| { | |
| "paper_id": "U14-1018", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T03:09:06.290840Z" | |
| }, | |
| "title": "Sinhala-Tamil Machine Translation: Towards better Translation Quality", | |
| "authors": [ | |
| { | |
| "first": "Randil", | |
| "middle": [], | |
| "last": "Pushpananda", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "Language Technology Research Laboratory", | |
| "institution": "University of Colombo", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "Language Technology Research Laboratory", | |
| "institution": "University of Colombo", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Mahesan", | |
| "middle": [], | |
| "last": "Niranjan", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Southampton", | |
| "location": { | |
| "country": "UK" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "Statistical Machine Translation (SMT) is a well-known and well established datadriven approach used for language translation. The focus of this work is to develop a statistical machine translation system for Sri Lankan languages, Sinhala and Tamil language pair. This paper presents a systematic investigation of how Sinhala-Tamil SMT performance varies with the amount of parallel training data used, in order to find out the minimum needed to develop a machine translation system with acceptable performance.", | |
| "pdf_parse": { | |
| "paper_id": "U14-1018", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "Statistical Machine Translation (SMT) is a well-known and well established datadriven approach used for language translation. The focus of this work is to develop a statistical machine translation system for Sri Lankan languages, Sinhala and Tamil language pair. This paper presents a systematic investigation of how Sinhala-Tamil SMT performance varies with the amount of parallel training data used, in order to find out the minimum needed to develop a machine translation system with acceptable performance.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Sri Lanka is a multi-ethnic, multi-lingual country. Sinhala and Tamil are the national languages of Sri Lanka. The majority of Sri Lankans do not have a good knowledge of languages other than their mother tongue. Therefore a language barrier between the Sinhala and Tamil communities exists. This language barrier and the problems that arose during the last 30 years in the country, encouraged us to a translation application using the SMT approach. This would reduce the language gap between these two communities and thereby help solve a burning issue in the country.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The choice of the Sinhala -Tamil language pair provides some opportunities as well as some challenges. The opportunity is that they share some affinity to each other, having evolved alongside each other in Sri Lanka. The challenges include the sparseness in the availability of data, and the limited research undertaken in them. Hence, developing a successful system with limited resources is our ultimate goal.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "There is very limited research reported in the literature for Sinhala-Tamil machine translation. Ac-cording to (Weerasinghe, 2003) , the Sinhala-Tamil language pair gives better performance compared to the Sinhala-English pair in SMT since they are more closely related to each other owing to their evolution within Sri Lanka. Some important factors to consider when building SMT for the Sinhala-Tamil language pair have been identified in (Sakthithasan et al., 2010) . The limited amount of data, and the restricted domain it represented, makes that word hard to generalize. Another study (Jeyakaran and Weerasinghe, 2011) , explored the applicability of the Kernel Ridge Regression technique to Sinhala-Tamil translation. This research resulted in a hybrid of classical phrase based SMT and Kernel Ridge Regression with two novel solutions for the pre-image problem.", | |
| "cite_spans": [ | |
| { | |
| "start": 111, | |
| "end": 130, | |
| "text": "(Weerasinghe, 2003)", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 440, | |
| "end": 467, | |
| "text": "(Sakthithasan et al., 2010)", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 590, | |
| "end": 623, | |
| "text": "(Jeyakaran and Weerasinghe, 2011)", | |
| "ref_id": "BIBREF1" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Background and Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Owing to the limited amount of parallel data available, it has been not possible to analyze how the results vary with increasing numbers of parallel sentences in Sinhala and Tamil for general purpose MT.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Background and Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Sinhala belongs to the Indo-Aryan language family and Tamil to the Dravidian family. Both Sinhala and Tamil languages are morphologically rich languages: Sinhala has up to 110 noun word forms and up to 282 verb word forms (Welgama et al., 2011) and Tamil has around 40 noun word forms and up to 240 verb word forms (Lushanthan, 2010). Also both these languages are syntactically similar. The typical word order of both these languages are Subject-Object-Verb. However both are flexible with the word order and variant word orders are possible with discourse -pragmatic effects (Liyanage et al., 2012; Wikipedia, 2014) .", | |
| "cite_spans": [ | |
| { | |
| "start": 222, | |
| "end": 244, | |
| "text": "(Welgama et al., 2011)", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 577, | |
| "end": 600, | |
| "text": "(Liyanage et al., 2012;", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 601, | |
| "end": 617, | |
| "text": "Wikipedia, 2014)", | |
| "ref_id": "BIBREF19" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sinhala and Tamil Languages", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "In addition there are some of the aspects of Tamil influence on the structure of the Sinhalese language. The most significant impact of Tamil on Sinhalese has been at the lexical level (Karunatilaka, 2011). \u0d85\u0db8\u0dca\u0db8\u0dcf (/amma/: mother), \u0d85\u0d9a\u0dca\u0d9a\u0dcf (/akka/: elder sister), \u0d85\u0dba\u0dd2\u0dba\u0dcf (/ayya/: elder brother) are some loan words out of more than thousand words borrowed from Tamil to Sinhala (Coperahewa and Arunachalam, 2011) .", | |
| "cite_spans": [ | |
| { | |
| "start": 374, | |
| "end": 408, | |
| "text": "(Coperahewa and Arunachalam, 2011)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sinhala and Tamil Languages", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "The open source statistical machine translation system: MOSES (Koehn et al., 2007) was used with GIZA++ (Och and Ney, 2004) using the standard alignment heuristic grow-diag-final for word alignments. Tri-gram language models were trained on the target side of the parallel data and the target language monolingual corpus by using the Stanford Research Institute language Modeling toolkit (Stolcke and others, 2002) with Kneser-Ney smoothing. The systems were tuned using a small extracted parallel dataset with Minimum Error Rate Training (MERT) (Och, 2003) and then tested with different test sets. Finally, the Bilingual Evaluation Understudy (BLEU) (Papineni et al., 2002) evaluation metric was used to evaluate the output produced by the translation system.", | |
| "cite_spans": [ | |
| { | |
| "start": 62, | |
| "end": 82, | |
| "text": "(Koehn et al., 2007)", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 104, | |
| "end": 123, | |
| "text": "(Och and Ney, 2004)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 388, | |
| "end": 414, | |
| "text": "(Stolcke and others, 2002)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 546, | |
| "end": 557, | |
| "text": "(Och, 2003)", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 652, | |
| "end": 675, | |
| "text": "(Papineni et al., 2002)", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Tools used", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "To build a good baseline system, we need to have a sentence-aligned parallel corpus to train the translation model and a (possibly larger) monolingual corpus of the target language to train the language model. We used the UCSC 1 10M words Sinhala Corpus (Weerasinghe et al., 2007) and the 4M words Tamil Corpus to build the Sinhala and Tamil language models respectively. Both these are open domain corpora mainly with newspaper articles and Technical writing. The characteristics of the Sinhala and Tamil corpora is shown in Table 1 .", | |
| "cite_spans": [ | |
| { | |
| "start": 254, | |
| "end": 280, | |
| "text": "(Weerasinghe et al., 2007)", | |
| "ref_id": "BIBREF15" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 526, | |
| "end": 533, | |
| "text": "Table 1", | |
| "ref_id": "TABREF2" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Data Collection and Data Preprocessing", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "Finding a good large Sinhala-Tamil parallel corpus was the main difficulty. For this purpose we collected a Sinhala-Tamil Parallel Corpus 1 University of Colombo School of Computing which consists of 25500 parallel sentences. This is also an open domain corpus which includes mainly newspaper texts and technical writing. The sentence length of sentences in this corpus was restricted to 8 -12 words. Both Sinhala to Tamil and Tamil to Sinhala translation models were built using this corpus. The characteristics of the Sinhala-Tamil parallel dataset is shown in Using the above parallel corpus, we trained two baseline systems: Sinhala to Tamil and Tamil to Sinhala. First, 500 parallel sentences were extracted randomly as the tuning dataset. Then of the remaining 25000 parallel sentences, 5000 sentences were extracted randomly as the initial dataset. By applying 10-fold cross-validation (Kohavi and others, 1995) (to get an unbiased result), we divided extracted 5000 sentences into 10 mutually exclusive partitions equally and then one of the partitions was used as the testing data and the other nine used as training data. Then we trained and evaluated the system iteratively for all combinations of the datasets and finally calculated the average performance of the results in order to obtain unbiased estimates of accuracy. We repeated the same procedure by adding 5000 more sentences to the initial dataset each time until the remaining dataset was empty.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data Collection and Data Preprocessing", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "Results Figure 1 shows the average BLEU score value variation against the number of parallel sentences in both Sinhala to Tamil and Tamil to Sinhala translation. However, it clearly indicates that much more data would be required to build an acceptable translation model for the Sinhala-Tamil language pair. The results of the Tamil to Sinhala translation system in figure 1 shows that the BLEU score approaches 12.9 when the dataset size reaches 25000. It also shows that results of the Sinhala to Tamil translation only approaches 10.1 for the full dataset of 25000 parallel sentences. The figure 1 shows that when the dataset size is increased from 5000 to 10,000 and 10,000 to 20,000, the increase in performance varies by Also, as shown in table 3, we can clearly see that as the number of sentences are increased, the average perplexity for both Sinhala and Tamil decreases. Sinhala and Tamil datasets were considered separately from the parallel corpus to calculate the perplexity values. These values are very high compared to those of the dominant European languages.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 8, | |
| "end": 16, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Data Collection and Data Preprocessing", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "Here we did an error analysis to identify the problems of the methods we used and to find new methodologies to improve the results.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data Collection and Data Preprocessing", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The BLEU scores for test sets of 5000 and 25000 data samples were taken for the error analysis. The process for the error analysis stated as follows.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error Analysis", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "\u2022 Calculate the number of total words (TotW) and unique words(UniW) in each training (Tr) and test (Te) datasets.", | |
| "cite_spans": [ | |
| { | |
| "start": 38, | |
| "end": 44, | |
| "text": "(TotW)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error Analysis", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "\u2022 Calculate the number of out-of-vocabulary (OOV) words in the test dataset (as a percentage of test dataset).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error Analysis", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "\u2022 Calculate the number of untranslated words (UntransW)(as a percentage of test dataset).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error Analysis", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "\u2022 Calculate the number of translated words which are not in the reference dataset (Targe-tOOV)(as a percentage of test dataset).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error Analysis", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "\u2022 Calculate the number of translated words which are not in the target language model (Target LM OOV) (as a percentage of reference dataset). we can see that the total number of words in the Tamil to Sinhala translation is lower than the Sinhala to Tamil translation in both training and testing datasets. However the unique number of words in the Tamil to Sinhala translation is much higher than the Sinhala to Tamil translation. This clearly shows the complexity of the Tamil language. However, as we expected OOV (unique word) rate is reduced by 8% -10%, when the dataset size is increasing. That is one of the reasons for the increment of BLEU score value. We have identified mainly two problems. According to table 4, 20% of unique words in the test set are not translated even they were in the training set and 17% to 19% of words which are not in the target reference set is in the translated output. Those are occurred due to phrase alignment problems and also the decoding problems. For an example if we need to translate \u0dd9\u0d9c\u0daf\u0dbb (Home) to Tamil, the phrase table consists only \u0dd9\u0d9c\u0daf\u0dbb \u0d91\u0db1\u0dca\u0db1 (Come home) and \u0dd9\u0d9c\u0daf\u0dbb \u0dba\u0db1\u0dca\u0db1 (Go home), then that word will not be translated even that word is in the training set. Since Sinhala and Tamil are lowresourced languages, we need to consider these issues to build a good translation system. We can clearly see that out-of-vocabulary rate and the untranslated word rate is much higher in Tamil to Sinhala Translation. Also when we consider the out-of-vocabulary words, we have found that those words consist of proper names, misspelled words, inflections, derivatives and honorifics. These are the main problems that we could identify from the error analysis. Since human evaluation is very costly, we used only the above technique to do the evaluation. According to the figure 1, we can see that even the OOV words are higher, BLEU score values of Tamil to Sinhala translation is higher. The main reason for this could be the size of the language model since words in the Sinhala monolingual corpus is more than twice as the words in the Tamil monolingual corpus. When consider the Target OOV and Target LM OOV in Tamil to Sinhala Translation is lower compared to the Sinhala to Tamil translation. That could be a another reason to get a higher BLEU score value for Tamil to Sinhala translation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error Analysis", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The purpose of this research was to find out how the SMT systems perform for Sinhala to Tamil and Tamil to Sinhala translation. We can conclude that while Tamil to Sinhala and Sinhala to Tamil translation is unable to produce intelligible output with parallel corpus of just 25000 sentence pairs of relatively short length, we can expect performance to approach usable levels by collecting a large parallel corpora. Using this experience, we are currently collecting a more balanced parallel corpus.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "However the error analysis shows that the sentence length limitations of the Sinhala-Tamil parallel corpus could not be the only reason for the comparatively lower BLEU scores, morphological richness may be the reason to get lower results since misspelled words and proper names are common to other languages too. Furthermore, a preliminary study shows that we can get better perplexity values for the same dataset we used for this research by stemming suffixes of the Sinhala and Tamil parallel sentences. In future, we are planning to investigate and find solutions to these problems and planning to implement a system capable of producing acceptable translations between Sinhala and Tamil for use by the wider community.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "5" | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "The authors would like to thank the anonymous reviewers for their helpful comments and suggestions. This work was supported by the National Research Council, ICT Agency and LK Domain Registry of Sri Lanka. The authors are grateful to past and current members of the Language Technology Research Laboratory of the UCSC, Sri Lanka for their significant contribution in developing the basic linguistic resources needed to carry out the research described above.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Acknowledgment", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "A Dictionary of", | |
| "authors": [ | |
| { | |
| "first": "Sandagomi", | |
| "middle": [], | |
| "last": "Coperahewa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sarojini", | |
| "middle": [], | |
| "last": "Arunachalam", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Godage International publishers", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sandagomi Coperahewa and Sarojini Arunachalam. 2011. A Dictionary of Tamil Word in Sinhala, vol- ume 2. Godage International publishers, Sri Lanka.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "A novel kernel regression based machine translation system for sinhala-tamil translation", | |
| "authors": [ | |
| { | |
| "first": "Mahendran", | |
| "middle": [], | |
| "last": "Jeyakaran", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Proceedings of the 4th Annual UCSC Research Symposium", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mahendran Jeyakaran and Ruvan Weerasinghe. 2011. A novel kernel regression based machine translation system for sinhala-tamil translation. In Proceedings of the 4th Annual UCSC Research Symposium.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Godage International publishers", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Ws Karunatilaka", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "WS Karunatilaka. 2011. Link. Godage International publishers, Sri Lanka.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Moses: Open source toolkit for statistical machine translation", | |
| "authors": [ | |
| { | |
| "first": "Philipp", | |
| "middle": [], | |
| "last": "Koehn", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hieu", | |
| "middle": [], | |
| "last": "Hoang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Alexandra", | |
| "middle": [], | |
| "last": "Birch", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chris", | |
| "middle": [], | |
| "last": "Callison-Burch", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Marcello", | |
| "middle": [], | |
| "last": "Federico", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nicola", | |
| "middle": [], | |
| "last": "Bertoldi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Brooke", | |
| "middle": [], | |
| "last": "Cowan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Wade", | |
| "middle": [], | |
| "last": "Shen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christine", | |
| "middle": [], | |
| "last": "Moran", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Richard", | |
| "middle": [], | |
| "last": "Zens", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "177--180", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, et al. 2007. Moses: Open source toolkit for statistical machine translation. In Pro- ceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions, pages 177-180. Association for Computational Lin- guistics.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "A study of cross-validation and bootstrap for accuracy estimation and model selection", | |
| "authors": [ | |
| { | |
| "first": "Ron", | |
| "middle": [], | |
| "last": "Kohavi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "IJCAI", | |
| "volume": "14", | |
| "issue": "", | |
| "pages": "1137--1145", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ron Kohavi et al. 1995. A study of cross-validation and bootstrap for accuracy estimation and model se- lection. In IJCAI, volume 14, pages 1137-1145.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Dulip Lakmal Herath, and Ruvan Weerasinghe", | |
| "authors": [ | |
| { | |
| "first": "Chamila", | |
| "middle": [], | |
| "last": "Liyanage", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Randil", | |
| "middle": [], | |
| "last": "Pushpananda", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Computational Linguistics and Intelligent Text Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "188--200", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chamila Liyanage, Randil Pushpananda, Dulip Lakmal Herath, and Ruvan Weerasinghe. 2012. A compu- tational grammar of sinhala. In Computational Lin- guistics and Intelligent Text Processing, pages 188- 200. Springer.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Morphological analyzer and generator for tamil language", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Sivaneasharajah Lushanthan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sivaneasharajah Lushanthan. 2010. Morphological analyzer and generator for tamil language. August.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "The alignment template approach to statistical machine translation", | |
| "authors": [ | |
| { | |
| "first": "Josef", | |
| "middle": [], | |
| "last": "Franz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hermann", | |
| "middle": [], | |
| "last": "Och", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Ney", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Computational linguistics", | |
| "volume": "30", | |
| "issue": "4", | |
| "pages": "417--449", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Franz Josef Och and Hermann Ney. 2004. The align- ment template approach to statistical machine trans- lation. Computational linguistics, 30(4):417-449.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Minimum error rate training in statistical machine translation", | |
| "authors": [ | |
| { | |
| "first": "Franz Josef", | |
| "middle": [], | |
| "last": "Och", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of the 41st Annual Meeting on Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "160--167", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Franz Josef Och. 2003. Minimum error rate training in statistical machine translation. In Proceedings of the 41st Annual Meeting on Association for Compu- tational Linguistics-Volume 1, pages 160-167. As- sociation for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Bleu: a method for automatic evaluation of machine translation", | |
| "authors": [ | |
| { | |
| "first": "Kishore", | |
| "middle": [], | |
| "last": "Papineni", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Salim", | |
| "middle": [], | |
| "last": "Roukos", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Todd", | |
| "middle": [], | |
| "last": "Ward", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Wei-Jing", | |
| "middle": [], | |
| "last": "Zhu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of the 40th annual meeting on association for computational linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "311--318", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting on association for compu- tational linguistics, pages 311-318. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Statistical machine translation of systems for sinhala-tamil", | |
| "authors": [ | |
| { | |
| "first": "Sripirakas", | |
| "middle": [], | |
| "last": "Sakthithasan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Weerasinghe", | |
| "middle": [], | |
| "last": "Ruvan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dulip Lakmal", | |
| "middle": [], | |
| "last": "Herath", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Advances in ICT for Emerging Regions (ICTer), 2010 International Conference on", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "62--68", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sripirakas Sakthithasan, ruvan Weerasinghe, and Dulip Lakmal Herath. 2010. Statistical machine translation of systems for sinhala-tamil. In Ad- vances in ICT for Emerging Regions (ICTer), 2010 International Conference on, pages 62-68. IEEE.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Srilm-an extensible language modeling toolkit", | |
| "authors": [ | |
| { | |
| "first": "Andreas", | |
| "middle": [], | |
| "last": "Stolcke", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "INTERSPEECH", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Andreas Stolcke et al. 2002. Srilm-an extensible lan- guage modeling toolkit. In INTERSPEECH.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Learning machine translation from in-domain and out-of-domain data", | |
| "authors": [ | |
| { | |
| "first": "Marco", | |
| "middle": [], | |
| "last": "Turchi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Cyril", | |
| "middle": [], | |
| "last": "Goutte", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nello", | |
| "middle": [], | |
| "last": "Cristianini", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proceedings of the 16th", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marco Turchi, Cyril Goutte, and Nello Cristianini. 2012. Learning machine translation from in-domain and out-of-domain data. In Proceedings of the 16th", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Annual Conference of the European Association for Machine Translation", | |
| "authors": [], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "305--312", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Annual Conference of the European Association for Machine Translation, pages 305-312.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Sinhala tamil parallel corpora subset with a total 1 million words", | |
| "authors": [ | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Randil", | |
| "middle": [], | |
| "last": "Pushpananda", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ruvan Weerasinghe and Randil Pushpananda. 2013. Sinhala tamil parallel corpora subset with a total 1 million words. Technical report, University of Colombo School of Computing.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Ucsc sinhala corpus -pan localization project-phase i", | |
| "authors": [ | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dulip", | |
| "middle": [], | |
| "last": "Herath", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Viraj", | |
| "middle": [], | |
| "last": "Welgama", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nishantha", | |
| "middle": [], | |
| "last": "Medagoda", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Asanka", | |
| "middle": [], | |
| "last": "Wasala", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eranga", | |
| "middle": [], | |
| "last": "Jayalatharachchi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ruvan Weerasinghe, Dulip Herath, Viraj Welgama, Nishantha Medagoda, Asanka Wasala, and Eranga Jayalatharachchi. 2007. Ucsc sinhala corpus -pan localization project-phase i.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Sri lankan tamil corpus", | |
| "authors": [ | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Randil", | |
| "middle": [], | |
| "last": "Pushpananda", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Namal", | |
| "middle": [], | |
| "last": "Udalamatta", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ruvan Weerasinghe, Randil Pushpananda, and Namal Udalamatta. 2013. Sri lankan tamil corpus. Tech- nical report, University of Colombo School of Com- puting and funded by ICT Agency, Sri Lanka.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "A statistical machine translation approach to sinhala-tamil language translation. Towards an ICT enabled Society", | |
| "authors": [ | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ruvan Weerasinghe. 2003. A statistical machine trans- lation approach to sinhala-tamil language transla- tion. Towards an ICT enabled Society, page 136.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Towards a sinhala wordnet", | |
| "authors": [ | |
| { | |
| "first": "Viraj", | |
| "middle": [], | |
| "last": "Welgama", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chamila", | |
| "middle": [], | |
| "last": "Dulip Lakmal Herath", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Namal", | |
| "middle": [], | |
| "last": "Liyanage", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ruvan", | |
| "middle": [], | |
| "last": "Udalamatta", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Tissa", | |
| "middle": [], | |
| "last": "Weerasinghe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Jayawardana", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Proceedings of the Conference on Human Language Technology for Development", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Viraj Welgama, Dulip Lakmal Herath, Chamila Liyan- age, Namal Udalamatta, Ruvan Weerasinghe, and Tissa Jayawardana. 2011. Towards a sinhala word- net. In Proceedings of the Conference on Human Language Technology for Development.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Tamil language -wikipedia, the free encyclopedia", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Wikipedia", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "", | |
| "volume": "30", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wikipedia. 2014. Tamil language -wikipedia, the free encyclopedia. [Online; accessed 30-October- 2014].", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF0": { | |
| "text": "Randil Pushpananda, Ruvan Weerasinghe and Mahesan Niranjan. 2014. Sinhala-Tamil Machine Translation: Towards better Translation Quality. In Proceedings of Australasian Language Technology Association Workshop, pages 129\u2212133.", | |
| "html": null, | |
| "content": "<table/>", | |
| "num": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF2": { | |
| "text": "Characteristics of Sinhala and Tamil Monolingual Corpora", | |
| "html": null, | |
| "content": "<table/>", | |
| "num": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF3": { | |
| "text": "", | |
| "html": null, | |
| "content": "<table><tr><td>Language</td><td>Total Words(TW)</td><td>Unique Words(UW)</td><td>UW/TW</td></tr><tr><td>Sinhala</td><td>252,101</td><td>37,128</td><td>15%</td></tr><tr><td>Tamil</td><td>219,017</td><td>53,024</td><td>24%</td></tr></table>", | |
| "num": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF4": { | |
| "text": "Characteristics of parallel dataset", | |
| "html": null, | |
| "content": "<table/>", | |
| "num": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF6": { | |
| "text": "Average perplexity values and out-ofvocabulary values of the Sinhala-Tamil Parallel Corpus", | |
| "html": null, | |
| "content": "<table/>", | |
| "num": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF8": { | |
| "text": "Results obtained from the error analysis of Sinhala to Tamil translationThe results obtained for the Sinhala to Tamil and Tamil to Sinhala translations are shown in", | |
| "html": null, | |
| "content": "<table><tr><td>Description</td><td colspan=\"2\">5000 TotW UniW</td><td colspan=\"2\">25000 TotW UniW</td></tr><tr><td>Training Dataset</td><td colspan=\"4\">39,044 16,328 194,784 49,402</td></tr><tr><td>Testing Dataset</td><td>4,336</td><td>2,968</td><td>21,462</td><td>10,381</td></tr><tr><td>OOV (%)</td><td>30.32</td><td>43.67</td><td>16.84</td><td>33.85</td></tr><tr><td>UntransW (%)</td><td>40.68</td><td>57.14</td><td>25.08</td><td>48.58</td></tr><tr><td>Reference Dataset</td><td>3,168</td><td>1,307</td><td>17,584</td><td>4,298</td></tr><tr><td>TargetOOV (%)</td><td>10.88</td><td>14.94</td><td>5.01</td><td>11.45</td></tr><tr><td colspan=\"2\">Target LM OOV (%) 0.04</td><td>0.07</td><td>0.15</td><td>0.44</td></tr></table>", | |
| "num": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF9": { | |
| "text": "", | |
| "html": null, | |
| "content": "<table><tr><td>: Results obtained from the error analysis</td></tr><tr><td>of Tamil to Sinhala translation</td></tr><tr><td>table 4 and 5 respectively. When considering</td></tr><tr><td>the 5000 and 25000 datasets in table 4 and 5,</td></tr></table>", | |
| "num": null, | |
| "type_str": "table" | |
| } | |
| } | |
| } | |
| } |