JP2003186872A - Language processing system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide such a system that situation information acquired by a language processing function in the course of the language processing can be used by other language processing function, in a language processing system having the plurality of language processing functions. <P>SOLUTION: This system, as its feature, has a first language processing means for performing first language processing for input data; a situation information management means for receiving the situation information acquired or generated in the course of the language processing performed with the first language processing means, and is a feature of the input data that the first language processing has processed from the first language processing means, and for transmitting the received situation information in response to a capture request from other language processing means; and a second language processing means for receiving the situation information by making the capture request for the situation information to the situation information management means, and for performing a second language processing in which the situation information is reflected. <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 system, and more particularly to a language processing system that mutually utilizes a plurality of language processing functions that can operate on separate computers.

Here, language processing means machine translation, automatic abstraction, information retrieval, kana-kanji conversion, voice recognition, character recognition (O
(CR), OCR function, full-text search, etc., is a generic term for information conversion and information recognition functions based on a plurality of language processes.

[0003]

2. Description of the Related Art In recent years, with the development of high speed and large capacity computer networks, various information systems sharing computer resources such as storage resources, CPU resources and device resources distributed on the computer network have been put into practical use. I'm out. In the future, heterogeneous computers such as personal computers, engineering workstations, and supercomputers, and heterogeneous operating systems will be connected to computer networks, and it will be possible to organically combine and use multiple information processing functions that operate on different computers. It is expected to become popular.

The advantage realized as a system operating in such a distributed computer environment is that it is easy to organically combine the information processing functions operating in different systems, and a new information processing application system can be realized. is there. Another major advantage is that it eliminates the labor and cost of porting the exact same functions to run on other computer systems or operating systems.

In such a situation, in the conventional language processing (function) application system of the type described in the present invention, since most of the systems were operated by a single computer, a computer system or an operating system that can be used. There was a problem that the system was limited. In addition, even in a system compatible with a distributed computer environment, there is a system in which only the graphical user interface part is separated as a network compatible component so that it can be operated on another computer.

However, the processing unit related to the core of language processing is closed in one computer system and has not reached the stage of organically connecting another language processing system.

By operating on different computers via a computer network, it is expected that the cost and labor of system development will be greatly reduced.

On the other hand, a language processing system such as machine translation, kana-kanji conversion, automatic abstraction, and document retrieval has heretofore been realized as an individual processing system, and it is necessary to individually manage dictionary information and user customization information. It was For example, taking an English-Japanese machine translation system having a post editor as an example, this language processing application system has a Kana-Kanji conversion system and an English-Japanese machine translation system as a plurality of language processing (function) systems. In this type of conventional system, even if a language-translated word pair is registered in the dictionary of the translation system, it is not registered in the dictionary of the Kana-Kanji conversion system. For this reason, the user needs to newly register the same word in the dictionary of the Kana-Kanji conversion system, which takes a lot of time and effort.

That is, in a language processing application system using a plurality of language processing systems, there is a problem that the functions cannot be sufficiently exhibited only by operating each language processing system separately.

[0010]

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and each language processing (system) can be operated by a different computer system, and the plurality of language processings can be performed. In the combined / used language processing system, the purpose is to improve the performance of each language processing function by sharing status information such as user customization information, dictionary information, and field information between language processings.

[0011]

According to the present invention, in a language processing system having two or more language processings, the situation information obtained in the first language processing step is used in the second language processing step. A language processing method and device characterized by the above. Here, the situation information is intermediate data acquired or generated in the language processing process, and is obtained by digitizing or symbolizing some of the characteristics of the input data processed by the language processing system.

[0012] Specifically, in a language processing system having a plurality of language processing means for performing a plurality of language processing, a first language processing means for performing a first language processing on input data; The information acquired or generated in the course of the language processing by the first language processing means, and the status information which is the characteristic of the input data processed by the first language processing means, is used as the first language processing means. And a status information management means for transmitting the received status information in response to a loading request from another language processing means, and sending a status information loading request to the status information management means and receiving the status information. However, the second language processing means for performing the second language processing reflecting the situation information is provided.

Further, when the status information having the same characteristics as the received status information is already stored in the storage means, the status information management means uses the low priority language processing means based on the predetermined priority order. It is characterized in that the generated status information is deleted.

A language processing system for executing machine translation processing as the first language processing and voice recognition processing as the second language processing, wherein the first language processing means is the machine. The translated word information obtained in the translation process in the translation process is transmitted to the situation information managing means as situation information, and the second language processing means receives the translated word information which is the situation information from the situation information managing means, It is characterized in that the voice recognition processing is performed by limiting the vocabulary to be recognized based on the received translated word information.

A language processing system for executing full-text search processing as the first language processing and character recognition processing as the second language processing, wherein the first language processing means is the full-text processing. The field information of the stored document obtained from the document database in the search process is transmitted to the status information management unit as status information, and the second language processing unit is the field that is the status information from the status information management unit. Character recognition processing is performed by receiving information and switching the recognition dictionary based on the received field information.

A language processing system for executing full-text search processing as the first language processing and machine translation processing as the second language processing, wherein the first language processing means is the full-text processing. The field information of the stored document obtained from the document database in the search process is transmitted to the status information management unit as status information, and the second language processing unit is the field that is the status information from the status information management unit. It is characterized in that information is received, and the machine translation processing is performed by switching the translation dictionary based on the received field information.

As described above, according to the present invention, the situation information obtained in one language process can be shared among a plurality of language processes, so that the performance of those language processes can be remarkably improved. As a result, it is possible to easily realize a high-performance and easy-to-use language processing system.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (Embodiment 1) An embodiment of the present invention will be described with reference to the drawings.

FIG. 1 illustrates the functional blocks of an embodiment of the present invention. Language processing server (1) 104 and language processing server (2) 1 that perform different language processing (functions)
05, the language processing server (3) 106, the status information storage unit 103 that stores status information, the status information management unit 102 that controls updating of the status information, and the language processing server to control the overall language processing application. It is composed of a language processing application control unit 101 that constitutes it. In the figure, thick lines represent data lines and thin lines represent control lines.

FIG. 2 shows an example of a device configuration for configuring the present invention. A plurality of computers are connected via a computer network.

The language processing application control unit 101 is a computer (1) 201, a language processing server (1) 104,
Language processing server (2) 105, language processing server (3) 1
Assuming that 06 is operating on the computer (2) 202, the language processing application is
The respective functions of the language processing server operating in No. 2 are properly used to realize a given function.

FIG. 3 is a diagram expressing the format of the status information sent from the language processing server to the status information management unit 102 as status information. The status information is the name of the language processing application using the language processing server that generated the status information,
User name using the language processing application,
It consists of the name of the language processing server itself and the status information body. The situation information management unit 102 stores / manages in the situation information storage unit 103, with one block as a unit of situation information. The status information is intermediate data acquired or generated by the language processing server during the language processing process, and defines a part of the characteristics of the input data processed by the language processing server, which is digitized or symbolized. I am trying.

FIG. 4 shows an example of a procedure when a language processing server sends status information to the status information management unit 102. On the other hand, FIG. 5 shows an example of a procedure when a language processing server fetches status information from the status information management unit 102.

When the language processing server sends the status information to the status information management unit 102, the language processing server first writes the status information to the status information management unit 102. When the writing is successful, the status information management unit 102 notifies the language processing server that the writing is successful, and finishes the writing process (“writing status information 1” and “successful writing 1” in FIG. 4 correspond to this procedure. ).

When writing fails due to a lack of storage area in the status information storage unit 103, the status information management unit 102 informs the language processing server that writing has failed. At this time, the language processing server resends the status information (“status information writing 2”, “writing failure 2”, and “status information resending 2” in FIG. 4 correspond to this procedure).

The procedure by which the language processing server fetches the status information from the status information management unit 102 is as shown in FIG.
First, when the status information management unit 102 receives new status information, the status information management unit 102 broadcasts that the status information has arrived to all connected language processing servers.
Next, the language processing server issues a request for status information fetching, and in response thereto, the status information managing unit 102 transfers the status information to the language processing server (“status information arrival notification 1” and “status information fetching request in FIG. 5”). 1 "and" acquisition of status information 1 "correspond to this procedure).

When the import fails due to the fact that the corresponding status information does not exist, the status information management unit 102 notifies the language processing server of the failure of importing the status information (“status information arrival notification 2” in FIG. 5). "Status information import request 2" and "Status information import failure 2" correspond to this procedure.

6 and 7 show the processing flow of the status information management unit 102. 6 and 7, the situation information storage unit 102
Is a process assuming that the queue structure is. Therefore, the latest status information is sequentially arranged at the beginning and the old status information is arranged at the end. When there is no free area in the status information storage unit 102, the oldest status information is deleted so that the writing fails.

When the status information is received, the processing shown in FIG. 6 is started. First, it is checked whether there is a space in the status information storage unit 103 (501). If there is no space, the oldest status information is deleted. Yes (502). Then, the received status information is written in the status information storage unit 103 (503). Next, inform the language processing server that wrote the status information that the writing was successful (504).
At the same time, it broadcasts that new status information has arrived (505).

When a request for importing status information is received, the processing of FIG. 7 is started, and the status information management unit 102 selects from the status information stored in the status information storage unit 103,
The status information has corresponding fields that match the names of the user of the language processing server that requested the acquisition of status information and the language processing application. It is checked whether or not the latest status information different from the name exists (701).

If found, the status information is transferred to the language processing server that has made the request (702). On the other hand, if it is not found, the language processing server is notified that the import has failed, and the processing ends (703).
8 and 9 show a part of the processing of the language processing server relating to the delivery of status information.

Processes other than those shown are not shown because they are unique to the function of the language processing server and are not the subject of the present invention.

When writing the status information in FIG. 8, the status information to be written first is the status information management unit 102.
(801). After the transfer, the status information management unit 102
The process waits for a response from (802), and if the writing is successful (803), the process ends. On the other hand, if the writing has failed (803), the status information is retransmitted.

The status information fetching shown in FIG. 9 is activated when a status information arrival notification is received from the status information management unit 102. In this process, first, a status information acquisition request is notified to the status information management unit 102 (90
1) Wait for a response from the status information management unit 102 (90
2). When the status information is sent from the status information management unit 102 (903), the status information is reflected in its own language processing (904). On the other hand, if the acquisition of the status information has failed (903), the process is terminated.

In this embodiment, it is assumed that the status information block shown in FIG. 3 is stored in the queue structure. However, it is also possible to provide a newly stored time field in the status information block. In this case, it is not necessary to store in the queue structure.

When the status information is managed by time, the method of deleting the status information in the process in which the status information management unit 102 receives the status information from the language processing server can be modified as follows. 1. 1. Delete the status information with the oldest stored time 2. Delete status information after a certain amount of time

In the case of 2, the status information is stored in the status information storage unit 10.
There is not always a vacant space when writing to No. 3. Therefore, when the writing fails, the notification needs to be sent to the language processing server.

Further, in the status information writing process of the language processing server shown in FIG. 8, the status information is retransmitted many times until the writing is successful, but this process also causes a deadlock. In order to avoid this, when the retransmission is performed a prescribed number of times, it is possible to modify so that the processing is terminated.

In the embodiment, the situation information management section for managing the situation information is provided, but it is easy to realize a modification in which the situation information is directly exchanged between the language processing servers.

(Second Embodiment) In the first embodiment, the language processing application is not limited to a specific language processing, but in the second and subsequent embodiments, a more specific embodiment will be shown and the status information will be described. Will also be specifically described.

FIG. 10 shows a voice dialogue system which is an embodiment of the present invention. In this embodiment, a speech recognition server 604 and a speech synthesis server 60 are used as language processing servers.
5, and a control unit 6 for controlling the entire voice dialogue system
01, status information management unit 602, status information storage unit 603,
A / D converter 606 that converts a voice analog signal input from a microphone into a digital signal, D / A that converts a digital signal from the voice synthesis server into a voice analog signal
It is composed of a converter 607.

The essential part of the processing for speech recognition and speech synthesis is not the gist of the present invention. For example, as for voice recognition, reference 1 (“TOSBURG II free-speaker free speech dialogue system-use of multimodal response and voice response cancellation-”, Yoichi Takebayashi et al.,
IEICE Transactions (A). Vol. J77-
A, No. 2, pp. For the method including the process of measuring the power of the voice input signal disclosed in H.241-250) and the like, and for the voice synthesis, the method capable of controlling the power of the voice synthesis disclosed in Document 1 is used. You can In addition, the voice dialogue system control unit 60
Also for No. 1, the method disclosed in Document 1 is used. Here, the description will be limited to the process of fetching the situation information and the use thereof, which is the main purpose of the present invention.

11 and 12 show the flow of processing of each language processing server for the transmission and reception of status information in this embodiment. FIG. 11 shows processing on the voice recognition side, and FIG. 12 shows processing on the voice synthesis side.

In the voice recognition server 604, an A / D converter 606 is input from the microphone 608 for voice recognition processing.
The power of the sound volume of the input signal converted by is measured, and the time when the power exceeds a given power is regarded as the start point of the voice input. Also, the time when the power becomes smaller than the given power is detected as the end of the voice input.

Then, recognition processing is performed on the voice section between the start end and the end. The flow charts shown in FIGS. 11 and 12 show a process for transmitting the power of background noise to the situation information management unit 602 as situation information, which is activated when the start edge is detected. That is, a process of transferring the average of the input power for a given section immediately before the start of the voice section to the situation information management unit 602 as the background noise power is performed.

Specifically, first, the average of the input powers for a given section immediately before the beginning of the speech section is obtained as background noise power, and this value is transferred to the status information management unit 602 as status information. (1001). Next, the response from the status information management unit 602 is waited for (1002), and if the writing is successful (1003), the process is ended.

When the voice synthesis server receives the status information arrival notice, the process shown in FIG. 12 is performed as an interrupt process. That is, after notifying the situation information management unit of the situation information acquisition request (1004), it waits for a response from the situation information management unit (1005). When the capture is successful (1006), it is determined that the captured situation information is background noise by checking the language processing phase field provided in the situation information. If it is background noise (1007), the power of voice synthesis is changed according to the size of the background noise (1008), and the process is ended.

FIG. 13 shows an example of status information written in the status information storage section by the voice recognition server.

In the figure, "sumita" is the user name in the user name field, and "voice IF" is displayed.
Is displayed in the field of the language processing application name as the language processing application name.
Is described in the language processing server name field as the language processing server name, “background Noise” is described in the language processing phase field as the language processing phase, and “60” is described in the situation information body field. This is because a user "sumita" uses a language processing server "voice Recog" in a language processing application "voice IF", and the "voice Rec" is used.
A language processing server called "og" uses "6" as background noise.
It means that the value of "0" is taken in.

In the present embodiment, the voice recognition server 604 takes in the power of background noise as situation information,
When background noise is large, the volume of voice synthesis can be increased. As a result, it is possible to prevent the background noise from being large and difficult to hear, and it is possible to construct a voice dialogue system with good intelligibility.

In this embodiment, the power of background noise is taken in as situation information. However, assuming a voice dialogue system assuming a telephone line, the power of voice input is modified as situation information to be written in the situation information management unit 602. It is possible to In this case, when the voice input server 605 has a large voice input sound volume, the voice synthesis sound can be reduced.

In voice recognition, when a large voice is generated,
The phonological characteristics change, and the recognition rate tends to decrease. By carrying out the above-described modification, a voice speaker can be expected to have a psychological effect of controlling the loudness of the voice, and thus, the recognition rate of voice recognition can be improved.

The words output by the speech synthesis server 605 are
The situation information can be modified so that it is managed by the situation information management unit 602. Under this modification, it is possible to modify the process so that the speech recognition server gives priority to a word synthesized by speech as a word at the time of recognition. Even in human-to-human dialogue, there are many cases in which the words said by the other party occur repeatedly. By performing such a modification, the voice recognition rate in the voice dialogue system is improved.

(Third Embodiment) In a third embodiment, a voice interactive machine translation system using two language processing servers, a voice recognition server and a machine translation system, will be described.

FIG. 14 illustrates the functional configuration of the voice interactive machine translation system of this embodiment. A machine translation system control unit 1101 that controls the entire application system,
The status information management unit 1102 manages status information, the status information storage unit 1103 stores status information, the voice recognition server 1104 performs voice recognition processing, and the machine translation server 1105 performs machine translation processing. There is.

The processing of machine translation itself in the machine translation server is not the gist of the invention of the present application. For example, the existing technology disclosed in Japanese Patent Publication No. 3-10984 may be used.

Regarding the voice recognition process itself in the voice recognition server, the voice word recognition is performed based on the vocabulary set which is a recognition candidate, not in the gist of the invention of the present application. For example, the existing technique disclosed in Document 1 is used. You can divert it.
Therefore, the use of the situation information, which is the basis of the invention of the present application, will be mainly described below.

FIG. 15 illustrates a processing flow of the machine translation system control unit 1102 of the interactive machine translation system.

In the processing of the control unit 1102, first, one original sentence is extracted (1201). Next, the original sentence is transferred to the machine translation server for translation into the second language, and the result is received (1202). For example, as shown in FIG. 16, this result generally has a plurality of translated word candidates, and the first candidate word for each translated word is displayed in the first translation sentence presentation (1203). Next, the control unit 1102
Requests the voice recognition server 1104 for voice input, and waits for a voice command input from the user (1204). When a voice command is input, it is checked whether the voice input word is "OK" (1205). "O
In the case of “K”, it means that the user is satisfied with the presented translated sentence, and thus the process proceeds to the translation of the next original sentence (1206).

On the other hand, when a word other than "OK" is input by voice, the input of voice is awaited. When the input voice is the displayed translation word, the translation word candidate corresponding to the translation word is displayed. Furthermore, at that point, the input of a voice command is awaited (1207). When a voice is input, or when the input word is one of the displayed translation candidates (120
8) Then, the indicated translated word candidate is set as the first candidate, and the displayed translated sentence is rewritten (1209).

For example, when the result shown in FIG. 16 is received from the machine translation server, the first translated sentence displayed is "I bought a computer". In this situation, the flow of the process will be specifically described in the case where the user wants to replace the translation of “buy” with “buy”.

In this case, the message "I bought a computer" is displayed in step 1203. At this time, when the user voice-inputs the word "correction", it is found in step 1205 that the word is not "OK", so the translation word selection mode (step 1207 and thereafter) is entered. While waiting for voice input in step 1207,
When the user inputs "Bought", it is understood that it is the translated word displayed in step 1208. Therefore, in step 1209, the candidates for "Bought", that is, "Bought", "Purchased", and "Handed" are obtained. "Entered" is displayed. Then, in step 1210, the next voice command is waited for. When the user voice-inputs the word "purchased" there, after checking in step 1211, step 1
Since 212 is executed, a desired translated sentence such as "I purchased a computer" is obtained.

In the voice recognition processing of this embodiment, the translated word to be replaced is designated by voice. Therefore, it is necessary to recognize an appropriate word from a large vocabulary. However, in general, in speech recognition, the recognition performance tends to deteriorate as the vocabulary to be recognized becomes large. In this embodiment, the situation information management unit 1102 can limit the vocabulary at the time of recognition to a small vocabulary.

17 and 18 show the flow of processing relating to the exchange of status information. The machine translation side is shown in FIG. 17, and the voice recognition side is shown in FIG.

The machine translation server starts the process shown in FIG. 17 when the translated text is obtained. That is, first, all the translated word candidates in the translated sentence are transferred to the status information management unit 1102 as status information (1301). Then, it waits for a response from the status information management unit 1102 (130
2) When the response that the writing is successful is received (1303), the processing is ended.

On the other hand, on the side of the voice recognition server (1104), when the status information arrival notification is received from the status information management unit 1102, the processing of FIG. 18 is performed as an interrupt processing. That is, after notifying the status information management unit 1102 of the status information acquisition request (1304), it waits for a response from the status information management unit 1103 (1305).

If the import is successful (1306), it is determined that the imported status information is a translation word candidate by checking the language processing phase field provided in the status information (1307). If it is a translation candidate,
The recognition vocabulary is limited to the translated word candidates and "OK" and "correction" (1308), and the process is ended.

Since the vocabulary for speech recognition is limited to a small vocabulary set of translated words + α, highly accurate speech recognition is possible. Furthermore, the vocabulary set is adaptively changed each time the translated sentence to be translated is changed, so that the user can virtually input a voice word with a very large vocabulary set, and It is possible to build a system that does not feel restrictions such as limitations.

Although the machine translation system for inputting voice commands is described in the present embodiment, the input means may be a keyboard. In English-Japanese machine translation, the translated word is displayed in Japanese. When correcting the displayed word, the user inputs it by kana-kanji conversion. In a machine translation system having a kana-kanji conversion server and a machine translation server that can be easily realized by modifying this embodiment, conversion accuracy is significantly improved by limiting conversion candidates in kana-kanji conversion processing. become.

(Fourth Embodiment) In the fourth embodiment, a full-text document search system having a document reader function, which has a character recognition server and a full-text document search server, will be described. In this embodiment, the situation information is field information obtained by analyzing a document database, and this field information is used for improving accuracy in character recognition.

The functional configuration of this embodiment is shown in FIG.
Overall control unit 1501, status information management unit 1502, status information storage unit 1503, full-text document search server 1504, character recognition server 1505, document database storage unit 1506 for storing documents, documents stored in the document database storage unit It comprises a field identification unit 1507 that identifies the field from the database.

Character recognition processing in the character recognition server is as follows:
It is not the gist of the present invention. A commercial document reader already exists, and the character recognition processing that realizes it can be used. As an algorithm for performing character recognition, it is possible to use the technology disclosed in, for example, Document 2 (“Pattern recognition”, Taizo Iijima, Corona Co.). The character recognition process has a recognition dictionary for character recognition, and this recognition dictionary stores standard character pattern information. At the time of recognition, the character pattern information of the input character is compared and collated with the character pattern information of the standard character stored in the recognition dictionary, and the most similar character is set as the recognition result. In the present embodiment, in this character recognition, the recognition dictionary is divided and held for each field.

The flow of processing of the control unit 1501 is shown in FIG. When the control unit is activated, it waits for a command input from the user (1601). When a command is input from the user, it is first checked whether the command is a search execution command (1602). If it is a search execution command, it waits for the search condition input from the user (160
3). When the search condition is input, the full-text document search server is activated and a document corresponding to the search condition is searched from the document DB 1506 (1604). Then, the search result is presented to the user (1605).

On the other hand, if the command input in step 1601 is a document read command, step 1606
The step 1607 is executed by the check. In step 1607, the character recognition server is started,
Read the document. Then, the full-text document search server is activated, and the read document is stored in the document DB 1506 (1
608).

Further, when the command input in step 1601 is the end command (1609), a check is performed and the processing of the control unit is ended.

The processing flow of the full text document search server 1504 is shown in FIGS. The full-text document search server 1504 started in step 1604 shown in FIG.
Perform a document search. That is, a document corresponding to the search condition is searched from the document DB 1506 (1701), the search result is returned to the control unit 1501 (1702), and the process ends. The full-text search process in step 1701 is described in, for example, Document 5 (“Text Search Processor”, Tsunesuke Takahashi,
The processing disclosed in (Institute of Electronics, Information and Communication Engineers) and the like may be used.

The full-text document search server activated in step 1608 shown in FIG. 20 executes document storage in FIG. The document input first is the document DB 15
After storing in 06 (1703), the remainder obtained by dividing the stored document by a predetermined constant N is 0 (170
4), Step 1705 is executed. That is, the field identification unit 1507 is activated to identify the field of the document stored in the document DB 1506.

The field identified in step 1705 is transferred to the status information management unit 1502 as status information (1706), and in step 1707, a response from the status information management unit 1502 is waited for. In step 1708, it is checked that the writing has been successful, and the process ends.

The field identification unit 1507 identifies fields of documents contained in the document database. In order to identify this field, existing techniques related to document classification (for example, Document 3 (“Automatic Document Classification Method Based on Learning Result of Semantic Attribute”, Atsuo Kawai, Transactions of Information Processing Society of Japan, V
ol. 33, No. 9, pp. 1114-112
2, 1933) are available. That is, the field of each document stored in the document database is determined. Then, the field having the larger number of the field determination results for all the documents may be set as the field of the document database.

Further, for example, by the following equation,
The score for each field is calculated for each document, and D
A weighted average of the scores is obtained by weighting determined by the elapsed time stored in B, and the field j having the maximum value is set as the field of the target document DB. Σiwi * Pij (1) where Pij is the score for field j of document i. wi
Is a weight determined by the elapsed time from the storage of the document i in the DB, and is normalized so that ΣiΣj wi * Pij becomes 1.

FIG. 23 shows the flow of processing relating to the acquisition of status information in the character recognition section. When the status information arrival notice is detected, the illustrated process is started. Notify the status information management unit of the request to import status information (180
1) Wait for a response from the status information management unit 1502 (18
02). If the status information is successfully imported (180
3) Then, it is checked whether the acquired status information is the field information (1804), and if it is the field information, the recognition dictionary corresponding to the acquired field information is switched (1805). If the acquisition of status information fails, or if it is not field information, the processing ends.

In character recognition, the smaller the set of recognized characters, the more reduction in recognition error can be expected. In the present embodiment, it is possible to reduce the set of recognized characters by switching and using dictionaries for each field. As a result, a significant improvement in recognition performance can be expected.

(Fifth Embodiment) In the fifth embodiment, a full-text document search system for other languages having a document reader function, which has a character recognition server, a full-text document search server, and a machine translation server function will be described. In this embodiment, the situation information is field information acquired by the full-text document search server, and is used for improving the recognition accuracy in the character recognition server and the machine translation accuracy by this field information.

The functional configuration of the present embodiment is shown in FIG. 24 (whole) control unit 1901, status information management unit 1902, status information storage unit 1903, full-text document search server 1904,
Character recognition server 1905, machine translation server 1906, document database storage unit 1907 for storing documents, field identification unit 1908 for identifying the field from the document database stored in the document database storage unit, field for character recognition It includes a separate recognition dictionary 1909 and a field-specific translation dictionary 1910 for machine translation.

The flow of processing of the control unit 1901 is shown in FIG. When activated, the control unit 1901 waits for a command input from the user (2001). When a command is input from the user, it is first checked whether the command is a search execution command (2002). If the command is a search execution command, it waits for the search condition input from the user (2003). When the search condition is input, the full-text document search server is started, and the document corresponding to the search condition is searched in the document DB 1
Search from 907 (2004). Then, after the machine translation server is activated to translate the search result (2005), the translation result is presented to the user (2006).

On the other hand, if the command input in step 2001 is a document read command, step 2007
Step 2008 is executed by the check in. In step 2008, the character recognition server is started,
Read the document. Then, the full-text document search server is activated, and the read document is stored in the document DB 1907 (2
009).

Furthermore, when the command input in step 2001 is the end command, the processing of the control unit ends after the check in step 2010.

Since the processing of the full-text document search server is the same as that of the fourth embodiment, its explanation is omitted.

Since the processing of the character recognition server is also the same as that of the fourth embodiment, its explanation is omitted.

The processing of machine translation itself in the machine translation server is not the gist of the present invention. For example, the existing technology disclosed in Japanese Patent Publication No. 3-10984 may be used.

FIG. 26 shows a flow of processing relating to status information fetching processing by machine translation. When the status information arrival notice is detected, the illustrated process is started. The status information management unit is notified of the status information acquisition request (2101) and waits for a response from the status information management unit (2102). If the acquisition of the status information is successful (2103), it is checked whether the acquired status information is the field information (2104), and if it is the field information, the translation dictionary corresponding to the acquired field information is switched (2105). ).

In this embodiment, it is possible to improve the processing performance of each language processing server by utilizing the field information acquired by the full-text document search server in a plurality of other language processing servers.

(Embodiment 6) A sixth embodiment of the present invention will be described. In the sixth embodiment, the status information management unit 1902
In, the feature is that mediation of interaction between situation information is performed. For example, when a plurality of language processing servers acquire the same kind of status information and the status information conflicts with each other, this adjustment is performed.

FIG. 27 shows the status information management section 190 of this embodiment.
Of the two processes, the flow of the process when the status information is received is illustrated. This is a modification of the process shown in FIG. 6 in the first embodiment. The difference from FIG. 6 is that after the newly received status information is written in the status information storage unit (step 2203), the status information having the same value as the value of the field of the language processing phase of the written status information is written in step 2204. Check whether or not it exists in the status information storage section (user name and language application name are the same), and if status information with the same value exists,
In step 2206, the status information with lower priority is deleted.

In determining the priority of the status information, for example, the priorities of a plurality of language processing servers are given in advance as follows (the inequality sign indicates the priority). Full-text search server> Character recognition server> Machine translation server Due to this, when competing status information exists in the status information storage unit, the status information issued by the language processing server having a lower priority may be deleted.

In determining the priority of status information,
It is also possible to provide a field for describing the certainty in the block of the situation information shown in FIG. 3 and modify it so that the situation information to be deleted is determined according to the magnitude of the certainty. This accuracy can be obtained when the language processing server that writes the status information acquires the status information. For example, in the case of the fourth embodiment, when the full-text document search server obtains a field from the documents stored in the document DB, the field that maximizes Expression (1) is obtained. The maximum value of the value of the expression (1) tends to be small when the fields of the stored document are dispersed, and thus this value can be adopted as the accuracy.

Although the display unit for displaying the processing result is not shown or described in all the above-mentioned embodiments, each control unit (for example, the language processing application control unit 101, etc.)
Shall be connected to and displayed.

[0098]

According to the present invention, the situation information acquired by a certain language processing server can be used by another language processing server to significantly improve accuracy and performance.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a first embodiment.

FIG. 2 is a device configuration diagram of the first embodiment.

FIG. 3 is a data structure diagram in the situation information storage unit of the first embodiment.

FIG. 4 is a diagram illustrating a communication protocol (when writing status information) between a status information management unit and a language processing server according to the first embodiment.

FIG. 5 is a diagram illustrating a communication protocol (at the time of capturing status information) between a status information management unit and a language processing server according to the first embodiment.

FIG. 6 is a diagram showing a flow of processing of a situation information management unit of the first embodiment.

FIG. 7 is a diagram showing a processing flow of a situation information management unit of the first embodiment.

FIG. 8 is a diagram showing a flow of processing relating to exchange of status information in the language processing server according to the first embodiment.

FIG. 9 is a diagram showing a flow of processing relating to exchange of status information in the language processing server according to the first embodiment.

FIG. 10 is a functional block diagram of the second embodiment.

FIG. 11 is a diagram showing a flow of processing relating to the exchange of status information in the voice recognition server and the voice synthesis server according to the second embodiment.

FIG. 12 is a diagram showing a flow of processing relating to the exchange of status information in the voice recognition server and the voice synthesis server according to the second embodiment.

FIG. 13 is a diagram showing an example of situation information according to the second embodiment.

FIG. 14 is a functional block diagram of a third embodiment.

FIG. 15 is a diagram showing the flow of processing of the control unit in the third embodiment.

FIG. 16 is a diagram showing an example of translation word candidates according to the third embodiment.

FIG. 17 is a diagram showing the flow of processing relating to the exchange of status information in the machine translation server and the voice recognition server of the third embodiment.

FIG. 18 is a diagram showing the flow of processing relating to the exchange of status information in the machine translation server and the voice recognition server in the third embodiment.

FIG. 19 is a functional block diagram of the fourth embodiment.

FIG. 20 is a diagram showing a processing flow of a control unit of the fourth embodiment.

FIG. 21 is a diagram showing a processing flow of the full-text document search server according to the fourth embodiment.

FIG. 22 is a diagram showing a processing flow of the full-text document search server according to the fourth embodiment.

FIG. 23 is a diagram showing the flow of processing relating to the exchange of status information in the character recognition server of the fourth embodiment.

FIG. 24 is a functional block diagram of the fifth embodiment.

FIG. 25 is a diagram showing a processing flow of a control unit of the fifth embodiment.

FIG. 26 is a diagram showing the flow of processing relating to the exchange of status information in the machine translation server of the fifth embodiment.

FIG. 27 is a diagram showing a flow of processing regarding reception of status information in the status information management unit in the sixth embodiment.

[Explanation of symbols]

101 Language processing application control unit 102 Status Information Management Department 103 status information storage unit 104 Language Processing Server (1) 105 Language Processing Server (2) 106 Language Processing Server (3)

Continued front page    (72) Inventor Hiroshi Kanazawa             1st Komukai Toshiba-cho, Sachi-ku, Kawasaki-shi, Kanagawa             Inside the Toshiba Research and Development Center F term (reference) 5B009 KB00 LA01 ME12 VA02 VA09                       VC03                 5B091 CB12 CC01 CD03 CD15 EA17

Claims (5)

[Claims] 1. A language processing system having a plurality of language processing means for performing a plurality of language processing, comprising: first language processing means for performing a first language processing on input data; From the first language processing means, status information, which is information acquired or generated in the course of language processing by the language processing means and is characteristic of the input data processed by the first language processing means, is received. Then, the status information management means for transmitting the received status information in response to the acquisition request from another language processing means, and the status information management means is requested to acquire the status information, and the status information is received, The second reflecting this situation information
And a second language processing means for performing the language processing of the above. 2. The status information management means, when the status information having the same characteristics as the received status information is already stored in the storage means, based on a predetermined priority order, the language processing means with a low priority order. The language processing apparatus according to claim 1, wherein the generated status information is deleted. 3. A language processing system for executing machine translation processing as the first language processing and voice recognition processing as the second language processing, wherein the first language processing means is the machine. The translated word information obtained in the translation process in the translation process is transmitted to the situation information management means as situation information, and the second language processing means receives the translated word information which is the situation information from the situation information management means, The language processing system according to claim 1, wherein the speech recognition processing is performed by limiting a vocabulary to be recognized based on the received translated word information. 4. A language processing system for executing full-text search processing as the first language processing and character recognition processing as the second language processing, wherein the first language processing means comprises the full-text processing. Field information of the stored document obtained from the document database in the search processing is transmitted to the status information management means as status information, and the second language processing means is the status information field from the status information management means. 2. The language processing system according to claim 1, wherein information is received, and character recognition processing is performed by switching a recognition dictionary based on the received field information. 5. A language processing system for executing full-text search processing as the first language processing and machine translation processing as the second language processing, wherein the first language processing means is the full-text processing. Field information of the stored document obtained from the document database in the search processing is transmitted to the status information management means as status information, and the second language processing means is the status information field from the status information management means. The language processing system according to claim 1, wherein the language processing system receives information and performs machine translation processing by switching a translation dictionary based on the received field information.