自然语言脚本生成动画脚本的关键技术研究
|
摘要
文景转换系统,主要分成三个模块:抽取自然语言脚本模块、自然语言脚本生成动画脚本模块、动画生成模块。本文是该项目从自然语言到动画的中间过渡模块。
本文的研究任务是从自然语言脚本生成动画脚本,主要分成二个步骤:场景识别和自然语言脚本到动画脚本的映射。
目前,场景识别多用于视频、图像、机器人及语音等领域,其中图像分类领域技术比较成熟,文本场景识别技术尚未形成完整的理论知识。本文在文本场景分类领域进行初步探索,提出了一种基于SVM的场景识别方法,并辅以简单的规则,验证了场景识别问题是以段落而非句子为基本单位的问题。同时,针对《一千零一夜》中部份语料实现了场景识别模块,并嵌入到自然语言脚本生成脚本系统中,为生成动画场景奠定基础。
另一方面,本文将脚本理解自然语言故事的方法应用于文景转换任务中,验证复合动作分解对自然语言脚本生成动画脚本的可行性。从而为动画模块生成必要的动画脚本序列。
本文的两个重要研究内容如下:
1.场景识别。主要研究内容:面向段落的场景识别,具体包括:场景类别划分、语料标注规则制定、初始语料加工、数据稀疏问题处理、后处理规则制定以及实验结果的评价。详细分析这些研究内容并给出具体算法和实验数据,结合不同的数据稀疏处理方法和后处理规则,分别对以句子为基本单位和以段落为基本单位场景识别实验比较。最后利用LIBSVM不同核函数,在最好的实验方法上,进行比较实验。
2.自然语言脚本到动画脚本的映射。主要研究内容:针对任务定义元动作、分解复合动作、构造实体间的等级关系及将自然语言脚本映射生成动画脚本,如:主语可以是事件的施事,也可以是动画脚本是的角色或开始位置等,从而验证该方法的可行性。
This research comes from the Nature Science Foundation "the 3D visualization of spatial relationships in text based on ontology" (Text to Scene, TTS for short). TTS system, mainly divided into three modules: extract natural language script module, natural language script based animation script generation module and animation production module. This article is from the natural language of the item to the middle of the transition animation module.
Our mission is natural language script based animation script generation, mainly including two parts: scene identification, natural language script and animation script mapping.
At present, the scene identification for video, images, robots and voice in which is stronger. But there isn't a complete theory for text scene identification. A method based on SVM tool has been explored. We also proved some rules greatly. Then we get known that a paragraph is the unit of scene identification. At the same time, some articles of the "Thousand and One Nights" provides text resource for scene identification module.
We also proved that natural language script to explore text meanings which can be translated to animation script feasibility.
The content of this paper is divided into the following two aspects:
1. Scene identification. Main elements: the scene identification for paragraphs. We got type of scene, made rules on identify scene for sentences and processed the corpus and sparse. Then we put some rules on the result which comes from models trained by SVM tool. Different methods on data sparse and rules explored more unsimilar results. With different LIBSVM kernel functions, different results can be abtained.
2. Natural language scrip based animation script generation. Main elements: we define unit actions, decompose compound actions , construct entity structure, and map natural language script to animaion script have been constructed. All that verify the feasibility of the method.
本文的研究任务是从自然语言脚本生成动画脚本,主要分成二个步骤:场景识别和自然语言脚本到动画脚本的映射。
目前,场景识别多用于视频、图像、机器人及语音等领域,其中图像分类领域技术比较成熟,文本场景识别技术尚未形成完整的理论知识。本文在文本场景分类领域进行初步探索,提出了一种基于SVM的场景识别方法,并辅以简单的规则,验证了场景识别问题是以段落而非句子为基本单位的问题。同时,针对《一千零一夜》中部份语料实现了场景识别模块,并嵌入到自然语言脚本生成脚本系统中,为生成动画场景奠定基础。
另一方面,本文将脚本理解自然语言故事的方法应用于文景转换任务中,验证复合动作分解对自然语言脚本生成动画脚本的可行性。从而为动画模块生成必要的动画脚本序列。
本文的两个重要研究内容如下:
1.场景识别。主要研究内容:面向段落的场景识别,具体包括:场景类别划分、语料标注规则制定、初始语料加工、数据稀疏问题处理、后处理规则制定以及实验结果的评价。详细分析这些研究内容并给出具体算法和实验数据,结合不同的数据稀疏处理方法和后处理规则,分别对以句子为基本单位和以段落为基本单位场景识别实验比较。最后利用LIBSVM不同核函数,在最好的实验方法上,进行比较实验。
2.自然语言脚本到动画脚本的映射。主要研究内容:针对任务定义元动作、分解复合动作、构造实体间的等级关系及将自然语言脚本映射生成动画脚本,如:主语可以是事件的施事,也可以是动画脚本是的角色或开始位置等,从而验证该方法的可行性。
This research comes from the Nature Science Foundation "the 3D visualization of spatial relationships in text based on ontology" (Text to Scene, TTS for short). TTS system, mainly divided into three modules: extract natural language script module, natural language script based animation script generation module and animation production module. This article is from the natural language of the item to the middle of the transition animation module.
Our mission is natural language script based animation script generation, mainly including two parts: scene identification, natural language script and animation script mapping.
At present, the scene identification for video, images, robots and voice in which is stronger. But there isn't a complete theory for text scene identification. A method based on SVM tool has been explored. We also proved some rules greatly. Then we get known that a paragraph is the unit of scene identification. At the same time, some articles of the "Thousand and One Nights" provides text resource for scene identification module.
We also proved that natural language script to explore text meanings which can be translated to animation script feasibility.
The content of this paper is divided into the following two aspects:
1. Scene identification. Main elements: the scene identification for paragraphs. We got type of scene, made rules on identify scene for sentences and processed the corpus and sparse. Then we put some rules on the result which comes from models trained by SVM tool. Different methods on data sparse and rules explored more unsimilar results. With different LIBSVM kernel functions, different results can be abtained.
2. Natural language scrip based animation script generation. Main elements: we define unit actions, decompose compound actions , construct entity structure, and map natural language script to animaion script have been constructed. All that verify the feasibility of the method.
引文
1 李晗静.基于自然语言描述的空间概念建模研究.哈尔滨工业大学博士学位论文.2007:1-13
2 R.C.Schank,C.K.Riesbeck.Inside Computer Understanding:Five Programs Plus Miniatures.Hillsdale.Linguistic Society of America.1982:494-495
3 R.C.Schank.Conceptual Information Processing.North-Holland Publishing Company,1975:96-102
4 姚天顺,朱靖波等.自然语言理解--一种让机器懂得人类语言的研究(第二版).清华大学出版社.2002:98-121
5 R.C.Schank,R.P.Abelson.Scripts Plans,Goals and Understanding:An Inquiry into Human Knowledge Structures Hillsdale.NJ:Erlbaum.1977:36-45
6 X.G.Zhang.Introduction to Statistical Learning Theory and Support Vector Machines.Acta Automatica Sinica.2000,26(1):32-42
7 蔡莉华,郑鹏.用熵阈值方法进行视频场景突变探测.计算机工程与应用.2003,39(10):113-115
8 代六玲,黄河燕等.中文文本分类中特征抽取方法的比较研究.中文信息学报.2003,18(1):26-32
9 周茜,赵明生等.中文文本分类中的特征选择研究.中文信息学报.2003,18(3):17-23
10 孙丽华,张积东等.一种改进的KNN方法及其在文本分类中的应用.应用科技.2002,29(2):25-27
11 赵世奇等.基于类别特征域的文本分类特征选择方法.中文信息学报.2005,9(6):21-27
12 W.L.Chen,X.Z.Chang et al.Automatic Word Clustering for Text Categorization Using Global Information.Asia Information Retrieval Symposium.2004:1-6
13 Q.Wang,X.L.Wang.A Study of Semi-Discrete Matrix Decomposition for LSI in Automated Text Categorization.First International Joint Conference on Natural Language Processing.2003:302-309
14 N.I.Badler,B.Webber and J.Kalita.Animation from instructions.Make Them Move,SanMateo,CA:Morgan Kaufmann,1991:51-93
15 P.Nugues,O.Bersot.A Conversational Agent to Help Navigation and Collaboration in Virtual Worlds.Virtual Reality.1998,3(1):71-82
16 G.Adorni,M.D.Manzo and F.Giunchigliad.Natural Language Driven Image Generation.Proceedings of COLING,Stanford,California.1984:495-500
17 S.R.Clay,J.Wilhelms.Put:Language-based Interactive Manipulation of Objects.IEEE Computer Graphics and Applications.1996,(3):31-39
18 R.Johansson,D.Williams and A.Berglund.Carsim:A System to Visualize Written Road Accident Reports as Animated 3D Scenes.ACL.2004:57-64
19 B.Coyne,R.Sproat.WordsEye:An Automatic Text-to-Scene Conversion System.Proceedings of the SIGGRAPH 2001 Annual Conference on Computer Graphics.2001:487-496
20 R.C.Schank,R.P.Abelson.SAM--A Story Understander.Yale AI Project Research Report.No.43,1975
21 汝钤,张松懋.从故事到动画片--全过程计算机辅助动画自动生成.自动化学报.2002,28(3):322-348
22 M.Szummer,R W.Picard.Indoor-outdoor image classification.In Proc.of IEEE Int.Workshop on Content-Based Access of Image and Video Database.1998:42-51
23 A.K.Jain,A.Vailaya and H.J.Zhang.On Image Classification:City vs.landscape.In IEEE Workshop on Content-Based Access of Image and Video Libraries,Santa Barbara,CA,1998:3-8
24 R.J,L.Guo and Y.T.Shen.Applying Multi-class SVMs into Scene Image Classification.Proceedings of the 17th International Conference on Innovations in Aapplied Artificial Intelligence.2004:24-34
25 孙志杰,许宏丽.一种图像低层视觉特征到高层语义的映射方法.计算机应用,2004,24(12):22-24
26 王艳妮等.一种基于语义的图像数据分类系统.计算机应用研究.2004,21(4):256-261
27 任建峰等.基于多类神经网络机的自然图像分类.西北工业大学学报.2005,23(3):295-298
28 J.Li,J.Z.Wang.Automatic Linguistic Indexing of Pictures by a Statistical Modeling Approach.IEEE Transactions on Pattern Analysis and Machine Intelligence.2003,25(9):1075-1088
29 R.Sproat.Inferring the Environment in A Text-to-Scene Conversion System.International Conference On Knowledge Capture archive,2001:147-154
30 V.Vapnik.The Nature of Statistical Learning Theory.New York:Springer,2000
31 http://www.csie.ntu.edu.tw/～cjlin/
32 Q.Luo.Study on Radial Basis Function Networks Based Reinforcement Learning in Robot Soccer.Journal of System Simulation.2002,14(8):1095-1096
33 董振东,董强.知网[EB/OL].http://www.keenage.com
34 Q.Liu,S.J.Li.Calculation of Semantic Similariby of Vocabulary Based on "Hownet".Computational Linguistics and Chinese Language Processing,2002,7(2):59-76
35 C.W.Hsu,C.J.Lin.A Comparison of Methods for Multi-class Support Vector Machines.IEEE Trans on Neural Networks,2002,13(2):415-425
36 Y.H.Li,A.K.Jain.Classification of Text Document.Computer Journal,1998,41(8):537-546
37 D.D.Lewis.An Evaluation of Phrasal and Clustered Representations on a Text Categorization Task.Proceedings of 15 ACM International Conference on Research and Development in Information Retrieva.1992:37-50
38 Y.Yang,J.O.Pedersen.A Comparative Study on Feature Selection in Text Categorization.Proc of the 14th Int'l Conf on Machine Learning(ICML' 97).San Francisco.Morgan Kaufmann,1997
39 赵妍妍等.基于多特征融合的句子相似度计算.全国第八届计算语言学联合学术会议.南京,2005:168-174
40 N.Chatterjee.A Statistical Approach for Similarity Measurement Between Sentences for EBMT.1999
41 周法国,杨炳儒.一种新改进的句子相似度计算方法.第七届中文信息处理国际会议论文集.2007:133-136
42 J.Thorston.A Probabilistic Analysis of the Rocchio Algorithm with TF/IDF for Text Categorization.Proc of 14th Int'l Conf on Machine Learing(ICML'97).1997:143-151
43 黄萱菁,石崎洋之.独立于语种的文本分类方法.中文信息学报.2000, 14(6):1-7
44 J.H.Holland.Concerning Efficient Adaptive Systems.In Yovits,M.C.,Self-Organizing Systems,1962:215-230
45 HollandJH.Adaptation in Natural and Artificial System.The Univ of Michigan Press,1975
46 李理等.基于遗传算法的多实体空间优化摆放与场景建模.哈尔滨工业大学硕士论文.2006:27-28
2 R.C.Schank,C.K.Riesbeck.Inside Computer Understanding:Five Programs Plus Miniatures.Hillsdale.Linguistic Society of America.1982:494-495
3 R.C.Schank.Conceptual Information Processing.North-Holland Publishing Company,1975:96-102
4 姚天顺,朱靖波等.自然语言理解--一种让机器懂得人类语言的研究(第二版).清华大学出版社.2002:98-121
5 R.C.Schank,R.P.Abelson.Scripts Plans,Goals and Understanding:An Inquiry into Human Knowledge Structures Hillsdale.NJ:Erlbaum.1977:36-45
6 X.G.Zhang.Introduction to Statistical Learning Theory and Support Vector Machines.Acta Automatica Sinica.2000,26(1):32-42
7 蔡莉华,郑鹏.用熵阈值方法进行视频场景突变探测.计算机工程与应用.2003,39(10):113-115
8 代六玲,黄河燕等.中文文本分类中特征抽取方法的比较研究.中文信息学报.2003,18(1):26-32
9 周茜,赵明生等.中文文本分类中的特征选择研究.中文信息学报.2003,18(3):17-23
10 孙丽华,张积东等.一种改进的KNN方法及其在文本分类中的应用.应用科技.2002,29(2):25-27
11 赵世奇等.基于类别特征域的文本分类特征选择方法.中文信息学报.2005,9(6):21-27
12 W.L.Chen,X.Z.Chang et al.Automatic Word Clustering for Text Categorization Using Global Information.Asia Information Retrieval Symposium.2004:1-6
13 Q.Wang,X.L.Wang.A Study of Semi-Discrete Matrix Decomposition for LSI in Automated Text Categorization.First International Joint Conference on Natural Language Processing.2003:302-309
14 N.I.Badler,B.Webber and J.Kalita.Animation from instructions.Make Them Move,SanMateo,CA:Morgan Kaufmann,1991:51-93
15 P.Nugues,O.Bersot.A Conversational Agent to Help Navigation and Collaboration in Virtual Worlds.Virtual Reality.1998,3(1):71-82
16 G.Adorni,M.D.Manzo and F.Giunchigliad.Natural Language Driven Image Generation.Proceedings of COLING,Stanford,California.1984:495-500
17 S.R.Clay,J.Wilhelms.Put:Language-based Interactive Manipulation of Objects.IEEE Computer Graphics and Applications.1996,(3):31-39
18 R.Johansson,D.Williams and A.Berglund.Carsim:A System to Visualize Written Road Accident Reports as Animated 3D Scenes.ACL.2004:57-64
19 B.Coyne,R.Sproat.WordsEye:An Automatic Text-to-Scene Conversion System.Proceedings of the SIGGRAPH 2001 Annual Conference on Computer Graphics.2001:487-496
20 R.C.Schank,R.P.Abelson.SAM--A Story Understander.Yale AI Project Research Report.No.43,1975
21 汝钤,张松懋.从故事到动画片--全过程计算机辅助动画自动生成.自动化学报.2002,28(3):322-348
22 M.Szummer,R W.Picard.Indoor-outdoor image classification.In Proc.of IEEE Int.Workshop on Content-Based Access of Image and Video Database.1998:42-51
23 A.K.Jain,A.Vailaya and H.J.Zhang.On Image Classification:City vs.landscape.In IEEE Workshop on Content-Based Access of Image and Video Libraries,Santa Barbara,CA,1998:3-8
24 R.J,L.Guo and Y.T.Shen.Applying Multi-class SVMs into Scene Image Classification.Proceedings of the 17th International Conference on Innovations in Aapplied Artificial Intelligence.2004:24-34
25 孙志杰,许宏丽.一种图像低层视觉特征到高层语义的映射方法.计算机应用,2004,24(12):22-24
26 王艳妮等.一种基于语义的图像数据分类系统.计算机应用研究.2004,21(4):256-261
27 任建峰等.基于多类神经网络机的自然图像分类.西北工业大学学报.2005,23(3):295-298
28 J.Li,J.Z.Wang.Automatic Linguistic Indexing of Pictures by a Statistical Modeling Approach.IEEE Transactions on Pattern Analysis and Machine Intelligence.2003,25(9):1075-1088
29 R.Sproat.Inferring the Environment in A Text-to-Scene Conversion System.International Conference On Knowledge Capture archive,2001:147-154
30 V.Vapnik.The Nature of Statistical Learning Theory.New York:Springer,2000
31 http://www.csie.ntu.edu.tw/～cjlin/
32 Q.Luo.Study on Radial Basis Function Networks Based Reinforcement Learning in Robot Soccer.Journal of System Simulation.2002,14(8):1095-1096
33 董振东,董强.知网[EB/OL].http://www.keenage.com
34 Q.Liu,S.J.Li.Calculation of Semantic Similariby of Vocabulary Based on "Hownet".Computational Linguistics and Chinese Language Processing,2002,7(2):59-76
35 C.W.Hsu,C.J.Lin.A Comparison of Methods for Multi-class Support Vector Machines.IEEE Trans on Neural Networks,2002,13(2):415-425
36 Y.H.Li,A.K.Jain.Classification of Text Document.Computer Journal,1998,41(8):537-546
37 D.D.Lewis.An Evaluation of Phrasal and Clustered Representations on a Text Categorization Task.Proceedings of 15 ACM International Conference on Research and Development in Information Retrieva.1992:37-50
38 Y.Yang,J.O.Pedersen.A Comparative Study on Feature Selection in Text Categorization.Proc of the 14th Int'l Conf on Machine Learning(ICML' 97).San Francisco.Morgan Kaufmann,1997
39 赵妍妍等.基于多特征融合的句子相似度计算.全国第八届计算语言学联合学术会议.南京,2005:168-174
40 N.Chatterjee.A Statistical Approach for Similarity Measurement Between Sentences for EBMT.1999
41 周法国,杨炳儒.一种新改进的句子相似度计算方法.第七届中文信息处理国际会议论文集.2007:133-136
42 J.Thorston.A Probabilistic Analysis of the Rocchio Algorithm with TF/IDF for Text Categorization.Proc of 14th Int'l Conf on Machine Learing(ICML'97).1997:143-151
43 黄萱菁,石崎洋之.独立于语种的文本分类方法.中文信息学报.2000, 14(6):1-7
44 J.H.Holland.Concerning Efficient Adaptive Systems.In Yovits,M.C.,Self-Organizing Systems,1962:215-230
45 HollandJH.Adaptation in Natural and Artificial System.The Univ of Michigan Press,1975
46 李理等.基于遗传算法的多实体空间优化摆放与场景建模.哈尔滨工业大学硕士论文.2006:27-28