JPH06332945A - Device and method for processing natural language - Google Patents

Abstract

(57) [Summary] [Purpose] To provide a natural language interface that judges the easiness of listening to a response sentence, and modifies the response sentence to make it easier for the user to output. [Structure] An input unit 101 for inputting a question sentence in a natural language, a sentence analysis unit 102 for analyzing the input question sentence, and a search unit 104 for searching information based on the analysis result in a natural language interface device. , A response sentence generation unit 106 that generates a response sentence in natural language from the search result
And a determination unit 1 that determines the easiness of listening to the generated response sentence.
07, a response sentence changing unit 112 that corrects the response sentence generated by the generating unit 106 based on the determination result, and a synthetic voice output unit 114 that outputs the corrected response sentence.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing device such as a natural language interface device for inputting a sentence in natural language, searching a database and outputting the result as synthetic speech.

[0002]

2. Description of the Related Art In recent years, a natural language interface device has been researched and developed which retrieves information from a database based on an input sentence in natural language and outputs the result as synthetic speech. The configuration and operation of such a conventional natural language interface device will be described with reference to FIGS.

FIG. 3 is a diagram showing the configuration of a conventional natural language interface device. In the figure, 301 is an input unit for receiving a question in a natural language sentence from a user. A sentence analysis unit 302 analyzes the natural language sentence obtained from the input unit 301 and analyzes the meaning of the sentence. A dialogue management unit 303 inquires of the user's question content obtained from the sentence analysis unit 302 to the database search unit 304, stores an answer to the inquiry as a history of topics, and sends it to the response sentence generation unit 306. Three
A database search unit 04 receives the user's question content from the dialogue management unit 303 and searches the database 305 for data corresponding to the question content. Thirty
5 is a database.

Reference numeral 306 is a response sentence generation unit for generating a response sentence, which is an answer to the user's question received from the dialogue management unit 303, together with its reading and accent information. Reference numeral 307 is an acoustic processing unit that generates a synthetic speech waveform from the sentence information generated by the response sentence generation unit 306.
A voice output unit 308 outputs the voice waveform generated by the sound processing unit 307 to a speaker and outputs a synthetic voice.

FIG. 4 is a flow chart showing a process in which the user interacts with the natural language interface device in the configuration of FIG. First, the input sentence receiving process step S12
Received the question text in natural language entered by the user at
It is passed to the sentence analysis processing step S13. In the sentence analysis processing step S13, the meaning of the natural language sentence is analyzed, and the content of the analyzed question of the user is passed to the database search processing step S14. In the database search processing step S14, appropriate data is selected with reference to the database 305, and is transferred to the topic management processing step S15.

The topic management processing step S15 is a question of the user at that time and the database search processing step S1.
The data obtained by the step 4 together with the data targeted by the question and answer are used as the history of the topic, and the dialogue management unit 3
Save in 03. This information is used as a context when the content of the question of the next user is analyzed, and the sentence analysis processing step S1.
Used in 3. The topic management processing step S15 is
After holding the history of topics, information of data as an answer is passed to the response sentence generation processing step S16. The response sentence generation processing step S16 is the topic management processing step S15.
A response sentence is generated on the basis of the data obtained from, and the acoustic processing step S1 together with the reading and accent information.
Pass to 7.

In the sound processing step S17, a voice waveform is generated from the sentence passed from the response sentence generation processing step S16, its reading and accent information, and passed to the synthetic voice output processing step S18. Synthetic voice output processing step S18
Outputs the passed voice waveform to the speaker as voice. Then, in step S19, it is determined whether or not the question answering is ended, and if it is ended, the processing is ended. If not ended, the process returns to the input receiving process of step S12. This repetitive process is a process flow of a process of answering a question by the natural language interface device.

[0008]

However, in the above-mentioned conventional example, the natural language interface device returns the response to the user by synthesizing the natural language response sentence generated based on the data retrieved from the database as it is. Since it outputs the output, there are the following drawbacks when the user hears the response from the system as a voice. (1) It is difficult to judge which part of the information presented by the system is important when heard by voice. (2) When the readings of a plurality of information that can be presented by the system are similar, it is difficult to distinguish the voices that are monotonically synthesized. (3) If the response sentence is always returned at the same speech rate, the user may be annoyed. (4) If it takes time to search the database, the system shuts down during that time, so that there is a time to make the user anxious.

[0009]

In order to solve the above-mentioned problems, a natural language processing apparatus of the present invention analyzes an input means for inputting a question text in natural language and a question text input by the input means. Analysis means, search means for searching information based on the analysis result by the analysis means, generation means for generating a response sentence in natural language from the search result by the search means, and response sentence generated by the generation means Judgment means for judging easiness of grasping, correction means for correcting the response sentence generated by the generation means based on the judgment result of the determination means, and output means for outputting the response sentence corrected by the correction means And with.

According to another aspect of the present invention, in the natural language processing method, an input step of inputting a question sentence in natural language, an analysis step of analyzing the question sentence input in the input step, A search step of searching information based on the analysis result of the analysis step, a generation step of generating a response sentence in natural language from the search result of the search step, and an easy grasp of the response sentence generated by the generation step. The determination step includes a determination step, a correction step for correcting the response sentence generated by the generation step based on the determination result of the determination step, and an output step for outputting the response sentence corrected by the correction step.

[0011]

In the natural language processing apparatus, the search means searches for information based on the result of the analysis means analyzing the question sentence in the natural language input from the input means, and the search result is generated by the generating means. The easiness of grasping the response sentence in natural language is determined by the determination means, and the generated response sentence is corrected by the correction means based on the determination result and output from the output means.

In the natural language processing method, information is searched for in the search step based on the result of analysis of the question sentence in the natural language input in the input step in the analysis step, and the information is generated in the generation step from the search result. The easiness of grasping the response sentence in natural language is determined in the determination step, and the generated response sentence is corrected in the correction step based on the determination result and output in the output step.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an apparatus according to an embodiment of the present invention. Reference numeral 101 denotes an input unit that receives a question sentence in natural language from a user, and includes a keyboard for inputting characters, a microphone for inputting voice, a voice recognition device, and the like. A sentence analysis unit 102 analyzes the natural language sentence obtained from the input unit 101 and analyzes the meaning of the sentence. Reference numeral 103 is a dialogue management unit that inquires the database search unit 104 about the question content of the user obtained from the sentence analysis unit 102, saves the answer to the inquiry as a history of topics, and sends the response sentence generation unit 106. 1
A database search unit 04 receives the user's question content from the dialogue management unit 103 and searches the database 105 for data corresponding to the question content. 10
5 is a database.

Reference numeral 106 denotes a response sentence generation unit for generating, from the dialogue management unit 103, a response sentence serving as an answer to the user's question together with its reading and accent information. 10
Reference numeral 7 indicates the generated response sentence based on the information of the sentence generated by the response sentence generation unit 106, the content of the dialogue managed by the dialogue management unit 103, and the history information of the dialogue between the user and the system. This is an easiness of hearing determination unit that determines easiness of hearing when the user listens to the synthesized voice. The easiness of hearing determination unit 107 includes a focus determination unit 108, a similar reading determination unit 109, an input speed determination unit 110,
It is composed of one of the timers 111. In the present embodiment, a case where the easiness of listening determination unit 107 is composed of the focus determination unit 108 will be described below.

The focus determination unit 108 is a database 105.
When the data retrieved from is generated as a response sentence, which data is important to the user is determined by using the history information of the dialogue held in the dialogue management unit 103. The easiness of listening determination unit 107 configured by the focus determination unit 108 receives the important data portion as the determination result and the information of the response sentence generated by the response sentence generation unit 106 from the response sentence changing unit 1
Pass to 12. Based on the determination result and the information of the response sentence passed from the easiness of listening determination unit 107, the response sentence changing unit 112 makes the sentence easy to hear when the response sentence is speech-synthesized, that is, the important data in this case is easy to understand. It is a response sentence changing unit that changes the response sentence so that the sentence becomes easy to hear.

Reference numeral 113 is an acoustic processing unit for generating a synthetic speech waveform from the sentence information changed by the response sentence changing unit 112. Reference numeral 114 is a voice output unit that outputs the voice waveform generated by the sound processing unit 113 to a speaker and outputs a synthetic voice.

FIG. 2 is a flowchart showing a process in which the natural language interface device configured as shown in FIG. 1 interacts with a user. First, in the input sentence receiving process step S1, the question sentence in the natural language input by the user is received and passed to the sentence analyzing process step S2. The sentence analysis processing step S2 analyzes the meaning of the natural language sentence and passes the content of the user's question to the database search processing step S3. The database search processing step S3 selects appropriate data with reference to the database 105,
It is passed to the topic management processing step S4.

In the topic management processing step S4, the dialogue management unit 103 stores the question of the user at that time and the data obtained in the database search processing step S3, together with the data targeted by the question answer, as a history of the topic.
Save to. This information is used in the sentence analysis processing step S2 as the context when the content of the question of the next user is analyzed. In the topic management processing step S4, after the history of the topic is stored, information of data as an answer is transferred to the response sentence generation processing step S5. The response sentence generation processing step S5 generates a response sentence based on the data obtained from the topic management processing step S4, and sends the generated response sentence to the easiness of hearing determination processing step S6 together with the reading and accent information. hand over.

In the easiness of hearing judgment processing step S6, a response sentence is extracted from the history of the topic obtained in the topic management processing step S4, the response sentence generated in the response sentence generation processing step S6, and its reading and accent information. It is determined which part is important to the user, and the result is passed to step S7 as focused information. In step S7, it is determined whether or not the information passed from the listening comprehension determination processing step S6 includes focus information that is an improvement point. If there is focus information, that is, if there is an improvement, the response sentence and the reading and accent information are passed to the response sentence change processing step S8 together with the focus information. If there is no improvement, the response sentence and the reading and accent information are passed to the acoustic processing step S9.

The response sentence change processing step S8 is determined by the determination of step S7, and the focus information in the response sentence, which is the information on the improvement point passed, and the response sentence, the reading and the accent information, are used as the focus in the response sentence. Generates a response sentence to which the accent of the word to which is applied is strengthened or information is added so that only that part is uttered slowly. Then, the response sentence is passed to the acoustic processing step S9. The sound processing step S9 generates a voice waveform from the sentence passed from the response sentence generation unit step S8, its reading and accent information,
The synthesized voice output processing is passed to step S10. In the synthesized voice output processing step S10, the delivered voice waveform is output to the speaker as voice. Then, in step S11, it is determined whether or not the question answering is ended, and if it is ended, the processing is ended. If not ended, step S12
Return to the input receiving process of. This repetitive process is a process flow of a process of answering a question by the natural language interface device.

In the natural language interface device of FIG. 1, the function of the sentence analysis unit 102 and the like stores a program corresponding to the flowchart of FIG.
It can also be realized by executing this program while the PU uses the work area on the memory.

FIGS. 5 and 6 show an example in which the natural language interface device having the configuration shown in FIG. In Figure 5, the question sentence of (User 1) "I want to know the hot springs around Izu that are effective for back pain."
On the other hand, the response sentence of (system 1) is generated based on the result of searching the database under the AND condition of the condition that the place is Izu and the condition that the effect is a hot spring with low back pain.
In this case, since the expression "per Izu" is used for the place, the easiness of hearing determination processing step S6
Determines that the condition of "hot spring" is the portion to be focused, and determines that the name of the hot spring shown by A and B in the figure is focused. As a result, the response sentence change processing step S8 adds the information indicating the strong accent, the information indicating that the speed is slowed down, or both of the information added to the reading of the names of the hot springs and the hotel of A and B. Generate a sentence. As a result, in the response of (system 1), the portions A and B are emphasized and speech synthesis is performed.

In addition, in FIG. 6, in response to the question sentence (user 2) "I want to know that it is effective for low back pain at a hot spring in Higashiizu, which can be reached from Tokyo immediately," the same response sentence (system 2) ) Is generated, but unlike the case of FIG. 5, since the expression “you can go from Tokyo in Higashiizu” is used, it is determined that the condition of the place should be focused,
A synthetic voice is generated by emphasizing the portion of the location "Atami" in C. By using the easiness of listening determination unit and the response sentence changing unit as described above, in the device of the present embodiment, it is possible to obtain an easy-to-hear synthetic voice in which a portion of important information in the response sentence is emphasized. became.

[Other Embodiments] In the above-described embodiment, the focus determination unit 108 is used as the audibility determination unit 107 to determine the easiness of hearing, but the present invention is not limited to this. As a reference, the similar reading determination unit 109 can be used. FIG. 7 is an example of a dialogue when the similar reading determination unit 109 is used. The response sentence of (system 3) is generated in response to the question (tell me the golf course in the Tama direction) from (user 3). In this case, the golf course "Higashitama Country Club" in the search result data is generated. We are responding to three things: "Tama Golf Club" and "Golf Course Minamitama".

Here, since the reading "Tama" is included in all of these three names, the similar reading determination unit 10
According to 9, it is determined that the reading part “Tama” is similar. If such a determination result is used, the response sentence changing unit reads the different reading portions with these three names with different strengths or speeds, and D
It becomes possible to perform speech synthesis by emphasizing the portions E, F and F. This allows the user to focus on similar readings.

Further, in the above embodiment, the easiness of listening is determined by using the similar reading determining unit 109 as the easiness of listening determining unit 107, but the present invention is not limited to this. The input speed determination unit 110 can be used. Depending on the user, it may be frustrating to perform question answering using synthetic voice at the same speed and strength. In the first place, short-minded users may get frustrated if they speak slowly.

In such a case, the input speed of the question sentence is
In case of keyboard input, the speed of the key input is measured, and in the case of voice input, the talking time of the user,
From the length of the sentence, it is possible to determine whether it is quick. If this condition is used, the result of the determination is that if the input speed is fast, the user who is originally short-minded or the user is irritated, and it is easier for the user to hear the response text It is possible to increase the vocalization speed of.

Further, in the above embodiment, the easiness of hearing is determined by using the input speed determination unit 110 as the easiness of listening determination unit 107. , The timer 111 can be used. When a question that requires a large amount of data retrieval is input, the system does not return the synthetic voice of the response for a while, which may cause anxiety to the user. In such a case, when the system cannot return a response for a certain period of time, the timer 111 can issue a response sentence for connecting the two.

FIG. 8 is an example of a dialogue when the timer 111 is used as the easiness of listening determination unit 107. (User 4)
While the response sentence of (system 5) is generated in response to the question, the timer detects that it takes a predetermined time from the input of the question. Then, in order to fill the time until the response, (system 4-1) or (system 4) prepared in advance
-2) and the response of (system 4-3) are generated. As a result, it is possible to realize an easy-to-listen dialogue without interruption of the conversation.

In the above-described embodiment, the easiness of listening determination unit 107 is used.
As the focus determination unit 108, the similar reading determination unit 109, the input speed determination unit 110, and the timer 111, the ease of listening is determined, but the present invention is not limited to this. The easiness of hearing can be determined by combining one or more of these determination units.

In the above embodiments, the case of voice output has been described, but the present invention is not limited to this, and even when a response sentence is displayed and output, the readability is determined in the same manner as the readability by determining the readability. Can also be created. That is, based on the determination result of the focus determination unit 108 or the similar reading determination unit 109, the portion to be emphasized is exactly as shown in FIG.
As shown in FIG. 7, it is possible to easily identify by underlining or changing the character color, character size, and brightness. Also, using the timer 111,
It is also possible to sequentially display response sentences as shown in FIG. 8 according to the passage of a predetermined time.

[0033]

As described above, by determining the easiness of grasping the response sentence and correcting the response sentence based on the determination result, there is an effect that the response sentence which the user can easily grasp is obtained.

In particular, the following effects can be obtained. (1) A response sentence in which an important part of the information that the system wants to convey is emphasized can be obtained. (2) In a plurality of answers having similar readings, a response in which different parts are emphasized is obtained so that the difference between them can be distinguished. (3) The speed of the output voice in the voice response can be changed according to the input speed of the question sentence from the user. (4) When it takes a long time for the system to search the database and return the result, by providing a response to fill the gap, seamless dialogue processing can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram of a natural language interface device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a processing procedure of the natural language interface device according to the embodiment of the present invention.

FIG. 3 is a block diagram of a conventional natural language interface device.

FIG. 4 is a flowchart showing a processing process of a conventional natural language interface device.

FIG. 5 is a diagram showing an example of interaction between a user and a system according to an embodiment.

FIG. 6 is a diagram showing an example of interaction between a user and a system according to an embodiment.

FIG. 7 is a diagram showing an example of interaction between a user and a system according to another embodiment.

FIG. 8 is a diagram showing an example of interaction between a user and a system according to another embodiment.

[Explanation of symbols]

 101 sentence input unit 102 sentence analysis unit 103 dialogue management unit 104 database search unit 105 database 106 response sentence generation unit 107 audibility determination unit 108 focus determination unit 109 similar reading determination unit 110 input speed determination unit 111 timer 112 response sentence change unit 113 sound processing unit 114 synthetic speech output unit

Claims (12)

[Claims] 1. An input unit for inputting a question sentence in natural language, an analyzing unit for analyzing the question sentence input by the input unit, and a searching unit for retrieving information based on an analysis result by the analyzing unit. Generating means for generating a response sentence in natural language from the search result of the searching means; determining means for determining easiness of grasping the response sentence generated by the generating means; and A natural language processing apparatus comprising: a correction unit that corrects a response sentence generated by the generation unit; and an output unit that outputs the response sentence corrected by the correction unit. 2. The output means outputs the response sentence by voice information, and the correction means corrects the accent, volume, or output speed of a part or all of the response sentence. The natural language processing device according to claim 1. 3. The storage device manages and saves the topic related to the question sentence and the response sentence based on the analysis result of the analysis device and the search result of the search device, and the determination device stores the contents of the storage device. Determining the important part in the response sentence generated by the generating means with reference, and modifying the important part determined by the determining means so as to be distinguishable from other parts in the response sentence. The natural language processing device according to claim 1. 4. The determining means determines the existence of similar portions in the response sentence generated by the generating means, and the correcting means can distinguish the similar portions determined by the determining means. The natural language processing apparatus according to claim 1, wherein the response sentence is modified so that 5. The determination means measures the search execution time by the search means to determine whether a predetermined time has passed, and the correction means determines that the predetermined time has passed by the determination means. The natural language processing device according to claim 1, wherein the response sentence is modified so that a predetermined response sentence is output prior to the response sentence generated by the generating means. 6. The determination means determines the input speed of the question sentence by the input means, and the correction means, based on the determination result of the determination means,
3. The natural language processing apparatus according to claim 2, wherein the response sentence is modified so that the voice output speed of the response sentence from the output unit is changed. 7. An input step of inputting a question text in natural language, an analysis step of analyzing the question text input in the input step, and a search step of searching information based on the analysis result of the analysis step. A generation step of generating a response sentence in natural language from the search result of the search step, a determination step of determining the easiness of grasping the response sentence generated by the generation step, and A natural language processing method comprising: a correction step of correcting a response sentence generated by the generation step; and an output step of outputting a response sentence corrected by the correction step. 8. In the output step, the response sentence is output by voice information, and in the correction step, a part or all of the response sentence is accented, volume, or
The natural language processing method according to claim 7, wherein the output speed is corrected. 9. A storage step of managing and storing topics related to the question text and the response text based on the analysis result of the analysis step and the search result of the search step, wherein the determination step is stored by the storage step. The important part in the response sentence generated by the generating step is referred to with reference to the content, and the correcting step is performed so that the important part determined by the determining means can be distinguished from other parts in the response sentence. The natural language processing method according to claim 7, which is modified. 10. The determining step determines the presence of similar parts in the response sentence generated by the generating step, and the correcting step can distinguish the similar parts determined by the determining step. 8. The natural language processing method according to claim 7, wherein the response sentence is modified so that 11. The determination step measures the search execution time by the search step to determine whether a predetermined time has passed, and the correction step determines whether the predetermined time has passed by the determination step. 8. The natural language processing method according to claim 7, wherein the response sentence is modified so that a predetermined response sentence is output prior to the response sentence generated by the generating step. 12. The determination step determines the input speed of the question sentence in the input step, and the correction step, based on the determination result of the determination step,
9. The natural language processing method according to claim 8, wherein the response sentence is modified so that the voice output speed of the response sentence from the output unit is changed.