KR101506757B1 - Method for the formation of an unambiguous model of a text in a natural language - Google Patents

Abstract

The method relates to the generation of a clear model in natural language. The basic notions for the meanings (contents) expressed by the natural language are defined, numbers are attached to each basic idea, or names and descriptions are used in the list of words that can represent the basic idea for each used natural language It is added. Clear models use only basic ideas. It is therefore possible for a machine to interpret a clear model, to input information and data in the base, or to generate text in a different natural language using a clear model. A text in an artificial language can also be generated as a language for programming.

Description

METHOD FOR THE FORMATION OF AN UNAMBIGUOUS MODEL OF A TEXT IN A NATURAL LANGUAGE BACKGROUND OF THE INVENTION [0001]

The present invention relates to information input in a machine using a natural language, and machine translation of a natural language.

Quot ;, " Method and Apparatus for Generating Structured Documents ", such as SATO YOSHIFUMI, and the predefined structure of the document is named " non-structured " US 6014680 documents are known which disclose a method for altering the manner in which structures are matched, made from keywords derived from the use of data in a complete change of structure and interpretation of the document according to the changed structure. Just as the defined documents can be interpreted in a single way, even though the facts that keywords can be taken into account can be taken into account, a recent document has the notion that for the generation of a clear model, Is different from the present invention because its generation is formed not only as a natural language but also as a tag and description of a natural language and a list of words and texts representing basic ideas of various natural languages. Lines and sentences), but refers to ideas found in text and interrelationships between them.
It also relates to a method and system for describing and identifying ideas in a natural language text for information retrieval and processing, such as FASS DANIEL C, which discloses concepts predefined by the user and composed of words and descriptions , The description is expressed in a concrete language that describes ideas, giving the opportunity to find synonyms, text, or search rules for a given word. The basic idea is that the proposed notion defined by the user can be found in the given text with the help of the description. The texts processed using this method can be displayed with ideas found for easier future retrieval. The latest notion is defined by the user and expressed in dependence on the employed search method.
The basic idea according to the present invention is that content or function understood by all people in one and the same way, regardless of the language they use. The basic idea is defined by a description in natural language. Natural language explanations are not used in the search, but while the tags of the basic idea provide the basic idea to be ignited while the human feedback, the notion of processing the text in natural language and the search for the generation of text in natural language Quot; is used as an attached list of words and texts representing concepts given in various languages to provide < / RTI > When basic notions are defined, a comparison is used between the descriptions of similar words of a given word as they are expressed in a dictionary. Even though they are "user" definitions, they can not be provided to look for ideas in a given text. The explanations are compared, the similar texts are defined as potentially explaining basic ideas, and the determination of whether the texts containing similar words are semantically similar will be obtained by the person making the basic idea, although it is not found. It is the most common system that machines interpret a large number of words given from natural languages, and these are actually all artificial languages. There has been an attempt to define the grammatical meaning of words. For example, in the case of a machine translation process, a semantic range (content) is input in order to define the preferred meaning of the words in the presented text, thereby achieving better results. There has also been an attempt to define the meaning of a given word based on a statistical analysis of the words accompanying it and the frequency of occurrence of such words in the text presented. There is an attempt to mark words in still presented languages and to make the same markings on words in different natural words in multiple words so that the words of two languages with one same meaning get equivalent .

The problem of clearly interpreting a natural language by a machine has not been solved, which is a hindrance in inputting information and data from a machine using natural language items. One is that the translation of official documents can not be machine dependent. No text can be compiled into natural language, which is especially necessary when writing on textbooks and patent applications that can be the subject of clear interpretation by others. A sentence in a natural language can have many acceptable meanings from a formal point of view, ie, no computer can be programmed by natural language, because grammatically correct sentences can be interpreted differently. Since there is no official tool that can process one piece of information written in natural language, accumulated human information can not be utilized optimally.

The interpretation of natural language always involves building a machine model of the interpreted information. The text of a natural language is processed in various meanings so that the meaning of the sentence, the word in it, and the grammatical elements of the word can be defined. The problem is that there is no feedback and the model can not be changed. This is due to the lack of a basis for comparing models and natural language texts. In other words, this model is a structure that can be interpreted in two more ways - once again. The technical nature of the proposal is a way to form a clear model. A model formed in this way will only be interpreted in one unique and clear way.
The method consists of five steps.
In the first stage, a number of languages are studied, the purpose of which is to define the set of basic ideas used by mankind. It should be noted that the words in the presented language can not be considered as a "basic" notion. The "basic" notion is a suggested semantic content (content) or sign of action, and in most cases one identical word in the presented natural language is used to represent various basic notions, ie words have different meanings. Technical suggestions for marking "sun = 1", neuioa = 1 (Bul.) May be used in the case of machine translation but will not contribute to us with a proper and clear translation. While this system creates translations in the form of "User rignts = rights of the drug addict (Bul.)", The word simply does not represent anything other than "user rights". This numbering of words does nothing but create a kind of intermediate language that can be interpreted in various ways. The proposal is not a word, but a number assigned to the meaning of the words. According to the method, these meanings are singular names that can be not only binary but also widely propagated natural language words. However, it should be noted here that words presented from natural language can be used only once to indicate the semantics presented. Thus, while the word "sun" can only represent a star, other words will be selected for all other meanings of the word "sun". It should be understood that the representation of the actual meaning of the words has no effect on the natural language. According to the method, the contents are characterized by their description, and these descriptions are made into natural language in the same way as the natural language dictionary at compiling. All the attached content is a list of words that can be expressed in the presented natural language, that is, a thesaurus of their meaning (content) contents rather than words.
The structure for displaying meaning contents having a unique label-name or number, a description and a word list, mixed words or texts representing the above-mentioned semantic content in a natural language can be further classified into basic notions It is called.
In this method, the second step is to structure the model of the natural language text using only basic ideas. At this stage, since all applicable technical methods are used, the grammatical and semantic meanings of the words can be defined, and individual models are formed. In the process of forming these models, global statistics on the use of words in their various meanings can be used, or local statistical data for each distinguishes users of the method. Similar texts with predefined meanings for words may also be used. Human translations of text given from one language to another are also employed for the purpose of defining basic concepts used in the text of a natural language through evaluation of selected words for translation and comparison of the meanings of the original texts and their meanings .
The third step of the method is feedback. At this stage, the model formed in the second step is used as the basis for generating the text in the same natural language in which the original text is constructed. The operator will have the opportunity to use a computer program that affects the changes in the generated model to follow the operator's understanding of the text. This can be accomplished, for example, through direct changes in the model as it interacts directly with the meaning (contents), represented by branching in addition to the relationships between the various contents. All of these may require considerable learning, or may be obtained through an attempt to explain to the computer whose meanings need to be changed. It is possible to compare the original text with the generated text to indicate the difference between them. A list of synonyms for each displayed word is obtained from the dictionary of synonyms, filtering of synonyms is rejected, and synonyms for inappropriate content are made. The operator will select from a list of synonyms, and this process is repeated in real time, i. E. There is a new generation process and a possible new collection. However, the choice of synonyms is not always sufficient to define the suggested meaning. As such, the semantics can be provided for changes to the interpretation of the interconnection between the two basic notions of a given text. Thus, using the visual meaning - for example, marking and underlying - the presented correlation can be defined. For example, you can define what the subject of the sentence is, what is the narrative and what the attribute is. It may be possible to form means by which the temporal relationship in the text is indicated. External features of the text can be given, so that interpretation and generation can be controlled more efficiently. For example, cases may be quoted differently from the standard interpretation, such as word play or satire, and both interpretations should be presented so that changes in standard and extrinsic characteristics are part of a clear model . A large number of such means can be used on average, because they want to help educated people express their thoughts on the computer. Here, the goal is to get a clear model that conveys the meaning of the message in the most accurate way.
In a fourth step of the method, a generated Unambiguous Model is attached to the file, including the natural language text. This makes interpretation of the natural language text clear, which is particularly useful in patent applications and machine translation. When the body of a textbook is created through the method of the explicit model attached, the computer program uses the definition of the meaning used in the text, as well as the definition of the content used to define the content at the higher level Perhaps you can generate a description with any level of complexity.
The fifth step of the method involves the use of explicit models of the natural language text for artificial formation of ideas and theories based on machine-based learning and formalized knowledge, derived from a clear model of the natural language text.

The present invention can be applied in machine translation in retrieving information in the case where the search is based on a similar and clear model of the text to be searched, rather than searching based on the words contained in the text at the present level of technology . A search may be made using a clear model analysis of the text, which allows the Explorer to answer questions such as searching information about transferring property to a foreign citizen in accordance with Bulgarian law.

An exemplary implementation of the first step of the present invention
With the aid of a computer program, the basic notion of language is defined and a list of synonyms is searched for each word of the natural language being examined. The description of each word from the language presented in the individual dictionary is compared with the description of the synonym presented in the same dictionary. Comparisons of descriptions are made through searches in text similar to simple comparisons of texts. The purpose is to define the various meanings of the presented words according to the synonyms of each meaning. This is done by comparing the description of the word with the description of the word's synonyms, from which similar texts can be defined, as suggested in the preface. They also form meaning (content). The description of this content is, in principle, formed by similar texts in the description of the two synonyms. When such content is found, a database check may be performed to determine whether the content has not already been registered, based on a comparison of the description of the registered content with the description of the newly discovered content. If the new content is not registered in the database, it will be registered slowly later.
After automatically forming a database of content and its description, the specialist will invite them to define labels for their content and to specify their descriptions. Regardless of the text - may be attached to the contents that may represent the contents under any condition depending on the text including, for example, scientific text, language play, words and features. It is possible to present a database of all contents in the presented language, and each content can be explained through a clear model of description in natural language. This can be accomplished by professional linguists capable of generating a clear model based on an automatically formed description of content in a natural language, using basic notions of the language. After finding the basic idea in the proposed natural language, the next natural language will use the formed database of the basic idea. The linguist will simply define how to express the registered contents in a separate language, and - perhaps - many of the contents have to be added to the database. Linguists who maintain a natural accord with the base should be made aware of each addition of new content so that they can propose and display the content appropriately in the language they are responsible for. The display of the contents can be made completely descriptive.
It is also possible to automate the investigation of the second, and subsequent natural language. In such a case, the same procedure as the first natural language is applied. Thus, a new database of registered contents will be created. The indications whose content is to be obtained from the second database are words from the second language. From the dictionary in the second language to the first language, possible translations are made for each representation of the content from the second base. For the translation of each of the words from the first language in the first base, the content that can be represented by this translation will be subtracted. Pseudo-translation will produce a description of its content from a second language by generating all combinations of the contents of each word in the descriptions with all possible translations of the first language. The physician translation for the description of the content in the second language will be compared to the description of the content subtracted from the first base. The best correlation is found and displayed. Each of these conformities must be approved by a linguist. After processing all matches, the remaining content in the second database will either be registered as new content in the first base or will find their correlation in the first database.
In the case of official documents, the unity of the generated text in a natural language from a clear model must be achieved. This can also be done in order to simplify the texts generated in natural language, that is, when the work of linguists is to add many features of the text to a definite model when there is sufficient achievement of clear production, It is possible for a large number of natural language texts to be conveyed and meaningful to convey information. Such an approach is particularly important in translating official documents from one language to another and especially in translating patent applications.
On the other hand, in the novel translation, it is better to generate a large number of texts in natural language from the definite model, in order to select the most appropriate structure for the actual language using the statistical data derived from the novel literature written in the presented language.
An exemplary implementation of the second stage of the present invention
The text can be represented as a branch-out list in which each tree represents a sentence from the text. Connection between different trees is possible. Each element in the tree is an object with features automatically derived from the body text or manually added by the operator. Some elements of the tree - sentences in the text - such as pronouns - can have links to elements of other trees. The arrangement of the trees in the list is important because it points to the body of the original body, as well as the array of sentences and the body from the explicit model.
An exemplary implementation of the third step of the present invention
The text editor superstructure is made up of additional features and support that allows changes to be made to the explicit text model automatically created with simplification of the body format. For example, the screen is divided into three zones. One section will contain the entire original body - a generic text editor. The second one will include feedback when building a clear model. It will contain the body of the processing statement generated by the machine. When the cursor of the mouse is fixed to the presented word of the machine-generated document, the supporting sentence (hint) will provide a description of the underlying meaning (content) that the individual word represents. The same sentence will be properly displayed in the original text. The third zone is a group of devices used to change a clear model applicable to the second zone. These devices include modifications to the interpreted content through synonym display of the word with the synonyms of other content named by the words of the question, and also by hints, representations of the underlying content named by the synonyms are displayed It is also possible. This includes the selection of specific texts such as language play, jokes, poetry, or science texts. It is also used to denote the exact meanings of the contents of pronouns actually used, such as her, her, or the like. The exact meaning can be given in the framework of the entire text by indicating the relationship between the previous sentences of the text and the given pronoun. The text is continuously analyzed from start to finish by entering all the necessary features and relationships to form a clear model. The sentence of the question will be processed until the machine creation does not create a text that has at least the same meaning as the original text. The process consists of a number of modifications and productions.
An exemplary implementation of the fourth stage of the present invention
The generated clear model of the proposed text is attached to the original file. This attachment can be done in various ways. Links can be attached to a clear model of the text in the original file. Files of the original text and files of the definite model can be registered in a single archive pack. In principle, it should be borne in mind that the text presented in natural language can have a large number of well-defined models. This is because the diversity of interpretation of the text presented in natural language is filtered by the human operator, who uses his or her own understanding in the process of translating the text from a human, that is, a natural language into a definite machine model. It is therefore possible to link the text presented in natural language to a number of distinct models. In the case of a patent application, the protected object is a single, definite model of the filed text, as is, of course, submitted.
An exemplary implementation of the fifth step of the present invention
A clear model of natural language text can be subject to formal processes. Various kinds of expressions of a definite model can be generated as suitable for various kinds of mechanical processes. Because formal interpretations can be applied to those models, explicit models can be viewed as a new kind of computer software. In this way, machine training can be perceived by extracting facts and relationships from a clear model of the natural language text. All mechanisms learned by artificial intelligence can be applied clearly and formally. In this way, conventional software can be replaced by an expert system that provides a service for the generation of application software in accordance with the user's needs, in a natural language, with the easy addition of a clear model.
The disclosed method is implemented by special computer software. Computer programs can be used by professionals to create and support databases with basic notions used by mankind. Other computer software is available to all users who create and use a clear model of natural language text. The final computer software should be able to form relationships between the basics and the database.
The application of the present invention may be in the area of machine translation in searching for knowledge that is perceived without using the words contained in the text, which is assumed by current skill levels, but is similar to a clear model found in the required text It may be better to find clear models. It is also possible to search with the help of analysis of definite models in text. In that way, the search engine can answer questions related to information about transferring property to foreign citizens under the Bulgari Act. The application of this method is not only for the possibility of automated searches and investigations, but also for the clear definition of protected objects, as well as the possibility of mechanical processes in the most up-to-date valuable information of mankind that will lead to the automatic generation of new information for mankind. It is especially important in the field of patent systems.

Claims (18)

CLAIMS What is claimed is: 1. A method of forming an unambiguous model of a natural language text, the creation of a data structure including words in a natural language text and semantic and grammatical relationships between them, silver,
a computer generates a clear model of the natural language text by selecting words and / or text associated with basic notions and selecting the basic notions named by the words or texts, ; Selecting one basic idea for each one word or text according to a predefined criterion; Establishing a relationship between the model of the text and the notions that includes only basic ideas, the basic notions being the markers of an entity or an action, A unique label, a description of said independent meaning in natural language, and an associated list of words in natural language and the text used in said method, said words or text being each said independent Lt; / RTI >
b) a defined criterion for the selection is applied in advance and uses the definite model on a basis of computer generation of new text in the same natural language or in another natural language, Displaying the basics in the model by one of the words or texts in a list of words or text associated with the word;
c) comparing the generated text using the computer to display the natural language text processed in step a) and the difference between two texts in a language;
d) applying user feedback for the selection of alternate basic notions when there is no semantic conformity between the explicit model and the natural language text processed in step a), and using the modified clear model b) Or proceeding to step e); And
e) using said computer, linking said generated clear model of said text to a file containing said text in said natural language. delete The method according to claim 1,
The list of semantic and grammatical properties, together with the type of text (science, prose or poem), as well as indicators of alternative basic ideas in the case of satire and language play, Lt; RTI ID = 0.0 > concept. ≪ / RTI > The method according to claim 1,
The application of the feedback,
Editing the clear model using an expression of alternative notions of basic notions having expressions in respective languages that are the same as the basic notions automatically selected; And / or
Editing the definite model using a synonym representation of a representation in each language of the automatically selected baseline; And / or
And displaying the explicit model using a representation of dependencies in the processed natural language text. The method according to claim 1,
Wherein the linking of the natural language text and the explicit model is performed using an archive package comprising a link or a file having the natural language text and a file having the data structure. The method according to claim 1,
And generating a body of a language different from the language of the original body. The method according to claim 1,
Wherein the artificial language comprises an information language used in various systems for programming, data processing, a formalized language for symbolic registration of scientific facts, And a language using the theories of the scientific domain. The method according to claim 1,
Wherein the natural language descriptions of the basic concepts also include representations by explicit models. CLAIMS What is claimed is: 1. A method of defining a set of basic notions in natural language, including the discovery of synonyms of each word from a natural language of a dictionary and storage in a database,
Comparing the definition of each word with the definition of the synonym of the word given in the dictionary and defined in the pseudo texts associated with the two definitions, the definitions being in different meanings And there is one basic notion for each independent entity;
Deriving an associated baseline for each pair of similar texts that includes similar words or word-synonyms higher than a given percentage of the total content;
For each related basic idea, a similar text found in a previous step is compared with a description of the basic ideas in the database to check in the database whether each of the basic ideas is already registered, and similar words Or if the word synonyms match more than a given percentage, deriving the basic notion as already registered, along with the explanations found for the basic notion and the other two similar explanations provided as a reason for the check ;
An application step of user feedback to check whether the derived texts match semantically due to the similarity of the texts, and if so, to add one or two word-synonyms to the registered representation of the basic idea; And
Adding a new baseline concept to the database when a check performed during the previous phase indicates failure, wherein one of the two similar texts is selected or a new description is entered. The method of claim 9,
In order to add a different language to the set of pre-established bases,
A first step of defining a new set of basic notions for the natural language when the basic notion can be automatically defined when the basic notion can be automatically defined;
In each of the basic notions defined in the previous step, from all possible translations for displaying each using a dictionary stored in a machine-readable memory, from the language to a language already present in a set of prefabricated preconceptions, Finding the second stage;
A third step of finding basic ideas having expressions attached to those matching the translations in the second step from the set of pre-formed basic ideas;
By generating each of the possible combinations of replacing each word from the above description with all possible translations for a language existing in the set of pre-formed bases using a dictionary in the machine-readable memory, A fourth step of generating pseudo-translations for the description from the language processed in the second step;
If the pseudo-translations are generated in the fourth step, the description of the basic ideas from the set of pre-formed basic ideas found in the third step, Comparing the pseudo-translations to the descriptions of the basic ideas from the fifth step;
A sixth step of arranging similar texts in descending order according to the degree of similarity;
A seventh step of applying user feedback to each correlation found in the fifth step to determine whether similar explanations of the similar words have semantic similarity;
Deleting the basic notions from the new set if semantic similarity is found in the previous description in the previous step and replacing the descriptions and representations in the language of the deleted basic notion from the new set into a set of pre- Adding to each existing existing notion and marking an added list of expressions in the language;
A ninth step of registering new basic ideas in the pre-formed set, or applying user feedback to find a correlation in a pre-formed set of all remaining basic ideas in the new set after applying the previous steps;
And a step of finding an appropriate expression in the language of all the basic ideas from the preformed set that has not received the expression after application of the previous steps using user feedback. A method of forming feedback in a natural language processing, the method comprising automatically suggesting alternate words and / or texts, the method comprising applying the feedback to determine the meaning of the words and generating speech and semantic, ≪ / RTI >
Using global statistics for use of the words in different meanings of the words, and / or using local statistics for each of the individual users of the method; And / or
Using a pseudo text having a predefined meaning of the words; And / or
Using a translation of the original text generated by the human in one or more languages. An apparatus comprising a definite model of a natural language text formed according to any one of claims 1 and 3 to 11, the apparatus for performing the mechanical processing of the same text in natural language. The method of claim 12,
Wherein the machine processing is machine translation from one natural language to another using a generation of a clear model of the first natural language text and a generation from a definite model of the second natural language text. The method of claim 12,
The mechanical processing not only replaces the basic ideas from the explicit model of the processed natural language text with descriptions but also replaces the basic ideas from the explicit model of the explanatory use repeatedly with the above description, A device that is a creation of a body having a predetermined complexity. The method of claim 12,
Wherein the machine processing is a textual creation in an artificial language. The method of claim 12,
Wherein the mechanical processing searches in the explicit model of natural language texts for the generation of a textual body in a natural or artificial language from a found definite model or to display a body match in a found definite model. The method of claim 12,
Wherein said clear model of the text in a natural language is processed and the results are input in an information base from which data, facts and correlations are extracted. The method of claim 12,
The mechanical processing may provide the user with an opportunity to operate one computer device while the mechanical processing is performed by another device and / or data from the machine processing is stored in a device other than the one on which the mechanical processing is performed A computer-readable medium having computer-readable instructions thereon.