Base_JSON/prefixO/json/O05/O05-2005.json

{
    "paper_id": "O05-2005",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T07:58:51.159258Z"
    },
    "title": "Aligning Parallel Bilingual Corpora Statistically with Punctuation Criteria",
    "authors": [
        {
            "first": "Thomas",
            "middle": [
                "C"
            ],
            "last": "Chuang",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Vanung University",
                "location": {
                    "addrLine": "No. 1 Van-Nung Road"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "Kevin",
            "middle": [
                "C"
            ],
            "last": "Yeh",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "National Tsing Hua University",
                "location": {
                    "addrLine": "101, Kuangfu Road, Hsinchu, 300",
                    "country": "Taiwan, ROC"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "We present a new approach to aligning sentences in bilingual parallel corpora based on punctuation, especially for English and Chinese. Although the length-based approach produces high accuracy rates of sentence alignment for clean parallel corpora written in two Western languages, such as French-English or German-English, it does not work as well for parallel corpora that are noisy or written in two disparate languages such as Chinese-English. It is possible to use cognates on top of the length-based approach to increase the alignment accuracy. However, cognates do not exist between two disparate languages, which limit the applicability of the cognate-based approach. In this paper, we examine the feasibility of exploiting the statistically ordered matching of punctuation marks in two languages to achieve high accuracy sentence alignment. We have experimented with an implementation of the proposed method on parallel corpora, the Chinese-English Sinorama Magazine Corpus and Scientific American Magazine articles, with satisfactory results. Compared with the length-based method, the proposed method exhibits better precision rates based on our experimental reuslts. Highly promising improvement was observed when both the punctuation-based and length-based methods were adopted within a common statistical framework. We also demonstrate that the method can be applied to other language pairs, such as English-Japanese, with minimal additional effort.",
    "pdf_parse": {
        "paper_id": "O05-2005",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "We present a new approach to aligning sentences in bilingual parallel corpora based on punctuation, especially for English and Chinese. Although the length-based approach produces high accuracy rates of sentence alignment for clean parallel corpora written in two Western languages, such as French-English or German-English, it does not work as well for parallel corpora that are noisy or written in two disparate languages such as Chinese-English. It is possible to use cognates on top of the length-based approach to increase the alignment accuracy. However, cognates do not exist between two disparate languages, which limit the applicability of the cognate-based approach. In this paper, we examine the feasibility of exploiting the statistically ordered matching of punctuation marks in two languages to achieve high accuracy sentence alignment. We have experimented with an implementation of the proposed method on parallel corpora, the Chinese-English Sinorama Magazine Corpus and Scientific American Magazine articles, with satisfactory results. Compared with the length-based method, the proposed method exhibits better precision rates based on our experimental reuslts. Highly promising improvement was observed when both the punctuation-based and length-based methods were adopted within a common statistical framework. We also demonstrate that the method can be applied to other language pairs, such as English-Japanese, with minimal additional effort.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Bilingual corpora are very important for building natural language processing systems [Moore 2002; Gey et al. 2002] , including data-driven machine translation [Dolan et al. 2002] , computer-assisted revision of translations [Jutras 2000] , and cross-language information retrieval [Chen and Gey 2001] . In order to develop NLP systems, it is useful to align bilingual corpora at the sentence level with very high precision [Moore 2002; Chuang et al. 2002, Kueng and Su 2002] . With aligned sentences, further analysis such as phrase and word alignment analysis [Ker and Chang 1997; Melamed 1997] , bilingual terminology [D\u00e9jean et al. 2002] and collocation [Wu 200 ] extraction analysis can be performed. Yang, C, Li, K. [2003] proposed an alignment method for bilingual title pairs on the Web for automatic generation of bilingual parallel corpora. The hybrid dictionary approach [Collier et al. 1998 ], text-based alignment [Kay and R\u00f6scheisen 1993] , part of speech-based alignment [Chen and Chen 1994] , and the lexical method [Chen 1993] are other examples of sentence alignment methods. While these methods presume little or no prior knowledge of source and target languages, they are relatively complex and require significant amounts of parallel text and language resources. sentences are analyzed. The correlations are 0.99 and 0.95 for the German-English and Chinese-English cases, respectively. The expected ratios and the corresponding standard deviations are (0.92, 0.1124) and (4.614, 0.84) for the German-English and Chinese-English cases, respectively. Furthermore, for English-Chinese alignment tasks, no orthographic, phonetic, or semantic cognates, that are readily recognizable by computer exist. Therefore, the inexpensive cognate-based approach is not applicable to Chinese-English tasks. We are thus motivated to find alternative evidence that two blocks of texts are mutual translations. It turns out that punctuation can be telling evidence, if we do more than hard matching of punctuation and take into consideration intrinsic sequencing and the statistical distribution of punctuation in ordered comparison. What is attractive about this approach is that it easily leads to sub-sentential alignment, which has been shown to be useful for statistical machine translation.",
                "cite_spans": [
                    {
                        "start": 86,
                        "end": 98,
                        "text": "[Moore 2002;",
                        "ref_id": "BIBREF17"
                    },
                    {
                        "start": 99,
                        "end": 115,
                        "text": "Gey et al. 2002]",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 160,
                        "end": 179,
                        "text": "[Dolan et al. 2002]",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 225,
                        "end": 238,
                        "text": "[Jutras 2000]",
                        "ref_id": null
                    },
                    {
                        "start": 282,
                        "end": 301,
                        "text": "[Chen and Gey 2001]",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 424,
                        "end": 436,
                        "text": "[Moore 2002;",
                        "ref_id": "BIBREF17"
                    },
                    {
                        "start": 437,
                        "end": 466,
                        "text": "Chuang et al. 2002, Kueng and",
                        "ref_id": null
                    },
                    {
                        "start": 467,
                        "end": 475,
                        "text": "Su 2002]",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 562,
                        "end": 582,
                        "text": "[Ker and Chang 1997;",
                        "ref_id": null
                    },
                    {
                        "start": 583,
                        "end": 596,
                        "text": "Melamed 1997]",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 621,
                        "end": 641,
                        "text": "[D\u00e9jean et al. 2002]",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 658,
                        "end": 665,
                        "text": "[Wu 200",
                        "ref_id": null
                    },
                    {
                        "start": 706,
                        "end": 728,
                        "text": "Yang, C, Li, K. [2003]",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 882,
                        "end": 902,
                        "text": "[Collier et al. 1998",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 927,
                        "end": 952,
                        "text": "[Kay and R\u00f6scheisen 1993]",
                        "ref_id": null
                    },
                    {
                        "start": 986,
                        "end": 1006,
                        "text": "[Chen and Chen 1994]",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 1032,
                        "end": 1043,
                        "text": "[Chen 1993]",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "Section 2 of this paper, we provide some information about the similarity in the use of punctuation in Chinese and English literature and also the differences. Our conclusion is that using punctuation as cognates to align disparate parallel texts will fail to provide adequate alignment results. Section 3, we define a punctuation compatibility factor as an indicator of mutual translation. A translation model that employs a punctuation probability function is proposed. In Section 4, we present experiments based on our novel approach of using the statistical properties of punctuation in parallel texts being analyzed to perform bilingual sentence alignment. We demonstrate that one can use punctuation alone to develop a high-precision sentence alignment program for distant parallel texts like those in Chinese-English corpora. Additionally, we examine the performance of sentence alignment by using punctuation in combination with length. In Section 5, we demonstrate that the proposed method is a very cost effective approach that can be effectively applied to other disparate bilingual languages like English-Japanese without a priori language knowledge of them. A brief conclusion is provided in Section 6.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "According to the Longman Dictionary of Applied-Linguistics [Richards et al. 1985] , a cognate is \"a word in one language which is similar in form and meaning to a word in another language because both languages are related.\" Although the ways in which different languages around the world use punctuation vary, symbols such as commas and full stops are used in most languages to demarcate writing, while question and exclamation marks are used to show interrogation and emphasis. However, these forms of punctuation can often look different or be used in different ways.",
                "cite_spans": [
                    {
                        "start": 59,
                        "end": 81,
                        "text": "[Richards et al. 1985]",
                        "ref_id": "BIBREF19"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation across languages",
                "sec_num": "2."
            },
            {
                "text": "The traditional Chinese writing system does not have punctuation, and it is up to the reader to demarcate the text while reading. With the influx of Western culture in the eighteenth century, punctuation systems similar to the one used with Roman script was adopted in China and Japan. The punctuation includes the period, comma, colon, dash, etc. Although most of those forms of punctuation look similar to Roman ones, they are usually coded as double-bytes and tend to be used differently. The full stop in Chinese and Japanese is a small empty circle, quite different in appearance from the Roman period. Quotes are also very different, shaped like a Greek letter \u0393, upright or upside down. There are forms of punctuation that have no counterparts in Roman text. For instance, \"\u3001\" is the pause symbol, which is used somewhat like the comma but only when separating items in a list. On the other hand, there are several uses of the Roman comma which do not occur in Chinese texts. A few examples are given below:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation across languages",
                "sec_num": "2."
            },
            {
                "text": "(Parenthetical expressions) (1e) Evolution, as far as we know, doesn't work this way. Yang [1981] described more punctuation marks in Chinese used in various ways that are similar or dissimilar to English punctuation. In summary, although Simard et al. [1992] considered the various forms of punctuation in English and French to be cognates, in general, punctuation forms are not cognates for many other language pairs.",
                "cite_spans": [
                    {
                        "start": 86,
                        "end": 97,
                        "text": "Yang [1981]",
                        "ref_id": null
                    },
                    {
                        "start": 239,
                        "end": 259,
                        "text": "Simard et al. [1992]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation across languages",
                "sec_num": "2."
            },
            {
                "text": "In both Chinese and English texts, the average ratio of the punctuation count to the total number of tokens available is low (less than 15%). But punctuation provides valid additional evidence, which can help one achieve a high degree of alignment precision. Our method can easily be generalized to other language pairs since minimal a priori linguistic knowledge is required.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation across languages",
                "sec_num": "2."
            },
            {
                "text": "In this section, we will describe how punctuation in two languages can be used to measure the likelihood of mutual translation in sentence alignment. We will use an example in the following to illustrate the method. A formal description also follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "Example 3 shows a Chinese sentence and its translation counterpart of two English sentences in a parallel corpus.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "(3c) \u9010\u6f38\u7684\uff0c\u6253\u9f13\uf967\u518d\u80fd\u6eff\u8db3\u4ed6\uff0c \u300c\u6253\u9f13\u539f\u662f\u6211\u6700\u559c\u6b61\u7684\uff0c\u5f8c\uf92d\u537b\u8b8a\u6210\u908a\u6253\u908a\u7761\uff0c\u4e00\u500b \u6708\uf9d1\u842c\u5143\u7684\u6b7b\u5de5\u4f5c\u300d \uff0c\u859b\u5cb3\u8868\u793a\u3002 (3e) Over time, drums could no longer satisfy him. \"Drumming was at first the thing I loved most, but later it became half drumming, half sleeping, just a job for NT$60,000 a month,\" says Simon.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "If we keep punctuations in the above examples in the original order and strip everything else out, we have ten pieces of punctuation from the English part (3e) and eight from the Mandarin part (3c) as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "(4c) \uff0c \uff0c \u300c \uff0c \uff0c \u300d \uff0c \u3002 (4e) , . \" , ,",
                        "eq_num": ", , \" ."
                    }
                ],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "They can be arranged into different match types as shown below.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "Match type (4c) (4e)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "1-1 \uff0c , 1-1 \uff0c . 1-1 \u300c \" 1-1 \uff0c , 0-1 , 1-1 \uff0c , 2-2 \u300d\uff0c , \"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "1-1 \u3002 .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Marks in English and Chinese",
                "sec_num": "3.1"
            },
            {
                "text": "There are several frequently used punctuation forms in Chinese text that are not available in English text, for example, the punctuation forms \"\u3001\" and \"\u3002\". These punctuation forms often correspond to the English punctuation forms \",\" and \".\", respectively. It is not difficult to see that the two punctuation strings above match up quite nicely, indicating that the corresponding texts are mutual translations. Roughly, the first two commas in Chinese correspond to the first two English punctuation marks (comma and period), while the Chinese open quote in the third position corresponds to the English open quote also in the third position. The two Chinese commas inside the quotes correspond to two of the four commas within the quotes in English. The two consecutive marks (\u300d \uff0c) correspond to (,\"), forming a 2-2 match. These correspondences can be unraveled via a dynamic programming procedure, much like sentence alignment. See Figure 2 for more details.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 934,
                        "end": 942,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Figure 2. The correspondence between two punctuation strings",
                "sec_num": null
            },
            {
                "text": "It is apparent that the punctuation in the two strings match up very consistently, and that the matching is somewhat continuous with respect to the alignment of regular words surrounding the punctuation (see the double-lined links in Figure 3 for details). Therefore, the example gives a convincing indication that the correspondence between punctuation across two languages can provide telling evidence that two texts are mutual translations.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 234,
                        "end": 242,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Figure 2. The correspondence between two punctuation strings",
                "sec_num": null
            },
            {
                "text": "Based on our initial observation, the portion of the identifiable punctuation matches between two parallel texts in Chinese and English is over 50%. Examining Figure 2 , we can identify institutively the matches between the Chinese punctuation and the equivalent English punctuation marks: (\u300c) corresponds to (\"), etc. This implies that although direct match information is useful, there is still a large discrepancy in the punctuation mappings between Chinese and English. We, therefore, define here a punctuation compatibility factor that can be used to further analyze the relationship between the punctuation found in parallel texts. The punctuation compatibility factor as an indicator of mutual translation is defined as",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 159,
                        "end": 167,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Punctuation marks as Good Indicators of Mutual Translation",
                "sec_num": "3.2"
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "max( , ) c n m \u03b3 = ,",
                        "eq_num": "(1)"
                    }
                ],
                "section": "Punctuation marks as Good Indicators of Mutual Translation",
                "sec_num": "3.2"
            },
            {
                "text": "where \u03b3 = the punctuation compatibility factor, c = the number of direct punctuation matches, n = the number of Chinese punctuation marks, m = the number of English punctuation marks.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation marks as Good Indicators of Mutual Translation",
                "sec_num": "3.2"
            },
            {
                "text": "We took aligned English-Chinese sentences that had the same punctuation count (which is the denominator of Equation 1), take ten for example, in order to determine how well punctuation works as an indicator of mutual translation of English and Chinese sentences. We also took the same English sentences and matched them up with randomly selected Chinese sentences to calculate the compatibility of punctuation marks in unrelated texts.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation marks as Good Indicators of Mutual Translation",
                "sec_num": "3.2"
            },
            {
                "text": "The results obtained indicated that the average compatibility of pairs of sentences, which were mutual translations, was about 0.67 (with a standard deviation of 0.170), while the average compatibility of random pairs of bilingual sentences was 0.34 (with a standard deviation of 0.167).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation marks as Good Indicators of Mutual Translation",
                "sec_num": "3.2"
            },
            {
                "text": "Figures 4 through 6 show the compatibility results based on punctuation counts of eight, ten and twelve respectively. These graphs were constructed by analyzing around 50,000 aligned sentences found in the Sinorama Magazine (1990 Magazine ( -2000 . 521, 259, and 143 sentences were selected to obtain values of n and m equal to 8, 10, and 12, respectively. The solid lines simply connect data points for easier observation. Intuitively, as the number of punctuation marks increases, the reliability of the compatibility function does also. Overall, if the punctuation marks are softly matched in ordered comparison across the two languages, they indeed provide useful information for effective sentence alignment. Analyzing the Sinorama corpus, we found that the percentage of matched sentences having the same number of punctuation marks was 21.42%. We selected and analyzed aligned sentences having different numbers of punctuation marks to get more insight into the distinction between matched and random sentences. The analysis also helped us to determine the proper use of the binomial distribution function for sentence alignment. Sentences with eight, ten, and twelve punctuation marks were arbitrarily chosen for analysis. It appears that the distinction between mutual translations and unrelated texts indeed becomes more prominent for sentences that have larger numbers of punctuation marks.",
                "cite_spans": [
                    {
                        "start": 215,
                        "end": 229,
                        "text": "Magazine (1990",
                        "ref_id": null
                    },
                    {
                        "start": 230,
                        "end": 246,
                        "text": "Magazine ( -2000",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 3. English punctuation across aligned sentences",
                "sec_num": null
            },
            {
                "text": "Instead of one-to-one hard matching of punctuation marks in parallel texts as used in the cognate approach of Simard et al. (1992) , we allow no match and one-to-several matching of punctuation matches. Our model of the probability of punctuation alignment is very similar to the word alignment model proposed by Brown et al. (1991) . In order to perform soft matching of punctuation, we define the probability that a sequence of punctuation marks i j P EP CP . We choose the punctuation alignment that maximizes the probability overall possible alignments, given a pair of punctuation sequences corresponding to a pair of parallel sentences, i.e., arg max ( , )",
                "cite_spans": [
                    {
                        "start": 110,
                        "end": 130,
                        "text": "Simard et al. (1992)",
                        "ref_id": null
                    },
                    {
                        "start": 313,
                        "end": 332,
                        "text": "Brown et al. (1991)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "i j P CP EP \u0391 \u0391 ,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "where A is a punctuation alignment. Assuming that the probabilities of the individually aligned punctuation pairs are independent and applying the Bayes' rule, we can make the following approximation: We observe that in most cases, the links of punctuation do not cross each other, much like the situation with sentence alignment. Therefore, it is possible to use the dynamic programming procedure to softly match punctuation across languages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( , )",
                        "eq_num": "( , ) ( ,"
                    }
                ],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "In order to explore the relationship between punctuation in pairs of Chinese and English sentences that are mutual translations, we selected a small set of manually aligned texts and investigated the characteristics and the statistics associated with the punctuation. Information from around 500 manually analyzed sentences was then used as the initial parameters to bootstrap a larger corpus. An unsupervised EM algorithm and dynamic programming were used to optimize the punctuation correspondence between a text and its translation counterpart. The steps in the standard EM algorithm which we used included initializing model parameters with manually analyzed punctuation matching probabilities, assigning probabilities to missing punctuation, estimating model parameters from completed data, and iterating the process until convergence was reached. The EM algorithm converged quickly after the second iteration of training.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "We observed that, in most cases, the links of punctuation did not cross each other, much like the situation with sentence alignment. Therefore, we were motivated to use the dynamic programming procedure to softly match punctuation across the languages by finding the Tables 1 to 4 , respectively. It should be noted that the calculated probability was the conditional probability of each punctuation mark, therefore, the sum of the probability in each table does not equal to one. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 267,
                        "end": 280,
                        "text": "Tables 1 to 4",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "\uff0c 1-1 12 0.666302 ; \u3001 1-1 11 0.422925 , \u300c 1-1 10 0.01497 ? \uff1f 1-1 9 0.236794 . \u3001 1-1 7 0.013698 \" \uff0c 1-1 7 0.018276 ; \uff0c 1-1 7 0.269134 . \uff1b 1-1 6 0.011742 \" \uff1a 1-1 6 0.015666 , \uff1a 1-1 5 0.007485 ? \uff0c 1-1 5 0.131552 , \uff1b 1-1 4 0.005988 . - 1-1 4 0.007828 : \uff1a 1-1 4 0.222101 ; \u3002 1-1 4 0.153791 ) ) 1-1 4 0.997159 , \uff0e 1-1 3 0.004491 . \uff0e 1-1 3 0.005871 . \ufe41 1-1 3 0.005871 ! \u3002 1-1 3 0.069751 ? - 1-1 3 0.078931",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "\uff0c 0-1 229 0.389455 - 0-1 58 0.098639 \u3002 0-1 52 0.088435 \u300d 0-1 50 0.085034 \u3001 0-1 45 0.076531 \u300c 0-1 41 0.069728 \u2026 0-1 39 0.066326 \uff1f 0-1 14 0.02381 \ufe42 0-1 14 0.02381 \uff1a 0-1 9 0.015306 \ufe41 0-1 7 0.011905",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "The punctuation match types (also known as the fertility functions) obtained through training are summarized in Table 5 . Notice that the counts shown in the table are not integers because the results of EM training were adjusted using the Good Turing Smoothing Method to improve them. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 112,
                        "end": 119,
                        "text": "Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Punctuation Alignment Model",
                "sec_num": "3.3"
            },
            {
                "text": "Unlike the method Simard et al. [1992] used to handle cognates, we model the compatibility of punctuation across two languages using the Binomial distribution. Each punctuation mark appearing in one language either has one to three punctuation counterparts across translation or does not. For each punctuation mark, the probability of it having a translation counterpart is independent with a fixed value of p. Our approach differs from Simard's in the following interesting ways:",
                "cite_spans": [
                    {
                        "start": 18,
                        "end": 38,
                        "text": "Simard et al. [1992]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "1. We use the Binomial distribution, while Simard et al. used a likelihood ratio.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "2. We go beyond hard matching of punctuation marks between parallel texts. We allow a punctuation mark in one language to match up with a number of compatible punctuation marks in another. The compatibility model is similar in structure to the lexical translation probability proposed by Brown et al. [1991] .",
                "cite_spans": [
                    {
                        "start": 288,
                        "end": 307,
                        "text": "Brown et al. [1991]",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "3. We take into consideration the intrinsic sequencing of punctuation marks in an ordered comparison. A flexible and ordered comparison of punctuation is carried out via dynamic programming.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "Following Gale and Church [1993] , we employ the Bayes Theorem to estimate the likelihood of aligning two text blocks E and C by calculating P(E, C|match) P(match).We adopt the same dynamic programming method, but use punctuation marks to measure the likelihood of mutual translation instead of lengths. The proposed sentence alignment method is based on a model in which each punctuation mark in L1 is responsible for generating a number of punctuation marks with a given matching probability in L2.",
                "cite_spans": [
                    {
                        "start": 10,
                        "end": 32,
                        "text": "Gale and Church [1993]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "We define the probability of mutual translation for a given alignment pattern P(A| C,E) as follows: Given two blocks of text E and C, we first strip off non-punctuation therein and determine the maximum number of punctuation marks n in either E or C.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "We employ punctuation-based sentence alignment, which maximizes the probability of overall possible alignment, given a pair of parallel texts, i.e., arg max ( , )",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "P C E \u0391 \u0391 ,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "where A is an alignment and C and E are the source and target texts, respectively.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "A further approximation encapsulates the dependence of a single parameter b, which is a function of CP and EP: P(A|Ci,Ej) = P(A|b(CP,EP)) .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "Since it is easier to estimate the distribution for the inverted form, we apply Bayes' Rule to further simplify the calculation: P(A|b) = P(b|A)P(A)/P(b) , where P(b) is a normalizing constant that can be ignored during minimization. P (A) is the match type, and its values are shown in Table 6 . We use a binominal distribution to estimate P(b): t = the total number of sentences that are aligned. From the data, we have found that about 66% of the time, a sentence in one language matches exactly one sentence in the other language (1-1). Three additional possibilities should be also considered: 1-0 (including 0-1), and many-1 (including 1-many). Chinese-English parallel corpora are quite noisy, reflecting from wider possibilities of the match types. Here, we used the same probabilistic figures as proposed by Chuang and Chang [2002] . Table 6 shows all eight possibilities used in our implementation.",
                "cite_spans": [
                    {
                        "start": 817,
                        "end": 840,
                        "text": "Chuang and Chang [2002]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 287,
                        "end": 294,
                        "text": "Table 6",
                        "ref_id": null
                    },
                    {
                        "start": 843,
                        "end": 850,
                        "text": "Table 6",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "( | , ) ( | , ) i j P A C E P A C E \u0391 \u2248 \u220f 1 ( ) ( , ) (1 ( , )) k k k t k r n r k k k k k k k n P A P Cp Ep P Cp Ep r \u2212 = \u239b \u239e \u2248 \u22c5 \u2212 \u239c \u239f \u239d \u23a0 \u220f ,",
                        "eq_num": "(3)"
                    }
                ],
                "section": "Punctuation-based Sentence Alignment Model",
                "sec_num": "3.4"
            },
            {
                "text": "1-1 1-0, 0-1 1-2 2-1 2-2 1-3 3-1 Chinese-English 0.64 0.0056 0.017 0.25 -0.056 -",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Table 6. The sentence alignment match type probability P(A) P(A)",
                "sec_num": null
            },
            {
                "text": "The Since the sentence lengths of the bilingual parallel texts of interest are highly correlated, ( | ) P match \u03b4 can be estimated using the Gaussian assumption following Gale and Church [1993] . Incorporating both the length-based and punctuation-based criteria, we can modify equation 3as follows:",
                "cite_spans": [
                    {
                        "start": 171,
                        "end": 193,
                        "text": "Gale and Church [1993]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Hybrid Punctuation-based and Length-Based Sentence Alignment Model",
                "sec_num": "3.5"
            },
            {
                "text": "1 ( | , ) ( ) ( | ) ( , )(1 ( , )) k k t k n r k k k k k k n P A C E P A P match P CP EP P CP EP r \u03b4 \u2212 = \u239b \u239e \u2248 \u22c5 \u22c5 \u2212 \u239c \u239f \u239d \u23a0 \u220f . (4)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Hybrid Punctuation-based and Length-Based Sentence Alignment Model",
                "sec_num": "3.5"
            },
            {
                "text": "The same dynamic programming optimization can then be used. Again, the computation and memory costs are very low when both the length-based and punctuation-based criteria are employed. The average slopes of c l and e l , and the associated standard deviations are estimated in an adaptive manner for each corpus being evaluated [Chuang et al. 2002] .",
                "cite_spans": [
                    {
                        "start": 328,
                        "end": 348,
                        "text": "[Chuang et al. 2002]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Hybrid Punctuation-based and Length-Based Sentence Alignment Model",
                "sec_num": "3.5"
            },
            {
                "text": "To explore the relationship between the punctuation marks in pairs of Chinese and English sentences that are mutual translations, we prepared a small set of 200 pairs of sentences aligned at the sentence and punctuation levels. We then investigated the characteristics of and the statistics associated with the punctuation marks. We derived estimates of the punctuation translation probabilities and fertility probabilities from the small set of hand-tagged data. This seed information was then used to train the punctuation translation model on a larger corpus via the EM algorithm. The probability of a punctuation mark having a translation counterpart was estimated as p = 0.670 with a standard deviation 0.170. For random pairs of bilingual sentences, p = 0.340, with a standard deviation 0.167. There appears to be marked differences between the two distributions, indicating that, indeed, soft and ordered comparison of punctuation marks across languages provide useful information for effective sentence alignment.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments and evaluation",
                "sec_num": "4."
            },
            {
                "text": "In order to assess the performance of punctuation-based sentence alignment, we randomly selected five bilingual articles from the Sinorama Magazine Corpus and Scientific American (US and Taiwan editions), and several chapters from the novel Harry Potter. These were subjected to an implementation of the proposed method. Some experimental results are shown in appendices A and B.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments and evaluation",
                "sec_num": "4."
            },
            {
                "text": "It should be noted that in Appendix A, the first English sentence and the first Chinese sentence are both title sentences, and that they are aligned based on the carriage return deliminater, even though no punctuation marks are found in the English sentence. We found that, in general, there were more periods in the English text than that in the Chinese text for a given bilingual corpus, especially in the case of a text translated from Chinese into English. As an example, 112 periods were found in a Chinese article [Sinorama 1988 ], whereas only 180 periods were found in the corresponding English translation. This phenomenon can be further seen by examining the punctuation distribution. The relationship between the number of sentences and the number of punctuation marks per sentence for the English-Chinese corpus was determined by analyzing 6,103 articles, including around 130,000 sentences, from Sinorama Magazine between 1976 and 2000. 1,138,447 punctuation marks were found in the English corpus and 2,056,675 punctuation marks in the Chinese corpus. Apparently, punctuation marks are used more sparingly in Chinese sentences. As shown in Figure 7 , there were two punctuation marks in most of the English sentences and three in most of the Chinese sentences. The figure also shows that a few long sentences had close to one hundred punctuation marks, but these were unuseful.",
                "cite_spans": [
                    {
                        "start": 520,
                        "end": 534,
                        "text": "[Sinorama 1988",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1154,
                        "end": 1162,
                        "text": "Figure 7",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Experiments and evaluation",
                "sec_num": "4."
            },
            {
                "text": "This observation prompted us to establish a special rule that the combination of a comma and an open quote in a Chinese sentence should be considered as being equivalent to a full stop. Applying this rule, we found that the sentence count increased from 112 to 126 for the Chinese text mentioned in the above example. This empirical rule helped to improve the precision of sentence alignment. These special cases can be found in both Appendixes A and B.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 7. The punctuation distribution for a bilingual corpus",
                "sec_num": null
            },
            {
                "text": "The precision rate of the length-based approach [Gale and Church 1993] is shown in Table 7 as a baseline for comparison. The precision rate is defined as the ratio between the number of correctly matched sentences in the system output and the number of matched sentences generated from the system output. The large variation observed in the alignment precision is primary due to the disparity in the lengths and match types. The experimental results obtained with the punctuation-based approach and the combination approaches are shown in Tables 8 and 9 . Overall, the punctuation-based approach outperformed the length-based approach, reducing the error rates consistently, and the improvement could exceed 50% at times. Additionally, we evaluated our method on a larger corpus the Scientific American Corpus. We used all of the English and Chinese articles from January 2003 to December 2003. There were 67 articles, 1523 English sentences, and 1599 Chinese sentences. Every article included both an English text and its corresponding Chinese text. The punctuation-based sentence alignment method achieved alignment precision rates of over 93%. Inferior performance was achieved when the hybrid punctuation and length-based method was used as compared with the punctuation-based method alone, as shown by the results listed in Tables 8 and 9 . This phenomenon may be attributed to the strong dependence of the length-based method on the average length of the sentences being analyzed. Apparently, length-based methods do not perform well in the case of a corpus that is composed of shorter sentences. Therefore, a length-based method may achieve poorer performance when it is combined with a punctuation-based method. Consequently, caution should be exercised in interpreting these precision rates.",
                "cite_spans": [
                    {
                        "start": 48,
                        "end": 70,
                        "text": "[Gale and Church 1993]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 83,
                        "end": 90,
                        "text": "Table 7",
                        "ref_id": null
                    },
                    {
                        "start": 539,
                        "end": 553,
                        "text": "Tables 8 and 9",
                        "ref_id": null
                    },
                    {
                        "start": 1329,
                        "end": 1343,
                        "text": "Tables 8 and 9",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Figure 7. The punctuation distribution for a bilingual corpus",
                "sec_num": null
            },
            {
                "text": "Our approach has been proven to be effective, and it has been used to construct a concordancer system called TotallRecall [Wu et al. 2003 ] in a Computer Assisted Language Learning (CALL) project. Based on the results of our experiments, it was also possible to speed up the corpus annotation and distribution efforts made by the Association for Computational Linguistics and Chinese Language Processing.",
                "cite_spans": [
                    {
                        "start": 122,
                        "end": 137,
                        "text": "[Wu et al. 2003",
                        "ref_id": "BIBREF22"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Table7. Baseline sentence alignment performance achieved using the length-based approach",
                "sec_num": null
            },
            {
                "text": "We achieved a striking improvement over the length-based baseline for bilingual text alignment when punctuation was used alone or in combination with lexical information. Combining punctuation and length information, we could get slightly better overall performance. However, the improvement was not entirely consistent. Thus we need to experiment on a longer parallel text in order to be more certain about it.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion",
                "sec_num": "5."
            },
            {
                "text": "Although word alignment links cross each other quite often, punctuation links do not. It appears that we can obtain sub-sentential alignment at the clause and phrase levels from the alignment of punctuation. For instance, after we align the punctuation in examples (3c) and (3e), we can extract the following finer-grained bilingual analyses: We have hand-coded a small English-Japanese punctuation mapping table and converted our alignment program to handle alignment of Japanese and English texts. It appears that the adapted program works with performance comparable to that of the original one. An example of aligning English-Japanese parallel texts based on punctuation is shown in Appendix C. Our Japanese-English program is a very preliminary one. Further and more rigorous investigation is needed.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion",
                "sec_num": "5."
            },
            {
                "text": "We have developed a very effective sentence alignment method based on punctuation. The probability of the finding matches between different punctuation marks in source and target texts is calculated based on a large bilingual corpus. The punctuation-based measure of mutual translation can be modeled by the binomial distribution. We have implemented the proposed method on the parallel Chinese-English Sinorama Magazine Corpus. The experimental results show that the punctuation-based approach outperforms the length-based approach with precision rates exceeding 93%.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "6."
            },
            {
                "text": "We have also demonstrated that the alignment method can be applied to other bilingual texts, without the need for a priori linguistic knowledge of the languages, like Japanese and English. This general approach has been found to be fast, easy to set up, and universal. We believe that this method can be easily applied to many different languages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "6."
            },
            {
                "text": "A number of interesting future directions for researches present themselves. First, punctuation alignment can be exploited to constrain word alignment and reduce error rates. Second, punctuation alignment makes possible a finer-grained level of bilingual analysis of sub-sentential alignment and can provide a strikingly more effective translation memory and bilingual concordance for more effective example-based machine translation (EBMT), computer assisted translation and language learning (CAT and CALL).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "6."
            },
            {
                "text": "Some experimental results of sentence alignment based on length and punctuation are presented here. Shaded parts indicate imprecision in alignment results. We calculated the precision rates by dividing the number of un-shaded sentences (counting both English and Chinese sentences) by the total number of sentences proposed. Since we did not exclude aligned pairs using a threshold, the recall rate should be the same as the precision rate. The experimental results indicate that when non 1-1 matches next to each other tend to fail the length-based aligner. However, the punctuation-based aligner appears to handle such cases more successfully.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "6."
            },
            {
                "text": "Type English text Chinese Text 11 Take note \u300c\u5171\u7b46\u300d\u602a\u73fe\u8c61 12 Allowing education to be led by the market may also lead to deficiencies in teaching practices.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Sentence alignment based on length",
                "sec_num": null
            },
            {
                "text": "\u5e02\u5834\uf9b4\u5c0e\u6559\u80b2\u9084\u53ef\u80fd\u5f15\u767c\u6559\u5b78\u4e0a\u7684\u5f0a \u75c5\u3002\u53f0\u5927\u6cd5\uf9d8\u7cfb\u6559\u6388\u8cc0\u5fb7\u82ac\uf96f\uff0c\u5c0d\u6cd5 \uf9d8\u7cfb\u5b78\u751f\uf92d\uf96f\uff0c\u8003\u4e0a\u53f8\u6cd5\u5b98\u3001\u9ad8\u8003\u662f \u6700\u597d\u7684\u51fa\uf937\uff0c 11 Professor He Te-fen of NTU's Department of Law say that for law students, the best opportunity for advancement is to pass the recruitment examinations for public prosecutors and judges, or the senior civil service exams.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Sentence alignment based on length",
                "sec_num": null
            },
            {
                "text": "31 \"In class, some students only want to learn specifically how to answer exam questions, and their choice of courses depends on whether the instructor's teaching method is helpful for passing the exams.\" Some instructors, seeing that some students do not take good notes, even designate one who does to give them to the others for reference. But this results in most of the students taking no notes at all, because after all they will get photocopies, paid for out of the class expenses fund.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u6709\u4e9b\u5b78\u751f\u4e0a\u8ab2\u53ea\u60f3\u5177\u9ad4\u77e5\u9053\u5982\u4f55\u7b54 \u8003\u984c\uff0c\u9078\u8ab2\u6a19\u6e96\u5c31\u662f\uf934\u5e2b\u7684\u6559\u66f8\u65b9\u5f0f \u662f\uf967\u662f\u5c0d\u8003\u8a66\u6709\u7528\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "\u751a\u81f3\u6709\uf934\u5e2b\u56e0\u70ba\u770b\u5b78\u751f\u7684\u7b46\u8a18\u8a18\uf967 \u597d\uff0c\u6307\u5b9a\u505a\u5f97\u597d\u7684\u540c\u5b78\u7d66\u5176\u4ed6\u4eba\uf96b \u8003\uff0c\u4ee5\u63d0\u9ad8\u7cfb\u4e0a\u7684\uf93f\u53d6\uf961\uff0c\u7d50\u679c\u8b8a\u6210 \u5b78\u751f\u4e5f\uf967\u505a\u7b46\u8a18\uf9ba\uff0c\u53cd\u6b63\u6709\u73ed\u8cbb\u53ef\u4ee5 \u5f71\u5370\u7d66\u5927\u5bb6\uff0c\u9019\u500b\u73fe\u8c61\u9084\u6709\u500b\u540d\u8a5e\u53eb \u300c\u5171\u7b46\u300d \u3002",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u6709\u4e9b\u5b78\u751f\u4e0a\u8ab2\u53ea\u60f3\u5177\u9ad4\u77e5\u9053\u5982\u4f55\u7b54 \u8003\u984c\uff0c\u9078\u8ab2\u6a19\u6e96\u5c31\u662f\uf934\u5e2b\u7684\u6559\u66f8\u65b9\u5f0f \u662f\uf967\u662f\u5c0d\u8003\u8a66\u6709\u7528\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "21 Two years ago, the CER completed a \"General Consultation Report on Educational Reform.\" One of its main proposals was that the past system of controlling the establishment, expansion and contraction of departments in higher education on the basis of estimates of personnel demand should be \"relaxed.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u6709\u4e9b\u5b78\u751f\u4e0a\u8ab2\u53ea\u60f3\u5177\u9ad4\u77e5\u9053\u5982\u4f55\u7b54 \u8003\u984c\uff0c\u9078\u8ab2\u6a19\u6e96\u5c31\u662f\uf934\u5e2b\u7684\u6559\u66f8\u65b9\u5f0f \u662f\uf967\u662f\u5c0d\u8003\u8a66\u6709\u7528\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "11 Education cannot be made merely to narrowly serve the economy.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\uf978\uf98e\u524d\ufa08\u653f\u9662\u6559\u6539\u6703\u5b8c\u6210\u300c\u6559\u80b2\u6539\u9769 \u7e3d\u8aee\u8b70\u5831\u544a\u66f8\u300d \uff0c\u5efa\u8b70\u7684\u91cd\u9ede\u4e4b\u4e00\u662f\uff0c \u904e\u53bb\u4ee5\u4eba\uf98a\u9700\u6c42\u7684\u63a8\u4f30\uff0c\u7ba1\u5236\u9ad8\u7b49\u6559 \u80b2\u79d1\u7cfb\u7684\u8a2d\uf9f7\u589e\u6e1b\uff0c\u61c9\u8a72\u300c\u9b06\u7d81\u300d \u3002",
                "sec_num": null
            },
            {
                "text": "\u4f46\u73fe\u5be6\u7684\uf9fa\u6cc1\u662f\uff0c 11 Yet in reality, \"the reason most parents are willing \u300c\u5927\u90e8\u5206\u5bb6\u9577\u4e4b\u6240\u4ee5\u80af\u82b1\u9322\u8b93\u5b69\u5b50\uf92d to pay to put their children through university is certainly not that they hope they will become passionate seekers after truth, but to enable them to find good careers,\" says Providence University president Li Chia-tung bluntly.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u6559\u80b2\uf967\u80fd\u300c\u7a84\u5316\u300d\u6210\u53ea\u70ba\u7d93\u6fdf\u670d\u52d9\uff0c",
                "sec_num": null
            },
            {
                "text": "Sentence alignment based on punctuation 11 Take note \u300c\u5171\u7b46\u300d\u602a\u73fe\u8c61 11 Allowing education to be led by the market may also lead to deficiencies in teaching practices.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\uf9a3\u5927\u5b78\uff0c\u7d55\uf967\u662f\u5e0c\u671b\u5b69\u5b50\u4ee5\u5f8c\u71b1\u8877\u65bc \u771f\uf9e4\u7684\u8ffd\u6c42\uff0c\u800c\u662f\u70ba\uf9ba\u4f7f\u5b69\u5b50\u5c07\uf92d\u80fd \u627e\u5230\u597d\u8077\u696d\uff0c\u300d\u975c\u5b9c\u5927\u5b78\u6821\u9577\uf9e1\u5bb6\u540c \u660e\u767d\u5730\uf96f\u3002",
                "sec_num": null
            },
            {
                "text": "\u5e02\u5834\uf9b4\u5c0e\u6559\u80b2\u9084\u53ef\u80fd\u5f15\u767c\u6559\u5b78\u4e0a\u7684\u5f0a \u75c5\u3002 11 Professor He Te-fen of NTU's Department of Law say that for law students, the best opportunity for advancement is to pass the recruitment examinations for public prosecutors and judges, or the senior civil service exams.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\uf9a3\u5927\u5b78\uff0c\u7d55\uf967\u662f\u5e0c\u671b\u5b69\u5b50\u4ee5\u5f8c\u71b1\u8877\u65bc \u771f\uf9e4\u7684\u8ffd\u6c42\uff0c\u800c\u662f\u70ba\uf9ba\u4f7f\u5b69\u5b50\u5c07\uf92d\u80fd \u627e\u5230\u597d\u8077\u696d\uff0c\u300d\u975c\u5b9c\u5927\u5b78\u6821\u9577\uf9e1\u5bb6\u540c \u660e\u767d\u5730\uf96f\u3002",
                "sec_num": null
            },
            {
                "text": "11 \"In class, some students only want to learn specifically how to answer exam questions, and their choice of courses depends on whether the instructor's teaching method is helpful for passing the exams.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u53f0\u5927\u6cd5\uf9d8\u7cfb\u6559\u6388\u8cc0\u5fb7\u82ac\uf96f\uff0c\u5c0d\u6cd5\uf9d8\u7cfb \u5b78\u751f\uf92d\uf96f\uff0c\u8003\u4e0a\u53f8\u6cd5\u5b98\u3001\u9ad8\u8003\u662f\u6700\u597d \u7684\u51fa\uf937\uff0c",
                "sec_num": null
            },
            {
                "text": "21 Some instructors, seeing that some students do not take good notes, even designate one who does to give them to the others for reference. But this results in most of the students taking no notes at all, because after all they will get photocopies, paid for out of the class expenses fund.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u6709\u4e9b\u5b78\u751f\u4e0a\u8ab2\u53ea\u60f3\u5177\u9ad4\u77e5\u9053\u5982\u4f55\u7b54 \u8003\u984c\uff0c\u9078\u8ab2\u6a19\u6e96\u5c31\u662f\uf934\u5e2b\u7684\u6559\u66f8\u65b9\u5f0f \u662f\uf967\u662f\u5c0d\u8003\u8a66\u6709\u7528\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "\u751a\u81f3\u6709\uf934\u5e2b\u56e0\u70ba\u770b\u5b78\u751f\u7684\u7b46\u8a18\u8a18\uf967 \u597d\uff0c\u6307\u5b9a\u505a\u5f97\u597d\u7684\u540c\u5b78\u7d66\u5176\u4ed6\u4eba\uf96b \u8003\uff0c\u4ee5\u63d0\u9ad8\u7cfb\u4e0a\u7684\uf93f\u53d6\uf961\uff0c\u7d50\u679c\u8b8a\u6210 \u5b78\u751f\u4e5f\uf967\u505a\u7b46\u8a18\uf9ba\uff0c\u53cd\u6b63\u6709\u73ed\u8cbb\u53ef\u4ee5 \u5f71\u5370\u7d66\u5927\u5bb6\uff0c\u9019\u500b\u73fe\u8c61\u9084\u6709\u500b\u540d\u8a5e\u53eb \u300c\u5171\u7b46\u300d \u3002 21 Two years ago, the CER completed a \"General Consultation Report on Educational Reform.\" One of its main proposals was that the past system of controlling the establishment, expansion and contraction of departments in higher education on the basis of estimates of personnel demand should be \"relaxed.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u6709\u4e9b\u5b78\u751f\u4e0a\u8ab2\u53ea\u60f3\u5177\u9ad4\u77e5\u9053\u5982\u4f55\u7b54 \u8003\u984c\uff0c\u9078\u8ab2\u6a19\u6e96\u5c31\u662f\uf934\u5e2b\u7684\u6559\u66f8\u65b9\u5f0f \u662f\uf967\u662f\u5c0d\u8003\u8a66\u6709\u7528\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "11 Education cannot be made merely to narrowly serve the economy.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\uf978\uf98e\u524d\ufa08\u653f\u9662\u6559\u6539\u6703\u5b8c\u6210\u300c\u6559\u80b2\u6539\u9769 \u7e3d\u8aee\u8b70\u5831\u544a\u66f8\u300d \uff0c\u5efa\u8b70\u7684\u91cd\u9ede\u4e4b\u4e00\u662f\uff0c \u904e\u53bb\u4ee5\u4eba\uf98a\u9700\u6c42\u7684\u63a8\u4f30\uff0c\u7ba1\u5236\u9ad8\u7b49\u6559 \u80b2\u79d1\u7cfb\u7684\u8a2d\uf9f7\u589e\u6e1b\uff0c\u61c9\u8a72\u300c\u9b06\u7d81\u300d \u3002",
                "sec_num": null
            },
            {
                "text": "\u4f46\u73fe\u5be6\u7684\uf9fa\u6cc1\u662f\uff0c 11 Yet in reality, \"the reason most parents are willing to pay to put their children through university is certainly not that they hope they will become passionate seekers after truth, but to enable them to find good careers,\" says Providence University president Li Chia-tung bluntly.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u6559\u80b2\uf967\u80fd\u300c\u7a84\u5316\u300d\u6210\u53ea\u70ba\u7d93\u6fdf\u670d\u52d9\uff0c",
                "sec_num": null
            },
            {
                "text": "More English-Chinese alignment results.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u5927\u90e8\u5206\u5bb6\u9577\u4e4b\u6240\u4ee5\u80af\u82b1\u9322\u8b93\u5b69\u5b50\uf92d \uf9a3\u5927\u5b78\uff0c\u7d55\uf967\u662f\u5e0c\u671b\u5b69\u5b50\u4ee5\u5f8c\u71b1\u8877\u65bc \u771f\uf9e4\u7684\u8ffd\u6c42\uff0c\u800c\u662f\u70ba\uf9ba\u4f7f\u5b69\u5b50\u5c07\uf92d\u80fd \u627e\u5230\u597d\u8077\u696d\uff0c\u300d\u975c\u5b9c\u5927\u5b78\u6821\u9577\uf9e1\u5bb6\u540c \u660e\u767d\u5730\uf96f\u3002",
                "sec_num": null
            },
            {
                "text": "31 \"The advocacy of core curriculum teaching is in itself a very important education for teachers.\" Lin Ku-fang says that when NHMC was set up it made broad-based education one of its founding principles, but discovered that attitudes were very hard to change, because \"people today feel they are respected for their profession rather than their personality.\" Although when first studying an academic discipline one starts from a general outline, nonetheless one must be very well versed in a subject to teach it well.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Sentence alignment based on length",
                "sec_num": null
            },
            {
                "text": "12 \"There is a great sense of challenge about core curriculum teaching, but many people make the mistake of thinking it is very simple,\" says Lin.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u901a\uf9fc\u672c\u8eab\u7684\u63d0\u5021\uff0c\u5c0d\uf934\u5e2b\u5c31\u662f\u5f88\u91cd \u8981\u7684\u6559\u80b2\uff0c\u300d\u6587\u5316\u8a55\uf941\u8005\uf9f4\u8c37\u82b3\u6307\u51fa\uff0c \u5357\u83ef\u6210\uf9f7\u6642\u5c31\u628a\u901a\uf9fc\u6559\u80b2\u8996\u70ba\u5275\u6821\uf9e4 \uf9a3\uff0c\u4f46\u9084\u662f\u767c\u73fe\u89c0\uf9a3\u554f\u984c\u6700\u96e3\u7a81\u7834\uff0c \u56e0\u70ba\u300c\u73fe\u4ee3\u4eba\u5e38\u89ba\u5f97\u81ea\u5df1\u88ab\u5c0a\u91cd\u662f\u56e0 \u70ba\u6211\u7684\u5c08\u696d\uff0c\u800c\uf967\u662f\u6211\u7684\u4eba\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "\uf941\uff0c\u4f46\u771f\u7684\u5f97\uf943\u901a\uff0c\u624d\u6559\u5f97\u597d\uff0c \u300c\u901a\uf9fc \u6311\u6230\u610f\u5473\u5f88\u6fc3\uff0c\u4f46\u5927\u5bb6\u90fd\u8aa4\u4ee5\u70ba\u5f88\u7c21 \u55ae\u3002\u300d 11 The scope of core curriculum teaching appears very broad, but it still has to start from the basics.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u96d6\u7136\u4e00\u9580\u5b78\u554f\u6700\u521d\uf95a\u6642\u662f\u67d0\u67d0\u5b78\u5c0e",
                "sec_num": null
            },
            {
                "text": "\u901a\uf9fc\u7bc4\u570d\u770b\u8d77\uf92d\u5f88\u5ee3\uff0c\u4f46\u9084\u662f\u7531\u57fa\u790e \u51fa\u767c\uff0c\uf9f4\u8c37\u82b3\u8a8d\u70ba\uff0c\u4efb\u4f55\u5b78\u79d1\u90fd\u53ef\u4ee5 \u5f9e\u300c\u4eba\u8207\u81ea\u7136\u300d \u3001 \u300c\u4eba\u8207\u4eba\u300d \u3001 \u300c\u4eba\u8207\u8d85 \u81ea\u7136\u6216\u81ea\u6211\u300d\u4e09\u500b\u5c64\u6b21\uf92d\u770b\u3002 11 In Lin Ku-fang's view, any branch of academic learning can be viewed on the three levels of \"man and nature,\" \"man and man,\" and \"man and the supernatural, or that which transcends self.\" \u5c31\u4ee5\u5730\u7403\u79d1\u5b78\u9019\u9580\u5c08\u696d\u5b78\u79d1\u70ba\uf9b5\uff0c\u4eba \u6240\u8a8d\uf9fc\u7684\u81ea\u7136\u662f\u5c08\u696d\uff0c\u63d0\u5347\u5230\u4eba\u8207\u751f \u614b\u7684\u4e92\u52d5\u3001\u4eba\u8207\u672a\uf92d\u751f\u547d\u8655\u5883\u5c31\u662f\u901a \uf9fc\u3002 10 To take the example of earth science, a very specialized discipline, man's cognitive knowledge of nature is its specialist content, but to go a step higher and investigate the interactive relationship between man and ecology or man and the future condition of life requires a broad-based, multidisciplinary approach. Sentence alignment based on punctuation. 21 \"The advocacy of core curriculum teaching is in itself a very important education for teachers.\" Lin Ku-fang says that when NHMC was set up it made broad-based education one of its founding principles, but discovered that attitudes were very hard to change, because \"people today feel they are respected for their profession rather than their personality.\" \u300c\u901a\uf9fc\u672c\u8eab\u7684\u63d0\u5021\uff0c\u5c0d\uf934\u5e2b\u5c31\u662f\u5f88\u91cd \u8981\u7684\u6559\u80b2\uff0c\u300d\u6587\u5316\u8a55\uf941\u8005\uf9f4\u8c37\u82b3\u6307\u51fa\uff0c \u5357\u83ef\u6210\uf9f7\u6642\u5c31\u628a\u901a\uf9fc\u6559\u80b2\u8996\u70ba\u5275\u6821\uf9e4 \uf9a3\uff0c\u4f46\u9084\u662f\u767c\u73fe\u89c0\uf9a3\u554f\u984c\u6700\u96e3\u7a81\u7834\uff0c \u56e0\u70ba\u300c\u73fe\u4ee3\u4eba\u5e38\u89ba\u5f97\u81ea\u5df1\u88ab\u5c0a\u91cd\u662f\u56e0 \u70ba\u6211\u7684\u5c08\u696d\uff0c\u800c\uf967\u662f\u6211\u7684\u4eba\u3002\u300d",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u96d6\u7136\u4e00\u9580\u5b78\u554f\u6700\u521d\uf95a\u6642\u662f\u67d0\u67d0\u5b78\u5c0e",
                "sec_num": null
            },
            {
                "text": "11 Although when first studying an academic \u96d6\u7136\u4e00\u9580\u5b78\u554f\u6700\u521d\uf95a\u6642\u662f\u67d0\u67d0\u5b78\u5c0e discipline one starts from a general outline, nonetheless one must be very well versed in a subject to teach it well.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u96d6\u7136\u4e00\u9580\u5b78\u554f\u6700\u521d\uf95a\u6642\u662f\u67d0\u67d0\u5b78\u5c0e",
                "sec_num": null
            },
            {
                "text": "11 \"There is a great sense of challenge about core curriculum teaching, but many people make the mistake of thinking it is very simple,\" says Lin.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\uf941\uff0c\u4f46\u771f\u7684\u5f97\uf943\u901a\uff0c\u624d\u6559\u5f97\u597d\uff0c",
                "sec_num": null
            },
            {
                "text": "21 The scope of core curriculum teaching appears very broad, but it still has to start from the basics. In Lin Ku-fang's view, any branch of academic learning can be viewed on the three levels of \"man and nature,\" \"man and man,\" and \"man and the supernatural, or that which transcends self.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u300c\u901a\uf9fc\u6311\u6230\u610f\u5473\u5f88\u6fc3\uff0c\u4f46\u5927\u5bb6\u90fd\u8aa4\u4ee5 \u70ba\u5f88\u7c21\u55ae\u3002\u300d",
                "sec_num": null
            },
            {
                "text": "11 To take the example of earth science, a very specialized discipline, man's cognitive knowledge of nature is its specialist content, but to go a step higher and investigate the interactive relationship between man and ecology or man and the future condition of life requires a broad-based, multidisciplinary approach.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "\u901a\uf9fc\u7bc4\u570d\u770b\u8d77\uf92d\u5f88\u5ee3\uff0c\u4f46\u9084\u662f\u7531\u57fa\u790e \u51fa\u767c\uff0c\uf9f4\u8c37\u82b3\u8a8d\u70ba\uff0c\u4efb\u4f55\u5b78\u79d1\u90fd\u53ef\u4ee5 \u5f9e\u300c\u4eba\u8207\u81ea\u7136\u300d \u3001 \u300c\u4eba\u8207\u4eba\u300d \u3001 \u300c\u4eba\u8207\u8d85 \u81ea\u7136\u6216\u81ea\u6211\u300d\u4e09\u500b\u5c64\u6b21\uf92d\u770b\u3002",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "We are indebted to Dr. Jason Chang for helpful discussions and suggestions. We acknowledge the support for this study provided through grants by the National Science Council and Ministry of Education, Taiwan (NSC-2213-E-238-015, NSC 90-2411-H-007-033-MC and MOE EX-91-E-FA06-4-4), and by the MOEA under the Software Technology for Advanced Network Application Project of the Institute for Information Industry.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgement",
                "sec_num": null
            },
            {
                "text": "\u3001\u3002\u3002 21 Wang Mang, who usurped the throne in 9 AD, named his daughter Jie (\"nimble and quick\"). The daughter of the emperor Huan Di (132-167 AD) was named Jian (\"solid and resolute\") while her mother, the empress Deng, had the even more emphatic name of Mengnu, which means \"fierce woman\"! \u738b\u83bd\u306e\u5a18\u306e\u540d\u306f\u300c\u5022\u300d \u3001\u5f8c\u6f22\u306e\u6853\u5e1d\u306e \u5a18\u306e\u540d\u306f\u300c\u5805\u300d\u3068\u3044\u3044\u3001\u6853\u5e1d\u306e\u6642\u306e\u7687 \u540e\u306e\u540d\u306f\u3001\u3088\u308a\u76f4\u63a5\u7684\u306a\u300c\u731b\uf981\u300d\u3068\u3044 \u3046\u3082\u306e\u3060\u3063\u305f\u306e\u3067\u3042\u308b\u3002 , , ( \" \" ) . ( ) ( \" \" ) , , , \" \" ! \u300c\u300d \u3001 \u300c\u300d \u3001\u3001 \u300c\u300d \u3002 11 Says Liu, \"These names show that society at that time had not yet come to hold the two sexes to such very different standards.\"\u300c\u3053\u306e\u73fe\u8c61\u306f\u3001 \u7537\u6027\u3068\uf981\u6027\u306e\u9053\u5fb3\ufa08\u70ba \u306b\u5bfe\u3059\u308b\u793e\u4f1a\u306e\u8981\u6c42\u304c\u3001 \u3042\u307e\u308a\u9055\u308f\u306a \u304b\u3063\u305f\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u300d \u3068\uf9c7\u5897\u8cb4 \u3055\u3093\u306f\u8a00\u3046\u3002 , \" . \" \u300c\u3001\u3001\u300d \u3002 11 Although they were gradually beginning to use specifically feminine names alluding to a gentle and submissive nature, such traits as a resolute spirit and an agile, tough body were also seen as virtues in a woman.\u5f53\u6642\u3001 \u3044\u308f\u3086\u308b\uf981\u6027\u7684\u306a\u540d\u524d\u3082\u3057\u3060\u3044 \u306b\u5897\u3048\u3066\u304a\u308a\u3001 \uf981\u6027\u3092\u4f4e\u304f\ufa0a\u308b\u3068\u3044\u3046 \u89b3\uf9a3\u3082\u78ba\u304b\u306b\u3042\u3063\u305f\u304c\u3001 \u305d\u308c\u3067\u3082\uf981\u6027 \u304c\u5f37\u304f\u305f\u304f\u307e\u3057\u304f\u3042\u308b\u3053\u3068\u3082\u80af\u5b9a\u3055 \u308c\u3066\u3044\u305f\u306e\u3067\u3042\u308b\u3002 , , .\u3001\u3001\u3001\u3002 11 \"The notion of the ideal woman being soft and weak was not so universally accepted then as it would later come to be.\" \u300c\uf981\u6027\u306f\u5f31\u304f\u3066\u304a\u3068\u306a\u3057\u3044\u65b9\u304c\uf97c\u3044 \u3068\u3059\u308b\u8003\u3048\u306f\u3001 \u5f8c\u306e\u6642\u4ee3\u306e\u3088\u3046\u306b\u7d76\u5bfe \u7684\u306a\u3082\u306e\u3067\u306f\u306a\u304b\u3063\u305f\u3088\u3046\u3067\u3059\u300d \u3068\uf9c7 \u5897\u8cb4\u3055\u3093\u306f\u8a00\u3046\u3002 \" . \" \u300c\u3001\u300d \u3002",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Sentence alignment based on punctuation",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Aligning sentences in parallel corpora",
                "authors": [
                    {
                        "first": "P",
                        "middle": [
                            "F"
                        ],
                        "last": "Brown",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "C"
                        ],
                        "last": "Lai",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [
                            "L"
                        ],
                        "last": "Mercer",
                        "suffix": ""
                    }
                ],
                "year": 1991,
                "venue": "Proceedings of the 29th conference on Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "169--176",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Brown, P. F., J. C. Lai and R. L. Mercer, \"Aligning sentences in parallel corpora,\" Proceedings of the 29th conference on Association for Computational Linguistics, Berkeley, CA, USA. 1991, pp. 169-176.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Translation Term Weighting and Combining Translation Resources in Cross-Language Retrieval",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Gey",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chen, A. and and F. Gey , \"Translation Term Weighting and Combining Translation Resources in Cross-Language Retrieval,\" TREC 2001.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "A Part-of-Speech-Based Alignment Algorithm",
                "authors": [
                    {
                        "first": "K",
                        "middle": [
                            "H"
                        ],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [
                            "H"
                        ],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 1994,
                "venue": "Proceedings of 15th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "166--171",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chen, K.H. and H.H. Chen, \"A Part-of-Speech-Based Alignment Algorithm,\" Proceedings of 15th International Conference on Computational Linguistics, Kyoto, 1994, pp. 166-171.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Aligning Sentences in Bilingual Corpora Using Lexical Information",
                "authors": [
                    {
                        "first": "S",
                        "middle": [
                            "F"
                        ],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 1993,
                "venue": "Proceedings of ACL-93",
                "volume": "",
                "issue": "",
                "pages": "9--16",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chen, S. F., \"Aligning Sentences in Bilingual Corpora Using Lexical Information,\" Proceedings of ACL-93, Columbus OH, 1993, pp. 9-16.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Adaptive Bilingual Sentence Alignment",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Chuang",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [
                            "N"
                        ],
                        "last": "You",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "S"
                        ],
                        "last": "Chang",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Lecture Notes in Artificial Intelligence",
                "volume": "2499",
                "issue": "",
                "pages": "21--30",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chuang, T., G.N. You and J.S. Chang , \"Adaptive Bilingual Sentence Alignment,\" Lecture Notes in Artificial Intelligence 2499, pp. 21-30.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "An Experiment in Hybrid Dictionary and Statistical Sentence Alignment",
                "authors": [
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Collier",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Ono",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Hirakawa",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "268--274",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Collier, N., K. Ono and H. Hirakawa, \"An Experiment in Hybrid Dictionary and Statistical Sentence Alignment,\" COLING-ACL 1998, pp. 268-274.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "The Misconception of High-Frequency Words in Scandinavian Translation",
                "authors": [
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Danielsson",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "M\u00fchlenbock",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "AMTA",
                "volume": "",
                "issue": "",
                "pages": "158--168",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Danielsson, P. and K. M\u00fchlenbock, Small but Efficient,\"The Misconception of High-Frequency Words in Scandinavian Translation,\" AMTA 2000, pp. 158-168.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Bilingual Terminology Extraction: An Approach based on a Multilingual thesaurus Applicable to Comparable Corpora",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "D\u00e9jean",
                        "suffix": ""
                    },
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Gaussier",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sadat",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the 19th International Conference on Computational Linguistics COLING",
                "volume": "",
                "issue": "",
                "pages": "218--224",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D\u00e9jean, H., \u00c9 Gaussier and F. Sadat, \"Bilingual Terminology Extraction: An Approach based on a Multilingual thesaurus Applicable to Comparable Corpora,\" Proceedings of the 19th International Conference on Computational Linguistics COLING 2002, Taipei, Taiwan, pp. 218-224.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "The Microsoft Research Machine Translation System",
                "authors": [
                    {
                        "first": "W",
                        "middle": [
                            "B"
                        ],
                        "last": "Dolan",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Pinkham",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "D"
                        ],
                        "last": "Richardson",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Msr-Mt",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "AMTA",
                "volume": "",
                "issue": "",
                "pages": "237--239",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dolan, W. B., J. Pinkham and S. D. Richardson, MSR-MT, \"The Microsoft Research Machine Translation System,\" AMTA 2002, pp. 237-239.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "A program for aligning sentences in bilingual corpora",
                "authors": [
                    {
                        "first": "W",
                        "middle": [
                            "A"
                        ],
                        "last": "Gale",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [
                            "W"
                        ],
                        "last": "Church",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Computational Linguistics",
                "volume": "19",
                "issue": "",
                "pages": "75--102",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gale, W. A. and K. W. Church , \"A program for aligning sentences in bilingual corpora,\" Computational Linguistics, vol. 19, pp. 75-102.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Translingual vocabulary mappings for multilingual information access",
                "authors": [
                    {
                        "first": "F",
                        "middle": [
                            "C"
                        ],
                        "last": "Gey",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [
                            "K"
                        ],
                        "last": "Buckland",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [
                            "R"
                        ],
                        "last": "Larson",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "455--456",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gey, F.C., A. Chen, M.K. Buckland and R. R. Larson , \"Translingual vocabulary mappings for multilingual information access,\" SIGIR 2002, pp. 455-456.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "The TransCheck System",
                "authors": [
                    {
                        "first": "J-M",
                        "middle": [],
                        "last": "Jutras",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Proc. of Applied Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "127--134",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jutras, J-M., An Automatic Reviser, \"The TransCheck System,\" Proc. of Applied Natural Language Processing, pp. 127-134.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Text-Translation Alignment",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Kay",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "R\u00f6scheisen",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Computational Linguistics",
                "volume": "19",
                "issue": "1",
                "pages": "121--142",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kay, M. and M. R\u00f6scheisen, \"Text-Translation Alignment,\" Computational Linguistics, 19:1, pp. 121-142.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "A class-based approach to word alignment",
                "authors": [
                    {
                        "first": "S",
                        "middle": [
                            "J"
                        ],
                        "last": "Ker",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "S"
                        ],
                        "last": "Chang",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Computational Linguistics",
                "volume": "23",
                "issue": "2",
                "pages": "313--344",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ker, S.J. and J.S. Chang, \"A class-based approach to word alignment,\" Computational Linguistics, 23:2, pp. 313-344.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "A Robust Cross-Domain Bilingual Sentence Alignment Model",
                "authors": [
                    {
                        "first": "T",
                        "middle": [
                            "L"
                        ],
                        "last": "Kueng",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [
                            "Y"
                        ],
                        "last": "Su",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the 19th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kueng, T.L. and K.Y. Su, \"A Robust Cross-Domain Bilingual Sentence Alignment Model,\" Proceedings of the 19th International Conference on Computational Linguistics, 2002.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "A portable algorithm for mapping bitext correspondence",
                "authors": [
                    {
                        "first": "I",
                        "middle": [
                            "D"
                        ],
                        "last": "Melamed",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "The 35th Conference of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Melamed, I. D., \"A portable algorithm for mapping bitext correspondence,\" In The 35th Conference of the Association for Computational Linguistics (ACL 1997), Madrid, Spain,1997.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Bitext Maps and Alignment via Pattern Recognition",
                "authors": [
                    {
                        "first": "I",
                        "middle": [
                            "D"
                        ],
                        "last": "Melamed",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Computational Linguistics",
                "volume": "25",
                "issue": "1",
                "pages": "107--130",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Melamed, I. D., \"Bitext Maps and Alignment via Pattern Recognition,\" Computational Linguistics, 25(1), pp.107-130, March, 1999.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Fast and Accurate Sentence Alignment of Bilingual Corpora",
                "authors": [
                    {
                        "first": "R",
                        "middle": [
                            "C"
                        ],
                        "last": "Moore",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "135--144",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Moore, R.C., \"Fast and Accurate Sentence Alignment of Bilingual Corpora,\" AMTA 2002, pp. 135-144.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Cognates Alignment",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Ribeiro",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Dias",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Lopes",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Mexia",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the Machine Translation Summit VIII (MT Summit VIII) -Machine Translation in the Information Age",
                "volume": "",
                "issue": "",
                "pages": "287--292",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ribeiro, A., G. Dias, G. Lopes and J. Mexia,\" Cognates Alignment,\" In Bente Maegaard (ed.), Proceedings of the Machine Translation Summit VIII (MT Summit VIII) -Machine Translation in the Information Age, Santiago de Compostela, Spain, 2001, pp. 287-292.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Longman Dictionary of Applied Linguistics",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Richards",
                        "suffix": ""
                    }
                ],
                "year": 1985,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Richards, J. et al., \"Longman Dictionary of Applied Linguistics,\" Longman, 1985.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Using cognates to align sentences in bilingual corpora",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Simard",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Foster",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Isabelle",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Proceedings of TMI92",
                "volume": "",
                "issue": "",
                "pages": "67--81",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Simard, M., G. Foster and P. Isabelle, \"Using cognates to align sentences in bilingual corpora,\" Proceedings of TMI92, Montreal, Canada, pp. 67-81.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "The New University: Breaking down the Departmental Barriers",
                "authors": [
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sinorama",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sinorama , The New University: Breaking down the Departmental Barriers, June, the 3 rd article, 1998.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "Bilingual Collocation Extraction Based on Linguistic and Statistical Analyses",
                "authors": [
                    {
                        "first": "",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wu, \"Bilingual Collocation Extraction Based on Linguistic and Statistical Analyses,\" Master thesis, National Tsing Hua University, Taiwan, 2003.",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Aligning a parallel English-Chinese corpus statistically with lexical criteria",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": 1994,
                "venue": "Proceedings of the 32nd Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "80--87",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wu, D., \"Aligning a parallel English-Chinese corpus statistically with lexical criteria,\" Proceedings of the 32nd Annual Meeting of the Association for Computational Linguistics, New Mexico, USA, 1994, pp. 80-87.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "TotalRecall: A Bilingual Concordance for Computer Assisted Translation and Language Learning",
                "authors": [
                    {
                        "first": "J",
                        "middle": [
                            "C"
                        ],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [
                            "C"
                        ],
                        "last": "Yeh",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [
                            "C"
                        ],
                        "last": "Chuang",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [
                            "C"
                        ],
                        "last": "Shei",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "S"
                        ],
                        "last": "Chang",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wu, J.C., K.C. Yeh, T.C. Chuang, W.C. Shei and J.S. Chang, \"TotalRecall: A Bilingual Concordance for Computer Assisted Translation and Language Learning,\" ACL2003 workshop.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Researches on Punctuation Marks",
                "authors": [
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yang, Y., \"Researches on Punctuation Marks,\" Tien-Chien Publishing, Hong Kong.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Automatic Construction of English/Chinese Parallel Corpora",
                "authors": [
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Journal of American Society of Information Science and Technology",
                "volume": "54",
                "issue": "8",
                "pages": "730--742",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yang, C. and K. Li, \"Automatic Construction of English/Chinese Parallel Corpora,\" Journal of American Society of Information Science and Technology, 54(8), 2003, pp 730-742.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "The relationships between German-English [Gale and Church 1993] and Chinese-English bilingual paragraph lengths [Chuang et al. 2002]. The correlations are 0.99 and 0.95 for the German-English and Chinese-English cases, respectively. The expected ratios and the corresponding standard deviations are (0.92, 0.1124) and (4.614, 0.84) for the German-English and Chinese-English cases, respectively."
            },
            "FIGREF1": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "His father, Tom, is a well-known scholar.(2c) \u4ed6\u7684\u7236\u89aa\u6e6f\u59c6\u662f\u4e00\u4f4d\u6709\u540d\u7684\u5b78\u8005\u3002"
            },
            "FIGREF3": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "Figure 4. Compatibility of translation pairs vs. random pairs with n=m=8"
            },
            "FIGREF4": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "Compatibility of translation pairs vs. random pairs with n=m = 12"
            },
            "FIGREF5": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": ". The translation probability functions corresponding to 1-1, 2-2, 1-0, and 0-1 English-Chinese punctuation matches are shown in"
            },
            "FIGREF6": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "maximum number of punctuation marks in either the English text or the Chinese text in the k th sentence to be aligned;"
            },
            "FIGREF7": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "length-based sentence alignment criterion involves a length-related probability distribution function ( | ) P match \u03b4 , where \u03b4 is a function of the sentence length of the source language c l and the sentence length of the target language e l"
            },
            "FIGREF9": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "Drumming was at first the thing I loved most (8c) \u5f8c\uf92d\u537b\u8b8a\u6210\u908a\u6253\u908a\u7761 (8e) but later it became half drumming, half sleeping (9c) \u4e00\u500b\u6708\uf9d1\u842c\u5143\u7684\u6b7b\u5de5\u4f5c (9e) just a job for NT$60,000 a month (10c) \u859b\u5cb3\u8868\u793a (10e) says Simon"
            },
            "TABREF2": {
                "num": null,
                "html": null,
                "content": "<table><tr><td colspan=\"5\">The frequency counts and the conditional probabilities of 1-1</td></tr><tr><td colspan=\"5\">English-Chinese punctuation matches, sorted according to count</td></tr><tr><td>E p</td><td>C p</td><td>Match type</td><td>Count</td><td>Prob.</td></tr><tr><td>,</td><td>\uff0c</td><td>1-1</td><td>541</td><td>0.809874</td></tr><tr><td>.</td><td>\u3002</td><td>1-1</td><td>336</td><td>0.657528</td></tr><tr><td>\"</td><td>\ufe41</td><td>1-1</td><td>131</td><td>0.34203</td></tr><tr><td>.</td><td>\uff0c</td><td>1-1</td><td>113</td><td>0.221133</td></tr><tr><td>\"</td><td>\ufe42</td><td>1-1</td><td>112</td><td>0.292423</td></tr><tr><td>\"</td><td>\u300c</td><td>1-1</td><td>65</td><td>0.16971</td></tr><tr><td>\"</td><td>\u300d</td><td>1-1</td><td>59</td><td>0.154044</td></tr><tr><td>,</td><td>\u3001</td><td>1-1</td><td>56</td><td>0.083832</td></tr><tr><td>,</td><td>\u3002</td><td>1-1</td><td>41</td><td>0.061377</td></tr><tr><td>!</td><td>\uff01</td><td>1-1</td><td>38</td><td>0.883508</td></tr><tr><td>.</td><td>\u2026</td><td>1-1</td><td>30</td><td>0.058708</td></tr><tr><td>?</td><td>\u3002</td><td>1-1</td><td>17</td><td>0.447277</td></tr><tr><td>:</td><td/><td/><td/><td/></tr></table>",
                "type_str": "table",
                "text": ""
            },
            "TABREF3": {
                "num": null,
                "html": null,
                "content": "<table><tr><td>2-2</td></tr></table>",
                "type_str": "table",
                "text": ""
            },
            "TABREF4": {
                "num": null,
                "html": null,
                "content": "<table><tr><td>Punctuation Match Type</td><td>Count</td><td>Probability</td></tr><tr><td>0-1</td><td>588.0005</td><td>0.225027</td></tr><tr><td>1-0</td><td>286.001</td><td>0.109452</td></tr><tr><td>1-1</td><td>1698.076</td><td>0.649852</td></tr><tr><td>1-2</td><td>2.466198</td><td>0.000944</td></tr><tr><td>2-1</td><td>0.965034</td><td>0.000369</td></tr><tr><td>2-2</td><td>37.19216</td><td>0.014233</td></tr></table>",
                "type_str": "table",
                "text": ""
            }
        }
    }
}