RU2509350C2 - Method for semantic processing of natural language using graphic intermediary language - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: method includes a step for syntax analysis of text. A step for extracting text components and relationships thereof in the text is then executed. A graph or graphic representation of the text is generated or used as representation of the meaning of the text independent of the language. That graph or graphic representation is used to perform modelling, knowledge presentation and processing in a language processing system. A judgment of the representation in the model of the semantic realm is made during the processing step, thereby checking consistency of the extracted text semantics.

EFFECT: improvement and further advancement of the method of processing natural language which enables to properly process text semantics or other data.

29 cl, 15 dwg

Description

The present invention relates to a method for processing a natural language using a language processing system, in particular an electronic translation system in which written or oral text is input into a language processing system. The present invention also relates to a translation system and, in particular, to an online translation system.

Natural language processing using language processing systems is a problem. Natural language consists of a sequence of words composed in a certain way to express a certain meaning. Simply put, language processing systems can analyze text by looking at the sequence word for word. Unfortunately, an isolated analysis of individual words does not make it possible to correctly identify the meaning of the sequence. In some cases, it leads to success, but very often analysis does not give anything, because the text is more than just a clustering of words. The sentence "colorless green ideas sleep furiously" is made up of words ordered syntactically correctly, i.e. the syntax (rules and principles that determine the structure of sentences in a given language) is applied correctly. However, it is easy to see that this sentence is meaningless. A system that only focuses on single words tries to process a sentence, while it is obvious that it cannot be reasonably processed.

For example, an electronic translation system may process an input text sequence according to the process shown in FIG. 1. At block 100, the user can enter an input text sequence for translation by, for example, a user interface, an electronic document, or the like. At block 102, the electronic translation system can analyze the sequence based on the syntax rules of the source language. At block 104, the electronic translation system can search the dictionary using the input language as an index into the dictionary of the output language for each word. At block 106, the electronic translation system can provide translated words based on the rules of the syntax of the output language, and at block 108, the electronic translation system can output the result to the user through, for example, a user interface, an electronic document, or the like.

Some systems known in the art use semantic verification. These systems use vocabulary that combine words with attributes. When performing semantic verification, the attributes must be consistent. For example, the word "animal" refers to the concept of "living", stone - to the concept of "inanimate" and "eats" - to the concept of "living". Using this kind of semantic verification, the sentence “stone eats grass” can be indicated as incorrect, since stones are not living beings, while the sentence “live eats grass” is true, because both the words “animal” and “eat” have the attribute “animal ".

These decisions are supposed to represent an understanding process in artificial intelligence. Unfortunately, this approach is very limited. When processing a natural language, sentences, in the general case, turn out to be much more complicated and cannot be processed by such systems. Great efforts have been made to solve these problems. On the one hand, some argue that programming semantics is impossible. On the other hand, there are companies that invest millions in semantics research. However, so far not a single system known in the art is capable of correctly processing natural languages.

In addition to the above, the development of electronic translation systems is a time-consuming process and provides separate language pairs. For example, in FIG. Figure 2 shows a conceptual diagram of language pairs for four (4) languages: English, French, Spanish, and German. However, in order to translate from any of the four languages to any other translation system, six (6) sets of language pairs are actually used, i.e. Language pairs English-German, English-French, English-Spanish, French-Spanish, Spanish-German and German-French.

The complexity of such a system increases significantly with the addition of additional languages. For example, adding a fifth language, Italian, adds four (4) additional language pairs, resulting in only ten (10) pairs. It is noteworthy that for each pair - complicated dictionaries, syntax and sets of semantic rules use large resources for development. Similarly, in such systems, each translation is carried out individually, even when translation into several languages is required.

Thus, it is an object of the present invention to improve and further develop a method for processing a natural language that can correctly process text semantics or other data, such as input speech and the like. Another objective of the invention is the improvement and further development of the language processing system for processing natural language, which avoids some or all of the above problems.

According to the invention, to solve the aforementioned problem, a method is proposed containing the features of claim 1. According to this paragraph, such a method is characterized by a step of analyzing the text with respect to its syntax and morphology, a step of extracting the components of the text and their relationship, a step of generating or using a graph or graphical representation of the text as a representation of the meaning of the text, independent of the language, and the stage of the text processing using a graph or graphical representation.

According to the invention, at first it became clear that the problem can be solved using the results obtained in the field of neurology. One main result is related to the fact that a person’s cognitive ability clearly distinguishes between syntax and semantics. If several people speaking different languages are in the same room where there is an umbrella, everyone “knows” that it is an umbrella. But this "knowledge" does not mean that the word "umbrella" is activated somewhere in the brain of the person present. However, for communication purposes, the umbrella object is tagged with a language-dependent word. Participants know the object without using the language. For example, if they want to leave the house when it rains outside, they activate the “tag” through a language-dependent dictionary for communication purposes. For example, they ask: "Can I take this umbrella?"

This clear distinction between syntax and semantics (or language-dependent information and language-independent information) underlies the method of the invention. At the first stage, the text entered into the language processing system is analyzed in relation to its syntax and morphology. At this stage, the grammatical structure is analyzed. This gives the first basic understanding of the text. The next step is to extract the individual components of the text. The text, in the General case, consists of sentences that contain the subject, supplement and predicate, respectively. Each component can be extracted and its role in the sentence revealed. These individual components and their relationships with each other are used in the next step in generating a graph or graphical representation of the text. The individual components form the vertices of the graph and the relations between the components are represented by edges. A graph is generally expressed by matrices. However, the logical structure can also be represented graphically for better understanding by people. It was found that this graph can be made completely independent of the language that is used when entering text into the system. The graph includes semantic information that is easy to use for further processing.

Instead of a graph and / or its graphical representation, other forms of graphical representation can be used. They, for example, include presentation using videos, icons, and the like.

Alternatively or in addition to the step of generating a graph or graphical representation, an existing graph or graphical representation can be used in the step of using a graph or graphical representation. This graph or graphical representation describes the knowledge already present in the language processing system. At this point, components extracted from the text are mapped to elements of an existing graph or graphical representation. In this way, a subset of an existing graph or graphical representation is determined.

According to a particularly preferred embodiment, the text input into the language processing system is modeled by visual-graphic or pictographic means. The result is a visual-graphic model that represents the text regardless of the language and which every user of the language processing system can understand. Therefore, users are not required to have knowledge of the languages used. This is also true if the user does not understand any of the languages used in the language processing system.

At the stage of the analysis of the text, information on the grammar of the language used in the input text is accessed. Each language has its own special grammar that defines the arrangement of words. So that users can, without knowing programming, prescribe grammars, grammar data can be entered using the grammar editor. Preferably, this grammar editor is language independent. Only a certain formalization of the possible structures of languages is required. This way you can avoid the lengthy development of different grammars for each individual language and instead get the opportunity to quickly and efficiently create prototypes. Thus, new languages can be quickly and directly integrated into the language processing system. Grammars generated by the grammar editor can be used for language analysis, as well as for language generation.

Preferably, the text analysis step is carried out by the syntax layer of the language processing system. The language processing system may have a modular configuration that provides reusability and modularity of the system. The syntax layer can segment and tokenize text. Segmentation means breaking the text into sentences, and tokenization means identifying specific word forms in a sentence. In the implementation of segmentation and tokenization, individual elements, as well as their relationship in the proposal, can be analyzed with respect to syntax and morphology.

To improve modularity and to provide a method that can be universally used, the syntax layer can be linked to a language processing system. This makes it possible to easily integrate different languages into the language processing system, adding a new syntax layer to the system. Since processing in the system is carried out using the presentation of text independent of the language, the method corresponding to the invention allows any language to be processed. Texts in new languages are converted into a language-independent representation by linking the new syntax layer to the language processing system. Thus, the method can be used quite universally.

Each language attached to the language processing system can be represented in a separate syntactic layer. In this way, syntax questions can be configured completely independently of each other.

In addition, individual languages may have common parts of the syntax layer. For example, High German, Swiss German, and Austrian German have much similar grammar. Only a few rules differ. In this case, the syntax layer may have a part that is common to several languages, and may have parts specific to a particular language. This shortens the work of changing the rules for individual languages and facilitates the input of data used in the syntax layer. Thus, language abstractions can be reused for individual syntactic layers.

Language-independent information can be retrieved in the relational layer. In general, language-independent information contains objects, actions, and attributes, as well as their relationships. Objects are usually represented by nouns in languages such as German, English, or Chinese. Actions are generally described by verbs of text. Adjectives can also represent action. For example, two companies can be tagged with “compete” or “being competitive”. Attributes can be attributes of sensations, such as color, temperature, size or quality, as well as attributes such as emotions. These objects, actions and attributes are extracted from the text by the syntactic and relational layer and transferred to the semantic layer.

At the stage of generating a graph or graphical representation, the objects, actions and attributes of a sentence or phrase of a text are connected with each other and presented in the form of a graph or graphically. A graphical representation (for example, in the form of a matrix or matrices) facilitates the processing of text in a language processing system. Although graphs can also be represented graphically, a purely graphical representation (not a graph, such as video or pictograms) can be more powerful because it provides greater flexibility with respect to presentation capabilities.

To achieve linguistic independence of the language processing system, objects, actions and attributes can be represented graphically or through pictograms. For example, a car can be represented by a car icon, a bench can be represented by a bench icon, the attribute “green” can be a green area, “to give” can be represented by pictograms of a person transferring an object to another person, or through video, and “to bark” can be represented by sound. Thus, everyone can understand the graphical representation of semantics without tagging objects, actions, or attributes in terms of a particular language.

The text processing step may include the step of understanding the extracted semantics of the text. This can be done by comparing the extracted semantics with the model or by determining the distance between the entities involved.

The central part of the method may be the semantic world. The semantic world represents the object world. The main objective of the object world is to present objects that are usually represented by nouns in languages such as German, English or Chinese. It consists of several spaces from two to n in size containing objects (or their prototypes), and organizes them into meaningful combinations.

Objects of the object world can be organized using structural trees or structural networks that logically link individual objects. It was found that people organize knowledge about the objects of the world and their relationships into a meaningful structure. This organization is not uniform. They use concepts and categories to store and sort information. Such categorization may exist for electronic devices (e.g., computers, printers and digital phones), paper (e.g. letters, documents and invoices), buildings (e.g. houses, museums and offices), etc. Individual category objects may be associated with other categories. For example, an office building has several rooms equipped with furniture, electronic devices, papers, etc. Furniture may contain tables, chairs or bookshelves. On the other hand, the chair can be an office chair, as well as a rocking chair. Both are chairs, but serve for completely different purposes. Thus, individual words are related to each other in categories.

The sense world additionally contains an action space that is responsible for representing actions. Actions can be associated with any other unit in the semantic world, for example, a unit tagged with the English word "withdraw" can be associated with objects "person", "money" and "cashpoint", which are participating actors. Such bonds are called molecules.

In addition, the semantic world may contain an attribute space that contains the attributes of the elements. The vast majority of attributes can be quantified in some natural way. The attributes of sensations, such as color, taste, size, or pressure, have a representation of one to three dimensions used in various contexts. Colors, for example, can be reproduced using a color spindle, which is specified by the hue, saturation and brightness of the desired color. Emotions can also be set using a multidimensional representation. According to the model proposed by psychologists, a six- or eight-dimensional emotional simplex can be used to combine all human emotions. Thus, emotions can also be represented regardless of language.

The language representation of the text may be ambiguous. For example, in the sentence "the chicken is ready to eat" the word "chicken" can be interpreted either as a eater or as a dish to be eaten. In the sentence "we saw the man with the telescope", either a person or "we" use a telescope. These ambiguities are resolved based on the context of the proposal. This context can be drawn from the semantic world. If the previous sentences relate to agriculture, the chicken is most likely a feeder. When previous suggestions are related to cooking, the chicken is most likely a meal. These contextual issues can be addressed with the help of the semantic world.

An ambiguous text will correspond to several graphs or graphic representations, and the number of representations is equal to the number of values that can be extracted from the text. Using the semantic world, one can determine the representation that is most likely to be true.

In a language processing system, there may be a relational layer that connects the syntactic layers and the semantic layer. This relational layer may contain abstractions about the possible relationships between objects in the layers. The relational layer receives the information output by the syntactic layer, and performs additional generalization and abstraction.

According to one embodiment of the invention, the method can be used in a translation system. In this case, the processing step comprises the step of generating a translation of the text into a language other than the original language of the text. Since a graph or graphical representation is language independent, it can serve as the basis for translation into any language. When implementing the steps of the method, the source text is first analyzed in relation to its syntax and morphology. The text components and their relationships are then extracted, and this information is used to generate a graph or graphical representation of the text, or to use an existing graph or graphical representation as a language-independent representation. After an optional semantic check, a language-independent representation is converted to a text representation. This conversion step can be carried out by the syntactic layer, since this layer uses the syntactic and morphological information of the target language. Thanks to the modular configuration of the system, it is theoretically possible to translate from each language into any other language. Since there is a language-independent platform between them, each language simply needs to be associated with a language-independent representation. Thus, a case is possible when dictionaries connecting individual languages with each other are not required. This greatly facilitates the development of an automatic translation system.

According to another embodiment of the invention, the method can be used for search, since it can significantly improve the results of search engines. A user enters a query on a search engine web page. This query is parsed syntactically and morphologically, resulting in the extraction of the text components and their relationships. This information is used to generate an internal graphical representation of the request. Ambiguities can be defined and resolved. In addition, moving away from the line approach, individual query words can be generalized using abstractions in structural trees. Thus, you can improve the quality of the results.

According to another embodiment of the invention, the method can be used in text analysis. It allows you to extract topics presented in the text. This can be used to automatically categorize text. In addition, it can be used to find logical chains or information about semantic structures in the text.

According to another embodiment of the invention, the method can be used to generate responses to text input into a language processing system. For example, a system may automatically generate a response to a request sent by a user requesting support. Unlike the methods known in the art, this method allows you to analyze and "understand" the text, and create an appropriate response to the request using the knowledge presented in the model of the semantic world.

Other embodiments are possible. Since the method provides for the presentation of the text, regardless of the language, the processing step can be replaced by a large number of other steps. Thus, the invention can be used very universally. In addition, if desired, the individual embodiments described herein may be arbitrarily combined.

According to each embodiment, the text generated in the processing step can be displayed to the user as a written or spoken language or as an image. If the processing step includes a text analysis step, the output may also contain statistics or a list of topics or search input.

To improve and facilitate the construction of the databases used in the system, the knowledge necessary at the stages of the method corresponding to the invention can be entered via the web interface. Knowledge may include lexicon tags, content of the model of the semantic world, grammatical information, presentation of attributes, etc. This information can be entered by an open group of users contributing information in a mode convenient for the user.

With respect to a language processing system for processing a natural language and, according to the invention, for solving the aforementioned problem, a method is proposed comprising the features of clause 16. Preferred embodiments of the invention are described in dependent claims 17-27.

To solve the above problems, a method for developing a language processing system according to claim 28 and its embodiments described in the dependent claims 29 and 30 are proposed.

There are several ways to preferentially build and further develop the concept of the present invention. To do this, refer to the claims subordinate to paragraphs 1, 16 or 28, on the one hand, and to the following explanation of a preferred example of an embodiment of the invention illustrated in the drawings, on the other hand. In connection with the explanation of a preferred example of an embodiment of the invention using the drawings, a preferred embodiment and further development of the concept will be generally explained. In the drawings:

FIG. 1 is an illustrative logical block diagram of a traditional translation process;

FIG. 2 is an illustrative conceptual diagram of an embodiment of language pairs used in the traditional translation process shown in FIG. one,

FIG. 3 - structure of the translation process,

FIG. 4 is a graphical representation of the proposal,

FIG. 5 - possible representation of actions,

FIG. 6 - representations of temperature (part a)) and emotions (part b)),

FIG. 7 is an illustrative logical block diagram of the translation process based on the semantic world,

FIG. 8 is an illustrative conceptual diagram of an embodiment of language pairs used in a translation process based on the semantic world shown in FIG. 7,

FIG. 9 is an illustrative block diagram of an embodiment of a translation system based on the semantic world,

FIG. 10 is an illustrative block diagram of an embodiment of a sense world system shown in FIG. 9,

FIG. 11 is an illustrative logical block diagram of an embodiment of a process for adding a language to add a language to a semantic world-based translation system shown in FIG. 9,

FIG. 12A is an illustrative flowchart of an embodiment of a process for adding a term to add a term to a language dictionary in a translation system based on the semantic world shown in FIG. 9,

FIG. 12B is an illustrative flowchart of another embodiment of a process for adding a term to add a term to a language dictionary in a translation system based on the semantic world shown in FIG. 9,

FIG. 13 is an illustrative block diagram of an embodiment of a system including one or more translation servers capable of implementing a semantic-based translation system shown in FIG. 9,

FIG. 14 is an illustrative block diagram of an embodiment of a computing system capable of implementing one or more components of the electronic system described herein.

In FIG. 3 shows an example of a translation process according to embodiments of the invention. Consider the translation of the sentence "die grüne Bank steht im Wald" using the invention. In FIG. 3 shows semantic layer 2, which is the core of the language processing system 1. The semantic layer is embedded in the relational layer 3. Several syntax layers 4, 5, 6 are attached to this relational layer 3. Each syntax layer represents a language: syntax layer 4 represents German, syntax layer 5 represents English, and syntax layer 6 represents Polish.

The text input into the language processing system 1 is supplied to the syntax layer 4. The syntax layer 4 analyzes the text with respect to its grammar and syntax. You can understand that "Bank" is the subject of the proposal. "Bank" has the attribute "grün". "Bank" carries out the action "stehen" and it takes place in "Wald". This information is obtained by syntactic and morphological analysis of the text.

The components of the text and their relationships can be extracted. This can be used to generate a universal representation of a sentence independent of the language, which is shown in FIG. 4 in the form of a graph. This graph can be converted to English or any other available language. At the first stage, the pictogram representing "Bank" is translated by the English word "bench". The bench attribute “grün” is translated with the word “green”, the action “stehen” is translated with the word “is”, and the representation “Wald” is tagged with the word “forest”. As a result of the translation of the presentation into English, the sentence "the green bench is in the forest" is obtained.

As follows from the above example, by tagging a graphical representation using another language and placing the word in grammatically correct order, each language can serve as the source and destination language, respectively.

In FIG. Figure 5 presents several possible actions carried out by the person depicted in the center. Showing actions of "think," "sit," "go," and "give."

In FIG. 6 illustrates illustrative representations of two attributes. 6a shows a temperature scale and associated attributes. In the general case, these representations are fuzzy and cannot reflect a specific meaning. Warm dishes will feel cold if they have a temperature of 10ºC or less. They will be designated as lukewarm at a temperature of 20ºC. A temperature of 70ºC will feel like hot.

In FIG. 6b shows a 4-dimensional space for representing emotions. The emotions that can be imagined are a superposition of the simplex: “scared,” “surprised,” “happy,” and “angry.” The emotion presented here is a point or region in this 4-dimensional space.

In many embodiments, relationships, models, syntactic requirements, etc., may form part of one or more interacting computer processes. Accordingly, according to embodiments of the present disclosure, a computer system employs a language-independent object world, thereby providing a central site for translations. According to an embodiment, the text or speech is translated from the source language into a language-independent interpretation before translating this representation into one or more final languages for output.

For example, the language systems of the present disclosure provide a mapping of the syntax and semantics of the input language, for example, into a graph in the semantic world, independent of the language. From this representation, independent of the language, you can make a translation into any language or several languages. According to an embodiment, a language independent graph may also be associated with and / or be the output of a graphical or multimedia presentation. The corresponding translation process analyzes the input text (or speech) in relation to its syntax and morphology, extracts the text components and their relationships, generates a text graph as a representation of the meaning of the text, which is independent of the language, and processes the text using the graph.

Such a system, in general, is consistent with the results of neurological studies. One major result includes an understanding of the fact that a person’s cognitive ability distinguishes between syntax and semantics. Let us return to the case considered above with an umbrella. This separation of syntax and semantics (or language-dependent information and language-independent information) is part of the translation process proposed in this disclosure. For example, when analyzing a text with respect to its syntax and morphology, the grammatical structure is analyzed. This provides a basic understanding of the text. The content of this text is retrieved. For example, the text, in the General case, includes sentences that may contain subject, addition and predicate. According to an embodiment, each component can be extracted and its role in the proposal can be revealed. These components and their relationships are used when converting text to graph. The components form the vertices of the graph and the relation between the components is represented by edges. According to an embodiment, this graph can be made partially or completely independent of the language used to input (or output) the text. The graph mainly includes semantic information that can be directly used for further processing.

During the analysis of the text, the system refers to the grammar information of the language used. Each language includes its own special grammar, which sets the rules for word placement. Another aspect of this disclosure provides a relatively direct, non-technical way of generating these grammar rules. So that users can, with little or no programming knowledge, prescribe grammar rules, grammar data can be entered using the grammar editor. Grammar rules include a certain formalization of the possible structures of a given language. Thus, you can avoid the lengthy development of various grammars for each individual language or reduce it and instead get the opportunity to quickly and efficiently create prototypes. Thus, it is possible to integrate new languages faster and easier into the language processing system disclosed here.

According to an embodiment, the text is analyzed by an analyzer operating in the syntax layer of the language processing system. In one aspect of the disclosure, the language processing system may have a modular configuration that provides reusability, adaptability and extensibility of the system. The analyzer can segment and tokenize text. Segmentation means breaking the text into sentences, and tokenization means identifying specific word forms in a sentence. After segmentation and tokenization, the elements, as well as their relationship in the proposal, can be analyzed with respect to syntax and morphology.

To improve modularity and to provide a process that can be universally used, syntax layer objects can be associated with a language processing system. This makes it possible to easily integrate different languages into the language processing system, adding a new analyzer and dictionary for each language. Since processing in the system is carried out using a representation of text that is independent of the language, any language can be processed. Texts in new languages are converted to a language-independent representation, and then pairing with any other existing language can be used for conversion. Thus, the process can be used quite universally.

Each language attached to the language processing system can be represented as a separate set of objects of the syntactic layer. In this way, syntax questions can be configured independently of one another. In addition, it is possible that individual languages may have common objects of the syntactic layer, for example, the analyzer or the main parts of the analyzer. For example, High German, Swiss German, and Austrian German have many common grammar rules. Only a few rules differ. In this case, a single analyzer can operate with each language, despite the fact that most of the rules are common for several languages and some rules depend on the language. This reduces the work of changing the rules for individual languages.

Language-independent information is retrieved in this syntax and / or optional relational layer. In general, language-independent information contains objects, actions, and attributes, as well as their relationships. Objects are usually represented by nouns in languages such as German, English, or Chinese. Actions are generally described by verbs of text. Adjectives can also represent action. For example, two companies can be tagged with the words “compete” or “competitive”. Attributes can be attributes of sensations, such as color, temperature, size or quality, as well as attributes such as emotions. These objects, actions and attributes are extracted from the text by objects of the syntactic or relational layer and are translated into representations in the semantic world (which are called the semantic layer here).

At the stage of processing the proposal, the objects, actions and attributes of the proposal or phrases of the text are associated with each other and presented in the form of a graph. A graph facilitates text processing in a language processing system, since graphs can be easily represented as matrices.

To ensure the linguistic independence of the system of the semantic world, objects, actions and attributes can be represented on the basis of a unique ID. However, it is preferable that each representation of the term in the semantic world also has an image or illustration of meaning to simplify work with the semantic world (especially for people who do not have programming skills). For example, a car can be represented by a car icon, a bench can be represented by a bench icon, the attribute “green” can be a green area, and “give” can be represented by pictograms of a person transferring an object to another person. Thus, graphs and objects of the semantic world can be understood by means of graphic representations, without tagging objects, actions or attributes in terms of a specific language.

The text processing step may include the step of understanding the extracted semantics of the text. This can be done by comparing the extracted semantics with the model of the semantic world or by determining the distance between the participating objects of the semantic world. "Distance" in this case indicates the relative relationship between different objects in the semantic world. The closer the objects are to each other, the closer they are connected or correlate more strongly. The more closely connected many objects in the semantic world, the more likely that the translation will be correct.

The semantic world contains language-dependent term objects (“LIT objects”). LIT objects are primarily intended to represent objects that are usually represented by nouns in languages such as German, English, or Chinese. It consists of several spaces ranging in size from 2 to n (2-n) containing objects (or their prototypes), and arranges them into meaningful combinations. Other parts of speech, such as verbs, can also be represented by objects.

In one aspect, the disclosure provides systems and processes for providing translation systems. In this case, a graph or other semantic representation of the input text does not depend on the language and can serve as the basis for translation into any language. In general, process steps include analyzing the source text with respect to its syntax and morphology using text components and their relationships to generate a graph of text as a language-independent representation. After an optional semantic check, a language-independent representation is converted to a text representation in the final language (s). This conversion step can be carried out by the syntactic layer, since this layer already includes the syntactic and morphological information of the final language (s). It is theoretically possible to translate from each language to any other language with minimal additional complication when each new language is added to the system. Since there is a language-independent platform between them, each language simply needs to be adapted to a language-independent representation. Thus, dictionaries connecting individual languages with each other are not required (in contrast to the above-described model corresponding to the prior art). This facilitates the development of an automatic translation system.

According to another aspect, the disclosure may provide a process that improves search in a search engine. For example, a user enters a query on a search engine web page. This query is analyzed with respect to its syntax and morphology, as a result of which the components of the text and their relationships are extracted. This information is used to generate an internal query graph. Through the use of a model of the semantic world, independent of language, ambiguities can be defined and resolved. In addition, moving away from the string approach to search queries, individual query words can be generalized using abstractions, for example, in structural trees, and the relationship between words. Thus, you can improve the quality of the results.

According to another aspect, the disclosure provides a process for analyzing the text and extracting information, for example, about the subject of the text. This can be used to automatically categorize text. In addition, it can be used to find logical chains or information about semantic structures in the text.

According to each embodiment, the text generated in the processing step can be displayed to the user as a written or spoken language or as an image. If the processing step includes a text analysis step, the output may also contain statistics or a list of topics or input for searching or other processing.

To improve and facilitate the construction of data structures, databases and representations used in the system, the knowledge used in the steps of the process corresponding to the invention can be entered via the web interface. Knowledge may include vocabulary tags, content of the semantic world, grammatical information, presentation of attributes, etc. This information can be entered by an open group of users entering information through an interface convenient for the user, rather than an interface adapted for programming.

To facilitate a complete understanding of the invention, in the remainder of the detailed description, the invention is described with reference to the drawings provided with an end-to-end notation.

In contrast to the disadvantages associated with FIG. 1 and 2, in FIG. 7 shows an illustrative flowchart of a semantic world-based translation process 300 according to embodiments of the present disclosure. According to FIG. 7, at block 310, text of any length, such as a sentence or paragraph, is entered into an electronic translation system according to the embodiments disclosed herein. For example, at block 310, you can enter "The boy is running to the park". At block 312, the system analyzes the sentence to extract the root form of key text concepts. Usually this will be at least the subject, predicate and, sometimes, the complement of the sentence. In the presented example, there are three key terms: (1) boy; (2) run; and (3) park. These terms are translated into the language-independent graph “semantic world” (Block 314). According to an embodiment, the graph includes a vertex for each of the key conceptual terms and an edge to illustrate their relationship with other terms. Key concepts are translated into the selected language (Block 316). In this example, the final language is German: (1) Junge; (2) laufen; and (3) Park. The language-dependent analyzer module converts the sentence using the appropriate articles, verb forms, etc. (Block 318), and the completed sentence "Der Junge läuft zum Park" is displayed to the user (Block 320).

Although in a simplified example, the process 300 shown in FIG. 7 illustrates the basic concepts of the semantic world and its graphic properties. Based on this disclosure, a specialist can understand for himself many more complex input scenarios in a natural language, and, as will be discussed below, representation in the semantic world provides considerable flexibility and power for solving such scenarios.

In FIG. 8 shows an illustrative conceptual diagram of language pairs according to the present disclosure. For example, as shown in FIG. 8, the four languages shown in FIG. 2, use four (4) language pairs, not six (6). In addition, the inclusion of an additional language - Italian, uses an additional language pair. Thus, for five languages, according to the present disclosure, five (5) language pairs are required as opposed to ten (10), as shown in FIG. 2.

Thus, according to FIG. 8, each language is attached to the central semantic world, and not to any other specific language. This provides a modular approach to the translation system, since no language needs to be tied to any other language individually. In addition, systems and processes consistent with this disclosure lead to a significantly less complex system, which is also, in general, much cheaper to develop than traditional systems.

Embodiments of the present system may, for example, be particularly useful in an international Internet chat or instant messaging session. An embodiment of the disclosed system can be used in the server instant messaging system, and each message can be translated into the preferred language of the individual end users when transmitting messages. There are many other uses of the disclosed translation system, which will be discussed in more detail below.

Embodiments of the translation system are distinguished by a modular design for multilingual processing of natural languages and multimodal interaction. Modules designed for different languages, and others that are independent of the language, can be combined into a working system that can analyze, reason, search, translate and generate a natural language. Embodiments of the system provide multimodal interaction: input and output of written and spoken natural language, as well as output as a language, speech, description or a combination thereof. In addition, the modules are preferably designed to be reusable by other programs. Whenever possible, the modules are language independent, which facilitates reuse. Strictly defined interfaces and common interface programs control the communication between system components. Thanks to this design, you can translate from every language into any other language. Languages to be translated can also be variations within a single language, for example, translation from Swiss German to High German or translation from colloquial style to formal style can be carried out. Signs of various embodiments may advantageously include, in whole or in part, the following:

modularity: ease of operation, reusability, configurability;

web orientation: accessibility from everywhere;

ergonomically-developed software: universal use;

social basis: the possibility of universal expansion;

universality: each language can be integrated; and

visual-graphic core: does not depend on language and is cognitively adequate.

Embodiments of this disclosure model and mimic human cognitive processing to optimize understanding of a natural language and the generation, translation, search engines, or other communication tasks.

The approach described here, based on the cognitive ability of a person, separates syntax and semantics, according to the processes of the human brain, and allows us to distinguish between multiple meanings of words. Special components operate on syntax rules or language-dependent word forms. Semantics is processed in a layer independent of the language, called the semantic world. This approach is based on the latest neurological findings. According to the example of the umbrella discussed above, the concept of "umbrella" when communicating is an object tagged with a word, depending on the language. Participants know the object without using the language. If they want to leave the house when it rains outside, they activate the “tag” through a language-specific dictionary for communicating with other people: “May I take this umbrella?” or "Könnte ich diesen Schirm nehmen?"

This helps to explain the advantage of this approach to language processing, which consists in the fact that the meaning is represented by human means and thus does not depend on the language. You can add all natural languages, because they use the same semantic world. This approach may not only be useful for translation, but it may also have many other useful uses. According to an embodiment of the semantic world, information can be added, processed and stored without the need for special language syntax. When an information unit is present in the semantic world, new languages can very easily be added by linking the syntactic representation to a unit independent of the language.

In FIG. 9 shows an illustrative block diagram of an embodiment of a translation system based on the semantic world. For example, FIG. 9 includes system components or modules that can be used to provide the disclosed language processing and translation system. System 522 of the semantic world includes representations of concepts that are independent of language. The semantic world system 522 also provides certain users with access to multimedia for visualizing or voicing representations of the terms and concepts stored therein. As shown, each semantic language system 524 is associated with a central semantic world 522. This is achieved using one or more linguistic tools 526. Semantic language systems 524 also contain one or more language dictionaries 528. Each language represented in the translation system usually has its own own dictionary 528 to provide specific terms for this language. Dictionary articles are associated with specific objects in the system 522 of the semantic world. However, in some cases, languages can be related enough to be able to share all or all of the linguistic tools 526. For example, different dialects can be represented as different languages, but in the general case they obey similar syntax rules that determine, for example, sentence structure and word order. In this case, a single linguistic toolkit can analyze each language based on all or almost all of the general rules of grammar. According to an embodiment, the linguistic toolkit provides an analyzer (parser) 530 for extracting terms from sentences to be translated, as well as formulating grammatical sentences from graphs of the object world. The analyzer 530 relies on grammar rules 532, classes 534 for changing the form of a word, templates 536, etc., for the correct construction and analysis of sentences in the corresponding language. Patterns 536 of linguistic tools help provide an immediate extension of terms in a language dictionary for building or modifying a language in a system. For example, templates can provide sentence fragments that help classify new terms correctly. In particular, if the user wants to add the word "tiger" to the language dictionary, for example, templates may be presented to him that help the system understand parts of speech or the coordinate system. A very simple example of many patterns would be “A tiger”, “I tiger” and “the tiger ball”. The user can choose one of them for application, and the system can learn to classify a new term. In this case, the system learns that "tiger" is a noun that can take the indefinite article, rather than a verb or an adjective, respectively. Similarly, a system can provide patterns for determining whether a verb conjugates correctly or incorrectly. Through this process, anyone can expand the system without the need for knowledge of linguistics or knowledge of other system languages.

As described above, analyzer 530 is a component for translating to or from graphs of the semantic world. However, in another embodiment, the relational processor 527 couples the semantic system 524 and the analyzer 530 to the semantic world system 522. In such an embodiment, the semantic system can generate an input text graph that is still associated with the source language. This graph can be further abstracted into its language-independent form using a relational processor 527. The relational processor 527 can extract verb tense, information from a suggested group, and other sentence details to help organize or expand a language-independent graph. For example, a relational processor may indicate a “definite article”, “a long verb form” or “directional information” for the example shown in FIG. 7. In various embodiments, as is apparent to one skilled in the art, the analyzer 530 and the relational processor 527 can be implemented as one or more modules, act together and simultaneously or individually and sequentially, and can share responsibility in any way other than described here. Those skilled in the art can also understand from this disclosure that other configurations may exist that provide the same or almost the same functions.

In FIG. 10 shows an embodiment of data structures representing the semantic world 522. In general, each LIT 638 object represents a specific language-independent term, for example, the terms "building", "room", "city", "house" and "office building "shown in the drawing. According to an embodiment, each object is a data structure including an object ID 640, a plurality of one or more relational relationships 644 and an optional set of one or more hierarchical relationships 646. An object ID 640 may be a number or code identifying a computer record in a computer where the object is stored, but, in the general case, will be unrecognizable to the user. According to an embodiment, LIT 638 objects are associated with other related terms through relational relationships 644. As shown, “city” and “building” are related because the city includes a plurality of buildings; likewise, building has a collection of rooms, so the two objects are interconnected. According to an embodiment, these relational relationships 644 may be weighted to indicate the strength of the relationships. Similarly, objects between which there is a class-type-subtype relationship can be connected by hierarchical relationships 646 and can form a tree structure. In FIG. 10 this relationship is illustrated by the objects "building", "house" and "office building". "Building" is a general term encompassing the concepts of "house" and "office building" as more specific types of buildings. Although not shown, the "house" object itself can be associated, for example, with the subtypes "cottage", "ranch" and "townhouse".

LIT objects 638 may also include vocabulary links 648. In addition, vocabulary objects 528 include relationships from specific language terms 650 to the corresponding LIT 638 object. For example, in FIG. 10 shows that the term “Bâtiment” is from object 526 of the French dictionary, the term “building” from the dictionary of English and other included languages is associated with object 638 “building”. Similarly, the terms “office building” in English, “Bürogebäude” in German, and “immeuble de bureaux” in French are associated with object 638 “office building”.

Each object may also include or be associated with one or more media presentations, such as visual presentation 642. Visual representations 642 can be used to illustrate the corresponding term in various situations. This is especially useful for helping users add a new language to the system, as they can be displayed so that the user can understand what term they should be associated with in the new language dictionary 528. In some embodiments, audio files, video files, image files, etc. For example, “to whistle” is better associated with a sound file or sound file and image than with a visual representation.

Attributes of objects can also be linked in the system 522 of the object world and can have special relational relationships. For example, an attribute may be an emotional scale, color representation, or physical attributes, such as temperature, size, or quality. Relational relationships allow you to arrange specific terms along the scale, which allows you to associate the appropriate terminology with specific or relative values along the scale. For example, "tiny", "small", "regular", "large", "huge", "enormous" and "infinite" can be placed on the size scale. The attribute space itself can be multidimensional. Attributes can also be represented as a structural tree, for example, "scarlet", "carmine" and "crimson" are subtypes of "red". Thus, the units of the semantic world are connected with each other in various ways on the network, which allows us to make complex inferences necessary for processing a natural language.

Furthermore, according to an embodiment, the sense world system 522 can be represented as a virtual world or a plurality of virtual worlds. For example, you can provide a user interface that allows the user to navigate the virtual representation of the system 522 of the semantic world. The user can, for example, first see a “city” consisting of “buildings”, and can zoom in to any particular building, for example, “home” or “office building”. After that, the user can enter the “room” of the house, and in each room there may be objects representing other terms, such as “sofa”, “chair”, “bed”, “table”, etc. Each object can also be tagged to display language representations of terms that are tied to vocabulary relationships 648 of this LIT 638 object. The virtual world can also include representations of people and actions, as well as modifying attributes. Thus, when moving to the "door" object in the virtual world, not only the "door" tag of the English language can be displayed, but also the color tag "red", the tag "wooden", etc.

In another embodiment, multiple "worlds" may exist and there may be connections between them. Immediate objects can open up new worlds. For example, a room scene may depict a window through which the moon is visible. By clicking on the image of the moon, you can go to another object world, which is dedicated to the cosmos. A person’s representation can lead to an object world that mimics the cellular basis or parts of a person’s body. If the user moves to the office in the office building, there may be a presentation of papers on the table, going to the papers (for example, clicking on them with the mouse), you can open the tree-like interface showing object 638 connected by hierarchical connections 646. For example, “paper” may refer to the words “advertisement”, “report”, “periodic”, etc. In turn, the word “periodic” may be associated with the words “newspaper” and “magazine”, etc.

Moving around such a world can be a useful tool for self-education, since the user can, optionally, view language tags in any languages connected to the semantic world. According to an embodiment, the user can optionally browse tags for a language that he wants to know, which helps him learn that language. Similarly, according to an embodiment, the terms of the primary language of the user and another language can be displayed so that the user can correlate them with a visual representation.

The various relationships between objects 638 in the semantic world 522 can also be modeled graphically. Spatial, temporal, causal, or metaphorical relationships between objects 638 of the semantic world (as well as other types of relationships) are ideal for graphical descriptions. In relation to translation, these types of relations are the basis for determining which structures and formulations serve for their verbal expression, since languages differ in the way they express these relations: some languages use prepositions, others realize them as morphemes attached to a noun, etc. It is best to generate adequate structures and formulations based on a neutral, abstract, and graphical representation. Thanks to this process, the generation component does not have to carry out a comprehensive restructuring of the input structure (as classical machine translation systems do), but just need to choose between the available structures of the final language using the mapping of relations into structures.

Returning to the translation process shown in FIG. 7, we explain additional details regarding the system 522 of the semantic world. When choosing the right translations, knowledge of the topics covered improves the translation by filtering ambiguous values that are not relevant to these topics. Topics can often be recognized from the relationship of the graphical representation of the translation text (see Block 314). For a number of topics, there are a large number of clusters in the N-dimensional semantic space. Efficient and fast clustering algorithms, for example, the k-means clustering algorithm, are used to find cluster centers. These cluster centroids represent text topics.

If there are ambiguous translations, the topic can be used to resolve them. For example, the input text may include "The dog was a Siberian Husky." The term "dog" actually has multiple meanings, including "a domesticated canine", "a despicable man" or "an iron bar driven into a stone or timber to provide a means of lifting it". Each of these definitions can have different translations into other languages and, thus, are also ambiguous for the system 522 of the object world. However, the context of the proposal allows you to select the correct object of the object world for use (one that corresponds to "a domesticated canine"), since other objects of the proposal, in particular "Siberian Husky", are more closely connected in the semantic world of objects with this object than with others. In principle, the Siberian Husky and the correct dog object will appear, for example, in a subset of the object world associated with animals or pets.

Text parsing often creates many syntactic graphs and some unresolved relationships between the vertices of the graph, as in the "dog" example above. A statistical approach to choosing the best graph is generally used to distinguish between ambiguities: Bayes theorem, according to an embodiment. Bayes's theorem states that the probability that a certain graph provides information (semantic entities) is proportional to the likelihood that semantic entities are present in this graph, multiplied by the a priori probability of the presence of the entity in this graph. Other algorithms and statistical analysis — known as or derived from standard statistical principles — can also be used to disambiguate translations from the source language to the interpretation of the LIT object, and are familiar to those skilled in the art.

One aspect of an embodiment of the disclosed language processing system provides a set of tools to help a user edit languages or add new ones. As explained earlier, users can navigate the virtual world, which helps to represent the terms included in the semantic world, independent of the language. It would be very helpful for multiple stakeholders to help add new language terms, correct incorrect terms, and even add new languages to the semantic world. Although experienced programmers can do all this, it would be faster and easier if a large number of users had the opportunity to develop and adjust the semantic world. This type of group activity has already been illustrated by the wiki- movement and websites such as Wikipedia. According to an embodiment, selected qualified users, such as linguists, language teachers, etc., could add languages or edit existing ones; on the other hand, you can allow any interested user to add and edit the language.

According to an embodiment, the system creates a certain formalization of possible language structures. Thus, you can avoid the long development of various grammars for each individual language and instead get the opportunity to quickly and efficiently create prototypes. Thus, you can quickly and easily connect new languages. Grammars are used for both language analysis and component generation.

Additionally, graphical user interfaces, which may be called Lexi-Wikis, allow users to enter words into language-specific dictionaries 528. Lexi-Wikis are designed for a wide range of users. From the corresponding words, the tools generate sample sentences that the user can easily select or modify. What forms and how many word forms need to be presented to the user is determined by various language-dependent inflection algorithms. The user-selected examples are translated into a complex representation that the program can process. According to an embodiment, the underlying morphological process uses linguistic knowledge and frequency information to determine the minimum information that the user must provide. Thus, he predicts the most likely word forms, so that the user is required as few word forms and as few actions as possible. Through this process, mental workload or intelligence is transferred from the user to the software side.

Very often, users are not able to position semantic entities absolutely, but they are quite able to recognize the difference from other semantic entities. Multidimensional scaling is used, an algorithm designed to place multidimensional points on the basis of a difference matrix, which is a matrix containing distances (or difference) to other semantic entities. These algorithms can be fuzzy, which is necessary because no two people will ever choose exactly the same distance. However, they can come to an agreement in general terms (for example, “far” or “very close”).

In FIG. 11 describes the process of adding a language to the system of the semantic world. At block 760, a user logs on to the system. In some embodiments, a user can optionally enter a special “user modification” mode that can prevent illegal or accidental system changes. In the menu, the user can choose to add a new language (Block 762). When creating a new language, a syntax analyzer must be created to parse and generate sentences. According to an embodiment, the system includes template rules that the user can select as necessary (Block 764). For example, a rule may indicate that adjectives usually change the noun that comes after them (for example, in English), or that they usually change the noun that stands before them (for example, in French). The change rules may explain exceptions to these rules, etc. After creating the analyzer, the terms can be added to the dictionary of the new language (Block 766). Each term is associated with objects of the semantic world (Block 768).

The binding of new terms can be done by any means including the processes described with reference to FIG. 12A and 12B. In one embodiment, the user logs into the system (Block 670) and chooses to add a term to a specific language dictionary (Block 872), for example, through a menu system. The user can enter a term (Block 874). The system may provide template requests to help ensure that the context is used correctly (Block 876). For example, templates can help the system categorize terms according to speech, correct or incorrect conjugation of verbs, etc. Answers can also help provide a specific context of the semantic world for directing the user to the appropriate semantic world or the region of the semantic world where the term’s object is independent of the language . The user can also browse the virtual semantic world (Block 878) and select the virtual representation with which the new dictionary entry should be associated (Block 880).

In an alternate embodiment of FIG. 12B shows another process by which a user can add words to dictionaries. According to FIG. 12B, the user enters the system (Block 882). The user can view the semantic world (Block 884). Selecting objects in a specific context (Block 886) may display terms associated with that object, for example, showing a leader in the virtual world. When a user selects an object in the absence of associated terms, the user can add a term to “tag” the object (Block 888). Similarly, the user can change tags to adjust or replenish language dictionaries. For example, a representation of the saxophone in the semantic world can be tagged with the term “instrument” or “musical instrument” in the English dictionary. The user can edit the tag to show a more precise term in the hierarchy by adding "saxophone".

In addition, external resources can be connected to the system so that knowledge representations available on the Internet, in public or private databases, etc., can be used in the components of the language system. Connected resources may include, for example, DBpedia, Wiktionary, Open Street Map, scientific taxonomies, ontologies from Semantic Web®, custom user taxonomies, etc. According to an embodiment, the consistency checking components verify the consistency of various representations and provide the correct calculation for potentially heterogeneous sources of knowledge. You can even integrate different types of media, such as graphics, video and audio.

In FIG. 13 shows an embodiment of a basic translation system and means of access thereto. Although such a translation system can take many forms, a network-based translation system can provide easy access for a large number of interested users. For example, in a computing system 994, for example, a program code can be stored, in whole or in part, on a server, which when executed provides some or all of the functions of the semantic world system 522, including semantic language systems 524. Server 994 can communicate electronically with a public or private, local or global network 992, such as the Internet. In turn, various users can electronically communicate with a computer translation system using other network devices 990a, 990b. Suitable user devices include personal computers, laptop computers, telephones connected to a data network, or other mobile devices (e.g., Blackberry® devices, Apple iPhone® devices, other PDAs, mobile phones, etc.). Users can access the translation system via a web interface using a browser or, in some embodiments, a special program installed on a user device.

One user can use a personal computer 990b to access the translation service, enter text for translation, select the source and destination languages, and receive the appropriate translated text, as described with reference to FIG. 7. At the same time, another user can edit and add languages to the system through a different interface on his 990a computer. Preferably, the translation system is scalable, so that multiple users can access the system at any given time. With this approach, multiple users can simultaneously try to edit languages. In a preferred embodiment, the translation system may provide a locking mechanism that allows only one user to edit, for example, a specific entry in dictionary 528 or a specific object LIT 638 of the semantic world at any given time.

Although this disclosure is mainly devoted to the use of the semantic world and the possibilities of analyzing the language to create a translation system, there are other options for using such a system of the semantic world. For example, the system of the semantic world allows you to expand the capabilities of the search engine. The embodiment of the system described here allows you to analyze a search query in a natural language, for example, a sentence or a question. The system can extract key terms and generate graphs and / or graphical equivalents that are language independent. Since these language-independent objects 638 are also associated with the corresponding LIT objects, the user's direct request can be expanded to include similar words, other word forms, semantically related words, etc.

For example, a user enters a query on a search engine web page: "What recent court decisions define qualifying income tax?" This query is analyzed in relation to its syntax and morphology, as a result of which the components of the text and their relations are extracted: "court", "decision", "define", "qualifying" and "income tax". Then this information can be used to generate a query graph, as when performing a translation. However, these terms or objects 638 of the semantic world per se may not provide all thematic results. Therefore, the relational and hierarchical relationships 644, 646 of the object world can be used to expand search terms. For example, “decision” can be associated with the terms “opinion” and “order”. Similarly, “court” can be associated with “judge”, and “income tax” can be associated with “IRS”. These additional terms can then be used to expand the final search. In this way, you can improve the quality of the results without forcing the user to expand their terminology or perform multiple search sessions.

Another aspect of this disclosure may be used to analyze text. To identify the topics presented in the text, you can use analyzers. This can be used to automatically categorize text. In addition, it can be used to find logical chains or information about semantic structures in the text. As part of the previous example, this text analysis can be performed using a search agent program that attempts to categorize new web pages for search purposes. When analyzing text, the output may contain statistics or a list of topics that can be used to tag web pages for a search query. Similarly, a similar system can be used in libraries to categorize new books, periodicals, articles, etc. to generate catalogs of thematic cards and search databases.

Similarly, identifying topics helps expand your search queries in the right environment. For example, the “decision” from the aforementioned search query can also be associated with “choice”, and “court” can be associated with “basketball” or “tennis”. Expanding the search with these terms will obviously expand the search into the wrong space of the object world. Thus, determining that the topic is “taxes” and / or “legal” can help the search engine expand terms in the appropriate context.

Other embodiments are possible. Since the disclosure provides a language-independent presentation of the text, the processing of this representation can take many different forms. Thus, the disclosed system can be used in a variety of applications. In addition, the various embodiments described herein may be combined in a desired manner.

In FIG. 13 shows a block diagram of one embodiment of a computing system 994 that can be used to implement some of the systems and processes described herein. For example, in one embodiment, the computing system 994 may be adapted to receive translation requests from another computer system (eg, user PC 990a, 990b), use the implementation of the semantic world to translate the request into the desired language, and return the translation. The functions provided by the components and modules of the 994 computing system can be combined in fewer components and modules or, conversely, distributed among additional components and modules.

Computing system 994 includes, for example, a server or personal computer compatible with IBM, Macintosh, Linux / Unix, and the like. In one embodiment, the computing device comprises, for example, a server, a laptop computer, a cell phone, a personal digital assistant, a kiosk, or an audio player. In one embodiment, exemplary computing system 994 includes a central processing unit (“CPU”) 1095, which may include a conventional microprocessor. Computing system 994 also includes memory 1097, such as random access memory ("RAM") for temporarily storing information and read-only memory ("ROM") for permanently storing information, as well as mass storage device 1098, such as a hard disk, diskette, or optical Memory device. Typically, modules of a 994 computing system are connected to a computer using a standard bus system. In various embodiments, for example, the Peripheral Component Interconnect (PCI), MicroChannel, SCSI, Industrial Standard Architecture (ISA), and Extended ISA (EISA) architecture can serve as a standard bus system.

Computing system 994 is generally managed and coordinated by an operating system such as Windows 95, Windows 98, Windows NT, Windows 2000, Windows XP, Windows Vista, Linux, SunOS, Solaris, or other compatible operating systems. On Macintosh systems, the operating system may be any available operating system, such as MAC OS X. In other embodiments, the computing system 994 may operate under a specialized operating system. Conventional operating systems control and dispatch computer processes for performing, performing memory operations, providing a file system, network functions, and I / O services and provide a user interface such as a graphical user interface ("GUI"), among others.

Illustrative Computing System 994 includes one or more publicly available devices and I / O interfaces 1096, such as a keyboard, mouse, touchpad, modem, Ethernet card, microphone, and / or printer. In one embodiment, devices and input / output interfaces 1096 include one or more display devices, such as a monitor, that allows you to visually present data to a user. In particular, the display device provides a representation of, for example, a GUI, application data, and multimedia presentations. Computing system 994 may also include one or more multimedia devices 1099, such as speakers, graphics cards, graphics accelerators, and microphones. According to an embodiment, the user enters the text to be translated or processed using the keyboard or touch panel representing the keyboard (input device 1096). On the other hand, a microphone (another input device 1096) perceives spoken text. Oral text can be stored in any of a variety of audio formats, such as WAV, MP3, or other formats. CPU 1095 can process this audio text and convert it to written text, such as an inline data object, plain text data file, Microsoft® Word document, etc.

In the embodiment shown in FIG. 13, 1096 devices and I / O interfaces provide a communication interface with various external devices. According to an embodiment, the computing system 994 is connected to a network 992, such as a LAN, WAN, or the Internet (see FIG. 13) via a wired, wireless, or combo link. Network 992 communicates with various computing devices and / or other electronic devices via wired or wireless communication lines. In the illustrative embodiment shown in FIG. 13, the network 992 is connected to one or more user terminals or computing devices 990a, 990b. Computing device 990b may transmit text input in audio or written text formats to computer system 994 for processing. In addition to the devices illustrated in FIG. 13, the network 992 may communicate with other data sources or other computing devices. In addition, data sources may include one or more internal and / or external data sources. In some embodiments, one or more databases or data sources can be implemented using a relational database, such as Sybase, Oracle, CodeBase, and Microsoft® SQL Server, as well as other types of databases, such as databases with a two-dimensional file, entity databases, relationships and an object-oriented database and / or record-based database.

According to the embodiment shown in FIG. 14, the computing system 994 also includes an application module that may be executed by the CPU 1095. According to the embodiment shown in FIG. 13, the application module manages the semantic world models and data. This module may include, for example, components, such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, subprocedures, program code segments, drivers, firmware , microcode, electronic circuits, data, databases, data structures, tables, arrays and variables.

In general, the word “module” used here refers to logic implemented by hardware and software or hardware, or to a set of software instructions, possibly having entry and exit points, written in a programming language, for example, Java, Lua, C, or C ++ . A program module can be compiled and linked into an executable program installed in a dynamically loaded library, or it can be written in an interpreted programming language, for example, BASIC, Perl or Python. Obviously, program modules can be called from other modules or from themselves and / or launched in response to registered events or interrupts. Software instructions can be wired in non-volatile memory, such as EPROM. It is also apparent that hardware modules may consist of connected logic devices, such as gates and triggers, and / or may consist of programmable devices, such as programmable gate arrays or processors. The modules described herein are preferably implemented as software modules, but can also be implemented in hardware or software / hardware. In general, the modules described here relate to logical modules that can be combined with other modules or divided into submodules regardless of their physical organization or storage.

According to another illustrative embodiment of the invention, the following are some basic features of the invention. They should be considered as an example illustrating the principles of the invention.

In the future, we will call the language processing system "Lingupedia", which is a trademark of Lingupedia Investments Sàri, Luxembourg. Lingupedia is a modular system for automated text translation.

Lingupedia uses a fully modular design for multilingual natural language processing and multimodal interaction. Modules of any type can be combined into a working system capable of analyzing, reasoning, searching, translating and generating a natural language. The system provides multimodal interaction: input and output of written and spoken natural language, as well as output as a language, speech, description or a combination thereof. The modules are designed to be reusable by various other programs, either in the Lingupedia system, for example, analysis and generation programs, or other programs. Whenever possible, the modules are language independent, thus providing reusability. Strictly defined interfaces and common interface programs control the communication between system components. Thanks to this design, you can translate from every language into any other language. Languages to be translated can even be variations within a single language, for example, translation from Swiss German to High German or translation from colloquial style to formal style can be carried out. Lingupedia has the following basic features:

• modularity: ease of operation, reusability, configurability

• web orientation: accessibility from anywhere

• ergonomically advanced software: versatile use

• social foundation: universal expansion

• versatility: each language can be integrated

• visual-graphic core: does not depend on language and is cognitively adequate

The basic principle of the Lingupedia system is an approach to modeling and simulating human cognitive processing to optimize the understanding of natural language and the generation, translation, search engines or other communication tasks.

Most algorithms are based on a spelling form, which simply means symbolism or a byte string without any meaning. Even ontologies use this “home is building” approach, sometimes with mathematical distances or spaces, but they always use these meaningless byte chains. The main disadvantage of byte chains is that they often really have multiple meanings, which may not make any sense at all: a dog may be a pet, a grab hook, a cramp iron, ...

An approach based on a person’s cognitive ability like Lingupedia clearly separates syntax and semantics, according to the processes of the human brain, and allows you to distinguish between multiple meanings of words. Special components operate on syntax rules or language-dependent word forms. Semantics is processed in a language-independent layer, Lingupedia Meaning World (LMW). This approach is based on the latest neurological findings. Example: if several people speaking different languages are in the same room where there is an umbrella, everyone “knows” that it is an umbrella. But this "knowledge" does not mean that the word "umbrella" is activated in any way in the brain of the person present. For communication purposes only, the umbrella object is tagged with a language-dependent word. Participants know the object without using the language. If they want to leave the house when it is raining outside, they activate the “tag” through a language-specific dictionary, but only to communicate with other people: “May I take this umbrella?” or "Könnte ich diesen Schirm nehmen?"

An advantage of the Lingupedia approach is that meaning is represented by human means and thus is language independent. Thus, it is possible to add all natural languages, since they use the same semantic world. Lingupedia declares: our approach allows us not only to translate, but also to optimize any work or program that uses a natural language. In the basic component of LMW, information can be added, processed and stored without the need for special language syntax. When an information unit is present in LMW, new languages can very easily be added by linking the syntactic representation to a unit independent of the language.

In addition, custom values can be stored regardless of the language: for example, a company producing a special printer can remove this printer from the template of this printer in LMW, adapt the output printer to specific parts and tag it with one or several languages. Images with descriptions of parts and features can be easily derived from LMW for a given language. Thus, multilingual product information (e.g., documentation, marketing information, or bug reports) can be automatically generated from a semantic world that is language independent. Communication with the consumer in various languages and in various forms (e-mail, letters, phone calls) can be automated, i.e. analyze, interpret, distribute to different departments and generate for the response to the consumer.

In addition to this CRM (Customer Relationship Management) application, LMW is useful as a fast and efficient information retrieval machine because, as far as mental representation is concerned, it is closer to the representation of human knowledge than other approaches. The Lingupedia method surpasses both the classical line search (requiring exact matches at the level of the spelling form) and the recently proposed semantic network search (requiring special annotation of texts in which information is supposed to be searched).

The work of LMW is based on the tagging method. To simplify navigation on LMW, you can activate the so-called tagging, depending on the language. If, for example, English tagging is activated and the user navigates to the umbrella, the tagging algorithm queries the English dictionary for an article that displays it for the user. In this way, users who speak another language receive help finding the right information.

Knowledge in LMW is represented by different worlds. In general, the “object world”, “structural trees / networks”, “action space” and “attribute space” are used.

The main objective of the object world is to present objects that are usually represented by nouns in languages such as German, English or Chinese. It consists of several spaces with dimensions from two to n containing objects (or their prototypes) and organizes them into meaningful combinations.

These objects are organized into structural trees or networks. People organize knowledge about the objects of the world and their relationships into a meaningful structure. This organization is not uniform. They use concepts and categories to store and sort information. Such grouping by categories can exist for “electronic devices” (computer, printer, digital telephone) or “papers” (letters, documents, invoices).

Part of the LMW action space is responsible for representing actions. Actions can be associated with any other unit in LMW, for example, a unit tagged with the English word “withdraw” or the German word “abheben” can be associated with the objects “person”, “money” and “cashpoint”, which are participating actors. Actions are optionally expressed in verbs: a molecule, through an action connected, for example, with two companies, can be tagged with "compete" or "being competitor". Such bonds are called molecules.

The attribute space is also structured directly with respect to utility. The vast majority of attributes can be quantified in some natural way. Attributes of sensations, such as color, taste, size or pressure already have a 1-3-dimensional representation used in various contexts.

Lingupedia also includes the following additional views and algorithms:

• integration of external knowledge representation resources

• descriptions for the natural representation of units in LMW

• graphic relationships between units in LMW

• clustering by topics in texts

• statistical analysis for disambiguation

• multidimensional scaling to calculate similarities

In addition to the components described above, the following parts of Lingupedia exist for modeling specific natural languages:

• language-independent grammar editor for defining grammars for each language

• Lexi-Wikis for determining the words of each language

• Dictionary with multi-purpose configurability

Typically, dictionaries do not provide an exact meaning: the dictionary gives the following English-German translations of the word "dog": Anschlag, Bauklammer, Finger, Gerüstklammer, Greifhaken, Hund and German-English translations of the word Hund: canine, dog, hound. Thus, several different meanings are suggested for one word. LMW can intelligently distinguish between these values. This means that first of all there are language-independent representations of meaning, for example, a woolly animal that is tagged in English with the word “dog”, or a special part of a goat that is also tagged with the same spelling form “dog” (in German) Gerüstklammer "). Thus, if a regular dictionary has 30,000 entries in English, LMW will require about 100,000 representations of meaning. A language-independent meaning can be understood from the context: is this object tagged with the word dog used in the construction field, or is it associated with an action tagged with the verb bark or walk? After clarifying the meaning, by finding the right unit in LMW, translation or further processing can be done better than in any existing system.

Lingupedia can a) use and integrate existing external resources from the network and b) open all Lingupedia components for public access - language-dependent syntax components, as well as the language-independent semantic domain LMW. In addition, right from the beginning, and even more with the development of LMW, it is very easy to integrate the syntactic part of the new language, since it requires only simple tagging. Lingupedia provides linguistic tools for quick and easy tagging for inexperienced users who do not require special knowledge, and encompassing every human language.

Below will be described some details of the components.

In the object world, all semantic entities are represented regardless of language. The representation is graphic, i.e. visualized in different forms. Semantic entities correspond to abstract or real objects that exist in a kind of "model" world. They are organized into spaces from two to n in dimension and into meaningful structures.

Simple objects can open up new worlds. For example, the moon can lead to another space, such as an orbit. Alternatively, a person’s view can lead into a space that mimics the cellular basis or parts of a person’s body. Relationships of objects can be represented, for example, in a city where there are buildings, parks and gardens. The building may be a private, public or office building. This building contains offices; Offices contain objects such as desks, computers, shelves, clocks or papers. Thus, objects are spatially or functionally related to the field of knowledge represented by the office or building. Objects may consist of parts; watches, for example, may consist of a mechanical mechanism and a dial with hour and minute hands. The Euclidean distance between objects in this model world represents the difference between two objects. Euclidean distance in semantic space is not equivalent to Euclidean distance in the real world. It is based on distinction or functional proximity.

LMW uses associative networks or oriented trees as a representation of knowledge. The user can switch from an object, for example, a document object, lying on a table in the graphic world, to the corresponding tree for searching, for example, a notification object. Each object can be associated with multiple structural trees, for example, a paper object with “paper”, as well as with the tree “materials”, where the vertices “tree”, “metal”, etc. are on the same tier.

In LMW, there are different types of relationships in networks: one type of relationship can be “is-a”. In this case, the object is a subtype of the parent vertex containing the supertype. Subtypes inherit the properties of their supertypes. Multiple inheritance is possible. This "is" tree is used to translate subtypes that do not have tagging in the target language. Instead of a specific term, a more general supertype is verbalized ("document" instead of "letter", "take" instead of "withdraw"), or a synonym or denied expression of the antonym is selected. In addition to hyponymy, other relations are also used for deduction and translation: affinity, connectedness, instanceism, membership, framework connectedness, likeness, synonymy, antonymy, meronymy etc. Languages are distinguished by their lexical inventory. This network of relationships provides the flexibility to generate a natural language in a system that must operate with each language, and where certain languages lack certain words, either due to language features or because they are not yet tagged in the Lingupedia system. LMW units can be artificial in the sense that they are only part of structural trees. This is true for some relationships or tops built.

In particular, with regard to actions, a visualized representation is cognitively adequate, since a verbalized definition is difficult to understand and less intuitive for users than a visual one. Films, graphics showing movements or schematic descriptions are used to illustrate various actions. The action space is also used to represent the thematic roles of verbs or other entities. Thematic roles relate the action to its agents, themes, goals, etc. Roles are either user-defined or inferred from the properties of the graphically displayed action. This is an elegant and intuitive way of assigning an internal, thematic structure to actions and events. This knowledge of the roles involved is used to disambiguate and to correctly generate the final sentence.

Attributes of objects can be an emotional scale, color representation, or physical attributes, such as temperature, size, or quality. For example, the word “exchange” can be associated with a two-dimensional space that represents a numerical scale representing monetary units. Other units can be associated with this space, for example, actions that can be tagged with the words “growth” or “decline”. Like the action space, the attribute space can be associated with other units in LMW. A color can be associated with an object that is tagged with the word "car." The attribute space itself can be multidimensional. Attributes can represent a structural tree, for example "scarlet", "carmine" and "crimson" are subtypes of "red". Thus, the units of the semantic world are connected with each other in various ways on the network, which allows us to make complex inferences necessary for processing a natural language.

External resources can be connected to the system so that knowledge representations available on the Internet can be used in system components. Connected resources are, for example, DBpedia, Wiktionary, Open Street Map, scientific taxonomies, ontologies from Semantic Web®, custom user taxonomies, etc. The highly intelligent consistency checking components verify the consistency of various representations and provide the correct calculation for a wide range of heterogeneous sources of knowledge. You can even integrate different types of media, such as graphics, video and audio. Different interpretation or translation algorithms allow you to operate with different types of representations.

Avatars represent humans or animals. Avatars, like all objects in LMW, are derived from other avatars. In this way, an internal hierarchy of objects formed by inference is defined. Human thinking works - this is the hypothesis of the authors - and in the final models of the world: when people imagine withdrawing cash from an ATM, they do not use the words "I, bank, cashpoint, withdraw". Instead, they use a “mental image” or “mental scene” that is independent of the language to imagine this process. They can even come up with whole stories, for example, in dreams, without the participation of their body. They imagine their body in an artificial environment simulated in the brain. In the future, LMW will serve as a platform for such modeling using artificial intelligence.

Different relationships are also modeled graphically. Spatial, temporal, causal, or metaphorical relationships between entities (as well as other types of relationships) are ideal for graphical descriptions. In relation to translation, these types of relations are the basis for determining which structures and formulations serve for their verbal expression, since languages differ in the way they express these relations: some languages use prepositions, others realize them as morphemes attached to a noun, etc. It is best, and cognitively adequate, to generate adequate structures and formulations based on a neutral, abstract and graphic representation. According to this method, the generation component does not have to carry out a comprehensive restructuring of the input structure (as classical machine translation systems do), but just need to choose between the available structures of the final language using the mapping of relations into structures. These mapping algorithms were developed to generate each integrable language.

Knowing about topics improves translation by filtering ambiguous values that are not relevant to these topics. For a number of topics, there are a large number of clusters in the N-dimensional semantic space. Efficient and fast clustering algorithms, for example, the K-mean clustering algorithm, are used to find cluster centers. These cluster centroids represent text topics. If there are ambiguous translations, the topic can be used to resolve them.

Text parsing often creates many syntactic graphs and some unresolved relationships between the vertices of the graph. A statistical approach to choosing the best graph is used: Bayes theorem. She argues that the likelihood that a certain graph provides information (semantic entities) is proportional to the likelihood that semantic entities are present in this graph, multiplied by the a priori probability of the entity's presence in this graph.

The Lingupedia system includes the world's first language-independent grammar editor: users can prescribe grammars without any programming knowledge. Only a certain formalization of the possible structures of the language is required. This way you can avoid the lengthy development of different grammars for each individual language and instead get the opportunity to quickly and efficiently create prototypes. This way you can quickly and easily connect new languages. Grammars are used for both language analysis and component generation. This concept of modularity and reusability of components applies to the following syntactic representations and processes:

• language independent, ie universal, abstract representation of grammatical structure

• grammars for analysis and generation

• syntactic-morphological rules for analysis and generation

Graphical user interfaces called Lexi-Wikis allow users to enter words in a language-specific vocabulary. Lexi-Wikis does not require any expert knowledge of the language, but is intended for a wide range of users. From the corresponding words, the tools generate sample sentences that the user can easily select or modify. What forms and how many word forms need to be presented to the user is determined by various language-dependent inflection algorithms. The user-selected examples are translated into a complex representation that the program can process. The underlying morphological method uses linguistic knowledge and frequency information to determine the minimum information that the user must provide. Thus, he predicts the most likely word forms, so that the user is required as few word forms and as few actions as possible. According to this method, mental load or intelligence is transferred from the user to the software side.

The vocabulary method is designed as a universal, multi-purpose main dictionary for any variety of natural language applications and for any type of language. The dictionary offers a new level of presentation: the level of phrases, which is located between individual words and complete sentences. This allows you to very flexibly operate the language units in the continuum of the word - phrase - sentence. Verbose expressions, which are still the main problem for most natural language systems, can be represented in the form of a more or less fixed structure: from completely unchanged and unmodifiable (having a fixed form and having no internal structure) to having an internal structure with certain restrictions (semantic, syntactic, lexical, pragmatic, stylistic, etc.) and up to providing modifications of any type.

The vocabulary method provides a mechanism for annotating articles with features useful for various natural language applications: morphological features for morphological analysis and generation, syntactical features for parsing and generation, semantic features for semantic processing, pragmatic features for pragmatic processing, and dialog features for effectively building dialogs on natural language. To explain a feature-based method: processing a natural language using its surface forms (strings) is not ideal, since each variant and equivalent or related form must be processed separately. This approach is not effective: time-consuming and error-prone for the programmer and not giving the user any flexibility when interacting with the software, for example, the possibility of dialogue: he should use only those lines that the program prepared; otherwise it will not be understood. Thanks to the use of features, a higher level of scientific abstraction is applied, which provides a more flexible and natural character of interaction.

In addition to the linguistic information used for interaction in the written language, the dictionary also stores information on the pronunciation of words, which is useful for sound input and output in speech recognition and synthesis. Built-in conversion algorithms. They translate the presentation of pronunciation from the internal form to another form for further processing by software of various types or for presentation to the user. Thus, this information can be flexibly used in different applications. The configuration tool allows you to select exactly those parts of the dictionary that are required by different applications.

Thanks to the storage of basic forms, instead of full forms (the latter are usually used by speech-related software), the dictionary uses an effective and flexible form of presentation and processing and allows the dynamic generation of all kinds of inflective, derivative and complex forms. The generation algorithm, which creates various word forms, while guaranteeing the correct pronunciation, derived from and adapted to the internal structure of words, is part of the system. The dictionary also provides a way of representing the various relationships between vocabulary entries. Relationships relate to different tasks of language processing, for example, for abbreviation, which is usually not used in speech, but only in written language. If it is not used in speech synthesis, its full form is presented to make it pronounced. Alternatively, if an article is to be searched using a search engine, its various spelling and inflective forms are indifferent to the search process, although they still have to be explicitly presented. Lingupedia's approach allows you to connect with them and easily find them.

Based on the principles set forth in the foregoing description and the annexed drawings, one of ordinary skill in the art to which the invention relates may provide various modifications and other embodiments of the invention set forth herein. Thus, it should be understood that the invention is not limited to the particular embodiments disclosed, and that modifications and other embodiments are to be included within the scope of the claims. Although specific terms are used here, they are used only in a general and descriptive sense, but not for purposes of limitation.

Claims (29)

1. A method of processing a natural language using a language processing system in which written or spoken text is entered into a language processing system, comprising stages in which:
analyze the text in relation to its syntax and morphology;
extract text components and their relationships;
generate or use a graph or graphical representation of the text as a language-independent representation of the meaning of the text; and
the text is processed using a graph or graphical representation, and at the processing stage, a judgment is formed on the above representation in the model of the semantic world, thus checking the consistency of the extracted text semantics. 2. The method according to claim 1, in which a visual-graphic model of the text is created, the visual-graphic model being independent of the language, which allows users to expand the language processing system without knowledge of the languages used. 3. The method according to claim 1 or 2, in which the grammatical data used in the analysis step is entered into the language processing system by means of a grammar editor independent of the language. 4. The method according to claim 1, in which the analysis step is carried out by the syntactic layer of the language processing system, which performs the segmentation and tokenization of the text, and the syntax layer can be attached to the language processing system. 5. The method according to claim 4, in which each language to be processed by the language processing system is represented in a separate syntactic layer, which allows you to reuse the abstractions of other languages in separate syntactic layers. 6. The method according to claim 4 or 5, in which additional abstraction and generalization of the data generated by the syntactic layer is carried out by the relational layer, the data preferably describing the relationship between objects and their abstractions. 7. The method according to claim 4, in which the text information independent of the language is extracted in the syntax and relational layers, the information independent of the language is transmitted to the semantic layer, and the information independent of the language contains objects, actions and attributes. 8. The method according to claim 1, in which at the stage of generating a graph or graphical representation of the objects, actions and attributes of the sentence or phrase are connected with each other. 9. The method of claim 8, wherein the objects, actions, and attributes are represented graphically. 10. The method according to claim 1, in which at the processing stage generate a translation of the text into a language different from the original language of the text, and the graph or graphical representation is the basis of the translation. 11. The method according to claim 1, in which at the processing stage, the text is analyzed for search or other language processing. 12. The method according to claim 10, in which at the processing stage generate a response to the text using the information specified in the model of the semantic world. 13. The method according to claim 1, in which the text generated at the processing stage is displayed to the user as a written or spoken language or as an image. 14. The method according to claim 1, in which the knowledge used in the individual steps is entered using a web interface for universal use, and the knowledge may include lexicon tags, content of the semantic world model, grammatical information, and presentation of attributes. 15. A system for processing natural language, containing:
a language-independent kernel module, this module manipulating a collection of objects representing terms and relationships between objects,
a collection of vocabulary modules depending on the language, each dictionary module having a collection of articles, each article of dictionaries associated with one of the collection of objects stored in a kernel module, independent of the language,
a text analyzer associated with one or more of the language-specific vocabulary modules, and
a sentence generator associated with one or more of the language-dependent dictionaries,
in which the text analyzer accepts input, extracts key terms from the input, uses the representation of key terms in the form of a graph based on a set of objects from a language module independent of the kernel module, and forms a judgment on the above representation in the semantic model of the world, thus checking the consistency of the extracted semantics mentioned input, and the sentence generator formulates the output text in one of the languages associated with it on the basis of a graph representation. 16. The system of clause 15, in which the input contains written text or spoken text. 17. The system according to clause 15, containing at least one linguistic syntax module associated with one or more of the dictionaries, depending on the language, and the linguistic syntax module includes the above text analyzer, many grammar rules and many templates. 18. The system of claim 17, wherein each language-dependent dictionary is associated with a separate linguistic syntax module. 19. The system of claim 17, wherein the language-dependent dictionaries of at least two closely related languages are associated with the same linguistic syntax module. 20. The system of clause 15, in which a multimedia presentation of the corresponding terms stored in this module is additionally stored in a language-independent kernel module. 21. The system of claim 20, wherein the multimedia presentations comprise images, sounds, or videos. 22. The system of claim 20, further comprising an editing component module, wherein the editing component module provides for changing dictionary entries and their relationships with one of a plurality of objects stored in a kernel module that is language independent. 23. The system of claim 22, wherein the editing component module is further configured to provide addition of articles to a set of language-dependent dictionaries. 24. The system of claim 22, wherein the editing component module is adapted to access it through a website. 25. The system according to paragraph 24, in which multimedia representations of terms can be displayed in the virtual world. 26. The system of claim 22, wherein access to the editing component module is limited to qualified users. 27. A method for developing a language processing system, comprising the steps of:
developing a core independent of the language, the core containing language terms objects, each language terms object containing multimedia representations of language terms and the relationships between the corresponding language terms;
add a dictionary object associated with a specific language;
add the words of a particular language to the dictionary object;
associate words with appropriate language terminology objects in the kernel; and
provide a web interface for universal use, and the web interface is used to enter the knowledge used in the operation of the language processing system, and the knowledge includes vocabulary tags, the contents of the model of the semantic world, grammatical information and the presentation of attributes. 28. The method according to item 27, in which the relationship between the corresponding terms of the language include relational relationships and hierarchical relationships. 29. The method according to item 27, containing the stage of creating a language analyzer for a particular language based on the rules of grammar and syntax.

Images