【摘 要】
：
Neural machine translation(NMT)has achieved great suc-cess under a great deal of bilingual corpora in the past few years.Howev-er,it is much less effective for low-resource language.In order to allevi
【机 构】
：
Institute of Intelligent Machines,Chinese Academy of Science,Hefei,China;University of Science and T
【出 处】
：
第十七届全国计算语言学学术会议暨第六届基于自然标注大数据的自然语言处理国际学术研讨会(CCL 2018)
论文部分内容阅读
Neural machine translation(NMT)has achieved great suc-cess under a great deal of bilingual corpora in the past few years.Howev-er,it is much less effective for low-resource language.In order to alleviate the problem,we present two approaches which can improve the perfor-mance of low-resource NMT system.The rst approach employs the weight sharing of decoder to enhance the target language model of low-resource NMT system.The second approach applies cross-lingual embed-ding and source sentence representation space sharing to strengthen the encoder of low-resource NMT.Our experiments demonstrate that the proposed method can obtain significant improvements on low-resource neural machine translation than baseline system.On the IWSLT2015 Vietnamese-English translation task,our model can improve the trans-lation quality by an average of 1.43 BLEU scores.Besides,we can also get the increase of 0.96 BLEU scores when translating from Mongolian to Chinese.
其他文献
藏文疑问句的句法分析在藏文问答系统、搜索引擎、信息的抽取和检索等领域有着广泛的应用前景.文章通过分析藏文疑问句的构成特点,对藏文疑问句进行了分类,归纳了各类藏文疑问句的结构特征,在此基础上利用PCFG对藏文疑问句进行了句法分析.经测试,在封测试集上的准确率、召回率和F值分别达97.6、97.3和97.4,在开放测试集上的准确率、召回率和F值分别达96.0、95.4和95.7.
跨语言信息检索指用户以一种语言提问,检索出另一种或几种语言描述的信息资源的检索技术,是信息检索领域重要的研究方向之一.近年来,跨语言词向量为跨语言信息处理提供了良好的表示形式,受到很多学者的关注.该文利用跨语言词向量实现从汉文查询词到蒙古文查询词扩展和映射,并利用该文提出的串联式查询扩展、串联式查询扩展过滤、交叉验证过滤三种查询扩展方法在进行词向量映射时对候选的蒙古文查询词进行筛选和排序,选择符合
文本蕴含是自然语言处理的难点,其形式类型复杂、知识难以概括.早期多利用词汇蕴含和逻辑推理知识识别蕴含,但仅对特定类型的蕴含有效.近年来,利用大规模数据训练深度学习模型的方法在句级蕴含关系识别任务上取得优异性能,但模型不可解释,尤其是无法标定引起蕴含的具体语言片段.本文研究文本蕴含成因形式,归纳为词汇、句法异构、常识三类,并以句法异构蕴含为研究对象.针对上述两个问题,提出句法异构蕴含语块的概念,定义
The conventional Chinese word embedding model is similar to the English word embedding model in modeling text,simply uses the Chinese word or character as the minimum processing unit of the text,witho
Word embeddings have recently been widely used to model words in Natural Language Processing(NLP)tasks including semantic similarity measurement.However,word embeddings are not able to cap-ture polyse
Existing sentence alignment methods are founded fundamentally on sentence length and lexical correspondences.Methods based on the former follow in general the length proportionality assumption that th
Metadata extraction for scientific literature is to automati-cally annotate each paper with metadata that represents its most valu-able information,including problem,method and dataset.Most existing w
For different language pairs,word-level neural machine translation(NMT)models with a fixed-size vocabulary suffer from the same problem of representing out-of-vocabulary(OOV)words.The common practice
Existing methods for knowledge graph embedding do not ensure the high-rank triples predicted by themselves to be as consistent as possible with the logical background which is made up of a knowledge g
In e-commerce websites,user-generated question-answering text pairs generally contain rich aspect information of products.In this paper,we address a new task,namely Question-answering(QA)aspect classi