Base_JSON/prefixO/json/O08/O08-3005.json

{
    "paper_id": "O08-3005",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T08:02:13.371134Z"
    },
    "title": "Improving Translation of Queries with Infrequent Unknown Abbreviations and Proper Names",
    "authors": [
        {
            "first": "Wen-Hsiang",
            "middle": [],
            "last": "Lu",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Cheng Kung University",
                "location": {
                    "addrLine": "No.1, University Road",
                    "postCode": "701",
                    "settlement": "Tainan City",
                    "country": "Taiwan (R.O.C"
                }
            },
            "email": "whlu@mial.ncku.edu.tw"
        },
        {
            "first": "Jiun-Hung",
            "middle": [],
            "last": "Lin",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Cheng Kung University",
                "location": {
                    "addrLine": "No.1, University Road",
                    "postCode": "701",
                    "settlement": "Tainan City",
                    "country": "Taiwan (R.O.C"
                }
            },
            "email": ""
        },
        {
            "first": "Yao-Sheng",
            "middle": [],
            "last": "Chang",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Cheng Kung University",
                "location": {
                    "addrLine": "No.1, University Road",
                    "postCode": "701",
                    "settlement": "Tainan City",
                    "country": "Taiwan (R.O.C"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Unknown term translation is important to CLIR and MT systems, but it is still an unsolved problem. Recently, a few researchers have proposed several effective search-result-based term translation extraction methods which explore search results to discover translations of frequent unknown terms from Web search results. However, many infrequent unknown terms, such as abbreviations and proper names (or named entities), and their translations are still difficult to be obtained using these methods. Therefore, in this paper we present a new search-result-based abbreviation translation method and a new two-stage hybrid translation extraction method to solve the problem of extracting translations of infrequent unknown abbreviations and proper names from Web search results. In addition, to efficiently apply name transliteration techniques to mitigate the problems of proper name translation, we propose a mixed-syllable-mapping transliteration model and a Web-based unsupervised learning algorithm for dealing with online English-Chinese name transliteration. Our experimental results show that our proposed new methods can make great improvements compared with the previous search-result-based term translation extraction methods.",
    "pdf_parse": {
        "paper_id": "O08-3005",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Unknown term translation is important to CLIR and MT systems, but it is still an unsolved problem. Recently, a few researchers have proposed several effective search-result-based term translation extraction methods which explore search results to discover translations of frequent unknown terms from Web search results. However, many infrequent unknown terms, such as abbreviations and proper names (or named entities), and their translations are still difficult to be obtained using these methods. Therefore, in this paper we present a new search-result-based abbreviation translation method and a new two-stage hybrid translation extraction method to solve the problem of extracting translations of infrequent unknown abbreviations and proper names from Web search results. In addition, to efficiently apply name transliteration techniques to mitigate the problems of proper name translation, we propose a mixed-syllable-mapping transliteration model and a Web-based unsupervised learning algorithm for dealing with online English-Chinese name transliteration. Our experimental results show that our proposed new methods can make great improvements compared with the previous search-result-based term translation extraction methods.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Many existing cross-language information retrieval (CLIR) systems [Ballesteros and Croft 1997; Hull and Grefenstette 1996] encounter great difficulties in dealing with unknown term translation since these systems rely mostly on general-purpose bilingual dictionaries, which usually lack translations of abbreviations and proper names. Moreover, according to the report in a previous work [Cheng et al. 2004] , even for frequent Web queries, about 64% of them are not covered in an English-Chinese lexicon with about 120K entries (provided by Linguistic Data Consortium). However, several automatic translation extraction methods based on parallel [Brown et al. 1993; Melamed 2000; Nie et al. 1999; Smadja et al. 1996] or comparable corpora [Rapp 1999; Fung and Yee 1998 ] eventually suffer from the problems of insufficient parallel texts and the shortage of translation accuracy of comparable corpora in various subject domains.",
                "cite_spans": [
                    {
                        "start": 66,
                        "end": 94,
                        "text": "[Ballesteros and Croft 1997;",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 95,
                        "end": 122,
                        "text": "Hull and Grefenstette 1996]",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 388,
                        "end": 407,
                        "text": "[Cheng et al. 2004]",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 647,
                        "end": 666,
                        "text": "[Brown et al. 1993;",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 667,
                        "end": 680,
                        "text": "Melamed 2000;",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 681,
                        "end": 697,
                        "text": "Nie et al. 1999;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 698,
                        "end": 717,
                        "text": "Smadja et al. 1996]",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 740,
                        "end": 751,
                        "text": "[Rapp 1999;",
                        "ref_id": "BIBREF23"
                    },
                    {
                        "start": 752,
                        "end": 769,
                        "text": "Fung and Yee 1998",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "The Web has been expanded with an enormous amount of multilingual hypertext resources in diverse subjects. Recently, a number of studies in natural language processing (NLP) have concentrated on the use of Web resources to complement insufficient text corpora [Cao and Li 2002; Kilgarriff and Grefenstette 2003 ]. To automatically collect huge amounts of parallel corpora from the Web in various domains, some researchers have developed feasible techniques of utilizing similar file names, text length, and link structures to extract parallel text pages from bilingual Web sites [Nie et al. 1999; Resnik 1999; Yang and Li 2003 ]. On the other hand, Lu et al. [2002] made the first attempt of mining unknown term translations from Web anchor texts. Both Cheng et al. [2004] and Zhang and Vines [2004] have explored language-mixed search-result pages for extracting translations of frequent unknown queries. Although these approaches have successfully enhanced the performance of frequent unknown query translation, they still suffer from the problems of data sparseness and indirect association errors in finding translations of infrequent unknown query terms, particularly for abbreviations and proper names [Melamed 2000 ].",
                "cite_spans": [
                    {
                        "start": 260,
                        "end": 277,
                        "text": "[Cao and Li 2002;",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 278,
                        "end": 310,
                        "text": "Kilgarriff and Grefenstette 2003",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 579,
                        "end": 596,
                        "text": "[Nie et al. 1999;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 597,
                        "end": 609,
                        "text": "Resnik 1999;",
                        "ref_id": "BIBREF24"
                    },
                    {
                        "start": 610,
                        "end": 626,
                        "text": "Yang and Li 2003",
                        "ref_id": "BIBREF30"
                    },
                    {
                        "start": 649,
                        "end": 665,
                        "text": "Lu et al. [2002]",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 753,
                        "end": 772,
                        "text": "Cheng et al. [2004]",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 777,
                        "end": 799,
                        "text": "Zhang and Vines [2004]",
                        "ref_id": "BIBREF31"
                    },
                    {
                        "start": 1208,
                        "end": 1221,
                        "text": "[Melamed 2000",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "In this paper, we focus on dealing with two kinds of translation of unknown query terms, including proper names and abbreviations. According to the report in Davis and Ogden [1998] , about 50% of unknown terms in queries are proper names. Most methods handling translations of proper names are based on name transliteration techniques [Knight and Graehl 1998; Lin and Chen 2002; Lin et al. 2003; Li et al. 2004] . One major drawback of these methods is that they do not consider semantic information. Lam et al. [2004] proposed a named entity matching model, which considers both semantic and phonetic information, and applied it in mining unknown named entity translations from online daily Web news. Huang et al. [2005] also presented a method to extract key phrase translations from the language-mixed search-result pages with phonetic, semantic and frequency-distance features. As for abbreviation translation, less attention has been put on this research topic in the past few years.",
                "cite_spans": [
                    {
                        "start": 158,
                        "end": 180,
                        "text": "Davis and Ogden [1998]",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 335,
                        "end": 359,
                        "text": "[Knight and Graehl 1998;",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 360,
                        "end": 378,
                        "text": "Lin and Chen 2002;",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 379,
                        "end": 395,
                        "text": "Lin et al. 2003;",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 396,
                        "end": 411,
                        "text": "Li et al. 2004]",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 501,
                        "end": 518,
                        "text": "Lam et al. [2004]",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 702,
                        "end": 721,
                        "text": "Huang et al. [2005]",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "Different from the above works, our major goal is to solve the problems of query translation to help users access English/Chinese information in cross-lingual Web searches. In this paper, therefore, we concentrate our attention on the challenge of dealing with the translations of infrequent unknown abbreviations and transliterated names in Web search queries, i.e., these unknown queries that appear infrequently in Web query logs. We present two new methods to effectively extract translations of these two kinds of infrequent unknown queries. First, we propose a search-result-based abbreviation translation method for handling bidirectional translation of abbreviations in Chinese/English. Second, a new two-stage hybrid translation extraction method, which combines Cheng et al.'s [2004] search-result-based term translation extraction method and a new Web-based transliteration method, is proposed to extract Chinese/English translations for infrequent unknown English/Chinese proper names. In addition, to train an effective transliteration model, we also present a Web-based unsupervised learning algorithm to automatically collect large amounts of diverse English-Chinese transliteration pairs from the Web. For application, we provide a real prototype website 1 for users to translate unknown terms in practice. Our experimental results show that the proposed new methods can make great improvements in extracting infrequent unknown term translation.",
                "cite_spans": [
                    {
                        "start": 772,
                        "end": 793,
                        "text": "Cheng et al.'s [2004]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "The remainder of this paper is organized as follows: Section 2 describes the problems of unknown term translation and our search-result-based term translation extraction approach. Section 3 evaluates the proposed approach. Section 4 provides a simple description and comparison with the related work. Section 5 gives our conclusions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "Cheng et al.'s search-result-based term translation extraction method (refer to Section 2.3) is effective in extracting translations for frequent unknown query terms. However, for a lot of infrequent abbreviations and proper names, their translations are still difficult to extract. For example, while submitting an English abbreviation \"AMIA\" to LiveTrans 2 , an incorrect Chinese translation \"\u7cfb\uf99c\" (series) is obtained. The reason might be that some abbreviations are semantically ambiguous and co-occur relatively infrequently with the correct Chinese translations of their full names (or original forms). However, we observe that for an English abbreviation, its full name may co-occur more frequently with its corresponding Chinese translation. Thus, to effectively extract correct translation for an infrequent abbreviation, our idea is to first identify its full name in search results, and then extract correct translation of its full name, using the search-result-based term translation extraction method mentioned above. Generally, it should be more feasible to extract the correct translation of an abbreviation via its full name. For example, if we can extract the full name of the abbreviation \"AMIA\", \"American Medical Informatics Association\", then we can get its correct Chinese translation \"\u7f8e\u570b\u91ab\u5b78\u8cc7\u8a0a\u5354\u6703\" via LiveTrans.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Problems",
                "sec_num": "2.1"
            },
            {
                "text": "On the other hand, an English proper name might have multiple Chinese transliterated names which often vary with different translators due to phonetic variation and the lack of standard transliteration rules [Gao et al. 2004] . In other words, there may be several Chinese transliterated names corresponding to an English name. For example, the name \"Disney\" has various Chinese transliterated names, including \"\u8fea\u58eb\u5c3c\", \"\u8fea\u65af\u5c3c\", \"\u8fea\u65af\uf90c\", \"\u72c4\u65af\uf90c\", and \"\u72c4\u58eb\u5c3c\"; the name \"Hussein\" also has several different Chinese transliterated names, including \"\u6d77\u73ca\", \"\u54c8\u73ca\", and \"\u4faf\uf96c\u56e0\". Obviously, it will be helpful for query translation in cross-lingual Web search if we can collect all possible transliterated names from the Web for each unknown proper name. However, it is a real challenge to find all the various transliterated names. Thus, we consider integrating name transliteration techniques into the process of translation extraction for infrequent unknown proper names. Our idea is that we first extract high-frequency terms from the search-result pages as transliteration candidates, and then filter out impossible candidates by using a name transliteration model. In fact, it is still challenging to build an effective transliteration model while lacking sufficient transliteration pairs for training. Therefore, we propose a Web-based unsupervised learning algorithm to automatically collect large amounts of English-Chinese transliteration pairs from Web search results. Figure 1 demonstrates the process of our search-result-based query translation method. First, an unknown term is determined by a general-purpose dictionary. Then, an unknown term is recognized as an abbreviated term using our search-result-based abbreviation translation extraction methods. If the unknown term does not belong to an abbreviated term, we have to examine whether the unknown term is a transliteration based on our two-stage hybrid translation extraction method. To deal with unknown term translation, we employ the search-result-based term translation extraction method (described in Section 2.3) to handle translation of frequent (popular) unknown query terms, and propose two new infrequent unknown translation methods, namely the search-result-based abbreviation translation extraction method (Section 2.4) and two-stage hybrid translation extraction method (Section 2.5), to solve the problems of translation of abbreviated terms (i.e., abbreviations) and transliterated terms (i.e., proper names). To recognize the abbreviated terms in queries, we",
                "cite_spans": [
                    {
                        "start": 208,
                        "end": 225,
                        "text": "[Gao et al. 2004]",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1459,
                        "end": 1467,
                        "text": "Figure 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Problems",
                "sec_num": "2.1"
            },
            {
                "text": "collected an abbreviation list containing about 4K entries from the Wikipedia 3 website and then generated some pre-defined abbreviation patterns like those used in Park and Byrd (2001) . Besides these, we used a Web-based transliteration model to recognize a transliterated term (Section 2.5).",
                "cite_spans": [
                    {
                        "start": 165,
                        "end": 185,
                        "text": "Park and Byrd (2001)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "In this section, we will describe Cheng et al.'s [2004] search-result-based term translation extraction method, which explores search-result pages utilizing co-occurrence relation and contextual information for extraction of translations of unknown query terms.",
                "cite_spans": [
                    {
                        "start": 34,
                        "end": 55,
                        "text": "Cheng et al.'s [2004]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Search-Result-Based Term Translation Extraction Method",
                "sec_num": "2.3"
            },
            {
                "text": "On the basis of co-occurrence analysis, chi-square test (\u03c72) is adopted to estimate semantic similarity between the source term E and the target translation candidate C. The similarity measure is defined as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(1) Chi-square Test Method",
                "sec_num": null
            },
            {
                "text": "2 2 ( ) ( , C) , ( ) ( ) ( ) ( ) N a d b c S E a b a c b d c d \u03c7 \u00d7 \u00d7 \u2212 \u00d7 = + \u00d7 + \u00d7 + \u00d7 + (1)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(1) Chi-square Test Method",
                "sec_num": null
            },
            {
                "text": "where a, b, c and d are the numbers of pages retrieved from search engines by submitting Boolean queries: \"E and C\", \"E and not C\", \"not E and C\", and \"not E and not C\", respectively; N is the total number of pages, i.e., N = a + b + c + d.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(1) Chi-square Test Method",
                "sec_num": null
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(1) Chi-square Test Method",
                "sec_num": null
            },
            {
                "text": "Due to the property of Chinese-English mixed texts often appearing in Chinese pages, the source term E and the target translation candidate C may share common contextual terms in the search-result pages. The similarity between E and C is computed based on their context feature vectors E cv and C cv in the vector-space model. The conventional tf-idf weighting scheme for each feature term t i in E cv and C cv , E cv = <w e1 , w e2 , \u2026, w em >, and",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2) Context-Vector Analysis Method",
                "sec_num": null
            },
            {
                "text": "C cv = <w c1 , w c2 , \u2026,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2) Context-Vector Analysis Method",
                "sec_num": null
            },
            {
                "text": "w cm >, is used and defined as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2) Context-Vector Analysis Method",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( , ) log( ), max ( , ) i i t j j f t p N w f t p n = \u00d7",
                        "eq_num": "(2)"
                    }
                ],
                "section": "2) Context-Vector Analysis Method",
                "sec_num": null
            },
            {
                "text": "where f(t i , p) is the frequency of term t i in the search-result page p, N is the total number of Web pages, and n is the number of the pages containing t i . Finally, we use the cosine measure to estimate the similarity between E and C as follows: (3)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2) Context-Vector Analysis Method",
                "sec_num": null
            },
            {
                "text": "To effectively extract correct translations for infrequent abbreviated terms, we propose an integrated method in which an abbreviated term is transformed to its full name first, and then we extract the correct translation of the full name using the search-result-based term translation extraction method described above (Section 2.3). In the following, we describe two new proposed methods exploiting search results to extract full names for English and Chinese abbreviations, respectively.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Search-Result-Based Abbreviation Translation Extraction Method",
                "sec_num": "2.4"
            },
            {
                "text": "To deal with the full names for a given English abbreviation, we designed an efficient process of identifying full names, which consists of three major steps based on the hybrid text mining approach proposed by Park and Byrd [2001] . First, we use the contextual terms around an abbreviated term in the search results to extract possible full name candidates. Second, we use occurrence frequency and Part-of-Speech (POS) information of full name candidates to filter out some impossible candidates. Finally, we propose a simple adaptive co-occurrence model",
                "cite_spans": [
                    {
                        "start": 211,
                        "end": 231,
                        "text": "Park and Byrd [2001]",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Extracting Full Names for English Abbreviations",
                "sec_num": "2.4.1"
            },
            {
                "text": "which utilizes several different augmenting and decaying factors in selecting the best full name candidate. More details are described in the following.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "To solve the problem of identifying full names without sufficient texts [Park and Byrd 2001] , we take advantage of Web search results as a corpus. Our idea is to take the given abbreviated term as a search term to fetch the top 200 search result snippets from Google. To extract possible full name candidates by exploring the search result snippets, we utilize contextual information of the abbreviated term in the snippets. These full name candidates must appear in the same snippets with the abbreviated term, and should have a minimum word length between |A|\u00d72 and |A|+5, where |A| is the length of characters of the abbreviated term. In addition, to select more reliable full name candidates, we put a constraint on the identification process in which the first character of the first word of each full name candidate should match the first character of the abbreviated term.",
                "cite_spans": [
                    {
                        "start": 72,
                        "end": 92,
                        "text": "[Park and Byrd 2001]",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Identifying Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Identifying Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "To reduce computation time while extracting many full name candidates, we first select the top 20 frequent full name candidates and then filter out some impossible candidates whose first word or last word are prepositions, be-verbs, modal verbs, conjunctions, or pronouns [Park and Byrd 2001] .",
                "cite_spans": [
                    {
                        "start": 272,
                        "end": 292,
                        "text": "[Park and Byrd 2001]",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2) Filtering Impossible Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "To select the best full name candidates, we propose an adaptive co-occurrence model by employing mutual information as well as four augmenting or decaying factors to compute the similarity between an abbreviated term A and its full name candidates F C .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "(A) Mutual Information: In this step, mutual information is used to compute the similarity between an abbreviated term A and its full name candidate F C . Mutual information is defined as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( , ) ( , ) ( , ) log( ) ( ) ( ) C C C C P A F MI A F P A F P A P F = \u00d7 \u00d7 .",
                        "eq_num": "(4)"
                    }
                ],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "Here P(A, F C ) is the probability of co-occurrence of A and F C . P(A) and P(F C ) are the probabilities of occurrence of A and F C in the Web, respectively. We can get the occurrence frequencies from search engines by submitting queries: \"A\", \"F C \", and \"A and F C \", respectively. To further determine correct full names, we add another augmenting factor to estimate the similarity between an abbreviated term and its full name candidates by adopting a fast and simple character matching method. We use two kinds of character matching: (1) first-letter matching is used to compute the total number N F of matching the first letter of each word in the full name candidate F C with each character in the abbreviated terms, and (2) non-first-letter matching is used to computer the total number N NF of matching the non-first letters of each word in the F C with each character in A. The score of character matching of A and F C is defined as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( ) ,",
                        "eq_num": "( 1 )"
                    }
                ],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "C F N F Overlap A F N N \u03b1 \u03b1 = * + \u2212 * .",
                        "eq_num": "(5)"
                    }
                ],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "Here, the weighting parameter \u03b1 is empirically set to 0.8. Basically, the first-letter matching should be reasonably assigned higher weight for each matching pair. The character similarity is defined as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( , ) ( , ) | | C C Overlap A F CharSim A F A = ,",
                        "eq_num": "(6)"
                    }
                ],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "where | A| is the number of characters of the abbreviated term A.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "The number N LD to represent the difference between character length |A| of the abbreviated term A and word length |F C | of the corresponding full name candidate F C as a decaying factor. N LD is defined as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(D) Difference of Length:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "| | | | LD C N A F = \u2212 .",
                        "eq_num": "(7)"
                    }
                ],
                "section": "(D) Difference of Length:",
                "sec_num": null
            },
            {
                "text": "(E) Number of Stop Words: The number N SW of stop words in the full name candidate F C is also used as a decaying factor.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(D) Difference of Length:",
                "sec_num": null
            },
            {
                "text": "We adaptively integrate the above two augmenting and two decaying factors into the basic co-occurrence model to compute the similarity between A and F C . Our adaptive co-occurrence model is defined as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( , ) ( , ) C Augument AC C Decay MI A F F S A F F \u00d7 = ,",
                        "eq_num": "(8)"
                    }
                ],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "where the augmenting factor F Augument is integrated as",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( ) ( ) 1 , Augument C SC F C h a r S i m AF N \u03b2 = \u00d7 + ;",
                        "eq_num": "(9)"
                    }
                ],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "Infrequent Unknown Abbreviations and Proper Names and the decaying factor F Decay is integrated as",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( ) 2 Decay LD SW F N N \u03b2 = + + .",
                        "eq_num": "(10)"
                    }
                ],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "To avoid the product being zero, here, \u03b2 1 and \u03b2 2 are the adaptable parameters and set to 1 heuristically.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(F) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "Due to language differences between Chinese and English, such as no space delimitation between Chinese words, it is more difficult to identify the full name for a given Chinese abbreviated term. Therefore, we designed a method slightly different from the method of extracting English full names described above. Our Chinese full name extraction method consists of three major steps. First, the possible full name candidates are extracted by using the PAT-tree-based keyword extraction method proposed by Chien [1997] . Second, we use the character similarity between an abbreviated term and its full name candidates to filter out some impossible candidates. Finally, to select the correct Chinese full name, we use the adaptive co-occurrence model (Equation 8) but slightly modify the decaying factors. The following description will explain the different points in more details.",
                "cite_spans": [
                    {
                        "start": 510,
                        "end": 516,
                        "text": "[1997]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Extracting Full Names for Chinese Abbreviations",
                "sec_num": "2.4.2"
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Extracting Full Names for Chinese Abbreviations",
                "sec_num": "2.4.2"
            },
            {
                "text": "To identify the possible full name candidates for a given Chinese abbreviated term A, we adopt a PAT-tree-based keyword extraction method [Chien 1997] to extract Chinese phrases in the search results related to the abbreviated term A as full name candidates. In addition, to select more reliable full name candidates, we put a length constraint on the candidates. These candidates should have more than (|A| +2) characters, where |A| is the number of characters of A.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Identifying Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "(2) Filtering Impossible Full Name Candidates",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Identifying Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "According to our observations, the Chinese full name candidates extracted by the PAT-tree-based keyword extraction method generally have higher reliability. Thus, we just use Equations (5) and (6) with a threshold of character similarity to filter out some impossible candidates.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Identifying Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Identifying Full Name Candidates",
                "sec_num": null
            },
            {
                "text": "Like the above method of selecting the best English full name candidates, we still use the proposed adaptive co-occurrence model (Equation (8)) to select the best Chinese full name candidates. Please note, though, that the processing of augmenting/decaying factors is a little different. For example, we remove the decaying factor of stopword number since most stopwords seldom appear in Chinese full names. Some different points will be described below. Here the Chinese cues \"\u6216\", \"\u4ee3\u8868\", \"\u7c21\u7a31\", \"\u7e2e\u5beb\" correspond to the English words \"or\", \"present\", \"short\", and \"acronym\", respectively.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "(B) Similarity of Character: First, we use the Chinese POS tagger to segment full name candidates. Then, we take character similarity (Equation 5and (6)) as an augmenting factor.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "(C) Difference of Length: Due to the fact that there is no space delimitation between Chinese words, we adopt a Chinese POS tagger 4 to do word segmentation for full name candidates. Then, we use the number N LD to represent the difference between character length |A| of the abbreviated term A and word length |F C | of the corresponding full name candidate F C ; this is considered a decaying factor (Equation 7).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "3) Selecting Best Full Name Candidate",
                "sec_num": null
            },
            {
                "text": "We adopt the same adaptive co-occurrence model (Equation (8)) with two augmenting factors and one decaying factor to compute the similarity between A and F C . The augmenting factors are the same as Equation (9), but the decaying factor in Equation (10) is modified adaptively by removing the stopword number as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(D) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( ) 3 Decay LD F N \u03b2 = + .",
                        "eq_num": "(11)"
                    }
                ],
                "section": "(D) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "To avoid the product being zero, here \u03b2 3 is an adaptable parameter and set to 1, heuristically.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(D) Adaptive Co-occurrence Model:",
                "sec_num": null
            },
            {
                "text": "To improve the performance of unknown term translation extraction for infrequent proper names, we consider integrating name transliteration techniques into the process of translation extraction in order to filter out impossible transliterated name candidates. Our idea is to first extract terms from the search-result snippets as translation candidates (see Section 2.3), and then filter out impossible transliterated name candidates based on the name transliteration model (described in Section 2.5.2). Therefore, in this section we propose a two-stage hybrid translation extraction method, a Web-based transliteration model to deal with transliteration mapping between an English proper name and its corresponding Chinese, and a Web-based",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Search-Result-Based Transliteration Name Extraction Method",
                "sec_num": "2.5"
            },
            {
                "text": "unsupervised learning algorithm to automatically collect diverse English-Chinese transliteration name pairs from Web search results for transliteration model training (Section 2.5.3).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "Our proposed two-stage hybrid translation extraction method is composed of two major steps. First, we use the search-result-based translation extraction method (Section 2.3) to extract k (k = 20) terms with higher similarity scores as transliteration candidates. Second, some impossible candidates included in general-purpose bilingual dictionaries are filtered out, and then the rest of the candidates are ranked according to transliteration similarity with the source proper name, which is computed based on the proposed Web-based transliteration model below (Equation (15)).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Two-Stage Hybrid Translation Extraction",
                "sec_num": "2.5.1"
            },
            {
                "text": "(A) English Syllable Segmentation: Wan and Verspoor [1998] have developed a fully rule-based algorithm to transliterate English proper names into Chinese names. We simplify their syllabification techniques to generate a few simple heuristic rules of segmenting an English name into a sequence of syllables. Each English syllable is regarded as an English transliteration unit (ETU) in this work and has at most one corresponding character of the Chinese transliterated name. Initially, we used only five rules for English syllable segmentation listed below:",
                "cite_spans": [
                    {
                        "start": 35,
                        "end": 58,
                        "text": "Wan and Verspoor [1998]",
                        "ref_id": "BIBREF28"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Filtering Impossible Candidates Using Web-Based Transliteration Model",
                "sec_num": "2.5.2"
            },
            {
                "text": "\u2022 a, e, i, o, u are vowels, and y is also regarded as a vowel if it appears behind a consonant. All other letters are consonants. \u2022 Separate two consecutive vowels except the following cases: ai, au, ee, ea, ie, oa, oo, ou, etc.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Filtering Impossible Candidates Using Web-Based Transliteration Model",
                "sec_num": "2.5.2"
            },
            {
                "text": "\u2022 Separate two consecutive consonants except the following cases: bh, ch, gh, ph, th, wh, ck, cz, zh, zk, ng, sc,ll, tt, etc.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Filtering Impossible Candidates Using Web-Based Transliteration Model",
                "sec_num": "2.5.2"
            },
            {
                "text": "\u2022 l, m, n, r are combined with the prior vowel only if they are not followed by a vowel.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Filtering Impossible Candidates Using Web-Based Transliteration Model",
                "sec_num": "2.5.2"
            },
            {
                "text": "\u2022 A consonant and a following vowel are regarded as an ETU. For example, \"Nokia\" (\uf95d\u57fa\u4e9e) is segmented into three ETUs \"no\", \"ki\", and \"a\", and \"Epson\" \u611b\u666e\u751f ( ) is segmented into three ETUs \"e\", \"p\", and \"son\". Currently, although some English names may be segmented incorrectly, it is easy to manually update new rules to improve English syllable segmentation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Filtering Impossible Candidates Using Web-Based Transliteration Model",
                "sec_num": "2.5.2"
            },
            {
                "text": "To avoid double errors of converting English phonetic representation to Chinese Pinyin and from Pinyin to Chinese characters, in this work, we adopted direct orthographic mapping for name transliteration. We use the probability P(e i , c i ) to estimate the possibility of the mapping between an ETU e i and a Chinese character c i . Additionally, to build an efficient online name transliteration model, we propose a more simple transliteration model. Our Web-based transliteration model is called forward-syllable-mapping transliteration model:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "( , ) ( , ) , ( , ) FSM FSM P EC S EC D E C = (12)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "where P FSM (E, C) is the co-occurrence probability of E and C and defined as",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( ) min( , ) 1 1 1 , [ ( 1 )(, ) ] , m n FSM i i i P EC Pe c \u03b3 \u03b3 = \u2248 \u2212 + \u220f",
                        "eq_num": "(13)"
                    }
                ],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "and \u03b3 1 is the smoothing weight. The decaying factor D(E, C) indicates the number of syllable difference between an English name E and a Chinese transliterated name C and is defined as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( ) , | | D E C m n \u03b5 = + \u2212 .",
                        "eq_num": "(14)"
                    }
                ],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "Here \u03b5 is a decaying parameter, m is the total number of ETUs, and n is the total number of Chinese characters.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "To improve incorrect transliteration mapping between ETUs and Chinese characters while an English-Chinese transliterated name pair with different numbers of transliteration unit, we propose the reverse-syllable-mapping transliteration model to assist in learning more correct mapping, which is defined below:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "( , ) ( , ) , ( , ) RSM RSM P EC S EC D E C = (15) where ( ) 2 ( ) 2 1 2 ( ) 2 1 [(1 ) ( , ) ], ; , [(1 ) ( , ) ]",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": ", . ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "\u23a7 \u2212 + \u2265 \u23aa \u23aa \u2248 \u23a8 \u23aa \u2212 + < \u23aa \u23a9 \u220f \u220f (16)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "Here \u03b3 2 is the smoothing weight and D(E, C) is the same as Equation 14. Our alternative transliteration model will combine the forward-syllable-mapping and reverse-syllable-mapping transliteration model, which is called mixed-syllable-mapping transliteration model, and defined as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( , )",
                        "eq_num": "( , ) ( , )"
                    }
                ],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": ".",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "MSM FSM RSM S EC S EC S EC = \u00d7",
                        "eq_num": "(17)"
                    }
                ],
                "section": "(B) Web-based Transliteration Model:",
                "sec_num": null
            },
            {
                "text": "To deal with the problems of the diversity of Chinese transliterated names to English proper names, we intend to take advantage of abundant language-mixed texts on the Web to collect various English-Chinese transliterated name pairs from the Web and build a an effective online transliteration model. Thus, we designed an unsupervised learning process for English-Chinese transliterated name mapping. The process is composed of three main stages:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Web-Based Unsupervised Learning Algorithm",
                "sec_num": "2.5.3"
            },
            {
                "text": "extraction of Chinese transliterated names, extraction of English original names, and learning of transliterated name mapping. More details are described below and the unsupervised learning algorithm is illustrated as well in Figure 2 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 226,
                        "end": 234,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Web-Based Unsupervised Learning Algorithm",
                "sec_num": "2.5.3"
            },
            {
                "text": "Input: initial transliterated name pair set V ec and a general-purpose bilingual dictionary D.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Web-based Unsupervised Learning Algorithm for Collecting English-Chinese Transliteration Pairs and Training a Transliteration Model",
                "sec_num": null
            },
            {
                "text": "Output: updating V ec and a transliteration model T. 3 Repeat from step1 until the desired number of transliteration pairs is reached.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Web-based Unsupervised Learning Algorithm for Collecting English-Chinese Transliteration Pairs and Training a Transliteration Model",
                "sec_num": null
            },
            {
                "text": "(1) Extraction of Chinese Transliterated Names: Xiao et al. [2002] have proposed a bootstrapping algorithm that uses only five frequent Chinese transliterated characters as initial seed character set: {\u963f, \u723e, \u5df4, \u65af, \u57fa} to automatically collect over 100,000 Chinese transliterated names by utilizing search-result pages. Inspired by this work, we further propose a bootstrapping algorithm to automatically find English-Chinese transliterated name pairs from search-result pages. Initially, we need at least one English-Chinese transliterated name pair containing two frequent Chinese transliterated characters as seed transliteration pair set V ec , e.g., V ec = {(Bush, \u5e03\u5e0c)}. We select two Chinese characters from the Chinese name of the seed pair, and then send them to search engines for getting search-results pages. To efficiently extract more Chinese transliterated names from search-result pages, we use the CKIP tagger (Section 2.4.2), which is a representative Chinese POS tagger and performs well in segmenting Chinese texts into meaningful words and extracting unknown words.",
                "cite_spans": [
                    {
                        "start": 48,
                        "end": 66,
                        "text": "Xiao et al. [2002]",
                        "ref_id": "BIBREF29"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 2. Web-based unsupervised learning algorithm for collecting English-Chinese transliterated name pairs and building a transliteration model",
                "sec_num": null
            },
            {
                "text": "(2) Extraction of English Original Names: We use the proposed two-stage hybrid translation extraction method described above (Section 2.5.1) to find possible English original names.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 2. Web-based unsupervised learning algorithm for collecting English-Chinese transliterated name pairs and building a transliteration model",
                "sec_num": null
            },
            {
                "text": "(3) Learning of Transliterated Name Mapping: On the basis of the rules of English syllable segmentation, we will gradually train an English-Chinese name transliteration model by computing the scores of the transliterated name mapping of the new extracted transliterated name pairs (Equation (17)).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 2. Web-based unsupervised learning algorithm for collecting English-Chinese transliterated name pairs and building a transliteration model",
                "sec_num": null
            },
            {
                "text": "We conducted the following experiments to evaluate the performance of our proposed search-result-based abbreviation translation extraction method and two-stage hybrid translation extraction method.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Results",
                "sec_num": "3."
            },
            {
                "text": "Evaluation Metric: For the following experiments on full name identification of abbreviations and translation of abbreviations, the average top-n inclusion rate is adopted as a metric. For a set of abbreviated terms to be expanded/translated, its top-n inclusion rate was defined as the percentage of the abbreviated terms whose correct full names/translations could be found in the first n extracted full name candidates/translation candidates [Cheng et al. 2004] .",
                "cite_spans": [
                    {
                        "start": 445,
                        "end": 464,
                        "text": "[Cheng et al. 2004]",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Results",
                "sec_num": "3."
            },
            {
                "text": "Correct Translation / Transliteration: The correct translation / transliteration or correct definition is judged by us according to more popular sense in general cases.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Results",
                "sec_num": "3."
            },
            {
                "text": "In this experiment, we intend to compare the performance of our proposed search-result-based abbreviation translation method with that of the search-result-based term translation extraction method proposed by Cheng et al. [2004] .",
                "cite_spans": [
                    {
                        "start": 209,
                        "end": 228,
                        "text": "Cheng et al. [2004]",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation for the Search-Result-Based Abbreviation Translation Extraction Method",
                "sec_num": "3.1"
            },
            {
                "text": "Test data: Four test sets of English abbreviated terms are prepared in the following.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Abbreviations",
                "sec_num": "3.1.1"
            },
            {
                "text": "FA-Dreamer-E: 28 frequent English abbreviated terms which have correct Chinese translations were manually selected from about 20K frequent queries with occurrence frequency over 10 in the Dreamer query log 5 which contains 228,566 unique queries. (The partial test data is listed in Appendix).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Abbreviations",
                "sec_num": "3.1.1"
            },
            {
                "text": "IA-Dreamer-E: 27 infrequent English abbreviated terms (frequency < 3 in Dreamer query log) which have correct Chinese translations were manually selected from infrequent English queries in the Dreamer query log (about 40K entries). (The partial test data is listed in Appendix).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Abbreviations",
                "sec_num": "3.1.1"
            },
            {
                "text": "FA-Wiki-E: 62 popular English abbreviated terms which have correct Chinese translations were manually selected from Wikipedia abbreviation list containing about 4k entries (Section 2.2). (The partial test data is listed in Appendix).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Abbreviations",
                "sec_num": "3.1.1"
            },
            {
                "text": "RA-Wiki-E: 25 English abbreviated terms which have correct Chinese translations were randomly selected from Wikipedia abbreviation list due to the list without frequency information. (The partial test data is listed in Appendix).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Abbreviations",
                "sec_num": "3.1.1"
            },
            {
                "text": "(1) Results for English Full Name Extraction Table 1 shows that our full name extraction method is effective for the test abbreviated terms with various subjects. Our method can achieve the top-1 inclusion rate of over 85% and the top-5 inclusion rate of over 92% for all test sets. Different from existing methods, our full name extraction method is very promising even for infrequent abbreviated terms by utilizing search results from Web search engines. However, some errors still result from the problem of data sparseness. For example, given the abbreviated term \"MPEG\", its correct full name \"Motion Picture Experts Group\" might appear quite rarely in the top 200 search results snippets. Therefore, the correct full name is filtered out by the filtering step and this causes trouble in extracting incorrect full names. Tables 2 to 5 show that the proposed search-result-based abbreviation translation extraction method actually performs better than the previous search-result-based translation extraction method proposed by Cheng et al. For example, for the infrequent English abbreviated queries from the Dreamer query log, the search-result-based abbreviation translation extraction method achieve the top-1 inclusion rate of 48% (see Table 3 ) but the search-result-based translation extraction method achieve the top-1 inclusion rate of 0%. Given the example query \"ISS\", the search-result-based term translation extraction method cannot obtain the correct Chinese translation \"\u570b\u969b\u592a\u7a7a\u7ad9\" among the top five extracted candidates. However, our search-result-based abbreviation translation extraction method can extract the correct full name \"International Space Station\", and then extract correct Chinese translation \"\u570b\u969b\u592a\u7a7a \u7ad9\" via the full name \"International Space Station\". As mentioned in Section 2.1, the reason might be that the abbreviated terms are semantically more ambiguous and co-occur relatively infrequently with the correct translations of their full names.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 45,
                        "end": 52,
                        "text": "Table 1",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 826,
                        "end": 839,
                        "text": "Tables 2 to 5",
                        "ref_id": null
                    },
                    {
                        "start": 1244,
                        "end": 1251,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Abbreviations",
                "sec_num": "3.1.1"
            },
            {
                "text": "To further improve the performance of our search-result-based abbreviation translation extraction method, we intuitively intend to combine our method and Cheng et al.'s method. We expect that such a combination would make both methods mutually complementary by extracting translations from abbreviations and their full names simultaneously. Tables 2 to 5 show that the linear combination method is effective in improving the top-5 inclusion rate. For example, for the abbreviated query \"AOL\", its correct full name \"America Online\" is correctly extracted via our abbreviation expansion method. It fails to find the correct translation among the top five extracted candidates using our search-result-based abbreviation translation method, but the correct translation \"\u7f8e \u570b\u7dda \u4e0a \" can be ranked at third place using the linear combination method. FA-Dreamer-C: 35 frequent Chinese abbreviated terms with correct English translations were manually selected from about 20K frequent queries with occurrence frequency over 10 in the Dreamer query log. (The partial test data is listed in Appendix).",
                "cite_spans": [
                    {
                        "start": 154,
                        "end": 176,
                        "text": "Cheng et al.'s method.",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 341,
                        "end": 354,
                        "text": "Tables 2 to 5",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "(3) Linear Combination Results",
                "sec_num": null
            },
            {
                "text": "IA-Dreamer-C: 28 infrequent Chinese abbreviated terms (frequency < 3 in Dreamer query log) with correct English translations were manually selected from infrequent Chinese queries in the Dreamer query log (about 115K entries). (The partial test data is listed in Appendix).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "(3) Linear Combination Results",
                "sec_num": null
            },
            {
                "text": "(1) Results for Chinese Full Name Extraction Table 6 shows that our Chinese full name extraction method is effective and can achieve top-1 inclusion rate of over 86% for the two test sets. We observed that some errors resulted from incorrect matching between the abbreviated query terms and their highly related full name candidates in the search results. For example, given the abbreviated term \"\u4e2d\u5f71\" (Central Motion Picture Corporation), our method extracted the incorrect full name \"\u4e2d\u570b\u96fb\u5f71\" (Chinese Movie) at first place. Since the correct full name \"\u4e2d\u592e\u96fb\u5f71\u516c\u53f8\" co-occurs infrequently with the abbreviated query term \"\u4e2d\u5f71\" in the search results, it can't be extracted by the PAT-tree-based keyword extraction method. As a result, our method extracted the incorrect full name \"\u4e2d\u570b\u96fb\u5f71\" because the abbreviated term \"\u4e2d\u5f71\" and the incorrect full name candidate \"\u4e2d\u570b\u96fb\u5f71\" have stronger correlation in the search results and higher character similarity. IA-Dreamer-C 86% 89% 89%",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 45,
                        "end": 52,
                        "text": "Table 6",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "(3) Linear Combination Results",
                "sec_num": null
            },
            {
                "text": "( Tables 7 and 8 show that, for the extraction of Chinese abbreviation translation, the proposed search-result-based abbreviation translation extraction method still performs better than the previous search-result-based translation extraction method proposed by Cheng et al. For example, for the infrequent Chinese abbreviated queries from the Dreamer query log, Cheng et al.'s method performs very poorly with a top-5 inclusion rate of 4%, but our method achieves great improvement with the top-5 inclusion rate of 29%. For example, given the Chinese abbreviated query \"\u570b\u5b89\u5c40\", Cheng et al.'s method cannot obtain the correct English translation \"National Security Bureau\" among the top five extracted candidates. However, our method can extract the correct Chinese full name \"\u570b\u5bb6\u5b89\u5168\u5c40\", and then extract the correct English translation \"National Security Bureau\", which is ranked at second place.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 2,
                        "end": 16,
                        "text": "Tables 7 and 8",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "(3) Linear Combination Results",
                "sec_num": null
            },
            {
                "text": "In addition, Table 8 shows that the linear combination method just achieves the same performance as our method, and is unable to further improve the top-n inclusion rates. In fact, we need larger amounts of test data to determine the effectiveness using the linear combination method in the future. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 13,
                        "end": 20,
                        "text": "Table 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "2) Performance Comparison between Search-Result-based Abbreviation Translation Extraction Method and Term Translation Extraction Method",
                "sec_num": null
            },
            {
                "text": "The following two experiments are focused on the evaluation of the performance of extracting translations for infrequent unknown English and Chinese proper names, respectively, using the proposed mixed-syllable-mapping transliteration model and the two-stage hybrid translation extraction method. IP-Dreamer-E: 41 infrequent unknown English proper names (frequency < 3 in the query log) are manually selected from the Dreamer query log. (The partial test data is listed in Appendix).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation for the Two-Stage Hybrid Translation Extraction Method",
                "sec_num": "3.2"
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Translation Extraction Results for English Proper Names",
                "sec_num": "3.2.1"
            },
            {
                "text": "According to the results shown in Tables 9 and 10, we can obtain the following findings. For the two test sets, the proposed two-stage hybrid translation extraction method made great improvements compared with the search-result-based translation extraction method and the general name transliteration method [Wan and Verspoor 1998; Knight and Graehl 1998; Lin and Chen 2002; Virga and Khudanpur 2003; Gao et al. 2004; Li et al. 2004] . In this work, we just use our proposed transliteration model as a \"Name Transliteration\" method for performance comparison. For example, the two-stage hybrid translation extraction method can achieve the top-1 inclusion rate of 41% (Table 10) for infrequent unknown English proper names, but the search-result-based translation extraction method only achieved 17%. The main reason is that most of the incorrect translation candidates extracted via the search-result-based translation extraction method can be filtered out based on our mixed-syllable-mapping transliteration model. For example, given the English proper name \"Pamela\", the correct Chinese transliterated name \"\u6f58\u871c\uf925\" can be extracted and ranked at second place (see Table  11 ).",
                "cite_spans": [
                    {
                        "start": 308,
                        "end": 331,
                        "text": "[Wan and Verspoor 1998;",
                        "ref_id": "BIBREF28"
                    },
                    {
                        "start": 332,
                        "end": 355,
                        "text": "Knight and Graehl 1998;",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 356,
                        "end": 374,
                        "text": "Lin and Chen 2002;",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 375,
                        "end": 400,
                        "text": "Virga and Khudanpur 2003;",
                        "ref_id": null
                    },
                    {
                        "start": 401,
                        "end": 417,
                        "text": "Gao et al. 2004;",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 418,
                        "end": 433,
                        "text": "Li et al. 2004]",
                        "ref_id": "BIBREF13"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 668,
                        "end": 678,
                        "text": "(Table 10)",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 1165,
                        "end": 1174,
                        "text": "Table  11",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "1) Two-Stage Hybrid Translation Extraction Method vs. Search-Result-based Term Translation Extraction Method",
                "sec_num": null
            },
            {
                "text": "(",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "1) Two-Stage Hybrid Translation Extraction Method vs. Search-Result-based Term Translation Extraction Method",
                "sec_num": null
            },
            {
                "text": "Tables 9 and 10 also demonstrate that the simple linear combination method obtained slight improvement on transliterated name performance since the general name transliteration method is still limited in generating correct transliteration candidates. However, note that for many English-Chinese transliteration pairs with different numbers of transliteration units, the mixed-syllable-mapping transliteration model is still effective to learn correct transliteration mapping between English syllables and Chinese characters. Table 12 shows that our two-stage hybrid translation extraction method obtains the top-1 inclusion rate of 64%. Surprisingly, it performs worse than the search-result-based translation extraction method at 70%. This means that our candidate filtering method based on our trained Web-based transliteration model is unable to improve the performance of extracting translations for frequent unknown Chinese proper names in Web queries. We will investigate the possible reasons in the following discussion. However, for the test set of infrequent unknown Chinese proper names, the two-stage hybrid translation extraction method made effective improvements compared with the search-result-based translation extraction method (Table 13 ). For example, the two-stage hybrid translation extraction method can achieve the top-1 inclusion rate of 46% for infrequent unknown Chinese proper names, whereas the search-result-based translation extraction method only achieved 27%. It shows that most of the incorrect translation candidates extracted via the search-result-based translation extraction method can be filtered out using our mixed-syllable-mapping transliteration model. For example, given the Chinese transliterated name \"\u827e\uf9f7\u514b\", its correct English original name \"Eric\" can be extracted and ranked at first place (Table 14) . According to our further analyses of the results shown in Tables 12 and 13, we obtain the following interesting findings.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 525,
                        "end": 533,
                        "text": "Table 12",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 1245,
                        "end": 1254,
                        "text": "(Table 13",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 1837,
                        "end": 1847,
                        "text": "(Table 14)",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "2) Linear Combination Results",
                "sec_num": null
            },
            {
                "text": "Our test set FP-Dreamer-C (frequent unknown Chinese transliterated terms) contains a number of company names, e.g., \"\u92b3\u8dd1\" (Reebok) and \"\u65b0\uf92a\" (Sina). In fact, these Chinese characters like \"\u92b3\", \"\u8dd1\", and \"\uf92a\" are rarely used as transliterated characters in general cases. Thus, these characters are certainly difficult to be matched with those possibly correct ETUs since they have never appeared in the training data of our collected English-Chinese transliterated name pairs from search-result pages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "The probabilities of some correct transliteration mapping between Chinese characters and English ETUs are lower than those of incorrect transliteration mapping trained from incorrect or partial matching transliteration pairs. However, our training data of about 10k potential transliterated name pairs extracted via our Web-based unsupervised learning algorithm should contain a number of incorrect transliteration mapping pairs and still be insufficient to build a good-quality transliteration model.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "The search-result-based term translation extraction method perform well for the test set of frequent unknown Chinese proper names while our two-stage hybrid translation extraction method is effective in improving the translation performance for infrequent unknown Chinese proper names. Therefore, we consider adding the information of term occurrence frequency in the query log into the process of unknown term translation. For a query with frequent Chinese proper names in the query log, we can use the previous search-result-based term translation extraction method to translate it. On the other hand, for queries with infrequent Chinese transliterated terms, we can use the proposed two-stage hybrid translation extraction method to translate them.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "However, utilizing Web search results to translate unknown terms would lead to only partial representative candidates, which are the most popular ones. Therefore, we should continuously collect much more English-Chinese transliterated name pairs for training a better transliteration model in the future, and at the same time improve the techniques of extracting and filtering English name candidates to further collect larger amounts of correct transliterated name pairs for building a high quality transliteration model. In addition, there are still a number of cases which are difficult to be dealt with by using the simple mixed-syllable-mapping transliteration model and need to be further improved in the future.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Infrequent Unknown Abbreviations and Proper Names",
                "sec_num": null
            },
            {
                "text": "In previous works on identifying full names of abbreviations, AFP (Acronym Finding Program) [Taghva and Gilbreth 1995] used free texts to find English abbreviations and their full names. Park and Byrd [2001] used contextual information around abbreviations to extract potential full name candidates based on their pre-defined rules. However, these methods might suffer from the problem of insufficient texts. Our proposed method exploiting search results can extract English full names for abbreviations in various domains, and then effectively extract correct Chinese translations via their full names.",
                "cite_spans": [
                    {
                        "start": 92,
                        "end": 118,
                        "text": "[Taghva and Gilbreth 1995]",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 187,
                        "end": 207,
                        "text": "Park and Byrd [2001]",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "4."
            },
            {
                "text": "Also, Leah et al. [2000] tried to find full name candidates from a small number of Web pages, and they used lots of syntax rules to select full name candidates of English acronyms. Instead of using many syntax rules, we propose an adaptive co-occurrence model to select the best full name candidates based on the co-occurrence relation and the integration of several augmenting and decaying factors.",
                "cite_spans": [
                    {
                        "start": 6,
                        "end": 24,
                        "text": "Leah et al. [2000]",
                        "ref_id": "BIBREF12"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "4."
            },
            {
                "text": "For name transliteration between Latin-alphabet languages and some Asian languages with different writing forms, such as English and Chinese, researchers have proposed phoneme-based mapping techniques [Knight and Graehl 1998; Lin and Chen 2002; Meng et al. 2001] . Lin et al. [2003] proposed a statistical transliteration model and apply the model to extract English proper names and their Chinese transliterated names in a parallel corpus with high average precision and recall rates. However, Li et al. [2004] pointed out that the transliteration precision of the phoneme-based approaches could be limited by two main constraints. First, Latin-alphabet foreign names from different origins have different phonic rules, such as French and English. Second, transforming English words to Chinese characters will need two steps: transforming from phonemic representation to Chinese Pinyin and from Pinyin to Chinese characters. Two cascaded transforming steps may cause double errors. To avoid this problem, we propose a Web-based mixed-syllable-mapping transliteration model for dealing with online English-Chinese name transliteration based on the concept of direct orthographic mapping.",
                "cite_spans": [
                    {
                        "start": 201,
                        "end": 225,
                        "text": "[Knight and Graehl 1998;",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 226,
                        "end": 244,
                        "text": "Lin and Chen 2002;",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 245,
                        "end": 262,
                        "text": "Meng et al. 2001]",
                        "ref_id": "BIBREF19"
                    },
                    {
                        "start": 265,
                        "end": 282,
                        "text": "Lin et al. [2003]",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 495,
                        "end": 511,
                        "text": "Li et al. [2004]",
                        "ref_id": "BIBREF13"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "4."
            },
            {
                "text": "Both Cheng et al. [2004] and Zhang and Vines [2004] have exploited language-mixed search-result pages for extracting translations of frequent unknown queries. Moreover, Huang et al. [2005] takes advantage of cross-language query expansion to retrieve more relevant search-result pages and then extract translations by combining with phonetic, semantic and frequency-distance features. However, these methods haven't solved the problems of translation extraction for infrequent unknown abbreviations and proper names. Currently, our search-result-based methods presented in this paper can effectively mitigate such kinds of translation problems.",
                "cite_spans": [
                    {
                        "start": 5,
                        "end": 24,
                        "text": "Cheng et al. [2004]",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 29,
                        "end": 51,
                        "text": "Zhang and Vines [2004]",
                        "ref_id": "BIBREF31"
                    },
                    {
                        "start": 169,
                        "end": 188,
                        "text": "Huang et al. [2005]",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "4."
            },
            {
                "text": "In this paper we presented two new search-result-based methods to extract unknown term translation based on the previous method proposed by Cheng et al., including the search-result-based abbreviation translation extraction method and the two-stage hybrid translation extraction method. Our experimental results demonstrate the effectiveness of improving translation extraction for infrequent unknown abbreviations and proper names. Additionally, our proposed adaptive co-occurrence model is effective in aiding the process of selecting the correct full name candidates for the best abbreviated terms. However, currently, Infrequent Unknown Abbreviations and Proper Names the search-result-based abbreviation translation extraction method can perform well in the first stage of extracting the full names of those test abbreviated terms but can hardly extract correct translations via the extracted full names in the second stage. In the future, we are investigating to integrate the cross-language query expansion techniques proposed by Huang et al. into our search-result-based abbreviation translation extraction method.",
                "cite_spans": [
                    {
                        "start": 140,
                        "end": 153,
                        "text": "Cheng et al.,",
                        "ref_id": null
                    },
                    {
                        "start": 1037,
                        "end": 1049,
                        "text": "Huang et al.",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "5."
            },
            {
                "text": "As for the two-stage hybrid translation extraction method, we will continuously collect larger amounts of English-Chinese transliterated name pairs via our proposed Web-based unsupervised learning algorithm to build a more reliable transliteration model. In the future, referring to the methods proposed by both Lam et al. [2004] and Huang et al. [2005] , we will extend our method by involving both semantic and phonetic information and expect that it can be more robust in extracting translations of unknown proper names.",
                "cite_spans": [
                    {
                        "start": 312,
                        "end": 329,
                        "text": "Lam et al. [2004]",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 334,
                        "end": 353,
                        "text": "Huang et al. [2005]",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "5."
            },
            {
                "text": "http://ws.csie.ncku.edu.tw/~jhlin/cgi-bin/index.htm 2 http://livetrans.iis.sinica.edu.tw/: This website is developed based on the search-result-based term translation extraction method by Web Knowledge Discovery lab of Academia Sinica, Taiwan.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "http://en.wikipedia.org/wiki/List_of_acronyms_and_initialisms",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "http://ckipsvr.iis.sinica.edu.tw/demo.htm, which is a Chinese POS tagger developed by Chinese Knowledge and Information Processing group of Academia Sinica.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "http://www.dreamer.com.tw, which was a popular Chinese search engine and is closed now.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Phrasal Translation and Query Expansion Techniques for Cross-Language Information Retrieval",
                "authors": [
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Ballesteros",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [
                            "B"
                        ],
                        "last": "Croft",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "Proc. of 20th International ACM SIGIR Conference on Research and Development in Information Retrieval",
                "volume": "",
                "issue": "",
                "pages": "84--91",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ballesteros, L. and W. B. Croft, \"Phrasal Translation and Query Expansion Techniques for Cross-Language Information Retrieval,\" In Proc. of 20th International ACM SIGIR Conference on Research and Development in Information Retrieval, 1997, Philadelphia, USA, pp. 84-91.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "The Mathematics of Machine Translation",
                "authors": [
                    {
                        "first": "P",
                        "middle": [
                            "F"
                        ],
                        "last": "Brown",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "A D"
                        ],
                        "last": "Pietra",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [
                            "D J"
                        ],
                        "last": "Pietra",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [
                            "L"
                        ],
                        "last": "Mercer",
                        "suffix": ""
                    }
                ],
                "year": 1993,
                "venue": "Computational Linguistics",
                "volume": "19",
                "issue": "2",
                "pages": "263--312",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Brown, P. F., S. A. D. Pietra, V. D. J. Pietra and R. L. Mercer, \"The Mathematics of Machine Translation,\" Computational Linguistics, 19(2), 1993, pp. 263-312.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Base noun phrase translation using Web data and the EM algorithm",
                "authors": [
                    {
                        "first": "Y.-B",
                        "middle": [],
                        "last": "Cao",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proc. of COLING",
                "volume": "",
                "issue": "",
                "pages": "127--133",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Cao, Y.-B. and H. Li, \"Base noun phrase translation using Web data and the EM algorithm,\" In Proc. of COLING, 2002, Taipei, Taiwan, pp. 127-133.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Translating unknown queries with web corpora for cross-language information retrieval",
                "authors": [
                    {
                        "first": "P.-J",
                        "middle": [],
                        "last": "Cheng",
                        "suffix": ""
                    },
                    {
                        "first": "J.-W",
                        "middle": [],
                        "last": "Teng",
                        "suffix": ""
                    },
                    {
                        "first": "R.-C",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "J.-H",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "W.-H",
                        "middle": [],
                        "last": "Lu",
                        "suffix": ""
                    },
                    {
                        "first": "L.-F",
                        "middle": [],
                        "last": "Chien",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proc. of 27th International ACM SIGIR Conference on Research and Development in Information Retrieval",
                "volume": "",
                "issue": "",
                "pages": "146--153",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Cheng, P.-J., J.-W. Teng, R.-C. Chen, J.-H. Wang, W.-H. Lu, L.-F. Chien, \"Translating unknown queries with web corpora for cross-language information retrieval,\" In Proc. of 27th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2004, The University of Sheffield, UK, pp. 146-153.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Free Resources and Advanced Alignment for Cross-Language Text Retrieval",
                "authors": [
                    {
                        "first": "M",
                        "middle": [
                            "W"
                        ],
                        "last": "Davis",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [
                            "C"
                        ],
                        "last": "Ogden",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proc. of the Sixth Text Retrieval Conference",
                "volume": "",
                "issue": "",
                "pages": "385--394",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Davis, M. W. and W. C. Ogden, \"Free Resources and Advanced Alignment for Cross-Language Text Retrieval,\" In Proc. of the Sixth Text Retrieval Conference (TREC 6), 1998, Gaithersburg, Maryland, pp. 385-394.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "An IR approach for translating new words from nonparallel, comparable texts",
                "authors": [
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Fung",
                        "suffix": ""
                    },
                    {
                        "first": "L.-Y.",
                        "middle": [],
                        "last": "Yee",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proc. of 36th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "414--420",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Fung, P. and L.-Y. Yee, \"An IR approach for translating new words from nonparallel, comparable texts,\" In Proc. of 36th Annual Meeting of the Association for Computational Linguistics, 1998, Montreal, Quebec, Canada, pp. 414-420.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Querying across Languages: A Dictionary-based Approach to Multilingual Information Retrieval",
                "authors": [
                    {
                        "first": "D",
                        "middle": [
                            "A"
                        ],
                        "last": "Hull",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Grefenstette",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "Proc. of 19th International ACM SIGIR Conference on Research and Development in Information Retrieval",
                "volume": "",
                "issue": "",
                "pages": "49--57",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hull, D. A. and G. Grefenstette, \"Querying across Languages: A Dictionary-based Approach to Multilingual Information Retrieval,\" In Proc. of 19th International ACM SIGIR Conference on Research and Development in Information Retrieval, 1996, Zurich, Switzerland, pp. 49-57.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Phoneme-based Transliteration of Foreign Name for OOV Problem",
                "authors": [
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "K.-F",
                        "middle": [],
                        "last": "Wong",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Lam",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proc. of the first International Joint Conference on Natural Language Processing (IJCNLP)",
                "volume": "",
                "issue": "",
                "pages": "274--381",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gao, W., K.-F. Wong and W. Lam, \"Phoneme-based Transliteration of Foreign Name for OOV Problem\" In Proc. of the first International Joint Conference on Natural Language Processing (IJCNLP), 2004, Hainan Island, China, pp. 274-381.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Mining Key Phrase Translations from Web Corpora",
                "authors": [
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Huang",
                        "suffix": ""
                    },
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Vogel",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proc. of HLT-EMNLP",
                "volume": "",
                "issue": "",
                "pages": "483--490",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Huang, F., Y. Zhang and S. Vogel, \"Mining Key Phrase Translations from Web Corpora,\" In Proc. of HLT-EMNLP, 2005, Vancouver, B.C., Canada, pp. 483-490.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Introduction to the special issue on the web as corpus",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Kilgarriff",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Grefenstette",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Computational Linguistics",
                "volume": "29",
                "issue": "3",
                "pages": "333--348",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kilgarriff, A. and G. Grefenstette, \"Introduction to the special issue on the web as corpus,\" Computational Linguistics, 29(3), 2003, pp. 333-348.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Machine Transliteration",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Knight",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Graehl",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Computational Linguistics",
                "volume": "24",
                "issue": "4",
                "pages": "599--612",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Knight, K. and J. Graehl, \"Machine Transliteration,\" Computational Linguistics 24(4), 1998, pp. 599-612.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Learning phonetic similarity for matching named entity translations and mining new translations",
                "authors": [
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Lam",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Huang",
                        "suffix": ""
                    },
                    {
                        "first": "P.-S",
                        "middle": [],
                        "last": "Cheung",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proc. of 27th International ACM SIGIR Conference on Research and Development in Information Retrieval",
                "volume": "",
                "issue": "",
                "pages": "281--288",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lam, W., R. Huang, P.-S. Cheung, \"Learning phonetic similarity for matching named entity translations and mining new translations,\" In Proc. of 27th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2004, The University of Sheffield, UK, pp. 281-288.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Acrophile: An Automated Acronym Extractor and Server",
                "authors": [
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Leah",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Ogilvie",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Price",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Tamilio",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proc. of the 5th ACM Digital Libraries Conference",
                "volume": "",
                "issue": "",
                "pages": "205--214",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Leah, L., P. Ogilvie, A. Price, and B. Tamilio, \"Acrophile: An Automated Acronym Extractor and Server,\" In Proc. of the 5th ACM Digital Libraries Conference, 2000, San Antonio, TX, pp. 205-214.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "A Joint Source-Channel Model for Machine Transliteration",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Su",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proc. of 42th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "160--167",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Li, H., M. Zhang and J. Su, \"A Joint Source-Channel Model for Machine Transliteration,\" In Proc. of 42th Annual Meeting of the Association for Computational Linguistics, 2004, Forum Convention Centre, Barcelona, pp. 160-167.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Word-Transliteration Alignment",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "C.-C",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "J.-S",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proc. of ROCLING XV",
                "volume": "",
                "issue": "",
                "pages": "1--16",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lin, T., C.-C. Wu, J.-S. Chang, \"Word-Transliteration Alignment,\" In Proc. of ROCLING XV, 2003, Hsinchu, Taiwan, pp. 1-16.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Backward machine transliteration by learning phonetic similarity",
                "authors": [
                    {
                        "first": "W.-H",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "H.-H",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proc. of CONLL",
                "volume": "",
                "issue": "",
                "pages": "139--145",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lin, W.-H. and H.-H. Chen, \"Backward machine transliteration by learning phonetic similarity,\" In Proc. of CONLL, 2002, Taipei, Taiwan, pp. 139-145.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Translation of Web Queries using Anchor Text Mining",
                "authors": [
                    {
                        "first": "W.-H",
                        "middle": [],
                        "last": "Lu",
                        "suffix": ""
                    },
                    {
                        "first": "L.-F",
                        "middle": [],
                        "last": "Chien",
                        "suffix": ""
                    },
                    {
                        "first": "H.-J",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "ACM Transactions on Asian Language Information Processing",
                "volume": "1",
                "issue": "2",
                "pages": "159--172",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lu, W.-H., L.-F., Chien, H.-J. Lee, \"Translation of Web Queries using Anchor Text Mining,\" ACM Transactions on Asian Language Information Processing, 1(2), 2002, pp. 159-172.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "A Bottom-up Merging Algorithm for Chinese Unknown Word Extraction",
                "authors": [
                    {
                        "first": "W.-Y",
                        "middle": [],
                        "last": "Ma",
                        "suffix": ""
                    },
                    {
                        "first": "K.-J",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proc. of ACL workshop on Chinese Language Processing",
                "volume": "",
                "issue": "",
                "pages": "31--38",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ma, W.-Y. and K.-J. Chen, \"A Bottom-up Merging Algorithm for Chinese Unknown Word Extraction,\" In Proc. of ACL workshop on Chinese Language Processing, 2003, pp. 31-38.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Models of translational equivalence among words",
                "authors": [
                    {
                        "first": "I",
                        "middle": [
                            "D"
                        ],
                        "last": "Melamed",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Computational Linguistics",
                "volume": "26",
                "issue": "2",
                "pages": "221--249",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Melamed, I. D., \"Models of translational equivalence among words,\" Computational Linguistics, 26(2), 2000, pp. 221-249.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Generate Phonetic Cognates to Handle Name Entities in English-Chinese Cross-Language Spoken Document Retrieval",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Meng",
                        "suffix": ""
                    },
                    {
                        "first": "W.-K",
                        "middle": [],
                        "last": "Lo",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Tang",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proc. of Automatic Speech Recognition and Understanding Workshop (ASRU)",
                "volume": "",
                "issue": "",
                "pages": "311--314",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Meng, H., W.-K. Lo, B. Chen and K. Tang, \"Generate Phonetic Cognates to Handle Name Entities in English-Chinese Cross-Language Spoken Document Retrieval,\" In Proc. of Automatic Speech Recognition and Understanding Workshop (ASRU), 2001, Italy, pp. 311-314.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Cross-language Information Retrieval Based on Parallel Texts and Automatic Mining of Parallel Texts from the Web",
                "authors": [
                    {
                        "first": "J.-Y",
                        "middle": [],
                        "last": "Nie",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Isabelle",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Simard",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Durand",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. of 22th International ACM SIGIR Conference on Research and Development in Information Retrieval",
                "volume": "",
                "issue": "",
                "pages": "74--81",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nie, J.-Y., P. Isabelle, M. Simard, and R. Durand, \"Cross-language Information Retrieval Based on Parallel Texts and Automatic Mining of Parallel Texts from the Web,\" In Proc. of 22th International ACM SIGIR Conference on Research and Development in Information Retrieval, 1999, University of California, Berkeley, pp. 74-81..",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Hybrid text mining for finding abbreviations and their definitions",
                "authors": [
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Park",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [
                            "J"
                        ],
                        "last": "Byrd",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proc. of Conference on Empirical Methods in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "126--133",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Park, Y. and R. J. Byrd, \"Hybrid text mining for finding abbreviations and their definitions,\" In Proc. of Conference on Empirical Methods in Natural Language Processing (EMNLP), 2001, pp. 126-133.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "Infrequent Unknown Abbreviations and Proper Names",
                "authors": [],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Infrequent Unknown Abbreviations and Proper Names",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Automatic identification of word translations from unrelated English and German corpora",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Rapp",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. of 37th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "519--526",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rapp, R., \"Automatic identification of word translations from unrelated English and German corpora,\" In Proc. of 37th Annual Meeting of the Association for Computational Linguistics, 1999, College Park, Maryland, USA, pp. 519-526.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "Mining the Web for Bilingual Text",
                "authors": [
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Resnik",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proceedings of the 37th Annual Meeting of the Association for Computational Linguistic",
                "volume": "",
                "issue": "",
                "pages": "527--534",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Resnik, P., \"Mining the Web for Bilingual Text,\" In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistic,.1999, College Park, Maryland, USA, pp. 527-534.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Translating collocations for bilingual lexicons: a statistical approach",
                "authors": [
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Smadja",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Mckeown",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Hatzivassiloglou",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "Computational Linguistics",
                "volume": "22",
                "issue": "1",
                "pages": "1--38",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Smadja, F., K. McKeown, and V. Hatzivassiloglou, \"Translating collocations for bilingual lexicons: a statistical approach,\" Computational Linguistics, 22(1), 1996, pp. 1-38.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Recognizing Acronyms and their Definitions",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Taghva",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Gilbreth",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "UNLV",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Taghva, K. and J. Gilbreth, \"Recognizing Acronyms and their Definitions. Technical Report 95-03,\" ISRI (Information Science Research Institute), UNLV, June, 1995.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Transliteration of Proper Names in Cross-Lingual Information Retrieval",
                "authors": [
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Virga",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Khudanpur",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "ACL 2003 workshop MLNER",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Virga, P. and S. Khudanpur, \"Transliteration of Proper Names in Cross-Lingual Information Retrieval,\" ACL 2003 workshop MLNER.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "Automatic English-Chinese name transliteration for development of multilingual resources",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Wan",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [
                            "M"
                        ],
                        "last": "Verspoor",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proc. of 36th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "1352--1357",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wan, S. and C. M. Verspoor, \"Automatic English-Chinese name transliteration for development of multilingual resources,\" In Proc. of 36th Annual Meeting of the Association for Computational Linguistics, 1998, Montreal, Quebec, Canada, pp. 1352-1357.",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "Extracting pronunciation-translated names from Chinese texts using bootstrapping approach",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Xiao",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "T.-S",
                        "middle": [],
                        "last": "Chua",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proc. of the 1st SIGHAN workshop on Chinese Language Processing",
                "volume": "",
                "issue": "",
                "pages": "1--6",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Xiao, J., J. Liu and T.-S. Chua, \" Extracting pronunciation-translated names from Chinese texts using bootstrapping approach,\" In Proc. of the 1st SIGHAN workshop on Chinese Language Processing, 2002, Taipei, Taiwan, pp. 1-6.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Automatic Construction of English/Chinese Parallel Corpora",
                "authors": [
                    {
                        "first": "C",
                        "middle": [
                            "C"
                        ],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [
                            "W"
                        ],
                        "last": "Li",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Journal of the American society for Information Science and Technology",
                "volume": "54",
                "issue": "8",
                "pages": "730--742",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yang, C. C. and K. W. Li, \"Automatic Construction of English/Chinese Parallel Corpora, \" Journal of the American society for Information Science and Technology, 54(8), 2003, pp. 730-742.",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Using the web for automated translation extraction in cross-language information retrieval",
                "authors": [
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Vines",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proc. of 27th International ACM SIGIR Conference on Research and Development in Information Retrieval",
                "volume": "",
                "issue": "",
                "pages": "162--169",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zhang, Y. and P. Vines, \"Using the web for automated translation extraction in cross-language information retrieval,\" In Proc. of 27th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2004, The University of Sheffield, UK, pp. 162-169.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF1": {
                "type_str": "figure",
                "uris": null,
                "text": "To augment the identification of full names, we utilize the information of orthographic and syntactic structure. N SC indicates the number of abbreviation-full-name pairs appearing in the same snippets. Several frequent patterns of abbreviation-full-name pair are used as the syntactic cues[Park and Byrd 2001], including: abbreviation (full name) full name (abbreviation) abbreviation, or full name full name, or abbreviation full name, abbreviation for short abbreviation \u2026 stands/short/acronym \u2026full name (C) Similarity of Character:",
                "num": null
            },
            "FIGREF2": {
                "type_str": "figure",
                "uris": null,
                "text": "A) Syntactic Cues: We also manually choose several syntactic patterns of Chinese abbreviation-full name pairs as the augmenting factor: abbreviation (full name) full name (abbreviation) abbreviation, \u6216 full name full name, \u6216 abbreviation abbreviation \u2026 \u4ee3\u8868/\u7c21\u7a31/\u7e2e\u5beb \u2026full name",
                "num": null
            },
            "FIGREF4": {
                "type_str": "figure",
                "uris": null,
                "text": "-Dreamer-C: 28 frequent unknown Chinese proper names are obtained from the transliterated terms of the frequent unknown English proper name set FP-Dreamer-E (described in Section 3.2.1). (The partial test data is listed in the Appendix). IP-Dreamer-C: 41 infrequent unknown Chinese proper names are obtained from the transliterated terms of the infrequent unknown English proper name set IP-Dreamer-E (described in Section 3.2.1). (The partial test data is listed in the Appendix). (1) Two-Stage Hybrid Translation Extraction Method vs. Search-Result-based Term Translation Extraction Method",
                "num": null
            },
            "TABREF0": {
                "content": "<table><tr><td>V c ,, perform the following sub-steps:</td></tr><tr><td>2.1 Two-Stage hybrid translation extraction</td></tr><tr><td>2.1.1 English name candidate extraction: use search-result-based term translation</td></tr><tr><td>extraction method to find English name candidates (see Section 2.3).</td></tr><tr><td>2.1.2 2.2 Learning of transliterated name mapping: update T by computing the scores of</td></tr><tr><td>transliterated name mapping of the new extracted transliterated name pairs (Equation</td></tr><tr><td>(17)).</td></tr></table>",
                "type_str": "table",
                "text": "English name candidate filtering: first filter out impossible English name candidates included in D; second, compute transliteration mapping scores based on the English syllable segmentation rules and the name transliteration model T; third, choose the candidates with the highest scores as the possible English original names. Update V ec by adding the new transliterated name pairs extracted.",
                "num": null,
                "html": null
            },
            "TABREF1": {
                "content": "<table><tr><td>Test Set</td><td>Top-1</td><td>Inclusion Rates Top-3</td><td>Top-5</td></tr><tr><td>FA-Dreamer-E</td><td>93%</td><td>96%</td><td>96%</td></tr><tr><td>IA-Dreamer-E</td><td>85%</td><td>96%</td><td>96%</td></tr><tr><td>FA-Wiki-E</td><td>90%</td><td>94%</td><td>94%</td></tr><tr><td>RA-Wiki-E</td><td>88%</td><td>88%</td><td>92%</td></tr></table>",
                "type_str": "table",
                "text": "",
                "num": null,
                "html": null
            },
            "TABREF2": {
                "content": "<table><tr><td>Translation Extraction Method</td><td colspan=\"3\">Inclusion Rates</td></tr><tr><td/><td colspan=\"3\">Top-1 Top-3 Top-5</td></tr><tr><td>Search-result-based Translation Extraction Method</td><td>43%</td><td>54%</td><td>57%</td></tr><tr><td>Search-result-based Abbreviation Translation Extraction Method</td><td>75%</td><td>82%</td><td>86%</td></tr><tr><td>Linear Combination</td><td>71%</td><td>82%</td><td>93%</td></tr></table>",
                "type_str": "table",
                "text": "Two test sets of Chinese abbreviated terms are prepared in the following.",
                "num": null,
                "html": null
            },
            "TABREF3": {
                "content": "<table><tr><td>Test Set</td><td/><td>Inclusion Rates</td><td/></tr><tr><td/><td>Top-1</td><td>Top-3</td><td>Top-5</td></tr><tr><td>FA-Dreamer-C</td><td>94%</td><td>100%</td><td>100%</td></tr></table>",
                "type_str": "table",
                "text": "",
                "num": null,
                "html": null
            },
            "TABREF4": {
                "content": "<table><tr><td>Translation Extraction Method</td><td colspan=\"3\">Inclusion Rates</td></tr><tr><td/><td colspan=\"3\">Top-1 Top-3 Top-5</td></tr><tr><td>Search-result-based Translation Extraction Method</td><td>17%</td><td>46%</td><td>54%</td></tr><tr><td>Search-result-based Abbreviation Translation Extraction Method</td><td>40%</td><td>66%</td><td>71%</td></tr><tr><td>Linear Combination</td><td>49%</td><td>63%</td><td>71%</td></tr><tr><td colspan=\"4\">Table 8. Inclusion rates on translation of infrequent Chinese abbreviations from</td></tr><tr><td>Dreamer query log</td><td/><td/><td/></tr><tr><td>Translation Extraction Method</td><td colspan=\"3\">Inclusion Rates</td></tr><tr><td/><td>Top-1</td><td>Top-3</td><td>Top-5</td></tr><tr><td>Search-result-based Translation Extraction Method</td><td>4%</td><td>4%</td><td>4%</td></tr><tr><td>Search-result-based Abbreviation Translation Extraction Method</td><td>11%</td><td>21%</td><td>29%</td></tr><tr><td>Linear Combination</td><td>11%</td><td>21%</td><td>29%</td></tr></table>",
                "type_str": "table",
                "text": "",
                "num": null,
                "html": null
            },
            "TABREF6": {
                "content": "<table><tr><td>Translation Extraction Method</td><td>Top-1</td><td>Inclusion Rates Top-3</td><td>Top-5</td></tr><tr><td>Search-result-based Translation Extraction Method</td><td>32%</td><td>71%</td><td>82%</td></tr><tr><td>Name Transliteration</td><td>11%</td><td>18%</td><td>21%</td></tr><tr><td>Linear Combination</td><td>32%</td><td>50%</td><td>86%</td></tr><tr><td>Two-Stage Hybrid Translation Extraction Method</td><td>61%</td><td>64%</td><td>68%</td></tr><tr><td colspan=\"4\">Table 10. Inclusion rates on translation of infrequent unknown English proper</td></tr><tr><td>names from Dreamer query log</td><td/><td/><td/></tr><tr><td>Translation Extraction Method</td><td>Top-1</td><td>Inclusion Rates Top-3</td><td>Top-5</td></tr><tr><td>Search-result-based Translation Extraction Method</td><td>17%</td><td>32%</td><td>37%</td></tr><tr><td>Name Transliteration</td><td>15%</td><td>15%</td><td>17%</td></tr><tr><td>Linear Combination</td><td>17%</td><td>37%</td><td>44%</td></tr><tr><td>Two-Stage Hybrid Translation Extraction Method</td><td>41%</td><td>46%</td><td>46%</td></tr></table>",
                "type_str": "table",
                "text": "",
                "num": null,
                "html": null
            },
            "TABREF7": {
                "content": "<table><tr><td/><td>Translation Extraction Method</td><td>Top-1</td><td>Inclusion rates Top-3</td><td>Top-5</td></tr><tr><td colspan=\"2\">Search-result-based Translation Extraction Method</td><td>71%</td><td>89%</td><td>93%</td></tr><tr><td colspan=\"2\">Name Transliteration</td><td>14%</td><td>21%</td><td>25%</td></tr><tr><td colspan=\"2\">Linear Combination</td><td>71%</td><td>82%</td><td>86%</td></tr><tr><td colspan=\"2\">Two-Stage Hybrid Translation Extraction Method</td><td>64%</td><td>71%</td><td>75%</td></tr><tr><td colspan=\"5\">Table 13. Inclusion rates on translation of infrequent unknown Chinese proper</td></tr><tr><td/><td>names from Dreamer query log</td><td/><td/></tr><tr><td/><td>Translation Extraction Method</td><td>Top-1</td><td>Inclusion rates Top-3</td><td>Top-5</td></tr><tr><td colspan=\"2\">Search-result-based Translation Extraction Method</td><td>27%</td><td>44%</td><td>51%</td></tr><tr><td colspan=\"2\">Name Transliteration</td><td>12%</td><td>22%</td><td>22%</td></tr><tr><td colspan=\"2\">Linear Combination</td><td>27%</td><td>47%</td><td>57%</td></tr><tr><td colspan=\"2\">Two-Stage Hybrid Translation Extraction Method</td><td>46%</td><td>51%</td><td>51%</td></tr><tr><td colspan=\"5\">Table 14. Effective results of translation extraction using the two-stage hybrid</td></tr><tr><td/><td colspan=\"4\">translation extraction method (underlined terms indicate correct</td></tr><tr><td/><td>translation)</td><td/><td/></tr><tr><td>Test Query</td><td>Translation Extraction Method</td><td/><td colspan=\"2\">Top 5 Translation Candidates</td></tr><tr><td/><td colspan=\"2\">Search-result-based Translation Extraction Method</td><td colspan=\"2\">Blog, Doll Edward, card, ebay, Eric Benet</td></tr><tr><td/><td>Name Transliteration</td><td/><td colspan=\"2\">Elic, Eddoc, Alic, Addoc, Eric</td></tr><tr><td>\u827e\uf9f7\u514b</td><td>Linear Combination</td><td/><td colspan=\"2\">card, ebay Blog, Doll Edward, Elic,</td></tr><tr><td/><td/><td/><td colspan=\"2\">Eric, Alex, Eric idle,</td></tr><tr><td/><td colspan=\"2\">Two-Stage Hybrid Translation Extraction Method</td><td colspan=\"2\">Clapton Eric, Eric Clapton</td></tr><tr><td/><td/><td/><td colspan=\"2\">Tears, KKBox Eric</td></tr></table>",
                "type_str": "table",
                "text": "",
                "num": null,
                "html": null
            }
        }
    }
}