JP2003271596A - Language processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a language processor for extracting a proper word as an important word even from dialogic data including wrong recognition or ambiguity. <P>SOLUTION: A text information storage means 2 stores, for example, the past text information in a dialogue, and a text information acquisition means 11 acquires, for example, the present text information in the dialogue concerned. A matched character string part detection means 3 detects a character string part matched between the text information acquired by the text information acquisition means 1 and the text information stored in the text information storage means 2, and an important word extraction means 4 extracts the important word from the character string part detected by the matched character string part detection means 3. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a language processing apparatus for extracting an important word from text information, and in particular, a language for extracting an appropriate word as an important word even from dialogue data including recognition error or ambiguity. Regarding a processing device.

[0002]

2. Description of the Related Art For example, according to a conventional method that has been used to extract an important word indicating a content or a topic from dialogue data, keyword spotting or morphological analysis means using a word dictionary is input from the inputted dialogue data. Words are extracted by.

In the "advertising device for electronic dialogue" described in Japanese Patent Laid-Open No. 11-3348, a word dictionary (advertising dictionary) is prepared in advance, and when a word registered in the dictionary appears in the dialogue data, Information related to the word (advertising information) is displayed with the word as an important word.
In the "automatic information providing method" described in Japanese Patent Laid-Open No. 6-236410, a database in which a word (topic) and a field to which the word belongs are recorded is used instead of a simple word dictionary.
It is performed to determine which field the dialogue belongs to, and to detect that the field to which the dialogue belongs has been converted.

In the "dialogue processing device" described in Japanese Patent Laid-Open No. 8-137874, a synonym dictionary, a word dictionary, and a contrast word dictionary are applied to input dialogue data in order to detect the presence or absence of topic change. With a morphological analyzer equipped with
A word list is created. JP-A-10-6
In the “topic processing device” described in Japanese Patent No. 9482, morphological analysis processing is performed on dialogue data (speech object) to extract a specific type of word as a keyword.

[0005]

However, in any of the above-mentioned conventional important word extraction methods, only words are extracted from dialogue data by using keyword spotting or morphological analysis with a word dictionary and then extracting only the words. It is held in a recording device to count the frequency and measure the distance between words on a space vector.
Such a conventional method has a problem in that once the first word extraction fails, a correct solution cannot be obtained again. For example, if the dialogue data includes noise, that is, a recognition error character or ambiguity, the word extraction is highly likely to fail.

[0006] As a concrete example, when there is the data "The big trash of this place is fish",
The word "dai" is incorrect, or "large garbage"
When the ambiguity between "and" is "real" is included, the correct recognition result of the dialogue is {this (adjective) / local (noun-
General) / of (particle-unification) large (adjective) / dai (error) / garbage (noun-general) / ha (partition particle) / fish (noun-general) / is (auxiliary verb)} This string is {this (adjective) / local (noun-general) / of (particle-adjective) large (adjective) / daigo (noun) / is (engagement particle) / fish (noun-general) / is (auxiliary verb) )} After being converted into a word list written in a mixture of kanji and kana,
In addition to being unable to extract the word "garbage" that was originally desired, there is a possibility that unnecessary words "the real pleasure" will be extracted as keywords.

The present invention has been made in view of such conventional circumstances. For example, even if text information includes an error or ambiguity, a suitable word can be extracted as an important word. The purpose is to provide such. More specifically, in the present invention, for example, even when the dialogue data obtained by the voice recognition device or the like contains an erroneous character string or an ambiguous character string, these errors and By keeping the ambiguity as it is and eliminating errors and ambiguity depending on the dialogue before and after, it is possible to properly extract the important words for expressing the contents of the dialogue.

[0008]

To achieve the above object, in a language processing apparatus according to the present invention, a text information storage means stores text information, a text information acquisition means acquires text information, and a matching character string is obtained. The part detecting means detects a matching character string part between the text information acquired by the text information acquiring means and the text information stored in the text information storing means, and the important word extracting means is detected by the matching character string part detecting means. The important word is extracted from the character string part.

Therefore, since the important word is extracted from the character string portion where the stored text information and the acquired text information match, an error or ambiguity may occur in the stored text information or the acquired text information. Even when it is included, an appropriate important word can be extracted. In other words, except when the stored text information and the acquired text information include exactly the same error, the detected matching character string portion does not include the error, so the error is not included. It is possible to reduce the ratio of extracting a word as an important word compared to the past,
This makes it possible to realize extraction of an appropriate important word.

In FIG. 3, as a schematic configuration example of the language processing apparatus according to the present invention, a text information storage section 31 having a function of text information storage means and text information having a function of text information acquisition means. An acquisition unit 32, a matching character string portion detecting unit 33 having a function of a matching character string portion detecting unit, and an important word extracting unit 34 having a function of an important word extracting unit are shown, and the text information storage unit 31 is shown. The text information 41 supplied from the to the matching character string portion detection unit 33 is shown.

Here, various information may be used as the stored text information or the acquired text information, for example, information including a sentence is used. The number of stored text information used to detect a matching character string portion or the number of acquired text information may be one, for example, but one of the other aspects is Alternatively, both may be plural, and in this case, the matching character string portion is detected in all of the three or more pieces of text information.

As the stored text information used to detect the matching character string portion and the acquired text information, as a preferred mode, for example, the topic contents are the same or similar but include different sentences. Text information is used. Specifically, for example, text information about past statements and text information about current statements about the same or similar topics in dialogues and lectures, or
It is possible to use various text information such as opinions of a plurality of people at the same place such as the same exhibition hall in a museum or text information of a questionnaire.

Further, the text information storage means can be constituted by using a memory, for example. Further, as the method of acquiring the text information by the text information acquisition means, not only the method of acquiring the text information in the format of the text information but also various methods such as the method of converting the format of the voice information into the format of the text information and acquiring the information are available. Methods may be used.

As the character string portion detected by the matched character string portion detecting means, for example, a common character string portion included in the stored text information and the acquired text information is detected. Further, not only a character string composed of two or more characters but also a mode in which one character is detected as a character string may be used. The number of character strings to be detected is not particularly limited, and various numbers may be used. Also, as the characters that make up the character string,
Not only hiragana, katakana, romaji, etc., but also various things such as symbols may be included.

Various methods may be used to detect the character string portion by the matching character string portion detecting means. In a preferred mode, for example, from the phonetic character before the text information is divided into words. It is possible to use a method in which a sequence of matching characters is detected as a sequence of matching character strings when viewed as a sequence of characters having no meaning, in order from front to back or from rear to front.

Further, various words may be used as the important words extracted by the important word extracting means. Further, various methods may be used to extract the important word by the important word extracting means, for example, by detecting a word having meaning in the detected matching character string portion, Is extracted as an important word. In addition, the extracted important word is stored in a storage unit such as a memory or displayed on a display unit such as a screen, for example.

In the language processing device according to the present invention, as one configuration example, the text information stored in the text information storage means is text information obtained by voice recognition of voice information. Further, in the language processing device according to the present invention, as one configuration example, in the text information acquisition means, the voice information input means inputs voice information, and the voice recognition means voice recognizes the voice information input by the voice information input means. And convert it to text information.

Therefore, for example, with respect to the voice information uttered by a person, the important word can be extracted. In general, the present invention is particularly effective because a recognition error or the like may occur in text information obtained by voice recognition of voice information. Here, various information may be used as the audio information. Further, as the voice recognition process, for example, a generally known technique can be used. The voice information input means can be configured by using a microphone, for example.

In the language processing apparatus according to the present invention, as a preferred example, the text information stored in the text information storage means and the text information acquired by the text information acquisition means have contents related to each other. Specifically, for example, text information that includes a common word in the stored text information and the acquired text information is used, and such a common word can be extracted as an important word.

Further, in the language processing apparatus according to the present invention, as an example of the configuration, the voice information input means of the text information acquisition means inputs voice information which is currently uttered by a dialogue performed by two or more people, and stores the text information. The text information stored in the means is text information obtained by voice-recognizing voice information issued in the past in the dialogue. Therefore, for example, with respect to the conversation contents talking about the same or similar topic, the important word can be extracted from the information of the words spoken by the speaker.

Further, in the language processing apparatus according to the present invention, as one configuration example, the text information storage means deletes the stored text information from the stored contents when a predetermined period has come. Therefore, it is possible to extract the important word for each predetermined period such as a break point of the dialogue.

Here, various periods may be used as the predetermined period, for example, a predetermined period such as 5 minutes or 10 minutes can be used. As a specific example of a preferable mode, 10 minutes can be used as a predetermined period in a 10-minute dialogue, and 5 minutes can be used as a predetermined period in a dialogue in which the content of a topic switches every 5 minutes. , The period until the end of the conference can be used as the predetermined period in the dialogue at the conference, and the period in which the painting is exhibited for the conversation about the predetermined painting in the museum is used as the predetermined period. be able to.

Further, in the language processing apparatus according to the present invention, the topic conversion detection means is converted into voice information input by the voice information input means of the text information acquisition means or text information converted by the voice recognition means of the text information acquisition means. Based on the detection of the topic change in the dialogue, the text information storage means deletes the stored text information from the stored contents in response to the topic change detection means detecting the topic change. Therefore, for example, the important word can be extracted for each period until a topic change occurs in the dialogue.

Here, as the conversion of the topic in the dialogue, detection of a predetermined word or phrase indicating a point of change of the topic, such as "By the way" or "Speak changes," is performed. It is possible to detect the silent time (the time during which no sound is emitted) by regarding it as the turning point of the topic.

Further, in the language processing apparatus according to the present invention, as one configuration example, the text information acquisition means acquires text information currently issued by a dialogue performed by two or more people using the text information on the network, The text information stored in the text information storage means is the text information issued in the past in the dialogue. Therefore, the important word can be extracted from the text information issued by chat on the Internet, for example.

Further, in the language processing apparatus according to the present invention, as one configuration example, in the important word extraction means, the morpheme analysis means performs morpheme analysis on the character string portion detected by the coincident character string portion detection means, and the part-of-speech information is added. And the predetermined part-of-speech word extraction means extracts the word of the predetermined part-of-speech as an important word from the word list acquired by the morpheme analysis means.

Here, as the morphological analysis processing, for example, a generally known technique can be used.
Further, as the word list with part-of-speech information, for example, the one obtained by converting the detected character string portion into a list of words with part-of-speech information is used. Various parts of speech such as nouns may be used as the part of speech of the word having a predetermined part of speech. Specifically, a mode can be used in which, for example, a word having a necessary part-of-speech such as a noun is extracted, while an unnecessary part-of-speech word such as a particle is not extracted.

Further, in the language processing apparatus according to the present invention, as one configuration example, the important word extracting means stores words that are candidates to be extracted as important words by the important word candidate word storing means, and the matching word extracting means A word that matches the word stored in the important word candidate word storage means is extracted as an important word from the character string portion detected by the matching character string portion detection means.

Various words may be used as the words stored in the important word candidate word storage means. The stored word is used as a keyword for extracting the important word, that is, if there is a word that matches the keyword in the detected character string portion, the word is extracted as the important word. It Further, various numbers may be used as the number of words stored in the important word candidate word storage means. Further, the important word candidate word storage means can be configured by using a memory, for example.

Further, in the language processing apparatus according to the present invention, the important word extracting means extracts the important words from the character string portion detected by the matching character string portion detecting means, excluding words satisfying a predetermined condition. Here, various conditions may be used as the predetermined condition. As a specific example, when the condition of one-character word is used, important words are extracted except for one-character words, that is, two or more important words are extracted.

Further, in the language processing apparatus according to the present invention, the importance degree giving means gives the importance degree to the important word extracted by the important word extracting means. Here, as the importance of the important word, for example, a numerical value in which the more important the value is, the larger the value can be used. Also, the importance of important words should be used, for example, in order to order these important words or to select some of these important words when multiple important words are extracted. You can

Further, in the language processing apparatus according to the present invention, as one configuration example, in the importance degree giving means, the importance degree appearance frequency calculating means calculates the appearance frequency of the important word extracted by the important word extracting means, The word appearance frequency information storage means stores the important word extracted by the important word extraction means and the information of the appearance frequency of the important word in association with each other, and the important word importance degree calculation means performs the operation based on the appearance frequency of the important word. Calculate the importance of important words.

Here, as the important word appearance frequency information storage means, for example, a memory can be used. As the appearance frequency of the important word, for example, the number of times the same important word is extracted can be used. Further, for example, a mode in which the importance of an important word is proportional to the frequency of appearance of the important word can be used.

Further, in the language processing apparatus according to the present invention, as one configuration example, the inter-word relevance information storage means stores the relevance information between words, and the important word relevance word acquisition means stores the inter-word relevance information. Another word related to the important word extracted by the important word extracting means is acquired based on the stored contents of the storing means. Note that the other words to be acquired may be other words inferred from the extracted important words, for example, in the case of providing important words for thinking of the next topic to those who are having a dialogue. Used as an important word.

Here, the inter-word degree-of-relationship information storage means can be configured by using a memory, for example. As information on the degree of association between words, for example, information on a plurality of related words and information on the degree of association thereof are used. Further, various words may be used as the words acquired by the important word related word acquiring means, for example, one word most relevant to the extracted plural important words may be acquired, or, for example, extraction may be performed. It is possible to obtain all the words related to the specified important word.

The technical idea according to the present invention as described above can be applied to, for example, a method, a program, a storage medium storing such a program, or the like. For example, in the language processing method according to the present invention, text information is acquired, a matching character string portion is detected between the acquired text information and the text information stored in the storage means, and an important word is detected from the detected character string portion. Extract. Here, for example, a memory is used as the storage means.

Further, in the program according to the present invention, the function of acquiring the text information, the function of detecting the matching character string portion between the acquired text information and the text information stored in the memory, and the detected character string portion The function of extracting important words from is realized by a computer. Here, various programs may be used as the program.

Further, in the storage medium according to the present invention, the program to be executed by the computer is stored so that it can be read by the input means of the computer, and the program stores the process of acquiring text information and the acquired text information. The computer is caused to execute a process of detecting a character string portion that matches the text information stored in the memory and a process of extracting an important word from the detected character string portion. Here, the storage medium is, for example, a floppy (registered trademark) disk or a CD (Compact Disk) -ROM.
Various types such as (Read Only Memory) may be used.

[0039]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described with reference to the drawings. In this example, a case is shown in which the present invention is applied to a dialogue processing device that extracts important words from dialogue data. FIG. 1 shows a configuration example of the dialogue processing apparatus of this example. The dialogue processing apparatus of this example includes a dialogue input unit 1 having a microphone device 11 and a dictation processing unit 12, and a dialogue recording unit 2 having a dialogue recording device 13 and an input data storage unit 14.
A dialogue comparison unit 3 having a character string comparison unit 15, a word dictionary unit 16, a morpheme analysis unit 17, and an important word extraction unit 4 having an important word extraction unit 18 are provided.

The dialogue input unit 1 recognizes a dialogue and converts voice data into text data. The dialogue recording unit 2 stores the text data, which has been compared by the dialogue comparing unit 3, in the dialogue recording device 13. The dialogue comparison unit 3 compares the text data in the past dialogue with the text data input at present, and outputs the matched character string. The important word extraction unit 4 performs morphological analysis on the matched character string to convert the character string into a word list with part-of-speech information, and outputs a word having a specific part-of-speech from the word list as an important word.

An example of the operation performed by the dialogue processing apparatus of this example will be shown below. In this example, an important word is extracted from voice data that includes an erroneous character string or ambiguity, for example. In the dialogue input unit 1, the voice in the dialogue is input from the microphone device 11, and the voice data input from the microphone device 11 is text data represented by phonetic symbols such as hiragana, katakana, and Roman letters by the dictation processing unit 12. And outputs to the dialogue comparison unit 3.

In the dialogue comparison unit 3, the text data received from the dialogue input unit 1 is compared with the past dialogue recording data 21 by the character string comparison unit 15, and the matching character string is output to the important word extraction unit 4. To do. Here, the past dialogue record data 21 is the past voice data in the dialogue as text data, and is supplied from the dialogue recording unit 2. The character string comparison unit 15 also outputs the text data received from the dialogue input unit 1 to the dialogue recording unit 2.

In the dialogue recording unit 2, the text data received from the dialogue input unit 1 and subjected to the comparison processing by the dialogue comparing unit 3 is added to the dialogue recording data held in the dialogue recording device 13 by the input data storage unit 34. Record in the form of As a specific example, the text data to be input is the text data of the sentence "The big trash of this place is fish", and the dialogue record data before the addition is "The trash of the trash is also good in this place." When the text data of the sentence is added, the dialogue record data after the addition is the text data of the sentence "The trash of the garbage is safe even in this area / The big garbage of this garbage is fish".

It is also possible to use a mode in which the dialogue data recorded in the dialogue recording device 13 is deleted, for example, after being held for a certain period. In addition, for example, Japanese Patent Laid-Open No.
The topic conversion is detected by using the existing technology described in Japanese Patent Publication No. 137874, and at the same time when the topic conversion is performed,
It is also possible to use a mode in which the dialogue data recorded in the dialogue recording device 13 is deleted.

In the important word extraction unit 4, first, the morpheme analysis unit 17 refers to the dictionary contents of the word dictionary unit 16 which stores the word dictionary in which part-of-speech information is described, and refers to the characters input from the dialogue comparison unit 3. By performing morphological analysis on the string, a word list in which the character string is divided into words and part-of-speech information is added to each word is generated and output to the important word extraction unit 18. Next, the important word extraction unit 18
A word having a specific part of speech or a specific word is extracted from the word list input from the morphological analysis unit 17, and these are output as the important word 22.

Next, referring to FIG. 2, the character string comparison unit 15
Thus, an example of a processing procedure for comparing the text data input from the dialogue input unit 1 with the dialogue data recorded in the dialogue recording unit 2 and outputting a matching character string will be described. First,
The text data input from the dialogue input unit 1 is String_c
It is read into a variable called urrent (step S1). In this example, it is assumed that the input text data is the text data of the sentence "This big trash is fish".

Next, the text data in the past dialogue (past dialogue data) is input from the dialogue recording device 13.
Read in String_log (step S2). In this example, it is assumed that the input conversational data is text data of a sentence, "The garbage and futoku are even better." Next, change the above String_current to String_o
Copy it to a variable called rg (step S3).

Here, as for the subsequent processing (the processing of steps S5 to S10), the processing of loop 1 is repeated until the length of the character string of String_org becomes 0 (step S4). As for the subsequent processing (the processing of step S7, step S8, and step S10), the processing of loop 2 is repeated until the length of the character string of String_current becomes 0 (step S5).

In the processing of loop 2 in the processing of loop 1, first, String_current and String_log are compared and
It is determined whether the entire String_current matches a part of the String_log (step S6). If they do not match, the last character of String_current is deleted (step S10).

[0050] In this example, first, the string of characters "This big trash is fish" is compared with the string of characters "The trash of the garbage is safe in this garbage". At this stage, there is no such partial matching relationship as described above, so String_current
The last character of the is deleted, and String_current becomes a string of characters "The big trash in this place is in the fish." Then, as a result of repeating such a character deletion process, in this example, String_current becomes a string of characters "this way", and at this stage, it matches a part of String_log.

In this way, the character string of String_current is St
If it partially matches the ring_log (step S6),
The character string of String_current is output to the important word extraction unit 4 (step S7). Next, delete the part of the string that matches the String_current from String_org, and
The character string of String_org is copied to String_current (step S8). In this example, the character string "Kono Chiho" is deleted from String_org, and as a result, String_org after the deletion is "Onodai Daibashi is a fish."
Will be a string of characters.

When the length of String_current becomes 0 as a result of repeating the processing of loop 2 as described above, the processing of loop 2 concerned is once terminated. And loop 1
As the processing of, the first character of String_org is deleted (step S9), and then String_org is replaced with String_curre.
After copying to nt, the above processing of loop 2 is performed again.
In this example, in the first loop 2 process, String_org becomes "large big trash is fish".

When the length of String_org becomes 0 as a result of repeating the processing of loop 1 including the processing of loop 2 as described above, the processing is terminated. In this example, by the processing by the character string comparison unit 15 as described above, “this way”,
Five character strings “no”, “ki”, “garbage”, and “da” are detected and output to the important word extraction unit 4.

Next, an example of a procedure of processing for extracting an important word by the important word extracting section 4 will be shown. First, the morphological analysis unit 1
7 refers to the word dictionary of the word dictionary unit 16 and converts the character string input from the dialogue comparison unit 3 into a word list with part-of-speech information. In this example, as the character string to be converted,
"This way", "no", "ki", "garbage", "is"
It is assumed that the following five character strings are input. In this case, in this example, as a result of converting these character strings, {this (adjective) / chiho (noun-general)}, {no (adnominal particle)}, {ki (noun-general)}, {garbage ( Noun-general)}, {is (auxiliary verb)} information is obtained as a word list with part-of-speech information.

Next, the important word word extraction unit 18 extracts a word having a specific part of speech from the word list obtained by the morpheme analysis unit 17. In this example, a noun is the extraction target. In this case, in this example, three words “chiho”, “ki”, and “garbage” are extracted as the important words 22.

Note that, for example, it is possible to perform a filtering process such as deleting the character string having a length of 1 such as "no" before performing the morphological analysis. In the configuration for performing such filtering processing, “no” and “ki” corresponding to a character string of one character are deleted from the above-mentioned four character strings, so that “chiho” is an important word. 2 called "garbage"
One word is extracted.

Further, for example, using the existing technology described in Japanese Patent Laid-Open No. 8-137874, the extracted important words are given a degree of importance based on their appearance frequency, and the given degree of importance is given. It is also possible to adopt a configuration in which a plurality of important words are narrowed down to an arbitrary number according to the above. In addition, for example, a structure that performs a process of predicting a new important word by inquiring a dictionary that describes the degree of association between words and selecting a word that is highly related to the extracted important word It is also possible to

As described above, in the dialogue processing apparatus of this example,
The past in which the dialogue input unit 1 receives voice data from a dialogue performed by, for example, two or more people, converts the voice data into text data, and records the input text data (conversation data) by the dialogue comparison unit 3. The text data of the dialogue (interaction data) is output, the character string of the matched portion is output, and the dialogue recording unit 2 stores the input dialogue data,
The important word extraction unit 4 extracts a specific word from the character string received from the dialogue comparison unit 3.

Further, in the dialogue processing apparatus of this example, for example, the dialogue recording unit 2 stores the inputted dialogue data only for a certain period, or, for example, the dialogue recording unit 2 switches the conversation data inputted into a topic. Is stored until is detected.

Further, in the dialogue processing apparatus of this example, for example, the important word extraction unit 4 converts the character string received from the dialogue comparison unit 3 into a word list with part-of-speech information by the morphological analysis unit 17, and the important word word extraction unit. 18 is used to extract a word having a specific part of speech, or, for example, the important word extraction unit 4 stores a word serving as a keyword by the word dictionary holding device, and the important word extraction unit 18 extracts a character string received from the dialogue comparison unit 3. A word that matches the keyword stored in the word dictionary holding device is extracted from the inside.

Further, in the dialogue processing apparatus of this example, for example, the important word extracting unit 4 uses the filtering function to make the dialogue comparing unit 3
Unnecessary character strings are deleted from the character strings received from to extract important words. As a specific example, a character string that meets a preset condition is deleted from the important word extraction target. For example, a character string that consists of one character, a character string that includes an unknown symbol, or "" (square brackets). ) Or a character string such as @ used in an email address is set to be deleted from the target of important word extraction.

Further, in the dialogue processing apparatus of this example, for example, the important word extraction unit 4 gives the importance degree to the word by the importance degree giving function. Further, in the dialogue processing apparatus of this example, as the processing of assigning the importance, for example, the frequency calculation function calculates the frequency of the extracted word, and the frequency information holding device holds the information of the word and the frequency, and the importance calculation function. Calculates the importance of a word based on frequency information.

In the dialogue processing apparatus of this example, for example, the important word extraction unit 4 holds the related word dictionary in which the degree of relevance between words is described by the related word dictionary holding device, and uses the important word prediction function. A new word is predicted based on the degree of association between words and output as an important word.

Further, in the dialog processing apparatus of this example, an example of the structure in which the voice data in the dialog is converted into the text data and the important word is extracted is shown. However, as another example of the structure, the text data in the chat etc. is inputted. It is also possible to extract an important word from the text data, and in this case, for example, the dialogue input unit 1 includes a character input terminal that allows two or more persons to have a dialogue with a character string on the network. As the character input terminal, various ones such as a personal computer having a keyboard can be used.

With the above-mentioned configuration, the dialogue processing apparatus of the present embodiment can reduce the influence of such recognition error even if it is, for example, from voice recognition data including recognition error or ambiguity, and it is appropriate. It is possible to extract important important words, and it is possible to increase the certainty of the extracted important words.

In this example, the function of the dialogue recording section 2 constitutes the text information storage means, the function of the dialogue input section 1 constitutes the text information acquisition means, and the function of the dialogue comparison section 3 The matching character string portion detecting means is constituted, and the function of the important word extracting portion 4 constitutes the important word extracting means. Further, in this example, the function of the microphone device 11 constitutes the voice information input means, and the function of the dictation processing section 12 constitutes the voice recognition means.

Further, in this example, the function of the morpheme analysis unit 17 constitutes a morpheme analysis unit, and the function of the important word word extraction unit 18 constitutes a predetermined part-of-speech word extraction unit. Further, in this example, for example, the function of an important word candidate word storage unit that stores words that are candidates to be extracted as important words and a word that matches the stored word is extracted as an important word from the character string portion. The function of the matching word extracting means may be provided in the important word extracting unit 4.

Further, in this example, for example, the dialogue recording section 2 or the like can be provided with a function of topic conversion detecting means for detecting conversion of a topic. Further, in the present example, for example, the function of the importance degree giving means may be provided in the important word extracting unit 4 and the like, and as the functions, for example, the function of the important word appearance frequency calculating means and the function of the important word appearance frequency information storing means. The function of the important word importance calculating means may be provided in the important word extracting unit 4 or the like. Further, in the present example, for example, the function of the inter-word degree-of-association information storage unit and the function of the important word-related word acquisition unit that acquires other words related to the important word extracted based on the information of the degree of association between words Can be provided in the important word extraction unit 4 or the like.

Here, the configuration of the language processing apparatus and the like according to the present invention is not necessarily limited to the one shown above,
Various configurations may be used. Further, the application fields of the present invention are not necessarily limited to those shown above, and the present invention can be applied to various fields.

As various processes performed in the language processing apparatus according to the present invention, the processor executes a control program stored in a ROM (Read Only Memory) in a hardware resource including a processor and a memory, for example. A configuration controlled by doing so may be used, and for example, each functional unit for executing the processing may be configured as an independent hardware circuit. The present invention also relates to a floppy (registered trademark) disc or a CD (Compact Disc) storing the above control program.
It can be understood as a computer-readable recording medium such as a ROM or the program (itself), and the processing according to the present invention is performed by inputting the control program into the computer from the recording medium and causing the processor to execute the control program. Can be made.

[0071]

As described above, according to the language processing apparatus of the present invention, for example, past text information is stored, for example, current text information is acquired, and the acquired text information and the stored text information are stored. Since the key word is extracted from the detected character string part by detecting the matching character string part, even when the stored text information or the acquired text information contains an error or ambiguity, , It is possible to extract appropriate important words.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration example of a dialogue processing apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a procedure of processing performed by a character string comparison unit.

FIG. 3 is a diagram showing a schematic configuration example of a language processing device according to the present invention.

[Explanation of symbols]

1 ... Dialogue input unit, 2 ... Dialogue recording unit, 3 ... Dialogue comparison unit, 4 ... Important word extraction unit, 11 ... Microphone device, 12 ... Dictation processing unit, 13 ... Dialogue recording device, 14 .. Input data storage unit, 15 ... Character string comparison unit, 16 ... Word dictionary unit, 17 ... Morphological analysis unit, 18 ... Important word extraction unit, 21 ... Dialog recording data, 22 ... Important word , 31 ... text information storage unit, 32 ... text information acquisition unit, 33 ... matching character string portion detection unit, 34 ... important word extraction unit, 41 ... text information,

Continued front page    F term (reference) 5B091 AA15 BA02 CA02 CB12 CC01                       CC15                 5D015 KK02

Claims (17)

[Claims] 1. A text information storage means for storing text information, a text information acquisition means for acquiring text information, text information acquired by the text information acquisition means, and text information stored in the text information storage means. A language comprising: a matching character string portion detecting means for detecting a matching character string portion; and an important word extracting means for extracting an important word from the character string portion detected by the matching character string portion detecting means. Processing equipment. 2. The language processing apparatus according to claim 1, wherein the text information stored in the text information storage means is text information obtained by voice recognition of voice information. apparatus. 3. The language processing apparatus according to claim 1 or 2, wherein the text information acquisition means recognizes voice information input means for inputting voice information and voice information input by the voice information input means. And a speech recognition means for converting the text information into text information, and a language processing device. 4. The language processing device according to claim 1, wherein the text information stored in the text information storage means and the text information acquired by the text information acquisition means are related to each other. A language processing device having the following contents. 5. The language processing apparatus according to claim 3, wherein the voice information input means of the text information acquisition means inputs voice information currently issued by a dialogue performed by two or more persons, and stores it in the text information storage means. A language processing device characterized in that the stored text information is text information obtained by voice recognition of voice information issued in the past in the dialogue. 6. The language processing apparatus according to claim 5, wherein the text information storage means deletes the stored text information from the stored contents when a predetermined period has elapsed. apparatus. 7. The language processing apparatus according to claim 5, wherein dialogue is performed based on voice information input by voice information input means of the text information acquisition means or text information converted by voice recognition means of the text information acquisition means. And a text information storage means for deleting the stored text information from the stored contents when the topic conversion is detected by the topic conversion detection means. Language processing device. 8. The language processing device according to claim 1, wherein the text information acquisition means acquires text information currently issued by a dialogue performed by two or more people on the network using the text information, and the text information is acquired. The language processing device, wherein the text information stored in the storage means is text information issued in the past in the dialogue. 9. The language processing device according to claim 1, wherein the important word extraction means performs morphological analysis on the character string portion detected by the matched character string portion detection means to perform part-of-speech. It is configured by using a morpheme analysis unit that acquires a word list with information, and a predetermined part-of-speech word extraction unit that extracts a word of a predetermined part-of-speech as an important word from the word list acquired by the morpheme analysis unit, A language processing device characterized by. 10. The language processing device according to claim 1, wherein the important word extraction means is an important word candidate word storage means for storing words that are candidates to be extracted as important words. A matching word extraction unit that extracts a word that matches a word stored in the important word candidate word storage unit from the character string portion detected by the matching character string portion detection unit as an important word, A language processing device characterized by the above. 11. The method according to any one of claims 1 to 10.
In the language processing device described in the paragraph 1, the important word extracting means extracts the important word from the character string portion detected by the matching character string portion detecting means, excluding words satisfying a predetermined condition. Processing equipment. 12. The method according to any one of claims 1 to 11.
The language processing device as described in the paragraph (1), further comprising an importance level assigning unit that assigns an importance level to the important word extracted by the important word extracting unit. 13. The language processing apparatus according to claim 12, wherein the importance degree assigning means calculates an appearance frequency of the important word extracted by the important word extracting means, and an important word appearance frequency calculating means.
An important word appearance frequency information storage unit that stores the important word extracted by the important word extraction unit and the information of the appearance frequency of the important word in association with each other, and the importance degree of the important word based on the appearance frequency of the important word. A language processing device comprising: an important word importance calculating means for calculating. 14. The method according to any one of claims 1 to 13.
In the language processing device described in the paragraph (1), an inter-word relevance information storage unit that stores information about relevance between words, and an important word extracted by the important word extraction unit based on the stored contents of the inter-word relevance information storage unit. A language processing apparatus, comprising: an important word-related word acquiring unit that acquires another word related to a word. 15. A method of acquiring text information, detecting a character string portion that matches the acquired text information and the text information stored in the storage means, and extracting an important word from the detected character string portion. Language processing method. 16. A function of acquiring text information, a function of detecting a character string portion that matches the acquired text information and the text information stored in the memory, and a function of extracting an important word from the detected character string portion. And a program that realizes 17. A storage medium, which stores a program to be executed by a computer so that the program can be read by an input means of the computer, the program includes a process of acquiring text information, the acquired text information, and text information stored in a memory. A storage medium characterized by causing the computer to execute a process of detecting a character string portion that matches with and a process of extracting an important word from the detected character string portion.