JP2002215661A - Interface knowledge response system in natural language - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a man-machine interface system in natural language, leading out a certain reply from data (knowledge data) already stored as knowledge, to reply to a sentence inputted in natural language. SOLUTION: This interface knowledge response system in natural language wherein a computer makes a language analysis to the input sentence inputted in natural language and performs a certain reply, data accumulation and execution, has (1) a means for making a morphological analysis, a syntactic analysis and a semantic analysis to the sentence inputted in. a natural language form, (2) a means for storing the input sentence as a knowledge sentence in a semantic model database (a knowledge model database) when the input sentence is information data on knowledge (a declarative sentence), (3) a means for temporarily storing the input sentence as a question sentence in the semantic model database when the input sentence is a question (the question sentence, verifying a meaning and outputting a certain reply sentence, and (4) a means for temporarily storing the input sentence as an instruction sentence in the semantic model database when the input sentence is an instruction (an instruction sentence), verifying a meaning and executing.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a knowledge response system that enables a man-machine interface in a natural language by allowing a computer to recognize a sentence in a sentence input in a natural language.

[0002]

2. Description of the Related Art A conversation between a human and a machine (computer) is performed by a limited number of commands. For example, if "PRINT FILE-A" is entered, the computer prints the contents of the file LIST-A. It uses the human-readable word PRINT, but the terms that follow are formalized. In order to use a computer, humans need to learn instructions that the computer can understand. That raises the barrier between humans and computers. In order to lower this barrier, menu-based and interactive man-machine interfaces have been devised and used in various fields. However, these methods also have many problems, such as having to learn operation procedures and taking a lot of time for system development.

[0003] If a computer can understand the words spoken by humans on a daily basis, the barriers existing between humans and computers as described above are removed. From such a thing,
In recent years, efforts have been made to make computers understand natural languages (words and languages that humans use every day).

In order to make a computer understand a natural language, analysis methods such as phonological analysis, morphological analysis, syntax analysis, and semantic analysis are being studied. Phonological analysis is speech recognition analysis at the time of speech input, and morpheme, syntax, and semantic analysis are speech input,
It is used to analyze words regardless of character input.

The word segmentation method of morphological analysis includes a longest match method, a character type cutting method, a minimum number of clauses method, and a CYK method. In the longest match method, a given character string is scanned in the order of input (such as English or Japanese, from right to left, but documents are ordered from right), and the longest match in the dictionary is found. It is commonly used in Japanese word processors because of its good word matching rate (hit rate). For example, "we wipe kimono here" is divided into clauses here, "/ kimono / wiping", and is converted to "take kimono here". However, this conversion is not perfect, too, and it can be separated from / footwear / wipe here. In this case, "Take off your footwear here"
Is converted to In the current Japanese word processor, the learning function also works on this phrase, and the conversion from the next time is performed on the phrase separated by the user. A higher hit rate is achieved by combining the longest match method and the phrase learning function.

The syntax analysis is an analysis method for recognizing a relationship between words from a morphological analysis result and generating a corresponding tree structure. Grammar plays an important role in this analysis, and the analysis differs depending on the language. Currently, parsing is basically performed according to grammatical rules written in the form of phrase structure rules.

[0007] Parsing alone is very ambiguous. In the case of English, parsing by unification of feature structure is used,
Although it is effective, it is not always an effective analytical method because the meaning of Japanese is ambiguous. For example, in the noun phrase "big girl with dark eyes"
It is said that more than 10 interpretations are possible. Generally speaking, the word "girl" can be interpreted as a "girl", but it can also be interpreted in the document as "a child owned (fostered) by a woman (usually a woman)." Such ambiguity is difficult to judge only by syntactic analysis, and requires semantic analysis.

In the semantic analysis, the meaning of a sentence is assembled using the meaning of each word and the result of the syntax analysis. The resulting meaning is usually described by a predicate logical expression. This resolves the ambiguity of the parsing. For example, "read" has a different meaning between "read a book" and "read your hand." The former is "read"
This is the original usage, but the latter means "guessing." Such meaning ambiguity is determined by co-occurring words. In the above example, the co-occurring words are "book",
"Hand". Co-occurrence dictionaries register the relationship between such co-occurring words and related words.

If a computer can understand a natural language,
The distance between humans and computers is much closer. For these reasons, natural language man-machine interfaces have been studied in various fields. A number of patents in this area are also filed.

As a syntax analysis, "Syntax analysis system"
(JP-A-10-91628). Here, a method is proposed in which a clause of the co-occurrence relation data having the highest frequency of the dependency relation among the co-occurrence relation data is recognized as a modification destination clause, thereby obtaining a more correct modification destination.

The semantic analysis is described in “Semantic Analysis Method and Apparatus and Medium Recording Semantic Analysis Program” (JP-A-2000-2000).
-148755). Here, the processing speed is increased by specifying the deep case necessary for the subsequent keyword extraction and generating the minimum necessary range of semantic structure without performing the semantic analysis inside the unnecessary unnecessary deep case. Advocating a way to As a co-occurrence dictionary,
"Co-occurrence dictionary construction / update method and co-occurrence / semantic analysis method" (JP-A-5-19712). Here, we propose a method of constructing a co-occurrence dictionary and a method of quantifying the validity of interpretation from the degree of each co-occurrence for semantic analysis using the co-occurrence dictionary.

[0012] On the other hand, the ones dealing with numerical expressions are:
There is a "quantity / time expression processing method" (JP-A-7-200583). Here, for the quantity / time expression of the input sentence (natural language description sentence), information about the criterion and range is extracted, and information about the quantity / time is complemented, and the content of the meaning complementation is evaluated based on a certain criterion. It proposes a method of determining the possibility of an error and inquiring the user about the content.

Even if syntax analysis and semantic analysis are performed, it is impossible to completely eliminate ambiguity. The ambiguity of Japanese, in particular, can be attributed to those who use the language (especially when speaking politically), but it exists as a structural problem. The more modifiers, the more ambiguity,
It is also a Japanese feature. As a method of evaluating this ambiguity, there is “Ambiguous search index information creating apparatus and ambiguous search index information creating method” (Japanese Patent Laid-Open No. 8-129559). Here, the time-series data is time-divided and evaluated for each period, the evaluation for each period is averaged, and the averaged evaluation result is reflected in the fuzzy search index information.
A method of performing the latest evaluation has been proposed.

[0014]

As seen in the prior art, computer processing using natural language has been studied in various fields and from various angles. However, there is no complete product at present. In one US lab, 10
It is said that there are places where more than 100,000 words are registered in the dictionary, but it is still not possible for natural conversation between humans and machines. It is so difficult to make a computer understand human words.

On the other hand, in the field of word processors and translations, natural language has been successfully applied to a considerable extent, if not completely. Document creation by voice input is also being put to practical use, although to a limited extent. Conversely, it has become possible to output an input document by voice. These are also one of the research results of natural language processing.

At present, in the document / document search performed on the Internet or the like, a search using a keyword (word) is mainly used. The logical relations used here are AND and OR
This makes it possible to narrow or broaden the search range. If you can search in natural language, you can search data more finely, but there are no successful cases so far.

In order for a human to talk to a machine, an interface using a natural language is required. In order to answer a question in a sentence with a sentence, it is necessary to understand the sentence, create a sentence, and answer it. However, answer sentences are not always prepared. How to overcome such obstacles remains as a future task.

[0018] In order to solve such a problem, the problem to be solved by the present invention is to solve a sentence input in a natural language.
An object of the present invention is to realize a natural language man-machine interface system that derives some answers from data (knowledge data) already stored as knowledge and gives answers.

[0019]

According to a first aspect of the present invention, a sentence is input in a natural language, a computer performs a language analysis on the input sentence, and a certain answer is input. In a knowledge response system that performs data storage and execution, (1) means for performing morphological analysis, syntax analysis, and semantic analysis on a sentence input in natural language form,
(2) When the input sentence is information data relating to knowledge (declaration sentence), means for storing it as a knowledge sentence in a semantic model database (knowledge model database). (3) When the input sentence is a question (question sentence). Means for temporarily storing in the semantic model database as a question sentence, verifying the meaning, and outputting some answer sentence. (4) When the input sentence is a command (command sentence), A natural language interface knowledge response system having means for temporarily storing in a database, verifying the meaning, and executing the means.

According to a second aspect of the present invention, the semantic analysis means of the first aspect uses a dictionary in which synonyms, derivatives, honorifics, etc. are registered based on the result of the syntax analysis, and the meaning of the input sentence is defined. Means for abstracting the input sentence to the extent that does not change, and means for performing data search on the abstracted sentence from the knowledge model database and outputting the search result as an answer sentence Natural language interface knowledge response system provided.

According to a third aspect of the present invention, in the semantic analysis means of the first aspect, based on the result of the syntax analysis, the input sentence includes a word including a numerical value, a quantity, or a logical relationship; If it is found that the above relational expression can be used, a symbol that abstracts the input sentence from another angle to the extent that the meaning does not change from the numerical value or the logical relation related to the item included in the knowledge model database A natural language interface knowledge response system including processing means (symbol processing rules) and means for tailoring and outputting an abstract sentence into an answer sentence.

The invention described in claim 4 is the first invention.
In the means for extracting answers to questions, after morphological analysis, syntactic analysis, and semantic analysis, if knowledge data corresponding to the question sentence is searched from the knowledge model database, the knowledge data is used as an answer. A natural language interface knowledge response system comprising means for sequentially comparing the abstracted knowledge data of claims 2 and 3 with a question sentence to derive the most suitable answer sentence and outputting it.

The invention described in claim 5 is the first invention.
In the means to elicit answers to questions, after searching for knowledge data corresponding to the question sentence from the knowledge model database after morphological analysis, syntax analysis, and semantic analysis, if the appropriate answer is not obtained, Means for requesting knowledge data required for answers (consultation means),
A natural language interface knowledge response system equipped with

According to a sixth aspect of the present invention, there is provided a natural language interface knowledge response system including a means for applying an answer to the above-mentioned question as a natural language instruction to a search engine or a robot on the Internet. I do.

According to the present invention, there is provided a morphological analysis, a syntax analysis, and a method for a sentence input in a natural language.
It performs semantic analysis, generates an output sentence through semantic verification (response model), and outputs an answer. For the question text, the knowledge data stored as knowledge is collated with the question text to elicit an answer. However, in many cases, there is often no answer corresponding to the question. In such a case, the range of the search is expanded by abstracting the question sentence, and the search of the knowledge data and the creation of the answer sentence are performed within a range where the meaning of the input sentence is not lost. When the input sentence is a declarative sentence (knowledge data), the sentence is stored as it is or is symbolized and stored in the knowledge model database as knowledge data, and is used as knowledge data when extracting a question answer later. Specifically, the above-mentioned search and answer are enabled by the following method.

In a knowledge response system in which a natural language is input and a knowledge response system performs a language analysis on the input sentence and gives some answer, the knowledge response system performs morphological analysis, syntax analysis, and semantic analysis on a sentence input in a natural language. Do. Next, when the input sentence is information data on knowledge (declaration sentence), it is stored in the knowledge model database as a knowledge sentence. When the input sentence is a question (question sentence), the sentence is temporarily stored in the knowledge model database as a question sentence, the meaning is verified, and some answer sentence is output. If the input sentence is a command (command sentence), the system executes the specified command.

It is an object of the present invention not only to realize a natural language man-machine interface, but also to respond to a question in some form. For example, for a question sentence entered in natural language, it is sufficient if knowledge data corresponding to the question is in the knowledge model database, but even if it is not, the answer of the question is not the answer "No corresponding answer" Even if the answer is not the answer itself, find an answer in a range that does not lose the gist of the question and respond, or ask for a re-question (when the meaning of the question cannot be understood) or give a consultation explained later. Therefore, if there is no appropriate answer in the knowledge model database, the input sentence is parsed, and then, using a dictionary in which synonyms, derived words, honorifics, etc. are registered, the input sentence is not changed in meaning within the original sentence. Abstract. An answer corresponding to the abstracted sentence is retrieved from the knowledge model database and output as an answer sentence.

The data registered in the knowledge model database may be in the form of a sentence, or may be in the form of only numerical values or items. For those requiring knowledge about numerical values or logical relationships, answer sentences are prepared as follows. That is, if it is found that the input sentence includes a word including a numerical value, a quantity, or a logical relationship based on the syntax analysis result, the input sentence includes a numerical value or a logical relationship related to the item included in the knowledge model database. ,
Abstract the input sentence from another angle and according to the symbol processing rules to the extent that the meaning does not change.

The search, creation, and output processing of the answer to the question is as follows. In other words, if the knowledge data corresponding to the question is in the knowledge model database, it is output as it is, but if not, the question is abstracted by the above-described abstraction processing, and the knowledge data is Semantic verification is performed against the sequentially abstracted sentences to derive the most suitable answer sentence and output it. The answer sentence not only outputs the knowledge data (data in the form of a sentence, word, numerical value, logical relationship, etc.) registered in the knowledge model database itself, but may be processed in some cases.

If even after the above efforts, a suitable answer cannot be found or the meaning of the question cannot be understood,
Ask a question again or give a consultation. Consultation is a request for knowledge data required for an answer. Knowledge data obtained from the user by the consultation is registered in the knowledge model database.

The natural language interface knowledge response system of the present invention having the above means can be applied as a man-machine interface such as a natural language command to a search engine or a robot on the Internet.

[0032]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a conceptual diagram of the processing of the natural language interface knowledge response system of the present invention. Input sentences handled in the present invention are knowledge input and question sentence input. Whether an input sentence belongs to knowledge data, question data, or instruction data can be determined by performing morphological analysis, syntax analysis, and semantic analysis. The input sentence is registered in the semantic model database, but if the input data is knowledge data, it is registered in the semantic model database (for knowledge, this is called the “knowledge model database”) and used as future knowledge data. You. If the input sentence is a question sentence,
It is temporarily stored in the semantic model database (for questions) and serves as a motivation for semantic verification.

When the input sentence is a command, request, or request sentence other than a knowledge sentence or a question sentence, some execution is involved. For example, for an input sentence "Please print file A", file A is printed out. In this case as well, it is temporarily registered in the semantic model database and is subjected to the same processing as the question sentence.

In the morphological analysis, an input sentence is analyzed using a word dictionary and divided into words. The grammatical composition of words divided by morphological analysis
Performed by parsing. The syntax analysis determines whether the input sentence is a declarative sentence or a question sentence. For example, "I like red flowers"
If so, it will be a declarative sentence, and if "Do you like red flowers?" The result of parsing is output as a parse tree. An output example of the result of the morphological analysis and the syntax analysis will be described in the embodiment.

In the semantic analysis, the meaning is abstracted using a thesaurus dictionary and a semantic dictionary based on the result of the syntax analysis. Synonyms and synonyms are registered in the thesaurus. For example, “birthday” is also referred to as “birthday”. The semantic abstraction rules in the semantic analysis include a noun abstraction rule, a verb abstraction rule, and a syntax abstraction rule. Specifically: ○ Noun abstraction rules ・ Prefixes (Polite) ・ Synonyms ○ Verb abstraction rules ・ Utilization (past, present, future) ・ Polite words ・ Synonyms ○ Syntax abstraction rules ・ Word order -Compound sentence-Inverted sentence-Connection rules.

For example, wording (a polite language, etc.): name → name, go → go word order (inverted sentence): he took his son to the park → he took his son : You, you, you, you are examples (more specific examples are given in the examples). The replacement of words is registered in the thesaurus dictionary, but the rules of its change and wording are registered in the semantic dictionary.

In the symbol processing rules of the semantic dictionary, the present invention is applied to a rule in which a numerical value or a logical relationship is established. For example, the following is an example using the four arithmetic operations. ○ Age rule If the date of the question is before the birthday, then age = this year-birth year-1 otherwise (after birthday), age = this year-birth year. The above example is a case where the date of birth (date of birth) is registered in the knowledge model database, and the inquiry is the age. On the other hand, the logical relationship is Pr
The log language can be described, for example, as follows. ○ Parent-child rule As a logical relationship, parent-child (A, B):-couple (A, C), parent-child (C, B). Knowledge statement: A and C are a couple. C and B are parents and children. Becomes It can be registered in the knowledge model database in the former form (logical relational expression) and used together with the latter knowledge sentence.

A sentence determined to be a declarative sentence by semantic analysis is registered in the knowledge model database. In this case, if the symbols related to numerical values and logical relationships are not the input sentence itself but simplified symbols as in the above example, the storage capacity can be saved. Conversely, a symbol can be used in an input sentence according to a symbol processing rule.

In the semantic verification, the knowledge or the semantic model of the question subjected to the semantic analysis is compared and examined, and an answer or consultation is output. The answer sentence is an answer to the question sentence, and the consultation is to ask a question that leads to the answer or request a re-question when the answer to the question sentence is not obtained. In the declarative sentence, the command sentence, the request sentence, and the request sentence can also be received as a question sentence from another viewpoint. For example, please tell us the stock price of Company A today. Tell me the stock price of Company A today. What is the stock price of Company A today? Each sentence is a request, an order, and a question, as in, but the contents can be treated as equivalent. Therefore, these sentences are often treated as question sentences at the stage of abstraction. However, in the case of "print file A" as in the example described above, execution on the system is involved.

The specific examples of the semantic analysis, the semantic verification, and the output sentence generation have been described above. In the semantic verification, the following is performed. ○ Semantic verification a. Compare semantic models to find answers to questions and point out any inconsistencies or inconsistencies in the content of the question. b. The uncertain value of the semantic model (question) is retrieved from the semantic model (knowledge) and answered. c. If no answer is obtained in b, the validity of the content of the question is evaluated. If there is a problem with the content of the question, a consultation or a request for a re-question sentence is issued to the user.

In preparing an answer sentence, in the case of a question that can give an answer of "yes" or "no", not only the answer of "yes" or "no" but also a knowledge model having a meaning close to the answer of the question sentence is obtained. Assuming that there is, answer with a supplementary explanation to the answer obtained by changing the content of the question. Let's look at the following example. Example 1: For the knowledge "He broke his leg", when he asked, "Have he broken his hand?" → Search for the knowledge "He broke his hand." → Denial → " Search for knowledge that? Has broken hand? → Negation (nobody has broken hand) → Search for knowledge that he has broken? From the above, the answer is "No, not hands but feet".

[0042]

FIG. 2 is a diagram showing a mechanism of a natural language interface knowledge response system according to the present invention. Basically, the mechanism is the same as in FIG. 1, except for morphological analysis, syntax analysis,
Semantic analysis, response model (semantic verification), semantic expression, syntax generation, and sentence generation are each incorporated into the system as modules. Among them, semantic analysis that forms the basis of the present invention,
Other than the response model and the semantic expression, they will be incorporated into the system as part of the computer OS in the future.

In the example of FIG. 2, knowledge and questions are input by voice,
Both keyboards are possible. In the case of voice input, the phoneme recognition module performs voice recognition, and converts the input voice word into a character string. For example, a character string of five characters “kaita” is generated from the uttered voice “Kaita”. In the case of key input, the input character string is directly subjected to language analysis.

The input sentence is decomposed into words by morphological analysis, and the structure is expressed in the form of a parse tree by syntactic analysis. Figures 3 and 4
Is an output example of the result of morphological analysis and parsing of the sentence "I am reading a topical book crazy". The suffix of the Japanese verb that is a sticky word changes. For example, "reading" in the input sentence is analyzed as [verb, independence], five steps, ma line, continuous connection, reading, reading. That is, it has been analyzed that “reading” is a continuous form of “reading”. By parsing the morphological analysis result based on the grammar, a parse tree as shown in FIG. 4 can be created.

A semantic analysis is performed based on the result of the syntax analysis. If the input sentence is knowledge data, it is registered as it is in the knowledge model database (for knowledge of the semantic model database), but if the same knowledge data or knowledge data with the same meaning is already registered, duplicate Then, the same data is registered. Therefore, in order to eliminate such inconvenience, abstraction of meaning in semantic analysis is important. The semantic abstraction includes a word-level abstraction and a sentence-level abstraction described below.

○ Abstraction of words Synonyms Synonyms Words with the same meaning Broad meaning words Same words as broad meanings Narrow meaning words Same words when interpreted narrowly Derivative words Words with prefixes and suffixes Words of the same reading written in different kanji, hiragana, katakana Foreign words Words expressing foreign words or English readings in Japanese in katakana Table 1 shows an example. It should be noted that “〜” in the table corresponds to “〜” above.

[0047]

[Table 1]

○ Sentence Abstraction Component Position Sentence where the positions of the main part and the modified part, and the positions of the modified part and the modified part are switched

Component Inversion A sentence preceded by a predicate that should be at the end of the sentence.

Omission of a component If the meaning is understood by the continuation of the story, a sentence omitting a certain component

Compound sentence A sentence in which a certain component of a sentence contains a clause Example: I think that such a thing is impossible. −− −−−−−− −−− −−−−− Subject predicate Subject predicate ↓ I think that is impossible. −−−−−− −−− −− −−−−− Subject predicate Subject predicate

Among the abstractions of meaning, examples of symbol processing are as follows. ○ Four arithmetic operations Example: Knowledge search by age rule If it is before birthday, age = this year-birth year-1. After birthday, age = this year-birth year. Suppose you have a question, "How old are you?"
On the other hand, the answer search is performed as follows. (1) What's your age? (2) Bring "date" from the system by applying "age rule" (3) Search for "birth date" from knowledge by applying "age rule" → failure (4) Ask the user for the "birth date" Answer text: Please tell us the target "birth date". If you want to cancel the question, please enter "return key"

○ Prolog language description processing Example: Knowledge search by parent-child rules Parent-child (A, B):-couple (A, C), parent-child (C, B). A and C are couples. C and B are parents and children. Suppose there is a question "A and B are parents and children?"
On the other hand, the answer search is performed as follows. (1) Search "A and B are parents and children" directly from the knowledge model database → Failure (2) Even if the direct relationship between A and B is not registered in the database, the knowledge of family relations related to A (see above) An example of
Couple (A, C). And parent and child (C, B). If) is registered, the answer is inferred from the relationship. That is, A and C
Is a couple, and C and B are parents and children. Therefore, it is related that the child of A is also the child of C. This is derived from the following relational expression. Parent and child (A, B):-couple (A, C), parent and child (C, B). (3) The answer is "Yes."

Knowledge Processing Using Frames A frame is a "framework for expressing knowledge".
Thereby, knowledge having a hierarchical unit can be expressed more efficiently and in a natural form.

Example: Biological frame ----------- Concept name :: Organism High-level concept name :: Natural breeding :: Yes Water :: Some respiration :: Yes

Animal Frame ----------- Concept Name :: Animal General Concept Name :: Organism Chlorophyll :: No Nutrition :: Organic Matter Athletic Organization :: Hold

Plant frame −−−−−−− Concept name :: Plant Top concept name :: Organism Chlorophyll :: Yes Root and leaf :: Yes Leaf color: Default: Green

From the above frames, it can be seen that living things include animals and plants. Conversely, by looking at the general concepts of animals and plants, it can be inferred that both are organisms.

As described above, the knowledge model database holds knowledge data in the form of a Prolog or a frame in addition to a sentence format. If the sentence subjected to the semantic analysis is knowledge data, it is registered in the database in the form of a sentence, Prolog, frame, or the like. Conversely, for a question sentence, the question sentence is compared with the knowledge data in the knowledge model database.
Will be considered. This process is performed using a response model.

FIG. 5 shows the positioning and contents of the response model. The external database is an information database relating to the response destination. Specific contents include a customer file and a company information file. Which types of files are included depends on the application. There are three execution patterns that the language understanding engine supports for natural languages:-declarative text-> memory-question text-> answer-imperative text-> execution. By setting the response control program, basic knowledge data, Web setting file, and image file, an application system on the Internet can be easily constructed.

FIG. 6 shows an example of a system in which the communication environment is constructed by the Internet and the development language is constructed by the Prolog language. The basic (inference) engine is a module that performs functions such as an answer search inference engine for questions, abbreviated sentence processing, and context understanding (semantic verification). When used on the Internet, a session management program for controlling the execution of the program must be introduced in order not to reload a large amount of dictionary data every time a session is interrupted.

[0062]

The most significant feature of the present invention is that it enables a man-machine interface in natural language. In the conventional man-machine interface,
Statements with special syntax and menu-style interfaces have been used. In the former case, the user had difficulty remembering the statement. In the latter method, clear instructions can be given to the machine, but the system is hard to convey the user's intention. On the other hand, since the present invention uses a natural language, the user has an advantage that the user can talk with the machine in everyday language, and special handling is not required.

In the present invention, since answers are inferred over a wide range by abstracting the meaning, it is not necessary to specifically prepare answer sentences for the question sentences. For example, when the knowledge data corresponding to the question is not in the knowledge model database, new knowledge can be naturally increased by requesting the user for knowledge necessary for answering the question (consultation). In addition, since it is possible to find and respond not only to the direct answer to the question but also to the meaning close to the question and the answer related to the question, some kind of answer to information that is not registered as knowledge as in the past You can answer it with a supplementary explanation.

Although related to the above-mentioned contents, in the present invention, maintenance of knowledge data is easy. The reason is that knowledge data can be created by text documents, and there is no need to learn how to use programs to create the data. The ability to express, the ability to propagate knowledge data through consultation, and the like.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram conceptually showing a processing flow of a natural language interface knowledge response system of the present invention.

FIG. 2 is a diagram for explaining a mechanism of a natural language interface knowledge response system of the present invention.

FIG. 3 is a diagram showing an example of an output result of a morphological analysis in the natural language interface knowledge response system of the present invention.

FIG. 4 is a structural analysis tree diagram showing an output result example of syntax analysis in the natural language interface knowledge response system of the present invention.

FIG. 5 is a diagram showing the positioning and contents of a response model in the natural language interface knowledge response system of the present invention.

FIG. 6 is a system configuration diagram for explaining a configuration of a natural language interface knowledge response system of the present invention.

Claims (6)

[Claims] 1. A knowledge response system in which a sentence is input in a natural language, a computer performs a language analysis on the input sentence, and performs some kind of answer, data storage, and execution.
Means for performing morphological analysis, syntax analysis, and semantic analysis on a sentence input in a natural language form; (2) when the input sentence is information data on knowledge (declaration sentence), a semantic model database as a knowledge sentence (Knowledge model database)
(3) When the input sentence is a question (question sentence), means for temporarily storing it as a question sentence in a semantic model database, verifying the meaning, and outputting some answer sentence, (4) When the input sentence is a command (command), a natural language interface knowledge response system includes means for temporarily storing the command as a command in a semantic model database, verifying the meaning, and executing the command. 2. The semantic analysis means according to claim 1, wherein a dictionary in which synonyms, derivatives, honorifics, etc. are registered is used to convert the input sentence to such an extent that the meaning of the input sentence does not change, based on the result of the syntax analysis. A nature characterized by comprising semantic abstraction means for abstracting (semantic abstraction rules), and means for performing data search on the abstracted sentence from the knowledge model database and outputting a search result as an answer sentence. Language interface knowledge response system. 3. The semantic analysis means according to claim 1, wherein based on the result of the syntax analysis, the input sentence includes a word including a numerical value, a quantity, or a logical relationship, or the relational expression can be used. If found, a symbol processing means (symbol processing rule) for abstracting the input sentence from another angle to the extent that the meaning does not change, based on a numerical value or a logical relationship related to the item included in the knowledge model database; And a means for tailoring the abstract sentence into an answer sentence and outputting the answer sentence. 4. The method according to claim 1, wherein, if the knowledge data corresponding to the question sentence is searched from the knowledge model database after the morphological analysis, the syntax analysis, and the semantic analysis, the knowledge data is extracted. If the answer is not true, means for sequentially comparing the abstracted knowledge data of claims 2 and 3 with a question sentence to derive the most appropriate answer sentence and outputting it is provided. A natural language interface knowledge response system. 5. A means for extracting an answer to a question according to claim 1, wherein after a morphological analysis, a syntax analysis, and a semantic analysis, knowledge data corresponding to the question sentence is searched from a knowledge model database to obtain an appropriate answer. A natural language interface knowledge response system comprising: means (consultation means) for requesting a user for knowledge data necessary for an answer when the user is not able to do so. 6. A natural language interface knowledge response system comprising: means for applying an answer to the above-mentioned question as a natural language instruction to a search engine or a robot on the Internet.