传统蒙古文和西里尔蒙古文分别是在中国和蒙古国使用的蒙古文，它们的口语基本相同，但是书写形式完全不同。结合传统蒙古文和西里尔蒙古文的构词特点，提出了基于联合序列模型的传统蒙古文和西里尔蒙古文相互转换方法，并做了大量的相互转换实验。实验中，传统蒙古文到西里尔蒙古文转换系统的词误识率和字母误识率分别达到了18.38%和6.75%，西里尔蒙古文到传统蒙古文转换系统的词误识率字母误识率分别达到了18.77%和7.14%，基本达到了实用要求。

Traditional Mongolian and Cyrillic Mongolian are both Mongolian languages and are widely used in China and Mongolia respectively. With almost the same pronunciations, their written forms are totally different. According to the characteristic of the two languages, this paper proposes a joint sequence model based approach and depicts in detail the corresponding experiments performed. In the experiments, the word error rate and letter error rate for the traditional Mon-golian to Cyrillic Mongolian conversion system are 18.38% and 6.75%, and that for Cyrillic Mongolian and traditional Mongolian conversion system are 18.77%and 7.14%. Experimental results show that the proposed approach can meet the basic requirements for practical use.

本文重点介绍了印刷体蒙古文的结构特征、印刷体蒙古文的切分方法.特征选择种类繁多,从大量模糊特征中选择出有效特征进行分类,再对印刷体蒙古文的切分方法(行切分、字元切分)进行了详细的说明.

This paper mainly introduces characteristics and segmentation method of printed Mongolian. There is many types of feature classification . Effective features were selected from a large number of fuzzy feature classification, seg-mentation of Printed Mongolian(line segmentation, character segmentation)has carried on the detailed explanation.

目前现代蒙古语语料库的标注虽然有《现代蒙古语语料库标注规范》为指导，但该规范在非蒙古文字符、专有名词、外来词的转写规则方面存在一定空白，非单字单位的标注问题尚未细化。本规范以信息处理用为目的，根据现代蒙古语的特点及规律，研究蒙古文语料库标记单位的合并或切分规则。本标注规范的研究遵循了中国电子技术标准化研究所等单位联合起草的《信息处理用蒙古文词语标记》，及内蒙古大学的《现代蒙古语语料库标注规范》。本研究今后需在大规模语料库基础上不断的完善。

Though the annotation for contemporary Mongolian corpus can de conducted under the guidance of Specifications for Contemporary Mongolian Corpus A nnotation ,there still is a blank for the transliteration of non-Mongolian characters ,proper nouns and loan words and there is no further division of the tagging of word units with more than one character in the specifications .For the purpose of information processing ,the specifi-cations study the rules of merging and segmentation for Mongolian corpus tagging units on the basis of Mongo -lian linguistic features and laws .The study follows the rules in Mongolian Word and Expression Marks for In-formation Processing jointly drafted by China Electronics Standardization Institution and other organizations and Specifications for Contemporary Mongolian Corpus Annotation drawn up by Mongolian University .It is sure that improvements need to make with the development of large -scale corpus in this study .

丝绸之路经济带为丝路沿线各国及地区带来的发展机遇。丝路沿线不同语言信息的交换及处理技术的提升将面临新的机遇与挑战。以蒙古文为实例，研究了丝路带相似多语言信息交叉处理通信问题，研讨了基于词典及语言学知识和基于统计的机器翻译（ SMT）两种方式的蒙文多语种文本的自动转写或翻译方法。试验结果显示，SMT方法对于相似语言转写的效果相比于语言学以及不同语系不同语族语言的翻译效果近提升一倍。

With the voice of developing the Silk Road economic belt ( SREB) , the communication between the multiple languages and their processing along SREB become a practical concern. Focus on Mongolian multi-graphic texts, widely used Mongolia, China and Russia today, in this research, we investigate two kinds of approaches be?tween their converting. One is interested in the linguistic knowledge applied a dictionary, and other is focus on the phrase-based statistical machine translation ( SMT ) . Results from the experiments demonstrate that the approach SMT is more convenient than the linguistic knowledge way. Further, we confirmed that SMT result for the similar language transformation is better than that for the different language family such as Chinese -Mongolian STM repor?ted in previous research.

词对齐研究是多文种信息处理工作的一项不容忽略的基础性研究内容。通过针对中文和蒙古文词对齐研究中存在的形态和词序不对称以及支撑词对齐研究的对齐语料缺乏问题,开展融合语言信息的中蒙混合词对齐模型构建方法研究。利用产生式词对齐结果以及中蒙两种语言的语言信息作为潜特征,建立高质量的融合语言信息的中蒙混合词对齐模型。实验结果证明,该文提出方法对于利用可比语料抽取对齐语料是可行的。

Generative model and discriminative model are taken as two mainly approaches in word alignment field. To solve the problems of Chinese-Mongolian word alignment such as asymmetric morphological structure and word order, is presented in a discriminative framework using the result of generative word alignment and Chinese and Mongolian contextual information as la-tent variables for Chinese-Mongolian word alignment task. This work provides a new method and some theoretical support for performance improvement of word alignment of smaller scale corpus.