JP7239959B2 - Method, computer program and apparatus for automatically synthesizing ideas from natural language text - Google Patents

Description

Application of Patent Act Article 30, Paragraph 2 April 21, 2017 Disclosure to the Mitou Advanced Business Secretariat

Applied Patent Act Article 30, Paragraph 2 Published at Awagasaki Business Salon held at Kanazawa City Monozukuri Center on April 27, 2017

Applied Patent Act Article 30, Paragraph 2 Published at Awagasaki Business Salon held at Kanazawa City Monozukuri Center on April 27, 2017

Application of Article 30, Paragraph 2 of the Patent Law Presentation at the Ishikawa Prefecture Venture Matching Forum held at the Yurakucho Denki Building on December 10, 2016

Article 30, Paragraph 2 of the Patent Act applies Published at the final examination of CVCK held at IT Business Plaza Musashi on March 8, 2017, and published on slide share on March 9, 2017

Application of Patent Law Article 30, Paragraph 2 January 20, 2017 Disclosure to Kanazawa City CVCK Secretariat

The present invention relates to a method, computer program and apparatus for synthesizing ideas from natural language text.

So-called "classical TRIZ" and its modern extensions are known as methods for synthesizing ideas such as inventions from natural language documents represented by patent documents (Non-Patent Document 1).
There, typical idea methods for problem solving extracted from past huge patent documents and the history of technological progress, typical direction of progress, methods of utilizing knowledge in other technical fields, other methods such as QC, etc. It includes software that makes it easy to use frequently used ideation tools (such as fishbone diagrams) introduced from
However, it is still not possible to combine, as necessary, the secret know-how of users and affiliated organizations written in "completely free natural language" and natural language texts (knowledge and information) in "multiple" technical fields. It is not possible to fully automatically synthesize new ideas in response to user requests (e.g., questions (open/closed), set of problem/issue situation descriptions (context), assumptions (hypotheses, etc.). .
On the other hand, Patent Literature 1 and Patent Literature 2 below show a mechanism for searching sound inference results from completely free natural language texts in multiple fields. , was not explained, and there were not sufficient preparations for the necessary or specialized system functions, UI (how to display user requests), knowledge expression methods, etc. (Problem 1) .
In addition, when it is necessary to exclude "science/technology/business/public order/ethical improbable" ideas from solution candidates, it is necessary to express and process constrained problem settings more accurately and efficiently. In many situations where it is necessary, ``NOT as a failure of proof'' is insufficient in its descriptive ability, and an interpretation that is easy to match with ``a proposition that has no facts in memory and cannot be automatically proved by rules'' is regarded as false. The policy has the harmful effect of unreasonably restricting the exploration of new ideas (especially ontologies and pragmatics (especially the synthesis of new concepts and ideas by combining words) that are free for users in natural language). (Problem 2).
In addition, although many desirable ideas to be automatically synthesized are selected with priority over other ideas according to the "values" of "general / specific field / organization / individual user", It has not been defined in a form that can be used for automatic synthesis, and its utilization method has not been established (Problem 3).

Patent Application No. 2013-087008 Multi-length matching of literals with variables embedded in natural language and management of infinite sets, etc. Patent Application No. 2016-009324 Reverse question from computer to user and addition of dynamic facts (temporary context) Patent Application No. 2017-092047 Addition of Dynamic Facts (Temporary Context) by Forward Inference Patent Application No. 2017-092057 Method for Realizing Negation in Logic Programming Japanese Unexamined Patent Application Publication No. 2004-252507 Method for recruiting knowledge and information http://www.proengineer-institute.com/triz.html "What is TRIZ?"

Therefore, the present invention provides a method, apparatus, and computer program for fully automatically synthesizing a completely new idea responding to a user's request from completely free natural language texts in multiple fields related to the user's request. intended to
Another object of the present invention is to provide a method for excluding "impossible" ideas in a manner that does not cause the above-mentioned problem of unreasonably restricting the exploration of ideas and the synthesis of solution candidates.
Another object of the present invention is to establish a method that can be used to automatically synthesize ideas, defining values ranging from general values to user-specific values in a form that can be used for automatic synthesis of ideas.

Claim 1 of the present invention, in order to solve such a problem,
A computer program that includes some or all of the functions of PROLOG,
By allowing the use of literals embedded with variables in natural language in logic programs,
Enabling natural language pattern matching coupled with recursive syllogisms,
Include in the logic program a rule (hereinafter referred to as a "synthesis rule") in which the variables included in each condition of the body on the right side of the rule corresponding to the rule of PROLOG do not include all types of variables included in the head on the left side,
Different literals (however, the scope of the included variable is common within the synthesis rule) that are each condition of the synthesis rule can separately match multiple texts in multiple fields,
allowing concepts from multiple fields to be combined by composing new concepts in the head literal of the composition rule;
Provide a computer program.
: Here, "multiple fields" may be different science and technology fields such as "mechanical technology", "computer technology", "educational technology", humanities, arts, etc., and more subdivided among them It may be an adjacent field, etc., or it may be another article in the same field (experimental results, historical facts, micro/macroeconomic phenomena and analysis, sports knowledge/results, etc.) .
> By this, one or more concepts (a term of each field or a phrase consisting of multiple terms) that satisfies each of the more abstract conditions of the body are placed in appropriate positions in the pattern of the head's literal (templates also fall under this category). (so-called "case" or positioning for constructing new functions and new values), it becomes possible to compose and synthesize new concepts and ideas (functions, values, etc.). Even if there are problems such as notation variations, a certain degree of comprehensiveness is ensured, and by targeting a larger number of articles, humans can read the same knowledge in multiple fields at almost the same time and invent ( It is also possible to enumerate ideas that should be possible (synthesis) without omission (although there may be some duplication due to notation variations).
In the demonstration example 1 below, the head of the 5th line, ``The new product $A is $B :- $C the new product is $A; All the types of variables ($A, $B) in , and neither condition ``$C new product is $A'' or ``$B is attractive'' alone, so natural language text knowledge limited to the company you belong to Combining "Our new Kutani product is Ochoko" and natural language text knowledge in the field of design and art "The shape of the lips is attractive", a new idea "New product Ochoko" in the head "New product $A shape is $B" The shape of is a lip-shaped synthesis.

2: ☆
3: * Logic program example 1
4: ☆
5: Shape of new product $A is $B :- $C New product is $A ; $B is attractive ;
6: CFRP is light and strong ///text in mechanical design field
7: Spheres are easy to roll ///text in the mechanical design field
8: Our company's new Kutani product is Ochoko.
9: Heart shape is warm /// Text in the field of design and art
10: The sphere is simple ///text in the field of design and art
11: Lip shapes are attractive /// Texts in the field of design and art: Embodiments such as detailed algorithms such as "pattern matching of natural language linked with recursive syllogism" have been published. Please refer to the above-mentioned patent document 1 by our company (but the main part is also excerpted in this specification).
:

In addition, claim 2 of the present invention is
each text that constitutes the plurality of texts in the plurality of fields is composed of one or more sentences that are separated from each other;
other sentences in the text to which the sentence that matches one of the above body conditions belongs,
2. The computer program according to claim 1, wherein control is performed so that another condition of the same body is preferentially matched.
: "Priority match" means that, for example, even if completeness and exhaustiveness are sacrificed, all articles to which matching sentences belong are cached in memory, and sentences that match other conditions are retrieved only from the cache. It is to seek. In addition, by increasing the number of recursively related articles based on user input, inference result sentences, keyword combinations of cached articles, etc. by breadth-first search, etc., articles in multiple fields are cached up to a predetermined capacity. Pattern matching with the condition may be performed only from the sentence. Combined with the policy of allowing users to freely use natural language, this can help avoid the frame problem to some extent.
In addition, since it is possible to synthesize new concepts and ideas by preferentially collecting elemental concepts from articles and contents with high relevance, affinity, and cohesion in terms of fields, cases, and instances, ideas that are easy to apply in reality. It is possible to synthesize and propose preferentially without leaking.
Note that "priority" does not exclude other articles or articles in other fields, and as the process progresses, elemental technologies, etc. will be collected from articles that are farther apart in a wider range of fields. (although a match is still required). For example, by simply constructing an algorithm with a certain degree of completeness as a breadth-first search, making the progress visible to the user, and making it possible to manually pruning and interrupting the visual and manual search, this ( Both priority and completeness) can be implemented.
In addition, claim 3 of the present invention is
For such uses (collect related sentences by field or by instance (article), and make it easier to search as a condition for natural language pattern matching with recursive syllogisms for automatic synthesis according to user requests) to the
In conjunction with the computer program of claim 1 or 2,
In order to prevent notational variations and duplication of pattern parts and element concepts,
Indicate the field, theme, pattern (template), or existing data to the user,
Provide a computer program that solicits contributions of meaningful data (knowledge or information) by users in its patterns (languages, templates, etc.) into its categories (fields, themes, etc.).
This makes it easy to collect and disperse knowledge of multiple fields or themes without notation variations, omissions or duplications.
For this basic mounting method, please refer to the aforementioned Patent Document 5 by our company.
Here, the categories and themes are narrowed down, and the extracted data (solution sets such as "($X, $Y) = (Taro, Hanako), (Jiro, Yumiko), ...") are constructed as complete sentences. The method of providing information to users and their client software is self-explanatory and different from the gist of the present invention, so it will not be described in detail in this specification.
Such a server may be a public directory server (web server + search function is also acceptable) or a directory server within an affiliated organization. It is advantageous to provide appropriate URLs to users. It is advantageous to manage the structure and the field or theme structure in conjunction with each other.
In addition, claim 4 of the present invention is
As requested by the user,
Accepts a literal expression in which variables are embedded in natural language,
4. The computer program according to any one of claims 1 to 3, characterized by automatically synthesizing an idea requested by a user by automatically finding a solution that binds embedded variables.
For a detailed implementation of this mechanism, please refer to published patent application WO 2004/020000 (although the main part is also excerpted in this specification).
This makes it possible to express simple and seamless user requirements by taking advantage of natural language pattern matching with recursive syllogism (fusion of search and inference), and to perform timely automatic synthesis at high speed. .
In addition, claim 5 of the present invention is
As requested by the user,
accepting a hypothesis consisting of one or more sentences in natural language;
By finding the difference set between the set of propositions that can be automatically proved by adding the assumption to the logic program and the set of propositions that could be proved automatically before adding the assumption to the logic program,
A computer program according to any one of claims 1 to 3 is provided for automatically synthesizing ideas requested by a user.
According to the technique of Patent Document 1, the entire proposition that can be automatically proved can be obtained by a question sentence (query) of the pattern "$A" (synonymous with asterisk) consisting only of variables with no restrictions on the character strings before and after. , the difference set of the automatically provable proposition set (solution of "$A") before and after adding the assumption (sentence set) from the user to the logic program (Figs. 1 and 2).
This makes it possible to synthesize new ideas corresponding to the assumptions indicated by the user without explicitly preparing a mechanism for forward inference in the execution system or logic program.
In addition, claim 6 of the present invention is
As requested by the user,
accepting a temporary context consisting of one or more sentences in natural language;
A computer program according to any one of claims 1 to 3, characterized in that it automatically synthesizes ideas requested by a user by forward inference from sentences contained in a temporal context.
Forward inference is a computer program (PROLOG processing system, etc.) that processes rules that consider a given proposition to be true when some or all of a plurality of conditions are true, such as the Horn clause of PROLOG. In this case, when new information is obtained that matches a part of a plurality of conditions of the rule, automatically search for the establishment of other conditions of the synthesis rule to automatically establish the predetermined proposition of the synthesis rule. It is clearly achievable if it is confirmed. For other implementations, see US Pat.
As a result, for example, it becomes possible to limit the rules to be the target of forward-looking reasoning to a set of promising rules rather than all of them, and to automatically synthesize ideas based on both the user's requests and the know-how such as the empirical rules of the knowledge provider. becomes possible.
In addition, claim 7 of the present invention is
In a question, fact, rule condition or rule effect entered by a user, or in a logic program fact, rule condition or rule effect,
In addition to the affirmation (true) as the existence of evidence facts or the success of automatic proof and its denial (absence of evidence facts and automatic proof failure/impossibility),
Permitting the use of the word "impossible" to imply the existence of evidence to the contrary or the success of automatic proof to the contrary;
When using an affirmative expression that does not include the variable from the beginning or does not include it due to variable binding,
Automatically confirming that the expression "impossible" corresponding to the affirmative expression does not exist and cannot be automatically proved,
7. A computer program as claimed in any one of claims 1 to 6, adapted to restrict the use of the corresponding affirmative expression if the "impossible" expression exists or can be automatically proved.
In a specific embodiment, knowledge expressions that mean "impossible" can be uniformly prioritized (or uniformly subordinated) to corresponding affirmative expressions.
In addition, logic translation can be performed with priority given to the latest user input.
Also, when translating into logic, expressions of the opposite meaning can be deleted from the memory storing the logic program or temporary context.
Also, when translating into logic, expressions of the opposite meaning can be commented out from the memory that stores the logic program or temporary context.
Moreover, it is also possible to automatically add a time stamp or cause event identification information to the comment.
For example, in the 15th line below, there is an expression as a rule of "impossible" ``Make $X eat $Y :- $X is livestock; $Y is poison;'' The value of $X known to be (
"goat") and a $Y value known to be poisonous (e.g. "arsenic paper")
is a variable set ($X, $Y) of "make $X eat $Y"
It is a knowledge representation that excludes (restricts) simultaneous binding of

12: ☆
13: * Logic program example 2
14: ☆
15: !feed $X $Y and make money :- $X is livestock; $Y is poison;
16: How to dispose of $Y's waste using $X :- Make $X eat $Y ;
17: Make $X eat $Y and make money :- $X eats $Y ;
18: Goats are Livestock
19: Paper with arsenic is poisonous
20: Goats eat paper with rapeseed oil
21: Goats eat arsenic paper
22: ☆
In this case, "Method for treating waste paper with rapeseed oil using goats" is generated as an idea, but the solution candidate "Method for treating waste paper with arsenic using goats" is generated in line 16. Since the grounds (sufficient condition) for "making money by feeding goats with arsenic paper" is restricted by the 15th line, it is not generated.
Such suppression (restriction) of generation is obvious at the time when the idea becomes a fixed character string by the literal binding process in which the variable is embedded in the natural language in the above-mentioned Patent Document 1, the calculation in a realistic time is clear. Since it is possible, it is advantageous to search for all logical programs at such timing and suppress/exclude them, for example, by line 15.
In the above example, the head (effect) of the rule is added with "!" to exclude the solution of poisoning livestock in the global scope. case, the fact that the attached condition is "impossible" exists in the logical program or the temporary context, or the success of the automatic proof that the attached condition is "impossible" (the logic program in memory or the temporary context using the rules of ).
A specific interpretation of "! ) is a simple string match, and its implementation is self-explanatory. On the other hand, with regard to automatic certification, the rule that omits the "!" in the rule with "!" A realization method such as satisfying the condition if possible may be used. Regarding the method of realizing this proof, please refer to Patent Document 1, which has already been published (the main part of which is also excerpted in this specification).
Constraints, meaning "cannot be" or "not", are extremely important knowledge representations for solving constraint satisfaction problems that express real problems as constraints. It will be greatly improved.
In general, such semantics (computer interpretation) could not be realized in conventional PROLOG, etc. It greatly simplifies
After all, the range of ideas that can be considered "impossible" for various reasons is
Compared to the conventional "NOT due to failure of proof",
without unduly restricting the scope of ideas,
And it is possible to exclude it with a more accurate and concise expression,
The utility of automatic idea synthesis can be further enhanced.
Moreover, claim 8 of the present invention is
Based on the content that defines the order of items (or item sets) in a given set of values,
A computer program as claimed in any one of claims 1 to 7, characterized in that it restricts the strings used to synthesize the concept requested by the user.
For example, in the example below, the general order of importance is defined as "family wedding>best friend's wedding>baby's birthday>pet's birthday."
Also, on line 31, this can be replaced with a more general predicate expression that can be referred to (however, the format is headed by a literal with variables embedded in natural language), and can be used as a condition for the body of the right side of other rules. I'm trying

22: ☆
23: ★Logic program example 3
24: ☆
25: <% generally important
26: Family wedding
27: Best friend's wedding
28: Baby's Birthday
29: Pet's Birthday
30: >% generally important
31: $X is generally more important than $Y : > % generally more important ; $X >$Y;
32: Prepare a month in advance for $X:- $X is generally more important than your best friend's wedding;
As a result, the range of ideas and idea components to be automatically synthesized can be set according to a predetermined sense of value (for example, importance) or the degree of characteristics suitable for being used as a component (for example, weight, hardness, flame as a substance). difficulty, etc.), leading to savings in the user's idea selection cost.
In addition, claim 9 of the present invention is
Based on the content that defines the order of items (or item sets) in a given set of values,
A computer program as claimed in any one of claims 1 to 8 for controlling the priority of strings used to synthesize user requested concepts.
: "Controlling the priority of strings" means, for example, in the above example, displaying "family wedding" earlier than "best friend's wedding" in terms of time, synthesizing speech, It means to display in a conspicuous upper part.
> This allows the user to get in touch with promising/valuable idea candidates more quickly, and enables flexible handling even when time is limited.
In addition, claim 10 of the present invention is
An apparatus using a computer program according to any one of claims 1 to 9 is provided.

In addition, claim 11 of the present invention is
A method using a computer program according to any one of claims 1 to 9 or according to figure 2 is provided.

A diagram illustrating the relationship between a logic program (fixed facts and rules), facts added as assumptions by the user, and content to be answered according to an embodiment of the present invention (when user requests are shown as assumptions). be. FIG. 2 is a diagram illustrating the most basic processing flowchart (steps as a method) of a computer program for implementing the embodiment of FIG. 1; It is an explanatory view showing a general configuration as a computer to implement.

The present invention is implemented as an automatic idea synthesizing system using the logical AI (subset of PROLOG + α) and knowledge expression (which can be described in almost natural language) of Patent Document 1.
Some simple examples described later are implemented by additionally implementing the above-described functions specific to the present invention in the embodiments of the following patent specification (excerpt from Patent Document 1 above) by the same inventor. can be implemented. In addition, the following citation of Patent Document 1 includes a supplementary purpose of the present inventor's feasibility and inventive step.

===
<Problem of Patent Document 1> It is possible to realize automatic processing of knowledge expressed in a form of embedding variables in a natural language by a computer in the same way as PROLOG, exhaustive deduction and solution search at the predicate logic level (automatic unification processing) , automatic derivation processing, backward reasoning for it, backtracking).
To make full use of the advantage that variables are not bound by a case structure, speed up processing, and display results in an easy-to-read manner for proof, learning, and the like.
<Solution of Patent Document 1 (summary)>
When humans input "facts", "rules" or "goals" in PROLOG, P
The constant part and the variable part of the contents corresponding to "literal" in ROLOG are distinguished by character type, delimiter or escape character and input into the computer,
The computer performs automatic unification and/or derivation of sentences contained in its input, treating variables as being able to cross boundaries of sentence subjects, predicates, etc.
The unification of patterns, which are statements that can contain variables, and constant strings, which are statements that do not contain variables,
For each new pattern obtained by substituting the process of finding the longest matching solution of the variable that first appears in the pattern and deleting the character string of the longest matching solution from the back until it becomes an empty character, the new pattern is While memorizing each state of the solution in the shaving process, which is the premise of being able to do it, it repeats the same recursively until there are no more variables. Get a unified answer.
Speed up by detecting whether the fixed character string in the pattern is included in the character string in the process of removing the longest matching solution.
Pattern-to-pattern unification is accelerated by focusing on word-initial and word-ending matches. We also provide a concrete method for exhaustive search.
<Solution of Patent Document 1 (full text, but some additions and corrections so as not to be confused with the present invention)>
(0009)
The present invention has been made in view of such problems, and claim 1 of Patent Document 1 is
A method by which a computer automatically performs some or all of the functions of the logic programming language PROLOG,
When a human inputs "facts", "rules" or "goals" in PROLOG, the constant part and variable part of the content corresponding to "literal" in PROLOG (hereinafter referred to as "sentence") are separated by character types, delimiters or escape characters. be entered into the computer separately by
The above "facts", "rules" or Automatic unification processing, automatic derivation processing, or both of sentences included in the input of "goals";
To provide a processing method characterized by:
Here, ``a human ... inputs a constant portion and a variable portion into a computer while distinguishing them by character type, delimiter, or escape character'' means that a human being enters the input field of the application software (processing system) according to the present invention, or Direct input after a command prompt (through or without a front-end processor, etc.), copy-paste or cut-paste from an Internet browser or text editing software, or a text file that has been separately input and saved including reading, quoting, referring to, etc.
In addition, "a computer treats a variable as something that can span the subject, predicate, subject, predicate, object, complement, predicate modifier, adnominal modifier, or sentence boundary of a sentence" As long as it is possible to cross the boundary (explicit division) of case structures with explicit divisions (including those corresponding to the hierarchical structure of syntactic analysis results and semantic analysis results) through particles, etc. and prepositions in English, etc. Instead of English, for example, S and V in SV format, SVO format, SVOO format, and SVOC format
, V and O boundaries, V and C boundaries, etc., including modes that handle them as being able to cross the boundaries of compound sentences and compound sentences (conversely, there may be modes that do not handle them)
. In these cases, candidate solutions that can be stored in variables include the entire sentence, arbitrary substrings extracted from the character strings that make up the entire sentence, ignoring the delimiters of the grammatical structure of the sentence, and even hierarchy. Complex sentences, compound sentences, longer papers, etc. with similar reference structures, as well as arbitrary substrings of them are targeted (In addition, when handling "sentence boundaries" as possible, When handling a set of knowledge sources (for example, collectively reading, writing, referencing a set of knowledge sources organized in a file, etc.), it is necessary to delimit knowledge sources at the boundaries of sentences. Therefore, line feed code, tab code, etc. can be used as such delimiters).
As a result, knowledge and information expressed in the form of embedding variables in natural language and knowledge and information in the form of ordinary sentences are unified knowledge that fully opens up the possibility of realistic mutual utilization between them. As a source, it becomes possible to automatically process them by computer without distinguishing between them (overcoming the above "difficulty A").
In order to open up the possibility of "realistic" interaction "sufficiently", automatic processing by computer should be provided as a search processing system 1) Simplicity of search formula (expression specifying a pattern),
2) Sufficient precision (accuracy of solution), 3) Sufficient recall (completeness in extracting solution candidates)
However, in the present invention, variables (multiple types or multiple times) can be embedded freely (that is, not limited to grammatical or semantic delimiters) in the sentence (simplicity Securing), PROLOG's knowledge processing procedures (and the underlying knowledge interpretation policy)
By basically following While avoiding risks, "the subject, predicate, subject, predicate, object
Treating variables as complements, predicate modifiers, adnominal modifiers, or as being able to cross sentence boundaries allows us to use natural language delimiters (comma, punctuation, parentheses, commas, periods, etc.) and modifier phrases ( Adjective phrases, adverbial phrases, non-essential case expressions, etc.) prevent unreasonable mismatches (regular expression unmatches) caused by the appearance of specified patterns, and between surface character string expressions where case structures are not specified ( It is possible to ensure a certain degree of reproducibility (completeness in extracting solution candidates) between the search formula and the knowledge source). In addition, it is possible to quote and extract integrally and continuously (as solution candidates for variables) without losing the accurate context (preceding and following character strings) of the natural language that is the source of knowledge. Simplicity makes it easy and convenient to confirm and verify the substantial soundness of reasoning (justification of grounds and reasoning background), and gives users a realistic sense of security that the grounds are always clear. Take full advantage of the specification that variables are not bound by the case structure (in addition, you do not need to know the rules of the case structure (such as the order of appearance), and you do not have to worry about the order of arguments being wrong). make it possible to
In addition, since the present invention is primarily aimed at Japanese (the amount of information per character is large and there is no space between words), there are many contents in which characters are used as units of processing and divisions, but English, etc. (The amount of information in one character is small, the number of characters in one sentence is large, but there are spaces between words.) It is also possible to perform similar processing on a unit basis of words separated by spaces.
(0010)
In addition, claim 2 of Patent Document 1 is
In the automatic unification process,
The unification of patterns, which are statements that can contain variables (constant parts and variable parts are distinguished by character types, delimiters, or escape characters), and constant strings, which are statements that do not contain variables,
The first (or last) occurrence of a variable (e.g. $X and $Y likes $Z) in a pattern (e.g.
$X), find the longest matching solution (e.g. Taro and Jiro), and delete the character string of the longest matching solution from the back (or front) until it becomes an empty character (e.g. Taro and Jiro > Taro and next > Taro and>Taro>Futo>Sora)
into the pattern (e.g. "Taro and Jiro and $Y like $Z">"Taro and Tsuji and $Y like $Z">"Taro and $Y like $Z">"Taro and $Y like $Z">"Futoshi and $Y like $Z">"and $Y like $Z") ($X = Taro and Jiro > Taro and next >
While remembering Taro and > Taro > Taro > Sora),
Repeat recursively until there are no more variables,
The state of each intermediate solution unified when there are no more variables (e.g. true ($X = Taro and Jiro,
$Y = Saburo), Shin ($X = Taro, $Y = Jiro and Saburo)) to combine each unification answer (e.g.:
Answer 1 = ($X = Taro and Jiro, $Y = Saburo) and Answer 2 = ($X = Taro, $Y = Jiro and Saburo) are obtained. Provide processing methods.
Here, "(or last)" corresponds to "(or before)", which means that when substituting from the variable that appears last, it is necessary to start shaving from the front.
This makes it possible to verify the state of each solution candidate, which reliably covers all possibilities from the longest matching solution to the shortest matching solution, and to provide an exhaustive answer (a further narrowed set of solution candidates). be possible.
(0011)
In addition, claim 3 of Patent Document 1 is
The processing method according to claim 2 of Patent Document 1 is provided, characterized in that, instead of gradually removing until an empty character is obtained, only a shortest matching solution obtained separately is removed gradually.
As a result, it is possible to save the calculation of the solution search from the length less than the shortest matching solution to the "empty" character, and it may be possible to speed up the automatic processing.
(0012)
In addition, claim 4 of Patent Document 1 is
When gradually scraping until it becomes an empty character,
If the fixed character C (immediately after (or immediately before)) that is the fixed character immediately after (or immediately before) the variable appears first (or last) in the pattern is another variable and the solution is not an empty character, it is in the other variable. If the first (or last) character C) of the solution candidate substituted as the solution candidate is included in the longest matching solution,
The processing method according to claim 2 or 3 of Patent Document 1, wherein a substring including the C is deleted from the end (or beginning) of the longest matching solution string to the first hit of the C. offer.
Here, "(or last)" corresponds to "(or immediately before)", which means that when substituting from the last appearing variable, the shaving must start from the front.
As a result, enumeration of solution candidates that do not match the fixed character immediately after (or immediately before) the variable and searching for those solution candidates can be saved, and the speed of automatic processing may be increased.
(0013)
In addition, claim 5 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When obtaining binding information that will be the "goal" (including "subgoal"; the same shall apply hereinafter),
Find binding information independently for each sentence that constitutes the "goal" and return the return value true and the product set if the intersection of each piece of binding information is not empty. to provide a method of processing as described in Section 1.
Here, the binding information is information indicating whether or not to substitute such solution candidates for each variable at the same time. The binding information that indicates how to bind each variable is ","
can be expressed in a list format such as "variable name 1 = value 1, variable name 2 = value 2, ..., variable name N = value N", while multiple bindings The method is as a set of such lists, or a form that expresses the AND / OR combination instead of "," to extract the common part between multiple bindings, or a conjunctive normal form that standardizes the whole of multiple bindings. or disjunctive normal form.
As a result, there are multiple patterns (sentences) to be satisfied at the same time, and in the case of a problem format in which the same solution should be entered in variables with the same name between the patterns (sentences) (for example, such an A
ND join queries, or cases where there are multiple sentences in the body of the rule), the pattern on the right side of the rule can wastefully repeat a large full-text search operation for each unification result of the pattern on the left side. to avoid the inefficiency of depth-first search while backtracking the body (which is more noticeable in the form of embedding variables in natural language), while at the same time simplifying the processing for each pattern to be satisfied. can be done. You can also display and verify the inference process (
completeness and soundness) are easy to understand and understand. In addition, orderliness (between sentences in the body part), which is superfluous for the intuition required for the meaning of rules and should not be defined, can be prevented from giving side effects to the reasoning process.
(0014)
Further, claim 6 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When obtaining the binding information that is the "goal",
empty the existing binding information first,
For the set of sentences included in the "goal", automatic calculation of additional binding information for whether the sentence can be true under the application of existing binding information and applying the additional binding information to the existing binding information While adding, remove the sentence and recursively make it smaller,
The processing method according to any one of claims 1 to 4 of Patent Document 1 is provided, in which the existing binding information when the sentence set becomes an empty set while it is possible to be true is set as the binding information as the "goal". do.
Here, "recursively decreasing" means that the deeper the recursion, the smaller the statement set and the more complex the binding information (the more variables are bound, and the range of binding values is narrowed).
For example, it means to implement equivalent recursive processing by making recursive calls in C language or by using a uniquely declared and defined stack structure for depth-first search (struct array and stack pointer, etc.). .
As a result, the problem of automatically processing an open question consisting of AND-joints of multiple patterns that must be satisfied at the same time, expressed in a form that embeds variables in natural language, is simplified into a simpler and recursive subproblem with the same structure. Transforms to facilitate presentation and verification (completeness and soundness) of reasoning processes.
(0015)
In addition, claim 7 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
If we unify the unvariable pattern with the unvariable fact,
The processing method according to any one of Claims 1 to 6 of Patent Document 1, wherein true is returned when the pattern and the fact match as a character string, or false is returned when the pattern and the fact do not match as a character string. offer.
This makes searching simple and fast.
In addition, when "facts not including variables" include multiple sentences separated by punctuation marks, etc. There may also be a mode in which "match as a string" may be that the string of the pattern matches part of the string of facts.
As a result, a plurality of pieces of information of the same type can be collectively expressed as a single fact and used as a search target. Also, some order among multiple elements (e.g., size, preciousness, antiquity, etc.)
etc.) will be easier to express.
(0016)
In addition, claim 8 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When unifying the variable-free pattern with the variable-free head of the Horn clause,
If the pattern and the head part match as a character string, return the result of applying the processing method according to claim 5 or 6 of Patent Document 1 with the body part of the Horn clause as a subgoal,
A processing method according to any one of claims 1 to 7 of Patent Document 1 is provided, which returns false if they do not match.
This enables a sound mechanical interpretation when the head of a rule that expresses variables in a form embedded in natural language does not contain variables, a seamless, easy-to-understand, simple and unified processing mechanism, and an explanation of the corresponding inference process. can give
(0017)
In addition, claim 9 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When unifying a pattern containing no variables with a Horn clause with a head containing variables,
The pattern is used as the "constant character string" in the processing method described in any one of claims 2 to 4 of Patent Document 1, and the head part is used as the "pattern" in the same processing method. For each binding information,
Returns true if the result of processing the processing method described in claim 5 or 6 of Patent Document 1 as the “subgoal” of the processing method described in claim 5 or 6 of Patent Document 1 is true if it is applied to the body of the horn clause, or returns true If not, return false. A processing method according to any one of claims 1 to 8 of Patent Document 1 is provided.
As a result, if the head of a rule that expresses a variable in a form embedded in a natural language contains variables, and the pattern that triggered this rule does not contain variables, the scope of binding of the variables in the body is restricted in advance, False judgment processing can be speeded up.
It can also provide a sound mechanical interpretation and a seamless, understandable, simple, and unified processing mechanism and description of the corresponding reasoning process.
(0018)
In addition, claim 10 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When unifying patterns containing variables with facts not containing variables,
"Pattern" in the processing method according to any one of claims 2 to 4 of Patent Document 1
and return the result of processing using the fact as a "constant string" in the same processing method,
A processing method according to any one of claims 1 to 9 of Patent Document 1 is provided.
As a result, all solution candidates (to class variables) between the pattern and the (static) fact are identified and returned as the "result", which is the basis for overcoming the above "difficulty A". In addition to the visualization and verification of the inference process, it also facilitates learning through the user's hands, such as modifying rules.
(0019)
In addition, claim 11 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When unifying a pattern containing variables with a Horn clause whose head part does not contain variables,
"Pattern" in the processing method according to any one of claims 2 to 4 of Patent Document 1
and there is binding information (C) that makes the result of processing using the head part as a "constant character string" in the same processing method as true, and the body part of the Horn clause as a "subgoal" of the processing method according to claim 5 or 6, if the result is true, return true together with the binding information (C),
A processing method according to any one of claims 1 to 10 of Patent Document 1 is provided.
As a result, all solution candidates (for each variable) between the pattern containing the variable and the (static) head (not containing the variable) are identified and returned as the "result". ”
The basic mechanism for overcoming the problem can be simply provided, and not only the visualization and verification of the reasoning process, but also the learning through the user's hands such as the modification of the rules can be facilitated.
In addition, it simply transforms into simpler and recursive (or mutually recursive) subproblems with the same structure, making it easier to display and verify (completeness and soundness) the reasoning process.
In addition, sound mechanical interpretation when the head of the rule that expresses variables in a form embedded in natural language does not contain variables, a seamless, easy-to-understand, simple and unified processing mechanism, and the inference process corresponding to these can give an explanation.
(0020)
Further, claim 12 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When unifying a pattern containing variables with a Horn clause with a head containing variables,
Returns false if the number of fixed characters that exist in the head, tail, or both of the pattern and the head part do not match,
A processing method according to any one of claims 1 to 11 of Patent Document 1 is provided.
As a result, it is possible to significantly save the amount of calculation and improve the processing speed as compared with a uniform search.
(0021)
In addition, claim 13 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
When unifying a first pattern containing variables with a Horn clause with a head containing variables,
If the number of fixed characters that exist in the head and tail of the pattern and the head part match,
Each obtained by binding the head part of the Horn clause with each constraint information that the result of applying the body part of the Horn clause as a "subgoal" in the processing method according to claim 5 or 6 of Patent Document 1 is true. Regarding the second pattern,
If the second pattern does not contain variables,
Processed using the first pattern as the "pattern" in the processing method according to any one of claims 2 to 4 of Patent Document 1 and using the second pattern as the "constant character string" in the same processing method return the result,
A processing method according to any one of claims 1 to 12 is provided.
Here, "when the second pattern does not contain variables" means that in the Horn clause (rule), all variables included in the head part are also included in the body part, and all variables in the body part As a result of binding variables with constants (fixed strings), all variables in the head part are also constants (
It means that the variable in the head part has disappeared and the head part can be regarded as a constant string by being bound (assigned) with a fixed string).
As a result, unification of patterns containing variables and Horn clauses (rules) with heads containing variables can be performed mechanically and automatically in a flow that is easy for the user to trace, using a simple computational model.
In other words, all candidate solutions (things to be assigned to each variable of the pattern) between the pattern that includes the variable and the head part that no longer includes the variable (by each binding information) (=converted to a constant string) are identified and " The original, more complex problem (unification of variables vs. variables)
By reducing (converting the problem), the basic mechanism for overcoming the above "difficulty A" is simply provided, and it is easy to express as a thought process. In addition, in this "reduction (problem conversion)", it is simply converted to a simpler and recursive (or mutually recursive) subproblem with the same structure, so it is possible to visualize and verify the inference process (complete It facilitates learning through user's hands such as correction of rules, etc.
After all, a sound mechanical interpretation when the head of the rule that expresses the variable in a natural language includes the variable, a seamless, easy-to-understand, simple and unified processing mechanism, and an explanation of the reasoning process corresponding to these can be easily given.
(0022)
In addition, claim 14 of Patent Document 1 is
A processing method according to any one of Claims 1 to 13 of Patent Document 1, characterized in that the computer restricts input of facts including variables.
Here, facts including variables (knowledge sources that are not rules) are, for example, "$X is $X." There is an unfounded fact that it becomes true if only superficial sentence patterns match).
Restricting these inputs to the computer simplifies the computation of automatic unification processing between patterns (sentences that may contain variables) and facts (sentences that do not contain variables (=fixed character strings)). , high-speed response by automatic processing is possible. It also makes it difficult to mix baseless facts into knowledge sources.
(0023)
In addition, claim 15 of Patent Document 1 is
The rule is in a form similar to a so-called Horn clause, and the computer restricts that the head part contains variables that do not appear in the body part. The described processing method is provided.
Here, "the computer restricts the head part from including variables that do not appear in the body part" means that the head part is "$X is nice" and the body part is "$Y is strong, $Y is $Y". Z-friendly, $Z is a child, $Y is clever," even if all the variables in the body ($Y and $Z in this example) are bound (assigned), the variables in the head are The computer restricts rules that leave (in this example, $X) unbound from being entered, stored, or activated as a knowledge source. Such rules lead to leaving unbound variables in the head part, as in the case of facts, so they tend to produce unsound inference results and also cause the speed of inference to decrease.
By restricting these rules from being entered into the computer, patterns (
Sentences that can contain variables) and head parts (which can be reduced to a set of sentences that do not contain variables (=fixed character strings) according to claim 13 of Patent Document 1) are simplified in automatic unification processing calculations, A high-speed response is possible through automatic processing. In addition, it makes it difficult to produce baseless inference results.
(0024)
In addition, claim 16 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
automatic unification of patterns with unbound variables remaining with each other and heads or facts,
Characterized by finding and returning a set of minimal solutions as a fixed string,
A processing method according to any one of claims 1 to 13 of Patent Document 1 is provided.
Here, as an example of "the set of solutions that are minimal as fixed strings", {($ A = "also", $B = <empty>, $
C=<empty>, $D=”was”), ($A=empty, $B=”was”, $C=”was”, $D=empty)}
can provide two solutions.
This makes it possible to provide an example solution (especially the smallest simple example) even when there are an infinite number of solutions.
(0025)
In addition, claim 17 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
automatic unification of patterns with unbound variables remaining with each other and heads or facts,
Characterized by finding and returning a set of minimal solutions as a string containing variables,
A processing method according to any one of claims 1 to 13 of Patent Document 1 is provided.
Here, as an example of "the set of solutions that are minimal as fixed strings", {($ A = "$C also", $B = <free>, $C = <free>, $D = "is $B"), ($A = <free>, $B = "also $D", $ C.
="$A is", $D=<free>)}, and so on. Assume that the sum (through solving) of the string lengths of values that bind variables (values that are assigned to variables) (however, if variables are included, each occurrence of the variable is regarded as one character) is the minimum be able to.
This allows us to provide a general form of the solution even when there are an infinite number of solutions.
(0026)
In addition, claim 18 of Patent Document 1 is
If the number of fixed characters that are commonly present at the beginning and end of each string of the pattern in which the unbound variable remains and the head part or fact do not match,
characterized by returning false,
A processing method according to claim 16 or 17 of Patent Document 1 is provided.
This saves automatic searches with no unification possibilities and speeds up automatic processing.
(0027)
In addition, claim 19 of Patent Document 1 is
When obtaining a minimum set of solutions as a string containing the fixed string or variable,
Each variable contained in both strings consisting of patterns and heads or facts (let it be the first string and the second string) with unbound variables remaining with each other is commonly bound within each string. and independently bound between strings (or while the scope of commonly binding variables with the same name is only within each string (or between strings is outside the scope of the variable)),
For each variable any substring (<empty> and (including the entire string) or by substituting itself and searching for a case where both strings match to find a set of minimal solutions,
A processing method according to claim 16 or 17 of Patent Document 1 is provided.
"Or substitute itself..." means to substitute "$X" itself for "$X" in the string, and it happens that both the first string and the second string When $X (with different scopes) is included, make sure to distinguish between $X in the first string and $X in the second string (for example, "$X1" and "$ X2", etc.) can also make an apparently sane calculation.
As a result, it is possible to ensure a certain degree of completeness (completeness depending on the nature of the problem (pattern pair)) of the solution set while ensuring soundness and termination (finiteness of computation).
(0028)
Further, claim 20 of Patent Document 1 is
Rather than just the strings consisting of the pattern and the head part or fact with the unbound variables remaining with each other,
Both strings corresponding to the number of fixed characters that exist in common at the beginning and end of each string of patterns and heads or facts in which unbound variables remain with each other, excluding the number of fixed characters that remain from both strings about,
A processing method for performing the processing described in claim 19 of Patent Document 1 is provided.
This makes it possible to reduce the space of the full search while maintaining a certain degree of coverage (or completeness in some cases) of the solution set, thereby speeding up the processing.
For example, as a process of automatic unification between "Ishikawa's $A likes $B." and "Ishikawa's $C likes $D ice cream." Delete the common part of "$A is $B"
and "$C and $D ice cream"
After that, all substrings of "$C and $D ice" are added to each of "$A" and "$B" by a full search (this is a variable such as $C and $D (${...・} format variables are the same)
is 6 characters, so while substituting 6 x (6 + 1) / 2 + 1 (<empty> part) + 1 (substituting itself) = 23 ways), similarly, "$C" and "$D" respectively all substrings of "$A is $B" (which is also 3 characters, so 3*(3+1)
/2 + 1 + 1 = 8 ways) can be substituted for all both strings (23 x 23 x 8 x 8
street), if both strings match (for example, (
``Moha Ice'',
”Conger eel ice cream”,
"$A hamo ice cream",
"$C moha ice cream",
"$A is $D ice cream"
*By the way, the set of assignment pairs {
($A = "also", $B = "ice", $C = <empty>, $D = "ha"),
($A=<Empty>, $B=”ice cream”, $C=”was”, $D=<Empty>),
($A = "$A", $B = "ice cream", $C = "$A is", $D = <empty>),
($A="$C mo", $B="ice", $C="$C", $D="wa"),
($A = "$A", $B = "$D ice cream", $C = "$A is", $D = "$D")
} or a set of partial substitutions for each substitution tuple (e.g. {
($A="mo", $B="ice"),
($A=<Empty>, $B=“ice cream”),
($A = "$A", $B = "ice cream"),
($A="$C mo", $B="ice"),
($A="$A", $B="mo$D ice")
})
can be answered as a solution.
In addition, $A = "$A" and $B = "$D ice" are free to be $A and $D, and any character string can be substituted for them (i.e., It is possible to show that there are infinite solutions and how the infinite number of solutions (character strings) should be constrained (in the form of regular expressions).
(0029)
In addition, claim 21 of Patent Document 1 is
Substitute any substring (including <empty> and the entire string) of the target string for each variable or itself, and search for cases where both strings match. when asking for
When a variable is included in a substring to be assigned to a variable, while replacing the variable included in the substring with a new variable that did not exist in both strings, a similar assignment to the new variable The processing method according to Claim 19 or 20 of Patent Document 1 is provided, characterized in that a full search is performed recursively thereafter to search for cases where both character strings match.
As a result, it is necessary to search for a solution in which a fixed character in the self string is reflexively substituted into a variable in the self string via a variable in the other string (for example, both strings = { "$X is $
X","i$Yu"}, as a solution candidate
can be enumerated), it is possible to further improve the comprehensiveness of the enumeration of solution candidates while maintaining the soundness of the solution search. In the example above, both strings (also called "pattern pairs") =
From {"$X is $X", "i$Yu"}, ($X="iu", $Y="yes") can be enumerated recursively by the present invention ($X ＝”I${Y1}U”)⇒(${
Y1}=<Empty>)⇒($Y=“Uhai”) search branch (a part of the search tree, which can be generated in a finite number of times in the case of breadth-first search according to the present invention). You can show it by doing it.
"Recursively ... exhaustive search" means, for example, that a matching substitution sequence (For example, the ($
X=“I${Y1}U”)⇒(${Y1}=<empty>)⇒($Y=“Uhai”), etc.). A variation of the first assignment is that if the types of variables in both strings are 2 and 3, respectively, and the length of the strings is 6 and 7 characters (variables are counted together as 1 character), then 2×(7×(7+1)/2+1+1)+3×(6×(6+1)/2+1+1
)=60+69=129. When performing breadth-first search so that the maximum depth of further branches derived from these branches is uniform, by setting a limiter (such as a program step that compares with the upper limit value) based on the number of assignments, the search is performed once can be restricted to an appropriate amount of computation (to secure termination and response speed that are no longer guaranteed by recursion). Note that even in the case of depth-first search, the amount of calculation can be limited by limiting the depth, the number of substitutions, and the like. In any case, an infinite loop due to self-recursion or mutual recursion is provided with a stack structure (array of structures), etc. that manages recursive function calls, and the call parameter (assignment sequence) of the ancestor branch (in the direction of the root) (may contain the state representation of both strings depending on the assignment context)
can be avoided by comparing with the current call parameters. In addition, it is possible to prune branches that cannot be unified depending on whether or not the fixed character strings existing at the beginning and end of both character strings match for the number of commonly existing characters. If no solution is found by searching with a limited amount of computation (for example, one million substitutions or a search depth of 20 or less), "No solution (
Accuracy: Substitution million times and search depth within 20)”, and if found, “($X = “Yes”, $Y = “Yes”), (...),... (However, , Accuracy: 1 million substitutions and a search depth of 20 or less)”, etc., while indicating the search accuracy. In addition, when the full search is clearly possible within the limit number of assignments and within the limit search depth (when there is completeness)
to that effect, "No solution (completely searched)" or "($X="yes", $Y="yes"
), (.
(0030)
In addition, claim 22 of Patent Document 1 is
Regarding the result of assigning to a variable, the remaining two strings after removing the number of fixed characters that exist in common at the beginning and end of each string are
Exists at the beginning or end of the string as the search progresses as the "partner's string" that is the target of "any substring of the partner's string (including <empty> and the entire string)" Characterized by reducing the number of fixed characters,
A processing method according to any one of claims 19 to 21 of Patent Document 1 is provided.
As a result, the number of branches decreases as the search progresses (as the search progresses toward the previous branch), and the amount of calculation can be saved.
(0031)
In addition, claim 23 of Patent Document 1 is
When obtaining a minimum set of solutions as a string containing the fixed string or variable,
Each variable contained in both strings consisting of patterns and heads or facts (let it be the first string and the second string) with unbound variables remaining with each other is commonly bound within each string. and independently bound between strings (or while the scope of commonly binding variables with the same name is only within each string (or between strings is outside the scope of the variable)),
A processing method for obtaining a set of solutions by recursively substituting a fixed character string (including an empty character) or a character string containing a variable into each variable,
Limit the variable to be substituted to the corresponding head or tail variable of the self string where the head or tail of the other string is a fixed character,
If the value is the head, the new variable or <empty> following the head fixed character or fixed character string, and if the tail, the tail fixed character or fixed string following the new variable or <Empty>,
Repeatedly removing fixed characters for the number of characters common to the beginning and end of each character string from the result of assigning to the variable,
There is provided a processing method according to claim 16 or 17 of Patent Document 1, characterized by:
This makes it possible to significantly limit the range of enumeration of solutions (without impairing the exhaustiveness of the search) by focusing on the presence of fixed characters at the beginning and end of the search. Become.
(0032)
In addition, claim 24 of Patent Document 1 is
When obtaining a minimum set of solutions as a string containing the fixed string or variable,
Each variable contained in both strings consisting of patterns and heads or facts (let it be the first string and the second string) with unbound variables remaining with each other is commonly bound within each string. and independently bound between strings (or while the scope of commonly binding variables with the same name is only within each string (or between strings is outside the scope of the variable)),
A processing method for obtaining a set of solutions by recursively substituting a fixed character string (including an empty character) for each variable,
Binding information that can be created by using only fixed character types (character types excluding variable identifiers and delimiters) in both strings, except for the number of common fixed characters that exist at the beginning and end of each string Obtained by enumerating in order from the smallest total string length of the value,
A processing method according to claim 16 or 17 of Patent Document 1 is provided.
This makes it possible to efficiently reach and enumerate solutions for problems with certain properties.
(0033)
In addition, claim 25 of Patent Document 1 is
Based on the obtained "set of solutions that are minimal as strings containing variables",
Provided is a processing method according to Claims 19 to 24 of Patent Document 1, in which a search is continued by a Horn clause that can be obtained by temporarily rewriting a variable of the Horn clause with a character string that includes the variable.
This results in a computed constraint that is necessary to unify the pattern and the head of the Horn clause (rule) (in the example above, the variable "$C" is a There is also a unification policy (for example, in the case of the last assignment tuple above) that has a constraint that unification can only be performed by binding (assigning) with a character string), the body part of the horn clause The search process can be propagated and the search can be made more efficient.
(0034)
In addition, claim 26 of Patent Document 1 is
Based on the obtained "set of solutions that are minimal as strings containing variables",
A set of fixed character string pairs obtained by substituting each element of a set of fixed character string pairs predefined as a set of solutions of the "fact including variable" into the variable part of "character string including variable" return,
A processing method according to any one of claims 19 to 25 of Patent Document 1 is provided.
As a result, a database (for example, "word equation (registered trademark)", "V
LANK (registered trademark)” can be efficiently linked to the solution search. Also, a large number of facts with similar sentence patterns can be efficiently managed in cooperation with the framework of the present invention.
(0035)
In addition, claim 27 of Patent Document 1 is
In the automatic unification process or automatic derivation process,
For automatic unification between a first pattern containing variables and a second pattern containing variables:
If the number of fixed characters that exist at the beginning and end of both patterns match, and the same variable appears only once in each pattern,
returns information indicating the existence of an infinite number of solutions,
A processing method according to any one of claims 1 to 26 of US Pat.
(0036)
In addition, claim 28 of Patent Document 1 is
An apparatus is provided for carrying out the method of any one of claims 1 to 27 of US Pat.
(0037)
In addition, claim 29 of Patent Document 1 is
A computer program for causing a computer to execute the method according to any one of claims 1 to 27 of Patent Document 1 is provided.
(Effect of the Invention of Patent Document 1)
(0038)
The effects of the present invention are as described above for the claims of each Patent Document 1.
(Brief description of the drawings in Patent Document 1)
(0039)
(FIG. 1) FIG. 1 is an overview of a network system that most effectively implements the present invention. (Example 1)
(FIG. 2) FIG. 2 is an explanatory diagram showing the hardware configuration of a computer that implements the present invention. (Example 1)
(FIG. 3) FIG. 3 is an explanatory diagram showing an example of screen display on a terminal implementing the present invention.
(Example 1)
(Mode for implementing the invention of Patent Document 1)
(0040)
The present invention can be implemented on any kind of computer (cloud server, large general-purpose machine, desktop PC, notebook PC, mobile terminal, dedicated machine for the present invention) connected/disconnected to a network, but in general A form implemented by a PC connected to the Internet and having a multitasking function and a window function will be described.
FIG. 1 is an overall view of the network system that most effectively implements the present invention. A content management device 1 allows a user PC 2 or a user/administrator PC 2 to manage content (rules and facts) handled by the present invention. A known file sharing server that accepts posts and distributes content while controlling sharing access rights and content provision priority according to a predetermined policy may be used. The user PC 2 inputs and stores contents (rules and facts) expressing the contents (or the contents to be considered) that the user believes (or the contents that the user wants to consider) in a natural language according to the method of the present invention. A processing computer, mobile terminal, or the like.
FIG. 2 is an explanatory diagram showing the hardware configuration of the computer used in the present invention. This is a general configuration of a computer, and the feature of the present invention is that a CP is generated using natural laws in accordance with the instructions and descriptions of the program 3 and content data 4 stored in the nonvolatile memory 2h.
It lies in the content of the processing processed by U2e and the nature of the interaction protocol (format) with the user.
FIG. 3 is an explanatory diagram showing an example of screen display on a terminal that implements the present invention. In the illustration, "
? In response to the question (in the open question format), "Do not do $A
=
In the same manner, the answer "$A = bullying/9/" is given by using the fact that the ninth line of the list L "does not bully".
(0041)
In such an embodiment, a computer program written in, for example, the C++ language can cause a computer (such as a PC) to perform any action possible. Therefore, the well-known basic operations of computers using the laws of nature (storing and retrieving to and from registers, memory, and storage, four arithmetic operations, comparison, etc.) and general functions that have already become standard libraries for development environments, etc. (Assignment/comparison/copy to variables and N-dimensional arrays, manipulation/comparison/display of character strings and numerical values, manipulation of list structures, queue structures, hash structures, vector types, etc.
Binary search and sorting operations, pattern matching operations using regular expressions, network communication operations, etc.) are different from the essence of the present invention (unique configuration leading to novelty and inventive step), and are well understood by those skilled in the art. Since it is known, I will not go into detail. On the other hand, the operation order and storage method (algorithm and data structure) for realizing the processing unique to the present invention by combining such general functions will be described in detail below. By providing the description of this algorithm, etc., the description of the above-described solution to the problem, and the description of the drawings of the present invention for reference in development, a standard-level professional programmer will be able to use standard application program development techniques ( GUI application integrated development environment by C++, etc.)
It is clearly possible to make a computer carry out the method of the present invention, and it is also possible to create an apparatus and a program for this method.
(0042)
The algorithm shown below is an algorithm based on standard structured programming in which a desired "function" is newly defined using C language/C++ language, etc., and the function is called. Processing means (program (or computer) can be realized.
(0043)
■ Processing algorithm specification of knowledge list L for inquiry P
================================================== =============================
======================================
Function Q1 (P: pattern that can include variables, L: knowledge list): *Example 1 of Patent Document 1 of the main function common to the claims of Patent Document 1*
{
1. Take the content B (fact or horn clause) of the first row of L.
However, if L is empty (if the first row cannot be retrieved), output false and terminate.
2. Set the initial value of the variable binding information C to empty, call the following function H (P, B, C), and output the binding information (=solution set) C that is true (=matches).
In the case of true (=matched) Output true (yes) (output the line number of the matched content B as well).
If false, do nothing.
3. If there is a content B' in the line next to L. ・Set B←B' and move to 2.
If not ・If true has never been output in 2, output false (no).
·end.
}
(0044)
■Processing algorithm specification of knowledge list L for inquiry P (Pre-fixed character string version)
================================================== ==============================
=====================================
Function Q2 (P: pattern that can include variables, L: knowledge list): *Embodiment 2 of the main function common to the claims of Patent Document 1 (prefixed character string version)*
{
(preliminary fixed string conversion)
1. Determine a fixed character string text T whose lines are {all character strings that can be deduced only from L/line number set of L used as the basis for deducing each character string}.
Specifically, the output series (A = Solution 1/Evidence 1, A = Solution 2
/Evidence 2, .
If an infinite loop occurs during the calculation, the deduction is partially omitted by an infinite loop detection function using an explicit call management stack or the like.
・For example, while the computer is checking the truth of the Horn clause under certain variable binding conditions, check whether the same Horn clause is being checked under the same variable binding conditions.
Manage a call management stack whose members are structures (Horn clause identifier being checked, current variable binding conditions),
Detects infinite loops by linearly searching for the same combination towards the root of the stack at call time.

(Since it is only a fixed character string, a simpler method than Q1 (function F) is used to find the solution set)
2. For each line t of text T, output all C for which the following function F(P, t, C) is true.
(output of final result)

3. Outputs true if there is at least one C, otherwise outputs false.
}
(0045)
■ A processing algorithm specification for one fact (fixed character string) or Horn clause B (however, all variables contained in the head also exist in the body) for the query P
================================================== =============================
======================================
Function H (P: pattern that can contain variables, B: one fact (fixed string) or Horn clause, C
: Variable binding information): Patent Document 1 of functions used in the inventions of claims 1 to 15 of Patent Document 1
Example 1-1 of
{
If P contains no variables {
If B is true {
Returns true if P=B.
Returns false otherwise.
}
If B is a Horn clause {
If the head part of B contains no variables {
If P=head part of B, returns true/false of function D(body part of B, E).
Returns false otherwise.
}
If the head part of B contains variables {
For each C1 for which F(head part of B, P, C1) is true,
Returns true if there is C1 for which the result of function D (applying C1 to the body of B, E) is true.
Returns false otherwise.
}
}
}
If P contains variables {
If B is true {
If there exists a C1 for which F(P,B,C1) is true, all such C1
returns true in addition to
If no such C1 exists, return false.
}
If B is a Horn clause {
If the head part of B contains no variables {
If there exists C1 for which F(P, head of B, C1) is true, and if function D(body of B, E) is true, add all such C1 to C return true.
Returns false otherwise.
}

If the head part of B contains variables {
★When implementing the invention of claim 12 of Patent Document 1★
//Ingenuity for speeding up (optional) start When fixed characters for the number of characters existing at the beginning and end of the head part of P and B do not match {
return false.
}
//Ingenuity for speeding up (optional) end

★ Implementation of the invention of claim 13 of Patent Document 1 ★
When fixed characters corresponding to the number of characters existing at the head and tail of the head part of P and B match {
Any binding information G in E for which the result of function D (body part of B, E) is true
, the function F (the head part of P and B bound by G (should be a fixed string),
If there is C1 that makes C1) true
Add all such C1 to C and return true.
Returns false if no such C1 exists.
}
}
}
}
}
(0046)
■ Enumeration algorithm specifications for the solution set of the body part of the Horn clause (or compound sentence queries that share variables)
================================================== =============================
======================================
Call example → function D ("$A and $B are friends, $A is human, $B is human", E);
Return example ← True (E = ((A = Hanako, B = Taro), (A = Taro, B = Jiro), (A = Jiro, B = Saburo)))
================================================== =============================
======================================
Function D (J: body part (or compound sentence query that shares variables), E: solution set): * Example 1-2A of Patent Document 1 of the function used in the invention of claim 6 of Patent Document 1 *
*However, the entire knowledge source L can be stored in a global variable and referenced {
1. Take the first sentence M in J. Returns true if the first sentence does not exist.
2. For the entire knowledge source L, call the function Q1(M,L) or Q2(M,L).
Returns false if the result is false.
when true if M contains no variables
do nothing.
If M contains variables
A set C of binding information that can be taken to make M true is stored.
* Variables that do not need to be bound to be true can also be explicitly indicated as such.
*If the bound value is an infinite set and requires a certain condition to be true, the condition can be added.
3. Fetch the next sentence M' in J. If it does not exist, return true (E=set C of binding information).
4. For the entire knowledge source L, call the function Q1(M',L) or Q2(M',L).
Returns false if the result is false.
when true if M' does not contain variables
do nothing.
If M' contains variables
A set C' of binding information that can be taken to make M' true and a stored set C
Let the new C be the intersection of
However, if the original set C is not an empty set but becomes an empty set by taking the product set, false is returned.
* Variables that do not need to be bound to be true can also be explicitly indicated as such.
*If the bound value is an infinite set and requires a certain condition to be true, the condition can be added.
Go back to 5.3.
}

Function D2 (J: body part (or compound sentence query that shares variables), E: solution set):
Another example 1-2B of Patent Document 1 of the function used in the invention of claim 6 of Patent Document 1 (improved speed by spreading binding information to other statements using recursive call of function) ★
{
1. Take the first sentence M in J. Returns true if M does not exist.
2. For the entire knowledge source L, call Q1(M,L) or Q2(M,L).
Returns false if the result is false.
If true, if M does not contain a variable, return true if there is no sentence following J. If so, set the next sentence as a new M and return to 2.
When M contains variables If there is no statement following J, possible binding information is stored in E and true is returned.
3. E is emptied, and for each binding information K in set C, a recursive call D2 (body parts J', E'
) if the result is true add to E that K and E' that became true 4. Returns false if it has never been true for each K, and returns true (E) if it has been true even once.}

Function D3 (J: body part (or compound sentence query that shares variables), E: solution set):
Example 1-3* of Patent Document 1 of the function used in the invention of claim 5 of Patent Document 1
{
1. Take the first sentence M1 in J. Returns true if the first sentence does not exist.
2. For the entire knowledge source L, call Q1(M1,L) or Q2(M1,L).
Returns false if the result is false.
when true if M1 does not contain variables
do nothing.
If M1 contains variables
A set C1 of possible binding information is stored.
3. Similarly, C2, C3, . . . are obtained for M2, M3, .
Returns false if false.
4. Find the intersection C for the remaining C1, C2, C3, . . . Cn and return true (C).
}

(0047)
■ Enumeration algorithm specifications for multiple length matching solutions (multiple possible) of fact S for query P
================================================== =============================
======================================
Calling example → Function F ("$A and $B are friends", "Taro, Jiro, and Saburo are friends", "");
Function F (P: pattern that can include variables, S: constant character string, C: variable binding information): * Invention of claim 2 of Patent Document 1 (however, "(or last)" and "or before" Example 1-4A of Patent Document 1 of the function used in the omitted implementation example)
{
(preparation)
1. Obtain the number of types of variables V in P and the names of appearing variables N1, N2, ... in order of appearance.

(Corresponding to the case where P originally does not contain variables)
2. When V is 0,
⇒ When P=S, return true and end.
⇒ When P≠S, return false and end.

(output)
3. When V is 1, no matter how many times N1 appears in P, there is only one solution R1 to the unification of P and S, so
⇒ If there is a match, output the unique solution as "C and N1=R1", return true, and end.
⇒ If there is no match, return false and exit.
* Unification itself can be performed as a match with a regular expression (with postfix reference if the same variable appears more than once)

4. When V is 2 or more Longest matching solution R1 for variable N1 Executes recursive call F(P', S, C and N1=R1) for P' created by filling all N1 in P ⇒ Longest for N1 A solution set that is considered to be a consistent solution is output.

5. R1'←Recursive call F(P'', S, C and N1=R1') for P'' created by filling all N1 in P with R1' as a character string excluding the last character of R1 Execution ⇒ For N1, a solution set is output as a solution candidate R1' that is one character removed from the longest matching solution.

6. When R1' is 1 or more characters, (in claim 3 of Patent Document 1, when it is greater than the number of characters of the shortest matching solution)
⇒ Return to 5 as R1←R1'.
In the case of 0 characters, (* in claim 3 of Patent Document 1, the number of characters in the shortest matching solution)
end.
}

(0048)
Function F (P: pattern that can include variables, S: constant character string, C: binding information of variables): ★Embodiment 1-4B of the function used in the invention of claim 4★
// Function F with speed-up device 1 focusing on the first character C of the fixed character string immediately after the variable first appears in P
{
(preparation)
1. Obtain the number of types of variables V in P and the names of appearing variables N1, N2, ... in order of appearance.

(Corresponding to the case where P originally does not contain variables)
2. When V is 0,
⇒ When P=S, return true and end.
⇒ When P≠S, return false and end.

(output)
3. When V is 1, no matter how many times N1 appears in P, there is only one solution R1 to the unification of P and S, so
⇒ If there is a match, output the unique solution as "C and N1=R1", and TR
Return UE and exit.
⇒ If there is no match, return false and exit.

4. (Improved speed 2. This process is covered in 5 and later, so you can skip it.)
This device 2 is not included in the claims of Patent Document 1★)
When V is 2 or more,
A solution that matches all variables (N1 (=N), N2, ...) in the longest matching mode and a solution that matches all variables (N1 (=N), N2, ...) in the shortest matching mode. When the seen solution and are the same (N1=R1, N2=R2, …) ⇒ output the unique solution as “C and N1=R1 and N2=R2 and …” and finish Match in the first place If you did not
⇒ Return false and exit.
If not the same ⇒ continue to 5

5. When V is 2 or more Longest matching solution R1 for variable N1 Executes recursive call F(P', S, C and N1=R1) for P' created by filling all N1 in P ⇒ Longest for N1 A solution set that is considered to be a consistent solution is output.

6. When variable N1 (self) follows immediately after variable N1 in P ⇒ processed after 7 When another variable N2 follows ⇒ processed after 7 is the first character of
When fixed characters continue ⇒ Process with 7 or later

7. One character C, which is a fixed character immediately after the first appearance of variable N1 in P (the first character C of R2 when the immediately following is another variable N2),
If it is included in the longest matching solution R1 ⇒ 1) R1' ← A string obtained by deleting the substring from the end of the string of R1 to the C ⇒ 2) P'' ← R1 ' for all N in P ⇒ 3) Execute recursive call F (P'', S, C and N1=R1')
⇒If there is a solution that is one step shorter for N1, it should be possible to output it. If it is not included in the longest matching solution R1 ⇒Since enumeration of shorter solutions for N1 is finished, end.

8. Return to 7 as R1←R1'.
}
================================================== =============================
======================================
The above is an excerpt from Patent Document 1 (Japanese Patent Application No. 2014-087008).

FIG. 1 illustrates a logic program (fixed facts and rules), input as assumptions by the user, in one embodiment where the user expresses the user's idea synthesis request in terms of assumptions (denoted by the "!!" symbol). FIG. 10 is a diagram for explaining the relationship between facts that are added, facts that are temporarily added as hypotheses by the user, and contents that are output as differences. The 5th to 13th rows of the in-memory array are shown, and the rest are omitted. The underline on the 5th line indicates the "synthesis rule" of claim 1, and the underlining on the 12th and 13th lines indicates the fact temporarily added as an assumption.

FIG. 2 is a diagram illustrating the main processing flowchart (steps as a method) of a computer program for the embodiment of FIG. 1 in which the user expresses the user's request through assumptions.
Step DS1 is a step in which the user inputs a set of assumed facts in this embodiment. "!!" indicates a hypothesis, and ";" indicates a sentence break.
Step DS2 is a step of listing all automatically provable propositions (fixed character strings) by "$A" in this embodiment.
Step DS3 is the step of temporarily adding the set of hypothesized facts to memory adjacent to the logic program.
Step DS4 is a step of enumerating all propositions (fixed character strings) that can be automatically proved by $A on the assumption that assumptions are temporarily added by DS3.
Step DS5 is a step of outputting a difference set between the set enumerated in DS4 and the set enumerated in DS2.
Step DS6 is the step of deleting the previously added set of hypothetical facts (for testing other hypotheses, etc.) from memory adjacent to the logic program.

FIG. 3 is an explanatory diagram showing a general configuration as a computer that implements. Since there are no unique features of the present invention here, it will not be described in detail.

4 General computer used in the present invention (user-dedicated personal computer, shared file server, web server)
4a-4k general components of a general computer 4 (user-dedicated personal computer, etc.) 21 logic program (fixed facts and rules, 5th line underlined "synthesis rule")
22 Facts added as assumptions 23 Facts temporarily added as assumptions (underlined)
24 Output difference set DS1 In one embodiment, step DS2 where the user inputs a set of hypothesized facts. Add a set of facts to the logic program Step DS4 After adding by DS3, $A lists all automatically provable propositions (fixed character strings) Step DS5 The set enumerated in DS4 and the set enumerated in DS2 Step DS6 of outputting a difference set of

Claims (8)

A computer program that includes some or all of the functions of PROLOG and causes a computer to execute processing according to a separate natural language-based logic program,
wherein the computer program allows the use of literals embedded with variables in natural language in the logic program;
the computer program enables natural language pattern matching in conjunction with recursive syllogisms;
The logic program includes a rule (hereinafter referred to as a "synthesis rule") in which the variables included in the literals of the body on the right side of the rule corresponding to the PROLOG rule do not include all types of variables included in the head on the left side,
The logic program is designed so that different literals in the body on the right side of the synthesis rule (however, the scope of the included variables is common within the synthesis rule) can separately match multiple texts in multiple fields. ,
A computer program that allows concepts from multiple fields to be combined to construct new concepts in head literals of said composition rules, comprising:
In a question, fact, rule condition or rule effect entered by a user, or in a logic program fact, rule condition or rule effect,
In addition to the affirmation (true) as the existence of evidence facts or the success of automatic proof and its denial (absence of evidence facts and automatic proof failure/impossibility),
the computer program allows the expression "impossible" to signify the existence of evidence to the contrary or the success of automatic proof to the contrary;
When the computer program uses as a basis an affirmative expression that does not contain the variable from the beginning or does not contain the variable due to variable binding,
The computer program automatically confirms that the "impossible" expression corresponding to the affirmative expression does not exist and cannot be automatically proved,
causing the computer program to limit the use of the corresponding affirmative expression if the "impossible" expression exists or can be automatically proven
computer program. each text that constitutes the plurality of texts in the plurality of fields is composed of one or more sentences that are separated from each other;
other sentences in the text to which the sentence that matches one of the above body conditions belongs,
2. The computer program according to claim 1, wherein control is performed so as to match other conditions of the same body with priority. a computer program
For the purpose of collecting related sentences by field or by instance (article) and making it easier to search as a condition for natural language pattern matching with recursive syllogism for automatic synthesis according to user requests,
In conjunction with the computer program of claim 1 or 2,
In order to prevent notational variations and duplication of pattern parts and element concepts,
Fields, themes, patterns (templates), or already provided data are specified to users Significant data (knowledge and information) in that pattern (language or template) for that category (field or theme) by users Solicit offers of computer programs. As requested by the user,
Accepts a literal expression in which variables are embedded in natural language,
4. The computer program according to any one of claims 1 to 3, wherein an idea requested by a user is automatically synthesized by automatically obtaining a solution that binds embedded variables. As requested by the user,
Accepts assumptions consisting of one or more sentences (fixed character strings) in natural language as facts ,
A set of sentences (fixed character strings) that can be automatically proved by adding the assumption to the logic program as a fact (the fact originally included in the logic program is also the "impossible" of claim 1 corresponding to the affirmative expression " is included in this set of sentences (fixed strings) unless the expression "" exists and can be automatically proved) and the set of sentences (fixed strings) that could be automatically proved before adding the assumption to the logic program as a fact (Facts originally included in the logic program are also included in this set of sentences (fixed character strings) unless the expression "impossible" in claim 1 corresponding to the affirmative expression does not exist and cannot be automatically proved.) By finding the difference set of
4. A computer program as claimed in any one of claims 1 to 3, characterized by automatically synthesizing ideas requested by a user. As requested by the user,
accepting a temporary context consisting of one or more sentences in natural language;
4. The computer program according to any one of claims 1 to 3, characterized in that it automatically synthesizes the ideas requested by the user by forward inference from sentences contained in a temporary context. Based on the content that defines the order of items (or item sets) in a given set of values,
7. A computer program as claimed in any one of claims 1 to 6 , characterized in that it restricts the strings used to synthesize the concepts requested by the user. Based on the content that defines the order of items (or item sets) in a given set of values,
8. A computer program as claimed in any one of claims 1 to 7 , for controlling the priority of strings used to synthesize user-requested concepts.

Images