C2mala2janani 111110022452-phpapp02
0 likes244 views
The document describes an animated story visualizer system for Tamil text. The system takes Tamil stories as input, pre-processes the text to extract features like named entities, emotions, environments and temporal constraints. These features are stored in a database along with animated movie clips. When a new input text is received, it is compared to the database to retrieve matching movie clips. These clips are assembled into an animation based on the extracted features and timelines. The system allows users to automatically generate animations from Tamil stories with minimal effort.
![Animated Story Visualizer for Tamil Text
M. Janani, Dr. T. Mala
Department of Information Science and Technology
Anna University, Chennai, India
janani_mrl@yahoo.com, malanehru@annauniv.edu
Abstract
Natural language is a straightforward and efficient medium for describing visual facts and mental
images. The System uses a novel approach to generate animation from Tamil texts such as stories.
Tamil text is pre-processed and the necessary features like named entities, environmental constraints,
temporal and emotion constraints for the given stories are extracted and placed in the database. The
system automatically generates a query based on the users input and compare it with features stored
in the database. Finally animation is dynamically generated using an external motion synthesis system.
Using this system, even greenhorn users can generate animation quickly and easily by giving the
Tamil text.
Keywords Computer animation, Natural Language Processing, Pre-processing, Feature Extraction,
Motion synthesis
I. INTRODUCTION
These days, animations are widely used in many applications, such as cartoons, web graphics, games,
and so on. Computer animation is one of the best methods for depicting the dynamic content. A
medium is necessary for the animation to be created in a convenient and natural manner. It should be
possible to describe the scenes directly from natural language.NLP is an easy and effective way to
analyze, understand and generate languages that humans use naturally.
The aim of this work is to generate an animation from Tamil texts such as movie scripts or stories.
Training input text is given to the pre-processing module. Here tokenization is performed and the
tokens are given to the morphological analyser which is used to convert the tokens into a POS tags.
Information related to named entities, temporal constraints, emotion and environment inference
features are extracted. A query is generated automatically from the input text which contains
information for the search process and compares it with the information already stored in the
database. Finally motion synthesis generates an animation. The interactions between characters are
handled by this module based on the information provided in the database.
II. RELATED WORK
Generating animation from natural language texts has been a challenge. WordsEye developed by
Coyne and Sproat [1] converts natural language texts to a scene. WordsEye focuses on generating a
still image, when a character motion is specified in a given text, the system simply prefer to pose for
118](https://image.slidesharecdn.com/c2mala2janani-111110022452-phpapp02-121115053403-phpapp01/85/C2mala2janani-111110022452-phpapp02-2-320.jpg)
![the action generated from the database. The Carsim system [2] describes a new version of text-to-scene
converter that handles texts describing car accidents using computer program and it is visualized in
the 3D environment. Storytelling System [3] illustrates a system called Interactive e-Hon, which
provides storytelling in the form of animation and conversation translated from original text. A
Constraint based scene conversion system [4] describes a Text2Scene conversion method which
automatically converts text into 3D scenes. A large database of 3D models is used by this method to
depict entities and actions.
III. SYSTEM OVERVIEW
In this section, overview of our system (Figure 1) is given, where the major components are identified.
When the Tamil text is given to the system, Tamil text is pre-processed and the information are
extracted and stored in the database along with the objects created. When an input text is given to the
system it automatically generates the query from the input text and compares it with the information
stored in the database. An animation is then generated using an external motion synthesis system.
Figure 1. SYSTEM OVERVIEW
PRE-PROCESSING
When the Tamil text is given to the system, natural language processes (Tokenization and
morphological analysis) are applied first.
Tokenization
The first step in NLP is to identify tokens, which decomposes the delimiters like punctuation and
whitespaces. Here Tamil text is given as the input to the tokenizer which breaks the text into
meaningful tokens. The tokens generated by the tokenizer are passed to the analysis engine.
Morphological Analysis
RCILTS [5] developed a tool called Atcharam, an analyser which performs Morphological Analysis for
Tamil text. The Morphological analyser takes a derived word as input and separates it into root word
and associated morphemes. It is the basic tool used in spell checker, grammar checker, parser and
machine translation systems. It has two major modules noun analyzer and verb analyzer.
119](https://image.slidesharecdn.com/c2mala2janani-111110022452-phpapp02-121115053403-phpapp01/85/C2mala2janani-111110022452-phpapp02-3-320.jpg)
![Finally, environmental constraints that specify location and actions are extracted.
ANIMATION GENERATION
Movie clips for the extracted information are created using Adobe Flash professional and stored onto
the database.
CONSTRAINT MATCHING
When an input text is given, information constraint should match with the information present in the
database to generate animation. String matching algorithm is used to compare the information
extracted and information stored in the database. Let P[1...M] and T[1...N] be the character array for
the given string. Pattern P is said to occur with shift s in text T. To find all valid shifts or possible
values of s so that P[1..m] = T[s+1..s+m] ; There are n-m+1 possible values of s.
Procedure String Matcher(T,P)
1. n length [T];
2. m length[P];
3. for s 0 to n-m
4. do if P[1..m] = T[s+1..s+m]
5. then shift s is valid
Find first match of a pattern of length M in a text stream of length N.
The extraction of Pattern 牀牀 牀 is done by,
牀牀鉦 M=4
牀 牆牀 牀 牀牀 牀 牀
牀牀 牀
牀牀 牀
牀牀 牀
牀牀 牀
牀牀 牀
牀牀 牀
By this method exact string is matched from the database for the given information.
STORY ASSEMBLER
Storyboards are the only way to convey rich information, viewing a particular order of events in a
most appealing way. Basically the system searches for noun and verb from the given input text then
automatically assemble and analyse the subsequence like background, named entities, temporal,
emotion and action movie clip from a database that matches the constraints.
MOTION SYNTHESIS
Animation is generated by motion synthesis by efficiently connecting the movie clips that are
assembled by the story assembler from the database. The character and objects interactions are
handled by this module based on the information that the movie clips have. Timeline specifies what
121](https://image.slidesharecdn.com/c2mala2janani-111110022452-phpapp02-121115053403-phpapp01/85/C2mala2janani-111110022452-phpapp02-5-320.jpg)
C2mala2janani 111110022452-phpapp02
- 2. Animated Story Visualizer for Tamil Text M. Janani, Dr. T. Mala Department of Information Science and Technology Anna University, Chennai, India janani_mrl@yahoo.com, malanehru@annauniv.edu Abstract Natural language is a straightforward and efficient medium for describing visual facts and mental images. The System uses a novel approach to generate animation from Tamil texts such as stories. Tamil text is pre-processed and the necessary features like named entities, environmental constraints, temporal and emotion constraints for the given stories are extracted and placed in the database. The system automatically generates a query based on the users input and compare it with features stored in the database. Finally animation is dynamically generated using an external motion synthesis system. Using this system, even greenhorn users can generate animation quickly and easily by giving the Tamil text. Keywords Computer animation, Natural Language Processing, Pre-processing, Feature Extraction, Motion synthesis I. INTRODUCTION These days, animations are widely used in many applications, such as cartoons, web graphics, games, and so on. Computer animation is one of the best methods for depicting the dynamic content. A medium is necessary for the animation to be created in a convenient and natural manner. It should be possible to describe the scenes directly from natural language.NLP is an easy and effective way to analyze, understand and generate languages that humans use naturally. The aim of this work is to generate an animation from Tamil texts such as movie scripts or stories. Training input text is given to the pre-processing module. Here tokenization is performed and the tokens are given to the morphological analyser which is used to convert the tokens into a POS tags. Information related to named entities, temporal constraints, emotion and environment inference features are extracted. A query is generated automatically from the input text which contains information for the search process and compares it with the information already stored in the database. Finally motion synthesis generates an animation. The interactions between characters are handled by this module based on the information provided in the database. II. RELATED WORK Generating animation from natural language texts has been a challenge. WordsEye developed by Coyne and Sproat [1] converts natural language texts to a scene. WordsEye focuses on generating a still image, when a character motion is specified in a given text, the system simply prefer to pose for 118
- 3. the action generated from the database. The Carsim system [2] describes a new version of text-to-scene converter that handles texts describing car accidents using computer program and it is visualized in the 3D environment. Storytelling System [3] illustrates a system called Interactive e-Hon, which provides storytelling in the form of animation and conversation translated from original text. A Constraint based scene conversion system [4] describes a Text2Scene conversion method which automatically converts text into 3D scenes. A large database of 3D models is used by this method to depict entities and actions. III. SYSTEM OVERVIEW In this section, overview of our system (Figure 1) is given, where the major components are identified. When the Tamil text is given to the system, Tamil text is pre-processed and the information are extracted and stored in the database along with the objects created. When an input text is given to the system it automatically generates the query from the input text and compares it with the information stored in the database. An animation is then generated using an external motion synthesis system. Figure 1. SYSTEM OVERVIEW PRE-PROCESSING When the Tamil text is given to the system, natural language processes (Tokenization and morphological analysis) are applied first. Tokenization The first step in NLP is to identify tokens, which decomposes the delimiters like punctuation and whitespaces. Here Tamil text is given as the input to the tokenizer which breaks the text into meaningful tokens. The tokens generated by the tokenizer are passed to the analysis engine. Morphological Analysis RCILTS [5] developed a tool called Atcharam, an analyser which performs Morphological Analysis for Tamil text. The Morphological analyser takes a derived word as input and separates it into root word and associated morphemes. It is the basic tool used in spell checker, grammar checker, parser and machine translation systems. It has two major modules noun analyzer and verb analyzer. 119
- 4. 牀 牀 牀 < noun > 牀 牀む 牀 < verb > 牀 < plural > < past tense marker> 牀 <case marker > < gender > Tamil Noun and Verb classification example By this method morphemes are generated and given to the learning process where the necessary informations are extracted. INFORMATION EXTRACTION OBJECT IDENTIFIERS Object Identifiers recognize named entities in text by Named Entity Recognition (NER). Rule based approach is used to extract named entities from the given text. Initially, root words say Noun, verb, adjective, pronoun, adverb from text file are extracted. Rules are created based on prefix and postfix of noun, i.e. noun that occurs between verb and noun, noun that occurs between noun and noun, noun that occurs between noun and verb and so on. If any of the above rules satisfies the input text named entities are extracted. Here is an example, Input: 牀 牀 牀む 牀 牀 牀む橿 Given input text is pre-processed and the root words are extracted. 牀 <Noun> 牀 <Noun> 牀 牀む橿<Verb> Now the rules are applied to this extracted root words. Here 牀 comes between noun and verb which satisfies the rule is extracted. TEMPORAL AND EMOTION EXTRACTION Temporal reasoning in NLP involves extraction, representation and reasoning with time and events in the natural language text. Here to extract temporal constraints, manually created dictionary is used. The root words are compared with the manually created dictionary and temporal constraints are extracted if the input text satisfies the inferences present in the dictionary. Similarly different emotion present in the text is also extracted using manually created dictionary. Figure 2 shows the different emotional constraints to be depicted. HAPPY 牀 牀 ,牀牀牀 牀牀, 牀牀 ,牀 牆牀む鉦 ,牀牀 牀 ,牀牀橿 ,牆牀 牀牀逗 ,牀牀逗鉦 ANGER 牀牀 牀牀鉦 ,牀牀逗 ,牆牀牀鉦 ,牆牀 ,牀牀鉦 牀 ,牀牆 牀 牀む鉦 SURPRISE 牀牀む逗牀 ,牀 牀牀鉦逗 , 牀牀逗萎牀 ,牀牆牀 , 牀朽逗 FEAR 牀 牀 ,牀牀 , ,牆牀牀 牀 ,牀朽逗む ,牀牀朽牀,牀 ,牀牀 牀 ,牀 牀 SADNESS 牆牀牀鉦 ,牀 牆牀,牆牀牀 牀, 牀牀 ,牀 牀 , 牀 , 牀 牀牆 , 牀 , 牀牀 , 牀牀 , 牀 ,牀 , 牀 牀 , 牀 , 牀 , 牀朽 牀 , 牀朽逗牀 Figure 2 Emotion Constraints 120
- 5. Finally, environmental constraints that specify location and actions are extracted. ANIMATION GENERATION Movie clips for the extracted information are created using Adobe Flash professional and stored onto the database. CONSTRAINT MATCHING When an input text is given, information constraint should match with the information present in the database to generate animation. String matching algorithm is used to compare the information extracted and information stored in the database. Let P[1...M] and T[1...N] be the character array for the given string. Pattern P is said to occur with shift s in text T. To find all valid shifts or possible values of s so that P[1..m] = T[s+1..s+m] ; There are n-m+1 possible values of s. Procedure String Matcher(T,P) 1. n length [T]; 2. m length[P]; 3. for s 0 to n-m 4. do if P[1..m] = T[s+1..s+m] 5. then shift s is valid Find first match of a pattern of length M in a text stream of length N. The extraction of Pattern 牀牀 牀 is done by, 牀牀鉦 M=4 牀 牆牀 牀 牀牀 牀 牀 牀牀 牀 牀牀 牀 牀牀 牀 牀牀 牀 牀牀 牀 牀牀 牀 By this method exact string is matched from the database for the given information. STORY ASSEMBLER Storyboards are the only way to convey rich information, viewing a particular order of events in a most appealing way. Basically the system searches for noun and verb from the given input text then automatically assemble and analyse the subsequence like background, named entities, temporal, emotion and action movie clip from a database that matches the constraints. MOTION SYNTHESIS Animation is generated by motion synthesis by efficiently connecting the movie clips that are assembled by the story assembler from the database. The character and objects interactions are handled by this module based on the information that the movie clips have. Timeline specifies what 121
- 6. kind of action occurs at particular time. Once the timeline has been set, animation is generated for the given Tamil text. Figure 3 shows the animation generated for the given Tamil Text Figure 3 Animation is generated from Tamil text PERFORMANCE ANALYSIS The performance analysis is used to monitor the functioning, efficiency, accuracy and other such aspects of a system. For the analysis performed for the learning process, the overall accuracy obtained is 83%. Figure 4 shows the Performance analysis for Animation generated from the Tamil text. Table 1 Test case for Animation Generation Figure 4 Performance analysis for Animation generation The overall accuracy obtained for the generation of animation is 80%. The performance can be further improved by generating rules and optimizing the learning process. CONCLUSION AND FUTURE WORK The system provides automated generation of animation from Tamil text which provides a new approach for users to create animation quickly. The proposed method takes Tamil text as the input and it is pre-processed and features like named entities, temporal constraints, emotion and environmental constraints are extracted and animation is generated dynamically by motion synthesis. Even non- professional people can rely on this system and they can generate animation quickly and easily by giving the Tamil text. Future work can be extended by generating animation via automatic speech recognition rather than text. 122
- 7. REFERENCES Coyne.B and Sproat.R, Wordseye: an automatic text-to-scene conversion system, In Proceedings of SIGGRAPH 2001, pp. 487-496. Johansson.R, Berglund.A, Danielsson.M and Nugues.N, Automatic Text-to-Scene Conversion in the Traffic Accident Domain, In Proceedings of The Nineteenth International Joint Conference on Artificial Intelligence, 2005, pp. 10731078. Kaoru Sumi, Mizue Nagata, Animated Storytelling System via Text, In Proceedings of SIGCHI International Conference on Advances in computer entertainment technology,2006. Liuzhou Wu and Zelin Chen,A Constraint-based Text-to-Scene Conversion System, In Proceedings of International Conference on Computational Intelligence and Software Engineering 2009. Anandan, P., Ranjani Parthasarathy & Geetha, T.V., Morphological Analyser for Tamil, ICON 2002, RCILTS-Tamil, Anna University, India. 123