JP2000099536A - Automatic summarizing device and storage medium recording automatic summarizing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an optimum summary sentence from a text and a summary rate. SOLUTION: A device 1 which morpheme-analyzes a text and outputs morpheme information, a device 2 calculating the minimum value of the number of characters reduced from a summary rate and the number of characters in the text, a device 3 adding a word string substituted for the word string of the text based on morpheme information and outputting word lattice structure to which substitution cost is added, a device 4 searching word lattice structure from a sentence end to a sentence head and outputting the largest reduction number of characters from the sentence end to respective nodes, a device 5 searching word lattice structure from the sentence head to the sentence end, calculating the reduction number of characters from the sentence head against respective paths, obtaining the sum of the reduction number of characters in the device 4, branch-cutting the path when the sum is smaller than the desired reduction number of characters and outputting the path when it is not smaller, a device 6 calculating the sum of substitution cost against the output path of the device 5 and outputting the path from the sentence head to the sentence end when a system reaches the sentence end and a device 7 outputting the path where the sum of substitution cost becomes minimum in the output path of the device 6 are installed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic summarizing apparatus for automatically summarizing an optimal summary sentence at a desired summarization rate when a desired summarization rate is given to an original sentence. And a recording medium on which an automatic summarizing program is recorded.

[0002] The present invention is suitable for automatically summarizing sentences such as TV news and the like, and can be applied to caption creation.

[0003]

2. Description of the Related Art Automatic summarization is to automatically reduce an original sentence by replacing the original sentence with a short word sequence. In order to implement automatic summarization, it is necessary to have a knowledge of which word string in the original sentence is to be replaced with what kind of word string. For example, the following is an example of replacement knowledge.

[Replacement knowledge 1] (Original word string → summary sentence word string) [Replacement knowledge la] Clarified → Expression, replacement cost = 0.6 [Replacement knowledge Ib] Clear → Expression, replacement cost = 0 .
3 [Replacement knowledge 1c] → φ (φ is empty, symbol for omission), replacement cost = 0.2 Here, the “replacement cost” attached to the replacement knowledge is the penalty for using the replacement knowledge. It is stated that as the replacement cost is closer to 0, the replacement knowledge is easier to use.

[0005] Using the above example of the permutation knowledge 1, consider summarizing the following original sentence 1 at the following summarization rate. It is assumed that the summarization rate = the number of characters in the summary sentence / the number of characters in the original sentence × 100. [Original text 1] "Clarified" (8 characters) [Summary rate 1] Summarization rate 60% or less.

To reduce the summarization rate to 60% or less,
The number of characters to be deleted is as follows.

8 characters × (100-60)% = 3.2 characters That is, the number of characters of 3.2 or more in the original text must be reduced.

In conventional automatic summarization, a summary sentence is obtained by sequentially applying replacement knowledge in the order of appearance. In addition, the number of replacement knowledge used there was very small because it was manually collected and created. (For example, Document 1:
"Yamamoto et al .:" Editorial sentence summarization system GREEN using compound structure of sentences "Natural language processing, Vo1.2, N
o. 2, pp. 39-55, I994], automatic summarization is performed by sequentially applying summary knowledge including replacement knowledge created manually. ).

In the example in which the summarization rate is 60% or less, as in the case of the above-mentioned document 1, when the number of reduced characters of 3.2 or more is sequentially applied in the order of appearance of the replacement knowledge 1, the replacement knowledge 1a is applied, and the following summary sentence 1a is obtained. can get.

[Summary Sentence la] "Announcement" (reduced number of characters = 6 characters, sum of replacement cost = 0.6)

[0011]

However, recently, a technique for automatically creating replacement knowledge has been developed, and a large amount of replacement knowledge can be easily obtained. (For example, Reference 2: "Naoto Kato:" Acquisition and Evaluation of Local Summary Knowledge for Summarizing News Sentences ", IEICE Technical Committee on Language Understanding and Communication, NLC 98-16, pp. 7)
-14, 1998 ").

When the replacement knowledge is greatly increased in this way, a plurality of replacement knowledges may be applied to a single original word string in some cases. In some cases, a conflict occurred, and an optimal summary could not be obtained.

Also in the above example, the case of the summary sentence 1a obtained by applying the replacement knowledge 1a and the case of the summary sentence 1b obtained by applying the replacement knowledge 1b and 1c conflict with each other. Is more appropriate because the sum of the replacement costs is closer to 0.

[Summary Sentence 1b] "Expressed" (reduced number of characters = 4 characters, sum of replacement costs = 0.5) Therefore, an object of the present invention is to provide an original sentence and a desired abstract in view of the above points. To be able to automatically obtain the optimal summary sentence of the original sentence by appropriately and efficiently selecting replacement knowledge using the replacement cost and the number of character reductions given the rate. It is in.

[0015]

In order to achieve the above object, an automatic summarizing apparatus according to the first aspect of the present invention morphologically analyzes an input original sentence, divides a character string included in the original sentence into words, and converts the part of speech into words. Morphological analysis means for outputting the assigned morphological information, character reduction number minimum value calculation means for calculating the minimum value of the number of characters to be reduced from the summarization rate of the input original text and the input number of characters of the original text, Based on the morphological information output from the analyzing means, the original word string is compared with the replacement knowledge obtained in advance from the head word string, and if the matching is successful, replacement is performed. A replacement knowledge search unit that adds a word string to the morphological information and outputs a word lattice structure to which the replacement cost is added, and that the word lattice structure output from the replacement knowledge search unit A backward character reduction number calculation means for searching for the maximum possible character reduction number from the end of the sentence of the word lattice structure to each node, and a word lattice structure output from the replacement knowledge search means; From the beginning to the end of the sentence, calculate the number of character reductions from the beginning of the sentence for each path, obtain the sum of the calculation result and the number of character reductions output from the backward character reduction number calculation means, and calculate the sum. If the number is smaller than the desired character reduction number, the path is pruned; if the sum is not smaller than the desired character reduction number, the forward character reduction number calculation means for outputting the path; A replacement cost calculating means for calculating the sum of the replacement costs of the paths output from the character reduction number calculating means and outputting a path from the beginning to the end of the sentence when the sentence end is reached; And a replacement cost minimum path calculating means for obtaining a path in which the sum of the replacement costs from the beginning to the end of the sentence is the minimum among the paths output from the path, and outputting a character string of the obtained minimum path as a summary sentence. Features.

[0016] Preferably, the replacement knowledge search means keeps an essential application replacement knowledge list for each path, and refers to the list when applying replacement knowledge.

[0017] Preferably, the backward character reduction number calculating means includes a sentence end node n (n
Is the number of words) and the node number i is 1
At each node i, the maximum possible number of character reductions from the last node n of the sentence to the current node i, that is, the maximum value m _b (i) of the number of character reductions, is obtained by dynamic programming.

[0018] Preferably, the forward character reduction number calculating means is configured such that the maximum value m _b (0) of the character reduction number from the last node to the first node is smaller than the minimum value m of the desired number of characters to be reduced. In this case, it is output that summarization cannot be performed at the input summarization rate.

Preferably, the forward character reduction number calculation means calculates the maximum value m _b (0) of the character reduction number from the last node of the sentence to the first node of the sentence as the minimum value m of the number of characters to be reduced.
If it is not smaller than 1, the node number i is 1 from the head node 0 of the word lattice structure to the end node n of the sentence.
For each path up to each node i, the character reduction number m _f (i) of the path is calculated, and the character reduction number m _{b at the} node i obtained by the backward character reduction number calculation means is calculated. (I) and the sum m _f
If (i) + _mb (i) is smaller than the minimum value m of the number of characters to be reduced, the path is pruned and the sum m _f
If (i) + _mb (i) is not smaller than the minimum value m of the number of characters to be reduced, the path is output.

[0020] Preferably, the forward character reduction number calculation means, when a word having no positive example is reached during the forward calculation processing, has the same condition (forward character reduction number, essential application replacement list, etc.) The candidate also prunes a path whose replacement cost is not the minimum.

According to a seventh aspect of the present invention, there is provided a recording medium comprising: an automatic summarization program for automatically obtaining an optimum summary sentence from an original sentence and a desired summarization rate by a computer; A recorded medium,
The automatic summarization program causes the computer to morphologically analyze the input original sentence to divide a character string included in the original sentence into words and generate morphological information to which the part of speech is added, and to calculate a summarization rate of the input original sentence and A minimum value of the number of characters to be reduced is calculated from the number of characters of the input original sentence and, based on the morphological information, the word sequence of the original sentence is sequentially compared with the replacement knowledge obtained in advance from the leading word sequence. When the collation is successful, the word string to be replaced is added to the morphological information when the collation is successful, and the word lattice structure with the replacement cost is obtained, and the word lattice structure is directed from the end of the sentence to the beginning of the sentence. The search is performed to calculate the maximum possible character reduction number (rearward character reduction number) from the end of the sentence of the word lattice structure to each node. To calculate the number of character reductions from the beginning of the sentence for each path, obtain the sum of the calculation result and the number of backward character reductions, and when the sum is smaller than the desired number of character reductions, Causes the path to be pruned,
If the sum is not less than the desired number of character reductions,
The path is selected, the sum of the replacement costs is calculated for the selected path, and when the end of the sentence is reached, the path from the beginning to the end of the sentence is output. A path that minimizes the sum of the replacement costs up to and including the obtained minimum path is output as a summary sentence.

Preferably, the automatic summarization program is input to the computer when the maximum value of the number of character reductions from the last node to the first node is smaller than the minimum value of the desired number of characters to be reduced. Output that the summarization rate cannot be summarized.

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Before describing embodiments of the present invention, the above-mentioned replacement cost required for automatic summarization according to the present invention will be described.

The summary knowledge required for automatic summarization is composed of two pieces: replacement knowledge and replacement conditions. As described above, the replacement knowledge is the knowledge that defines the word string to be replaced with the word string of the original sentence. For example, knowledge of omitting the adnominal particle "no". On the other hand, the replacement condition is a numerical value indicating whether or not the replacement knowledge is applied, that is, the replacement cost described above. The success or failure of the replacement knowledge is determined by the word strings before and after it. For example, "Japanese Bank"
Cannot be omitted.

Therefore, the replacement cost is calculated by calculating the distance between the word string before and after the replacement knowledge and the replacement condition acquired in advance. That is, the replacement cost of replacing the i-th to j-th word strings w _ij with the word strings x _ij is
If it is expressed as distsub (w _ij → x _ij ), it is defined by equation (1) (for further details, see the above reference 2).

[0026]

(Equation 1)

In the above equation (1), if there is a positive example, the value of g _low (= 0.01) to 1.0 (0.0 ≦ g (w _ij → x _ij 1.
0, positive example) ≤ 1.0)), and it is defined that the substitution can be made closer to 0.0. If there is no positive example (there is no permutation knowledge to be applied), take 0.0.

The outline of the automatic summarization algorithm according to the present invention will be described. For the sake of simplicity, a case where one sentence is summarized below will be considered. If it spans multiple sentences, it may be simply connected.

Now, it is assumed that the original text is to be summarized below a certain summarization rate. At this time, characters that are equal to or more than m (= the number of characters in the original text × the summarization rate) must be deleted. Furthermore, I want the best summary. Here, the “optimal summary” is defined as a case where the sum of the costs of the applied replacement knowledge (replacement cost) is minimized. Therefore, automatic summarization is
This is to obtain a path (optimal path) with the minimum replacement cost from the beginning of the sentence to the end of the sentence by deleting the number of characters equal to or more than the character. Formulated as (2).

[0030]

Argmin Σ distsub (w _ij → x _ij ) (2) x∈X x X = {(x ₀ ,…, x _ij ,…, x. _N ) | Σ (| w _ij |-| x _ij |) ≧ m)｝ The algorithm for finding the solution of equation (2) will be described later with reference to FIGS. In the present invention, two evaluation functions of the number of character reductions and the replacement cost are used, but a heuristic function (predicted value of the evaluation function from the current node to the goal) is used when calculating the former.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 shows an apparatus configuration of an embodiment of the present invention. In FIG. 1, il is a terminal for inputting an original sentence, and i2 is a terminal for inputting a summary rate. ol is a terminal for outputting a summary sentence.

The morphological analyzer 1 morphologically analyzes the original sentence input to the terminal of i1, divides a character string included in the original sentence into words, and outputs morphological information to which the parts of speech are added.

The character reduction number minimum value calculation device 2 calculates the minimum value of the number of characters to be reduced from the summarization rate of the original sentence input to the terminal i2 and the number of characters of the original sentence input to the terminal i1. Based on the morphological information output from the morphological analysis device 1, the replacement knowledge search device 3 checks the word sequence of the original sentence in order from the leading word sequence with replacement knowledge that has been obtained in advance. If successful, a word string to be replaced is added to the morphological information, and a word lattice structure with the replacement cost is output. In this example, it is assumed that the replacement cost is stored in advance in an internal memory (not shown) together with the replacement knowledge along with each replacement knowledge.

The backward character reduction number calculator 4 searches the word lattice structure output from the replacement knowledge search device 3 from the end of the sentence to the beginning of the sentence, and reduces the maximum possible character from the end of the word lattice structure to each node. Calculate the number.

The forward character reduction number calculation device 5 searches the word lattice structure output from the replacement knowledge search device 3 from the beginning of the sentence toward the end of the sentence, and calculates the number of character reductions from the beginning of the sentence for each path. When the sum of the calculation result and the character reduction number output from the backward character reduction number calculation device 4 is obtained, and the sum is smaller than the desired character reduction number (calculated from the input original text and the summarization rate), The path is pruned, and if the sum is not less than the desired number of character reductions, the path is output.

The replacement cost calculator 6 calculates the sum of the replacement costs of the paths output from the forward character reduction number calculator 5, and outputs the path from the beginning to the end of the sentence when the end of the sentence is reached.

The replacement cost minimum path calculation device 7 finds a path that minimizes the sum of the replacement costs from the beginning to the end of the sentence among the paths output from the replacement cost calculation device 6, and obtains the character string of the obtained minimum path. Is output from the terminal o1 as a summary sentence.

FIG. 2 is a flowchart showing an example of a procedure for performing automatic summarization processing by the apparatus configuration of FIG. The procedure of automatic summarization according to the present invention will be described below with reference to FIG.

First, when the original text is input to the terminal of i1,
In step S1, the original sentence is morphologically analyzed by the morphological analysis device 1, and its word division and part of speech are output. When the digest rate is input to the terminal of i2, in the next step S2, the minimum value of the number of characters to be reduced is calculated from the number of characters of the original text and the digest rate.

In the following step S3, step S1
Based on the morphological analysis result obtained in step 1, the matching is performed with the replacement knowledge obtained in advance from the head word string, and if the matching is successful, the word string to be replaced is added to the morphological analysis result. , Giving the replacement cost to create a word lattice structure.

In steps S4 to S7, backward calculation is performed. That is, in steps S4, S6, and S7, the node number i is decremented by one from the sentence end node n (n is the number of words) of the word lattice structure toward the beginning node 0, and the processing of step S5 is executed at each node i. . Step S
In step 5, the maximum possible character reduction number (backward character reduction number) from the last node n to the current node i, that is, the maximum value m _b (i) of the character reduction number is obtained by dynamic programming (DP) and stored.

In the next step S8, if the maximum value m _b (0) of the number of character reductions from the last node of the sentence to the first node of the sentence is smaller than the minimum value m of the desired (specified) number of characters to be reduced, After executing step S9, the automatic summarization process ends. In step S9, the fact that summarization is not possible at the input summarization rate is output. On the other hand, if the maximum value m _b (0) of the number of reduced characters from the last node to the first node is not smaller than the minimum value m of the number of characters to be reduced, step S10 is executed.

In steps S10 to S16, a forward calculation is performed. In steps S10, S15, and S16, the node number i is incremented by one from the head node 0 of the word lattice structure to the end node n of the sentence, and the processing of steps S11 to S14 is executed on all paths up to each node i. First, in step S11, the number of character reductions (forward character reduction number) m _f (i) of the path is calculated, and the number of character reductions m _b in node i obtained in step S5 described above.
(I) and the sum m _{f in} step S12
If (i) + _mb (i) is smaller than the minimum value m of the number of characters to be reduced, this path cannot be the final solution, so the path is pruned in step S13. The number of possible path candidates increases as the sentence ends, but the number of candidates can be reduced by such pruning. On the other hand, if the sum m _f (i) + _mb (i) is not smaller than the minimum value m of the number of characters to be reduced, step S14 is executed. In step S14, the replacement cost cost of the path
The sum of (i) is obtained, and when reaching the end of the sentence, a path from the beginning to the end of the sentence is output.

In the next step S17, a path which minimizes the sum of the replacement costs is obtained from the paths obtained in step S14, and a summary sentence is output from the terminal o1.

Further, a specific example of the processing of FIG. 2 will be described using a case where the following original text 2 is summarized as follows.

[Original 2] "Prime Minister Fukuzawa clarified his policy at the special committee." (24 characters) [Summary rate 2] Summarization rate of 70% or less.

It is also assumed that the following has been obtained in advance as replacement knowledge.

[Replacement Knowledge 2] [Replacement Knowledge 2a] Prime Minister / Normal Noun → Prime Minister / Common Noun, Replacement Cost = 0.01 [Replacement Knowledge 2b] Committee / Common Noun → Commission / Common Noun, Replacement Cost = 0 .02 [Replacement knowledge 2c] Obvious / adjective noun to / case particle / replacement / replacement polite / replacement past → representation / replacement noun, replacement cost = 0.6 [replacement knowledge 2d] Noun / case particle → assertion / sa-variant noun, replacement cost = 0.3 [replacement knowledge 2c] →→ φ (φ is a symbol representing the sky), replacement cost = 0.2 where the word is “surface expression / Part of speech ”.

When an original sentence is input to the terminal of i1, in step S1, the original sentence is morphologically analyzed. Then, as a result of the morphological analysis, the original sentence 2 is divided into 13 words (the number is represented by n) as shown in FIG. In FIG. 3A, the numbers in the upper row are the node numbers assigned between words in order from the beginning of the sentence. At this time,
The number of the sentence start node is 0, and the number of the end node of the sentence is the number of words n (= 13).

When the digest rate of 70% is input to the terminal of i2, in step S2, since the number of characters in the original text is 24, the minimum value of the number of characters to be reduced (this is represented by m, and the desired character reduction (Called a number) is calculated as follows:

[Desired character reduction number] m = 24 characters × (100−70)% = 7.2 characters In step S3, based on the result of the morphological analysis, collation with replacement knowledge 2 is performed in order from the first word string. If the collation is successful, the word string to be replaced is added to the morphological analysis result, and a word lattice structure to which the replacement is added is created. Then, a word lattice structure as shown in FIG. 3B is obtained. In FIG. 3B, for example, “0.01” written next to “Prime Minister” of the replacement candidate indicates the replacement cost.

In steps S4 to S7, backward calculation is performed. In steps S4, S6 and S7, the node number i is decreased from i = 13, 12,..., 0 by 1 from the end node 13 of the word lattice structure to the head node 0, and step S4 is executed in each node i. . In step S4, the maximum possible character reduction number from the end-of-sentence node 13 to the current focused node i (this is represented as m _b (i) and referred to as the backward character reduction number) is calculated. At this time, processing is efficiently performed by dynamic programming. Then
As shown in FIG. 3C, the backward character reduction number at each node is calculated.

In FIG. 3 (c), for example, the backward character reduction number m _b (8) at the node 8 is the case 8a, the replacement knowledge 2d “better → φ” and the replacement knowledge 2e.
Apply “Clearly → Statement”. (Number of character reduction 4) Case 8b Replacement knowledge 2c "Clarified → assertion"
Apply. (Character reduction number 6) There are two cases, but in case 8b, the number of character reduction is the largest, so that m _b (8) = 6 is obtained.

In step S8, the maximum character reduction number m _b (0) = 10 from the sentence end node to the head node is not smaller than the desired character reduction number m = 7.2 (m _b (0) = 1).
Since 0 ≧ m = 7.2), step S10 is executed without executing step S9.

In steps S10 to S16, a forward calculation is performed. In steps S10, S15, and S16, the node number i is incremented by one from i = 0, 1,..., 13 from the head node 0 of the word lattice structure to the end node 13 of the word lattice structure. Steps S11 to S14 are executed. In step S11, the number of character reductions in each path from the beginning of the sentence to the node i of interest is calculated (referred to as m _f (i) and referred to as the number of forward character reductions). Taking the case of node 5 (i = 5) as an example, four paths are obtained as shown in FIG.

Next, the sum with the backward character reduction number m _b (i) obtained in step S5 is calculated. In the example of the node 5, when the sum of the backward character reduction number m _b (5) = 6 is obtained, FIG.
The result is as shown in FIG.

The sum (m _f (i) + m shown in FIG.
_b (i)) is smaller than the desired character reduction number m (m
_{If f} (i) + _mb (i) <m), step S13 is executed, and if not smaller ( _mf (i) + _mb (i) ≧ m)
, Execute step S14. In step S13, the path is pruned. In step S14, the sum of the replacement costs of the path is obtained. In the example of the node 5, the path 5a
Is a path in which the sum of the forward character reduction number and the backward character reduction number is smaller than the desired character reduction number (m = 7.2), and is pruned in step S13, and is not used in the subsequent processing. Since the sum of the paths 5b, 5c, and 5d is not smaller than the desired character reduction number (m = 7.2), step S
14 is executed. In step S14, the sum of replacement costs in each path is obtained. FIG. 4C shows an example of the node 5.

Similarly, in step S14, the sum of the replacement costs in each pass from the beginning to the end of the sentence is obtained. When reaching the end of the sentence, four summary candidates as shown in FIG. 5A, which are paths from the beginning of the sentence to the end of the sentence, are obtained.

In step S17, among the summary candidates,
The path l3d with the minimum replacement cost is selected, and the next summary sentence obtained by connecting the word strings is output from the terminal o1 (see FIG. 5B).

[Summary Sentence 2] "Fukuzawa announces policy at special committee" (Other Embodiments) In the above description, it is assumed that the permuted knowledge applied is independent. However, there is some replacement knowledge, such as “Prime Minister → Prime Minister”, that must be applied once after applying it. To deal with this, keep a list of required applied replacement knowledge for each path,
A process of referring to the list when applying the replacement knowledge may be added to the algorithm of FIG.

During the forward calculation in steps S10 to S16, when a word having no positive example (for example, “de” in FIG. 4A) is reached, a condition (forward character reduction count) is set. , The essential application replacement list, etc.) are the paths whose replacement cost is not the minimum (for example, path 5 in FIG.
It is also possible to add the improvement of pruning b).

The morphological analyzer 1, the character reduction number minimum value calculator 2, and the like shown in FIG. 1 may be constituted by hardware such as a module circuit (device) using a ROM or the like.
It may be configured by software operating under computer control, such as a morphological analysis routine and a character reduction number minimum value calculation routine. Further, the present invention can be applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.).
The present invention may be applied to a dedicated device including one device (for example, an automatic summarizing device, an automatic subtitle creating device, and the like).

Another object of the present invention is to provide a recording medium (storage medium) recording software program codes for realizing the functions of the above-described embodiments to a system or an apparatus, and to provide a computer for the system or the apparatus. Needless to say, the present invention is also achieved when the CPU (or the CPU or the MPU) reads out and executes the program code stored in the recording medium. In this case, the program code itself read from the recording medium realizes the function of the above-described embodiment, and the recording medium (for example, CD-ROM, MD,
Floppy, etc.) would constitute the present invention.

[0065]

As is clear from the above description, according to the present invention, the original sentence is optimally summarized at a given summarization rate or less by using two evaluation functions of the character reduction number and the replacement cost. By simply inputting the original sentence and the summarization rate, the optimum summarization can be automatically obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an automatic summarizing apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating a procedure of automatic summarization performed by the apparatus of FIG. 1;

FIG. 3 is a diagram for sequentially explaining the processing contents and results of each stage of one embodiment of the present invention when a specific original sentence is input, (a) a morphological analysis result, (b) a word lattice structure,
(C) The calculation of the number of backward character reductions is shown by a specific example.

4A and 4B are explanatory diagrams continuing from FIG. 3; (a) calculation of the number of forward character reductions, (b) calculation of the sum of the number of forward character reductions and backward character reduction, and (c) calculation of the sum of the replacement cost. The calculation is shown by a specific example.

FIG. 5 is an explanatory diagram that is continuous with FIG. 4;
(B) A summary sentence is shown by a specific example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Morphological analyzer 2 Character reduction number minimum value calculation device 3 Replacement knowledge search device 4 Backward character reduction number calculation device 5 Forward character reduction number calculation device 6 Replacement cost calculation device 7 Replacement cost minimum path calculation device

Claims (8)

[Claims] 1. A morphological analysis means for morphologically analyzing an input original sentence, dividing a character string included in the original sentence into words, and outputting morphological information to which the part of speech is added, and a summarization rate of the input original sentence. A character reduction number minimum value calculating means for calculating a minimum value of the number of characters to be reduced from the number of characters of the original text, and a head of the word string of the original text based on the morphological information output from the morphological analysis means. In order from the word string of
A replacement knowledge search means for adding a replacement word string to the morphological information when the collation is successful, and outputting a word lattice structure provided with the replacement cost; and the word lattice output from the replacement knowledge search means. Backward character reduction number calculation means for searching the structure from the end of the sentence to the beginning of the sentence and calculating the maximum possible number of character reductions from the end of the sentence of the word lattice structure to each node; and the word output from the replacement knowledge search means The lattice structure is searched from the beginning to the end of the sentence, the number of character reductions from the beginning of each path is calculated for each path, and the sum of the calculation result and the number of character reductions output from the backward character reduction number calculation means is obtained. If the sum is less than the desired number of character reductions, the path is pruned; if the sum is not less than the desired number of character reductions, the forward Character replacement number calculation means, and a replacement cost calculation means for calculating the sum of the replacement costs of the paths output from the forward character reduction number calculation means, and outputting the path from the beginning to the end of the sentence when the sentence is reached. A path that minimizes the sum of the replacement costs from the beginning to the end of the path output from the replacement cost calculation means, and outputs a character string of the determined minimum path as a summary sentence; And an automatic summarizing apparatus. 2. The replacement knowledge search means stores an essential application replacement knowledge list for each path and refers to the list when applying replacement knowledge.
The automatic summarizing apparatus according to claim 1. 3. The backward character reduction number calculation means decreases a node number i by one from a sentence end node n (n is the number of words) of the word lattice structure toward a sentence head node 0,
At each node i, the last node n to the current node i
3. The automatic summarizing apparatus according to claim 1, wherein the maximum possible number of character reductions, that is, the maximum value m _b (i) of the character reduction number is obtained by dynamic programming. 4. The forward character reduction number calculation means includes a maximum value m _b of character reduction numbers from a sentence end node to a sentence start node.
4. The method according to claim 1, wherein when (0) is smaller than a minimum value m of a desired number of characters to be reduced, a message that summarization cannot be performed at the input summarization rate is output. Automatic summarization device. 5. The forward character reduction number calculation means includes a maximum value m _b of character reduction numbers from a sentence end node to a sentence start node.
If (0) is not smaller than the minimum value m of the number of characters to be reduced, the node number i is increased by one from the head node 0 of the word lattice structure to the end node n of the word lattice structure, and each node i is increased. , Calculate the character reduction number m _f (i) of the path, and sum with the character reduction number m _b (i) at the node i obtained by the backward character reduction number calculation means, The sum m _f (i) + _mb
If (i) is smaller than the minimum value m of the number of characters to be reduced, the path is pruned and the sum m _f (i) + _mb
The automatic summarizing apparatus according to claim 4, wherein when (i) is not smaller than the minimum value m of the number of characters to be reduced, the path is output. 6. The forward character reduction number calculation means, when a word without a positive example is reached during the forward calculation processing, the same candidate of the condition (forward character reduction number, mandatory application replacement list, etc.) 6. The automatic summarizing apparatus according to claim 5, wherein a path whose replacement cost is not minimum is also pruned. 7. A recording medium recording an automatic summarization program for automatically obtaining an optimum summary sentence from an original sentence and a desired summarization rate by a computer from the summarization rate,
The automatic summarization program causes a computer to morphologically analyze the input original sentence to divide a character string included in the original sentence into words and generate morphological information to which the part of speech is added, and to calculate a summarization rate of the input original sentence and A minimum value of the number of characters to be reduced is calculated from the number of characters of the input original text, and based on the morphological information, the word sequence of the original text is sequentially replaced with the replacement knowledge obtained in advance from the leading word sequence. When the matching is successful, the word string to be replaced is added to the morphological information when the matching is successful, and a word lattice structure to which the replacement cost is given is obtained.The word lattice structure is directed from the end of the sentence to the beginning of the sentence. Search and calculate the maximum possible character reduction number (rearward character reduction number) from the end of the sentence of the word lattice structure to each node. By searching towards the end of the sentence, to calculate the number of characters reductions from beginning of a sentence for each path,
The sum of the calculation result and the backward character reduction number is obtained. If the sum is smaller than the desired character reduction number, the path is pruned, and if the sum is not smaller than the desired character reduction number. , The path is selected, the sum of the replacement costs is calculated for the selected path, and when the end of the sentence is reached, the path from the beginning to the end of the sentence is output. A recording medium on which an automatic summarization program is recorded, wherein a path that minimizes the sum of replacement costs from the beginning to the end of a sentence is obtained, and a character string of the obtained minimum path is output as a summary sentence. 8. The automatic summarization program according to claim 1, wherein when the maximum value of the number of character reductions from the last node of the sentence to the first node of the sentence is smaller than the minimum value of the desired number of characters to be reduced, 8. A recording medium on which the automatic summarization program according to claim 7 is output, indicating that summarization is impossible.