学术会议｜第一届数据科学与二语研究国际研讨会

Scott Crossley

Title: Semantic embedding approaches as state-of-the-art in second language studies

Abstract: Semantic embedding approaches commonly used in natural language processing such as transformer models have rarely been used to examine second language (L2) learning. Importantly, the performance of semantic embedding approaches has not been contrasted with more traditional annotation approaches to assessing L2 learning. This study used NLP techniques related to lexical annotations and semantic embedding approaches to model the receptive vocabulary of L2 learners based on their lexical production during a writing task. The goal of the study is to examine the strengths and weaknesses of both approaches in modeling and understanding L2 lexical knowledge. Findings indicate that semantic word-embedding approaches based on transformer models outperform linguistic annotations and Word2vec semantic embedding approaches in the task of predicting L2 learners’ vocabulary scores. The findings help to support the strength and accuracy of transformer models in L2 studies as well as their generalizability across tasks when compared to linguistic feature models and other semantic embedding approaches. Limitations to transformer approaches, especially explainability, are discussed.

Biography: Scott Crossley is a professor of special education at Vanderbilt University. His research focuses on using data science and artificial intelligence to explore cognition, with an emphasis on natural language processing. He is specifically interested in language and literacy acquisition.

Luke Plonsky

Title: Questioning quantitative methods in L2 research: A look behind the curtain

Abstract: For better or for worse, it wasn’t until fairly recently that we as a field began to reflect on and systematically examine our methodological practices (see Gass et al., 2021). This work has led to real progress on fronts ranging from sampling and design to data analysis, researcher training, and open science (e.g., Marsden & Plonsky, 2018; Norouzian, 2020; Sudina, 2021; Veríssimo, 2021). But a wide range of problems persist. In this talk, I discuss several major methodological challenges we face as well as several advances—the “methodological turn”, as Heidi Byrnes (2013) put it—that has been taking place in recent years. In particular, I’ll address the need for stronger validity arguments behind our instruments, and I will encourage a healthy skepticism toward quantitative approaches which (still) dominate in the field. I will also explain why, despite these concerns and reservations, I am generally (if cautiously) optimistic about the future of research methods in the applied language sciences.

Biography: Luke Plonsky (PhD, Michigan State) is Associate Professor of Applied Linguistics at Northern Arizona University, where he teaches courses in second language acquisition and research methods. His work in these and other areas has resulted in over 100 articles, book chapters, and books. Luke is Editor of Studies in Second Language Acquisition, Managing Editor of Foreign Language Annals, Co-Editor of De Gruyter Mouton's Series on Language Acquisition, and Co-Director of the IRIS Database (iris-database.org). In addition to prior appointments at Georgetown and University College London, Luke has lectured in China, Japan, The Netherlands, Poland, Spain, and Puerto Rico.

Clarence Green

Title: Data science and second language vocabulary: coverage, lexical profiling, question modeling.

Abstract: Data science has made some of the most important contributions the study of second language vocabulary over the past decades. In many ways, corpus linguistics has been exploring the benefits of ‘big’ data science since before big data came to prominence across such a wide range of fields. Text-based data mining has helped us understand the vocabulary demands of multiple subject areas and educational contexts, and has helped us model learning and acquisition. There is probably no other way in which complex lexical profiles could have been developed without data science, telling us the vocabulary that students need to master for proficient comprehension. Data science in second language vocabulary research has produced resources such as high quality word lists, and linked vocabulary size research to text comprehension estimates, giving the field a goal for general second language proficiency, e.g. around 10, 000 word families. This talk reviews how corpus research is a unique type of data mining that provides unparalleled insights  for our understanding of vocabulary across the curriculum. The talk reports on a model built from a corpus representing Extensive Reading/ER and Extensive Viewing/EV that simulates incidental vocabulary learning potential. Incidental learning makes a robust contribution to vocabulary, whereby vocabulary develops as a by-product of meaning-focused tasks, e.g. reading comprehension. ER/EV involve students self-selected pleasure reading/viewing and a large amount of comprehensible input over prolonged periods (Renandya & Jacobs, 2016). Over the past 10-15 years, a research stream has developed in corpus linguistics modelling how much (and what) vocabulary ER/EV might be able to contribute. The presentation illustrates both the value of ER/EV and also the contribution that corpus linguistics can make to ER/EV research.   

Biography: Dr Clarence Green is a Senior Lecturer in the School of Education of Federation University Australia. He holds a PhD in linguistics (University of Melbourne) and a Master of Applied Linguistics TESOL. He has taught and published widely in corpus linguistics, first and second language acquisition, literacy, psycholinguistics, and English grammar. He currently coordinates the Master of Specialist Teaching, Master of Education and Speech Pathology courses in the School of Education, and is on the editorial board of the Journal of English for Academic Purposes and the Asia Pacific Journal of Education.

陈莹

标题：巴基斯坦和越南学习者汉语声调和语调产出的声学研究

摘要：巴基斯坦的官方语言乌尔都语是语调语言，词层面上没有声调；越南语是声调语言，但是其调型和调值异于汉语的四声。本研究招募了通过HSK四级考试的中等汉语水平的巴基斯坦学习者和越南学习者各20名参与了两项汉语普通话语音产出实验。实验一比较两组二语学习者和汉语母语者在单双音节词、三四音节短语以及五六音节句子中产出的声调，实验二增加了陈述和疑问两种句子类型，比较二语学习者和汉语母语者在句子层面的语调产出。本研究对时长、音高、调域和基频斜率四种声学参数进行分析，同时分析了二声和三声的拐点出现时间和拐点与起点之间的基频差。混合效应模型的结果显示，两组学习者都未能产出具有统计学意义的接近汉语母语者的声调和语调，且母语为语调语言的巴基斯坦学习者产出的汉语声调和语调特征，不同于母语为声调语言的越南学习者。本研究将基于多个二语语音习得理论模型讨论这些声学分析结果。

个人简介：陈莹，南京理工大学外国语学院教授、博士生导师，外国语言文学学科带头人，语言信息智能处理及应用工信部重点实验室主任，美国俄勒冈大学语言学系博士，美国罗切斯特大学语言学系和脑认知科学系访问学者，研究方向为实验语音学、认知心理语言学、语言信息处理。近年来主持国家自然科学基金面上项目、国家社会科学基金一般项目、教育部人文社科基金规划项目以及相关横向课题多项，发表SCI、SSCI、EI、CSSCI收录论文30余篇，担任10余个SCI、SSCI期刊及国际学术会议审稿人，研究成果获江苏省哲学社会科学优秀成果二等奖，获评江苏省高校青蓝工程中青年学术带头人、江苏省333高层次人才工程中青年学术技术带头人、福建省闽江学者讲座教授。

蔡金亭

标题：使用产出语料及有声思维数据研究双外语使用中的语言迁移

摘要：外语习得与使用中的语言迁移受到了普遍重视，但其研究方法尚未达成一致意见。多语产出中迁移研究的比较——归纳方法框架（蔡金亭2022）提出，可以依据学习中的产出语料及有声思维数据分析多语使用中的语言层迁移和概念迁移。为了进一步展示该方法框架，笔者以时间和空间域为对象，通过分析中国汉-英-日三语者的汉、英、日作文语料及其有声思维数据，先判断学习者英、日语个人层面迁移的发生情况，然后再归纳出组群层面迁移的出现概率。希望本发言能进一步激发学界对语言迁移研究中数据及方法的深入讨论。

个人简介：蔡金亭，上海财经大学外国语学院教授，外国语言学及应用语言学方向学术带头人。兼任集刊《第二语言学习研究》主编，外研社应用语言学核心话题系列丛书语言习得系列主编，中国二语习得专业委员会和教育语言学专业委员会副会长，西安交通大学等校的兼职教授。主要研究方向为第二语言习得。已发表论文60多篇；出版《语篇因素对英语过渡语中使用一般过去时的影响》《中国学生英语过渡语研究》《语言迁移研究》等专著7部；论著3次获得省社科优秀成果奖。主持完成国家社科基金项目“中国大学生英语产出中的母语迁移历时研究”和教育部人文社科项目“母语迁移与语内因素对中国学生使用英语动词论元构式的影响研究”；目前主持国家社科重点项目“双外语学习中语言迁移的多维动态研究”。

陈 飞

标题：跨模态视听发音训练的有效性验证——基于气流模型的二语语音教学

摘要：跨模态视听发音教学是计算机辅助二语语音训练的有效方法之一。本研究针对普通话辅音发音中送气音的发音难点，开发了基于3D虚拟说话人的气流模型，来进行普通话送气音的二语发音训练。眼动实验结果表明：在线学习过程中，非汉语母语者也能够分配足够的注意力在非母语视觉表征的气流模型上。面向日语母语者的发音训练结果表明，基于视听跨模态的气流模型训练效果要远远好于听觉单模态发音训练，并且这种视听跨模态训练优势还可以进一步扩展到改变承载声调或者改变后接韵母的新词汇之中。

个人简介：陈飞，湖南大学外国语学院教授、博士生导师、博士后合作导师，湖南大学岳麓学者。中国英汉语比较研究会语音教学研究专业委员会理事，美国语言学协会（Linguistic Society of America）会员，国际自闭症研究协会（International Society for Autism Research）会员。在Journal of Child Language、Autism Research、Journal of Autism and Developmental Disorders、《语言科学》等国内外期刊发表SSCI/SCI/CSSCI/EI检索的论文30余篇。主要研究方向为一语及二语习得、病理语言学、神经语言学。

丁红卫

标题：语音声学数据分析及其在二语习得等研究领域中的应用

摘要：以语音的方式进行交流是人类最直接、最高效、最频繁的沟通方式。语音信号当中不仅包含了文字语义所表达的语言信息，而且还承载着说话人情感态度所传达的副语言信息，并反映出其年龄性别健康状况等非语言信息。人们对语音的理解远不止于文字内容的“言内之意”，还有更多反映心理生理状态的“弦外之音”。这使得语音数据不仅可以用于语言分析，也可以用于情感计算，还可以用于疾病诊断与司法鉴定。但是在所有的语音数据研究中，语言相关的研究一直居于主导地位。

报告首先介绍语音声学数据分析在自然科学及社会科学中的各类应用。接着将聚焦语音数据在跨语言中的比较，特别是在二语语音产出中的音段与超音段的相关研究。

个人简介：丁红卫，上海交通大学外国语学院教授、博导。获德国语言学博士与工学教授资格。国家社科重大项目“精神障碍人群语料库建设及面向脑科学和人工智能的语言研究”首席专家，国际言语韵律协会理事会Officer、中国语言学会语音学分会副主任委员，担任Springer出版社两部系列丛书主编与SCI/SSCI期刊审稿人，发表国际论文100余篇。于德国电子信息学院从事语音技术研究近20年中，主持或参与十多项德国或欧盟纵向项目及国际著名公司横向项目，包括多语种语音合成系统建设与计算机辅助语音学习系统研发等。回国后主要从事二语语音习得研究，听力及精神障碍人群语音韵律的认知神经科学探究。并继续与国际著名公司合作研发语音智能及言语健康方面的技术产品。

龚箭

标题：汉语母语者英语嘶擦音/s, ʃ/产出的声学分析

摘要：嘶擦音是语言中常见的一类辅音，这类辅音在发音时舌头拉长，将气流引导至齿尖处，形成湍流，带有“嘶嘶”的噪声。在英语中，有两对以清浊作为区别特征的嘶擦音，分别为/s, z/和/ʃ, ʒ/。在汉语中，则有三个清嘶擦音/s, ʂ, ɕ/。本文对汉语母语者所产出的英语清嘶擦音/s, ʃ/和汉语清嘶擦音/s , ʂ/进行了声学分析，通过对这些辅音的频谱重心（COG）进行测量和对比，发现汉语母语者产出的英语/s/的频谱重心与英语母语者非常接近，而在英语/ʃ/的产出上，汉语母语者则直接迁移了汉语/ʂ/的发音，显示汉语母语者对于英语/s/的习得优于对英语/ʃ/的习得。此外，汉语母语者在汉语/s, ʂ/的产出上体现出了显著的性别区别，而在英语/s, ʃ/的产出上则没有显著的性别区别。这一结果表明嘶声产出上的性别差别可能与具体语言相关，受到具体语言音系结构的影响。

关键词：嘶擦音，二语语音产出，声学分析，性别差别

金檀

Title: Human-machine teaming up for text simplification: What should pre-service teachers learn?

Abstract: While a large body of natural language processing (NLP) studies have provided valuable insights into the development of automated text simplification (ATS), understanding of how teachers utilize ATS to prepare teaching materials for differentiated instruction (DI) remains limited. In this connection, this presentation reports on a project that examines the practice of pre-service teachers in the preparation of teaching materials for DI, focusing on the knowledge and skills needed to perform ATS. A qualitative case study, consisting of three stages, is undertaken to investigate the types of knowledge and skills pre-service teachers require to effectively perform ATS. First, semi-structured interviews are conducted to elicit pre-service teachers’ perceptions toward the knowledge and skills required. Second, a training workshop tailored to pre-service teachers’ reported needs toward ATS is carried out. Third, post-hoc online queries and written reflections collected from pre-service teachers are analyzed to gain additional insights into their learning needs. Overall, the study provides robust evidence supporting the argument that pre-service teachers can learn and benefit from ATS, and highlights certain knowledge and skills that pre-service teachers need to acquire when performing ATS. The findings also provide implications that suggest how pre-service teachers can further develop their corpus literacy in text simplification through human-machine collaboration.

Biography: Tan Jin is Professor in the School of Foreign Studies at South China Normal University. His research interests include corpus linguistics, language testing and assessment as well as computer assisted language learning. He has published extensively in journals such as Language Learning & Technology, Language Teaching Research, Language Testing, Modern Language Journal and TESOL Quarterly.

金檀，华南师范大学外国语言文化学院教授、博士生导师，研究方向为语料库语言学、语言测试与评估及计算机辅助语言学习。代表成果有“分层阅读文本智能改编研究”“自动写作反馈合作加工研究”及“课外学习资源混合设计研究”。创建语言数据网（LanguageData.net），主持研制“阅读分级指难针”“写作表达范文库”及“外语研究方法课”。

雷蕾

Title: A large-scale longitudinal study of syntactic complexity development in EFL writing: A mixed-effects model approach

Abstract: This study examined syntactic complexity with a large-scale corpus of EFL (English as a foreign language) compositions written by the same group of 1081 intermediate Chinese university students over a time span of four consecutive semesters in two academic years. The compositions were analyzed for the longitudinal trajectories of syntactic complexification over time and for proficiency level difference, and genre differences. It was found that length-based measures, measures of coordination, and those of phrasal sophistication significantly increased while measures of clausal subordination significantly decreased from the first to the fourth semesters. Interestingly, students’ writing style was found to become more formal and exhibited more prominent features of academic prose over time. Furthermore, students’ proficiency level and genre showed significant main effects on most of the syntactic complexity measures. Our findings complemented existing research on writing development patterns with large-scale and longitudinal evidence. Some possible explanations in relation to these findings were discussed. In addition, implications for teachers and researchers of EFL writing were offered.

Biography: Lei Lei is Professor in Applied Linguistics at the Institute of Corpus Studies and Applications, Shanghai International Studies University. His research interests include academic English, second language writing, digital humanities in linguistics. He has published extensively in journals such as Applied Linguistics, Language Teaching, International Journal of Corpus Linguistics, and Journal of English for Academic Purposes.

雷蕾，博士，上海外国语大学教授、博士生导师。研究兴趣涉及语料库语言学、学术英语、计量语言学等领域。在剑桥大学出版社等出版专著5部，在Applied Linguistics、Language Teaching、International Journal of Corpus Linguistics、Journal of English for Academic Purposes等SSCI期刊发表研究性论文近40篇、书评10余篇，其中两篇论文入选ESI高被引论文；在CSSCI期刊发表论文或书评10余篇。主持完成国家社科基金等项目多项。兼任Journal of English for Academic Purposes (SSCI) 等国内外期刊编委。

卢春明

标题：自然语言交流的认知神经机制

摘要：语言不仅与个体的认知活动密切相关，而且在社会互动中发挥着重要的交流功能。经过数万年的演化，语言交流与人脑也建立了密切关联。但是，由于理论和方法技术的限制，语言交流的认知神经机制却并不清楚。在过去十多年里，我们聚焦语言交流，先后发展了基于近红外光谱成像的超扫描技术和高生态效度的自然语言交流研究范式，围绕语言加工与交流的认知神经机制开展了系统研究，研究了语言交流的认知层级结构及其神经基础，探讨了语言交流中的计算推理机制，以及基于语言交流的课堂教学机制。

个人简介：卢春明，北京师范大学认知神经科学与学习国家重点实验室教授，IDG/麦戈文脑科学研究院副院长，课题组长，博士生导师，主持国家自然科学基金项目多项。研究方向为语言交流和社会互动，在Nature Communications、PNAS等期刊发表论文50余篇。目前担任国际近红外光学成像协会通讯理事、中国神经科学学会认知神经生物学分会秘书长、中国神经科学学会科普与继续教育工作委员会委员、中国教育学会脑科学与教育分会理事等。

王敏

Title: Effect of Interactive Intensity on Lexical Alignment and L2 Writing Quality

Abstract: This study examined the relationships among interactive intensity, lexical alignment and L2 writing quality under two variations (multi-turn vs. single-turn) of a story continuation task. The multi-turn version entailed more interaction between the learner and the text than the single-turn version. One-hundred and nine Chinese undergraduates were assigned to either the multi- or the single-turn group. The continuation task was preceded by a pretest one week earlier and followed by a posttest two weeks later. Both versions of the continuation task resulted in lexical alignment but the magnitude was stronger in the multi-turn mode. Compared with the pretest, participants’ use of content and function words was more similar to the input when completing the story and during the posttest, and increases in the similarity of content words were more marked in the multi-turn group. Furthermore, the effect of alignment on writing quality was influenced by interactive intensity: the multi-turn group improved more remarkably from the pretest to the posttest than the single-turn group. By employing the algorithmic indices Latent Semantic Similarity (LSS) and Language Style Matching (LSM), this study was the first to examine L2 lexical alignment beyond exact repetition of content words and the first to associate interactive intensity with L2 writing quality.

Biography: Wang Min is Professor in Applied Linguistics at the Institute of Applied Linguistics, Zhejiang University. Her research interests include L2 interactive alignment, structural priming and L2 construction learning and the Xu-argument. She has published in journals such as Modern Language Journal, Applied Linguistics, Journal of Corpus Linguistics and Linguistic Theory, and Journal of English for Specific Purposes.

王敏，浙江大学外语学院教授、博士生导师，中国英汉语比较研究会第二语言习得研究专业委员会常务理事。主要研究兴趣包括：基于使用理论视域下的语言习得研究、二语互动协同研究、续论、学术英语等。曾在Applied Linguistics、Modern Language Journal、English for Specific Purposes、System、Corpus Linguistics & Linguistic Theory,《外语教学与研究》、《现代外语》等国内外知名语言学期刊发表论文。主持国家社科基金等项目多项，出版专著2部。曾获教育部高校人文社科优秀研究成果二等奖1项、国家优秀教学成果二等奖1项。

吴燕京

标题：情绪效价如何影响二语加工和双语控制

摘要：情绪效价是语言表达和词汇理解中的重要组成部分。以往的双语研究主要关注情绪效价本身的加工过程。针对情绪效价作为变量如何影响双语者语言加工和控制（如，语言转换）方面的研究很少。本研究从语言产出和语言理解两个加工层面，采用行为学和功能核磁共振（fMRI）相结合的方法，探究了情绪效价影响二语加工和语言转换的认知过程以及其神经机制，提出了负面启动效应（negative priming effect）假设，为完善双语语言加工与控制的理论提供了实证研究的基础。 

个人简介：吴燕京，宁波大学外国语学院教授，中国社会科学院合作博士生导师。毕业于英国威尔士班戈大学心理学专业。曾任英国谢菲尔德大学任职终身讲师，浙江省千人计划特聘教授。研究方向为神经语言学的理论，方法和应用。发表SCI／SSCI收录论文30余篇，包括《自然•人类行为》，《美国科学院院刊》等国籍著名学术期刊。曾主持英国科学院（British Academy）人文社科项目2项，其中1项获优秀科研成果奖。现主持1项国家社科基金重点课题。是2021年爱思唯尔中国高被引学者。曾获浙江省哲学社会科学成果奖二等奖。

徐晓东

标题：母语和外语语境下的违实推理研究

摘要：违实句（counterfactuals）是指能够传递与事实相反意义的语句。一般认为违实句的理解需要构建双重心理表征，这一过程会受到句法、语义、语用、语境等多种因素的影响。本文采用脑电和行为测试手段考察了前提知识的真假以及外语水平的高低如何影响违实推理。本研究发现，在母语环境下，当前提为真时，违实推理可以顺利展开；而当前提为假时，违实推理遇阻而无法正常进行。在外语环境下，前提真假对违实推理的影响受被试外语水平的影响。对高水平中国外语学习者来说，当前提命题为真时，违实推理能够顺利展开，而当前提命题为假时，违实推理无法正常进行；对低水平外语学习者来说，无论前提真假与否，违实推理均无法正常展开。这表明，无论是母语还是外语环境，违实推理均会在某种程度上受制于前提命题的真假。

个人简介： 徐晓东，南京师范大学外国语学院教授、博导。目前主要采用眼动、脑电和脑成像等心理学和神经科学手段，探讨母语及二语加工中，句法、语义、语用以及个体认知特质差异对语言理解的影响。在Second Language Research, Brain and Language, Journal of Experimental Psychology: LMC, 外语教学与研究等期刊发表论文40余篇。获评江苏省“333工程”中青年学术带头人，省“社科优青”（2019）等；主持国家自然科学基金、国家社会科学基金各1项；获得江苏省哲学社会科学优秀成果奖三等奖2项、江苏省教育科学研究成果二等奖1项。

于翠红

标题：认知翻译视角下的方向性效应研究

摘要：本研究基于E-prime实验设计，考察两组不同英语水平和不同口译经验中国大学生（本科生和已经通过口译2级的学生）在执行四项视译任务（方向性：正向口译与反向口译，句子复杂度：低复杂度和高复杂的）时的反应时和正确率，从认知翻译视角探讨中国学生在进行英-汉与汉-英视译过程中的方向性效应及其与口译经验和文本复杂性的相关性。结果表明，前向口译比后向口译对认知要求更高，两水平在低复杂度文本翻译上比较接近，但在高复杂度文本翻译上差异更加显著。研究同时证实了口译员优势假设，即口译经验在不同复杂性文本翻译中发挥重要调节作用，缺乏口译经验的学习者在与语言转换、注意力资源分配和监控相关的执行功能中花费更多的认知努力。但通过口译经验和实践培养的特定任务认知技能对于提高视译员的执行能力具有正向作用。

个人简介：于翠红，博士、教授，研究方向为认知加工、二语习得。联系方式：13863899826 yucuihong2009@126.com

张晓鹏

Title: Modeling Effects of Linguistic Complexity on Second Language Processing Effort

Abstract: The evaluation of text difficulty is core in language teaching practice. However, little research has been done to explore what makes texts difficult for second language (L2) learners. This talk will report on a study that examined how linguistic complexity features contributed to text processing effort during L2 reading. The processing effort was operationalized as L2 learners’ mean fixation duration, regression rates, skipping rates, and mean saccade amplitude, since these measures have been found to reflect the overall text difficulty. The eligible linguistic complexity indices were regressed against these eye-movement measures based on 272 English texts. Results indicate that the linguistic complexity indices incorporated in the one-component partial least square regression model could explain approximately 26.2% of the variance in the four eye-movement measures and outperformed the oft-used readability formulas in predicting L2 processing efforts. These findings communicate that the model obtained in this study can be used to predict processing effort involved in L2 text reading and potentially provide useful implications for developing systems to assess text difficulty for L2 learners.

Keywords: linguistic complexity; text difficulty; processing effort; eye-movement; English-L2 learners

个人简介：张晓鹏，博士，教授，博士生导师，西安交通大学青拔学者A岗，中国二语习得研究专业委员会理事，广东外语外贸大学外国语言学及应用语言学研究中心“续”论研究室兼职研究员。近年来，主持国家社科基金项目2项，省部级课题3项，在Applied Linguistics, Language Learning, The Modern Language Journal,《外语教学与研究》《现代外语》等SSCI/CSSCI语言学期刊上发表论文三十余篇，出版专著1部。