Despite mounting evidence that multilinguality can be easily weaponized against language models (LMs), works across NLP Security remain overwhelmingly English-centric. In terms of securing LMs, the NLP norm of "English first" collides with standard procedure in cybersecurity, whereby practitioners are expected to anticipate and prepare for worst-case outcomes. To mitigate worst-case outcomes in NLP Security, researchers must be willing to engage with the weakest links in LM security: lower-resourced languages. Accordingly, this work examines the security of LMs for lower- and medium-resourced languages. We extend existing adversarial attacks for up to 70 languages to evaluate the security of monolingual and multilingual LMs for these languages. Through our analysis, we find that monolingual models are often too small in total number of parameters to ensure sound security, and that while multilinguality is helpful, it does not always guarantee improved security either. Ultimately, these findings highlight important considerations for more secure deployment of LMs, for communities of lower-resourced languages.
Bhaasha, Bhasa, Zaban: A Survey for Low-Resourced Languages in South Asia -- Current Stage and Challenges
Dept of Computer Science & Engineering, West Bengal University of Technology(计算机科学与工程系,西孟加拉理工大学);Department of Computer Science, University of Memphis(计算机科学系,孟菲斯大学)
Participatory Research for Low-resourced Machine Translation: A Case Study in African Languages
Masakhane;University of Pretoria(比勒陀利亚大学);Niger-Volta LTI;African Masters in Machine Intelligence(非洲机器智能硕士项目);Federal University of Technology, Akure(阿库雷联邦理工大学);University of Johannesburg(约翰内斯堡大学);Bayero University, Kano(卡诺贝罗大学);Jomo Kenyatta University of Agriculture and Technology(乔莫·肯雅塔农业技术大学);UESTC(电子科技大学);Siseng Consulting Ltd(Siseng咨询有限公司);African Leadership University(非洲领导力大学);InstaDeep Ltd(InstaDeep有限公司);Sapienza University of Rome(罗马第一大学);Udacity(优达城);Parliament, Republic of South Africa(南非共和国议会);Explore Data Science Academy(Explore数据科学学院);UCD, Konta;AI4D
Attentive fine-tuning of Transformers for Translation of low-resourced languages @LoResMT 2021
Indian Institute of Information Technology Tiruchirappalli(印度特里奇拉帕利信息技术学院);ULTRA Arts and Science College(ULTRA文理学院);Sri Sivasubramaniya Nadar College of Engineering(斯里·西瓦苏布拉马尼亚·纳达尔工程学院);Kongu Engineering College(孔古工程学院);Sultan Idris Education University(苏丹依德里斯教育大学);National University of Ireland Galway(爱尔兰国立大学戈尔韦分校)
本文报告了IIITT团队为英语-马拉地语和英语-爱尔兰语语言对LoResMT 2021共享任务提交的机器翻译(MT)系统。该任务的重点是为爱尔兰语和马拉地语等资源相对较低的语言提供出色的翻译。我们微调IndicTrans,这是一个针对英语->马拉地语的预训练多语言NMT模型,使用外部并行语料库作为额外训练的输入。我们对后一种语言对使用了预先训练的Helsinki-NLP Opus MT English->Irish模型。我们的方法在BLEU指标上产生了相对有希望的结果。在团队名称IIITT下,我们的系统分别在英语->马拉地语,爱尔兰语->英语和英语->爱尔兰语中排名1,1和2。
Native vs Non-Native Language Prompting: A Comparative Analysis
Qatar Computing Research Institute(卡塔尔计算研究所);Qatar University(卡塔尔大学);Qatar University of Doha for Science and Technology(多哈科学技术卡塔尔大学);Liverpool John Moores University(利物浦约翰摩尔大学);Carnegie Mellon University in Qatar(卡塔尔卡内基梅隆大学)
Corpora deduplication or duplication in Natural Language Processing of few resourced languages ? A case of study: The Mexico's Nahuatl
LIFAT, Université de Tours(图尔大学计算机科学、人工智能与图像处理实验室);Instituto de Investigaciones en Educación, Universidad Veracruzana(韦拉克鲁斯大学教育研究所);Trading Central(Trading Central公司);Independent Researcher(独立研究员)
BenLLMEval: A Comprehensive Evaluation into the Potentials and Pitfalls of Large Language Models on Bengali NLP
Islamic University of Technology(伊斯兰理工大学);York University(约克大学);University of Alberta(阿尔伯塔大学);Nanyang Technological University(南洋理工大学);Dialpad Inc.(Dialpad公司)
DeepHateExplainer: Explainable Hate Speech Detection in Under-resourced Bengali Language
Fraunhofer FIT(弗劳恩霍夫应用信息技术研究所);RWTH Aachen University(亚琛工业大学);Noakhali Science and Technology University(诺亚克利科技大学);Begum Rokeya University(贝古姆·罗凯亚大学);The University of Alabama(阿拉巴马大学);National University of Ireland, Galway(爱尔兰国立高威大学);Islamic University of Technology(伊斯兰科技大学)
Steklov Mathematical Institute at St. Petersburg(圣彼得堡斯捷克洛夫数学研究所);St. Petersburg State University(圣彼得堡国立大学);Kazan (Volga Region) Federal University(喀山(伏尔加地区)联邦大学);Kyrgyz State Technical University n. a. I. Razzakov(吉尔吉斯斯坦伊·拉扎科夫国立技术大学);The Cramer Project(克拉默项目);Kyrgyz AI Research Institute(吉尔吉斯斯坦人工智能研究院)
Classification Benchmarks for Under-resourced Bengali Language based on Multichannel Convolutional-LSTM Network
Fraunhofer FIT(弗劳恩霍夫应用信息技术研究所);RWTH Aachen University(亚琛工业大学);Insight SFI Research Centre for Data Analytics(Insight SFI数据分析研究中心);Data Science Institute(数据科学研究所);National University of Ireland Galway(爱尔兰国立大学戈尔韦分校);Vrije Universiteit Amsterdam(阿姆斯特丹自由大学)
From Automation to Collaboration: Human-in-the-Loop Methods for Safe and Trustworthy NLP
Department of Computer Science and Engineering, BRAC University(布拉克大学计算机科学与工程系);Department of Electrical and Electronic Engineering, Rajshahi University of Engineering and Technology(拉贾沙希工程与技术大学电子与电气工程系)
Human-in-the-Loop Generation of Adversarial Texts: A Case Study on Tibetan Script
Institute of National Security, Minzu University of China(中国民族大学国家安全学院);School of Information Science and Technology, Xizang University(西藏大学信息科学与技术学院);National Language Resource Monitoring & Research Center | Minority Languages Branch(国家语言资源监测与研究中心 | 少数民族语言分支);The State Key Laboratory of Tibetan Intelligence(藏语智能 state key laboratory)
CreoleVal: Multilingual Multitask Benchmarks for Creoles
Aalborg University(奥尔堡大学);KU Leuven(鲁汶大学);National Institute of Information and Communications Technology(日本信息通信研究机构);University of Copenhagen(哥本哈根大学);University of Electronic Science and Technology of China(电子科技大学);Meta;Zamboanga State College of Marine Sciences and Technology(三宝颜加国立海洋科学与技术学院);University of Newcastle(纽卡斯尔大学);University of Surrey(萨里大学);Kean University(肯恩大学);Linköping University(林雪平大学);Université Paris Nanterre(巴黎第十大学);Massachusetts Institute of Technology(麻省理工学院)
An assessment of orthographic similarity measures for several African languages
University of Cape Town(开普敦大学)
众所周知,自然语言界面和工具,如拼写检查和以自己的语言进行的网络搜索,在以信通技术为媒介的通信中是有用的。然而,南部非洲的大多数语言资源不足。因此,如果通用的和少数特定于语言的NLP工具都可以重用或轻松地跨语言调整,这将是非常有用的。这取决于语言之间相似性的概念和程度。我们从正字法和语料库的角度来评估这一点。12个版本的《世界人权宣言》(UDHR)进行了检查,显示集群的语言,从而或多或少地适合跨语言适应的NLP工具,这不符合格思里区。为了检验这些结果的普遍性,我们放大isiZulu定量和定性与其他四个语料库和不同类型的文本。结果表明,《世界人权宣言》是一个典型的文本文件的正字法。研究结果还提供了对词汇多样性和体裁等典型指标可用性的深入了解,以及相同的统计数据在不同的文档中可能意味着不同的东西。虽然NLTK for Python可以用于文本的基本分析,但它和类似的NLP工具需要大量的定制。