JP3772214B2 - Natural sentence ambiguity elimination device and natural sentence ambiguity elimination program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a device and a program for resolving the ambiguity of the meaning represented by a natural sentence in text data composed of a Japanese natural sentence already input to a computer.
[0002]
[Prior art]
In Japanese before the Middle Ages, there was a usage of “moment knot” in which the counselor and the utilization form at the end of the sentence had a morphological correspondence, but in the case of modern Japanese sentences where “moment knot” disappeared, predicate Is placed at the end of the sentence, so the contents of the sentence cannot be finalized unless it reaches the end of the sentence. For this reason, it is not clear that a sentence with a long and complex content does not read until the end of the sentence whether the content is positive or negative, or represents a question. That is, when multiple words that become “knots” corresponding to a single word that is related to a single “student” appear in the same natural sentence, which “knot” word is truly “study” for understanding the meaning of the sentence. It can be said that it is difficult to accurately grasp whether a word is supported. Here, in the study of sentence structure in contemporary Japanese, it has been suggested that certain adverbs in modern languages play a role similar to the counsel in ancient words, and foresee subsequent elements (for example, Non-Patent Document 1). For example, combinations such as “maybe ... probably” and “probably… probably” can be predicted to some extent by introspection and intuition as a pair of preceding element (calling element) and succeeding element (response element) that form a responsive relationship. It is pointed out that it is considered to be attached (for example, refer nonpatent literature 2).
[0003]
[Non-Patent Document 1]
Satoshi Ohno, “Study on Ties”, 1st Edition, Iwanami Shoten, January 12, 1993, p350-351
[Non-Patent Document 2]
Masashi Takashi, "Grammar of Modality", 1st Edition, Kuroshio Publishing, May 25, 1991, p29-46
[0004]
[Problems to be solved by the invention]
However, no systematic research has been conducted on such responsive pairs, and a small number of responsive pairs are exemplified in the above-mentioned documents and textbooks. In other words, in the study of “relationship” in modern Japanese, there is a lack of responsive pairs that cannot be predicted by introspection and intuition, so a word and a word that appears (co-occurs) with it are really responsive. It is necessary to say that there is too little basic data to clarify whether they are related. Therefore, it is possible to verify whether the Japanese text entered by the user himself / herself into the computer expresses the correct “relationship” expression, or the meaning of the natural Japanese text received by the user from another person. The only way to understand is to rely on the user's own intuition and introspection. Such a problem becomes more prominent as the input natural Japanese sentence is longer.
[0005]
Therefore, in view of the above problems, the present invention uses a database that accumulates responsive pairs that form a “knot” in a Japanese sentence. It is intended to find the responsive relationship that is presumed to be correct and to resolve the ambiguity of the natural sentence.
[0006]
[Means for Solving the Problems]
That is, the present invention provides a set of response pairs in which a call element and a response element forming a knot among two words co-occurring in a Japanese natural sentence are paired in the original natural sentence including the response pair. A natural computer consisting of a computer that resolves ambiguity of responsive relationships in the input natural Japanese text by searching the responsive pair database stored with the average inter-word distance between the call elements and response elements added A program for causing a sentence ambiguity eliminating apparatus and the computer to function as a natural sentence ambiguity eliminating apparatus.
[0007]
As shown in FIG. 1, the basic functional configuration diagram is shown by a solid line, the natural sentence ambiguity resolving apparatus A includes an input receiving unit 1 that receives input of Japanese text data, and a Japanese text received by the input receiving unit 1. Morpheme analysis means 3 for performing morpheme analysis on data, word number assignment means 4 for assigning word numbers to each morpheme analyzed by morpheme analysis means 3 in order from the top, and morpheme analyzed by morpheme analysis means 3 Call element extraction means 5 for extracting a predetermined call element from the call, and a response element corresponding to the call element extracted by the call element extraction means 5 from the response pair database DB1, and morpheme analysis means based on the search result A response element extraction means 6 for extracting a morpheme corresponding to a response element from among the morphemes analyzed in step 3, a call element extracted by the call element extraction means 5 and the response element extraction means The term distance calculating means 8 for calculating on the basis of the word spacing distance between all response elements extracted in the word number assigned by the word number providing means 4, all of the operation result by the inter-word distance calculating means 8 Response element specifying means 9 for specifying response elements having an interword distance in a predetermined relationship with an average distance between words obtained from the average from the responsive pair database DB1 is provided. .
[0008]
Here, the call element means a word corresponding to “relation” among two words constituting “relation” in a modern Japanese sentence. For example, according to the classification of “Basic Japanese Grammar Revised Edition” (Takashi Masuoka, Yukinori Takubo, Kuroshio Publishing, 1992), the words that can be call elements are: “Proposed particle”, “Training particle”, “Declaration particle” "Adverb" corresponds. The specific target words are “same”, “shika”, “even”, “ha”, “mo”, “just”, “only”, “even”, “if”, “about” (about) , "Just", "what", "never", "probably", "probably", "maybe", "by all means", "just", "if", "surely", etc. Words, but not necessarily limited thereto. The “response element” corresponding to “knot” is a word that appears at the same time when “call element” appears in a modern Japanese natural sentence. In the present invention, it is defined that “call element” and “response element” appear in the same sentence at the same time as “co-occur”, and the “call element” and “response element” to be “co-occur” A combination is called a “co-occurrence pair”, and “call element” and “response element” appearing in this order in the “co-occurrence pair” are defined as “response” and “response”. A combination of “call element” and “response element” is particularly referred to as “response pair”. The responsive pair database DB1 stores a “responsive pair” in which a call element and a response element obtained by extracting a large amount of modern Japanese natural sentence data as original text data are paired. Furthermore, the data of these answering pairs are classified for each call element, and each answering pair includes a word space between the call element and the answering element in a number of texts that form the basis for generating the answering pair database. The average value of distance is given. This inter-word distance is obtained by setting the distance between adjacent morphemes as 1 when the original text is analyzed by morpheme.
[0009]
Accordingly, by causing the computer to function as the natural sentence ambiguity eliminating apparatus A in this way, words having “relationships”, that is, responsive relationships are selected from the inputted Japanese natural sentences. Then, each inter-word distance between the call element and one or more response elements corresponding to the call element is compared with the average distance between the call element and the response element stored in the response pair database DB1, and the corresponding 1 By selecting one response element, it is possible to clarify the response relationship in the natural sentence and assist the user in verifying and understanding the sentence. Here, “input” means inputting character data into a computer using an input device such as a keyboard, mouse, or tablet, or receiving input from an external device such as another computer. In addition, “a response element having a distance between words having a predetermined relationship with the average distance between words” means a response element having a distance between words that is the same as or similar to the average distance between words, and a constant calculation with the average distance between words. It means a response element having an inter-word distance in a formula relationship.
[0010]
In particular, in order to make it easier for the user who uses the natural sentence ambiguity eliminating apparatus A to assist in understanding and verifying sentences, the Japanese text data received by the input receiving means 1 is used. An output means 2 for outputting, a relevance information output means 7 for outputting relevance information indicating a relevance between the call element extracted by the call element extraction means 5 and the response element extracted by the response element extraction means 6; It is preferable to further function as response element clarification means 10 for outputting response element clarification information that clearly specifies the response element specified by the element specifying means 9, and to further improve the visual usability, the output means 2, The sex information output means 7 and the response element clarification means 10 may each display and output Japanese text data, relevance information, and response element explicit information on a display device such as a display.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0012]
The natural sentence ambiguity resolving apparatus A according to this embodiment, as shown in the functional configuration of FIG. 1, receives responsive pair data that is a pair of a call element and a response element that form a “relationship” from a Japanese natural sentence. Using the collected response pair database DB1, find one call element and one or more response elements in the text data representing the natural Japanese sentence already entered, and identify one response element indicating the meaning of the sentence. This is to assist the verification of the Japanese natural sentence and the understanding by the user. This natural sentence ambiguity resolving apparatus A is constituted by a computer that has a built-in responsive pair database DB1 or is connected to the outside and can be searched. This computer has internal devices such as a CPU 101 and a memory 102 electrically connected by a bus line or the like, as shown in a schematic device configuration diagram in FIG. 2, and a storage device 103 such as a hard disk. A display device 104 such as a display, an input device 105 such as a keyboard and a mouse, and a normal personal computer equipped with external devices such as various communication interfaces 106. For example, the responsive pair database DB1 is connected to the outside via a communication line. It is. The data stored in the responsive pair database DB1 can also be stored in a storage device such as a hard disk.
[0013]
Then, the CPU 101 reads out the natural sentence ambiguity elimination program recorded in the storage device 103 and stores it in the memory 102. The CPU 101 performs processing according to the program, and the memory 102, the storage device 103 such as a hard disk, a monitor, etc. By driving devices such as the display device 104, the input device 105 such as a keyboard and a mouse, and various communication interfaces 106, the computer functions as the natural sentence ambiguity eliminating device A. Here, as shown in FIG. 1, the function of the natural sentence ambiguity resolving device A is as follows: input receiving means 1, output means 2, morpheme analyzing means 3, word number assigning means 4, call element extracting means 5, response element It refers to element extraction means 6, relevance information output means 7, interword distance calculation means 8, response element specification means 9, and response element clarification means 10. In particular, the input receiving unit 1 has a function of receiving data input by the user from the input device 105 such as a keyboard and a mouse and storing the data in the memory 102. The output unit 2, the semantic information output unit 4, and the fixed meaning The information output means 6 has a function of displaying characters on the display device 104 such as a display.
[0014]
In addition, the responsive pair database DB1 morphologically analyzes a large amount of Japanese natural sentence data collected from newspaper articles or the like as original text data, and a pair of co-occurrence in the same sentence based on the data after the morphological analysis. It is obtained by calculating the word (co-occurrence pair) by calculating the complementary similarity and calculating the reliability of the co-occurrence pairs that appear in the original text data in the order of “call element” and “response element”. It also stores “calling pairs”. Here, “reliability” refers to the ratio of the number of original sentences in which the corresponding response pair appears to the number of original sentences in which a certain call element appears, paying attention to each answering pair. In the present embodiment in which the reliability indicates a predetermined value or more, “Daily Newspaper Article Data” and “Nikkei Newspaper Article Data” for 10 years are used as original text data. Here, in this embodiment, the call element is a word classified as “proposed particle”, “collecting particle”, or “declarative adverb”, specifically, “same”, “shika”, “even”, “Wa”, “Wo”, “Just”, “Only”, “Even”, “If”, “About”, “Just”, “What”, “Never”, “Probably (probably ) ”,“ Maybe ”,“ Make sure ”,“ Made ”,“ Moshi ”,“ Surely ”, and have been described in the progressive sentence interpretation support program in advance. Response pairs are classified for each call element. Then, in this responsive pair database DB1, a word number is assigned to each morpheme obtained by morphological analysis of each original text data, and for each responsive pair, the corresponding element in the original text data in which the corresponding responsive pair appears. The difference between the word number and the word number of the call element is the inter-word distance, and the data is stored in the state where the average value of the inter-word distance in all original text data in which the corresponding pair appears is given as the "average inter-word distance" is there. Here, the distance between words between adjacent morphemes is 1. FIG. 3 shows a part of the contents of the responsive pair database DB1 regarding the call element “never”. In the figure, the response element corresponding to the call element “Never” is described, and in the right column of each response element, the average inter-word distance between the call element “Never” and the response element is described. ing.
[0015]
The operation of the natural sentence ambiguity resolving apparatus A is as follows: input receiving means 1, output means 2, morpheme analyzing means 3, word number assigning means 4, call element extracting means 5, response element extracting means 6, relevance information output means. 7. Corresponding to the inter-word distance calculation means 8, the response element specifying means 9, and the response element specifying means 10 is performed as follows.
[0016]
As shown in the flowchart of FIG. 4, the natural sentence ambiguity eliminating apparatus A first receives an input of text data of a Japanese natural sentence from the input device 105 such as a keyboard by the function of the input receiving means 1 (S1). Then, the text data is temporarily stored in the memory 102, and the character string is immediately displayed on the display device 104 such as a display by the function of the output means 2 (S2). Next, the morphological analysis unit 3 reads the text data that has been accepted from the memory 102 and executes morphological analysis, and the text data is decomposed into parts of speech (S3). Here, the morpheme analysis program includes, for example, “JUMAN” (http://www.lab25.kuee.kyoto-u.ac.jp/nl-resource/juman.html) or “tea bowl” as a Japanese morpheme analysis program. (Http://chasen.aist-nara.ac.jp/) can be used. And the word number is given to each morpheme in order from the sentence head by the function of the word number assigning means 4 (S4). Further, the function of the call element extraction unit 5 extracts morphemes corresponding to the above-described call elements from the morphemes obtained from the morpheme analysis results of the morpheme analysis unit 3 (S5). The extracted morpheme is temporarily stored in the memory 102. Further, the function of the response element extraction means 6 searches the response pair database DB1 for the previously extracted call element, and one or more morphemes that become response elements corresponding to the response element corresponding to the call element are obtained. The morpheme analyzed in step S3 is extracted from the morpheme (S6). This extracted response element is also temporarily stored in the memory 102. Here, the function of the relevance information output means 7 reads the call element and the response element from the memory 102 and clearly indicates the call element as relevance information (for example, a mark surrounding the call element with a circle or The underline attached to the call element) and an arrow mark extending from the mark toward the response element are output to the display device 104, and displayed superimposed on the character string displayed on the display device 104 in step S2. (S7). On the other hand, the function of the interword distance calculation means 8 performs an operation of subtracting the word number of the response element from the word number of each response element stored in the memory 102 to calculate the distance between words for each response element. (S8). The distance between words as a result of this calculation is also stored in the memory 102 in association with the corresponding response element. Further, the function of the response element specifying means 9 compares the inter-word distance related to each response element with the average inter-word distance of the response element stored in the responsive pair database DB1, and here, the average inter-word distance is the largest. The response element corresponding to the inter-word distance indicating a close value is determined as the response element corresponding to the call element extracted in step S5 (S9). Finally, the response element specifying information for specifying the response element determined in step S9 is displayed on the display device 104 by the function of the response element specifying means 10 (S10). Here, as an example of the process of displaying the response element explicit information, the response element is underlined, and in order to distinguish the arrow mark extending from the call element to the response element and the arrow mark extending to another response element, Although the method of representing the arrow mark to be represented by a solid line and the other arrow mark by a broken line is employed, other processing such as color-coding the arrow mark may be performed.
[0017]
Here, with reference to FIG. 5, a specific example of disambiguation in a Japanese natural sentence displayed on a display device 104 such as a display after input will be described. The following example relates to the call element “never”. First, as shown in FIG. 5 (a), when a Japanese natural sentence “The road is never easy” is input using the input device 105 and displayed on the display device 104, the natural sentence ambiguity described above. According to the flowchart of the operation of the feasibility canceling apparatus A, the morpheme analysis is performed on the natural sentence, a word number is assigned to each morpheme, and the call element “never” is extracted. In addition, "|" (vertical line) shown in the figure indicates a break between adjacent morphemes. Then, referring to the responsive pair database DB1 as shown in FIG. 3, when “Easy”, “De”, “Ha” and “None” are extracted as response elements corresponding to the call element “Never”, the relevance information From the call element “never”, arrows corresponding to each response element “Easy” “De” “Ha” “None” are displayed on the display device 104. Next, the inter-word distance between the call element “never” and each response element is calculated, and the response element “no” having the average inter-word distance closest to the inter-word distance is determined from the response pair database DB1. In the relevance information, an arrow from “never” to “never” is displayed on the display device 104 with bold lines as response element explicit information. Therefore, the user who uses this natural sentence ambiguity resolution device A uses the correct “relationship” expression in the input or input natural sentence, and among them, “never” and “never” are correct responses. You can see that there is a relationship. It should be noted that not only such a short natural sentence but also a relatively long natural sentence such as that shown in FIG. Can clearly indicate that they are in the correct responsive relationship.
[0018]
As described above, in the present embodiment, a plurality of responsive relations are specified by clearly indicating call elements and response elements that are appropriate to respond truly in the sentence from the text data of the input Japanese natural sentence. Can eliminate the ambiguity of natural sentences. Therefore, a user who uses this natural sentence ambiguity eliminating apparatus A verifies whether a sentence based on the text data input by himself / herself has a correct responsive relationship, or based on text data received from another person. It will be very helpful in understanding the true meaning of the text. This means that the natural sentence ambiguity eliminating device A and the natural sentence ambiguity eliminating program can be applied as a part of a natural sentence verification device or verification program in a Japanese input device or a Japanese input program. Become.
[0019]
It should be noted that the present invention is not limited to the above-described embodiment, and call elements and response elements that are in a correct “relationship” relationship from Japanese natural sentences after input are similarly applied to call elements other than “never”. Can be extracted, and the ambiguity of the Japanese natural sentence can be resolved. Also, for example, a response element “no difference” composed of a plurality of morphemes combining a response element “difference” corresponding to the call element “probably” and a response element “no” is generated, and the average word interval If the distance is stored in the responsive pair database, the combination of the true call element and the response element can be extracted from the input text data more practically as in the above-described embodiment.
[0020]
In addition, the specific configuration and function of each part are not limited to the above embodiment, and various modifications can be made without departing from the spirit of the present invention.
[0021]
【The invention's effect】
As described above in detail, according to the natural sentence ambiguity eliminating apparatus or the natural sentence ambiguity eliminating program according to the present invention, a plurality of responsive relationships appear in a Japanese natural sentence based on input text data. If the meaning of the sentence becomes ambiguous, the average distance between the call element and the response element and the predetermined relationship among the inter-word distances between the call element and the response element in the natural sentence. Because the response elements with the distance between words and the responsive relationship are mechanically selected, the most probable responsive relationship is revealed, the ambiguity is resolved, and the user understands the meaning of the natural sentence And is very useful for verifying whether the correct responsiveness is used in natural sentences. Further, by applying the present invention to a Japanese language input program, a Japanese language verification program, etc., the present invention can also improve their reliability.
[Brief description of the drawings]
FIG. 1 is a diagram schematically showing a functional configuration of a natural sentence ambiguity eliminating apparatus according to the present invention and an embodiment thereof.
FIG. 2 is a schematic device configuration diagram of a computer configuring the natural sentence ambiguity eliminating device according to the embodiment.
FIG. 3 is a view showing a part of responsive pair data used in the embodiment.
FIG. 4 is a flowchart showing an overview of the operation of the natural sentence ambiguity eliminating apparatus.
FIG. 5 is a diagram showing a specific example of disambiguation of text data using the natural sentence disambiguation apparatus.
[Explanation of symbols]
A ... Natural sentence ambiguity eliminating device DB1 ... responsive pair database 1 ... input receiving means 2 ... output means 3 ... morpheme analyzing means 4 ... word number assigning means 5 ... call element extracting means 6 ... responsive element extracting means 7 ... relevance information Output means 8 ... Word distance calculation means 9 ... Response element specifying means 10 ... Response element specifying means

Claims (6)

A set of response pairs in which a call element and a response element forming a knot among two co-occurring words in a Japanese natural sentence are defined as a call element and response element in the original natural sentence including the response pair. It is composed of a computer that resolves the ambiguity of the responsiveness relationship in the input Japanese natural sentence text by searching the responsive pair database stored with the average inter-word distance added.
An input receiving means for receiving input of Japanese text data;
Morphological analysis means for performing morphological analysis on the Japanese text data received by the input receiving means;
Word number assigning means for sequentially assigning word numbers from the beginning to each morpheme analyzed by the morpheme analyzing means;
Call element extraction means for extracting a predetermined call element from the morpheme analyzed by the morpheme analysis means;
A response element corresponding to the call element extracted by the call element extraction means is searched from the response pair database, and a morpheme corresponding to the response element is extracted from the morpheme analysis means by the morpheme analysis means based on the search result. Responding element extraction means,
Inter-word distance calculation means for calculating the inter-word distance between the call element extracted by the call element extraction means and all response elements extracted by the response element extraction means based on the word number assigned by the word number assigning means. When,
Response element specifying means for specifying one response element having an interword distance in a predetermined relationship with an average word distance obtained by averaging all calculation results by the word distance calculation means from the response pair database; A natural sentence ambiguity resolving apparatus characterized by comprising: Output means for outputting Japanese text data received by the input receiving means;
Relevance information output means for outputting relevance information indicating the relevance between the call element extracted by the call element extraction means and the response element extracted by the response element extraction means;
The natural sentence ambiguity resolving apparatus according to claim 1, further comprising response element specifying means for outputting response element specifying information specifying the response element specified by the response element specifying means.   3. The natural sentence according to claim 2, wherein the output means, the relevance information output means, and the response element explicit means display and output Japanese text data, relevance information, and response element explicit information, respectively, on a display device such as a display. Disambiguation device. A set of response pairs in which a call element and a response element forming a knot among two co-occurring words in a Japanese natural sentence are defined as a call element and response element in the original natural sentence including the response pair. A computer that resolves the ambiguity of responsiveness in the input natural Japanese text by searching the responsive pair database stored with the average distance between words added,
An input receiving means for receiving input of Japanese text data;
Morphological analysis means for performing morphological analysis on the Japanese text data received by the input receiving means;
Word number assigning means for sequentially assigning word numbers from the beginning to each morpheme analyzed by the morpheme analyzing means;
Call element extraction means for extracting a predetermined call element from the morpheme analyzed by the morpheme analysis means;
A response element corresponding to the call element extracted by the call element extraction means is searched from the response pair database, and a morpheme corresponding to the response element is extracted from the morpheme analysis means by the morpheme analysis means based on the search result. Responding element extraction means,
Inter-word distance calculation means for calculating the inter-word distance between the call element extracted by the call element extraction means and all response elements extracted by the response element extraction means based on the word number assigned by the word number assigning means. When,
Response element specifying means for specifying one response element having an interword distance in a predetermined relationship with an average word distance obtained by averaging all calculation results by the word distance calculation means from the response pair database; A natural sentence ambiguity resolving program characterized by causing a natural sentence ambiguity resolving apparatus to function. The computer further
Output means for outputting Japanese text data received by the input receiving means;
Relevance information output means for outputting relevance information indicating the relevance between the call element extracted by the call element extraction means and the response element extracted by the response element extraction means;
5. The natural sentence ambiguity elimination program according to claim 4, wherein the natural sentence ambiguity elimination program functions as a natural sentence ambiguity elimination device provided with response element clarification means for outputting response element clarification information that specifies the response element specified by the response element specification means. 6. The nature according to claim 5, wherein the output means, the relevance information output means, and the response element explicit means function to display and output Japanese text data, relevance information, and response element explicit information on a display device such as a display. A sentence ambiguity resolution program.

Images