JP2012248142A - Sentence creation support natural sentence processing method for displaying meaning in detail of input sentence - Google Patents

Abstract

PROBLEM TO BE SOLVED: To construct semantic structure for expressing meanings of an input natural sentence, to subdivide the semantic structure, to generate a natural sentence from the subdivided semantic structure, and to represent meanings in details of the natural sentence.SOLUTION: Morphological analysis and syntactic analysis are performed to an input natural sentence, semantic construction processing is performed to analysis results to construct partial semantic structure, and the whole semantic structure of the input natural sentence is constructed while coupling the partial semantic structure by a semantic root. When the semantic structure is subdivided by the semantic root, a natural sentence is generated from the subdivided semantic structure and clearly indicated, meanings of the input natural sentence are verified in the details. A writer creates a sentence by using a grammar and knowledge, however, a reader does not necessarily share the same knowledge with the writer. In the present invention, semantic analysis is mainly performed only by grammar information by standing at a place of the reader who does not have the knowledge. Thus, the writer is supported so as to create a sentence which can be accurately understood even by the reader who does not have the knowledge. Such a sentence creation support natural sentence processing method is provided.

Description

When morphological analysis and syntax analysis are performed on a natural sentence input by a word processor or the like, a morphological analysis result and a syntax analysis result are obtained. A partial semantic structure can be constructed by applying a semantic construction rule to the morphological analysis result and the syntax analysis result. By combining these partial semantic structures with semantic roots, the entire semantic structure of the input natural sentence can be constructed. If a natural sentence generation path is formed in the entire semantic structure, an entire natural sentence can be generated from the semantic structure. Since the semantic structure is connected by the semantic root, if the semantic structure is divided by the semantic root and a natural sentence generation path is formed in the divided partial semantic structure, the partial natural sentence that expresses the meaning of the partial semantic structure is displayed. Can be generated. The whole natural sentence and partial natural sentence can verify the meaning of the input natural sentence in detail. By referring to this, there is provided a natural sentence processing method for assisting a writer to create a natural sentence expressing an accurate meaning.

Before performing semantic analysis, first perform morphological analysis and syntactic analysis. Morphological analysis is a fairly difficult technique, but it is almost established and is available for free. In the present invention, a free morphological analysis program JUMAN (Non-Patent Document 3) is used. Parsing is still said to have problems, but quality parsing programs are already available for free. In the present invention, a syntax analysis program KNP (Non-patent Document 4) used by many researchers is used. In the present invention, the phrase analysis result obtained in the middle is used instead of directly using the syntax analysis result. Morphological analysis and phrase analysis are well-known techniques and will not be described in detail.

The semantic structure of the present invention has already been approved as a patent in Japan (Patent Document 1), the United States (Patent Document 2), and Europe (Patent Document 3). The semantic structure is constructed based on the semantic frame shown in Japanese Patent No. 3811902. Based on this semantic structure, we have proposed a semantic word as a universal language of universal language. A book that explains the role of universal grammar in an easy-to-understand manner regarding Japanese and English semantic words has already been published as "Universal Japanese and English Grammar from the Viewpoint of Intelligence (Non-patent Document 1)".

The semantic structure for describing the natural language that is common throughout the world has not yet been questioned in the field of natural language processing, so there is no reference. However, a document widely cited by natural language processing researchers is “Natural Language Processing” (Non-Patent Document 2) edited by Shin Nagao. Those who provide morphological analysis programs and syntax analysis programs for free have written, so it will be helpful for morphological analysis and syntax analysis.

Japanese Patent No. 3811902 United States Patent 6126306 European Patent No. 0532338 14.08.02

Sadafumi Ando "Universal Japanese and English Grammar from the Viewpoint of Intelligence" Screenplay, Inc., published on March 28, 1995 Masao Nagao "Natural Language Processing" Iwanami Shoten September 25, 2003 JUMAN Japanese Morphological Analysis System Media Laboratory, Graduate School of Informatics, Kyoto University Japanese Language Parsing System KNP Media Laboratory, Graduate School of Informatics, Kyoto University

Sentence writers create sentences on the assumption that the reader shares the same knowledge as the writer. However, readers do not always share the same knowledge as writers. At that time, it may be misread. For example, in the sentence “Keep your heart on the flowing clouds and twittering wild birds”, an ordinary person has the knowledge that “clouds flow” and “wild birds sing”. "The wild birds tweet. Keep in mind the clouds and the tweet." However, in the next sentence that “Clouds”, “Wild Birds”, and “Twitter” do not understand, the following interpretation 2 holds. Interpretation 2: “A (clouds) and B (wild birds) flow. The A (clouds) and B (wild birds) do C (tweet). The D (heart) remains in C (tweet).” Since it is difficult to think of “birds flowing” and “clouds tweet”, it can be seen that interpretation 2 does not hold. Interpretation 1 is valid because people generally have the knowledge described above.

However, if the following sentence “Let company C of company A or company B who undertakes undertake D debt”, two interpretations hold. In other words, Interpretation 1: “A company or C company undertakes D debt” Interpretation 2: “C company undertakes D debt”. The only question is whether the phrase “Company B, Company C” is combined as meaning first, or “Company A or Company B” is combined as meaning first. However, it is a serious problem for Company A whether it is “Company A and Company C” or “Company C only” that bears the debt. If there is knowledge that “Company A was the only company that supported it, and Company C was an affiliated company of Company B”, it would be interpretation 1, but “Company A and Company B supported Interpretation 2 is established if there is knowledge that “Company C is an affiliate of Company A and Company B”.

    In some cases, a long sentence may include a quotation that is not directly related to the text. Furthermore, in Japanese, the subject may be related to the last verb, such as “I think ………”. The interpretation changes depending on whether the context can be understood and whether there is knowledge about the content of the sentence.

    “New Zealand's Key Prime Minister said at a regular press conference on the 15th that a New Zealander who was a member of Sea Shepherd, an anti-whaling group in the Antarctic Ocean, invaded a Japanese research whaling ship. "It doesn't help. Their activities on the high seas are very dangerous and they seek to be calm." "

    From the context, it is readily apparent that Prime Minister Key is involved in “criticism” far behind. In other words, "Key Prime Minister ... criticized." However, if the context cannot be read, it is considered that the subject is related to the invasion of the latest verb, so “Key Prime Minister ... has entered the research whaling vessel”.

In addition, in the sentence, “Change of job and re-employee are disadvantaged at that time”, many people add letters appropriately with their own knowledge, “At that time, the change of job and re-employed are disadvantageous. It can be interpreted to mean "changed job at that time and re-employed become disadvantaged". In this way, there are many sentences in which the semantic interpretation differs depending on whether there is an ambiguous expression that the writer does not notice or whether the reader has knowledge about the input natural sentence. Moreover, the writer often does not notice at all. This is not allowed in official documents. It is important to create sentences that do not cause misunderstandings in sentences involving damages. The present invention provides information that accurately presents the meaning of an input sentence and can consider whether the meaning intended by the writer is accurately expressed. This helps to create sentences that accurately express the meaning intended by the writer. It is to provide a natural sentence processing method for that purpose.

When the reader is presented and has no knowledge of the sentence, he or she tries to decipher the words by connecting them together. In addition, compound words are interpreted as one word, but phrases tend to be preferentially interpreted after compound words. Such a tendency is incorporated into the semantic analysis rules to construct the meaning. Company A, Company B and Company C can be added as organization names, debts as monetary nouns, and to some extent grammatical knowledge, but they do not become full-fledged semantic information.

Semantic analysis begins with grammatical information. Basically, a semantic structure is constructed on the basis of semantic construction rules based on the semantic interpretation of ordinary people. That is, as in Interpretation 1, the semantic structure is constructed by prioritizing phrases such as “Bird's Twitter” and “Company B of Company B” over “Clouds and Wild Birds”. In addition, since the noun to be the subject is generally related to the first predicate (verb), it is analyzed to mean “the prime minister… invaded the investigation whaler”. In this case, no words or sentences are added at all. In this case, it is analyzed that it means that the job is changed and the re-employed is disadvantaged. The writer asks, “It may be interpreted in this way, but is that okay?” If you want to prevent such interpretation, the writer will look for a different way of expressing. The present invention only draws attention to the writer. In the present invention, the input natural sentence is once converted into a semantic structure, the semantic structure is decomposed into various parts, the natural sentence is expressed from the decomposed semantic structure, and the meaning of the details of the input sentence is expressed in the natural sentence. Because it is displayed with, it is possible to find a place where the reader is easily misunderstood. However, there are cases where it is unavoidable to insert a quote that is not related to the meaning of the text in the input sentence. For that purpose, a function to specify that semantic analysis is performed independently is provided. In other words, ((...)), <<<< >>>, double brackets such as "" ... …… "", etc. ―――. Specify the semantic range for semantic analysis by combining the punctuation and punctuation marks as in,. Also,, and to specify the boundary of the semantic analysis range in the sentence. . Like this, it also has a function to block the semantic relations before and after by putting double punctuation marks and punctuation marks. “Key Prime Minister criticized << …… and >> criticized at the regular press conference on the 15th”, and if we analyzed semantics independently within the scope of << …… >>, The key prime minister is related to “criticism”. In addition, the parentheses such as <<<< >>>> are deleted from the input sentence and excluded from the target of semantic analysis.

“In the Tokyo stock market on the 12th, Japan Airlines stocks that are attracting attention on the future of management reconstruction were flooded with sell orders, and the market price fell to 37 yen, a 30 yen reduction compared to the weekend, without a successful transaction. What are the main points in the sentence? What is the basic meaning that you want to present as knowledge, what background information is the background, which expressions are easily misunderstood, and which part should be considered to simplify the sentence? Or, it is difficult to understand what must be accurately presented as information. The present invention finely divides the semantic structure of an input natural sentence, generates a partial semantic natural sentence from the finely divided semantic structure, and displays the meaning of the partial semantic natural sentence in detail. A semantic structure symbol indicating the semantic structure is added to the sentence generated from the semantic structure to clarify the semantic structure of the natural sentence. In addition, since the root meaning from which branches and leaves have been removed and the branch and leaf meanings behind the root meaning are shown separately, the meaning structure including the meaning and partial meaning of the whole sentence can be understood. When performing machine translation, it is assumed that the exact sentence intended by the writer has been created. Even if a bad sentence is translated, the bad sentence is a bad sentence. First of all, it is important to create correct sentences in Japanese before translation. In addition, it is important to create a sentence that expresses the correct meaning in a public sentence that causes liability. It can also be used for intellectual training while creating sentences.

It is a figure which shows the structure of the whole apparatus for implement | achieving this invention. It is a block diagram which shows the whole flow of a program. FIG. 3 is a diagram illustrating a configuration of a joint of the MW element 30. It is a figure which shows the joint which PS element 31 has. It is a figure which shows the data structure of MW element 30, PS element 31, and semantic root collection element 32. It is a figure which shows the semantic frame 40 of three basic simple sentences, ie, PS_EX, PS_IS, and PS-DO. It is a figure which shows the meaning structure of "Taro floats a flower in water." It is a figure which shows the semantic structure of "the flower which Taro floats in water." It is a figure which shows the morphological analysis result of the sentence "I want to cut off speed and noise, and to stop for a while and to keep in mind to the flowing cloud and the chirping of wild birds" in the autumn day. It is a figure which shows a phrase analysis result among the results obtained by performing a syntax analysis on the morphological analysis result of FIG. A semantic sentence generation path 56 for generating a natural sentence from this semantic structure is displayed by hierarchically displaying the semantic structure of the sentence “Autumn day, cut off speed and noise, ...” in a hierarchical manner. FIG. It is a figure showing the meaning of the details expressed by the sentence "Turn off speed and noise, one day of autumn ...". “In the Tokyo stock market on the 12th, Japan Airlines, which has been attracting attention for the future of management reconstruction, rushed to sell orders, and the deal was not completed, but the price range of Nikko shares on the same day would be limited by 30 yen compared to the weekend 37 It is a figure which shows the natural sentence production | generation path | route 56 for producing | generating a natural sentence from this semantic structure by displaying the semantic structure of "the quotation price has fallen to the circle." It is a figure which shows the simplified meaning of the detail which cut off the natural sentence production | generation path | route with the meaning structure of the input sentence "In the Tokyo stock market on the 12th ...", deleted the branch leaf meaning, and simplified. It is a figure which shows the detailed meaning displayed without leaving the natural sentence production | generation path | route with the meaning structure of the input sentence "In the Tokyo stock market on the 12th ...", leaving the branch leaf meaning. It is a program list figure which shows the flow of the basic program of a natural sentence production | generation. It is a figure which shows the relationship between the meaning structure of the sentence "I want to keep in mind the flowing clouds and the chirping of wild birds," and simple sentences and horizontal combined words.

The overall configuration of the apparatus is shown in FIG. The central control device 1, the internal memory 2, the display device 3, the external storage device 4, and the input device 5 are connected to each other via a bus 8 with a memory hub 6 and an IO hub 7 as relay devices. The external storage device 4 includes rules such as a Japanese dictionary, morpheme analysis rules, syntax analysis rules, morpheme semantic construction rules, clause semantic construction rules, a dictionary search program, a morpheme analysis program, a syntax analysis program, and a morpheme semantic construction program. In addition, programs such as phrase meaning construction programs and natural sentence generation programs are stored. The central controller 1 retrieves the program from the external storage device 4 to the internal memory 2 and performs various processes according to the instructions of the program.

The overall process flow is shown in FIG. Input natural sentences do not necessarily need to be divided, but in order to provide functions such as independently analyzing the meaning of quoted text inserted in the sentence and preferentially analyzing the specified meaning range. The input sentence is divided into several parts. For this purpose, natural sentence division 20 is performed. The morpheme analysis 21 is performed on the divided input natural sentence, and the morpheme analysis 21 searches all character strings from the beginning of the character string of the input natural sentence. For the retrieved character string, morpheme analysis 21 is executed according to the morpheme analysis rules that define the relationship between the preceding and following character strings, and only relevant morphemes are selected to obtain a morpheme analysis result as shown in FIG. A parse analysis 22 is executed on the morphological analysis result obtained as a result of the parse analysis rule to obtain a parse analysis result. In the present invention, only the phrase analysis result as shown in FIG. 10 which is a part of the syntax analysis result is used. A morpheme semantic construction rule is applied to the previous morpheme analysis result to perform a morpheme semantic construction 24 to obtain a partial semantic structure of the input natural sentence. If the semantic structure obtained by the morpheme semantic construction 24 can be combined with the already constructed semantic structure, the semantic root 41 connects them. Next, the phrase meaning construction rules are applied to the phrase analysis result to perform the phrase meaning construction 23 to obtain the semantic structure. If the semantic structure obtained by the morpheme semantic construction 24 or the semantic structure constructed by the phrase semantic construction 23 can be combined, these semantic structures are combined by the semantic root 41. Such semantic construction is repeated to obtain an overall semantic structure 58 as shown in FIG. 11 expressed by the input natural sentence. Further, when constructing a semantic structure, the semantic root 41 that is the base point of the coupling of the semantic structure is collected and stored in Bun_ixrt of the semantic root collection element 32. The semantic structure is arbitrarily divided by these semantic roots 41. If a natural sentence generation path 56 is created in the divided semantic structure, a natural sentence expressing the meaning of the partial semantic structure 57 can be obtained. Details of the semantic analysis program will be described later, focusing on natural sentence generation.

Describe the semantic structure expressed in natural language with the following two elements. That is, PS (Primitive S entence) elements 31 of MW (Meta Word) elements 30 and 4 of FIG. The data structure is shown in FIG. MW mainly stores semantic information about words. This is indicated by (). MW basically has four joints: U (upper Upper), L (lower Lower), B (previous Before), and N (next Next). The coupling information of these MW elements 30 is stored in the MW coupling element 50 as follows. That is, the element numbers of these binding partners are stored in Prc_mw [mw_num] .N, Prc_mw [mw_num] .B, Prc_mw [mw_num] .U, and Prc_mw [mw_num] .L. Basically, there is only one element connected to the top, but in complex sentences, it may be combined with multiple elements. For this purpose, Prc_mw [mw_num] .amw is provided as an auxiliary. If more elements need to be combined above, Prc_mw [mw_num] .bmw, Prc_mw [mw_num] .cmw, etc. can be provided. mw_num is the number of the MW element 30. The type of the coupling partner such as MW or PS is also stored in Prc_mw [] .FLG as the MW coupling element 50. Presence / absence of expression prohibition is stored in Prc_mw [] .KAK as an MW expression permission / inhibition element 52, and blocking of a natural sentence generation path is stored as an MW path blocking element 53. FLG and KAK are expressed as 8 hexadecimal digits.

The character string of the word is stored in the next MW character string element 51. That is, the word headword character string is mainly in Prc_mw [] .moji, articles and prefixes are mainly in Prc_mw [] .moji_jar, and particles are mainly in Prc_mw [] .moji_jcs. The symbols are mainly stored in Prc_mw [] .moji_ronri. As shown in Fig. 4, the PS elements are basically the main case (Agent case), time case (Time case), space case (Space case), object case (Object case), frequency case (V case), reason case. It consists of seven cases such as (Why case) and Predicate case. This is indicated by []. PS element 31 consists of a total of 8 bonds, A, T, S, V, W, O, P, and L (lower) for combining with lower MW. have. The PS coupling information of the PS element is stored in the PS coupling element 55. It is stored as the data structure shown in FIG. For example, the combination information of the A case is stored in Prc_mw [] .A. When the PS element 31 is combined with the MW element as shown in FIG. The semantic frame 40 shown in FIG. 6 is a minimum unit of meaning that cannot be further decomposed, and is called a basic semantic unit. Since this basic semantic unit expresses one sentence, it is also called a basic simple sentence. In the present invention, a semantic structure of an input natural sentence is constructed by combining such semantic frames with the semantic structure of the input natural sentence. The Japanese character string is stored on it, but the semantic structure is divided into parts, a natural sentence generation path is created on the partial semantic structure, and the corresponding meaning is displayed in Japanese. . When using semantic words to answer natural language questions and perform multilingual machine translation, details in the semantic structure must be accurately described.In the present invention, Japanese is added to the semantic structure. Therefore, there is no need to specify the internal structure of the semantic structure in detail.

As the meaning frame of the verb, the meaning frame shown in FIG. 7 is stored in the dictionary, and is extracted and used when necessary. When taken out, the verb string is written in the P2 case. For character strings other than verbs, it is not necessary to specify the type of case. Rather, in order to make the order of the Japanese character string generated from the semantic structure the same as the character string of the input sentence, the semantic frame is written in the slot of the semantic frame in the word order of the input natural sentence. PS_EX shown in FIG. 6A is used for a semantic frame for storing a noun compound word. Write the first word in the S case and the next word in the semantic root. However, there is no need to specify the case to store other than that. The number of words to be written in the semantic frame is estimated to be about 10 at maximum, and the semantic frame as shown in FIG. 7 is used as the general-purpose semantic frame. However, it is not necessary to specify the case shown in FIG. In this case, it is only necessary to secure 10 slots.

The input natural sentence is once converted into a semantic structure. Semantic structures can be quite complex, but we start with a simple semantic structure. Combining simple semantic structures with semantic roots to build up more complex semantic structures. In the semantic construction process, the semantic root 41 that becomes the bond base point is classified into several types and collected. The element for storing these semantic roots is the semantic root collection element 32, which is stored in Bun_ixrt as follows. The data structure is shown in FIG. The semantic root 41 that is the base point of the single sentence constituting the main point is stored in Bun_ixrt [Bun_num] .syu_shi []. Bun_num is the number of the input natural sentence. Among input natural sentences, the semantic roots 41 of simple sentences expressed in simple sentences are collected in Bun_ixrt []. Tan_bun []. Since the word to be written to the slot is grouped by clause, it is written to Bun_ixrt [] .setu [Setu_num] .slot []. Setu_num is a section number. In addition, the vertically combined compound word that is vertically connected is Bun_ixrt []. Setu [Setu_num] .goh_tate_kon [], and the horizontally combined word is horizontally connected to Bun_ixrt []. Setu [Setu_num] .goh_yoko_kon []. The phrase is written in Bun_ixrt [] .setu [Setu_num] .phrase_kon [], and the word taken out from the sentence, that is, the root word is written in Bun_ixrt [] .bun_tate_kon [].

The meaning is universal. Semantic words were devised to realize this on a computer. In order for a semantic word to be used as a global common language, languages from around the world must be described on the computer with exactly the same semantic structure and the same meaning. The present invention is an application of a very elementary technique of semantic words. Briefly explain very basic ideas.

There are three types of basic sentences that cannot be disassembled any more: PS_EX “EXist ~ is; existence”, PS_IS “IS is; state” that represents the state, and PS_DO “DO to do; act” that represents the action. is there. This is shown in an easy-to-understand manner as shown in FIG. In other words, in the P case of PS_EX, there is “Yes” meaning existence, in the P case of PS_IS “Yes” meaning state, and in the P case of PS_DO, “Yes” meaning action is written. The MW element 30 has an implicit value: “Anyone” or “What” (WHO WHAT) for the A rating, “Someday” (WHEN) for the T rating, “Where” for the S rating, In the O case, “what” (WHAT), in the V case “some times”, and in the W case “why” (WHY) are written. The basic semantic unit is basically composed of 1H5W. In addition to these three basic simple sentences, five quasi-basic simple sentences expressing meanings such as “movement”, “change”, “progress”, “occurrence”, “disappearance” are set. Semantic structures expressed by phrases such as verbs, auxiliary verbs, adjectives, suffixes, and judgments are composed of these 8 basic simple sentences and semi-basic simple sentences combined vertically and horizontally. This can represent any complex semantic structure.

“Verbs, auxiliary verbs, adjectives, judgments, suffixes” and the like are predicates. A predicate in which a plurality of predicates are combined is called a composite predicate. Here, a sentence expressed by a single predicate or a compound predicate is a simple sentence. As an example, a semantic structure of “Taro floats a flower in the water” is shown in FIG. That is, the PS_EX (existence) of the basic semantic frame 40 is vertically combined with the O2 rating of the PS_DO (act) of the basic semantic frame 40, and “is” in the P1 rating of the PS_EX, but “floating” in the P2 rating of the PS_DO. "Is written. The remaining cases are called slots, and word strings are written in them. In this case, a character string and semantic information such as “Taro is”, “Flower” is assigned to the A1 slot, and “Water is” is written to the S1 rank. The meaning of “Taro floats a flower in the water” is considered to mean “Taro does” in the state of “flower is”, “in the water”, “is”. In such a state, “do” is expressed by a verb (predicate) “floating”. The small characters in the figure are thought to exist as implicit values in the human brain, but are not represented as character strings in natural sentences. In order not to express this even on computers, the expression is prohibited. A mark (*) was added to indicate this in natural text.

The MW element at the lowest level is a MW element with a special role. Since it is at the root of the semantic structure, it is called a semantic root 41. The word written in the semantic root is called the root word. All semantic structures connect this as a connecting point. In order to generate a natural sentence from the semantic frame 40 by a program, a natural sentence generation path is generated using a semantic root as a base point. If you collect only expressible character strings along that line, you can easily generate natural sentences. A natural sentence can be easily generated as follows. Simply embed the upper PS_EX in the O2 case of the lower PS_DO. Only the case element of FIG. 7 is shown as follows.
([A2 T2 S2 V2 W2 ([A1 T1 S1 V1 W1 O1 P1]) O2 P2])
Among these, when only the MW elements 30 having character strings such as “Taro ga”, “Hana wa”, “In the water”, “Float” are arranged, the result is as follows. The symbol A2 indicating the A2 case is shown as a suffix.
([(Taro) A2 (flower) A1 (in water) S1 (floating) P2]))
In the present invention, the symbol () of the MW element 30 and the symbol [] of the PS element 31 are used as described above in order to express the semantic structure in a natural sentence. When the symbol of the MW element 30 inside the PS element 31 is deleted, the result is as follows.
([Taro floats a flower in the water])
Further, when the symbol [] of the PS element is omitted, the following is obtained.
(Taro floats a flower in the water)
The semantic structure symbol 42 inside is omitted, but this is the semantic structure of the sentence “Taro floats a flower in the water”. If this MW element is omitted,
Taro floats a flower in the water
It becomes. In order to suggest a semantic structure in natural sentences, symbols that do not feel uncomfortable with character strings are used. Here, the meaning frame 40 of “floating” has been introduced. There are various types of meaning frames 40 in the meaning frame 40. In the present invention, since the internal structure of the semantic frame 40 is not a problem, the semantic frame 40 is used as a general-purpose semantic frame 40 in the description of the present invention.

Next, the semantic structure of the sentence “Taro floats in the water” will be described. this is
"Taro floats a flower in the water. That flower"
It means that. This is shown as a semantic frame 40 in FIG. As shown in Fig. 7, it has the same semantic structure as “Taro floats a flower on the water”, but “Hanaga” written in the A1 rating of the semantic frame 40 is prohibited, and it means that it is at the bottom. The state is written in the root 41. If a slot in which a word string is not written is omitted, the semantic structure shown in Table 1 is obtained.

Table 1
[(Taro is) (* flower) (in the water) (floating)]
↑ ↓
(Flower) If the semantic structure above the semantic root 41 is embedded before the semantic root 41,
([(Taro is) (* flower) (in the water) (floating)] ↑ ↓ flower)
It looks like this. If it is embedded after the semantic root 41, it is in English word order. ↑ ↓ are marks indicating that the PS element [[Taro] (* flower) (in the water) (floating)] and the MW element (flower) are vertically connected. This is a vertically connected symbol of the semantic structure symbol 42. I am trying to make a semantic structure in a natural sentence emerge by writing a symbol indicating the semantic structure in the natural sentence. We decided to use “::” instead of “↑ ↓” in order to use symbols that are as familiar as possible. Therefore ([(Taro is) (* flower) (in the water) (floating)] :: flower)
Write like this. If it is not necessary to delete the prohibited character string and emphasize the MW element 30 as a semantic structure, delete the MW symbol in the outer () and change it as follows ([Taro floats on the water]: :flower)
Is described. If it is not necessary to emphasize [] of the semantic frame 40, deleting it will result in the following.
(Taro floats in the water :: flowers)
This is “Taro floats a flower in the water” and “flower” is taken out and written in the element MW of the semantic root 41. The symbol :: indicates a semantic structure that is vertically connected to the sentence.

The “reason” of “Taro ’s floating flower in the water” and “The reason why Taro floated the flower in water” are not taken from the preamble. It has a slightly more complicated semantic structure. However, since the difference in level is content that can be understood naturally by the reader, the slight difference is ignored and the difference is not shown.

The semantic structure of the compound word will be described. A word in which the words “management” and “reconstruction” are combined like “management reconstruction” is called a compound word. In the present invention, such a compound word is not regarded as one word, but is handled as the following sentence, that is, “Management has a reconstruction. Its reconstruction”. To express this as a semantic structure, PS_EX representing existence is used. Table 2 shows this semantic structure as a character string.

Table 2
([(Slot) A1 (write prohibited) T1 (slot) S1 (write prohibited) V1 (write prohibited) W1 (write prohibited) O1 (present)) P1] :: ())
Write “Management” in slot S1 and “Rebuild” in slot A1. Since this is extracted and written in the semantic root 41, the semantic structure is as shown in Table 3. Prohibit expression of slot A1.

Table 3
([(* Reconstruction) A1 (In Management) S1 (* Yes)]: :( Reconstruction))
Here, when the particle “ni” and the verb “a” are prohibited from being expressed and unnecessary symbols and character strings are deleted, the following is obtained. (Management :: Reconstruction) In other words, it is a semantic structure in which the sentence “Management has reconstruction” and the word “Reconstruction” are vertically connected. In other words, “reconstruction” means not “reconstruction” but “reconstruction related to management”.

By the way, “total 300 km” is a compound word, but its meaning is “300 km is total. That 300 km”. Therefore, it is described as a semantic structure as shown in Table 4.

Table 4
([(* 300km) (total) (*)]] :: (300km))
When the prohibited character string is deleted, it becomes as follows. (Total: 300 km). In this case, the basic sentence of PS_IS should be used because it means "~ is ~", but PS_EX is used. Ignore this difference.

Next, the semantic structure of the phrase will be described. The phrase “Birds of the Birds” is thought to be a vertical combination of the semantic structure of “There is a Bird in Wild Birds” and the semantic structure of “That Twitter”. This means that “Twitter is not just twitter, but“ birds ”. It is expressed as a semantic structure as shown in Table 5. Changed the particle "Ni" to "No" for "Wild Bird" and deleted the prohibited text.

Table 5
(Bird's :: Twitter)

Just as “economic reconstruction” and “economic reconstruction” have almost the same meaning, compound words and phrases have almost the same semantic structure. However, there is a slight difference in the binding power between words. The combination of words is stronger. Therefore, the meaning is constructed by combining words from left to right first. For example, in the case of “Tokyo Stock Market on the 12th”, “Tokyo” and “Stock” are first combined to become (Tokyo :: Stock), then “Stock” and “In the Market” are combined, (Stock :: In the market), then “12th” and “In Tokyo Stock Market” are combined to become (12 :: In Tokyo Stock Market). Table 6 shows the internal structure of the semantic structure reflecting the above-described semantic construction result.

Table 6
(12th :: ((Tokyo :: Stock) :: In the market))
In other words, the compound word has a stronger binding force, so it is combined first. After that, the semantic structure of the phrase is built.

The difference between a sentence and a compound word or phrase is whether or not semantic information such as a character string can be freely written in a slot of a semantic structure. The semantic frame 40 describing compound words and phrases has no slots in which the writer can freely write semantic information, and therefore can express only a limited meaning. On the other hand, the use of verbs, adjectives, judgments and other predicates with slots in which semantic information can be freely written, allows complex meanings to be expressed freely and accurately. In general, semantic information as a background is often expressed using compound words and phrases. Moreover, it is assumed that the reader has knowledge as a matter of course, but the meaning is not necessarily accurate. It is thought that it is at least the level that we want to show as background knowledge information. The basic idea is to express the information that you want to present clearly using a predicate, and the rest of the background knowledge as a compound word or phrase. The present invention is intended to provide a natural sentence creation support method capable of such a purpose.

In the semantic language which is the basis of the present invention, the meaning of a sentence is basically composed of a semantic structure of a single sentence vertically and horizontally. As shown in Table 7, the meaning structure of the sentence "I refuse noise for a day, stop for a while, and keep my heart in mind" is as shown in Table 7, and the sentence "I cut the noise for a day" The sentence 102 and the sentence “I want to keep in mind” are combined horizontally.

Table 7
(Suppress the noise for a day) → (stop for a while) → (keep on twitter)
↑
() Meaning Is the relationship of if_then, but it means that if you turn off the noise and stop for a while, the twitter will stay in your heart. It is a semantic structure in which three sentences are connected by a logical relationship. However, for now, we will treat it as a horizontal connection. This is indicated by “→”. “Speed”, “Noise”, “Cloud”, “Twitter” are words, but the words are combined horizontally, so they are described as shown in Table 8 below.

Table 8
(With speed) → and (* noise) (wild bird) → or (* Twitter)
↑ ↑
(Noise) Semantic Root (twitter) Semantic Root Semantic Language treats sentences and words in the same way. Since words are fixed as meanings, they do not have a meaning frame, but sentences can express a variety of meanings by writing various words in the predicate slots. “Speed” and “Noise” in “Speed and Noise” have a logical product relationship, and “Wild Bird” and “Twitter” in “Wild Bird and Twitter” have a logical sum relationship. Here, only a logical relationship is implied, and it is assumed that it is simply connected horizontally as in the case of sentences. In this way, when several meanings are laterally coupled, a semantic root 41 is always set at the lowermost layer to represent the laterally coupled semantic structure. This combines with another semantic structure. The last word joined horizontally is written in the semantic root 41 as a representative, but in order to prohibit double expression, the word at the end point is prohibited.

Next, the meaning root will be described. As already explained, the word written in the semantic root is the root word, but one word is itself the root word. It has already been described that a semantic root 41 is provided as a representative of a semantic structure in a semantic structure in which several semantic frames 40 and MW elements 30 are coupled horizontally. Only this semantic root 41 is combined with other semantic structures. In this way, the point of attachment is limited, making it easier to construct semantic structures. The semantic structure of the natural sentence can be divided by the semantic root 41 by giving the semantic root 41 the function of whether or not the natural sentence generation base point and the natural sentence generation path can be blocked. Providing a root of meaning is a key point of the present invention.

The setting of the semantic range will be described. The semantic analysis will be explained with an example sentence: “I want to cut off speed and noise during the autumn day, stop for a while, and keep my mind in the clouds and wild birds.” In the present invention, the input sentence is divided into several parts, and semantic analysis is performed independently at each part. In the conventional natural sentence analysis method, syntax analysis 22 and the like are proceeding with a focus on a dependency relationship such as how a certain word is related to another word. The dependency is performed over the entire input natural sentence. Whether or not there is a dependency relationship is often based on the knowledge that a writer would naturally have. However, in the present invention, the semantic analysis is performed on the assumption that the person has little knowledge that is naturally considered to be possessed. Without knowledge, words can only be related to words that are very close. On the contrary, the purpose of this is to perform semantic analysis, show how a reader with no knowledge interprets the input sentence, and call the writer's attention. In natural sentence processing by a computer, the possibility of dependency is determined by statistical processing. The meaning of important sentences with liability cannot be determined by statistical processing. Rather, it is important to indicate to the writer what is easily misunderstood.

The semantic range division of the input sentence applies the semantic range division rule (Bsyori_irange.rule) to the phrase analysis result. The semantic range may be divided in any way, but here, it is conditional on the verb inflection form being either the basic continuous form or ending in the ta series joint form and with a reading. This is because the meaning of each single sentence is independent and seems to be combined horizontally to present knowledge to the reader. This is the rule shown in Table 9. There is only one example illustrated. The semantic range division rule is applied to the phrase analysis result shown in FIG. As shown in FIG. 9, in the result of morphological analysis, mrph_no = 9 ([verb * * basic continuous form]) and mrph_no = 12 ([verb * * data system joint form]) are bmst_n = 4, and bnst_n = 6, features such as prescription: relation: continuous reading and so on. “*” Is a symbol for ignoring the condition. Create many such rules and register them in Bsyori_irange.rule.

Table 9
Semantic range division rule Bsyori_irange.rule
(() Mae (Phrase: Action: Continuous reading) hon () ato prog_no: 8000 rule_no: 8020)
Such a rule is applied to the phrase analysis result of FIG. mae (front), hon (book), and ato (rear) are suffixes for () and are shown for reference, and are not described in the actual rules. hon = bnst_no, mae = bnst_no−1, and ato = bnst_no + 1. In other words, bnst_no is searched from the beginning, mae means before, and ato means after. That is, in FIG. 10, the semantic range division rule and the phrase analysis result are matched from bnst_no = 0, but it is checked whether or not the matching is performed before and after bnst_no. In the rules shown above, mae also does not specify the ato condition, so only hon needs to be matched. 8000 of prog_no is a program number, and the function prog_8000 () is executed. rule_no: 8020 is an identification number of the rule. Based on the phrase analysis result of FIG. 10, bnst_no is traced to check which rule matches. In the case of the above-mentioned rule, bnst_no = 4 “breaks” and bnst_no = 6 “stops”, and matches the feature “phrase: movement: continuous readings” of the phrase analysis result. Therefore, the semantic analysis range is divided into three ranges of bnst_no = 0-4, 5-6, and 7-12. This is defined as semantic range 0 (imi_range_0), semantic range 1 (imi_range_1), and semantic range 2 (imi_range_2). The semantic analysis is performed by dividing into these three semantic ranges.

Next, semantic construction using morphological analysis results will be described. When a noun is attached to a noun, such as (autumn) or (speed) or (noise), the semantic structure is very simple. In the present invention, semantic construction starts with a simple and reliable semantic structure. Semantic construction proceeds with analysis based on the phrase analysis results shown in FIG. The morphemes corresponding to the clauses in FIG. 10 are shown in mrph_from and mrph_to in FIG. For example, morphemes corresponding to “autumn” of bnst_no = 0 are mrph_from = 0 of “autumn” and mrph_to = 1 of “no”. The morpheme semantic construction rule Msyori.rule is applied to this range. Table 10 shows the morpheme semantic construction rule Msyori.rule used in the description of the present invention.

Table 10
Morphological Semantic Building Rules Msyori.rule
(() Mae ([noun (common noun temporal phase noun)]) hon ([particle (case particle connection particle)]) ato prog_no: 2000 rule_no: 1900)
(([Noun Numeral]) mae ([Suffix noun noun suffix]) hon () ato prog_no: 1400 rule_no: 1400)
(([Suffix noun noun suffix]) mae ([special (punctuation)] hon () ato prog_no: 2400 rule_no: 2405)
(([Verb * * (basic continuous form)) mae ([special (punctuation)] hon () ato prog_no: 2002 rule_no: 2414)
(([Verb * * Basic Conjunction]) mae ([Suffix Adjective Predicate Suffix]) hon () ato prog_no: 4600 rule_no: 4413)
(() Mae ([verb * * basic form]) hon () ato prog_no: 2000 rule_no: 2400)
(() Mae (adverb) hon () ato prog_no: 2001 rule_no: 2401)

Describe the search conditions in the order of [part of speech fine classification utilization type utilization form string]. If the usage type and the usage type are not specified, use “*” as follows. [Part-of-speech sub-category * * character string] No search condition is specified if no search condition is described, such as (). Items in parentheses, such as [] (common noun time phase noun), mean that either a common or time phase noun may be used. If there is no matching condition after the fine classification, the condition does not have to be described as [noun fine classification]. In the semantic range 0 (imi_range_0), matching starts from mrph_no = 0 in FIG. The part of speech of [autumn] is a noun and the subcategory is a temporal noun. The part of speech of mrph_no = 1 is a particle and the subcategory is a connected particle. Therefore, when hon is mrph_no = 0, rule number rule_no: 1900 matches. The function prog_2000 () is executed. One MW element 30 is taken out of the memory and set as mw_1, and the character string “autumn” as the headword is written in Prc_mw [1] .moji of the MW element 30 and the particle “no” is written in Prc_mw [1] .moji_jcs. Since the noun “autumn” is not used in future semantic analysis, it is deleted from the morphological analysis result as processed. FIG. 11 shows the semantic structure hierarchically so that the relationship with the input sentence can be intuitively understood. In FIG. 11, the MW element 30 is shown as mw_1 (autumn).
Since the rule of rule_no: 1900 also matches “speed and” and “noise”, MW elements 30 of mw_3 (speed and) and mw_4 (noise) are created as shown in FIG. “One day” builds meaning with two rules. Rule number rule_no: 1400 creates mw_2 (one day), and rule number rule_no: 2405 adds a reading as mw_2 (one day).

Since “Decline” matches rule_no: 2414, the function prog — 20002 () extracts the semantic frame ps — 2 from the semantic frame dictionary, and the character m in the element moji of the Pw mw — 22 (), that is, Prc_mw [mw — 22] .moji Write "severe". Then, processing such as writing a punctuation mark “,” in the element moji_jcs of mw_23 which is the semantic root 41, that is, Prc_mw [mw_23] .moji_jcs, is performed to construct a semantic structure as shown in Table 11 and FIG.
Table 11
ps_2 [mw_22 (cut off)]
↑
mw_23 (,)
This completes the construction of the semantic structure of the semantic range 0 (imi_range_0). Next, the meaning is constructed for the meaning range 1 (imi_range_1). As shown in FIG. 11, mw_24 (for a while) is created. The character string “stop,” is mrph_no = 12, and matches the rule number rule_no: 2414. Therefore, the function prog_2002 () is executed, and the semantic frame (ps_4) is fetched from the semantic frame dictionary and set in the memory. It is the same method as the above-mentioned “cut off”. This semantic structure is shown in Table 12.

Table 12
ps_4 [mw_38 (stop)]
↑
mw_39 (,)
Next, for the meaning range 2 (imi_range_2), semantic construction such as “clouds”, “wild bird”, “twitter”, “heart” is performed. The morpheme semantic construction rule Msyori.rule is applied to this range. mw_5 (clouds), mw_6 (wild birds), mw_7 (twits), mw_8 (hearts), etc. can be constructed with the morpheme semantic construction rule rule_no: 1900. This is shown in FIG.

“Flowing” with mrph_no = 14 matches the morpheme semantic construction rule rule_no: 2400, executes the function prog_2000 (), and fetches the semantic frame ps_6 into the memory. This is shown in FIG. As shown in FIG. 11, the frame 40 meaning “fasten” is taken out to the memory in the same manner as described above. “I want to keep” is a compound meaning frame 40 in which the meaning frame 40 of the verb “Stuck” and the meaning frame 40 of the suffix “Tai” are vertically connected. In the present invention, the whole semantic structure of the natural sentence is changed. The purpose is to explain in an easy-to-understand manner, and the semantic structure inside the semantic frame will not be discussed in detail. For this purpose, a virtual meaning frame 40 of “Fix_I want” is created and used. “Fix” and “Tai” execute the function prog — 4600 () because the morpheme semantic construction rule rule_no: 4413 matches with mrph_no = 25. That is, the character string “stop_tai” obtained by connecting the basic consecutive form “stop” and the suffix adjective predicate suffix “tai” with “_” is the P2 MW element 30 of the meaning frame 40 of ps_8 ( mw — 68), ie, Prc_mw [mw — 68] .moji. This is shown in Table 13.

Table 13
ps_8 [mw_68 (stop_want)]
↑
mw_69 (.)
Next, semantic construction is performed using the phrase analysis results.

In the parsing 22, some feature features are given to indicate what role the clause has. Semantic construction is performed using these features. For example, as shown in FIG. 10, regarding [autumn] where bnst_no = 0, [word] is given in feature 2 and [engagement: no case] is given in feature 4, and [every day, bnst_no = 1] ], [Description] is given in Feature 1, and [Quantity] is given in Feature 2. By combining these, the phrase meaning construction rule Bsyori.rule as shown in Table 14 is provided.

Table 14
Clause semantic construction rule Bsyori.rule
((Speaking person: no) mae (speaking quantity) hon () ato prog_no: 6100 rule_no: 6220)
((Speaking person: no case) mae (sentence particle case element combination element) hon () ato prog_no: 6100 rule_no: 6221)
((Participant particle clerk: G) mae (Body particle grammar element element) hon () ato prog_no: 6200 rule_no: 7044)
((Participant particle clerk: Conjunction) mae (Partial particle Participant case element Conjunctive element) hon () ato prog_no: 6200 rule_no: 6460)
((Phrase: Action: Conjunction Conjunction Modification Conjunction Parallel Condition) mae (Conjunction Particle) hon () ato prog_no: 6000 rule_no: 6060)

Since the rule number rule_no: 6220 of the clause meaning construction rule matches with bnst_no = 1, the function prog_6100 () is executed. That is, as shown in FIG. 11, a semantic frame ps_10 is created, the character string “autumn” is added to the mw_83 of the S case of the semantic frame 40, the semantic information of the character string “one day,” etc. of mw_2, etc. Copy to the slot of mw_81 of the A case and mw_88 of the semantic root 41. Since "A day," in the A rating is prohibited, it is omitted in FIG. This semantic root 41 is stored in Bun_ixrt [Bun_num] .setu [Setu_num] .phrase_kon [i] = mw_88 as phrase (phrase) _kon (root) as follows. Here, Bun_num is a sentence number, Setu_num is a section number, and i is the number of semantic roots 41 already registered. That is, the sentence numbers and the section numbers are classified and stored. Since “autumn” with bnst_no = 0 is no longer involved in the semantic structure at this stage, it is deleted as processed from the phrase analysis result of FIG. Since “speed and” of bnst_no = 2 and “noise” of bnst_no = 3 match rule_no: 7044 of the clause semantic construction rule, the following meaning is constructed by the function prog_6200 (). Combine mw_3 and mw_4 horizontally. Then, another mw_70 is newly provided under mw_4, and is combined with mw_4 in the vertical direction. The situation is shown in Table 15.

Table 15
mw_3 (with speed) → mw_4 (* noise)
↑
mw_70 (noise)
Since this becomes a connection point with other semantic structures, goh word) _yoko (horizontal) _kon (root), that is, Bun_ixrt [Bun_num] .setu [Setu_num] .goh_yoko_kon [i ] = Mw — 70; Here, in order to store the coupling information as described above, mw_3, mw_4, and mw_70 store the element numbers of the coupling partners as follows. Prc_mw [mw_3] .N = mw_4; Prc_mw [mw_4] .B = mw_3; Prc_mw [mw_4] .L = mw_70; Prc_mw [mw_70] .U = mw_4; where mw_4 is the element number of the MW element mw_4. In addition, information such as the type of element to be combined, that is, whether it is PS or MW, is stored in Prc_mw [mw_num] .FLG. Since “speed and” of bnst_no = 2 will not be involved in the construction of the semantic structure in the future, it is deleted as processed from the phrase analysis result of FIG.

Next, in imi_range_0, the character strings “one day” and “noise” are written in the MW element 30 of the slot of the frame 40 meaning “cut off”. An empty slot of the semantic frame 40 is found, and semantic information such as a character string is copied to the empty slot in that order. This can be achieved with a simple program. By copying the semantic information of mw_88 (one day) to the empty slot mw_17 of the semantic frame ps_2 that has already been created, and storing the element number of the coupling partner in the MW coupling element 50 of the slot mw_17 and the PS coupling element 55 of ps_10 , Store the binding information. In this case, a character string is written in the empty slot according to the word order of the input sentence so that the word order of the input natural sentence does not change during output. In this way, the semantic structure of “one day” is combined with the semantic frame 40 of “cut off”. In a semantic language, a word with a specified meaning is written in a slot with a specified meaning. If this is done strictly, the word order of the writer and the natural sentence generated from the semantic structure will be changed. Since it changes slightly and places a psychological burden on the writer, here, in order to output in the word order of the writer, it is written in the slot in the word order of the input sentence. The semantic information of the semantic root mw_70 of (speed and) → (noise) is copied to the empty slot mw_18, and the combined information is stored with the element numbers of the coupling partners in mw_18 and mw_4. Again (speed and noise) will have two semantically connected MWs, mw_70 and slot mw_18. With the slots mw_17 and mw_18, whether or not the natural sentence expression is possible and whether or not the natural sentence generation path is blocked is controlled by the MW expression permission element 52 and the MW path blocking element 53, Prc_mw [mw_num] .KAK. KAK is an 8-digit hexadecimal number that can be expressed by the first digit from the top, and can be blocked by the fourth digit from the top. Since the semantic root 41 is already collected and classified and stored in Bun_ixrt [Bun_num], it can be controlled at any time. If the route can be cut off, the sentence reads “Turn off noise for a day,” otherwise, the sentence “Turn off speed and noise for a fall day,” is obtained.

The semantic construction of imi_range_1 by the phrase semantic construction rules is simple. As shown in FIG. 11, the semantic information and the combined information are simply written in mw_24 (for a while) in the empty slot mw_33 of the semantic frame ps_4 [mw_33 (slot) mw_38 (stop)). The semantic construction of imi_range_2 is performed as follows. When bnst_no = 10 in FIG. 10 matches the clause semantic construction rule rule_no: 6221, the function prog_6100 () is executed, and the semantic structure as shown in Table 16 is obtained.

Table 16
mw_73 mw_79
(Bird's :: Twitter)
At this stage, “wild bird” will not be involved in semantic construction in the future, and is deleted from FIG. 10 as processed. “Flowing cloud” matches bnst_no = 8 and the phrase semantic construction rule rule_no: 6060 matches, and the function prog_6000 () causes mw_5 (clouds or clouds) to the semantic root mw_54 of the “flowing” semantic frame ps_6 as shown in FIG. ) Write semantic information. Since “cloud” is vertically joined with the phrase “flowing”, it is stored in Bun_ixrt [Bun_num] .setu [Setu_num] .bun_tate_kon []. Based on this, a character string “flowing cloud” can be generated. At this stage, since the semantic analysis of “flowing” is finished, the item “flowing” is deleted from FIG. 10 as being processed.

In the previous processing, “wild bird” with bnst_no = 9 has been removed from the processing target, so “cloud” with bnst_no = 8 and “twitter” with bnst_no = 10 are next to each other. rule_no: Matches 6460. Therefore, a semantic structure is constructed by the function prog — 6200 () as shown in Table 17 and FIG.

Table 17
mw_71 (clouds) → mw_79 (* twitter)
↑
mw_80 (to twitter)

The meaning information of mw_80 (to twitter) is copied to the empty slot mw_63 of the meaning frame 40 of “Still_Tai” and combined with the above semantic structure. In this case, mw_80 is not deleted only by writing the semantic information into the slot. When mw_8 (heart) is written into the slot mw_64, the semantic construction of imi_range_2 is completed. If the natural sentence generation path is interrupted in the slot mw_63, a sentence “I want to keep in mind” is obtained based on mw_69. If you don't cut it off, you will get a natural sentence that says, “I want to keep my mind on the flowing clouds and the singing birds.”

The natural sentence generation process will be described in detail with reference to the program shown in FIG. Functions are shown with parentheses () and; after the character string, as in writing support (); The program begins with main (); Perform morphological analysis (); and parsing (); read rules (); and then execute sentence support (); Sentence support (); divides the input sentence into several parts by dividing the meaning range of the input natural sentence (); Thereafter, the meaning of the natural sentence is constructed by means of constructing the meaning of the input natural sentence (); and then natural sentence generation (mw_root); is executed using the designated semantic root mw_root as a base point. In the semantic construction (); of the input natural sentence, the semantic structure is constructed by applying the morphological semantic construction rules to the morphological analysis result in the semantic construction (); based on the morphological analysis result. Then, in the semantic construction based on the phrase analysis result () ;, the semantic structure is constructed by applying the phrase semantic construction rules to the phrase analysis result.
In natural sentence generation (mw_root) ;, a natural sentence generation path of a semantic structure is traced to collect character strings for creating a natural sentence. Depending on whether the element connected above is PS or MW, path generation is possible. Different. First, MW generation route creation and character string collection (mw_root) are executed starting from mw_root. MW generation route creation and character string collection (); As shown in MW, if the combined element is an MW element, MW generation route creation and character string collection (); Create a PS generation path and collect character strings (); create a generation path and collect character strings that can be expressed by that path. In the PS generation path and character string collection () ;, it branches into seven cases: A, T, S, V, W, O, and P. The creation route creation order is determined by kaku specified in the case order table. In the figure, the Japanese case order is shown, but the case order is different for each language. The basic order is APOST for English and ATSPO for Chinese. Any order can be indicated in the order table. MW generation route generation and character string collection (); is a MW element that can be expressed and the route is blocked, and if the bypass route amw is specified, the generation route generation and character string in the bypass route By collecting () ;, these elements are traced, a natural sentence generation path is formed using the designated semantic root 41 as a base point, and expressible character strings are collected. When the MW element cannot be expressed or the route is blocked, the natural sentence generation 25 is performed via the MW element designated by N of the MW element without taking in the character string of the MW element.

The character string of the MW element number mw_num is collected as follows. That is,
Strings such as articles, prepositions, prefixes, etc.
strcat (word string [moji_num] [1], Prc_mw [mw_num] .moji_jar);
The string of headwords is
strcat (word string [moji_num] [2], Prc_mw [mw_num] .moji);
Particle string,
strcat (word string [moji_num] [3], Prc_mw [mw_num] .moji_jcs);
Logical symbols etc.
strcat (word string [moji_num] [4], Prc_mw [mw_num] .moji_ronri);
Is written in the character string [moji_num]. These character strings are output in the order specified in the word order table by natural sentence generation (); from the collected character strings. The word order shown in FIG. 16 is in the case of Japanese and is shown in the order of prefix, headword, suffix, and logical symbol, but the order can be freely changed in the word order table. Therefore, it can be generated in the word order of each language. MW generation route creation and character string collection (); and PS elements in the case of PS elements, PS generation route creation and character string collection (); Collect with (), and finally collect the character string of the route indicated by Prc_mw [] .N.
The output order of these character strings can be easily changed depending on the position of the program within the creation route of MW and character string collection (); In English, relative pronouns are expressed before clauses. To achieve this, "strcat (word string [moji_num] [2], Prc_mw [mw_num] .moji)" is in the program MW generation path creation and string collection () It should be moved above "if (when the element to be connected to is PS)". Thus, according to the present invention, case order, word order, simple sentence order, relative pronoun and clause order can be expressed not only in Japanese but also in word order of languages of various countries.

Explain the semantic structure and natural sentences. In the present invention, the input natural sentence is once converted into a semantic structure. Various parts of the semantic structure are specified, and partial semantic natural sentences are expressed from the semantic structure of each part. The partial meaning natural sentence can verify in detail whether the meaning intended by the writer is accurately expressed. The semantic structure is shown in (a) of FIG. 11, and the input sentence 100 is shown in (b). In FIG. 11, since the character strings of the semantic roots 41 that combine the semantic structures are arranged in a hierarchical manner and shown hierarchically, it can be seen at a glance what structure the meaning of the sentence is. This way of displaying the semantic structure can be done by a computer, but it becomes difficult to understand if it becomes a complicated long sentence. Therefore, normally, a natural sentence is output as shown in FIG. Here, how the partial semantic natural sentence is generated from the partial semantic structure will be described with reference to the semantic structure shown in FIG.

Natural sentences generated from this semantic structure are classified into two types. One is to divide the semantic structure into parts such as trunks, branches, and leaves, and display the meaning of the partial semantic structure in natural sentences. These can be generated easily if the natural sentence generation path is cut off at the semantic root. The phrase “simplified” is attached to the sentence, word or phrase generated in this way. The other is a natural sentence that is generated without blocking the natural sentence generation path at the semantic root, and is distinguished by adding a “detail” pillow. In FIG. 12, a simplified semantic expression is shown in the first half and a detailed semantic expression is shown in the second half. In the simple expression, the natural sentence generation path is blocked and only the minimum necessary character string is displayed, so it is convenient to see the basic semantic expression. In most cases, a simplified expression is sufficient, but if you want to examine it more closely, you will see a detailed semantic expression.

Although the gist of the input natural sentence is shown in FIG. 12, it is simple and easy to understand. This is a collection of expressible character strings by blocking the natural sentence generation path with mw_17 (one day), mw_18 (noise) and mw_63 (Twitter). Natural sentence generation in this case is simple. In the semantic frame 40 above the mw_23, mw_39, and mw_69 of the semantic root 41, a natural sentence can be obtained by arranging character strings that can be expressed from left to right. The main sign is appended with the horizontal link symbol →, which indicates that the main sign is a semantic structure in which three simple sentences are combined horizontally. The simple simple sentence in FIG. 12 includes a sentence having a horizontal join symbol “→” at the end of the sentence, a sentence having a sentence end symbol “.”, And a sentence that is vertically connected to other sentences. A sentence that is vertically combined has a vertical combination symbol “::” in front of the root word that is a connected word of meaning. In the simple simple sentence, only the characters written in the slot of the simple sentence are shown, so that the state is as if the branches and leaves are removed from the tree, and the semantic structure is easy to understand. The simple written word indicates the word written in the slot of the semantic frame 40 with the branches and leaves removed. The word written in the slot of the simplified simple sentence semantic frame 40 above is displayed for reference. In addition, simple horizontal combined words and simple phrases are displayed. Horizontally combined words have a semantic structure in which words are combined horizontally with →. Simplified horizontal combined word 1 is “speed and noise →”, and simplified horizontal combined word 2 is “cloud and → twitter”. Simplified phrase 1 is “Autumn :: One day”, and simplified phrase 2 is “Wild bird's :: twitter”. The phrase has the particle "no", but has the same semantic structure as the compound word. Vertically connected words are indicated by detailed vertically connected words that do not block the natural sentence formation path. In this sentence 100, there is no vertical combination word, but in FIG. 14 described later, a detailed vertical combination word is shown.

The character string “detailed semantic expression” in the figure is a character string generated when the natural sentence generation path is not blocked. The detailed simple sentence 1 is a generated natural sentence based on the semantic root mw_23, the detailed simple sentence 2 is generated based on the semantic root mw_39, the detailed simple sentence 3 is generated based on mw_71, and the detailed simple sentence 4 is generated based on mw_69. In FIG. 11, the natural sentence generation path generated by the program is indicated by a dotted line. However, by embedding the upper semantic structure in the lower semantic structure and arranging them in a line, such a natural sentence portion can be obtained.

Since the principle of the semantic language is simple and clear, a hierarchical semantic structure can be easily constructed from a simple sentence as follows. One word is treated as the root word. In a compound word or phrase composed of two words, the subsequent word is the root word. If the words are combined horizontally, the last word is the root word. In the case of a simple sentence, it consists of a semantic frame and a root word. The word after the vertical joint symbol :: is the root word. When these root words are specified in the sentence and combined in the order in accordance with the combination symbol, a hierarchical semantic structure as shown in FIG. 17 is automatically obtained. The detailed simple sentence 4 will be described as an example. In the detailed simple sentence 4 in FIGS. 11 and 12, the MW element symbol () is omitted, but in the detailed simple sentence 4 in FIG. 17, in order to clearly indicate which word is combined with which word, The meaning structure symbols such as “()”, “::” and “→” are added to the table. Take out (flowing :: cloud) from the detailed single sentence 4 and write it down. (Wild Bird's :: Twitter) is also written down. When “clouds” of (flowing :: clouds) and “twitters” of (wild birds :: twiddles) are arranged in the same row, it becomes the simplified horizontal combined word 2 (clouds and → twiddles). As a result, it can be seen that “clouds” and “twitter” are constructed in a semantic structure. If you place a meaning frame (same_want) below it, and write the words “twitter” and “heart” of the root words (in the cloud and → twitter) in that meaning frame, the simplified sentence 4 is obtained. As a result, “clouds” in “I want to keep in mind the flowing clouds and twittering of wild birds” are “clouds” in “flowing clouds”, which is next to “twittering in the wild birds” It becomes clear that it is connected (to clouds and → twitter), and that “twitter” is “twitter” (we want to keep in mind twitter). That is, it can be seen that “cloud” is related to “simplified sentence 3” and is combined with simplified sentence 4 through simplified horizontal composition 2. From this semantic structure, it can be seen that “clouds” have nothing to do with “wild birds”. In this way, when an input sentence is expressed on a semantic structure, it becomes clear that a word is associated with which word. Explain the relationship between semantic structures and natural sentences with a slightly more complicated input sentence.

The input sentence 200 says, “In the Tokyo stock market on the 12th, Japan Airlines stocks that are attracting attention on the future of management reconstruction are inundated with sell orders, and the price limit of Nikko shares on the same day will be the lower limit without a successful transaction. The quotes have dropped to 37 yen, which is 30 yen lower than the previous weekend. " In FIG. 13, the simple simple sentence 2, the simple simple sentence 3, and the simple simple sentence 5 are arranged side by side in the lowermost layer, and are combined with → as follows. That is,
(The stock is inundated with orders.) → (Transaction is not successful: Remain) → (The price has dropped to 37 yen.)
However, these simple simple sentences are simple sentences generated by blocking the natural sentence generation path with mw_83, mw_84, mw_24, and mw_25. This is the main point of this input sentence.

In the present invention, the gist can be easily obtained as described above. First, from this sentence, it can be determined whether the input sentence accurately reflects the intention of the writer. In the section of simple simple sentences, not only sentences that are horizontally coupled as described above, but also simple simple sentences that are vertically coupled, such as simple single sentence 1 and simple single sentence 4, are shown. Simplified simple sentence 1 and simplified simple sentence 2 are vertically connected by the root word “to stock”, and therefore, as shown in FIG. 13, the semantic structure is as shown in Table 18.
Table 18
(In the market, whereabouts are attracting attention :: * to stocks) ps_18
Ps_11
Similarly, the simple simple sentence 4 and the simple simple sentence 5 are vertically coupled with “up to 37 yen”, so the semantic structure is as shown in Table 19.
Table 19
(Lower limit: Up to * 37 yen) ps_22
(The price has dropped to 37 yen.) Ps_4
Since the root words "stock" and "up to 37 yen" are duplicated when generating a natural sentence, "*" is added to the character string of the upper semantic root and expression is prohibited. Become a sentence.
Table 20
“In the market, attention is focused on whereabouts :: stocks are inundated with orders,”
“Lower limit: The value has dropped to 37 yen.”
It can be seen that the root sentence behind “::” is two sentences vertically connected. The simple simple sentence is easier to understand than the detailed simple sentence because the character string not directly related to the simple sentence is deleted. In FIG. 14, the character string written in the slot of the simple simple sentence is shown as a simple writing word and a simple phrase. The semantic structure of “12th :: Tokyo :: stocks :: in the market” of the detailed writing word 1 in FIG. 14 can be understood by following the natural sentence generation path extending upward from mw_144 in FIG. It can be seen that “Tokyo” relates to “stock”, “stock” relates to “in the market”, and “12 days” relates to “in the market”. That is, it is clear that “12th day” is related to “in the market”, not related to “Tokyo”. Detail writing word 8 (Same day :: Nikko :: Stock :: Value :: Width :: Limit :: Lower limit :: Last weekend :: Ratio :: 30 :: Yen depreciation :: 37 yen Can be understood by following the natural sentence generation path extending upward from mw_24 in FIG. The “limit” of mw_214 relates to the “stock” of mw_209, the “width” of mw111 and the “lower limit” of mw_238, and the “width” of mw_111 relates to the “value” of mw_103 and the “limit” of mw_116. ing. “Yen depreciation” is related to “ratio”, “30” and “up to 37 yen”. In this way, it is complicated to express in words, but it is easy and clear in the semantic structure of FIG. From the figure, it can be seen that “yen depreciation” is not related to “previous weekend” but related to “ratio” of “previous weekend”. This means that “30 :: Yen depreciation” of the detailed vertical combination word 9 in FIG. 15 and “Last weekend :: ratio :: Yen depreciation” of the detailed vertical combination word 12 and “Yen of simplification 6 of FIG. It can be confirmed even from “An :: Up to 37 yen”.

As described above, when the input sentence is converted into a semantic structure, the semantic structure is decomposed into details, and the semantic structure is reconstructed based on the semantic structure, the semantic structure as shown in FIGS. 11 and 13 is obtained. It is done. When natural sentences are generated from the respective semantic structures, the natural sentences shown in FIGS. 12, 14, and 15 are obtained. From these semantic natural sentences, the meaning of the input natural sentence can be verified in detail. The words in the sentence are not vaguely related to the whole meaning, but the meaning defined by the clear semantic structure is combined to make the meaning robust. From this, if the meaning of the input sentence is shown, it can be determined whether or not the meaning intended by the writer is accurately expressed, so that the writer can assist in creating a correct sentence.

When performing machine translation, it is assumed that the exact sentence intended by the writer has been created. Even if a bad sentence is translated, the bad sentence is a bad sentence. First of all, it is important to create correct sentences in Japanese before translation. In addition, it is important to create a sentence that has an accurate meaning in official sentences that cause liability. It can also be used for intellectual training while creating sentences.

1 Central control unit,
2 internal memory,
3 display device,
4 external storage device,
5 input devices,
6 Memory hub,
7 IO hub,
8 Bus,
20 Natural sentence division program,
21 Morphological analysis program,
22 syntax analysis program,
23 phrase meaning construction program,
24 Morphological semantic construction program,
25 Natural sentence generation program,
30 MW element,
31 PS element 32 Semantic root collection element, Bun_ixrt [Bun_num]
40 meaning frames,
41 Meaning root,
42 Semantic structure symbols, (), [], ::: →
50 MW coupling element, Prc_mw []. U, L, B, N
Prc_mw []. FLG
51 MW character string element, Prc_mw []. moji_jar [], moji []
moji_jcs [], moji_ronri []
52 MW representation availability element, Prc_mw []. KAK
53 MW path blocking element, Prc_mw []. KAK
55 PS coupling element Prc_ps []. A, T, S, V, W.M. O. P, L
56 Natural Sentence Generation Path 57 Partial Semantic Structure 58 Whole Semantic Structure

Claims (2)

A central control device 1 connected by a bus 8 with a memory hub 6 and an IO hub 7 as a relay device, an internal memory 2, an external storage device 4, a display device 3, and an input device 5, and a Japanese dictionary and morphological analysis Rules such as rules, syntax analysis rules, morpheme semantic construction rules, clause semantic construction rules, dictionary search programs, morphological analysis programs, syntax analysis programs, semantic range dividing programs of input natural sentences, morphological semantic construction programs, clause semantic construction In a natural language processing apparatus that stores programs such as programs and natural sentence generation programs in the external storage device 4,
The MW element 30 stores at least the MW combination element 50 that stores the element number of the upper / lower / left / right combination partner and the type of the MW element or the PS element, and the MW that stores the character string obtained from the morphological analysis result and the phrase analysis result. A character string element 51, an MW expression availability element 52 for storing information for controlling the possibility of expression of the character string, and an MW path blocking element 53 for storing information for controlling whether to block the natural sentence generation path,
The PS element 31 has a single and a plurality of MW elements 30 that are vertically coupled above and a PS coupling element 55 that stores the numbers of the MW elements 30 that are vertically coupled below.
Semantic frames are composed of single and multiple MW elements and single and multiple PS elements combined, and single and multiple semantic frames and single and multiple MW elements are combined vertically and horizontally to construct a semantic structure. The MW element at the lowest layer is a semantic root 41 that combines these semantic structures, and the semantic root 41 is stored in the semantic root collection element 32.
The central control unit 1 takes out the morpheme analysis program 21 into the internal memory 2, performs the morpheme analysis by the morpheme analysis program 21, obtains the morpheme analysis result, and the parse analysis program 22 To get the parsing result
The morpheme semantic construction program 24 applies the morpheme semantic construction rules to the morpheme analysis results, and the program specified by the rules combines the MW elements and the semantic frames to construct partial semantic structures, and simultaneously performs the semantic analysis. The target character string is stored in the MW character string element 51, and the semantic structure is expanded by combining the existing semantic structure with the existing semantic structure having a connection relationship with the existing semantic structure. Store the semantic root in the semantic root collection element 32;
Further, the phrase semantic construction program 23 applies a phrase semantic construction rule to the phrase analysis result obtained by the syntax analysis result, and combines the MW elements and the semantic frames according to the program specified by the rule to obtain partial semantics. Constructing the structure, and simultaneously storing the character string subject to semantic analysis in the MW character string element 51,
The semantic structure is combined with the semantic structure that has already been created and connected with the semantic structure to construct the semantic structure, and the semantic root is stored in the semantic root element 32. Repeat to build a semantic structure with strings written on the semantic structure,
A natural sentence generation path is created based on the MW representation availability element 52 and the MW path blocking element 53, and a natural sentence representing the entire meaning of the input natural sentence is generated, or the natural sentence generation path is interrupted to partially Divide into semantic structures, generate natural sentence generation from the divided semantic structures, display the meaning of the input natural sentence in multiple directions, verify the input natural sentence in detail, and express the correct meaning based on this A sentence-supporting natural sentence processing method that supports the creation of natural sentences. The meaning of the input natural sentence is once converted into a semantic structure, and a natural sentence is generated from the semantic structure. The semantic structure in the natural sentence is displayed by adding the semantic structure symbol 42 to the generated natural sentence. The sentence creation support natural sentence processing method according to claim 1, wherein: