Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678171
Ei Phyu Phyu Mon, Ye Kyaw Thu, Than Than Yu, Aye Wai Oo
Spell checker is a crucial language tool of natural language processing (NLP) and becomes important due to the increase of text-based communication at work, information retrieval, fraud detection, search engines, social media and research areas. In this paper, automatic spelling checking for Burmese is studied by applying Symmetric Delete Spelling Correction Algorithm (SymSpell). We experimented by using an open source SymSpell python library and applied our developing Burmese spelling training corpus together with four frequency dictionaries on ten error types. For the error detection phase, the N-gram language model is used to check our developing spelling training corpus against a dictionary. For the correction phrase, SymSpell is applied to propose candidate corrections within a specified maximum edit distance from the misspelled word. After generating candidates, the best correction in the given context is automatically chosen according to the highest frequency with a minimum edit distance. We investigated the performance of each error type and studied the importance of the dictionary depending on the average term length and maximum edit distance for Burmese spell checker based on SymSpell. Moreover, we observed that syllable level segmentation with a maximum edit distance of 3 gives faster and higher quality results compared with word level segmentation results.
{"title":"SymSpell4Burmese: Symmetric Delete Spelling Correction Algorithm (SymSpell) for Burmese Spelling Checking","authors":"Ei Phyu Phyu Mon, Ye Kyaw Thu, Than Than Yu, Aye Wai Oo","doi":"10.1109/iSAI-NLP54397.2021.9678171","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678171","url":null,"abstract":"Spell checker is a crucial language tool of natural language processing (NLP) and becomes important due to the increase of text-based communication at work, information retrieval, fraud detection, search engines, social media and research areas. In this paper, automatic spelling checking for Burmese is studied by applying Symmetric Delete Spelling Correction Algorithm (SymSpell). We experimented by using an open source SymSpell python library and applied our developing Burmese spelling training corpus together with four frequency dictionaries on ten error types. For the error detection phase, the N-gram language model is used to check our developing spelling training corpus against a dictionary. For the correction phrase, SymSpell is applied to propose candidate corrections within a specified maximum edit distance from the misspelled word. After generating candidates, the best correction in the given context is automatically chosen according to the highest frequency with a minimum edit distance. We investigated the performance of each error type and studied the importance of the dictionary depending on the average term length and maximum edit distance for Burmese spell checker based on SymSpell. Moreover, we observed that syllable level segmentation with a maximum edit distance of 3 gives faster and higher quality results compared with word level segmentation results.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"84 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121224281","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Online channels, e.g., Facebook Messenger and Line, are widely used especially in COVID-19 pandemic. To quickly respond to their customer, chatbot system are implemented in many companies or organizations, connected to those channels. The Office of Registrar, Thammasat University also implements a chatbot to answer questions from students. An important step in the chatbot system is to know an intention of a question message. A bidirectional LSTM model is employed to classify a question message from the chatbot system into five intention classes. The experimental results shows that the obtained model yields an accuracy of 0.80 on our validation dataset.
{"title":"A Bidirectional LSTM Model for Classifying Chatbot Messages","authors":"Nunthawat Lhasiw, Nuttapong Sanglerdsinlapachai, Tanatorn Tanantong","doi":"10.1109/iSAI-NLP54397.2021.9678173","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678173","url":null,"abstract":"Online channels, e.g., Facebook Messenger and Line, are widely used especially in COVID-19 pandemic. To quickly respond to their customer, chatbot system are implemented in many companies or organizations, connected to those channels. The Office of Registrar, Thammasat University also implements a chatbot to answer questions from students. An important step in the chatbot system is to know an intention of a question message. A bidirectional LSTM model is employed to classify a question message from the chatbot system into five intention classes. The experimental results shows that the obtained model yields an accuracy of 0.80 on our validation dataset.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123533040","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678166
Sawittree Jumpathong, Kanyanut Kriengket, P. Boonkwan, T. Supnithi
It takes a long time to build vocabularies and their definitions because they must be approved only by the experts in the meeting of building vocabularies and the definitions are also unstructured. To save time, we applied three techniques of classification to the experiments that are one-class SVMs, isolation forests, and local outlier factors, and also observed how well the method can suggest word definition status via the accuracy. As a result, the local outlier factors obtained the highest accuracy when they used vectors that were produced by USE. They can recognize the boundary of the approved class better and there are several approved clusters and outliers are scattered among them. Also, it is found that the detected status of definitions is both identical and opposite to the reference one. For the patterns of definition writing, the approved definitions are always written in the logical order, and start with wide or general information, then is followed by specific details, examples, and references of English terms or examples. In case of the rejected definitions, they are not always written in the logical order, and their definition patterns are also various - only Thai translation, Thai translation with related entries, parts of speech (POS), Thai translation, related entries, and English term references followed by definitions, etc.
{"title":"Anomaly Detection in Lexical Definitions via One-Class Classification Techniques","authors":"Sawittree Jumpathong, Kanyanut Kriengket, P. Boonkwan, T. Supnithi","doi":"10.1109/iSAI-NLP54397.2021.9678166","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678166","url":null,"abstract":"It takes a long time to build vocabularies and their definitions because they must be approved only by the experts in the meeting of building vocabularies and the definitions are also unstructured. To save time, we applied three techniques of classification to the experiments that are one-class SVMs, isolation forests, and local outlier factors, and also observed how well the method can suggest word definition status via the accuracy. As a result, the local outlier factors obtained the highest accuracy when they used vectors that were produced by USE. They can recognize the boundary of the approved class better and there are several approved clusters and outliers are scattered among them. Also, it is found that the detected status of definitions is both identical and opposite to the reference one. For the patterns of definition writing, the approved definitions are always written in the logical order, and start with wide or general information, then is followed by specific details, examples, and references of English terms or examples. In case of the rejected definitions, they are not always written in the logical order, and their definition patterns are also various - only Thai translation, Thai translation with related entries, parts of speech (POS), Thai translation, related entries, and English term references followed by definitions, etc.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129288123","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678183
Isada Sukprapa, Nguyen Duy Hung, T. Supnithi
Current approaches to text summarization are not genuinely interested in how competent readers perform the task often by re-constructing the arguments in the text then arriving at the summary from conclusions of acceptable arguments. This paper aims to mimic this natural path using formal argumentation techniques. Assuming the availability Argumentative Discourse Unit (ADU) graph of the given text, we build structured argumentation frameworks called S-ASPIC+ and ABA representing the text. Then we use ABA proof procedures to re-construct arguments in the text and evaluate their acceptabilities. Finally, we aggregate the conclusions of acceptable arguments. We demonstrate our approach using a dataset of argumentative micro-texts and report the results, describing comparisons to other methods.
{"title":"Text Summarization using Formal Argumentation","authors":"Isada Sukprapa, Nguyen Duy Hung, T. Supnithi","doi":"10.1109/iSAI-NLP54397.2021.9678183","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678183","url":null,"abstract":"Current approaches to text summarization are not genuinely interested in how competent readers perform the task often by re-constructing the arguments in the text then arriving at the summary from conclusions of acceptable arguments. This paper aims to mimic this natural path using formal argumentation techniques. Assuming the availability Argumentative Discourse Unit (ADU) graph of the given text, we build structured argumentation frameworks called S-ASPIC+ and ABA representing the text. Then we use ABA proof procedures to re-construct arguments in the text and evaluate their acceptabilities. Finally, we aggregate the conclusions of acceptable arguments. We demonstrate our approach using a dataset of argumentative micro-texts and report the results, describing comparisons to other methods.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127740487","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678176
Mako Komatsu, Teruhiko Unoki, M. Shikida
The need for presentation skills is increasing year by year, and presentation lectures are being held in companies and universities. In particular, it is important to communicate interactively with the audience during the presentation. Currently, due to the influence of COVID-19, there are more and more opportunities for presentations in a hybrid with face-to-face and remote audiences. In a hybrid presentation, it is difficult to communicate with both audiences in the same way because there is a difference in awareness between face-to-face and remote audiences due to the influence of presence information. In this paper, we propose a method to support the awareness of remote audiences by sending vibration notifications to the presenter during the presentation in order to promote communication and support the improvement of presentation skills, and confirm the usefulness of the method.
{"title":"Presentation Skills Training System Using Vibration Notification in a HyFlex Workshop","authors":"Mako Komatsu, Teruhiko Unoki, M. Shikida","doi":"10.1109/iSAI-NLP54397.2021.9678176","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678176","url":null,"abstract":"The need for presentation skills is increasing year by year, and presentation lectures are being held in companies and universities. In particular, it is important to communicate interactively with the audience during the presentation. Currently, due to the influence of COVID-19, there are more and more opportunities for presentations in a hybrid with face-to-face and remote audiences. In a hybrid presentation, it is difficult to communicate with both audiences in the same way because there is a difference in awareness between face-to-face and remote audiences due to the influence of presence information. In this paper, we propose a method to support the awareness of remote audiences by sending vibration notifications to the presenter during the presentation in order to promote communication and support the improvement of presentation skills, and confirm the usefulness of the method.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"89 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134109484","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678159
Transformers are the current state-of-the-art type of neural network model for dealing with sequences. Evidently, the most prominent application of these models is in text processing tasks, and the most prominent of these is machine translation. Recently, transformer-based models such as the Edit-Based Transformer with Repositioning (EDITOR) and Levenshtein Transformer (LevT) models have become popular in neural machine translation. To the best of our knowledge, there are no experiments for these two models using under-resourced languages. In this paper, we compared the performance and decoding time of the EDITOR model and the LevT model. We conducted the experiments for under-resourced language pairs, namely, Thai-to-English, Thai-to-Myanmar, English-to-Myanmar, and vice versa. The experimental results showed that the EDITOR model outperforms the LevT model in English-Thai, Thai-English and English-Myanmar language pairs whereas LevT achieves better score than EDITOR in Thai-Myanmar, Myanmar-Thai and Myanmar-English language pairs. Regarding the decoding time, EDITOR model is generally faster than the LevT model in the four language pairs. However, in the case of English-Myanmar and Myanmar-English pairs, the decoding time of EDITOR is slightly slower than the LevT model. At last, we investigated the system level performance of both models by means of compare-mt and word error rate (WER).
{"title":"A Study of Levenshtein Transformer and Editor Transformer Models for Under-Resourced Languages","authors":"","doi":"10.1109/iSAI-NLP54397.2021.9678159","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678159","url":null,"abstract":"Transformers are the current state-of-the-art type of neural network model for dealing with sequences. Evidently, the most prominent application of these models is in text processing tasks, and the most prominent of these is machine translation. Recently, transformer-based models such as the Edit-Based Transformer with Repositioning (EDITOR) and Levenshtein Transformer (LevT) models have become popular in neural machine translation. To the best of our knowledge, there are no experiments for these two models using under-resourced languages. In this paper, we compared the performance and decoding time of the EDITOR model and the LevT model. We conducted the experiments for under-resourced language pairs, namely, Thai-to-English, Thai-to-Myanmar, English-to-Myanmar, and vice versa. The experimental results showed that the EDITOR model outperforms the LevT model in English-Thai, Thai-English and English-Myanmar language pairs whereas LevT achieves better score than EDITOR in Thai-Myanmar, Myanmar-Thai and Myanmar-English language pairs. Regarding the decoding time, EDITOR model is generally faster than the LevT model in the four language pairs. However, in the case of English-Myanmar and Myanmar-English pairs, the decoding time of EDITOR is slightly slower than the LevT model. At last, we investigated the system level performance of both models by means of compare-mt and word error rate (WER).","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133951787","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678185
Phuriphan Prathipasen, Pitisit Dillon, P. Aimmanee, Suree Teerarungsigul, Sasiwimol Nawawitphisit, S. Keerativittayanun, Jessada Karnjana
In analyzing 3D point clouds obtained from a terrestrial LiDAR scanner for rockfall detection, a widely-used clutter removal algorithm is Nearest Neighbor Clutter Removal (NNCR). However, there is a critical problem regarding computational complexity of NNCR. Subsequently, we presented a new algorithm for clutter removal based on grid density as a solution to this problem. Nevertheless, the previously proposed method showed that data points were lost. This study proposes a multi-scale grid-density-based method, assuming that the clutter is normally distributed. Outcomes from the experiment indicate that a proposed method could retrieve data points lost in the previous method. The balanced accuracies, recalls, and F-scores of the proposed method were improved by approximately 13, 33, and 17 percent, respectively, compared with the previously proposed method. Also, the proposed method is about 19 times faster than NNCR.
{"title":"Clutter Removal Algorithm Based on Grid Density with a Recursive Approach for Rockfall Detection in 3D point clouds from a Terrestrial LiDAR Scanner","authors":"Phuriphan Prathipasen, Pitisit Dillon, P. Aimmanee, Suree Teerarungsigul, Sasiwimol Nawawitphisit, S. Keerativittayanun, Jessada Karnjana","doi":"10.1109/iSAI-NLP54397.2021.9678185","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678185","url":null,"abstract":"In analyzing 3D point clouds obtained from a terrestrial LiDAR scanner for rockfall detection, a widely-used clutter removal algorithm is Nearest Neighbor Clutter Removal (NNCR). However, there is a critical problem regarding computational complexity of NNCR. Subsequently, we presented a new algorithm for clutter removal based on grid density as a solution to this problem. Nevertheless, the previously proposed method showed that data points were lost. This study proposes a multi-scale grid-density-based method, assuming that the clutter is normally distributed. Outcomes from the experiment indicate that a proposed method could retrieve data points lost in the previous method. The balanced accuracies, recalls, and F-scores of the proposed method were improved by approximately 13, 33, and 17 percent, respectively, compared with the previously proposed method. Also, the proposed method is about 19 times faster than NNCR.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133625268","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678168
Surawat Pothong, N. Facundes
This paper addresses the application and integration of coreferences resolution tasks in a legislative corpus by using SpanBERT, which is an improvement of the BERT (Bidirectional Encoder Representations from Transformers) model and semantic extraction by Abstract Meaning Representation (AMR) for reducing text complexity, meaning preservation and further applications. Our main processes are divided into four subparts: legal text pre-processing, coreference resolution, AMR, evaluation for meaning preservation, and complexity reduction. Smatch evaluation tool and Bilingual Evaluation Understudy (BLEU) scores are applied to evaluate overlapped meaning between resolved and unresolved coreference sentences. The AMR graphs after complexity have been reduced can be applied for further processing tasks with Neural Network such as legal inferencing and legal engineering tasks.
{"title":"Coreference Resolution and Meaning Representation in a Legislative Corpus","authors":"Surawat Pothong, N. Facundes","doi":"10.1109/iSAI-NLP54397.2021.9678168","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678168","url":null,"abstract":"This paper addresses the application and integration of coreferences resolution tasks in a legislative corpus by using SpanBERT, which is an improvement of the BERT (Bidirectional Encoder Representations from Transformers) model and semantic extraction by Abstract Meaning Representation (AMR) for reducing text complexity, meaning preservation and further applications. Our main processes are divided into four subparts: legal text pre-processing, coreference resolution, AMR, evaluation for meaning preservation, and complexity reduction. Smatch evaluation tool and Bilingual Evaluation Understudy (BLEU) scores are applied to evaluate overlapped meaning between resolved and unresolved coreference sentences. The AMR graphs after complexity have been reduced can be applied for further processing tasks with Neural Network such as legal inferencing and legal engineering tasks.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"68 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130233770","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678157
Bilingual word embeddings (BWEs) represent the lexicons of two different languages in a shared embedding space, which are useful for cross-lingual natural language processing (NLP) tasks. In particular, bilingual word embeddings are extremely useful for machine translation of low-resource languages due to the rare availability of parallel corpus for that languages. Most of the researchers have already learned bilingual word embeddings for high-resource language pairs. To the best of our knowledge, there are no studies on bilingual word embeddings for low resource language pairs, Myanmar-Thai and Myanmar-English. In this paper, we present and evaluate the bilingual word embeddings for Myanmar-Thai, Myanmar-English, Thai-English, and English-Thai language pairs. To train bilingual word embeddings for each language pair, firstly, we used monolingual corpora for constructing monolingual word embeddings. A bilingual dictionary was also utilized to alleviate the problem of learning bilingual mappings as a supervised machine learning task, where a vector space is first learned independently on a monolingual corpus. Then, a linear alignment strategy is used to map the monolingual embeddings to a common bilingual vector space. Either word2vec or fastText model was used to construct monolingual word embeddings. We used bilingual dictionary induction as the intrinsic testbed for evaluating the quality of cross-lingual mappings from our constructed bilingual word embeddings. For all low-resource language pairs, monolingual word2vec embedding models with the CSLS metric achieved the best coverage and accuracy.
{"title":"Supervised Bilingual Word Embeddings for Low-Resource Language Pairs: Myanmar and Thai","authors":"","doi":"10.1109/iSAI-NLP54397.2021.9678157","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678157","url":null,"abstract":"Bilingual word embeddings (BWEs) represent the lexicons of two different languages in a shared embedding space, which are useful for cross-lingual natural language processing (NLP) tasks. In particular, bilingual word embeddings are extremely useful for machine translation of low-resource languages due to the rare availability of parallel corpus for that languages. Most of the researchers have already learned bilingual word embeddings for high-resource language pairs. To the best of our knowledge, there are no studies on bilingual word embeddings for low resource language pairs, Myanmar-Thai and Myanmar-English. In this paper, we present and evaluate the bilingual word embeddings for Myanmar-Thai, Myanmar-English, Thai-English, and English-Thai language pairs. To train bilingual word embeddings for each language pair, firstly, we used monolingual corpora for constructing monolingual word embeddings. A bilingual dictionary was also utilized to alleviate the problem of learning bilingual mappings as a supervised machine learning task, where a vector space is first learned independently on a monolingual corpus. Then, a linear alignment strategy is used to map the monolingual embeddings to a common bilingual vector space. Either word2vec or fastText model was used to construct monolingual word embeddings. We used bilingual dictionary induction as the intrinsic testbed for evaluating the quality of cross-lingual mappings from our constructed bilingual word embeddings. For all low-resource language pairs, monolingual word2vec embedding models with the CSLS metric achieved the best coverage and accuracy.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131338715","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-12-21DOI: 10.1109/iSAI-NLP54397.2021.9678153
Khaing Zar Mon, Reenu, Ye Kyaw Thu
Nowadays, speaker recognition has become one of the important application area of digital signal processing. Speech corpus is important in developing the speech processing and the development of the corpus is essential for low-resourced languages. Burmese (Myanmar language) can be recognized as a low-resourced language because of lack of available resources for speech processing research. Speaker Recognition (SReg) is an approach used to automatically recognize a speaker from their speech utterance. The main concept of SReg is to recognize the information of the speaker’s identity. In SReg, various features have been extracted to reflect the characteristics of the speakers. In this paper, an effective multi-feature combination and comparison of performance between the different size of data are proposed. In this work, weather news from Department of Meteorology and Hydrology, Myanmar is collected. The total size of the implemented Burmese speech corpus is over 10 hours and it contained 13 females and 3 males. The dataset is split into training data and testing data in 4:1 ratio. The experimental results on 16 speakers show that the proposed Burmese speaker recognition based on multi-feature combination achieved 99.16% accuracy and high applicability.
{"title":"Speaker Recognition by Combining Features for Myanmar Weather Forecast Domain","authors":"Khaing Zar Mon, Reenu, Ye Kyaw Thu","doi":"10.1109/iSAI-NLP54397.2021.9678153","DOIUrl":"https://doi.org/10.1109/iSAI-NLP54397.2021.9678153","url":null,"abstract":"Nowadays, speaker recognition has become one of the important application area of digital signal processing. Speech corpus is important in developing the speech processing and the development of the corpus is essential for low-resourced languages. Burmese (Myanmar language) can be recognized as a low-resourced language because of lack of available resources for speech processing research. Speaker Recognition (SReg) is an approach used to automatically recognize a speaker from their speech utterance. The main concept of SReg is to recognize the information of the speaker’s identity. In SReg, various features have been extracted to reflect the characteristics of the speakers. In this paper, an effective multi-feature combination and comparison of performance between the different size of data are proposed. In this work, weather news from Department of Meteorology and Hydrology, Myanmar is collected. The total size of the implemented Burmese speech corpus is over 10 hours and it contained 13 females and 3 males. The dataset is split into training data and testing data in 4:1 ratio. The experimental results on 16 speakers show that the proposed Burmese speaker recognition based on multi-feature combination achieved 99.16% accuracy and high applicability.","PeriodicalId":339826,"journal":{"name":"2021 16th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-12-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132557019","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：