ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)最新文献

{"title":"Unsupervised Neural Machine Translation for Similar and Distant Language Pairs","authors":"Haipeng Sun, Rui Wang, M. Utiyama, Benjamin Marie, Kehai Chen, E. Sumita, T. Zhao","doi":"10.1145/3418059","DOIUrl":"https://doi.org/10.1145/3418059","url":null,"abstract":"Unsupervised neural machine translation (UNMT) has achieved remarkable results for several language pairs, such as French–English and German–English. Most previous studies have focused on modeling UNMT systems; few studies have investigated the effect of UNMT on specific languages. In this article, we first empirically investigate UNMT for four diverse language pairs (French/German/Chinese/Japanese–English). We confirm that the performance of UNMT in translation tasks for similar language pairs (French/German–English) is dramatically better than for distant language pairs (Chinese/Japanese–English). We empirically show that the lack of shared words and different word orderings are the main reasons that lead UNMT to underperform in Chinese/Japanese–English. Based on these findings, we propose several methods, including artificial shared words and pre-ordering, to improve the performance of UNMT for distant language pairs. Moreover, we propose a simple general method to improve translation performance for all these four language pairs. The existing UNMT model can generate a translation of a reasonable quality after a few training epochs owing to a denoising mechanism and shared latent representations. However, learning shared latent representations restricts the performance of translation in both directions, particularly for distant language pairs, while denoising dramatically delays convergence by continuously modifying the training data. To avoid these problems, we propose a simple, yet effective and efficient, approach that (like UNMT) relies solely on monolingual corpora: pseudo-data-based unsupervised neural machine translation. Experimental results for these four language pairs show that our proposed methods significantly outperform UNMT baselines.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124897737","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}

{"title":"A Cascaded Unsupervised Model for PoS Tagging","authors":"Necva Bölücü, Burcu Can","doi":"10.1145/3447759","DOIUrl":"https://doi.org/10.1145/3447759","url":null,"abstract":"Part of speech (PoS) tagging is one of the fundamental syntactic tasks in Natural Language Processing, as it assigns a syntactic category to each word within a given sentence or context (such as noun, verb, adjective, etc.). Those syntactic categories could be used to further analyze the sentence-level syntax (e.g., dependency parsing) and thereby extract the meaning of the sentence (e.g., semantic parsing). Various methods have been proposed for learning PoS tags in an unsupervised setting without using any annotated corpora. One of the widely used methods for the tagging problem is log-linear models. Initialization of the parameters in a log-linear model is very crucial for the inference. Different initialization techniques have been used so far. In this work, we present a log-linear model for PoS tagging that uses another fully unsupervised Bayesian model to initialize the parameters of the model in a cascaded framework. Therefore, we transfer some knowledge between two different unsupervised models to leverage the PoS tagging results, where a log-linear model benefits from a Bayesian model’s expertise. We present results for Turkish as a morphologically rich language and for English as a comparably morphologically poor language in a fully unsupervised framework. The results show that our framework outperforms other unsupervised models proposed for PoS tagging.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121561747","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}

{"title":"Special Issue on Deep Structured Learning for Natural Language Processing","authors":"Gunasekaran Manogaran, H. Qudrat-Ullah, Qin Xin","doi":"10.1145/3436206","DOIUrl":"https://doi.org/10.1145/3436206","url":null,"abstract":"The landscape of deep structured learning has changed in recent years. Many new trends and techniques have entered the arena. This editorial on “Deep Structured Learning for Natural Language Processing” brings out the innovative applications of deep structured learning for various dimensions of natural language processing (NLP). To this point, this editorial gives the researcher and practitioner’s orientation on recent developments in NLP and provides an opportunity for relevant and qualitative research. This special section provides research articles that were selected following a rigorous review process focusing on various applications of deep learning for NLP. Overall, the special section contains 5 articles. Specifically, the first article. From a practical perspective, today’s crime rate increases exponentially, leading to numerous serious threats in society. Due to this research, interest in cyberbullying has acquired increased research attention in recent days. Forefronts in machine learning and NLP are the two most important cutting-edge techniques that reduce the risk of cyberbullying in contemporary modern society. A new detection algorithm based on fast text detection and word similarity is introduced. It further analyzes the cyberbullying characteristics from the lexical as well as the syntactic viewpoints. From the experimental observation, it is envisioned that detection accuracy and recall measures are improved compared to the existing schemes. A scheme called “deep interactive memory network for aspect-level sentiment analysis” is explored to effectively study the subjective information. It makes use of deep learning techniques to extract opinions from the massively large datasets. Imbalanced Twitter sentiment analysis is the systematic issue deals with class imbalances and its effects. A research article is proposed to analyze the emotions in a big corpus of imbalanced tweets using deep learning techniques. Another important research area is big data analytics and NLP applications. Against this background, research articles emphasis on transnational happiness study on big data using artificial intelligence and deep learning techniques. Next, knowledge discovery plays a vital role when dealing with big data. Effective ways of retrieving knowledge from news texts are analyzed in depth using NLP and deep learning methods. Finally, citation recommendation in massively generated big data is often a crucial task. This process is made easier with model-enhanced graph modeling techniques. This approach reduces the complexity rate and assists in effective citation process. This special issue emphasis on various aspects NLP using deep learning techniques. The articles published in this special section contribute to the application of NLP in various domains, which includes NLP and deep learning for big data analytics, recommendation, and linguistic analysis. We thank the authors for submitting good quality research contributions reflecting ","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"364 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124572938","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}

{"title":"Real-time Assistive Reader Pen for Arabic Language","authors":"Mohammad A. Alzubaidi, Mwaffaq Otoom, Nouran S. Ahmad","doi":"10.1145/3423133","DOIUrl":"https://doi.org/10.1145/3423133","url":null,"abstract":"Disability is an impairment affecting an individual's livelihood and independence. Assistive technology enables the disabled cohort of the community to break the barriers to learning, access information, contribute to the community, and live independently. This article proposes an assistive device to enable people with visual disabilities and learning disabilities to access printed Arabic material in real-time, and to help them participate in the education system and the professional workforce. This proposed assistive device employs Optical Character Recognition (OCR) and Text To Speech (TTS) conversion, using concatenation synthesis. OCR is achieved using image processing, character extraction, and classification, while Arabic speech synthesis is achieved through concatenation synthesis, followed by Multi Band Re-synthesis Overlap-Add (MBROLA). Waveform generation in the second phase produces vocal output for the disabled user to hear. OCR character and word accuracy tests were conducted for nine Arabic fonts. The results show that six fonts were recognized with over 60% character accuracy and two fonts were recognized with over 88% accuracy. A Mean Opinion Score (MOS) test for speech quality was conducted. The results showed an overall MOS score of 3.53/5 and indicated that users were able to understand the speech. A real-time usability testing was conducted with 10 subjects. The results showed an overall average of agreements scores of 3.9/5 and indicated that the proposed Arabic reader pen meets the real-time constraints and is pleasant and satisfying to use and can contribute to make printed Arabic material accessible to visually impaired persons and people with learning disabilities.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129954843","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}

{"title":"Optimisation of the Largest Annotated Tibetan Corpus Combining Rule-based, Memory-based, and Deep-learning Methods","authors":"M. Meelen, Élie Roux, N. Hill","doi":"10.1145/3409488","DOIUrl":"https://doi.org/10.1145/3409488","url":null,"abstract":"This article presents a pipeline that converts collections of Tibetan documents in plain text or XML into a fully segmented and POS-tagged corpus. We apply the pipeline to the large extent collection of the Buddhist Digital Resource Center. The semi-supervised methods presented here not only result in a new and improved version of the largest annotated Tibetan corpus to date, the integration of rule-based, memory-based, and neural-network methods also serves as a good example of how to overcome challenges of under-researched languages. The end-to-end accuracy of our entire automatic pipeline of 91.99% is high enough to make the resulting corpus a useful resource for both linguists and scholars of Tibetan studies.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133802812","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}

{"title":"Improving Semantic Coherence of Gujarati Text Topic Model Using Inflectional Forms Reduction and Single-letter Words Removal","authors":"Uttam Chauhan, Apurva Shah","doi":"10.1145/3447760","DOIUrl":"https://doi.org/10.1145/3447760","url":null,"abstract":"A topic model is one of the best stochastic models for summarizing an extensive collection of text. It has accomplished an inordinate achievement in text analysis as well as text summarization. It can be employed to the set of documents that are represented as a bag-of-words, without considering grammar and order of the words. We modeled the topics for Gujarati news articles corpus. As the Gujarati language has a diverse morphological structure and inflectionally rich, Gujarati text processing finds more complexity. The size of the vocabulary plays an important role in the inference process and quality of topics. As the vocabulary size increases, the inference process becomes slower and topic semantic coherence decreases. If the vocabulary size is diminished, then the topic inference process can be accelerated. It may also improve the quality of topics. In this work, the list of suffixes has been prepared that encounters too frequently with words in Gujarati text. The inflectional forms have been reduced to the root words concerning the suffixes in the list. Moreover, Gujarati single-letter words have been eliminated for faster inference and better quality of topics. Experimentally, it has been proved that if inflectional forms are reduced to their root words, then vocabulary length is shrunk to a significant extent. It also caused the topic formation process quicker. Moreover, the inflectional forms reduction and single-letter word removal enhanced the interpretability of topics. The interpretability of topics has been assessed on semantic coherence, word length, and topic size. The experimental results showed improvements in the topical semantic coherence score. Also, the topic size grew notably as the number of tokens assigned to the topics increased.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131855393","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}

{"title":"A Survey of Offensive Language Detection for the Arabic Language","authors":"Fatemah Husain, Özlem Uzuner","doi":"10.1145/3421504","DOIUrl":"https://doi.org/10.1145/3421504","url":null,"abstract":"The use of offensive language in user-generated content is a serious problem that needs to be addressed with the latest technology. The field of Natural Language Processing (NLP) can support the automatic detection of offensive language. In this survey, we review previous NLP studies that cover Arabic offensive language detection. This survey investigates the state-of-the-art in offensive language detection for the Arabic language, providing a structured overview of previous approaches, including core techniques, tools, resources, methods, and main features used. This work also discusses the limitations and gaps of the previous studies. Findings from this survey emphasize the importance of investing further effort in detecting Arabic offensive language, including the development of benchmark resources and the invention of novel preprocessing and feature extraction techniques.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127684716","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}

{"title":"Hybridization between Neural Computing and Nature-Inspired Algorithms for a Sentence Similarity Model Based on the Attention Mechanism","authors":"P. Zhang, Xingzhe Huang, Maozhen Li, Yu Xue","doi":"10.1145/3447756","DOIUrl":"https://doi.org/10.1145/3447756","url":null,"abstract":"Sentence similarity analysis has been applied in many fields, such as machine translation, the question answering system, and voice customer service. As a basic task of natural language processing, sentence similarity analysis plays an important role in many fields. The task of sentence similarity analysis is to establish a sentence similarity scoring model through multi-features. In previous work, researchers proposed a variety of models to deal with the calculation of sentence similarity. But these models do not consider the association information of sentence pairs, but only input sentence pairs into the model. In this article, we propose a sentence feature extraction model based on multi-feature attention. In addition, with the development of deep learning and the application of nature-inspired algorithms, researchers have proposed various hybrid algorithms that combine nature-inspired algorithms with neural networks. The hybrid algorithms not only solve the problem of decision-making based on multiple features but also improve the performance of the model. In the model, we use the attention mechanism to extract sentence features and assign weight. Then, the convolutional neural network is used to reduce the dimension of the matrix. In the training process, we integrate the firefly algorithm in the neural networks. The experimental results show that the accuracy of our model is 74.21%.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"91 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117171774","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}

{"title":"Hate Speech Detection in Roman Urdu","authors":"Muhammad Moin Khan, K. Shahzad, M. K. Malik","doi":"10.1145/3414524","DOIUrl":"https://doi.org/10.1145/3414524","url":null,"abstract":"Hate speech is a specific type of controversial content that is widely legislated as a crime that must be identified and blocked. However, due to the sheer volume and velocity of the Twitter data stream, hate speech detection cannot be performed manually. To address this issue, several studies have been conducted for hate speech detection in European languages, whereas little attention has been paid to low-resource South Asian languages, making the social media vulnerable for millions of users. In particular, to the best of our knowledge, no study has been conducted for hate speech detection in Roman Urdu text, which is widely used in the sub-continent. In this study, we have scrapped more than 90,000 tweets and manually parsed them to identify 5,000 Roman Urdu tweets. Subsequently, we have employed an iterative approach to develop guidelines and used them for generating the Hate Speech Roman Urdu 2020 corpus. The tweets in the this corpus are classified at three levels: Neutral-Hostile, Simple-Complex, and Offensive-Hate speech. As another contribution, we have used five supervised learning techniques, including a deep learning technique, to evaluate and compare their effectiveness for hate speech detection. The results show that Logistic Regression outperformed all other techniques, including deep learning techniques for the two levels of classification, by achieved an F1 score of 0.906 for distinguishing between Neutral-Hostile tweets, and 0.756 for distinguishing between Offensive-Hate speech tweets.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"6 5","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120912139","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}

{"title":"Multimodal News Feed Evaluation System with Deep Reinforcement Learning Approaches","authors":"S. Rakesh Kumar, S. Muthuramalingam, F. Al-turjman","doi":"10.1145/3414523","DOIUrl":"https://doi.org/10.1145/3414523","url":null,"abstract":"Multilingual and multimodal data analysis is the emerging news feed evaluation system. News feed analysis and evaluations are interrelated processes, which are useful in understanding the news factors. The news feed evaluation system can be implemented for single or multilingual language models. Classification techniques used on multilingual news analysis require deep layered learning techniques rather than conventional approaches. In this proposed work, a hierarchical structure of deep learning algorithms is implemented for making an effective complex news evaluation system. Deep learning techniques such as the Deep Cooperative Multilingual Reinforcement Learning Model, the Multidimensional Genetic Algorithm, and the Multilingual Generative Adversarial Network are developed to evaluate a vast number of news feeds. The proposed tech-niques collaborate in a pipeline order to build a deep news feed evaluation system. The implementation details project that the newly proposed system performs 5% to 12% better than the other news evaluation systems.","PeriodicalId":144107,"journal":{"name":"ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-03-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116628400","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}