power bands On Natural Language Generation Technol

[powerband6i]

2.4
visual grounding technology process: 1) data preparation, since before the birth of the natural language, the computer will show a variety of different target objects, so as much as possible to record all acts related to the target object. 2) The term clustering, term clustering cluster consists of three basic ways: distributed clustering, when the two words appear in the same statement, the two mutually exclusive terms, not the same type of words, and vice versa was established. Clustering based on semantic features of the connection, this clustering is at the same time ignored the word appears in a statement, the main features of the semantics of words and their links, the main purpose of this clustering method for those with visual characteristics are ground contact of the word cluster. Mixed cluster, the word clustering is whether the words should also be considered a common occurrence and semantic characteristics of words and visual contact. 3) feature selection, the characteristics assigned to each individual word, the characteristics of a class of words to obtain other types of word choice as a feature of the connection. 4) a semantic model, the characteristics of the words get quantify different characteristics of the Gaussian model for the establishment and in different regions indicate the corresponding Gaussian state of the target object. When a new target object appears, the first feature extraction of its value, then the Gaussian region where the characteristic value can determine the status of the new target object. 5) produces an object description language, which is a visual grounding the final stage, we hope to form a grounded language model to generate the object description language. This language is necessary to generate a constraint into the search problem. There are three constraints need to integrate the search, namely: grammar constraints, semantic constraints and context constraints. Syntax constraints mainly used to produce the words and syntax compatible; semantic constraints can describe the characteristics of the target object; context of the search constraints can reduce the target object relative to other objects ambiguity. Constraints as long as we integrate these three elements to search, you can achieve the object of the language description.
Natural Language Generation (Natural Language Generation) is one of the two major areas of natural language processing, which is a branch of artificial intelligence and computational linguistics is the computer system generates understandable text. Natural language generation technology is artificial intelligence (AI) technology in one of the most active, it is to study how a computer program based on information generated inside the computer that a high-quality natural language text. Many scholars have to the NLG Technology.

ground refers to the visual language and the context of the user to link things related to treatment process. Natural language generation system using computer vision to form a description of the object scene. This system is to adopt a This description method is based on a learning algorithm, the learning algorithm will be a probability structure that visual semantic structure of words, parts of speech and individual words to be encoded. Using this probability structure, one that contains the syntax, semantics and context of the algorithm can limit the formation of natural and unambiguous descriptions of objects. Formation of natural language generation system consisting of adjectives and nouns noun phrases, but also the formation of the space clauses. Linguistic structures will need to generate training data, we can also handle the training process there have been no new word sequence. The output of the language generation system through comprehensive scanning of the original training corpus based on words to achieve. Evaluated from the human judgments, which automatically generates the language to describe the performance of human language can describe the properties comparable.
above four traditional natural language generation technology development in natural language generation plays an important role, until now, there are many important fields of science use these Language Generation technologies. However, with the scientific and technological progress, the existing natural language generation system does not have the ability to learn has been thoroughly exposed, so people put forward a higher its requirements: ability to learn.
2.1 template generation technology (Template-based Generation)
2.3 phrases / Rules Expansion Technology (Phrase / Plan Expansion)
natural language generation as the common theoretical linguistics and computer linguistics research topic in recent decades has been considerable concern people, and the corresponding natural language generation technology is gradually moving from traditional technology to the new technology forward.
1 Introduction
the automatic generation of natural language that people in the simulation process is carried out in the article unconscious generation process, the article generated in the process, there is always first in people's minds some kind of In most cases, these things would like to say, no more than two or three basic concepts and can be set up between them a relationship. To make them speak out or take them out of the stage of writing, consider the words express these concepts should have the grammatical features. In the field of computational linguistics, we describe the above process can be automatically generated during the text is divided into two stages: the decision phase and the presentation layer content generation phase. Usually people development and application of NLG system has two main purposes: first, as the communication tool in people's lives, the use of linguistic knowledge and domain knowledge to generate the text, analysis reports, help messages and so on. Second, as a test of a theory of language-specific techniques.
2 Natural Language Generation
visual grounding not only has the advantages of learning and memory function, but also in the process of language production, the information on to express the semantic and syntactic aspects of polymerization, which is the current focus of the study. The disadvantage is that the visual grounding in the data preparation stage to manually record the behavior of the target object state, which would be a waste of manpower and resources.
advantage of the method is obvious: for efficient means of achieving simple. The disadvantage is the low quality of the generated text, it is difficult to meet the changing needs of the people, can not generate body problem specific analysis of the text. Secondly, the use of modeling techniques for system maintenance, modification or expansion is very difficult.
Posted by: Si Chang
2.5 Learning-based natural language generation technology - visual grounding
Generation property features
RST (Rhetorical Structure Theory) generation technology is based on Mann and Thompson's description of text structure on Rhetorical Structure Theory, RST theory article, clause or whether the various parts of a larger number of constituent units are made between one of the few, the relationship between repeated according to a certain level of cohesion together. Most NLG generation system includes a set of rhetorical relations, and relations with the specific application of the corresponding set is a subset. RST relations Nucleus-Satellite and Multi-Nucleus two modes. Nucleus-Satellite model which includes the core (Nucleus) and subsidiary part of the (Satellite), the expression of the core of the basic proposition, a subsidiary of the subsidiary part of the proposition expressed, more used to describe the purpose, cause and effect, transition, background and other relationships; Multi-Nucleus model involves a or more of the discourse, it is not affiliated parts, more used to describe the sequence, parallel and other relations.
concept of this method has the advantage of simple,balance bands, any kind of different types of language can be easily added as a feature into it. Generated text is quite flexible. Its weakness is difficult to maintain the content of the relationship between attributes is difficult to control the feature set selection.
template using a first generation technology, this generation and filling method similar to the principle, the system had been designed several possible scenarios, the corresponding structures of several templates, each template includes a number of constants and a number of variables, when the user input certain information, the text generator to the information embedded in the template as a string substitution variable. This generator is called non-language text generator, because it's handled only in a string level, not at a deeper level for language processing. Although the idea of ​​this technique is simple, but the resulting text quality is not high. But it still has a very wide range of uses. For example: If many applications are using the technical processing error messages, warning messages.
pattern generation technology is NGL technology, so it has good maintenance, high quality output to the text. It sure is only a paragraph for fixed structure, the resulting text is not flexible.
Mckeown's Schema technology is based on linguistics, rhetoric predicate (Predicate) to express the text structure of a method, which uses Predicat law to describe the structure of the text is the text of the skeleton as that of the representation is also clear expression of the theme of discourse in order. Text is composed by the proposition,power balance, the proposition is the Predicate of a sentence or a clause in the proposition based on text classification, each proposition are summarized as specific Predicate. In terms of the same type of text there are some combination of the standard model to represent the Predicate structure of the text, this model is called Schema. Its corresponding node in the tree is usually divided into five types: Root, Schema, Predicate, Argument,power bands, and Modifier. Which, Root is the root of the tree that article. Root number below each child node Schema, Schema represents a paragraph or a sentence group, Schema following child nodes can continue to Schema, can also be a Predicate. Predicate is the root of a tree to represent a sub-tree of the sentence, which is the basic unit of the article, the basic semantics of the sentence for each component are a subset of Predicate nodes, with the Argument said. If the Argument with modifiers, the use of child Modifier flag. Argument or Modifier is a tree leaf node, each node in the tree contains a number of slots used to mark produce all kinds of information for use.
3 Conclusion
sense will, the property features generation Shisheng Cheng is the most difficult techniques. In these systems, every possible part of the minimum that can change the properties by a simple feature that out. For example: the tone of a sentence is active or passive, it is the problem or the tone of command action or a statement. The output of each unit with the unique attributes of a particular feature set is connected. Output process is to increase the production of each property characteristic information section, has been the only way to be able to determine an output results so far. Then a linear process to a string of properties feature set into a linear string of symbols. The level of the sentence level, grammatical features characteristic attribute is the output to the symbol vocabulary.
natural language generation has experienced nearly four decades of development, natural language generation technology has been tremendous growth with each passing day. In time, the domestic and foreign experts in this field have been put forward new theories and methods to design a new generation model, the language generation of new progress. Language generation for future research is mainly in the following areas: natural language generation research will focus on research prescriptive grammar language form go to focus on revealing the reasonable effectiveness in communication. Recent research has focused on the participants to talk dialogue, and its planning process as the information the main problem should be solved. The language of the existing planning procedures and information generation program lacks the necessary interaction between, from now until the next period of time, text messages, focus on planning is still one of the topics. At the same time, to test a particular theory of language and language generation model developed will be developed. Vocabulary and syntax in information planning to compress the combined two levels of study can not be underestimated importance. Natural language generation research will continue in a number of language learning, computer areas and close cooperation with other disciplines to obtain new results.
Health technology is the people
Because the current generation system in terms of natural language grammar vocabulary, or in the information processing aspects of planning do not have ability to learn. To compensate for this deficiency, we introduce a learning-based natural language generation technology - visual grounding.
technology, compared with the Schema, Phrase / Plan technology has more flexibility in the tree generation process, while also generating the overall structure of the text. Its main basic data to determine its structure, the text is often more difficult to establish the rule base, because the internal relations between sentences restrictions must be carefully considered, should be to prevent inappropriate expansion.
2.2 Pattern Generation (Schema-based Generation)