Shao-han can be transliterated to \"\uc7a5\uc0ac\uc624\ud55c\" Jang-sa-o-han or \"\uc7a5\uc0e4\uc624\ud55c\" Jang-sya-o-han), making it very difficult to choose the most widely used one.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Difficulties in Chinese-Korean Named Entity Translation", |
| "sec_num": "2" |
| }, |
| { |
| "text": "NEs originated from Japan may contain Hiraganas, Katakanas, or Kanjis. For each character type, J-C translation rules may be similar to or very different from J-K translation rules. Some of these rules are based on Japanese pronunciation, while some are not. For NEs composed of all Kanjis, their Chinese translations are generally exactly the same as their Kanji written forms. In contrast, Japanese NEs are transliterated into Hangul characters. Take \"\u5c0f\u6cc9 \u7d14\u4e00\u90ce\" (Koitsumi Junichiro) for example. Its Chinese translation \"\u5c0f\u6cc9\u7d14\u4e00\u90ce\" is exactly the same as its Kanji written form, while its pronunciation (Xiao-quan Chun-yi-lang) is very different from its Japanese pronunciation. This is different from its Korean translation, \"\uace0\uc774\uc988\ubbf8 \uc900\uc774\uce58\ub85c\" (Ko-i-jeu-mi Jun-i-chi-ro). In this example, we can see that, because the translation rules in Chinese and Korean are different, it is ineffective to utilize phonetic similarity to find the Korean translation equivalent to the Chinese translation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Difficulties in Chinese-Korean Named Entity Translation", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In this section, we describe our C-K NET method for dealing with the problems described in Section 2. We observed that an NE and its translation may co-exist in a sentence. Such sentences may have structural similarity. For example, for \"\u674e\u660e\u535a\" and its Korean translation \"\uc774\uba85\ubc15\", the following sentences both contain the structure \"NE ( translation )\":", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Pattern-Based Named Entity Translation", |
| "sec_num": "3" |
| }, |
| { |
| "text": "\"2007\ub144 12\uc6d419\uc77c \ubc24 \u674e\u660e\u535a NE ( \uc774\uba85\ubc15 translation ) \ud6c4\ubcf4\uc758 \ub2f9\uc120\uc774 \uc0ac\uc2e4\uc0c1\" \" \u674e\u660e\u535a NE ( \uc774\uba85\ubc15 translation ) \ub300\ud1b5\ub839 \uc911\uad6d \ubc29\ubb38 \uc758\ubbf8\uc640 \uacfc\uc81c\"", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Pattern-Based Named Entity Translation", |
| "sec_num": "3" |
| }, |
| { |
| "text": "These local structures can be treated as surface patterns. In previous work, [4] employ five hand-crafted patterns to extract NE translations. However, it is time-consuming to manually create most patterns. Therefore, we aim to develop an approach that learns patterns automatically.", |
| "cite_spans": [ |
| { |
| "start": 77, |
| "end": 80, |
| "text": "[4]", |
| "ref_id": "BIBREF3" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Pattern-Based Named Entity Translation", |
| "sec_num": "3" |
| }, |
| { |
| "text": "To generate translation patterns (TP), we need to prepare sentences containing at least one Chinese NE and its Korean translation. For each pair of sentences x and y, we apply the Smith-Waterman local alignment algorithm [5] to find the longest common string. During the alignment process, positions where x and y share the same word are counted as a match. x's ith character and y's jth character are denoted as x i and y i , respectively. The algorithm firstly constructs an |x| \u00d7 |y| matrix S. Each element S i,j represents the similarity score of an optimal local alignment ending at x i and y j , which can be calculated by the following formula:", |
| "cite_spans": [ |
| { |
| "start": 221, |
| "end": 224, |
| "text": "[5]", |
| "ref_id": "BIBREF4" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "S i,j = max \uf8f1 \uf8f4 \uf8f4 \uf8f2 \uf8f4 \uf8f4 \uf8f3 0 S i\u22121,j\u22121 + s(x i , y i ) S i\u22121,j \u2212 d S i,j\u22121 \u2212 d ,", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "where are selected, a wild card will be attached in front of \u03c4 . This process stops until it arrives at the first zero-valued element and \u03c4 is output.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "s(x i , y i ) is", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "The following is an example of a pair of sentences that contains \"\u8a00\u627f\u65ed\" (Jerry Yen) and its Korean translation, \"\uc5b8\uc2b9\uc6b1\" (Eon Seung-uk) :", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "\u2022 \ub300\ub9cc \ubc30\uc6b0 \uc5b8\uc2b9\uc6b1 (\u8a00\u627f\u65ed) \uc694\uc57d\uc815\ubcf4.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "\u2022 \ubc30\uc6b0 \uc5b8\uc2b9\uc6b1(\u8a00\u627f\u65ed)\uc774 \ucde8\uc7ac\uc9c4\uc758 \uc9c8\ubb38\uc5d0 \ub2f5\ud558\uace0 \uc788\ub2e4.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "After alignment, the pattern is generated as:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "\ubc30\uc6b0 <Korean NE slot>(<Chinese NE slot>)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "This pattern generation process is repeated for each NE-translation pair.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Translation Patterns", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "After learning the patterns, we have to filter out some ineffective patterns and determine each TP's weight for ranking translation candidates. Each TP \u03c4 is evaluated by employing it to extract all possible Korean translations for each training-set NE e in from the sentences used to generate \u03c4 . In extracting e's translations, \u03c4 's <Chinese NE slot> is replaced with e. \u03c4 's extraction F-score over all training-set NEs is treated as its weight and calculated as follows: ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Weighting Translation Patterns", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "To find a Chinese NE's Korean translation, we can apply the TPs to extract possible Korean translations. For the given input Chinese NE e, our system sends e to AltaVista and limits the result pages to Chinese and Korean. The snippets are collected and break them into sentences.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracting and Ranking Translation Candidates", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "Only the sentences containing e are retained.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracting and Ranking Translation Candidates", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "We then use all C-K TPs, whose <Chinese NE slot> are replaced with e, to match the retained sentences. The strings matched by <Korean NE slot> are extracted as e's translation candidates. Each candidate c is scored by summing up the weights of TPs extracting c. This score is used to rank all candidates.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracting and Ranking Translation Candidates", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "In this section, we conduct an experiment to evaluate our pattern-based C-K NET system on different NE types.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation and analysis", |
| "sec_num": "4" |
| }, |
| { |
| "text": "In this section, we illustrate how to prepare the experimental data for learning and testing TPs. As mentioned in Section 3, TP learning requires sentences containing a Chinese NE and its Korean translation.To prepare them, we firstly collect a list of NEs in Chinese, which comprises of 120 NEs originated in Chinese such as \u8a00\u627f\u65ed (Jerry Yan) and \u5468\u6770\u502b (Jay Chou) as well as NEs originated in Korea such as \u88f4\u52c7\u4fca (Bae Young-Jun) and \u5f35\u5a1c\u62c9 (Jang Na-Ra). These NEs are divided into five types based on its Wikipedia page's categorization, including person, location, organization, architecture, and others. Secondly, to acquire these NEs' Korean translation, each NE is sent to the Chinese Wikipedia, and the title of the matched article's Korean version is treated as the NE's translation in Korean. Thirdly, each NE and its Korean translation are attached in a query and then the query is sent to AltaVista. For instance, \"\u8a00\u627f \u65ed\" (Jerry Yan) and its Korean translation \"\uc5b8\uc2b9\uc6b1\" are used to produce a query \"+\u8a00\u627f\u65ed +\uc5b8 \uc2b9\uc6b1\". The returned snippets in the top 20 pages are split into sentences. Only the sentences that contain at least one NE and its Korean translation are retained in the test set.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Data Sets", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "The preparation of test data is similar to that of training data. 40 NEs (other than the 120 training NEs) are collected from the Wikipedia. The distribution among the five categories are exactly the same as that of the training NEs. Then, each NE is directly sent to Altavista. The returned snippets in the top 20 pages are split into sentences. Only the sentences that contain the NE are retained.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Data Sets", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "We use the Mean Average Precision (MAP) [6] and Top-1 Precision at the Top-1 to measure our NET system's performance. For evaluating the performance of translating each NE, we can calculate the average precision (AP), which is the average of precisions computed after ", |
| "cite_spans": [ |
| { |
| "start": 40, |
| "end": 43, |
| "text": "[6]", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation Methodology", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "AP = \u2211 N r=1 (P (r) \u00d7 rel(r)) number of correct translations ,", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation Methodology", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "where r is the rank, N the number of extracted candidates, rel() a binary function on the correctness of a given rank, and P () precision at a given cut-off rank. For evaluating the performance of translating all NEs, we can calculate the mean average precision (MAP), which is mean value of all the average precisions. Top-1 precision, used for evaluating the quality of our system's Top-1 candidates, is computed as the ratio of numbers that the correct translation is the top 1 candidate divided by the total number of NE queries. For evaluating the translation result, our Korean expert manually checked if the candidates are correct translations. A candidate is judged as correct regardless it is generated based on the Sino-Korean or Mandarin pronunciation. Table 1 shows the categorical and overall performance of our pattern-based NET method. The results show that our method can achieve close to 0.85 in both metrics. The scores of translating person names are even higher. Both metrics are over 0.9 and the Top-1 precision is equal to 1, that is, all the Top-1 candidates output by our system are correct. This indicates that our system's ranking for translation candidates is very accurate and can be further exploited in other applications that incorporate our NET system.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 764, |
| "end": 771, |
| "text": "Table 1", |
| "ref_id": "TABREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Evaluation Methodology", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "From the evaluation results, we find that our method can translate most Chinese NEs effectively. However, there are still some cases dropping the performance. In the following subsections, we discuss TP's effectiveness and analyze error cases.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discussion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "For most test NEs, our method can extract their correct translations and put them in the forepart of the candidate list. For example, for \"\u91d1\u6cf0\u7199\" (Jin Tai-xi), the Top-1 candidate is \"\uae40\ud0dc\ud76c\" (Kim Tae-Hee), which is exactly the correct translation. As mentioned in section 2, Korean transliterates Chinese NEs according to their Sino-Korean or Mandarin pronunciation. Our method can extract translations based on both methods. For example, for \"\u8b1d\u9577\u5ef7\" (Xie Changting), both the Sino-Korean transliteration \"\uc0ac\uc7a5\uc815\" (Sa-jang-jeong) as well as the Mandarin transliterations such \"\uc170\ucc3d\ud305\" (Sye-chang-ting), \"\uc138\ucc3d\ud305\" (Se-chang-ting), and \"\uc2dc\uc5d0\uce20\uc559 \ud305\" (Si-e-cheu-ang-ting) are extracted. It shows that our method can extract most Chinese NEs' Korean translations regardless of how the Korean translations are generated.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Effectiveness of Pattern-based NET", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "Our method may extract an NE's relevant phrases in addition to its translations. For instance, for \"\u89aa\u6c11\u9ee8\" (People First Party), both the correct translation \"\uce5c\ubbfc\ub2f9\" (Chin-min-dang) and the relevant phrase \"\uc479\ucd94\uc704\" (Ssung Chu-ui, \u5b8b\u695a\u745c, the chairman of People First Party) are extracted. Although relevant phrases are not exact the translations, they might improve the performance of some NET applications such as cross-language information retrieval (CLIR). This is because the effect of adding relevant phrases in queries is similar to that of query expansion.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Relevant Terms", |
| "sec_num": "5.2.1" |
| }, |
| { |
| "text": "Different phraseology in Chinese and Korean might make our NET method extract false positives. For example, the First Sino-Japanese War (1894-1895) is called Jiawu Zhanzheng (\u7532 \u5348\u6230\u722d) by Chinese but is called Cheong-il-jeong-jaeng (\u6e05\u65e5\u6230\u722d) by Koreans. The Top-1 candidate output by our system is \"\uac11\uc624\uc804\uc7c1\" (kap-o-jeon-jaeng), which is only annotated for Koreans to understand its pronunciation. The most widely used Korean translations for the First Sino-Japanese War is \"\uccad\uc77c\uc804\uc7c1\" (Cheong-il-jeong-jaeng, \u6e05\u65e5\u6230\u722d). The other example is the Chinese query \"\u6d6a\u6f2b\u6eff\u5c4b\" (Lang-man-man-u), a Korean drama's name. The Top-1 candidate is \"\ub791\ub9cc\ub9cc\uc6b0\" (rang-man-man-u), which is the transliteration of the Chinese characters \"\u6d6a\u6f2b\u6eff\u5c4b\" based on the Mandarin pronunciation. However, the correct Korean translation is \"\ud480\ud558\uc6b0\uc2a4\" (Full House), which is the transliteration based on the English pronunciation. These two queries show that different phraseology may rank the incorrect translation candidates higher. However, the correct translations, such as \"\uccad\uc77c\uc815\uc7c1\" and \"\ud480\ud558\uc6b0\uc2a4\", are also extracted by our method. For some applications, such as CLIR, the inaccurate ranking does not influence the performance a lot [7] .", |
| "cite_spans": [ |
| { |
| "start": 1166, |
| "end": 1169, |
| "text": "[7]", |
| "ref_id": "BIBREF6" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Different Phraseology in Chinese and Korean", |
| "sec_num": "5.2.2" |
| }, |
| { |
| "text": "In this paper, we have demonstrated several advantages of our pattern-based NE translation method. Our pattern-based method achieves higher recall because it extracts NE translations from the Web, which contains most of human knowledge. Even translations of novel NEs can be found. Second, our method can extract most translations for each NE. This feature makes similar effects of query expansion and very helpful for cross-language information retrieval because documents containing frequent or infrequent translations can be retrieved. Finally, the high MAP over all five domains establishes our method's generality.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "6" |
| }, |
| { |
| "text": "In the future, we plan to apply our method to other language pairs. We also hope to extract not only the translations but also relevant information to them. We believe these new features can be applied to other applications, such retrieving multimedia contents on the Web 2.0 platform whose tags are written in different languages from queries.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "6" |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Korean wave as a sign of global shift", |
| "authors": [ |
| { |
| "first": "Cho Hae-Joang", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Korea Journal", |
| "volume": "", |
| "issue": "", |
| "pages": "167--172", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Cho Hae-Joang, \"Reading the \"Korean wave as a sign of global shift\"\", Korea Journal, pp. 167-172, 2005.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Proper name translation in cross-language information retrieval", |
| "authors": [ |
| { |
| "first": "Sheng-Jie", |
| "middle": [], |
| "last": "Hsin-Hsi Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Yung-Wei", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "" |
| }, |
| { |
| "first": "Shih-Cbung", |
| "middle": [], |
| "last": "Ding", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Tsai", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Proceedings of 17th COLING and 36th ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "232--236", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hsin-Hsi Chen, Sheng-Jie Huang, Yung-Wei Ding, and Shih-Cbung Tsai, \"Proper name translation in cross-language information retrieval\", Proceedings of 17th COLING and 36th ACL, pp. 232-236, 1998.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Translating named entities using monolingual and bilingual resources", |
| "authors": [ |
| { |
| "first": "Yaser", |
| "middle": [], |
| "last": "Al-Onaizan", |
| "suffix": "" |
| }, |
| { |
| "first": "Kevin", |
| "middle": [], |
| "last": "Knight", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proceedings of the 40th Annual Meeting of the Association of Computational Linguistics (ACL)", |
| "volume": "", |
| "issue": "", |
| "pages": "400--408", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Yaser Al-Onaizan and Kevin Knight, \"Translating named entities using monolingual and bilingual resources\", Proceedings of the 40th Annual Meeting of the Association of Com- putational Linguistics (ACL), pp. 400-408, 2002.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "NTCIR-6 experiments using pattern matched translation extraction", |
| "authors": [ |
| { |
| "first": "Dong", |
| "middle": [], |
| "last": "Zhou", |
| "suffix": "" |
| }, |
| { |
| "first": "Mark", |
| "middle": [], |
| "last": "Truran", |
| "suffix": "" |
| }, |
| { |
| "first": "Tim", |
| "middle": [], |
| "last": "Brailsford", |
| "suffix": "" |
| }, |
| { |
| "first": "Helen", |
| "middle": [], |
| "last": "Ashman", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Proceedings of NTCIR-6 Workshop Meeting", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dong Zhou, Mark Truran, Tim Brailsford, and Helen Ashman, \"NTCIR-6 experiments using pattern matched translation extraction\", Proceedings of NTCIR-6 Workshop Meeting, 2006.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Identification of common molecular subsequences", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [ |
| "F" |
| ], |
| "last": "Smith", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "S" |
| ], |
| "last": "Waterman", |
| "suffix": "" |
| } |
| ], |
| "year": 1981, |
| "venue": "Journal of Molecular Biology", |
| "volume": "147", |
| "issue": "", |
| "pages": "195--197", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T. F. Smith and M. S. Waterman, \"Identification of common molecular subsequences\", Journal of Molecular Biology, vol. 147, pp. 195-197, 1981.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "A study of information seeking and retrieving", |
| "authors": [ |
| { |
| "first": "Tefko", |
| "middle": [], |
| "last": "Saracevic", |
| "suffix": "" |
| }, |
| { |
| "first": "Paul", |
| "middle": [], |
| "last": "Kantor", |
| "suffix": "" |
| }, |
| { |
| "first": "Alice", |
| "middle": [ |
| "Y" |
| ], |
| "last": "Chamis", |
| "suffix": "" |
| }, |
| { |
| "first": "Donna", |
| "middle": [], |
| "last": "Trivison", |
| "suffix": "" |
| } |
| ], |
| "year": 1988, |
| "venue": "Journal of the American Society for Information Science", |
| "volume": "39", |
| "issue": "3", |
| "pages": "161--176", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tefko Saracevic, Paul Kantor, Alice Y. Chamis, and Donna Trivison, \"A study of informa- tion seeking and retrieving\", Journal of the American Society for Information Science, vol. 39, no. 3, pp. 161-176, 1988.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Learning patterns from the web to translate named entities for cross language information retrieval", |
| "authors": [ |
| { |
| "first": "Yu-Chun", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Richard", |
| "middle": [], |
| "last": "Tzong-Han Tsai", |
| "suffix": "" |
| }, |
| { |
| "first": "Wen-Lian", |
| "middle": [], |
| "last": "Hsu", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Proceedings of the third International Joint Conference on Natural Language Processing (IJCNLP)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Yu-Chun Wang, Richard Tzong-Han Tsai, and Wen-Lian Hsu, \"Learning patterns from the web to translate named entities for cross language information retrieval\", Proceedings of the third International Joint Conference on Natural Language Processing (IJCNLP), 2008.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF0": { |
| "text": "the similarity function of x i and y i ; d is the gap penalty. After S are calculated, we backtrack from the optimal element to generate the output TP \u03c4 . \u03c4 is initialed to be empty. The backtracking process iterates as follows. Suppose the current element is S ij , our algorithm selects the largest one from {S i\u22121,j\u22121 , S i,j\u22121 , S i\u22121,j } as the next element. If S i\u22121,j\u22121 is selected, that is, x i\u22121 and y i\u22121 are identical, x i\u22121 will be attached in front of \u03c4 . If either S i,j\u22121 or S i\u22121,j", |
| "type_str": "table", |
| "content": "<table/>", |
| "num": null, |
| "html": null |
| }, |
| "TABREF2": { |
| "text": "Evaluation Results", |
| "type_str": "table", |
| "content": "<table><tr><td>NE Type</td><td colspan=\"2\">MAP Top-1 Precision</td></tr><tr><td>Location</td><td>0.7854</td><td>0.8571</td></tr><tr><td>Person</td><td>0.9458</td><td>1.0000</td></tr><tr><td colspan=\"2\">Organization 0.8333</td><td>0.6667</td></tr><tr><td colspan=\"2\">Architecture 0.7736</td><td>0.8333</td></tr><tr><td>Others</td><td>0.8702</td><td>0.7500</td></tr><tr><td>All</td><td>0.8402</td><td>0.8500</td></tr><tr><td colspan=\"3\">truncating the list after each of the correct translations in turn:</td></tr></table>", |
| "num": null, |
| "html": null |
| } |
| } |
| } |
| } |
