JP2021068218A - Headline generation device and program - Google Patents

Abstract

To provide a head line generation device capable of generating a headline divided into a plurality of sentences as needed without losing an element of essential information by summation.SOLUTION: A head line generation device 1 includes: a division unit 21 for acquiring a modification tree by analyzing a modification relation of a clause about an input sentence to divide it into divided sentences corresponding to partial trees of the modification tree; a key phrase determination unit 22 for selecting a divided sentence including a predetermined key phrase from among the divided sentences; an element determination unit 23 for determining whether or not a sentence satisfies an element condition that is a condition concerning the element of the sentence; a sentence expansion processing unit 24 for generating an expansion sentence by adding an element of the sentence based on the divided sentence selected by the key phrase determination unit 22 to repeatedly add an element until the expansion sentence satisfies the element condition; and an output unit 30 for outputting information represented by the expansion sentence as a head line when the expansion sentence satisfies the element condition.SELECTED DRAWING: Figure 1

Description

The present invention relates to a headline generator and a program.

There is a service that delivers only the outline of the news content in text (called headlines, etc.). If the above headlines can be automatically generated based on sentences such as news and announcements, labor saving and cost reduction can be expected. Much research has already been done on techniques for extracting summaries from text.

For example, Patent Document 1, Patent Document 2, Patent Document 3, Non-Patent Document 1, and Non-Patent Document 2 describe techniques for summarizing sentences and the like.

Japanese Unexamined Patent Publication No. 2017-151863 Japanese Patent No. 4333318 Japanese Patent No. 6205466

Shun Hasegawa, Tsutomu Hirao, Manabu Okumura, Masaaki Nagata, "Headline Proceedings Using Sentence Compression", Proceedings of the 23rd Annual Meeting of the Natural Language Processing Society, March 2017, URL: https://www.anlp .jp / proceedings / annual_meeting / 2017 / pdf_dir / A1-3.pdf Shun Tanaka, Ryohei Sasano, Daiya Takamura, Manabu Okumura, "Summary length, sentence length, news article summary with limited number of sentences", Proceedings of the 22nd Annual Meeting of the Natural Language Processing Society, March 2016, URL: https //www.anlp.jp/proceedings/annual_meeting/2016/pdf_dir/D1-3.pdf

However, all of the prior arts listed above discard some of the essential information contained in the original text for summarization. However, there are areas where it is desirable to maintain essential information when summarizing, such as news of transportation interruptions and delays. As an example, when summarizing the news that the sections of railways A, B, C, and D from D1 to D2 are currently normal, these listed line names or It is worthwhile to maintain all the information such as the information that identifies the section (two station names). In other words, it is desirable to maintain information and maintain completeness, rather than throwing away information to make it easier to understand. Further, when performing such a type of summarization, it may be superior in terms of easy-to-understand information to output a plurality of headlines (summary texts) based on one input sentence. However, none of the listed prior art documents summarizes one sentence by dividing it into a plurality of sentences according to the meaning.

The present invention has been made based on the above-mentioned problem recognition, and it is possible to generate a headline divided into a plurality of sentences as needed without losing essential information elements by summarizing. It is intended to provide a headline generator and a program.

[1] In order to solve the above-mentioned problems, the headline generator according to one aspect of the present invention obtains a dependency tree by analyzing the dependency relationship of clauses with respect to the input sentence, and obtains the dependency tree to obtain the sentence. A division part that divides into a division sentence corresponding to a subtree of the dependency tree, a key phrase determination unit that selects a division sentence including a predetermined key phrase among the division sentences, and a condition that the sentence is related to a sentence element. Sentence extension processing that generates an extended sentence by adding sentence elements based on the element determination unit that determines whether or not the element condition is satisfied and the divided sentence selected by the key phrase determination unit. A unit and an output unit that outputs information represented by the extended sentence as a headline when the extended sentence satisfies the element condition are provided, and the sentence expansion processing unit includes a unit when adding the element. A clause corresponding to a subtree including the element in the dependency tree is added, and the sentence extension processing unit adds the element condition to the extended sentence when the extended sentence does not satisfy the element condition. If it is satisfied, the addition of the element is repeated until it is determined by the element determination unit.

[2] Further, one aspect of the present invention is a normalization in which, in the above-mentioned headline generator, when the extended sentence satisfies the element condition, the extended sentence is normalized so as to match a predetermined normal pattern. The output unit further includes a normalization unit, and the output unit outputs the expression normalized by the normalization unit as the headline.

[3] Further, in one aspect of the present invention, in the headline generator, the key phrase is a route name in a means of transportation, and the element condition is that the sentence includes the line name and the sentence includes the line name. The sentence includes information for specifying the section of the route name, and the sentence includes information indicating the state of the route name in the section.

[4] Further, in one aspect of the present invention, a dependency tree is obtained by analyzing the dependency relationship of clauses with respect to the input sentence, and the sentence is divided into a divided sentence corresponding to a subtree of the dependency tree. A division unit that divides into the above, a key phrase determination unit that selects a division sentence including a predetermined key phrase among the division sentences, and whether or not the sentence satisfies the element condition which is a condition related to the element of the sentence. An element determination unit for determining, a sentence extension processing unit that generates an extended sentence by adding a sentence element based on the divided sentence selected by the key phrase determination unit, and the extended sentence set the element condition. When satisfied, the sentence extension processing unit includes an output unit that outputs the information represented by the extended sentence as a headline, and the sentence expansion processing unit includes a subtree including the element in the dependency tree when the element is added. The sentence expansion processing unit adds a clause corresponding to, and when the extended sentence does not satisfy the element condition, until the element determination unit determines that the extended sentence satisfies the element condition. This is a program for operating a computer as a headline generator that repeats the addition of the above elements.

According to the present invention, the headline generator can automatically generate a headline based on the input sentence. Moreover, the headline generator can generate a set of headlines having necessary and sufficient elements according to the elements of the input sentence (in many cases, a set having a plurality of headlines as elements).

It is a block diagram which shows the schematic functional structure of the headline generator by embodiment of this invention. It is a schematic diagram which shows an example of the result which the division part by the same embodiment performed the parsing processing about the input sentence. It is a schematic diagram which shows the example which the division part by the same embodiment divided an input sentence based on the syntactic analysis result. It is the schematic which shows the example of the element condition for use in the element determination part by the same embodiment. It is the schematic which shows the example of the procedure which the sentence extension processing part by the same embodiment expands a key sentence. It is the schematic which shows the example of the process which the headline generator by the same embodiment determines the remaining element. It is a flowchart which shows the procedure of the whole processing of the headline generation apparatus 1 by the same embodiment. It is a schematic diagram which shows the example (in the case of other than railway information) of the text data (input sentence) acquired by the text acquisition unit 10 by the same embodiment. It is a schematic diagram which shows the example (when the case other than railway information) that the division part by the same embodiment divided an input sentence based on the syntactic analysis result. It is a schematic diagram which shows the example (in the case of other than railway information) of the procedure which the sentence extension processing part by the same embodiment expands a key sentence.

An embodiment of the present invention will be described below with reference to the drawings. The headline generator according to the present embodiment parses the input sentence (sentence to be headlined) by a syntactic analyzer and divides it into a plurality of semantic unit sentences (also referred to as "divided sentences"). The headline generator selects a key sentence including a key phrase from the divided sentences. The headline generator combines syntactically close or adjacent clauses into the above key sentences until the required elements of the output are satisfied (the element condition can determine whether or not they are satisfied). To do. A sentence extended by combining clauses is called an extended sentence. When the extension sentence meets the element condition, the headline device generates a headline by fitting (that is, normalizing) the element to the template. If the output element remains in the remaining sentence after removing the element output as the headline, the used element is masked so that the remaining element is output, and the headline is generated again. In this way, the headline generator generates a plurality of headlines.

FIG. 1 is a block diagram showing a schematic functional configuration of the headline generator 1 according to the present embodiment. As shown in the figure, the headline generation device 1 includes a text acquisition unit 10, a processing unit 20, an output unit 30, a control unit 40, and a data storage unit 50. Further, the processing unit 20 includes a division unit 21, a key phrase determination unit 22, an element determination unit 23, a sentence extension processing unit 24, and a normalization unit 25. Each of these functional units can be realized by, for example, a computer and a program. In addition, each functional unit has a storage means, if necessary. The storage means is, for example, a variable on the program or a memory allocated by executing the program. Further, if necessary, a non-volatile storage means such as a magnetic hard disk device or a solid state drive (SSD) may be used. Further, at least a part of the functions of each functional unit may be realized not as a program but as a dedicated electronic circuit.

The text acquisition unit 10 acquires text data input from the outside. This text data may be passed from another external device, for example, or may be input by a person using an input means such as a keyboard. This text data includes sentences such as news. This text data is a target to be processed by the processing unit 20.

The output unit 30 outputs the headline generated by the processing unit 20 to the outside. The headline is text data including information on an extended sentence output by the sentence expansion processing unit 24, which will be described later. The extended sentence at this time is determined by the element determination unit 23, which will be described later, to satisfy the element condition. The headline may be the extended sentence output by the sentence expansion processing unit 24, or may be a headline that has been normalized by the normalization unit 25 described later. The output unit 30 can also distribute the generated headline via, for example, a communication network.

The control unit 40 controls the execution of processing by the entire headline generation device 1.

The data storage unit 50 stores data for reference by the processing unit 20. The content of the data stored in the data storage unit 50 depends on the field of the sentence to be processed by the headline generator 1. In the present embodiment, the data storage unit 50 has a route name data storage unit and a station name data storage unit (both not shown) inside. The line name data storage unit stores data of a set of railway line names. In addition, the station name data storage unit stores data of a set of railway station names. In addition, the station name data storage unit stores the relationship between the line name and the station name belonging to the line. In other words, the station name data storage unit holds information indicating whether or not a station with a certain station name belongs to a line with a certain line name.

The processing unit 20 generates headline data by processing the text data acquired by the text acquisition unit 10. More detailed functions of each unit constituting the processing unit 20 will be described below.

The division unit 21 performs a syntactic analysis process (dependency analysis process) of the text data acquired by the text acquisition unit 10, and divides a sentence by using the syntactic analysis result. In addition, the divided sentence may always include a phrase corresponding to the root of the dependent tree. Further, the divided sentence may correspond to, for example, one branch that is sequentially engaged from each leaf included in the dependent tree toward the above-mentioned root. That is, the division unit 21 can output a plurality of sentences (referred to as divided sentences) obtained by dividing the input sentence. The division unit 21 may use an existing technique for parsing. As an example, the parser 21 can use the parser CaboCha.

The key phrase determination unit 22 determines whether or not each of the divided sentences output by the division unit 21 includes a key phrase. The key phrase determination unit 22 determines that the divided sentence including the key phrase is a key sentence. Further, the key phrase determination unit 22 determines that the divided sentence that does not include the key phrase is a non-key sentence. The set of key phrases may be stored in the data storage unit 50 in advance, for example. The key phrase determination unit 22 can determine the key phrase by referring to the data storage unit 50. In this embodiment, as an example, a railway line name is used as a key phrase. That is, in the present embodiment, the data storage unit 50 has a set of railway line names in advance. Key phrases are not limited to railway line names. What is used as a key phrase may be determined in advance and stored in a storage means as needed.

The element determination unit 23 determines whether or not the sentence satisfies the element condition which is a condition related to the element of the sentence. The determination by the element determination unit 23 is for determining whether or not a sentence that is a candidate for a headline (a divided sentence or an extended sentence extended based on the divided sentence) has a sufficient element as a headline. Is. The element conditions will be further described later.

The sentence expansion processing unit 24 generates an extended sentence by adding sentence elements based on the divided sentence (key sentence) selected by the key phrase determination unit 22. When adding the element to the sentence, the sentence expansion processing unit 24 adds a clause corresponding to the subtree including the element in the dependency tree. When the generated extended sentence does not satisfy the element condition, the sentence expansion processing unit 24 sequentially repeats the addition of elements until the element determination unit 23 determines that the extended sentence satisfies the element condition. You can.

When the sentence extended by the sentence extension processing unit 24 meets the element conditions, the normalization unit 25 normalizes the sentence so as to match a predetermined normal pattern. The normal pattern may be defined in advance and stored in a storage means, for example. By normalizing, it is possible to give a sense of unity to the style and tone of the headline. A specific example of the normalization process will be described later.

FIG. 2 is a schematic diagram showing an example of the result of parsing the input sentence (text data) by the dividing unit 21. In the illustrated example, the input sentence is "For the Tokaido Sanyo Shinkansen, the up line is between Tokuyama and Kyoto in Yamaguchi Prefecture, and the down line is between Kyoto and Shin-Kobe." In addition, the figure shows the result of parsing the above input sentence in the form of a tree genealogy (the above-mentioned dependency tree). The result of this parsing shows the dependency relationship of the clauses that make up the sentence. That is, the sections related to the section "I'm not waiting" are "Tokaido Sanyo Shinkansen is", "between", "down line", "between Kyoto and Kobe", and "driving". Each of them. In addition, the sections related to the above section "between" are "up line ga", "tokuyama in Yamaguchi prefecture", and "in Kyoto" respectively. In the figure, for convenience, each section is provided with reference symbols A to I. The root in the tree structure is the node A in this figure. A leaf is a terminal node on the opposite side of the root. In this figure, the B, G, H, I, D, E, and F nodes are leaves.

FIG. 3 is a schematic diagram showing an example in which the division unit 21 divides the input sentence based on the above syntactic analysis result. Here, the sentence is divided into seven sentences based on the result of parsing. Each of these sentences follows the syntax tree shown in FIG. 2 from the leaf side (lower side) to the root side (upper side), that is, from the relevant node side to the associated node side. Is generated by. The first split sentence is "The Tokaido Sanyo Shinkansen is forgotten", which was obtained by tracing the section in the syntax tree shown in Fig. 2 as BA. The second split sentence is "the up lines are in between and are forgotten", and is obtained by tracing the clause in the syntax tree shown in FIG. 2 as GCA. The third split sentence is "I have forgotten between Tokuyama and Tokuyama in Yamaguchi Prefecture", which was obtained by tracing the section in the syntax tree shown in Fig. 2 as HCA. The fourth split sentence is "We are forgotten between Kyoto", which was obtained by tracing the section in the syntax tree shown in Fig. 2 as ICA. The fifth split sentence is "the down line is forgotten", which is obtained by tracing the clause in the syntax tree shown in FIG. 2 as DA. The sixth split sentence is "Forgotten between Kyoto and Shin-Kobe", which was obtained by tracing the section in the syntax tree shown in Fig. 2 as EA. The seventh split sentence is "Driving is suspended", which is obtained by tracing the section in the syntax tree shown in FIG. 2 as FA.

Here, the split sentence obtained by the split portion 21 is a split sentence including the root in the dependent tree and corresponding to the minimum subtree. A minimum subtree is a subtree that does not have branch branches at a node in the middle from the root to the leaf. It should be noted that the division unit 21 does not necessarily have to generate only the division sentence corresponding to the minimum subtree.

Of the split sentences shown in FIG. 3, the first split sentence is a sentence selected as a key sentence. Other split statements are non-key sentences. A key sentence is a sentence that can be a key for the headline generator 1 to generate a headline. The key phrase determination unit 22 determines whether or not the division sentence is a key sentence depending on whether or not each division sentence output by the division unit 21 includes a key phrase. In this embodiment, the key phrase is a railway line name. That is, since the first divided sentence includes the route name "Tokaido Sanyo Shinkansen" among the seven divided sentences shown in FIG. 3, the key phrase determination unit 22 determines this divided sentence as a key sentence. Since each of the other divided sentences from the second to the seventh does not include the route name, the key phrase determination unit 22 determines these divided sentences as non-key sentences. What is used as the key phrase shall be appropriately determined for each application field of the device. The key phrase determination unit 22 determines whether or not the expression (word or the like) in the sentence is a route name by referring to the route name data storage unit described above.

The condition for the key phrase determination unit 22 to determine may be defined as one of the element conditions (key condition) described below.

FIG. 4 is a schematic view showing an example of element conditions for the element determination unit 23 to use for determination. As shown in the figure, in this embodiment, four conditions from No. 1 to No. 4 are used. Of these, the first condition is a key condition. The key condition is a condition used by the key phrase determination unit 22 to determine whether or not the divided sentence is a key sentence. The first condition is that there is <Line> (there is a line name). That is, the key phrase determination unit 22 determines that the divided sentence having the route name is a key sentence as described above. The conditions from No. 2 to No. 4 are that there is <Status>, there is <Area>, and if <Area> is not "all lines", there are two or more <Stations> on the <Line> line. That is.

Here, <Status> is a state. The expression corresponding to <Status> is, for example, an expression such as "restart", "suspension", or "normal". Further, <Area> represents a designated line. The expression corresponding to <Area> is, for example, an expression such as "all lines", "upper and lower lines", "up line", and "down line". In addition, <Station> is the station name.

The element condition is a condition for determining whether or not the headline (or a candidate for the headline) has a sufficient element as a headline. The element condition is stored in a dedicated storage area, for example, and is referred to by the key phrase determination unit 22 and the element determination unit 23.

FIG. 5 is a schematic diagram showing an example of a process in which the sentence expansion processing unit 24 expands the key sentence. Along with this figure, the process in which the sentence expansion processing unit expands the sentence will be described.

In the first stage, the sentence is "The Tokaido Sanyo Shinkansen is forgotten." This sentence is a key sentence selected by the key phrase determination unit 22. The element determination unit 23 determines the element condition of this sentence. The element determination unit 23 determines that <Area> is insufficient in this sentence based on the element conditions. The sentence expansion processing unit 24 expands the sentence according to the determination result. Specifically, the sentence extension processing unit 24 adds “up line” and “between”, which are branches of GC of the syntax analysis result (FIG. 2).
In the second stage, the sentence is, "The Tokaido Sanyo Shinkansen is not on the up line." The element determination unit 23 determines the element condition of this sentence. The element determination unit 23 determines that <Station> is insufficient in this sentence based on the element condition. The sentence expansion processing unit 24 expands the sentence according to the determination result. Specifically, the sentence extension processing unit 24 adds H's "Tokuyama in Yamaguchi Prefecture" related to C's "between" in the syntax analysis result (FIG. 2).
In the third stage, the sentence is, "The Tokaido Sanyo Shinkansen has an up line with Tokuyama in Yamaguchi Prefecture." The element determination unit 23 determines the element condition of this sentence. When <Area> is not "all lines", the element determination unit 23 has insufficient (insufficient) number of <Stations> based on the element condition that there are two or more <Stations> on the <Line> line. ). The sentence expansion processing unit 24 expands the sentence according to the determination result. Specifically, the sentence extension processing unit 24 adds "Kyoto" of I related to "between" of C in the syntactic analysis result (FIG. 2).
In the fourth stage, the sentence is, "The Tokaido Sanyo Shinkansen has an up line between Tokuyama and Kyoto in Yamaguchi Prefecture." The element determination unit 23 determines the element condition of this sentence. The element determination unit 23 determines that all the element conditions are satisfied. Of the station names (characters that match the station names registered in the dictionary) that appear in this fourth-stage sentence, Tokuyama and Kyoto are included in the set of stations on the Tokaido / Sanyo Shinkansen. In addition, Yamaguchi is not included in the set of stations on the Tokaido / Sanyo Shinkansen.

Since the sentence extended to the fourth stage satisfies the element condition, the normalization unit 25 normalizes this sentence. The normalized pattern is predefined and stored. According to the pattern, the normalization unit 25 normalizes the sentence of the fourth stage above and converts it into "railway information, Tokaido / Sanyo Shinkansen Tokuyama-Kyoto up line operation suspension". In the example of normalization here, "railroad information" is always added to the beginning of the headline. Then, the route name follows. The route name was "Tokaido Sanyo Shinkansen" in the original sentence, but it has been converted to the expression "Tokaido / Sanyo Shinkansen" (with a midpoint) by normalization. The normalization of such names (route names) is based on a dictionary stored in advance. Then, when <Area> is not "all lines", the target section is then represented by two station names. These two station names are expressed by connecting them with "~" in the order according to the direction of the train, depending on whether <Area> is an "up line" or a "down line". Here, it is expressed as "Tokuyama-Kyoto" using the station names of Tokuyama and Kyoto. Then, further, the information of <Area> and <Status> follows. In addition, <Status> is expressed by using an uninflected word as appropriate. Here, it is converted into the expression "suspend driving on the up line".

The expression normalized in this way is the headline at which the headline generator 1 is stationary. The output unit 30 outputs this normalized headline to the outside.

FIG. 6 is a schematic view showing an example of a process in which the headline generator 1 determines the remaining elements. The control unit 40 of the headline generator 1 determines the presence or absence of the remaining elements by the procedure shown in FIG. Further, when there is a remaining element, the control unit 40 makes a clear copy of the headline generation device 1 so as to further perform sentence expansion processing using the remaining element. An example of a specific processing process is as described below.

In step S51, the control unit 40 compares the elements included in the original input text with the elements included in the already output headline. Here, the original input sentence was "The Tokaido Sanyo Shinkansen is not operating between Tokuyama and Kyoto in Yamaguchi Prefecture on the up line and between Kyoto and Shin-Kobe on the down line." In addition, as explained above, the headline, which is the first output (already output), was "Railway information, Tokaido / Sanyo Shinkansen Tokuyama-Kyoto up line operation suspension".

In step S52, the control unit 40 removes all the elements already output from the original sentence. If the element that has already been output is only the first output described above, when the already output element is removed, "<Line> is a down line between <Station> and <Station> in Yamaguchi Prefecture where <Area> is. Is driving between Kyoto and Shin-Kobe. " In other words, here, the element "Tokaido Sanyo Shinkansen" that has already been output has been removed and replaced with the tag <Line>. In addition, the elements "Kyoto" and "Tokuyama" have been removed and replaced with the tags <Station>, respectively. In addition, the element "forgotten" has been removed and replaced with the tag <Status>. In other words, the element output as the headline is masked with tags such as <Line>, <Area>, <Station>, and <Status>.

In step S53, the element determination unit 23 determines whether or not the elements are satisfied with respect to the result of the removal process in step S52 based on the element conditions. In this example, the determination result that <Line> and <Status> are insufficient can be obtained. In the result of the removal process in step S52, "Kyoto" and "Shin-Kobe" remain, so the condition that two <Stations> are required is satisfied.

In step S54, the control unit 40 returns the missing element to the sentence based on the determination result in step S53. That is, here, the control unit 40 returns the expression "Tokaido Sanyo Shinkansen" which is a <Line> element and the expression "forgotten" which is a <Status> element to the result of step S52. The resulting sentence is, "On the Tokaido Sanyo Shinkansen, <Area> is between <Station> and <Station> in Yamaguchi Prefecture, and the down line is between Kyoto and Shin-Kobe."

In step S55, the control unit 40 controls the headline generator 1 so as to execute the sentence division process and the sentence extension process again based on the sentence obtained in step S54.

As a result of the second execution, in this example, the headline generator 1 generates a second output (headline) of "railway information, operation suspension on the Tokaido / Sanyo Shinkansen Kyoto-Shin-Kobe down line". .. As described above, the headline generator 1 can also generate a plurality of headlines while excluding the elements that have already been output.

FIG. 7 is a flowchart showing the overall processing procedure of the headline generator 1. Hereinafter, the processing procedure of the headline generator 1 will be described with reference to this flowchart.

First, in step S1, the division unit 21 performs a syntactic analysis process of the input sentence. Then, the division unit 21 divides the sentence based on the syntactic analysis result, and obtains n sentences S1, S2, ..., Sn.

The following processing within the range of steps S2 to S6 is repeatedly executed for each divided sentence Si (1 ≦ i ≦ n).

In step S2, the key phrase determination unit 22 determines whether or not the sentence Si has a key phrase. The key phrase is, for example, the name of a railway line. However, other key phrases may be used depending on the field of the content of the sentence. If the sentence has a key phrase (step S2: YES), the process proceeds to the next step S3. If the sentence does not have a key phrase (step S2: NO), the process for the sentence Si is finished, and the process proceeds to the next sentence (if any).

Next, in step S3, the element determination unit 23 refers to the condition and determines whether or not the sentence Si satisfies the element. If the elements are satisfied (step S3: YES), the process proceeds to the next step S4. If the elements are not satisfied (step S3: NO), the process proceeds to step S5.

When the process proceeds to step S4, the normalization unit 25 normalizes the expression of the sentence (Si). Then, the output unit 30 outputs the normalized sentence to the outside. When the processing of this step is completed, the processing for the sentence Si is completed, and the process proceeds to the processing of the next sentence (if any).

When the process proceeds to step S5, the sentence extension processing unit 24 attempts a process of extending the sentence in order to satisfy the missing elements in the sentence at that time. After this step, the process proceeds to step S6. In step S6, it is determined whether or not the sentence extension processing in step S5 by the sentence expansion processing unit 24 is successful. If the sentence expansion is successful (step S6: YES), the process returns to step S3 to determine whether or not the elements are satisfied. If the sentence expansion fails (step S6: NO), the process for the sentence Si is finished, and the process proceeds to the process for the next sentence (if any). Note that the sentence extension fails, for example, when the element condition is still not satisfied even if the extension using all the remaining elements is performed.

When the processing from sentences S1 to Sn (steps S2 to S6) is completed, the process proceeds to step S7. In step S7, under the control of the control unit 40, the headline generator 1 determines the presence or absence of the remaining elements. The process of determining the presence or absence of the remaining elements is as shown in steps S51 to S54 in FIG. If there are remaining elements (step S7: YES), the process returns to step S1 for the process of generating the headline by the remaining elements. If there are no remaining elements (step S7: NO), the processing of the entire flowchart is terminated.

[Example of application to other than railway information]
Above, a processing example when the headline generator 1 is applied to railway information has been described. In the following, as an application example other than railway information, an example in which the headline generator 1 processes highway information will be described.

FIG. 8 is a schematic diagram showing an example of text data (input sentence) acquired by the text acquisition unit 10. Here, the input sentence is "Because the Tomei Expressway is snowy, the up line between the Numazu interchange in Shizuoka prefecture and the Oi-Matsuda interchange is closed from 3:10 am." As in the case of railway information, the division unit 21 performs a syntax analysis process (dependency analysis process) of this input sentence to generate a syntax tree. Here, the illustration of the syntax tree is omitted.

FIG. 9 is a schematic diagram showing six sentences (divided sentences) generated by the dividing unit 21 based on the syntactic analysis processing result of the input sentence shown in FIG. According to the parsing result of the input sentence, all the clauses that make up this sentence are directly or indirectly related to the clause "is". The division section 21 uses the original input sentence as the first sentence "Tomei Expressway is now", the second sentence "Because of snow", and the third sentence "Numazu, Shizuoka Prefecture". "It consists of an interchange", the fourth sentence "The up line between the Oi-Matsuda interchange is", the fifth sentence "It consists of 3:10 am", and the sixth sentence " It is closed to traffic. " Of the six split sentences shown here, only the first sentence "Tomei Expressway is" is a key sentence, and the other sentences are non-key sentences. In this example, the expressway name is a key phrase, and of the first to sixth divided sentences, only the first sentence includes this key phrase.

FIG. 10 is a schematic diagram showing a process of starting from a key sentence and sequentially performing sentence expansion processing. In this example, the tag <Line> represents an expressway name. This highway name is a key phrase, as mentioned above. The tag <Station> represents any one of an interchange name, a junction name, and a tollhouse name. The interchange may be described as "IC", and the junction may be described as "JCT". Further, the tag <Area> represents any one of "all lines", "upper and lower lines", "up line", and "down line". Further, the tag <Status> represents either "closed road" or "released closed road".

As shown in FIG. 10, the sentence expansion processing unit 24 sequentially expands the sentence. Specifically, it is as follows. In the first stage, the sentence is "Tomei Expressway is now" (selected key phrase). At this time, the element determination unit 23 determines that <Area> and <Status> are insufficient. In the second stage, the sentence is "The Tomei Expressway is made of snow." Also at this time, the element determination unit 23 determines that <Area> and <Status> are insufficient. In the third stage, the sentence is "Because the Tomei Expressway is snowy, it consists of the Numazu interchange in Shizuoka prefecture." Also at this time, the element determination unit 23 determines that <Area> and <Status> are insufficient. In the fourth stage, the sentence is, "Because the Tomei Expressway is snowy, there is an up line between the Numazu interchange in Shizuoka prefecture and the Oi-Matsuda interchange." At this time, the element determination unit 23 determines that <Area> and <Status> are insufficient. In the fifth stage, the sentence is, "Because the Tomei Expressway is snowy, the up line between the Numazu interchange in Shizuoka prefecture and the Oi-Matsuda interchange starts at 3:10 am." Also at this time, the element determination unit 23 determines that <Area> and <Status> are insufficient. In the sixth stage, the sentence is, "Because of the snow on the Tomei Expressway, the up line between the Numazu interchange in Shizuoka prefecture and the Oi-Matsuda interchange is closed from 3:10 am." At this time, the element determination unit 23 determines that all the necessary elements are satisfied.

The normalization unit 25 normalizes the above sentence in the sixth stage. The normalized result is "Tomei Expressway Oi Matsuda IC-Numazu IC up line closed". In this normalization process, for example, the word "interchange" is converted to the expression "IC". The output unit 30 outputs this normalized expression.

The present embodiment is summarized as follows.

(1) The headline generation device 1 includes at least a division unit 21, a key phrase determination unit 22, an element determination unit 23, and a sentence expansion processing unit 24. The division unit 21 obtains a dependency tree by analyzing the dependency relationship of the clauses on the input sentence, and divides the sentence into a division sentence corresponding to a subtree of the dependency tree. The dependency tree represents the dependency relationship between the nodes of a sentence as a tree structure. As an example, the above-mentioned divided sentence has a common root in the dependency tree with the original input sentence. Further, as an example, the division sentence obtained by dividing the division portion 21 is a subtree containing the above-mentioned roots in the dependent tree and being a minimum subtree. A minimum subtree is a subtree that does not have branch branches at a node in the middle from the root to the leaf. The key phrase determination unit 22 selects a divided sentence (that is, a key sentence) including a predetermined key phrase from the divided sentences. Here, the key phrase is a phrase that can be a key in the field to which the input sentence handled by the headline generator 1 belongs. For example, for railways and highways, the key phrase is the line name (the highway name is also a type of line name). The element determination unit 23 determines whether or not the sentence satisfies the element condition which is a condition related to the element of the sentence. Element conditions are stored in advance, for example. The element determination unit 23 may specify which element is missing in the sentence to be determined. The sentence expansion processing unit 24 generates an extended sentence by adding sentence elements based on the divided sentence (that is, the key sentence) selected by the key phrase determination unit 22. That is, the sentence extension processing unit 24 adds an element that is insufficient in light of the element condition. The output unit 30 outputs the information represented by the extended sentence as a headline when the extended sentence created by the sentence expansion processing unit 24 satisfies the element condition. When the element is added, the sentence expansion processing unit 24 adds a clause corresponding to the subtree including the element in the dependency tree. Further, the sentence expansion processing unit 24 repeats the addition of the element until the element determination unit determines that the extended sentence satisfies the element condition when the extended sentence does not satisfy the element condition. ..

In the headline generator 1 described above, the normalization unit 25 may be omitted. In that case, the headline generator 1 does not perform the normalization process and outputs an extended statement that satisfies the element conditions.

(2) The processing unit 20 of the headline generator 1 may be configured to include a normalization unit 25. When the extended sentence satisfies the element condition, the normalization unit 25 performs a process of normalizing the extended sentence so as to match a predetermined normal pattern. In this case, the output unit 30 outputs the expression normalized by the normalization unit 25 as a headline.

(3) As an example, the key phrase may be a route name in a means of transportation. Here, the means of transportation includes railways, aviation, road networks (for example, highways), ships, and the like. In this case, an example of the element condition is that a) the sentence includes the line name, b) the sentence contains information for specifying a section in the line name, and c) the sentence includes the section in the line name. It contains information that represents the state. Here, the information for specifying the section is, for example, the expressions such as "whole line", "upper and lower lines", "up line", and "down line", and specifying between those bases by giving the names of two bases. Information. The base includes, for example, a railway station, an interchange on a road, a junction, a tollhouse, an airport, a port, and the like. Further, here, the information indicating the state includes expressions such as "normal operation", "operation (operation) suspension", "traffic closure", and "traffic closure cancellation".

As described above, according to the present embodiment, the headline can be automatically generated based on the input sentence without relying on human hands. Further, in the present embodiment, the number of headlines generated may be a plurality. Then, in the present embodiment, the set of generated headlines can include all the elements of the original sentence without losing the elements of the information. That is, the generated headline information is exhaustive.

It should be noted that at least a part of the functions of the headline generator in the above-described embodiment can be realized by a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by a computer system and executed. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. The "computer-readable recording medium" is a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, a DVD-ROM, or a USB memory, or a storage device such as a hard disk built in a computer system. Say that. Furthermore, a "computer-readable recording medium" is a device that temporarily and dynamically holds a program, such as a communication line when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. , In that case, it may include a program that holds a program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or a client. Further, the above-mentioned program may be a program for realizing a part of the above-mentioned functions, and may be a program for realizing the above-mentioned functions in combination with a program already recorded in the computer system.

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs and the like within a range that does not deviate from the gist of the present invention.

The present invention can be used, for example, in businesses such as broadcasting and information distribution. However, the scope of use of the present invention is not limited to those exemplified here.

1 Headline generator 10 Text acquisition unit 20 Processing unit 21 Dividing unit 22 Key phrase judgment unit 23 Element judgment unit 24 Sentence expansion processing unit 25 Normalization unit 30 Output unit 40 Control unit 50 Data storage unit

Claims (4)

A division part that obtains a dependency tree by analyzing the dependency relation of clauses on the input sentence and divides the sentence into a division sentence corresponding to a subtree of the dependency tree.
A key phrase determination unit that selects a split sentence containing a predetermined key phrase from the split sentences, and a key phrase determination unit.
An element judgment unit that determines whether or not the sentence satisfies the element condition, which is a condition related to the element of the sentence.
A sentence expansion processing unit that generates an extended sentence by adding sentence elements based on the divided sentence selected by the key phrase determination unit, and a sentence expansion processing unit.
An output unit that outputs the information represented by the extended statement as a headline when the extended statement satisfies the element condition.
Equipped with
When the element is added, the sentence expansion processing unit adds a clause corresponding to the subtree containing the element in the dependency tree.
When the extended sentence does not satisfy the element condition, the sentence expansion processing unit repeats the addition of the element until it is determined by the element determination unit that the extended sentence satisfies the element condition.
Headline generator. A normalization unit that normalizes the extended sentence so as to match a predetermined normal pattern when the extended sentence satisfies the element condition.
With more
The output unit outputs the expression normalized by the normalization unit as the headline.
The headline generator according to claim 1. The key phrase is a route name in the means of transportation.
The element condition includes a sentence including the line name, the sentence including information specifying a section of the line name, and the sentence including information representing a state of the line name in the section.
The headline generator according to claim 1 or 2. A division part that obtains a dependency tree by analyzing the dependency relation of clauses on the input sentence and divides the sentence into a division sentence corresponding to a subtree of the dependency tree.
A key phrase determination unit that selects a split sentence containing a predetermined key phrase from the split sentences, and
An element judgment unit that determines whether or not the sentence satisfies the element condition, which is a condition related to the element of the sentence.
A sentence expansion processing unit that generates an extended sentence by adding sentence elements based on the divided sentence selected by the key phrase determination unit, and a sentence expansion processing unit.
An output unit that outputs the information represented by the extended statement as a headline when the extended statement satisfies the element condition.
Equipped with
When the element is added, the sentence expansion processing unit adds a clause corresponding to the subtree containing the element in the dependency tree.
When the extended sentence does not satisfy the element condition, the sentence expansion processing unit repeats the addition of the element until it is determined by the element determination unit that the extended sentence satisfies the element condition.
A program that allows a computer to function as a headline generator.

Images