JP3329353B2 - Topic Word Selection Method and Topic Structure Recognition Device in Text Topic Structure Recognition - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for topic structure recognition in natural language analysis, and more particularly to a topic word selection method for selecting a topic word and a topic structure recognition apparatus to which the topic word selection method is applied. .

[0002]

2. Description of the Related Art When a human being is presented with text or conversation data and given the task of "seeking the same thing in a block in which the same is written from these text or conversation data", there is no individual difference. The ability to answer the same structure has been experimentally confirmed. For the experiment, for example, "Takeshita et al .:" Research on Human Communication from the Viewpoint of Topic Structure Recognition "IEICE 1
99-93 Autumn Meeting D-62 (p.6-64) ”. Such a structure grasped by a human is called a “topic structure”. Since the topic structure forms a nested structure, each topic consists of a "topic word" indicating the topic, a "topic level" indicating the depth of the nesting, and a sentence of the sentence in the text or dialog data. Can be expressed by the "topic scope" of whether to continue. In the following, text and conversation data for which the topic structure is analyzed will be referred to as language data.

FIG. 1 shows an example of a topic structure for language data having contents related to a telecommunications policy. The language data starts from the 0th sentence and continues to at least the 770th sentence. The topic level of the topic having the topic word “communication service” is 1, and the topic scope is in the range from the 0th sentence to the 770th sentence. For the sake of simplicity, in the following, the topic will be referred to using a topic word, such as "topic of" communication service "".

[0004] Among the topics of "communication services", there are topics of "new service" having a topic level of 2 and "conventional service". The topics of "new service" are from the 125th sentence to the 431rd sentence. Has a topic scope up to the sentence,
The topic of "traditional services" is from the 432rd sentence to the 77th sentence
It has a topic scope of up to 0 sentences. Also, among the topics of "new service", a child topic of "service A"
"Service B" in the topic of "traditional services"
Child topic exists, and each topic scope is 3rd
It is from the 01st sentence to the 431st sentence, the 521st sentence and the 770th sentence.

Recognition of such a topic structure by a computer is called topic structure recognition. Several methods have been proposed for recognizing topic structures. Here, a brief description of the topic structure recognition method described in "Takeshita:" Video Retrieval System Using Topic Structure Recognition ", Information Processing Society of Japan 94-IM-15-1". FIG. 2 is a block diagram showing a configuration of an example of a topic structure recognition device used in this recognition method. FIG. 3 is a flowchart showing a topic structure recognition process in this recognition method. It is a figure showing an example of the flow of processing after topic structure recognition pre-processing.

The conventional topic structure recognition apparatus shown in FIG. 2 includes a data input unit 701 for inputting language data, a processing unit 702 for executing various processes, a display unit 703 for displaying results, a processing result and The storage unit 704 stores data required during processing, and the dictionary / rule unit 705 stores dictionaries and rules used in the topic structure recognition processing. The storage unit 704 includes a language data storage unit 710 that stores preprocessed language data, and a topic structure storage unit 71 that holds intermediate processing results and final processing results.
1 is provided. Further, the topic structure storage unit 711 includes a base development storage unit 712, a semantic development storage unit 713, and an integrated topic storage unit 714. On the other hand, a dictionary
The rule unit 705 is provided with a preprocessing dictionary 721, a semantic expansion processing rule 722, a base expansion processing rule 723, and an integration processing rule 724.

When a topic structure recognition process is performed using this topic structure recognition apparatus, first, as shown in FIG. 3, a topic structure recognition preprocessing 74 for input language data 730 is performed.
Perform 0. The first step of the topic structure recognition pre-processing 740 is a morphological analysis process 741 for the input language data 730. In the morphological analysis process 741, the character string of the input language data 730 is divided into words to form a word string, and the part of speech of each word, the inflected form of the inflected word, and the like are identified. Subsequently, as a second step of the preprocessing 740,
Using the result of the morphological analysis as an input, a single sentence separation process 742
Perform The single sentence delimiting process 742 is a process of dividing a sentence including a plurality of predicates, such as an embedded sentence or a multiple sentence, into a single sentence including only one predicate. As a third step of preprocessing 740, salient noun phrase extraction 743 is executed. The prominent noun phrase extraction 743 is a process of extracting the noun phrase that is most emphasized in each simple sentence by using the result of the simple sentence separation process 742 as an input. These processes belonging to the topic structure recognition pre-processing 740 are executed by the processing unit 702 using the pre-processing dictionary 721 in the dictionary / rule unit 705,
The result is stored in the language data storage unit 710 in the storage unit 704.
Is stored in

When the topic structure recognition pre-processing 740 is completed,
The topic development process is separated into a base development process 750 and a semantic development process 760 and executed. Here, infrastructure development
Topic development explicitly indicated by clue phrases such as "first" and "next", chapters, bullet points, etc. It is the development of a topic that is presented and progressed in a non-target form.

First, as shown in FIG. 3, in the base development process 750, a topic establishment section determination 751, a topic word determination 752, a topic scope and a topic level determination 753 are performed.
Are sequentially performed. Here, the topic establishment section is a section in which a topic is presented and established. In the topic word determination 752, prominent noun phrases in each topic establishment section are set as topic word candidates, and the topic word with the highest priority among these topic word candidates is selected. In the topic scope and topic level determination 753, processing is performed based on a structure such as an itemized list. The base development process 750 is performed by
The processing is performed by the processing unit 702 using the base expansion processing rule 723 in the rule unit 705, and the result is stored in the base expansion storage unit 712 included in the topic structure storage unit 711 in the storage unit 704.
Is stored in

FIG. 4 shows a specific example of the processing in the base development processing 750. First, the starting point of the language data (text) and each item of the list numbered in (1) and (2) are determined as the topic establishment section of the base development. Then, in the topic word determination 752, “communication service” from the first topic establishment section, “new service” from the second topic establishment section, and “conventional service” from the third topic establishment section. , Respectively, have been selected as topic words.

After the execution of the base development process 750, a semantic development process 760 is performed. The semantic development process 760 is similar to the base development process 750, and is used to determine the topic establishment section 76
1. It is composed of three processes, namely, topic word determination 762, topic scope and topic level determination 763. This semantic expansion processing 760 uses the semantic expansion processing rule 722 in the dictionary / rule unit 705 and the base expansion processing 7
The processing is also executed by the processing unit 702 using the result of 50, and the result is stored in the semantic development storage unit 713 included in the topic structure storage unit 711 in the storage unit 704.

In the example shown in FIG. 4, paragraphs longer than a certain length are selected as topic establishment sections, and "service A" and "service B" are selected as topic words in these sections. As the topic scope, the range from the start point of the above-described topic establishment section to the start point of the next topic establishment section in the base development is required. The topic levels are all set to the same level, that is, level 1 in the case of the semantic development of the text.

Finally, an integration process 770 of base development and semantic development is performed, and as a result, a topic structure 780 of the entire language data is output. This integration processing 770
The respective topic structures of the base development process 750 and the semantic development process 760 are input, and the processing is executed by the processing unit 702 using the integrated processing rules 724 in the dictionary / rule unit 705. In the example shown in FIG. 4, as a result of the integration processing, FIG.
The topic structure 780 similar to that shown in FIG.

In each of the basic development and the semantic development, the rules for determining the topic establishment section, topic word, topic scope, and topic level (semantic development processing rule 722 and base development processing rule 723) are based on language data. It depends on the form of communication, such as dialogue, monologue, or written text. For the differences between topic development styles and topic structure recognition rules depending on the transmission form, and the results of topic structure recognition experiments, see "Takeshita et al .:" Study of Human Communication from the Viewpoint of Topic Structure Recognition "IEICE 1993 Fall Meeting D -62 (p.6-64)].

[0015]

However, in the above-described conventional topic structure recognition method, it is difficult to correctly recognize a topic word for a written word text including a complicated adjunct modification, and a long topic scope is used. It is difficult to recognize topic words of topics having a long topic scope because complex adjunct modification often appears around the topic word of the topic having. For this reason, for example, when the recognized topic structure is used as a chaptering / table-of-contents structure for a human, there is a problem that it is difficult to grasp the entire outline because a large item of the chaptering is mistaken.

[0016] As texts including complicated association modification, for example, there are texts that appear in newspaper reports. FIG. 5 shows an example of such a text. In FIG. 5, sentence numbers such as [s0] and [s1] are assigned to each sentence. The subject of this text is clearly "Cambodia's general election". FIG. 6A shows an example of a topic structure recognized by a human for this text, and FIG. 6B shows a topic structure obtained by a conventional topic structure recognition method. If the topic structure by a human is treated as a table of contents, the items described in the original text shown in FIG. 5 can be estimated. On the other hand, in the topic structure based on the conventional topic structure recognition method, it is difficult to understand what is described in the first place because the biggest topic “UNTAC cooperation” is inappropriate. It is difficult to estimate.

An object of the present invention is to provide a method and an apparatus capable of correctly recognizing a topic word even for a text including a complex adjunct modification and providing an accurate topic structure.

[0018]

Topic word selection method of the present invention, in order to solve the problem] is the topic terms a method of determining the in recognizing the topic structure recognition apparatus topic structure of the language data using the previously prepared rules, the topic structure recognition device Excellent for topic word candidates
Topic word candidate priority rules for assigning priorities, and
A set of correction conditions and correction contents for correcting the priority
A word including the described topic word candidate priority correction table
Comprising a write-rule section, the priority assigned according to topic word candidate precedence rules for the language topic word candidates extracted from the data, the topic word candidate priority modification table
Based on a set of a correction condition and a correction content given in advance, when the correction condition is satisfied, the priority of the topic word candidate is corrected according to the corresponding correction content; The topic word is determined by comparing the priorities assigned to the candidates.

The topic structure recognition apparatus of the present invention were used: an input unit for inputting language data, and the dictionary-rule section for storing rules such for topic structure recognition, rules such the dictionary specification and regulations section A processing unit that performs processing, a storage unit that stores the result of the processing unit, and a display unit that displays the processing result of the processing unit, wherein the dictionary / rule unit assigns priority to the topic word candidates. A topic word candidate priority rule for assigning, and a topic word candidate priority correction table describing a set of a correction condition and a correction content for correcting the priority, wherein the storage unit receives an input from an input unit. A language data storage unit for storing information about the generated language data, and a topic structure storage unit for storing information about the topic structure, wherein the language data storage unit stores a character string and a part of speech of each word included in the language data. And the word information table that stores the broadcast,
A single sentence information table for storing information on words, salient noun phrases, and types of salient noun phrases included in each simple sentence of the language data, wherein the topic structure storage unit establishes a topic within a range in which topics are presented and established; It includes a table that stores information including sections, topic words, topic levels, and topic scopes.

[0020]

After the priority is assigned to each topic word candidate by obtaining the topic word candidates, the priority is corrected if a predetermined condition is satisfied, and the topic word is determined according to the priority. , So you can select topic words more accurately. More specifically, complex adjacency modification is performed by determining adnominal modification relations such as whether the topic word candidate is not adjoined to another simple sentence and correcting the priority of the topic word candidates according to the adjoint modification relation. It is possible to correctly recognize the topic word of the text including. In particular, it is possible to correctly recognize a topic word having a long topic section.

[0021]

Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 7 is a block diagram showing a configuration of a topic structure recognition device according to one embodiment of the present invention. This topic structure recognition apparatus is different from the conventional topic structure recognition apparatus shown in FIG. 2 particularly in the configuration of dictionaries and tables included in the dictionary / rule section 105. That is, in the topic structure recognition apparatus of the present embodiment, a data input unit 101 for inputting language data, a processing unit 102 for executing various processes, a display unit 103 for displaying results, a process result and a process It comprises a storage unit 104 for storing necessary data and a dictionary / rule unit 105 for storing dictionaries and rules used in the topic structure recognition processing. The storage unit 104 has a language data storage unit 110 that stores the language data after the preprocessing.
And a topic structure storage unit 111 for holding intermediate processing results and final processing results. The language data storage unit 110 includes a simple sentence information table 115 and a word information table 116, and the topic structure storage unit 111 includes a base development storage unit 112, a semantic development storage unit 113, and an integrated topic storage unit 114. Have been. On the other hand, a dictionary
The rule unit 105 includes a preprocessing dictionary 121, a semantic expansion processing rule 122, a base expansion processing rule 123, an integration processing rule 124, a topic word candidate priority rule 125, a question expression dictionary 126, and a topic word candidate priority correction table. 127 are provided. Topic word candidate priority correction table 127
Describes in advance a pair of an adjacency modification relationship indicating whether a topic word candidate has been modified by another simple sentence and a change to a topic word candidate priority.

When the topic structure of language data is analyzed using this topic structure recognition apparatus, the processing is performed in the same manner as the conventional processing flow shown in FIG. Different. In the case of using this topic structure recognition device, the priority as a topic word candidate is assigned to a prominent noun phrase with reference to the topic word candidate priority rule 125, and then the topic word candidate is assigned to another simple sentence. A topic word is determined by finding a modification relation such as whether the modification is performed and correcting the priority order according to the result. Hereinafter, a procedure of analyzing a topic structure by the topic structure recognition apparatus of the present embodiment will be described, centering on a topic word determination method.

[Language Data Storage Unit, Base Expansion Storage Unit, and Semantic Expansion Storage Unit] The topic structure analysis performed by the topic structure recognition apparatus of this embodiment is shown in FIG. 3 until the topic word is determined. The procedure proceeds as described above. And the result of topic structure recognition pre-processing, base expansion processing and semantic expansion processing is
Language data storage unit 110 and base development storage unit 1
12 and the semantic expansion storage unit 113. At the time when all of the base expansion processing and the determination of the topic establishment section of the semantic expansion processing are completed, each of these storage units 110,
The states of 112 and 113 are shown in FIG.

As described above, the linguistic data storage unit 110 stores the simple sentence information table 115 and the simple sentence information table 116.
It is included. The word information table 116 has a field for a word number indicating the order of appearance of the word in the text, a field for describing a character string of the word as a result of the morphological analysis, and information for describing information such as the part of speech of the character string. A field is provided. In the example of FIG. 8, the first word of the text to be recognized is “communication” of a sa-variable noun (a noun that can be combined with “suru” to become a sa-variable verb).

On the other hand, in the simple sentence information table 115, a field for describing the order of appearance of the simple sentence in the text as a simple sentence number, a word range field for describing the start and end word numbers of the simple sentence, , A field for describing the type of salient noun phrase, a field for describing the type of salient noun phrase, a field for describing the question expression contained in the simple sentence, When a noun phrase becomes a topic word candidate in base expansion or semantic expansion, a field for describing the priority order is included. In the example of FIG. 8, the first simple sentence ranges from word numbers 0 to 8, and the prominent noun phrase included in the simple sentence is word number 0, that is, “communication”, which is an explicit type prominent noun phrase. is there. When the salient noun phrase is composed of a plurality of words, a plurality of word numbers such as 0 and 1 are designated. The value of the field of the first single sentence question expression is "-1", which means that there is no question expression. If a question expression described in the question expression dictionary 126 included in the dictionary / rule unit 105 is detected in a single sentence, the question expression number is stored in the single sentence information table 1.
Described in 15 question expression fields. For example, if the question expression "question" is detected, the question expression number 0 is described in the question expression field of the record corresponding to the simple sentence. Also, the topic candidate priority of the first simple sentence is 2, and it will be described later in which topic establishment section this is a topic candidate.

The base development storage unit 112 stores, for each topic number, a field (topic establishment section field) for describing a single sentence number indicating the start and end of a topic establishment section in which a topic is presented and established, and a topic word. A field for describing, a field for describing a topic level, and a field for describing a topic scope by its start and end word numbers are included. In addition, the semantic development storage unit 113 includes the same fields as the base development storage unit 112. In the example of FIG. 8, in the base development storage unit 112, the topic establishment section of the first topic is in the range of simple sentence numbers 0 to 10, and the topic word is the word number 0, that is, “communication”. Here, when a topic word is composed of a plurality of words, a plurality of word numbers such as 0 and 1 are designated. The topic level of this topic is 1, and the topic scope is in the range of single sentence numbers 0 to 3029.

[Overall Flow of Topic Word Decision Processing] A salient noun phrase included in each topic establishment section in the basic development and the semantic development is set as a topic word candidate. FIG. 9 shows a flow of processing for selecting a topic word from topic word candidates in each topic establishment section.
Is shown in the flowchart of FIG. Although the rules used for selecting a topic word are actually different between the basic development and the semantic development, both of the processing flows are as shown in FIG. First,
Each topic word candidate is given a priority as a topic word candidate based on the topic word candidate priority rule 125 (step 201). Then, the union modification relation regarding the topic word candidate is detected (step 202), and the priority of each topic word candidate is corrected according to the topic word candidate priority correction table 127 (step 203). The word is determined, and the process ends.

In the case of determining a topic word in the basic development, a prominent noun phrase included in each topic establishment section in the basic development is set as a topic word candidate in the topic establishment section, and is determined according to the topic candidate priority for the basic development. Priorities are assigned, and the priority of each topic word candidate is corrected using the topic word priority correction table according to the union modification relation of the topic word candidates. From the topic word candidates in each topic establishment section, the highest priority is given. If you select a high candidate and only one candidate is selected, the candidate is the topic word. If there are multiple candidates,
If the topic establishment section is the short type whole type for establishing the topic of the whole itemized list, the candidate that appears latest in time is used. If it is the other chapter type, the candidate that appears first in time is used. Is selected as a topic word in the base development.

In the case of topic word determination in semantic development,
The prominent noun phrase included in each topic establishment section in the semantic expansion is set as a topic word candidate in each topic establishment section, and priority is given to each candidate according to the topic candidate priority for the semantic expansion. Similarly, the priority of each topic topic candidate is corrected using the topic word candidate priority correction table, the topic word candidate in each topic establishment section having the highest priority is selected, and one selected candidate is selected. When there is only one candidate, the candidate is set as a topic word, and when there are a plurality of selected candidates, the candidate that appears earliest in time is set as a topic word in a semantic expansion.

[Priority Assignment by Topic Candidate Priority Rule] Here, the assignment of priority by the topic word candidate priority rule 125 will be described. FIG. 10 shows topic word candidate priority rules used in base development, and FIG. 11 shows topic word candidate priority rules used in semantic development. A case where topic word candidate priority rules are applied to the example shown in FIG. 8 will be described. Semantic expansion storage unit 11
Since the topic establishment section of the first topic (topic number 0) recorded in No. 3 is in the range of simple sentence numbers 80 to 83, the salient noun phrases in the single sentence in that range on the simple sentence information table 115 are topic candidates. . According to the simple sentence information table 115, the prominent noun phrases included in the simple sentences of the simple sentence numbers 80 and 81 are implicit types, but the word information table 116.
According to the above, since all prominent noun phrases include proper nouns, the priority is 2 according to the priority rule shown in FIG. On the other hand, the simple sentences of the simple sentence numbers 82 and 83 have no prominent noun phrases, and thus do not have the topic word candidate priority. FIG. 12 shows the state of the single sentence information table 115 after the topic word candidate priorities are assigned as described above for the first topic establishment section of this semantic development. Since the single sentence numbers 82 and 83 have no priority, a value of -1 is recorded in the topic word candidate priority field.

[Detection of Adjunct Modification Relationship] As shown in the flowchart of FIG. 9, after giving the priority of topic word candidates, the association modification of the topic word candidates is detected. Here, the union modification is a modification to a noun. For example, in the noun phrase “camera he brought”, a simple sentence “he brought” modifies the noun “camera”. An example of a process for detecting the noun modification relation will be described with reference to the flowchart in FIG. The process of detecting the noun modification relation is performed for each topic word candidate.

First, the last word in the topic word candidates is A
And the word immediately before A in the original text is B (step 211). It is determined whether or not B is included in the current topic word candidate (step 212). If B is included in the topic word candidate, the process proceeds to step 215. If B is not included in the topic word candidate, It is determined whether B is a continuous form of the inflected word (step 213). If it is determined in step 213 that it is not a continuous form, the process proceeds to step 215. If it is a continuous form, it is assumed that the simple sentence including B modifies the topic word candidate A in a continuous form (step 21).
4) Go to step 215. That is, the word B immediately before the last word A in the topic word candidate is not included in the topic word candidate, and the word B is a series of inflected words whose endings change like verbs or adjectives. If it is a body shape, a simple sentence containing B is regarded as a subject word candidate that has been modified continuously.

At step 215, it is checked whether the word A is at the earliest position in the topic word candidate. If the word is at the earliest position, the processing is terminated. The word before A in the word candidate is set as A again, and the word immediately before A after the update in the original text is set as B (step 216).
Then, the process returns to step S12, and the process is repeated from the process of determining whether or not B is included in the topic word candidate.

Consider the example shown in FIG. Since only the word with the word number 1059 is the topic word candidate of the salient noun phrase included in the single sentence with the single sentence number 80, the word with the word number 1059 is A and the word with the word number 1058 is B. The word B is not included in the topic word candidate but is not an inflected word, so that the union modification relation is not detected. Also, since A is the earliest word in the topic word candidate, the process of detecting the union modification relation for this topic word candidate ends here.

Next, topic word candidates of salient noun phrases included in the simple sentence with the simple sentence number 81 are examined. Since this topic word candidate is composed of two words having the word numbers 1063 and 1064, the word of the word number 1064 is assumed to be A, and the word of the word number 1063 is assumed to be B. Then, since word B is included in the topic word candidate, the union modification relation is not detected at this time. At this time, since A is not the earliest word in the topic word candidates, the word number 1063 is newly added.
Is assumed to be A and word B of word number 1062 is assumed. Then, since B is not included in the topic word candidate and B is a conjunctive form of the conjugate word, it is determined that the simple sentence 80 including B modifies the current topic word candidate by the union. At this point, A is the earliest word in the topic word candidates, so the processing is terminated here.

[Correction of Priority and Selection of Topic Word] As shown in the flowchart of FIG. 9, after detecting the union modification relation, the topic word candidate priority correction table 12
The priority of each topic word candidate is corrected according to 7. FIG. 14 shows a configuration example of the topic word candidate priority correction table 127. In this example, the priority of the topic word candidate modified by the unary sentence with another simple sentence is corrected to 0.5 to increase the priority. In addition, as a method of correcting the priority using the topic word candidate priority correction table, according to the characteristics of the text data to be analyzed,
For example, it is conceivable to operate the priority order of the topic word candidates included in the simple sentence that modifies other topic word candidates. In some cases, instead of giving an absolute value such as 0.5 as a priority after correction, it may be better to add or subtract a value up to that value such as -0.5. The correction method is appropriately selected according to the properties.

In the example shown in FIG. 8, since the topic word candidates of the salient noun phrase included in the simple sentence with the simple sentence number 81 are modified by the unary sentence 80, the topic word candidate priority shown in FIG. According to the correction table, its priority is corrected to 0.5. FIG. 15 shows the state of the single sentence information table 115 after the priority is corrected.

When a topic word is selected based on the simple sentence information table 115 (see FIG. 12) before the priority is corrected based on the union modification relationship, the priority of both the single word topic word candidates of the simple sentence numbers 80 and 81 are changed. Since it is 2, the topic word candidate included in the simple sentence with the simple sentence number 80 appearing earlier in time is selected as the topic word. That is, the word “Company P” with the word number 1059 is selected as a topic word. On the other hand, when a topic word is selected based on the single sentence information table 115 (see FIG. 15) after the priority is corrected according to the noun modification relation as described above, the topic word candidate of the simple sentence with the simple sentence number 81 is selected. Since the priority is the highest at 0.5, “product Q” including the words with the word numbers 1063 and 1064 is selected as the topic word.

FIG. 16 shows an example of the result of topic structure recognition for the text example of FIG. 5 using the method of the present embodiment. Comparing with FIG. 6B showing the result using the conventional method, it can be seen that the result of the method of the present embodiment makes it possible to easily estimate the contents of the original text.

[Results of Evaluation Test on Topic Structure Recognition] Next, actual text data was obtained using a topic structure recognition system incorporating the above-described topic word selection method of the present embodiment and a system not incorporating this method. The result of recognizing the topic structure will be described. In the experiment, a total of 63 newspaper articles were used as text data. Then, the topic structure recognized by a human and the topic structure recognized by the computer are compared with respect to the text data to calculate a recall and a precision, and the system according to the present embodiment and the system according to the present embodiment are not used. Both systems (conventional systems) were evaluated.
Here, the recall is the topic structure recognized by humans.
The precision is a measure indicating how much is recognized by the computer, and the relevance is a measure indicating how much of the topic structure recognized by the computer is also recognized by a human. If the topic structures recognized by the human and the computer match, both the recall and the precision are 100%. The precision and recall were obtained for each topic establishment section, topic word, and topic scope. The relevance and recall for the topic scope are the weights of the topic scopes that evaluate the correct topic words.
For example, if a large topic with a long topic scope is correctly recognized, the precision and recall will improve, and conversely, even if a small topic with a short topic scope is misrecognized, the precision and recall will not decrease so much . Table 1 shows the results.

[0041]

[Table 1] As is clear from Table 1, by considering the noun modification relationship according to the present invention, both the relevance and the recall are improved for the topic scope as compared with the conventional method not considering the noun modification relationship. Thus, it was confirmed that topic words having a larger topic scope can be correctly recognized by using the present invention.

[0042]

As described above, according to the present invention, the priorities are assigned to the topic word candidates and then the priorities are corrected by the topic word candidate priority correction table.
You will be able to correctly determine topic words. In particular, by correcting the priority based on the noun modification relation, there is an effect that a topic word of a written word text including a complicated noun modification can be correctly recognized. This is especially useful for topics with a long topic scope. Since a large topic structure is correctly recognized, it becomes easy for the user to understand when used as a chapter / table of contents structure.

[Brief description of the drawings]

FIG. 1 is an example of topic structure recognition by a human.

FIG. 2 is a block diagram showing the structure of an example of a conventional topic structure recognition device.

FIG. 3 is a flowchart showing a conventional process for topic structure recognition.

FIG. 4 is an example after the pre-processing in the conventional topic structure recognition.

FIG. 5 is a diagram illustrating an example of a text.

6A is a diagram showing an example of a topic structure recognized by a human for the text shown in FIG. 5, and FIG. 6B is a diagram showing a case where a conventional topic structure recognition method is applied to the text shown in FIG. FIG. 14 is a diagram showing a result of extracting a topic structure by using FIG.

FIG. 7 is a block diagram illustrating a configuration of a topic structure recognition device according to an embodiment of the present invention.

FIG. 8 is a diagram showing a relationship between each table and each storage unit.

FIG. 9 is a flowchart showing a process for determining a topic word performed using the apparatus of FIG. 7;

FIG. 10 is a diagram showing an example of a topic candidate priority order rule for base development.

FIG. 11 is a diagram illustrating an example of topic candidate priority rules for semantic development.

12 is a diagram showing a state of a single sentence information table at the time when priorities are assigned to respective topic candidates according to the topic candidate priority rules of the semantic development of FIG. 11;

FIG. 13 is a flowchart showing a process for detecting a modification of a union.

FIG. 14 is a diagram illustrating a configuration example of a topic candidate priority correction table.

FIG. 15 is a diagram illustrating a state of the single sentence information table after the topic candidate priority is corrected.

FIG. 16 is a diagram showing an example of a result when the method of the present invention is applied to the text example of FIG. 5;

[Explanation of symbols]

 Reference Signs List 101 data input unit 102 processing unit 103 display unit 104 storage unit 105 dictionary / rule unit 110 language data storage unit 111 topic structure storage unit 112 base development storage unit 113 semantic development storage unit 114 integrated topic storage unit 115 single sentence information table 116 words Information Table 121 Preprocessing Dictionary 122 Semantic Expansion Processing Rule 123 Base Expansion Processing Rule 124 Integration Processing Rule 125 Topic Word Candidate Priority Rule 126 Question Expression Dictionary 127 Topic Word Candidate Priority Correction Table 201-203, 211-216 Step

────────────────────────────────────────────────── ─── Continuation of front page (56) References JP-A-7-160710 (JP, A) JP-A-7-160711 (JP, A) JP-A-7-160712 (JP, A) JP-A-6-160712 236410 (JP, A) JP-A-6-139276 (JP, A) JP-A-4-332084 (JP, A) Atsushi Takeshita, "Video Retrieval System Using Topic Structure Recognition", Information Processing Society of Japan Research Report 94- IM-15, Japan, March 11, 1994, Vol. 94, no. 24, p. 1-p. 8. Atsushi Takeshita, et al., "Detection of topic introductions in D-67 monologue" 1994 IEICE Autumn Conference, Japan, September 5, 1994, p. 70 (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/30

Claims (8)

(57) [Claims] 1. A method for determining a topic word in a topic structure recognition apparatus for recognizing a topic structure of language data using rules prepared in advance, wherein the topic structure recognition apparatus assigns priorities to topic word candidates.
Topic word candidate priority rule for giving, and the priority
That describes a set of correction conditions and correction contents for correcting
Dictionary / rule part including title candidate priority correction table
And assigns priorities to topic word candidates extracted from the language data in accordance with the topic word candidate priority rules. Based on the topic word candidate priority correction table, correction conditions and corrections given in advance are provided. According to a set of contents, when the correction condition is satisfied, the priority of the topic word candidate is corrected according to the corresponding correction content, and the priority assigned to a plurality of topic word candidates is compared. A topic word selection method characterized by determining a topic word by using 2. The topic structure recognizing device divides the linguistic data into simple sentences which are units each having only one predicate, and for each simple sentence, a prominent noun phrase which is a noun phrase most emphasized in the simple sentence. To perform topic structure pre-recognition processing including identifying the type of the prominent noun phrase, and then develop the topic into an explicitly indicated base expansion and a semantic expansion developed in the base expansion Using a base expansion processing rule and a semantic expansion processing rule, respectively, for the base expansion, determining a topic establishment section in which a topic is presented / established; determining a topic word in the topic establishment section; The topic level and topic scope representing the nesting of topics are sequentially determined, and then, regarding the semantic expansion, the topic establishment section is determined, the topic word is determined, and the topic scope and topic level are sequentially determined. After using the integrated processing rules, if rows the process of integrating with respect to each of the processing result of the semantic development based deployment,
Topic word selection method as claimed in 請 Motomeko 1. 3. The topic word selection method according to claim 1, wherein the set of the correction condition and the correction content includes, as the content, a priority order correction corresponding to a noun modification relation to the topic word candidate. 4. The topic word selecting method according to claim 3, wherein the set of the correction condition and the correction content includes a content of raising a priority for a topic word candidate which is subject to continuous modification. 5. An input unit for inputting language data,
A dictionary / rule part for storing rules for topic structure recognition;
Includes a processing unit that performs processing using the dictionary specification and regulations of the rules include, a storage unit for storing the result of the processing unit, and a display unit for displaying the processing result by the processing unit, said dictionary and regulations section A topic word candidate priority rule for assigning a priority to a topic word candidate, and a topic word candidate priority correction table that describes a set of correction conditions and correction contents for correcting the priority. The storage unit includes: a language data storage unit that stores information about language data input from an input unit; and a topic structure storage unit that stores information about a topic structure. A word information table that stores information on the character strings and parts of speech of each included word, and information on words, prominent noun phrases, and types of prominent noun phrases included in each simple sentence of the language data. A sentence information table, wherein the topic structure storage unit includes a table for storing information including a topic establishment section, a topic word, a topic level, and a topic scope, which are ranges in which topics are presented and established. Topic structure recognition device. 6. The topic structure recognition pre-processing, wherein the dictionary / rule unit divides the language data into simple sentences, extracts salient noun phrases from each simple sentence, and identifies the type of the salient noun phrases. Dictionary for preprocessing, base expansion processing rules for performing processing for base expansion, semantic expansion processing rules for performing processing for semantic expansion, and integration for integrating base expansion and semantic expansion A processing rule, wherein the topic structure storage unit stores a base development information, a semantic development storage unit that stores information about the semantic development, and information after the integration of the base development and the semantic development. , And each of the semantic expansion storage unit and the base expansion storage unit includes a topic establishment section in which a topic is presented and established, a topic word, a topic level, and a topic scope. Comprising a table for storing information including, the topic structure recognition apparatus according to claim 5. 7. The topic structure recognition apparatus according to claim 5, wherein the modification condition of the topic word candidate priority correction table includes a continuous modification relation to the topic word candidate. 8. A set of the correction condition and the correction content of the topic word candidate priority correction table includes a correction condition that the topic word candidate is modified by the union by another simple sentence and a correction content that the priority is increased. The topic structure recognition device according to claim 7, wherein the set includes: