| { |
| "paper_id": "L18-1036", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T11:39:18.013790Z" |
| }, |
| "title": "Building Parallel Monolingual Gan Chinese Dialects Corpus", |
| "authors": [ |
| { |
| "first": "Fan", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University", |
| "location": { |
| "postCode": "330022", |
| "settlement": "Nanchang", |
| "country": "China" |
| } |
| }, |
| "email": "xufan@jxnu.edu.cn" |
| }, |
| { |
| "first": "Mingwen", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University", |
| "location": { |
| "postCode": "330022", |
| "settlement": "Nanchang", |
| "country": "China" |
| } |
| }, |
| "email": "mwwang@jxnu.edu.cn" |
| }, |
| { |
| "first": "Maoxi", |
| "middle": [], |
| "last": "Li", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Jiangxi Normal University", |
| "location": { |
| "postCode": "330022", |
| "settlement": "Nanchang", |
| "country": "China" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "Automatic language identification of an input sentence or a text written in similar languages, varieties or dialects is an important task in natural language processing. In this paper, we propose a scheme to represent Gan (Jiangxi province of China) Chinese dialects. In particular, it is a two-level and fine-grained representation using Chinese character, Chinese Pinyin and Chinese audio forms. Guided by the scheme, we manually annotate a Gan Chinese Dialects Corpus (GCDC) including 131.5 hours and 310 documents with 6 different genres, containing news, official document, story, prose, poet, letter and speech, from 19 different Gan regions. In addition, the preliminary evaluation on 2-way, 7-way and 20-way sentence-level Gan Chinese Dialects Identification (GCDI) justifies the appropriateness of the scheme to Gan Chinese dialects analysis and the usefulness of our manually annotated GCDC.", |
| "pdf_parse": { |
| "paper_id": "L18-1036", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "Automatic language identification of an input sentence or a text written in similar languages, varieties or dialects is an important task in natural language processing. In this paper, we propose a scheme to represent Gan (Jiangxi province of China) Chinese dialects. In particular, it is a two-level and fine-grained representation using Chinese character, Chinese Pinyin and Chinese audio forms. Guided by the scheme, we manually annotate a Gan Chinese Dialects Corpus (GCDC) including 131.5 hours and 310 documents with 6 different genres, containing news, official document, story, prose, poet, letter and speech, from 19 different Gan regions. In addition, the preliminary evaluation on 2-way, 7-way and 20-way sentence-level Gan Chinese Dialects Identification (GCDI) justifies the appropriateness of the scheme to Gan Chinese dialects analysis and the usefulness of our manually annotated GCDC.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Automatic language identification of an input sentence or a document is an important task in Natural Language Processing (NLP), especially when processing speech or social media messages. Currently, the interest in language resources and its computational models for the study of similar languages, varieties and dialects has been growing substantially in the last few years (Zampieri et al, 2014 (Zampieri et al, , 2015 (Zampieri et al, , 2017 Malmasi et al., 2016) . Meanwhile, an increasing number of dialect corpus and corresponding computational models have been released for Catalan, Russian, Slovene, etc. However, no free corpus has been released for the similar, varieties or dialects in Mandarin Chinese. Therefore, in this paper, we focus on the corpus building and its computational model design for the closely related parallel monolingual Gan Chinese languages. As we all know, Chinese is spoken in different regions, with distinct differences among regions. There are different expressions for a same concept among the closely related Gan Chinese languages, varieties and dialects. For example, \u4eca\u91cc jin li 'today', \u4eca\u5bb6 jin jia 'today', \u4eca\u5b81 jin ning 'today', \u4eca\u515c jin dou 'today', \u4eca\u671d jin zhao 'today' are the valid expressions in Nanchang, Yichun, Jian, Fuzhou and Yingtan district in Jiangxi province (Gan in short) of China, respectively. Although these expressions are different, they have the same semantic meanings. They all refer to \u4eca\u5929 jin tian 'today' in Mandarin Chinese (called \u666e\u901a\u8bdd Putonghua 'common language' in Mainland China). More specifically, firstly, we present a scheme to handle Gan Chinese dialects which is a fine-grained representation using Chinese character, Chinese Pinyin and Chinese audio forms. Based on the scheme, we manually annotate a parallel Gan Chinese Dialects Corpus (GCDC) consists of 310 documents with 6 different genres (news, official document, story, prose, poet, letter and speech) from 19 different Gan regions. As a byproduct, the corpus contains the parallel Gan Chinese audio with 131.5 hours. Besides, we conduct a preliminary experiment on the proposed GCDC through the sentence-level Gan Chinese Dialects Identification (GCDI) task. The simple but effective character Chinese Pinyin uni-gram yields a strong baseline on 2-way, 7-way and 20-way Gan dialects discrimination. The overall accuracy can reach to 78.64% on the fine-grained 20-way classification, which shows the automatic Gan Chinese dialects identification should be feasible. The evaluation result justifies the appropriateness of the scheme to Gan Chinese dialects analysis and the usefulness of our manually annotated GCDC. The rest of this paper is organized as follows. Section 2 overviews related work. In Section 3, we present the scheme to deal with Gan Chinese dialects. Section 4 describes the annotation and an annotation instance of the GCDC. In Section 5, we present our preliminary experiment for the sentence-level Gan Chinese dialects identification on the proposed GCDC, and we conclude this work in Section 6 and present future directions.", |
| "cite_spans": [ |
| { |
| "start": 375, |
| "end": 396, |
| "text": "(Zampieri et al, 2014", |
| "ref_id": "BIBREF26" |
| }, |
| { |
| "start": 397, |
| "end": 420, |
| "text": "(Zampieri et al, , 2015", |
| "ref_id": "BIBREF27" |
| }, |
| { |
| "start": 421, |
| "end": 444, |
| "text": "(Zampieri et al, , 2017", |
| "ref_id": "BIBREF25" |
| }, |
| { |
| "start": 445, |
| "end": 466, |
| "text": "Malmasi et al., 2016)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1." |
| }, |
| { |
| "text": "In this section, we describe the representative dialect corpus and its corresponding discrimination models.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "2." |
| }, |
| { |
| "text": "In the past decade, several parallel corpora among different languages have been proposed, e.g. Chinese-English (Ayan and Dorr, 2006) , Japanese-English (Takezawa et al., 2002) and French-English (Mihalcea and Pedersen, 2003) . They are annotated either at wordlevel or phrase-level alignment between two different languages (bilingual). Recently, many researchers pay attention to the parallel corpora only in the closely related languages (monolingual), varieties and dialects (Zampieri et al, 2014 (Zampieri et al, , 2015 (Zampieri et al, , 2017 Malmasi et al., 2016) which containing Bulgarian, Macedonian, etc. and a group containing texts written in a set of other languages. However, none parallel corpora in the closely related languages in the Gan dialects has been freely released so far. The representative certain scale parallel corpora is the Greater China Region (GCR) corpora (Xu et al., 2015) which focus on Mandarin with simplified and traditional scripts.", |
| "cite_spans": [ |
| { |
| "start": 96, |
| "end": 133, |
| "text": "Chinese-English (Ayan and Dorr, 2006)", |
| "ref_id": null |
| }, |
| { |
| "start": 153, |
| "end": 176, |
| "text": "(Takezawa et al., 2002)", |
| "ref_id": "BIBREF18" |
| }, |
| { |
| "start": 196, |
| "end": 225, |
| "text": "(Mihalcea and Pedersen, 2003)", |
| "ref_id": "BIBREF13" |
| }, |
| { |
| "start": 479, |
| "end": 500, |
| "text": "(Zampieri et al, 2014", |
| "ref_id": "BIBREF26" |
| }, |
| { |
| "start": 501, |
| "end": 524, |
| "text": "(Zampieri et al, , 2015", |
| "ref_id": "BIBREF27" |
| }, |
| { |
| "start": 525, |
| "end": 548, |
| "text": "(Zampieri et al, , 2017", |
| "ref_id": "BIBREF25" |
| }, |
| { |
| "start": 549, |
| "end": 570, |
| "text": "Malmasi et al., 2016)", |
| "ref_id": null |
| }, |
| { |
| "start": 891, |
| "end": 908, |
| "text": "(Xu et al., 2015)", |
| "ref_id": "BIBREF21" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Parallel Corpus", |
| "sec_num": "2.1" |
| }, |
| { |
| "text": "Generally speaking, language identification among different languages is a task that can be solved at a high accuracy. For example, Simoes et al. (2014) achieved 97% accuracy for discriminating among 25 unrelated languages. However, it is generally difficult to distinguish between related languages or variations of a specific language (see Zampieri et al, 2014, 2015 for example). To be more specific, Ranaivo-Malancon (2006) proposed features based on frequencies of character n-grams to identify Malay and Indonesian. Zampieri and Gebre (2012) found that word uni-grams gave very similar performance to character n-gram features in the framework of the probabilistic language model for the Brazilian and European Portuguese language discrimination. Tiedemann and Ljubesic (2012) ; Ljubesic and Kranjcic (2015) showed that the Na\u00ef ve Bayes classifier with uni-grams achieved high accuracy for the South Slavic languages identification. Grefenstette (1995) ; Lui and Cook (2013) found that bag-of-words features outperformed the syntax or character sequences-based features for the English varieties. Besides these works, other recent studies include: Spanish varieties identification (2014), Arabic varieties discrimination (Elfardy and Diab,2013; Zaidan and Callison-burch, 2014; Salloum et al.,2016; Tillmann et al.,2014) and Persian and Dari identification (Malmasi and Dras, 2015) ; Indian languages identification (Murthy and Kumar, 2006) .", |
| "cite_spans": [ |
| { |
| "start": 132, |
| "end": 152, |
| "text": "Simoes et al. (2014)", |
| "ref_id": "BIBREF17" |
| }, |
| { |
| "start": 404, |
| "end": 427, |
| "text": "Ranaivo-Malancon (2006)", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 522, |
| "end": 547, |
| "text": "Zampieri and Gebre (2012)", |
| "ref_id": "BIBREF24" |
| }, |
| { |
| "start": 753, |
| "end": 782, |
| "text": "Tiedemann and Ljubesic (2012)", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 785, |
| "end": 813, |
| "text": "Ljubesic and Kranjcic (2015)", |
| "ref_id": "BIBREF6" |
| }, |
| { |
| "start": 939, |
| "end": 958, |
| "text": "Grefenstette (1995)", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 961, |
| "end": 980, |
| "text": "Lui and Cook (2013)", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 1227, |
| "end": 1250, |
| "text": "(Elfardy and Diab,2013;", |
| "ref_id": null |
| }, |
| { |
| "start": 1251, |
| "end": 1283, |
| "text": "Zaidan and Callison-burch, 2014;", |
| "ref_id": "BIBREF23" |
| }, |
| { |
| "start": 1284, |
| "end": 1304, |
| "text": "Salloum et al.,2016;", |
| "ref_id": null |
| }, |
| { |
| "start": 1305, |
| "end": 1326, |
| "text": "Tillmann et al.,2014)", |
| "ref_id": "BIBREF20" |
| }, |
| { |
| "start": 1363, |
| "end": 1387, |
| "text": "(Malmasi and Dras, 2015)", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 1422, |
| "end": 1446, |
| "text": "(Murthy and Kumar, 2006)", |
| "ref_id": "BIBREF14" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Dialects Identification Models", |
| "sec_num": "2.2" |
| }, |
| { |
| "text": "In this section, we present the scheme to Gan Chinese dialects which has two-level partitions and three forms.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Annotation Scheme", |
| "sec_num": "3." |
| }, |
| { |
| "text": "Gan Chinese is spoken in different regions in Jiangxi province of China, with distinct differences between two regions. To be specific, Table 2 shows a two-level Gan dialects partition is provided. The first level contains six regions of Gan dialects (Yan Sen, 1986) , such as \u660c\u9756\u7247 'chang jing region', \u5b9c\u840d\u7247 'yi ping region', \u5409\u83b2\u7247 'ji lian region', \u629a\u5e7f\u7247 'fu guang region', \u9e70\u5f0b\u7247 'yin yi region', \u5ba2\u5bb6\u8bdd 'Hakka'. The six regions are further divided into 19 sub-regions in the second level. For example, \u660c \u9756\u7247 'chang jing region' contains 5 subregions, such as \u65b0\u5efa 'xinjian', \u5357\u660c 'nanchang' and so on.", |
| "cite_spans": [ |
| { |
| "start": 251, |
| "end": 266, |
| "text": "(Yan Sen, 1986)", |
| "ref_id": "BIBREF22" |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 136, |
| "end": 143, |
| "text": "Table 2", |
| "ref_id": "TABREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Two-level Partitions", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Chinese Pinyin: Pinyin is basically the alphabet for the Chinese language. The Pinyin system was invented to help people pronounce the sound of the Chinese characters. It is a Romanization system used to learn Mandarin. It transcribes the sounds of Mandarin using the Western (Roman) alphabet. It is well know that pronunciation is vital for any language. Therefore, we annotate Chinese Pinyin into our corpus. Figure 1 and Table 1 show pitch contours of lexical tones in Mandarin (Chen et al., 2016) . In our corpus annotation, we annotate the pitch height with 1, 2, 3 and 4 accordingly. We observe that the same concept can be expressed using different linguistic expressions for the different region of Gan dialects as mentioned in the Introduction section. Chinese audio: Furthermore, we present Chinese audio as a byproduct in this corpus. It consists of the parallel Gan Chinese audio and mandarin Chinese sound for each document.", |
| "cite_spans": [ |
| { |
| "start": 481, |
| "end": 500, |
| "text": "(Chen et al., 2016)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 411, |
| "end": 419, |
| "text": "Figure 1", |
| "ref_id": "FIGREF0" |
| }, |
| { |
| "start": 424, |
| "end": 431, |
| "text": "Table 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Three Forms", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "In this section, we address the key issues with the GCDC annotation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Gan Chinese Dialects Corpus", |
| "sec_num": "4." |
| }, |
| { |
| "text": "The annotator team consists of a Ph.D. in Chinese linguistics as the supervisor (senior annotator) and 19 undergraduate students from different 19 Gan regions in Chinese linguistics as annotators. An annotator of a given region works only in data of his/her area. The annotation is done in four phases. In the first phase, the annotators spend 1 month on learning the principles of scheme. In the second phase, the annotators spend 1 month on independently annotating the same 30 documents, and another 1 month on crosschecking to resolve the difference and to revise the guidelines. In the third phase, the annotators spend 2 months on annotating the remaining 280 documents. In the final phase, the supervisor spends 1 month carefully proofread all 310 documents.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Annotator Training", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Currently, the GCDC corpus consists of the representative 19 sub-regions of Gan dialects and their statistics as shown in Table 2 . Given the above scheme, we annotate parallel 310 XML-style documents with 6 different genres (news, official document, story, prose, poet, letter and speech), containing 218 newswire documents from Chinese Treebank 6.0 with Linguistic Data Consortium (LDC) catalog number LDC2007T36, and other 92 documents for the remaining genres from the internet using Baidu search engine with official document, story, prose, poet, letter and speech as key words. Specifically, we don't have parallel sentences for each variant of each sentence in all documents, and the documents included differ among the dialects but are all parallel with respect to a Mandarin translation. We require the annotators to annotate the documents included differ among the dialects but are all parallel with respect to a Mandarin translation. As a byproduct, it has the 131.5-hour audios, wherein 69.0 hours Gan dialects sound and 62.50 hours Putonghua sound, and the total number of sentence in the corpus is 3,878. ", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 122, |
| "end": 129, |
| "text": "Table 2", |
| "ref_id": "TABREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Corpus Statistics", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "It is very challenging to check the agreement between annotators. We focus on \u9e70 \u5f0b \u7247 'yin yi region', and require another 2 annotators from this region to annotate 30 documents, 173 sentences, from the corpus. We calculate the annotation consistency value which is 0.93 for Pinyin. Due to the homophone phenomenon is obvious in Chinese character, we don't calculate the agreement for Chinese character. The high inter-annotator consistency in Chinese Pinyin guarantees the corpus's quality. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Quality Assurance", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "As mentioned in the Introduction section, automatic language identification of an input text is an important task in Natural Language Processing (NLP) because somebody must determine the language of the text before applying tools trained on specific language. For the sentence-level language identification, a user is given a single sentence, and the user needs to identify the language. Below, we recast the sentence-level dialects identification in the Gan dialects as a multi-class classification problem. Firstly, we will describe some features. Then, these features are fed into a classifier to determine the dialect of a sentence.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Preliminary Experimentation", |
| "sec_num": "5." |
| }, |
| { |
| "text": "In this section, we represent the character-level N-gram features.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Features", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "Chinese Character Pinyin N-gram: According to the related work (Nikola and Denis, 2015; Cagri and Taraka, 2016), n-grams with n \uf0a3 3 are effective features for discriminating general languages. Also, Cagri and Taraka (2016) showed their simple linear SVM model with ngram feature is quite useful and hard to beat by current neural network models. Compared with English, no space exists between words in Chinese sentence. Therefore, we use character uni-grams, bi-grams and tri-grams in Chinese Pinyin as features. We take Pinyin with lexical tones or without it as different two kinds of features. Chinese Character N-gram: While our corpus provides both Chinese character and Chinese Pinyin simultaneously, we also present Chinese Character unigrams, bi-grams and tri-grams as features. This is because sometime Pinyin is not available in a specific situation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Features", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "Classifier: After extracting the above proposed features, we train a single multi-class linear kernel support vector machine using LIBLINEAR (Fan et al., 2008) for Gan Chinese dialects identification. They adopt the default parameters such as verbosity level with 1, trade-off between training error and margin with 0.01, slack rescaling, zero/one loss. Evaluation Metric: We report system's performance using accuracy, which is the ratio of the number of the correctly predicted sentence divided by the total number of sentence for Gan dialects.", |
| "cite_spans": [ |
| { |
| "start": 131, |
| "end": 159, |
| "text": "LIBLINEAR (Fan et al., 2008)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Classifier and Evaluation Metric", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "For the Gan dialects dataset, we generate three scenarios using 5-fold cross validation:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Classifier and Evaluation Metric", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "(1) 2-way detection: We try to distinguish between two groups of dialects, the ones is \u666e \u901a \u8bdd 'Putonghua', and the others are the left 19 subregions of Gan dialects;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Classifier and Evaluation Metric", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "(2) 7-way detection: The level-1 Gan dialects of \u660c \u9756\u7247 'chang jing region', \u5409\u83b2\u7247 'ji lian region', \u629a\u5e7f\u7247 'fu guang region', \u5b9c\u840d\u7247 'yi ping region', \u9e70\u5f0b\u7247 'yin yi region', \u5ba2\u5bb6\u8bdd 'Hakka' and \u666e\u901a \u8bdd 'Putonghua' as shown in Table 2 are considered; (3) 20-way detection: We detect both Mandarin and other level-2 19 Gan dialects of \u65b0\u5efa xinjian, \u5357 \u660c nanchang, \u6e56\u53e3 hukou, etc. as shown in Table 2 are all considered.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 208, |
| "end": 232, |
| "text": "Table 2 are considered;", |
| "ref_id": "TABREF1" |
| }, |
| { |
| "start": 369, |
| "end": 377, |
| "text": "Table 2", |
| "ref_id": "TABREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Classifier and Evaluation Metric", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "In this section, we report the experiment results for the Gan Chinese dialects identification on our dataset. Table 4 shows the performance on Chinese character Pinyin feature. As can be seen, the character uni-gram Pinyin feature yields the best performance on both news and other types of genres. Obviously, the performance of 2-way classification is higher than both 7-way and 20-way language discrimination. Strangely, the performance of the more fine-grained 20 different dialect labels task achieves higher results than the identification of only 7 labels. We attribute it to the parallel nature of the corpus. Basically, the performance is increased with the increment of the number of training data. Table 4 : The performance of sentence-level Gan dialects identification using Chinese character Pinyin. 'Y' stands for the corpus with lexical tones, 'N' indicates none.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 110, |
| "end": 117, |
| "text": "Table 4", |
| "ref_id": null |
| }, |
| { |
| "start": 708, |
| "end": 715, |
| "text": "Table 4", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Experimentation Results", |
| "sec_num": "5.3" |
| }, |
| { |
| "text": "In addition, lexical tones in uni-gram Pinyin reflect the fine-grained characteristic of Gan dialects. Using lexical tones is better than without it. We also conduct the trigram case, but the performance is lower than bi-gram about 20%. Compared with uni-gram, there are much noise in both bi-gram and tri-gram features. The proposed uni-gram features significantly outperforms the bi-gram ones with p<0.05 using paired t-test for significance. It shows the effectiveness of the proposed Chinese character Pinyin feature. More specifically, the accuracy for each level-1 Gan dialects for news domain with Chinese character uni-gram Pinyin feature is reported in Table 5 . As shown, we gain the best identification performance for \u666e \u901a \u8bdd 'Putonghua', while the accuracy of \u9e70\u5f0b\u7247 'yin yi region' is the worst one. The reason is that the difference between \u9e70\u5f0b\u7247 'yin yi region' and \u666e\u901a\u8bdd 'Putonghua' is not obvious enough as shown in Table 5 : Accuracy of each level-1 Gan dialects on news domain. To be more specific, we report the confusion table for each level-1 Gan dialect using Chinese character uni-gram Pinyin in Table 6 . As can be seen, most instances have been correctly classified. Due to the challenge of discrimination for the closely related languages in the Gan dialects, some instances still have been falsely classified. For example, we can know that some instances falsely classified from \u660c\u9756 \u7247'chang jing region' to'\u666e\u901a \u8bdd ' Putonghua (20) is similar to those from \u9e70\u5f0b\u7247'yin yi region' to \u666e\u901a\u8bdd 'Putonghua' (15). The reason is that the \u660c\u9756\u7247 'chang jing region' and \u9e70\u5f0b\u7247 'yin yi region' are closed to \u666e\u901a\u8bdd 'Putonghua'. Table 6 : Confusion table of each level-1 Gan dialects using uni-gram Pinyin with tones on news domain. Remark: L 1 stands for \u660c \u9756 \u7247 'chang jing region', L 2 indicates \u629a\u5e7f\u7247'fu guang region', L 3 donates \u5ba2\u5bb6\u8bdd 'Hakka', L 4 refers to \u5409\u83b2\u7247'ji lian region', L 5 means \u666e \u901a\u8bdd 'Putonghua', L 6 represents \u5b9c\u840d\u7247'yi ping region', L 7 embodies \u9e70\u5f0b\u7247'yin yi region'. Table 7 shows the performance on Chinese character. Again, the character uni-gram feature yields best performance on both news and other type of genres. It also yields promising results on the extremely difficult fine-grained 20-way language classification. Table 7 : The performance of sentence-level Gan dialects identification using Chinese character.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 662, |
| "end": 669, |
| "text": "Table 5", |
| "ref_id": null |
| }, |
| { |
| "start": 925, |
| "end": 932, |
| "text": "Table 5", |
| "ref_id": null |
| }, |
| { |
| "start": 1112, |
| "end": 1119, |
| "text": "Table 6", |
| "ref_id": "TABREF5" |
| }, |
| { |
| "start": 1619, |
| "end": 1626, |
| "text": "Table 6", |
| "ref_id": "TABREF5" |
| }, |
| { |
| "start": 1966, |
| "end": 1973, |
| "text": "Table 7", |
| "ref_id": null |
| }, |
| { |
| "start": 2224, |
| "end": 2231, |
| "text": "Table 7", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Domain", |
| "sec_num": null |
| }, |
| { |
| "text": "Predicted label L 1 L 2 L 3 L 4 L 5 L 6 L 7 True label L 1 48 2 0 0 20 0 0 L 2 2 35 2 1 6 0 0 L 3 0 0 39 0 19 0 0 L 4 1 1 4 58 8 0 0 L 5 1 2 2 4 212 0 1 L 6 1 0 1 0 3 28 0 L 7 1 0 1 0 15 0 5", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Domain", |
| "sec_num": null |
| }, |
| { |
| "text": "In this paper, we annotate a parallel Gan Chinese Dialects Corpus (GCDC) based on different levels of modularity (written and spoken data) with different layers of annotations and transcription. Meanwhile, we conduct a preliminary experiment on the proposed GCDC through sentence-level Gan Chinese dialects identification task on different levels of granularity. The simple but effective character Chinese Pinyin and character uni-gram yields a strong baseline, especially on the 20-way Gan dialects discrimination, which shows the fine-grained automatic Gan Chinese dialects identification should be feasible. In future work, we would like to explore more features without the need of using the Pinyin notation, enlarge the scale of the corpus, and test other classifiers. Furthermore, we will finally investigate how dialect identification can help other NLP tasks.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusions", |
| "sec_num": "6." |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": "The authors would like to thank the anonymous reviewers for their comments on this paper. This research was supported by the National Natural Science Foundation of China under Grant No. 61772246, 61402208, 61462044, 61462045, 61662031 ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Acknowledgments", |
| "sec_num": "7." |
| } |
| ], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Large-scale characterization of nonnative mandarin Chinese spoken by speakers of European origin: Analysis on iCALL", |
| "authors": [ |
| { |
| "first": "Nancy", |
| "middle": [ |
| "F" |
| ], |
| "last": "Chen", |
| "suffix": "" |
| }, |
| { |
| "first": "Darren", |
| "middle": [], |
| "last": "Wee", |
| "suffix": "" |
| }, |
| { |
| "first": "Rong", |
| "middle": [], |
| "last": "Tong", |
| "suffix": "" |
| }, |
| { |
| "first": "Bin", |
| "middle": [], |
| "last": "Ma", |
| "suffix": "" |
| }, |
| { |
| "first": "Haizhou", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "Speech Communication", |
| "volume": "84", |
| "issue": "216", |
| "pages": "46--56", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Nancy F. Chen, Darren Wee, Rong Tong, Bin Ma, Haizhou Li. (216). Large-scale characterization of non- native mandarin Chinese spoken by speakers of European origin: Analysis on iCALL. Speech Communication, 84:46-56.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Discriminating similar languages : experiments with linear SVMs and neural networks", |
| "authors": [ |
| { |
| "first": "Cagri", |
| "middle": [], |
| "last": "Coltekin", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Rama", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the third workshop on NLP for similar languages, varieties and dialects", |
| "volume": "", |
| "issue": "", |
| "pages": "15--24", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Cagri Coltekin, taraka Rama. (2016). Discriminating similar languages : experiments with linear SVMs and neural networks. In Proceedings of the third workshop on NLP for similar languages, varieties and dialects, pages 15-24, Osaka, Japan, December 2016.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Sentence level dialect identification in Arabic", |
| "authors": [ |
| { |
| "first": "Heba", |
| "middle": [], |
| "last": "Elfardy", |
| "suffix": "" |
| }, |
| { |
| "first": "Mona", |
| "middle": [ |
| "T" |
| ], |
| "last": "Diab", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Proceedings of ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "456--461", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Heba Elfardy and Mona T Diab. (2013). Sentence level dialect identification in Arabic. In Proceedings of ACL, pages 456-461.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "LIBLINEAR: A library for large linear classification", |
| "authors": [ |
| { |
| "first": "Kai-Wei", |
| "middle": [], |
| "last": "Rong-En Fan", |
| "suffix": "" |
| }, |
| { |
| "first": "Cho-Jui", |
| "middle": [], |
| "last": "Chang", |
| "suffix": "" |
| }, |
| { |
| "first": "Xang-Rui", |
| "middle": [], |
| "last": "Hsieh", |
| "suffix": "" |
| }, |
| { |
| "first": "Chih-Jen", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Lin", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Journal of Machine Learning Research", |
| "volume": "9", |
| "issue": "", |
| "pages": "1871--1874", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xang-Rui Wang, and Chih-Jen Lin. (2008). LIBLINEAR: A library for large linear classification. Journal of Machine Learning Research, 9: 1871-1874.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Going beyond aer: An extensive analysis of word alignments and their impact on MT", |
| "authors": [ |
| { |
| "first": "Bonnie", |
| "middle": [ |
| "J" |
| ], |
| "last": "Necip Fazil Ayan", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Dorr", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Proceedings of the 21 st International Conference on Computational Linguistics and the 44 th annual Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "9--16", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Necip Fazil Ayan and Bonnie J Dorr. (2006). Going beyond aer: An extensive analysis of word alignments and their impact on MT. In Proceedings of the 21 st International Conference on Computational Linguistics and the 44 th annual Meeting of the Association for Computational Linguistics, pages 9-16.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Comparing two language identification schemes", |
| "authors": [ |
| { |
| "first": "Gregory", |
| "middle": [], |
| "last": "Grefenstette", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "Proceedings of Analisi Statistica dei Dati Testuali (JADT)", |
| "volume": "", |
| "issue": "", |
| "pages": "263--268", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Gregory Grefenstette. (1995). Comparing two language identification schemes. In Proceedings of Analisi Statistica dei Dati Testuali (JADT), pages 263-268.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Discriminating between closely related languages on twitter", |
| "authors": [ |
| { |
| "first": "Nikola", |
| "middle": [], |
| "last": "Ljubesic", |
| "suffix": "" |
| }, |
| { |
| "first": "Denis", |
| "middle": [], |
| "last": "Kranjcic", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Informatica", |
| "volume": "39", |
| "issue": "1", |
| "pages": "1--8", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Nikola Ljubesic and Denis Kranjcic. (2015). Discriminating between closely related languages on twitter. Informatica, 39(1):1-8.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Classifying English documents by national dialect", |
| "authors": [ |
| { |
| "first": "Marco", |
| "middle": [], |
| "last": "Lui", |
| "suffix": "" |
| }, |
| { |
| "first": "Paul", |
| "middle": [], |
| "last": "Cook", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Proceedings of Australasian Language Technology Workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "5--15", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marco Lui and Paul Cook. (2013). Classifying English documents by national dialect. In Proceedings of Australasian Language Technology Workshop, pages 5- 15.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Automatic identification of close languages-case study: Maly and Indonesian", |
| "authors": [ |
| { |
| "first": "Bali", |
| "middle": [], |
| "last": "Ranaivo-Malancon", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "ECTI Transactions on Computer and Information Technology", |
| "volume": "2", |
| "issue": "", |
| "pages": "126--134", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bali Ranaivo-Malancon. (2006). Automatic identification of close languages-case study: Maly and Indonesian. ECTI Transactions on Computer and Information Technology, 2:126-134.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Language variety identification in Spanish tweets", |
| "authors": [ |
| { |
| "first": "Wolfgang", |
| "middle": [], |
| "last": "Maier", |
| "suffix": "" |
| }, |
| { |
| "first": "Carlos", |
| "middle": [], |
| "last": "Gomez-Rodriguez", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of the LT4CloseLang Workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "25--35", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Wolfgang Maier and Carlos Gomez-Rodriguez. (2014). Language variety identification in Spanish tweets. In Proceedings of the LT4CloseLang Workshop, pages 25- 35.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Automatic Language Identification for Persian and Dari texts", |
| "authors": [ |
| { |
| "first": "Shervin", |
| "middle": [], |
| "last": "Malmasi", |
| "suffix": "" |
| }, |
| { |
| "first": "Mark", |
| "middle": [], |
| "last": "Dras", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of PACLING", |
| "volume": "", |
| "issue": "", |
| "pages": "59--64", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Shervin Malmasi and Mark Dras. (2015). Automatic Language Identification for Persian and Dari texts. In Proceedings of PACLING, pages 59-64.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Discriminating Between Similar Languages and Arabic Dialect Identification: A Report on the Third DSL Shared Task", |
| "authors": [], |
| "year": null, |
| "venue": "Proceedings of the COLING VarDial Workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "1--14", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Discriminating Between Similar Languages and Arabic Dialect Identification: A Report on the Third DSL Shared Task. In Proceedings of the COLING VarDial Workshop, pages 1-14.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "An Evaluation Exercise for Word Alignment", |
| "authors": [ |
| { |
| "first": "Rada", |
| "middle": [], |
| "last": "Mihalcea", |
| "suffix": "" |
| }, |
| { |
| "first": "Ted", |
| "middle": [], |
| "last": "Pedersen", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "Proceedings of the HLT-NAACL 2003 Workshop on Building and Using Parallel Texts: Data Driven Machine Translation and Beyond", |
| "volume": "", |
| "issue": "", |
| "pages": "1--10", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Rada Mihalcea and Ted Pedersen. (2003). An Evaluation Exercise for Word Alignment. In Proceedings of the HLT-NAACL 2003 Workshop on Building and Using Parallel Texts: Data Driven Machine Translation and Beyond, pages 1-10.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Language identification from small text samples", |
| "authors": [ |
| { |
| "first": "Kavi", |
| "middle": [], |
| "last": "Narayana Murthy", |
| "suffix": "" |
| }, |
| { |
| "first": "G Bharadwaja", |
| "middle": [], |
| "last": "Kumar", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Journal of Quantitative Linguistics", |
| "volume": "13", |
| "issue": "01", |
| "pages": "57--80", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kavi Narayana Murthy and G Bharadwaja Kumar. (2006). Language identification from small text samples. Journal of Quantitative Linguistics, 13(01):57-80.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "A Simple Baseline for Discriminating Similar Languages", |
| "authors": [ |
| { |
| "first": "Matthew", |
| "middle": [], |
| "last": "Purver", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the LT4VarDial", |
| "volume": "", |
| "issue": "", |
| "pages": "1--5", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Matthew Purver. (2015). A Simple Baseline for Discriminating Similar Languages. In Proceedings of the LT4VarDial, pages 1-5.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Sentence level dialect identification for machine translation system selection", |
| "authors": [ |
| { |
| "first": "Wael", |
| "middle": [], |
| "last": "Salloum", |
| "suffix": "" |
| }, |
| { |
| "first": "Heba", |
| "middle": [], |
| "last": "Elfardy", |
| "suffix": "" |
| }, |
| { |
| "first": "Linda", |
| "middle": [], |
| "last": "Alamir-Salloum", |
| "suffix": "" |
| }, |
| { |
| "first": "Nizar", |
| "middle": [], |
| "last": "Habash", |
| "suffix": "" |
| }, |
| { |
| "first": "Mona", |
| "middle": [], |
| "last": "Diab", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "772--778", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Wael Salloum, Heba Elfardy, Linda Alamir-Salloum, Nizar Habash, and Mona Diab. (2014). Sentence level dialect identification for machine translation system selection. In Proceedings of ACL, pages 772-778.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Language identification: a neural network approach", |
| "authors": [ |
| { |
| "first": "Alberto", |
| "middle": [], |
| "last": "Simoes", |
| "suffix": "" |
| }, |
| { |
| "first": "Jose", |
| "middle": [ |
| "Joao" |
| ], |
| "last": "Almeida", |
| "suffix": "" |
| }, |
| { |
| "first": "Simon D", |
| "middle": [], |
| "last": "Byers", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of the 3rd Symposium on Languages, Applications and Technologies, SLATE'14", |
| "volume": "", |
| "issue": "", |
| "pages": "252--265", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Alberto Simoes, Jose Joao Almeida, and Simon D Byers. (2014). Language identification: a neural network approach. In Proceedings of the 3rd Symposium on Languages, Applications and Technologies, SLATE'14, pages 252-265.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "Toward a Broad-coverage Bilingual Corpus for Speech Translation of Travel Conversations in the Real World", |
| "authors": [ |
| { |
| "first": "Toshiyuki", |
| "middle": [], |
| "last": "Takezawa", |
| "suffix": "" |
| }, |
| { |
| "first": "Eiichiro", |
| "middle": [], |
| "last": "Sumita", |
| "suffix": "" |
| }, |
| { |
| "first": "Fumiaki", |
| "middle": [], |
| "last": "Sugaya", |
| "suffix": "" |
| }, |
| { |
| "first": "Hirofumi", |
| "middle": [], |
| "last": "Yamamoto", |
| "suffix": "" |
| }, |
| { |
| "first": "Seiichi", |
| "middle": [], |
| "last": "Yamamoto", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proceedings of the 3 rd International Conference on Language Resources and Evaluation (LREC)", |
| "volume": "", |
| "issue": "", |
| "pages": "147--152", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Toshiyuki Takezawa, Eiichiro Sumita, Fumiaki Sugaya, Hirofumi Yamamoto, and Seiichi Yamamoto. (2002). Toward a Broad-coverage Bilingual Corpus for Speech Translation of Travel Conversations in the Real World. In Proceedings of the 3 rd International Conference on Language Resources and Evaluation (LREC), pages 147-152.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Efficient discrimination between closely related languages", |
| "authors": [ |
| { |
| "first": "Jorg", |
| "middle": [], |
| "last": "Tiedemann", |
| "suffix": "" |
| }, |
| { |
| "first": "Nikola", |
| "middle": [], |
| "last": "Ljubesic", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proceedings of COLING", |
| "volume": "", |
| "issue": "", |
| "pages": "2619--2634", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jorg Tiedemann and Nikola Ljubesic. (2012). Efficient discrimination between closely related languages. In Proceedings of COLING, pages 2619-2634.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Improved Sentence-Level Arabic Dialect Classification", |
| "authors": [ |
| { |
| "first": "Christoph", |
| "middle": [], |
| "last": "Tillmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Yaser", |
| "middle": [], |
| "last": "Al-Onaizan", |
| "suffix": "" |
| }, |
| { |
| "first": "Saab", |
| "middle": [], |
| "last": "Mansour", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of the First Workshop on Applying NLP Tools to Similar Languages", |
| "volume": "", |
| "issue": "", |
| "pages": "110--119", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Christoph Tillmann,Yaser Al-Onaizan, and Saab Mansour. (2014). Improved Sentence-Level Arabic Dialect Classification. In Proceedings of the First Workshop on Applying NLP Tools to Similar Languages, pages 110- 119.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "Building Monolingual Word Alignment Corpus for the Greater China Region", |
| "authors": [ |
| { |
| "first": "Fan", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "Xiongfei", |
| "middle": [], |
| "last": "Xu", |
| "suffix": "" |
| }, |
| { |
| "first": "Mingwen", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Maoxi", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the Joint Workshop on Language Technology for Closely Related Languages,Varieties and Dialects", |
| "volume": "", |
| "issue": "", |
| "pages": "85--94", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Fan Xu, Xiongfei Xu, Mingwen Wang, Maoxi Li. (2015). Building Monolingual Word Alignment Corpus for the Greater China Region. In Proceedings of the Joint Workshop on Language Technology for Closely Related Languages,Varieties and Dialects, pages 85-94, Hissar, Bulgaria, September 10, 2015. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "The Partitions of Jiangxi Dialects", |
| "authors": [ |
| { |
| "first": "Sen", |
| "middle": [], |
| "last": "Yan", |
| "suffix": "" |
| } |
| ], |
| "year": 1986, |
| "venue": "Dialect", |
| "volume": "", |
| "issue": "1", |
| "pages": "19--38", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sen Yan. The Partitions of Jiangxi Dialects. Dialect, 1986,(1):19-38 (in Chinese).", |
| "links": null |
| }, |
| "BIBREF23": { |
| "ref_id": "b23", |
| "title": "Arabic dialect identification", |
| "authors": [ |
| { |
| "first": "F", |
| "middle": [], |
| "last": "Omar", |
| "suffix": "" |
| }, |
| { |
| "first": "Chris", |
| "middle": [], |
| "last": "Zaidan", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Callison-Burch", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Computational Linguistics", |
| "volume": "40", |
| "issue": "1", |
| "pages": "171--202", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Omar F Zaidan and Chris Callison-Burch. (2014). Arabic dialect identification. Computational Linguistics, 40(1): 171-202.", |
| "links": null |
| }, |
| "BIBREF24": { |
| "ref_id": "b24", |
| "title": "Automatic identification of language varieties: The case of Portuguese", |
| "authors": [ |
| { |
| "first": "Marcos", |
| "middle": [], |
| "last": "Zampieri", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Binyam Gebrekidan Gebre", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proceedings of KONVENS", |
| "volume": "", |
| "issue": "", |
| "pages": "233--237", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marcos Zampieri and Binyam Gebrekidan Gebre. (2012). Automatic identification of language varieties: The case of Portuguese. In Proceedings of KONVENS, pages 233-237.", |
| "links": null |
| }, |
| "BIBREF25": { |
| "ref_id": "b25", |
| "title": "Findings of the VarDial Evaluation Compaign", |
| "authors": [ |
| { |
| "first": "Marcos", |
| "middle": [], |
| "last": "Zampieri", |
| "suffix": "" |
| }, |
| { |
| "first": "Shervin", |
| "middle": [], |
| "last": "Malmasi", |
| "suffix": "" |
| }, |
| { |
| "first": "Nikola", |
| "middle": [], |
| "last": "Ljubesic", |
| "suffix": "" |
| }, |
| { |
| "first": "Preslav", |
| "middle": [], |
| "last": "Nakov", |
| "suffix": "" |
| }, |
| { |
| "first": "Ahmed", |
| "middle": [], |
| "last": "Ali", |
| "suffix": "" |
| }, |
| { |
| "first": "Jorg", |
| "middle": [], |
| "last": "Tiedemann", |
| "suffix": "" |
| }, |
| { |
| "first": "Yves", |
| "middle": [], |
| "last": "Scherrer", |
| "suffix": "" |
| }, |
| { |
| "first": "Noemi", |
| "middle": [], |
| "last": "Aepli", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the EACL VarDial workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "1--15", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marcos Zampieri, Shervin Malmasi, Nikola Ljubesic, Preslav Nakov, Ahmed Ali, Jorg Tiedemann, Yves Scherrer, Noemi Aepli. (2017). Findings of the VarDial Evaluation Compaign. In Proceedings of the EACL VarDial workshop, pages 1-15.", |
| "links": null |
| }, |
| "BIBREF26": { |
| "ref_id": "b26", |
| "title": "A report on the DSL shared task", |
| "authors": [ |
| { |
| "first": "Marcos", |
| "middle": [], |
| "last": "Zampieri", |
| "suffix": "" |
| }, |
| { |
| "first": "Liling", |
| "middle": [], |
| "last": "Tan", |
| "suffix": "" |
| }, |
| { |
| "first": "Nikola", |
| "middle": [], |
| "last": "Ljubesic", |
| "suffix": "" |
| }, |
| { |
| "first": "Jorg", |
| "middle": [], |
| "last": "Tiedemann", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proceedings of the COLING VarDial Workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "58--67", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marcos Zampieri, Liling Tan, Nikola Ljubesic, and Jorg Tiedemann. (2014). A report on the DSL shared task 2014. In Proceedings of the COLING VarDial Workshop, pages 58-67.", |
| "links": null |
| }, |
| "BIBREF27": { |
| "ref_id": "b27", |
| "title": "Overview of the DSL shared task 2015", |
| "authors": [ |
| { |
| "first": "Marcos", |
| "middle": [], |
| "last": "Zampieri", |
| "suffix": "" |
| }, |
| { |
| "first": "Liling", |
| "middle": [], |
| "last": "Tan", |
| "suffix": "" |
| }, |
| { |
| "first": "Nikola", |
| "middle": [], |
| "last": "Ljubesic", |
| "suffix": "" |
| }, |
| { |
| "first": "Jorg", |
| "middle": [], |
| "last": "Tiedemann", |
| "suffix": "" |
| }, |
| { |
| "first": "Preslav", |
| "middle": [], |
| "last": "Nakov", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proceedings of the RANLP LT4VarDial workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "1--9", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marcos Zampieri, Liling Tan, Nikola Ljubesic, Jorg Tiedemann, and Preslav Nakov. (2015). Overview of the DSL shared task 2015. In Proceedings of the RANLP LT4VarDial workshop, pages 1-9.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "uris": null, |
| "text": "Pitch contours of lexical tones in Mandarin. The vertical axis denotes pitch height, whereas the horizontal staves indicate different tone levels within one's comfortable vocal range.", |
| "type_str": "figure", |
| "num": null |
| }, |
| "TABREF1": { |
| "num": null, |
| "text": "shows the statistics in detail with the number of non-news genre are shown in parentheses.", |
| "html": null, |
| "content": "<table><tr><td>Dialects</td><td colspan=\"2\">Dialects</td><td>Number of</td><td>Number of</td></tr><tr><td>region</td><td colspan=\"2\">location</td><td>document</td><td>sentence</td></tr><tr><td>(level-1)</td><td colspan=\"2\">(level-2)</td><td/><td/></tr><tr><td>\u660c\u9756\u7247</td><td>\u65b0</td><td>\u5efa</td><td>10(4)</td><td>-</td></tr><tr><td>chang jing</td><td>xinjian</td><td/><td/><td/></tr><tr><td>region</td><td colspan=\"2\">\u5357\u660c</td><td>10(3)</td><td>-</td></tr><tr><td/><td colspan=\"2\">nanchang</td><td/><td/></tr><tr><td/><td>\u6e56</td><td>\u53e3</td><td>10(2)</td><td>-</td></tr><tr><td/><td>hukou</td><td/><td/><td/></tr><tr><td/><td colspan=\"2\">\u90fd\u660c</td><td>6(6)</td><td>-</td></tr><tr><td/><td colspan=\"2\">douchang</td><td/><td/></tr><tr><td/><td colspan=\"2\">\u9756\u5b89</td><td>16(0)</td><td>-</td></tr><tr><td/><td>jingan</td><td/><td/><td/></tr><tr><td/><td>#total</td><td/><td>52(15)</td><td>353(101)</td></tr><tr><td>\u5409\u83b2\u7247</td><td colspan=\"2\">\u5409\u5b89 jian</td><td>10(2)</td><td>-</td></tr><tr><td>ji lian</td><td colspan=\"3\">\u5409\u6c34 jishui 8(6)</td><td>-</td></tr><tr><td>region</td><td>\u6c38</td><td>\u4e30</td><td>21(12)</td><td>-</td></tr><tr><td/><td colspan=\"2\">yongfeng</td><td/><td/></tr><tr><td/><td>#total</td><td/><td>39(20)</td><td>362(138)</td></tr><tr><td>\u629a\u5e7f\u7247</td><td colspan=\"2\">\u8fdb\u8d24</td><td>10(2)</td><td>-</td></tr><tr><td>fu guang</td><td>jingxian</td><td/><td/><td/></tr><tr><td>region</td><td colspan=\"2\">\u4e1c\u4e61</td><td>14(2)</td><td>-</td></tr><tr><td/><td colspan=\"2\">dongxiang</td><td/><td/></tr><tr><td/><td colspan=\"2\">\u629a\u5dde</td><td>17(12)</td><td>-</td></tr><tr><td/><td>fuzhou</td><td/><td/><td/></tr><tr><td/><td>#total</td><td/><td>41(16)</td><td>230(105)</td></tr><tr><td>\u5b9c\u840d\u7247</td><td colspan=\"2\">\u4e30\u57ce</td><td>10(4)</td><td>-</td></tr><tr><td>yi ping</td><td colspan=\"2\">fengcheng</td><td/><td/></tr><tr><td>region</td><td colspan=\"2\">\u5b9c\u4e30</td><td>5(3)</td><td>-</td></tr><tr><td/><td>yifeng</td><td/><td/><td/></tr><tr><td/><td colspan=\"2\">\u840d\u4e61</td><td>7(5)</td><td>-</td></tr><tr><td/><td colspan=\"2\">pingxiang</td><td/><td/></tr><tr><td/><td>#total</td><td/><td>22(12)</td><td>165(99)</td></tr><tr><td>\u9e70\u5f0b\u7247</td><td>\u4f59\u5e72</td><td/><td>10(4)</td><td>-</td></tr><tr><td>yin yi</td><td>Yugan</td><td/><td/><td/></tr><tr><td>region</td><td colspan=\"2\">\u4e0a\u9976</td><td>8(8)</td><td>-</td></tr><tr><td/><td colspan=\"2\">shangrao</td><td/><td/></tr><tr><td/><td>#total</td><td/><td>18(12)</td><td>110(63)</td></tr><tr><td>\u5ba2\u5bb6\u8bdd</td><td colspan=\"2\">\u8d63\u5dde</td><td>24(12)</td><td>-</td></tr><tr><td>Hakka</td><td>ganzhou</td><td/><td/><td/></tr><tr><td/><td colspan=\"2\">\u5174\u56fd</td><td>12(2)</td><td>-</td></tr><tr><td/><td>xinguo</td><td/><td/><td/></tr><tr><td/><td colspan=\"3\">\u5927\u4f59 dayu 10(3)</td><td>-</td></tr><tr><td/><td>#total</td><td/><td>46(17)</td><td>294(120)</td></tr><tr><td>\u666e\u901a\u8bdd</td><td/><td/><td>156(92)</td><td>1113(625)</td></tr><tr><td>Putonghua</td><td/><td/><td/><td/></tr><tr><td>#total</td><td/><td/><td>310</td><td>3878</td></tr><tr><td/><td colspan=\"3\">Table 2: Corpus statistics.</td><td/></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF2": { |
| "num": null, |
| "text": "", |
| "html": null, |
| "content": "<table><tr><td>describes an annotation instance of GCDC for</td></tr><tr><td>clarity.</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF3": { |
| "num": null, |
| "text": "An annotation instance for Gan dialects.The example comes from file chtb2946 of CTB (Chinese Tree Bank) released by the LDC. The <voiceInfo> section describes the detail information, such as region, location, sex and age of annotator, genre type, record channel, duration and file. The <sentence> section demonstrates the specific contents including Chinese character and Chinese Pinyin, containing <putongContent> section refers to the Chinese character in Mandarin, <ganContent> section represents the Chinese character in specific Gan dialect. <putongPinyin> section indicates the Chinese Pinyin in Mandarin, while <ganPinyin> section means the Chinese Pinyin in specific Gan dialect. The whole corpus are available through the LREC 2018 repository.", |
| "html": null, |
| "content": "<table/>", |
| "type_str": "table" |
| }, |
| "TABREF5": { |
| "num": null, |
| "text": "", |
| "html": null, |
| "content": "<table><tr><td colspan=\"2\">, also we have enough</td></tr><tr><td>training data for \u666e\u901a\u8bdd 'Putonghua'.</td><td/></tr><tr><td>Dialect</td><td>acc. (%)</td></tr><tr><td>\u660c\u9756\u7247 chang jing region</td><td>68.57</td></tr><tr><td>\u629a\u5e7f\u7247 fu guang region</td><td>76.09</td></tr><tr><td>\u5ba2\u5bb6\u8bdd Hakka</td><td>67.24</td></tr><tr><td>\u5409\u83b2\u7247 ji lian region</td><td>80.56</td></tr><tr><td>\u666e\u901a\u8bdd Putonghua</td><td>95.50</td></tr><tr><td>\u5b9c\u840d\u7247 yi ping region</td><td>84.85</td></tr><tr><td>\u9e70\u5f0b\u7247 yin yi region</td><td>22.73</td></tr></table>", |
| "type_str": "table" |
| } |
| } |
| } |
| } |
