JP2000331002A - Search device, search method, recording medium storing search control program - Google Patents

Abstract

(57) [Summary] [Problem] Even when a sentence containing ambiguous information is input, the input sentence can be accurately searched, and there is no sense of incongruity with the sentence input by the user Generate and output such a statement. SOLUTION: A CPU refers to a word information table stored in a word information table ROM and performs a search using an input word as a key (step S3). The CPU refers to the chain rule table stored in the chain rule table ROM to search for a chain pattern for the "type" of the input word (step S4), and further sets the first reference date and time from the corresponding interpretation rule. Update (Step S
5).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a search device, a search method, and a recording medium on which a search control program is recorded, in which a numerical range can be managed and searched for an ambiguous input.

[0002]

In recent years, speech recognition technology has entered the stage of practical use, and systems using speech recognition technology such as navigation devices have begun to be commercialized. Along with this, attempts have been made to create an electronic secretary that can answer and answer by voice.

[0003] This electronic secretary aims at interpreting schedule information and the like input by a user in voice, for example, to adjust the schedule and appropriately present the schedule to the user.

In general, when a user inputs a schedule, an appointment (reservation), or the like to a computer, the user is required to specifically indicate a date and time from the computer. This is because if you try to indicate with a word that has a numerically ambiguous meaning, the word may have a different numerical meaning for different people, and the computer cannot understand it correctly.

[0005] In this case, the user can easily understand when the calendar or the calendar of the day is visible on the screen, but is very difficult to use when there is no large screen capable of displaying such information on the hardware.

For example, the words "go golf on the weekend" and "stop at a camera shop before returning home" are natural for users, but specific dates and times are unknown for computers, so they are treated as schedule information. It is hard. Conversely, the terms such as "go golf at 6:00 am on September 19" and "depend on the camera shop today at 18:00" can not be said that the date and time have not been specifically determined in advance, so it is easy for the user It is difficult to use.

In the case of "going golf on the weekend", a sentence having an ambiguous meaning in the numerical range such as "weekend" is not simply scheduled data but merely memo data, regardless of the date and time. There is, however, no convenience because search, automatic notification using it, and duplication check of schedule cannot be performed.

It is also conceivable to replace ambiguous words and phrases input by a user with general concrete values that can be easily handled by the computer. For example, "Weekend" is replaced with "Friday to Sunday", and "Going golf on weekend" is treated as "Going golf from this Friday to Sunday this week".

[0009] However, even if this conversion result is presented to the user by means of a schedule overlap check, search output, automatic notification, etc., the sentence and expression input by the user are significantly different, and in some cases, the It may be inevitable that the user will feel great discomfort because it may be different from the calculated value.

The present invention has been made in view of such circumstances, and even when a sentence containing ambiguous information is input, it is possible to accurately search for the input sentence. It is another object of the present invention to provide a search device, a search method, and a recording medium that stores a search control program that generates and outputs a sentence that does not cause a sense of incongruity with a sentence input by a user.

[0011]

According to the present invention, there is provided a retrieval apparatus comprising: an input unit for inputting a sentence; and a numerical comparison indicating a numerical range of the entire sentence from each word constituting the sentence input to the input unit. A data indicating means for generating numerical data, and a word indicating a predetermined numerical range for a problem area targeted by the system among words constituting the sentence input to the input means, Output configuration data generating means for converting into numerical information indicating a numerical range and generating output configuration data composed of the numerical information and other words excluding those converted into the numerical information; and inputting to the input means A first storage unit that associates the numerical comparison data generated by the numerical comparison data generation unit with the output configuration data generated by the output configuration data generation unit in association with the sentence. Searching means for searching a sentence within the search range based on a sentence indicating the search range input to the input means, and generating a search result sentence indicating the selected sentence as a result of the search; Output means for outputting a search result sentence generated by the means, wherein the search means includes a numerical comparison data generated by the numerical comparison data generation means for a sentence indicating a search range inputted to the input means. For the data, a numerical comparison data having a numerical range overlapping is searched from the storage means, and among the output configuration data corresponding to the overlapping numerical comparison data, numerical information constituting the output configuration data is obtained. The above-mentioned problem is solved by converting the word into a word corresponding to the current situation of the user and generating output configuration data including the converted word as a search result sentence.

According to the above retrieval apparatus, the numerical comparison data having a numerical range overlapping with the numerical comparison data generated by the numerical comparison data generating means for the sentence indicating the search range input to the input means. From the storage means. Then, among the output configuration data corresponding to the overlapping numerical comparison data, the numerical information configuring the output configuration data is converted into a word corresponding to the current situation of the user,
Output composition data including the converted word is generated as a search result sentence and output.

According to the search method of the present invention, numerical comparison data indicating a numerical range of the entire sentence is generated from each word constituting the input sentence, and a corresponding one of the words constituting the input sentence is generated. A word indicating a predetermined numerical range with respect to a problem area targeted by the system is converted into numerical information indicating the predetermined numerical range, and other than the converted numerical information and those converted into the numerical information. Is generated and the first sentence is stored in the first storage unit in association with the input sentence, in which the generated numerical comparison data and the generated output configuration data are associated with each other. The first storage means searches the first storage means for numerical comparison data having a numerical range overlapping with the numerical comparison data generated for the sentence indicating the input search range, and corresponds to the overlapping numerical comparison data. You Of the output configuration data, the numerical information forming the output configuration data is converted into a word corresponding to the current situation of the user, and the output configuration data including the converted word is generated as a search result sentence. The above-described problem is solved by outputting the generated search result sentence.

According to the above retrieval method, for the input sentence, the generated numerical comparison data and the generated output configuration data are stored in the first storage means in association with each other. Then, the first storage means searches the first storage means for numerical comparison data whose numerical range overlaps with the numerical comparison data generated for the sentence indicating the input search range, and corresponds to the overlapping numerical comparison data. Of the output configuration data to be converted, the numerical information constituting the output configuration data is converted into a word corresponding to the current situation of the user, and the output configuration data including the converted word is generated as a search result sentence. Then, the generated search result sentence is output.

[0015] The recording medium storing the search control program according to the present invention generates numerical comparison data indicating a numerical range of the entire sentence from each word constituting the input sentence,
A word indicating a predetermined numerical range with respect to a problem area targeted by the system among words constituting the input sentence is converted into numerical information indicating the predetermined numerical range, and the converted numerical information is converted. And other words other than those converted to the numerical information are generated, and
For the input sentence, the generated numerical comparison data and the generated output configuration data are stored in the first storage unit in association with each other, and the sentence indicating the input search range is generated. The first storage means searches the first storage means for numerical comparison data having a numerical range overlapping with the numerical comparison data, and outputs the output configuration data among the output configuration data corresponding to the overlapping numerical comparison data. A search control program that converts the constituted numerical information into words corresponding to the current situation of the user, generates output configuration data including the converted words as a search result sentence, and outputs the generated search result sentence The recording solves the above-mentioned problem.

According to the recording medium on which the above-mentioned search control program is recorded, by installing this search control program in a control device such as a computer, the control device makes the generated numerical comparison data And the generated output configuration data are stored in the first storage means in association with each other. Then, the first storage means searches the first storage means for numerical comparison data whose numerical range overlaps with the numerical comparison data generated for the sentence indicating the input search range, and corresponds to the overlapping numerical comparison data. Of the output configuration data to be converted, the numerical information constituting the output configuration data is converted into a word corresponding to the current situation of the user, and the output configuration data including the converted word is generated as a search result sentence. Then, the generated search result sentence is output.

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

1. Configuration of Schedule Search Apparatus The present invention can be applied to, for example, the schedule search apparatus 1 having the configuration shown in FIG.

The schedule search device 1 includes a voice input unit 10 for recognizing an input voice and a control unit 20 for controlling a schedule search or the like based on the voice recognition result.
And an audio output unit 30 that outputs schedule search results and the like by audio.

The voice input unit 10 includes a microphone 11 for generating a voice signal from an input voice, and an analog / digital (A / D) converter 1 for converting an analog signal into a digital signal.
2, a voice recognition circuit 13 for performing voice recognition based on voice data, and a first RAM (Ra
ndom Access Memory) 14 and a first ROM (Read Only Memory) 15 for storing application programs
And a voice recognition data ROM 16 for storing recognition target words for voice recognition.

The microphone 11 converts an input voice into a voice signal, which is an electric signal, and supplies the voice signal to the A / D converter 12. The A / D converter 12 samples an audio signal, which is an analog signal from the microphone 11, in accordance with a predetermined sampling clock, and converts it into digital audio data by quantizing the audio signal. 13.

The speech recognition circuit 13 temporarily stores the speech data supplied from the A / D converter 12 in the first RAM 14 and uses the recognition target words stored in the speech recognition data ROM 16 to perform the first A predetermined process according to an application program stored in the ROM 15 is executed.

Here, the first RAM 14 stores voice data supplied from the A / D converter 12 via the voice recognition circuit 13, data necessary for the operation of the voice recognition circuit 13, and the like. . The first ROM 15 stores, for example, an application program for causing the voice recognition circuit 13 to perform voice recognition.

The voice recognition circuit 13 reads a plurality of recognition target words stored in the voice recognition data ROM 16 under the control of the control unit 20, and stores a dictionary (hereinafter, a recognition dictionary) composed of the recognition target words. create. Then, when receiving the voice data from the A / D converter 12, the voice recognition circuit 13 stores the voice data in the first RAM 14. When, for example, one word of voice data is stored in the first RAM 14, the voice recognition circuit 13 determines, for example, a sound diameter (frequency), a strength (amplitude), a speed (utterance reach), and the like of the voice data. Based on the result of the acoustic analysis, the likelihood (likeliho) for the speech (word) input to the microphone 11 for each of the recognition target words stored in the recognition dictionary based on the analysis result.
od) Perform the calculation. The speech recognition circuit 13 supplies the likelihood corresponding to the recognition target word to the control unit 20 together with the recognition target word. Also,
The speech recognition circuit 13 performs acoustic analysis and likelihood calculation in the same manner even when speech data of a plurality of words is input, and determines the most likelihood based on the chain probability data and likelihood of each word stored in the speech recognition data ROM 16. The likely combination of the recognition target words is supplied to the control unit 20.

The control unit 20 includes a CPU (Central Processing Unit) 21 for controlling the entire apparatus, and a CPU 2.
A second RAM for temporarily storing data necessary for the operation of the first RAM
22, a second ROM 23 for storing a system program and an application program for performing predetermined processing,
Word information table ROM2 for storing word information table
4 and a chain rule table ROM 25 for storing a chain rule table. The word information table and the chain rule table will be described later in detail.

The control unit 20 generates text data for speech synthesis using the speech recognition result of the speech input unit 10 and the word information table and the concatenation rule table, and supplies the text data to the speech output unit 30. In addition, CPU
Reference numeral 21 includes a timer (not shown) and always counts the current date and time.

The voice output unit 30 includes a voice synthesis data ROM 31 for storing a voice synthesis dictionary which is a dictionary for converting text data into voice data, and a third ROM 32 for storing an application program for voice synthesis.
And a third RAM 33 for temporarily storing text data for speech synthesis and other data necessary for operation from the control unit 20.
, A voice synthesis circuit 34 that performs voice synthesis, a D / A converter 35 that converts a digital signal into an analog signal, and a speaker 36 that outputs voice based on the voice signal.

The voice synthesizing circuit 34 performs a predetermined process according to an application program in the third ROM 32. The voice synthesis circuit 34 is, specifically, a third RAM 3
3 is compared with the speech synthesis dictionary stored in the speech synthesis data ROM 31 to generate speech data corresponding to the speech synthesis text data. A converter 35
Is supplied to the speaker 36 via the.

Here, the word information table stored in the word information table ROM 24 will be described. The word information table includes an input word that is a speech recognition word that may be received from the speech recognition circuit 13 and information necessary for interpreting each word. Specifically, as shown in FIG. 2, “input word”, “type”, “value”,
It consists of a “word holding flag”.

The "input word" is a word indicating a time such as a year, a month, a date, and a time among the words recognized by speech. The “type” indicates the type in the interpretation of the word with respect to the “input word”. For example, year, month,
Week, day of week, day, hour, etc. correspond.

"Value" indicates a specific value in the type. For example, when the “type” is “month”, it specifically specifies which month from January to December, and when the “type” is “hour”, what time it is.

The "word holding flag" indicates whether the "input word" itself is used when constructing an output sentence.

For example, the input word "Shumatsu" has a type of "day of the week", a value of "Friday to Sunday", and a word holding flag of "Yes". Similarly, the human word "raisyu" is of type "week" and has a value of "current week +1"
“No” is attached to the word holding flag. The input word “about” has a type of “before and after” and has no value, and the word holding flag is attached with “Yes”. The same applies to other cases.

Next, the chain rule table stored in the chain rule table ROM 25 will be described.
The chain rule table is composed of a sequence of a plurality of words recognized by speech and information necessary for proceeding with the interpretation from the chain pattern of the words. Specifically, as shown in FIG. "," Post-interpretation scale ", and" interpretation rule ".

The "chain pattern" indicates a continuous pattern of "type" in the above-mentioned word information table. That is, it is a chain pattern of the “type” of the “input word” recognized by the voice.

"Post-interpretation scale" means "chain pattern"
Is a time unit when interpreting, that is, a scale, which indicates one of a month, a date, and an hour.

The "interpretation rule" indicates a rule for calculating the value of the interpreted date and time. Here, the first "reference" is the current date and time.

For example, the chain pattern “week + day” of the type of the input word has a scale of “day” after interpretation, and the interpretation rule is attached with “the applicable day of the week as a reference”.
Similarly, the chain pattern “day of the week + before and after” has an interpreted scale of “day” and an interpretation rule of “standard day ± 1 day”. Also, the chain pattern "human power start + week"
This is applied when the type of the first input word is “week”, the scale after interpretation is “week”, and the interpretation rule is “applicable week based on the present”. The same applies to other data.

2. Generation of Numerical Data for Comparison In the conventional schedule management, it is necessary to specify a date and time when there is an errand when inputting a schedule. However, for example, when it is only known that there will be golf next weekend, a specific date and time cannot be specified.

Therefore, the schedule search device 1
Even if there is an ambiguous voice input such as "golf golf next weekend", data of a predetermined format is generated from the result of the voice recognition so that the schedule can be managed. I have. The generated data converted in this manner is called numerical comparison data. Below, "Raishu Shumatsu Matsuro"
The generation of numerical comparison data from the voice input described above will be described, and the description of “Golfu” will be omitted.

CPU 21 of schedule search device 1
Receives the speech recognition result from the speech input unit 10 and executes the processing from step S1 shown in FIG.

In step S1, the CPU 21 sets the current date and time as the first reference date and time, and proceeds to step S1.
Proceed to 2. That is, the current date and time is set to be September 21, 1998.
If it is a day, the first reference date will also be September 21, 1998. Here, the first reference date and time is date and time data that the CPU 21 holds in the process of interpreting the input sentence,
12 is stored. Numerical comparison data is obtained as a result of updating the first reference date and time by interpreting the input word. At this time, the first reference date and time is, as shown in FIG. 5, the portion "199" where the input word is "input start".
8/9/21 ".

In step S2, the CPU 21 obtains one input word from the sentences recognized by the voice input unit 10 and proceeds to step S3. For example, for an input sentence "Raishu Shumatsu Guro"
"Laishu" is acquired as an input word.

In step S3, the CPU 21 refers to the word information table stored in the word information table ROM 24, performs a search using the input word as a key, and proceeds to step S4. According to the word information table,
For example, for the input word “raisyu”, the “type” is “week” and the “value” is “current week + 1”.

In step S4, the CPU 21 refers to the chain rule table stored in the chain rule table ROM 25, searches for a chain pattern for the "type" of the input word, and proceeds to step S5. For example, the input word “raisyu” is at the beginning of the sentence, so “chain pattern” is “input start + week”, “post-interpretation scale” is “week”, and “interpretation rule” is “standard day”. Of the week.

In step S5, the CPU 21 updates the first reference date and time from the corresponding interpretation rule. Specifically, the CPU 21 updates the first reference date and time using the “value” and the “interpretation rule” for the input word, and proceeds to step S6. For example, when the first reference date and time is September 21, 1998, the “value” is “current week + 1”, and the “interpretation rule” is “reference week”, FIG. As shown in FIG.
Weeks up to 4 are updated as the first reference date and time. The date and time scale after this interpretation is week.

In step S6, the CPU 21 determines whether or not the input word has been completed. When it is determined that the input word has been completed, the process proceeds to step S7, and when it is determined that the input word has not been completed, the process returns to step S2.

For example, when the processing of the input word of “Raishu” is completed for the sentence of “Raishu Shumatsu Goro”, the flow returns to step S2 to return to the step S2. Execute the process.

Similarly, when the CPU 21 completes the processing from step S2 to step S5 for the input word of “Shu Matsu”, the CPU 21 sets the first reference date and time to “1998/1998”.
10/2 to 10/4 ", and the scale becomes" day "after interpretation. Further, when the process is also performed for “about”, as shown in FIG. 5, the “type” is “back and forth”, the “chain pattern” is “day + back and forth”, the “post-interpretation scale” is “day”, and “interpretation”. The “rule” is “reference date ± 1”, and the first reference date is “1998/10 / 1-1 to 10/98”.
5 ". Then, when the processing is completed for each word of the sentence "Laishu Shu Matsuro", the process proceeds to step S7.

In step S7, the CPU 21 stores the first reference date and time obtained by ending the processing of step S6 as numerical comparison data in the second RAM 22, and ends the processing.

As described above, when there is a voice input sentence in the voice input unit 10, the CPU 21 performs the processing in steps S1 to S7 to generate numerical comparison data for the voice input sentence. It is stored in the second RAM 22. Further, when there is another voice input sentence, similarly, numerical comparison data is generated for the other voice input sentence.

These numerical comparison data are data that can be compared with other numerical comparison data, and are data indicating a time range for schedule management.

3. Generation of Output Configuration Data If the above-mentioned input sentence is output as it is, there is a problem that the content is suitable when the actual date and time has elapsed. For example, the sentence "Raishu Matsuro" can be left as it is if it is output during the week, but it will be unnatural if the next week does not output "Conscious Matsuro".

It is also conceivable to create output data from the data for numerical comparison in accordance with the date and time at the time of output. However, for example, "Laishu Shumatsu Matsuro"
Is input in a numerically ambiguous sentence as in the numerical comparison data (for example, 1998/10 / 1-199).
Even if the output sentence is reconstructed from (8/10/5), only a sentence such as "from the beginning to the end" can be generated at most, and a sentence completely different from the original input sentence will be output. .

Therefore, the schedule search device 1
Output configuration data is generated for the voice input sentence so that a search result corresponding to the voice input sentence can be output. This processing may be performed after the above-described processing from step S1 to step S7 shown in FIG. 4 or may be performed in parallel with the processing.

Specifically, C of the schedule search device 1
When receiving the speech recognition result of the input speech sentence by the speech input unit 10, the PU 21 executes the processing of step S11 and the subsequent steps shown in FIG. Here, as in the case described with reference to FIG. 4, it is assumed that a voice input of “Raishu Matsuguro” has been made.

In step S11, the CPU 21
The current date and time are set as the second reference date and time, and the process proceeds to step S12. The second reference date is a variable for creating output configuration data. This is similar to the first reference date and time, but does not reflect the value of the ambiguous input word, and determines only the value of the numerically specific input word. Different from the reference date and time. For example, assuming that the current date and time are September 21, 1998, the second reference date and time is also September 21, 1998 (September 21, 1998).

In step S12, the CPU 21
The speech input unit 10 acquires one input word from the sentences recognized by speech, and proceeds to step S13. For example,
For the input sentence "Raishu Shu Matsuro", "Raishu" is acquired as an input word.

In step S13, the CPU 21
Referring to the word information table stored in the word information table ROM 24, a search is performed using the input word as a key, and the process proceeds to step S14. According to the word information table, for example, for the input word "raisyu",
The “type” is “week”, the “value” is “current week + 1”, and the “word holding flag” is “Yes”. "Word holding flag"
Is a flag indicating whether the input word is ambiguous.
Is ambiguous, and No means not ambiguous. Note that an input word whose word holding flag is Yes is ambiguous, and it is necessary to reflect that in the output sentence.

In step S14, the CPU 21
It is determined whether the “word holding flag” of the input word is Yes. If it is determined that the input word is “Yes”, the process proceeds to step S19. If it is not “Yes”, that is, if it is determined that the input word is No, the process proceeds to step S15.

In step S15, the CPU 21
With reference to the chain rule table stored in the chain rule table ROM 25, a chain pattern for the "type" of the input word is searched, and the process proceeds to step S16. For example, the input word “raisyu” is at the beginning of the sentence, so “chain pattern” is “input start + week”, “post-interpretation scale” is “week”, and “interpretation rule” is “standard day”. Of the week.

In step S16, the CPU 21
The first reference date and time is updated from the corresponding interpretation rule. Specifically, the CPU 21 updates the second reference date and time using the “value” and the “interpretation rule” for the input word, and proceeds to step S17. For example, when the second reference date is September 21, 1998, the “value” is “current week + 1”, and the “interpretation rule” is “reference week”, The week from September 28 to October 4 is updated as the second reference date and time. The date and time scale after this interpretation is week.

In step S17, the CPU 21
It is determined whether the input word has been completed. If it is determined that the input word has been completed, the process proceeds to step S18. If it is determined that the input word has not been completed, the process returns to step S12.

For example, when the processing for the input word of “Raishu” is completed for the sentence of “Raishu Shumatsuro”, the process returns to step S2, and the processing for the input word of “Ryushu Matsu” is performed. Execute.

On the other hand, in step S14, the word holding flag
In step S19 when it is determined to be Yes,
The CPU 21 proceeds to step S20 using the second reference date and time and its scale up to now as output configuration data.

For example, as for the voice input of “Raishu Matsumatsuro”, the processing of steps S11 to S18 is first performed for “raishu”, and then the processing of steps S11 to S14 is performed for “shumatsu”. Is performed and the process proceeds to step S19, as shown in FIG. 7, the second reference date and time and its scale, that is, “(September 2, 1998)
8 to October 4) "as output configuration data.

In step S20, the CPU 21
The input word is added to the output configuration data, and step S1
Go to 7. At this time, since the input word is “Shumatsu”, the output configuration data is “(19
Week from September 28 to October 4, 1998).

When the CPU 21 returns to step S12 via step S17, it processes the input word "about" and proceeds to step S18. Since the word holding flag is Yes, this input word is added to the output configuration data in step S20. That is, as shown in FIG. 7, the output configuration data is “(September 28, 1998 to October 4, 1998) Shugo Matsuro”.

In step S18, the CPU 21
The remaining second reference date and time are added as output configuration data. That is, if the second reference date and time not reflected in the output configuration data remains, the second reference date and time are added to the output configuration data, and the process ends. In the case of the above example, there is no second reference date and time to be added.
The data for output configuration is "(10
Week 4 of the month).

When the CPU 21 performs the processing from step S11 to step S20 on one voice input sentence to generate output configuration data, the CPU 21 links this to the numerical comparison data and stores it in the second RAM 22. Let it. Therefore, the second RAM 22 stores the data for numerical comparison and the data for output configuration linked to one voice input sentence.

4. Schedule Search (4-1) Search Process Next, a schedule search using numerical comparison data will be described. When there is a voice input instructing the search range, the schedule search device 1 generates numerical comparison data for the voice input, and stores the range indicated by the numerical comparison data and each of the stored numerical comparison data. The data is compared with the range indicated by the data, and the data for numerical comparison that partially or entirely overlaps is selected as an output candidate of the search result. Here, a description will be given of a process of searching for a schedule when there is a voice input of "Laishu Nakaba".

First, it is assumed that the second RAM 22 has already stored, for example, numerical comparison data corresponding to three word strings of input sentence numbers W1, W2, and W3 as shown in FIG.

For example, as for the input sentence number W1, “1998/10”
/ 1-1998 / 10/5 "is stored. As for the input sentence number W2, "1998/9 /
30 13:00 ”is stored. Regarding the input sentence number W3, “1998/9/22 10: 30-1” for “Tomorrow Hiro”
998/9/22 13:30 "is stored. Then, when there is a voice input sentence "Laishu Nakaba", it is searched whether any of W1, W2, and W3 has a portion that overlaps therewith.

Next, the CPU 21 performs the above-described processing from step S1 to step S7 to generate numerical comparison data for the input sentence number W4 "Raishu Nakaba". Then, a search is made as to whether there is numerical comparison data that overlaps with the input sentence number W4 among the input sentence numbers W1, W2, and W3.

As shown in FIG. 8, the numerical comparison data of the input sentence number W4 is "1998/9 / 30-1998 / 1".
0/1 ". The numerical comparison data of the input sentence number W1 is “1998/10 / 1-1-1998 / 10/5”, which is duplicated for the date of October 5, 1998. Therefore, the CPU 21 selects the data for numerical comparison of the input sentence number W1 as an output candidate of the search result.
The CPU 21 similarly compares the input sentence numbers W2 and W3 with the input sentence number W4. The numerical comparison data of the input sentence number W2 has a range overlapping with the numerical comparison data of the input sentence number W4. On the other hand, the numerical comparison data of the input sentence number W3 has no overlapping range with the numerical comparison data of the input sentence number W4. As a result, the CPU 21 selects numerical comparison data of the input sentence numbers W1 and W2 as output candidates of the search result.

In other words, for example, when the input sentence number W4 “Raishu Nakaba” is input as a search sentence, the input sentence number W1 “Raishu Nakamatsuro” and the input sentence number W2 “ "History of Sanju" is retrieved as date-related data. Also, when the input sentence number W4 is input as a sentence for schedule registration, it can be suggested that there is a possibility of double booking with the schedules of the input sentence numbers W1 and W2.

(4-2) Creation of Output Sentence After obtaining the above output candidates, the CPU 21 of the schedule search device 1 needs to output them. However, if output is performed as it is, there is a problem that the output sentence becomes unnatural depending on the current date and time. Therefore, the CPU 21
Is designed to create an output sentence corresponding to the current date and time using the already generated output configuration data.

This output sentence is dynamically created from the output configuration data according to the date and time at the time of output. The CPU 21
Specifically, a portion reflecting the reference date and time in the output configuration data is converted into a word expressing an appropriate date and time by referring to the current date and time to create an output sentence. An appropriate date expression based on the current date and time is to eliminate redundancy to the extent that the user does not misunderstand, and to use a relative expression (such as "raigetsu" or "tomorrow") as close as possible on the date and time scale.
The expression using.

After the search process, for example, when creating an output sentence for outputting the input sentence number W1, the CPU 21
While checking the current date and time with a timer not shown,
The time information in parentheses of the output configuration data "<week> (September 28, 1998) Shu Matsuro" of the input sentence number W1 stored in the second RAM 22 is compared with the current date and time. The CPU 21 is, for example,
If it is a day, the time information in the parentheses above is converted to "raisyu", and an output sentence "raisyu matsuro" is generated. If today is September 30, 1998,
The CPU 21 uses the time information in the parentheses as “consult”.
To create an output sentence "Consciousness" as shown in FIG.

When the CPU 21 outputs the input statement number W3, the CPU 21 outputs the time information in parentheses of the output configuration data "<date> (September 22, 1998) Hiragoro" of the input statement number W3. Compare with current date and time. The CPU 21
When the current date and time is September 22, the time information in the parentheses is converted into “today”, and as shown in FIG.
Generates an output sentence "Kyo Hiru Goro".

The CPU 21 determines that the input word string is
In the case of the output structure data, the output configuration data “<time> (1998/9/26 22:00) Zengo” is generated. When outputting the input sentence number W3, the CPU 21 compares the time information in parentheses of the output configuration data with the current date and time. CPU 21
Converts the time information in the parentheses into "Kyojujuji" when the current date and time is, for example, September 26, and as shown in FIG. 9, "Kyojujujuji"
Generates an output sentence "Zengo".

The CPU 21 of the control unit 20 outputs the output sentence data (hereinafter referred to as “output sentence data”) created as described above.
That. ) To the audio output unit 30.

The voice synthesis circuit 34 of the voice output unit 30 temporarily stores the output sentence data (text data for voice synthesis) from the control unit 20 in the third RAM 33, and stores the output text data in the voice synthesis data ROM 31. By collating with the stored speech synthesis dictionary, speech data corresponding to the speech synthesis text data is generated. The voice synthesis circuit 34 converts the voice data into a D / A converter 35
Is supplied to the speaker 36 via the. Thereby, the speaker 36 outputs the output sentence of the search result by voice.

As described above, the schedule search device 1 generates numerical comparison data indicating the temporal range of the input voice, and if there is a voice input whose temporal range is ambiguous, the ambiguous data is further obtained. Output configuration data for converting and outputting the input voice in accordance with the output date and time.

Thus, the schedule search device 1
Can perform search processing using numerical comparison data even for ambiguous input, and when outputting search results, the meaning of the original word is used for the ambiguous part of the input sentence. An output sentence corresponding to the current date and time can be generated using the held output configuration data. Therefore, a search result can be output according to the degree of ambiguity of the voice input sentence, and a system that is very easy for the user to use can be configured.

In the above embodiment, the first and second reference dates and times as date and time data have been described as examples of data having a mathematical spatial range, but the present invention is not limited to this. The present invention can be applied to, for example, data in a real space such as latitude / longitude and place name data, or in a virtual space in which, for example, the influence range of a person or an object is numerically expressed.

The schedule search device 1 stores the schedule management control program in the second RO from the recording medium recorded by the schedule management control program.
The above-described processing can be performed by installing the software in M15, the second ROM 23, the third ROM 32, and the like.

[0088]

As described above in detail, according to the retrieval apparatus, the retrieval method, and the recording medium on which the retrieval control program is recorded according to the present invention, the numerical value comparison of the sentence indicating the retrieval range inputted to the input means is performed. With respect to the numerical comparison data generated by the data generation unit, the numerical comparison data whose numerical range overlaps is searched from the storage unit, and among the output configuration data corresponding to the overlapping numerical comparison data, An ambiguous sentence was input by converting the numerical information constituting the output composition data into words corresponding to the current situation of the user and generating output composition data including the converted words as a search result sentence. Even in such a case, the search can be effectively performed, and an output sentence to be a result of the search can be generated while the ambiguity of the input sentence remains.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a schedule search device to which the present invention has been applied.

FIG. 2 is a diagram showing an example of a word information table stored in a word information table ROM of the schedule search device.

FIG. 3 is a diagram showing an example of a chain rule table stored in a chain rule table ROM of the schedule search device.

FIG. 4 is a flowchart illustrating an operation when generating numerical comparison data relating to an input speech sentence.

FIG. 5 is a diagram specifically illustrating the generation of the numerical comparison data using an example.

FIG. 6 is a flowchart illustrating an operation when generating output configuration data relating to an input speech sentence.

FIG. 7 is a diagram specifically illustrating the generation of the output configuration data using an example.

FIG. 8 is a diagram for explaining a search using data for numerical comparison.

FIG. 9 is a diagram illustrating the relationship between an input word string, output configuration data, and an output sentence.

[Explanation of symbols]

1 schedule search device, 10 voice input unit, 20
Control unit, 21 CPU, 22 second RAM, 24 word information table ROM, 25 chain rule table RO
M, 30 audio output unit

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Masakazu Hattori 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Yasuhiko Kato 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo No. Sony Corporation F term (reference) 5B075 NR14 PP07 PP13 PP24 PQ04 5D015 AA05 HH11

Claims (13)

[Claims] An input unit for inputting a sentence; and a numerical comparison data generating unit for generating numerical comparison data indicating a numerical range of the entire sentence from each word constituting the sentence input to the input unit. Converting a word indicating a predetermined numerical range to a problem area targeted by the system among words forming the sentence input to the input means, into numerical information indicating the predetermined numerical range, Output configuration data generation means for generating output configuration data consisting of numerical information and other words excluding those converted into numerical information; and for the sentence input to the input means, the numerical comparison data generating means First storage means for storing the data for numerical comparison generated in step 1 in association with the data for output configuration generated by the data generation means for output configuration, and a search range input to the input means. Searching means for searching a sentence within the search range based on the sentence indicating the search result, and generating a search result sentence indicating the selected sentence as a result of the search, and outputting the search result sentence generated by the search means Output means, wherein the search means has a numerical range overlapping with the numerical comparison data generated by the numerical comparison data generation means for the sentence indicating the search range input to the input means. Numerical comparison data is retrieved from the storage means, and among the output configuration data corresponding to the overlapping numerical comparison data, the numerical information constituting the output configuration data is converted into a word corresponding to the current situation of the user. And generating output configuration data including the converted word as a search result sentence. 2. The numerical comparison data generating means generates numerical comparison data indicating a temporal range of the entire sentence from each word constituting the sentence input to the input means, and the output configuration data The generation unit converts a word indicating a predetermined time range with respect to the input time among the words constituting the sentence input to the input unit into time information indicating the predetermined time range, and Generating output configuration data consisting of other words excluding those converted to time information, wherein the search unit is the numerical comparison data generation unit for a sentence indicating a search range input to the input unit; With respect to the generated numerical comparison data, the storage means is searched from the storage means for numerical comparison data having an overlapping time range, and of the output configuration data corresponding to the overlapping numerical comparison data, Converts the time information constituting the use data word corresponding to the current time, the search apparatus according to claim 1, wherein the generating output configuration data including the converted word as a search result statement. 3. The numerical value comparison data generating means includes, for a word whose time range is ambiguous, a second storage means for storing a value indicating a time range indicated by the word with respect to a current time, and Of the words constituting the sentence input to the means, words whose time range is ambiguous are described in the second section.
3. The retrieval device according to claim 2, wherein the comparison unit refers to a value indicating a time range stored in the storage unit, and generates numerical comparison data indicating a time range of the entire input sentence. . 4. The retrieval apparatus according to claim 1, wherein said input means comprises speech recognition means for performing speech recognition processing on an input speech sentence and outputting a sentence after the recognition processing. 5. The search device according to claim 1, wherein the output unit synthesizes a voice based on the search result sentence generated by the search unit, and outputs the search result sentence by voice. . 6. Generating numerical comparison data indicating a numerical range of the entire sentence from each word constituting the input sentence, and a problem which is targeted by the system among the words constituting the input sentence. A word indicating a predetermined numerical range for an area is converted into numerical information indicating the predetermined numerical range, and an output including the converted numerical information and other words excluding those converted into the numerical information is output. A configuration data is generated, and for the input sentence, the generated numerical comparison data and the generated output configuration data are stored in the first storage unit in association with each other, and the input search is performed. Numerical comparison data having a numerical range overlapping with the numerical comparison data generated for the statement indicating the range is retrieved from the first storage means, and the output configuration data corresponding to the overlapping numerical comparison data is retrieved. That is, the numerical information forming the output configuration data is converted into a word corresponding to the current situation of the user, and the output configuration data including the converted word is generated as a search result sentence. A search method characterized by outputting a sentence. 7. A method for generating numerical comparison data indicating a temporal range of the entire sentence from each word constituting an input sentence, wherein a predetermined time period corresponding to an input time of each word constituting the input sentence is generated. The word indicating the range is converted into time information indicating the predetermined time range, and output configuration data including the converted time information and other words excluding those converted into the time information is generated. For the input sentence, the generated numerical comparison data and the generated output configuration data are stored in the first storage unit in association with each other, and the input sentence indicating the search range is generated. The first storage means retrieves the numerical comparison data whose time range overlaps with the obtained numerical comparison data, and among the output configuration data corresponding to the overlapping numerical comparison data, De Converting the time information constituting the data into words corresponding to the current time, generating output configuration data including the converted words as a search result sentence, and outputting the generated search result sentence. The search method according to claim 6, wherein the search is performed. 8. For a word whose time range is ambiguous, a value indicating a time range indicated by the word with respect to the current time is stored in the second storage means, and among words constituting the input sentence, For a word whose time range is ambiguous, reference is made to the value indicating the time range stored in the second storage means, and numerical comparison data indicating the time range of the entire input sentence is generated. The search method according to claim 7, wherein 9. The retrieval method according to claim 6, wherein a recognized sentence obtained by subjecting an input speech sentence to speech recognition processing is used as the input sentence. 10. The search method according to claim 6, wherein a speech is synthesized based on the generated search result sentence, and the search result sentence is output by voice. 11. A method for generating numerical comparison data indicating a numerical range of the whole sentence from each word constituting an input sentence, and a problem which is targeted by the system among the words constituting the input sentence. A word indicating a predetermined numerical range for the region is converted into numerical information indicating the predetermined numerical range,
Generating output configuration data including the converted numerical information and other words excluding those converted to the numerical information, and generating, for the input sentence, the generated numerical comparison data; The first data is stored in the first storage means in association with the output configuration data, and the numerical comparison data having a numerical range overlapping with the numerical comparison data generated for the sentence indicating the input search range is described above. Retrieving from the first storage means, of the output configuration data corresponding to the overlapping numerical comparison data, converting the numerical information constituting the output configuration data into words corresponding to the current situation of the user, A recording medium on which a search control program for generating output configuration data including the generated word as a search result sentence and outputting the generated search result sentence is recorded. 12. Generating numerical comparison data indicating a temporal range of the entire sentence from each word constituting an input sentence, and generating a predetermined time-based data corresponding to an input time among the words constituting the input sentence. The word indicating the range is converted into time information indicating the predetermined time range, and output configuration data including the converted time information and other words excluding the words converted into the time information is generated. For the input sentence, the generated numerical comparison data and the generated output configuration data are stored in the first storage means in association with each other, and the input sentence indicating the search range is generated. The first storage means retrieves the numerical comparison data whose time range overlaps with the obtained numerical comparison data, and among the output configuration data corresponding to the overlapping numerical comparison data,
The time information forming the output configuration data is converted into a word corresponding to the current time, the output configuration data including the converted word is generated as a search result sentence, and the generated search result sentence is output. The recording medium according to claim 11, wherein the search control program is recorded. 13. For a word whose time range is ambiguous, a value indicating the time range indicated by the word with respect to the current time is stored in the second storage means, and among words constituting the input sentence, For words whose time range is ambiguous, the second
A search control program for generating numerical comparison data indicating the time range of the entire input sentence with reference to the value indicating the time range stored in the storage means. Item 13. A recording medium on which the search control program according to Item 12 is recorded.