JP2000090087A - Interpreter and record medium recording program for effectively presenting function of interpreter - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an interpreter capable of smooth conversation between a user, who comprehends a first language, and a speaking party who comprehends a second language. SOLUTION: Concerning this interpreter, a question sentence translation data base 101 stores question sentences in the first language and translated question sentences in the second language while pairing them. Corresponding to the input of a question sentence from an input operating part 103 of the user, a question sentence selecting part 104 selects any question sentence out of the question sentence translation data base 101. A voice output part 106 outputs the paired translated question sentence. A response sentence data base 102 stores plural response sentences in the second language corresponding to the translated question sentences and translated response sentences in the first language while pairing them. A voice response from the speaking party is received by a voice input part 109 and a voice recognizing part 110 recognizes the response in voice as a character response sentence. A response sentence extracting part 107 retrieves the response sentence similar to the character response sentence from the response sentence data base 102 and clearly displays the difference of character response sentence and similar response sentence on a display part 105.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an interpreter for translating conversation between a first language and a second language.

[0002]

2. Description of the Related Art In a conventional portable interpreter, when a user selects a menu or a keyword which is classified according to a use scene and selects a word or one sentence (question sentence) in a first language, the user is prompted to enter the menu. The second corresponding to the selected word or sentence
A method in which a word or a sentence (a question sentence) of a language is displayed is mainly adopted.

In order for a user to get a reply from a conversation partner,
For example, in the technology disclosed in Japanese Patent Application Laid-Open No. H7-105220, when displaying the above-described question text in the second language to a conversation partner,
A method is shown in which several types of response sentences in the second language are displayed, and one of the response sentences is selected by a conversation partner.

In the technique disclosed in Japanese Patent Application Laid-Open No. 9-319750, responses from conversation partners are assumed, responses are classified hierarchically to facilitate selection, and numerals or symbols are directly input to allow the communication from the other party. The manner in which the response is obtained is shown.

[0005]

However, the above-mentioned 2
In one technique, in order to obtain a response sentence from a conversation partner, the user has to teach the conversation partner how to use the interpreter and leave the operation of the device to the other party. Difficult conversation.

SUMMARY OF THE INVENTION In view of the above problems, an object of the present invention is to provide an interpreter having excellent operability, which does not require a conversation partner to be proficient in handling the interpreter.

[0007]

In order to solve the above-mentioned problems, the present invention provides a question sentence database storing a pair of a question sentence in a first language and a question sentence in a second language translated from the first sentence. A response sentence database storing a plurality of standard second language response sentences responding to the second language question sentence, and a first language response sentence translated therefrom; input operation means;
Voice input means, and question sentence extraction means for searching the question sentence database for the first language question sentence input by the user with the input operation means and extracting a paired second language question sentence therefrom; And a voice for recognizing the input voice as a text response in the second language when the user's conversation partner inputs the voice in the second language in response to the output of the extracted question in the second language. Recognizing means; response sentence extracting means for extracting a response sentence in a first language corresponding to a response sentence in a second language similar to the character response sentence recognized by the voice recognition means from the response sentence database; Display means for displaying a response sentence in the second language, a response sentence in the first language, and a character response sentence recognized by the voice recognition means, extracted by the sentence extraction means.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the communication device according to the present invention will be described below with reference to the drawings.

(Embodiment 1) FIG. 1 is a configuration diagram of an interpreter according to Embodiment 1 of the present invention.

This interpreting apparatus includes a question sentence translation database 101, a response sentence database 102, an input operation unit 1
03, a question sentence selection unit 104, a display unit 105, a speech output unit 106, a response sentence extraction unit 107, a speech recognition dictionary 108, a speech input unit 109, and a speech recognition unit 110.

The question translation database 101 is a ROM
The first sentence and the second language question sentence translated into the second language are stored as a pair with an identification number. FIG. 2 shows this question sentence translation database 1
01 shows an example of the content. In the question sentence translation database 101, the first language is Japanese and the second language is English. In this embodiment, the interpreter is used in a situation where a user who uses the interpreter understands Japanese and a conversation partner understands English.

The identification number "0002" of the question translation database 101 contains the Japanese question "How long does it take to get to the airport?"
does it take to the airport? " Further, the number “5” of standard response sentences responding to this question message and the identification number “T1001” of the response message,
"T1002",... "T1005" are stored. The identification number of this response statement is stored in the response statement database 1
02 is stored as an identification number for identifying the response sentence.

The response sentence database 102 is composed of a ROM or the like, and stores a list of standard response sentences that answer English question sentences in the question sentence translation database 101. FIG.
FIG. 4 is a diagram showing an example of contents stored in a response sentence database 102. Each response sentence 301 stores therein a Japanese translation 302 of the response sentence and an identification number 303 of the response sentence.

The response sentence database 102 stores a response sentence "30dollars" identified by the identification number "T0999" and a translation into Japanese of "30 dollars."

The input operation unit 103 includes a keyboard or the like, and receives a user's input of a question sentence. At this time, the user may input the full text of the question text, but, for example, if he wants to ask for the time required for the airport, he inputs "airport" and "time". The input operation unit 103 notifies the question text selection unit 104 of the input question text or the input words “airport” and “time”.

Further, upon receiving an input of a question instruction from the user, the input operation unit 103 notifies the question sentence selection unit 104 of that fact.

The question selecting section 104 includes an input operating section 103
When the notification of the question sentence is received from, the question sentence translation database 101 is searched, and the matching question sentence, the translated question sentence, the identification number and the identification number of the response sentence are extracted.

When a word is received as a question sentence, the word is used as a keyword, a question sentence including the keyword is selected, and the translated question sentence, the identification number, and the identification number of the response sentence are paired. And extract For example,
When notified of “airport” and “time”, the user selects the question “How long does it take to get to the airport” with the identification number “0002” including these two keywords? The question sentence selecting unit 104 displays the extracted Japanese question sentence and the English question sentence in a predetermined area of the display unit 105, and
Response statement identification number "T1001, T1002, T100
3, T1004, T1005 "to the response sentence extraction unit 107.

The question selecting section 104 includes an input operating section 103
When the notification of the question instruction is received from, the voice output unit 106 is notified of the identification number of the question sentence.

The display unit 105 is composed of a liquid crystal display or the like, and displays a question sentence or a response sentence under the control of the question sentence selection unit 104 and the response sentence extraction unit 107.

FIG. 4 is a diagram showing an example of the display contents of the display unit 105. This figure is a display example in a state where the question sentence is selected by the question sentence selection unit 104 and the translated question sentence is output from the voice output unit 106 as voice. A Japanese question sentence 401 and its English question sentence 402 are displayed by the question sentence selection unit 104, and a standard answer sentence as an answer to the English question 402 and a translated answer sentence translated into Japanese The response example 403 is a list of the response sentence extraction unit 1
07. With such a display, the user can know the content of the question he or she wants to answer and the content of the response sentence that the conversation partner will answer.

FIG. 5 is a diagram showing another example of the display contents of the display unit 105. This figure shows the display contents after receiving a voice response sentence to the question sentence from the user's conversation partner. The question sentences 401 and 402 are displayed in the same manner as in FIG. The response example 501 shows a list of response sentences similar to the recognition result of the voice response from the conversation partner recognized by the voice recognition unit 110. Words that match the recognition result in this response sentence are highlighted. The lower part of the display unit 105 shows the voice response recognition result 502. If the user looks at the contents of the display unit 105, it can be estimated that the required time to the airport is "20 minutes by taxi".
In other words, since the part in which the word corresponding to the recognition result 502 is highlighted is shown in the response sentence, it is displayed in the response sentence.
By comparing with 2, it can be inferred from the fact that the numbers are different.

The voice output unit 106 stores the identification number of the question text and the voice pattern of the translated question text. When the identification number is notified from the question text selection unit 104,
The voice pattern of the identification number is extracted, converted into a voice signal, and a question sentence is output as voice via a speaker.

The answer sentence extracting unit 107 is provided for the question sentence selecting unit 10.
When the notification of the identification number of the response sentence is received from 4, the response sentence having the identification number stored in the response sentence database 102 and the translated response sentence are extracted. A list of the extracted response sentences and their translated response sentences is displayed in a predetermined area of the display unit 105 as shown in FIG.

When the response sentence extracting unit 107 receives a notification of the character response sentence as the recognition result from the speech recognition unit 110, the response sentence extraction unit 107 selects a response sentence similar to the character response sentence from the extracted response sentences. The similar response sentence, its translation response sentence, and the character response sentence are displayed in a predetermined display area of the display unit 105.

At the time of selecting a similar response sentence, a word constituting the notified character response sentence is compared with a word constituting the extracted response sentence. Select a sentence as a similar response sentence. In addition, the display attribute of the matching word in the response sentence is changed so that the display attribute of the matching word is reversely displayed in white as shown in FIG.

In the present embodiment, the similar response sentence includes the words constituting the character response sentence in the response sentence, but the similar response sentence may be selected as follows. The response sentence extraction unit 107 calculates the ratio of the number of words matching the words constituting the character response sentence to the total number of words included in the response sentence as the similarity, and resembles the response sentences having the similarity of a predetermined value or more. Select as a response sentence.

The speech recognition dictionary 108 stores word standard patterns and words in pairs. FIG. 6 is a diagram showing an example of the contents of the speech recognition dictionary 108.

The word standard pattern 601 is obtained by converting a normally pronounced word "twenty" 602 to a pronunciation time corresponding to 5 minutes.
It is represented by the feature amount of 0 frame. Each frame corresponds to a voice pattern for a predetermined time, which sounds the word “twenty” in 0.5 seconds and is shifted by a time interval of 10 ms. This voice pattern is subjected to Fourier transform, the frequency band is divided into 16, and the intensity of each band is represented by a 16-value sequence as a feature amount of one frame. Note that the voice pattern 604
Describes the division into frames. The word mi
Also for "nutes" and the like, the word standard pattern is represented by a numerical sequence of 16 in a predetermined number of frames, but is omitted in the figure.

The voice input unit 109 includes a microphone or the like, receives input of a voice response sentence of a conversation partner, and notifies the voice recognition unit 110 of the voice signal.

The voice recognition unit 110 analyzes the voice signal notified from the voice input unit 109, divides the voice signal into segments in word units, and checks the voice pattern of each segment against the word standard pattern in the voice recognition dictionary 108. . A word stored as a pair with the word standard pattern having the highest similarity to the voice pattern is used as a recognition result, and the words of the recognition result are arranged to obtain a character response sentence. The obtained character response sentence is notified to the response sentence extraction unit 107.

In this embodiment, although the voice response sentence "20 minutes by taxi." Input from the voice input unit 109 is recognized by the voice recognition unit 110, depending on the input voice response sentence, the similarity degree is the highest. Selecting a word paired with a high standard word pattern does not always result in a character response sentence that is a correct recognition result.

The speech recognition section 110 of the present embodiment further has the following configuration.

The case where the voice response sentence "You can go on foot." Shown in FIG.

The voice recognition unit 110 performs voice analysis processing as shown in FIG. 7B to divide the voice pattern 701 into segments 702 to 706 corresponding to each word.
The voice patterns 702 to 706 of each segment are collated with the word standard patterns stored in the voice recognition dictionary 108, and the similarity is calculated using, for example, a DP (dynamic programming) matching method. As a result, the word candidates “show”, “you”, “how”,... Are sequentially selected from the words in the word standard pattern having the highest similarity from the voice pattern of the segment 702. Similarly, word candidates “can”, “and”,... Are selected from the voice pattern of the segment 703.

Now, the words constituting the response sentence extracted by the response sentence extraction unit 107 include "you", "can", "go", and "o".
"n", "foot", and the like. Voice recognition unit 1
10 checks whether or not each word candidate is present in a word in the response sentence extracted by the response sentence extraction unit 107 for each segment, and if so, determines the similarity, for example,
Each word candidate is rearranged in the order of similarity as “double”.
As a result, the order of each word candidate is rearranged from the voice pattern 701 of the voice response sentence input from the voice input unit 109 based on the corrected similarity as shown in FIG. Note that, in FIGS. 7B and 8, description of specific numerical values of the similarity is omitted. The voice recognition unit 110 arranges the candidate characters having the highest similarity in the order of the segments, and “You can go
on foot. "is notified to the response sentence extraction unit 107 as a character response sentence as a recognition result.

As described above, since the response sentence to the question sentence is prepared in the response sentence extraction unit 107 as a standard response sentence, the accuracy of voice recognition of an unspecified conversation partner is improved by using the response sentence. Let me.

Next, the operation of this embodiment will be described with reference to FIGS.
This will be described with reference to the flowchart of FIG.

First, the user inputs a Japanese question sentence from the input operation unit 103 (S902).

The question sentence selecting unit 104 includes an input operation unit 103
A question sentence that matches the question sentence accepted in (1) is found by searching the question sentence translation database 101 (S904).
The found question text and the English translated question text translated from the question text are displayed on the display unit 105 (S906).

Next, the response sentence extraction unit 107 searches the response sentence database 102 based on the identification number of the response sentence notified from the question sentence selection unit 104, and extracts the response sentence and its translation. (S908). A list of the extracted response sentences and their translations is displayed on the display unit 105 (S910).

The question sentence selection unit 104 includes an input operation unit 103
In, it is determined whether there is a question instruction from the user (S912). If not, the process returns to S902, and if so, the voice output unit 106 is notified of the question instruction. Audio output unit 106
Outputs the English translation question sentence (S914).

Next, the sound input unit 109 is connected to the sound output unit 1.
The voice input of the English response sentence of the conversation partner who answered the English translation question sentence output from 06 is accepted (S916).

The voice recognition section 110 analyzes the voice signal notified from the voice input section 109 and performs voice analysis.
8 and recognize the voice as a character response sentence (S91).
8).

The response sentence extracting unit 107 selects a response sentence similar to the character response sentence from the list of response sentences (S920).
A similar response sentence (English), its translated sentence (Japanese), and a character response sentence (English) as a recognition result are displayed on the display unit 105. At this time, a part that matches the character response sentence of the similar response sentence Is changed (S922). By doing so, the user can understand the answer to the question in Japanese even if he cannot understand English.

Next, the operation of step S918 will be described in detail with reference to the flowchart of FIG.

First, the voice recognition unit 110 is connected to the voice input unit 1
Speech analysis of the speech signal input from S09 is performed to divide the speech signal into speech patterns of segments in word units (S100).
2). One of the divided voice patterns is taken out, and the similarity is calculated by pattern matching with the standard word pattern of the voice recognition dictionary 108 (S1004). A word of the standard word pattern having a similarity value equal to or greater than a certain value is set as a word candidate (S1006). This collation is performed for the voice patterns of all the segments (S1008).

Subsequently, it is determined whether or not the word candidate matches a word constituting the response sentence in the list of response sentences extracted by the response sentence extraction unit (S1010). If they do not match, the process moves to S1014, and if they do match, the similarity of the word candidate is corrected to twice (S1012).

Next, it is determined whether or not there is an unprocessed word candidate (S1014). If there is, the process returns to S1010. If not, the word candidates having the highest similarity of each segment are arranged and the character response sentence as the recognition result is obtained. (S1016).

As described above, since a standard response sentence to a question sentence is prepared in advance, when a voice response is recognized by speech, the words included in the response sentence constitute a character response sentence as a recognition result. You can take advantage of the high probability of being a word. As a result, the accuracy of voice recognition is improved even if the voice response is from an unspecified conversation partner.

(Embodiment 2) FIG. 11 is a configuration diagram of an interpreter according to Embodiment 2 of the present invention.

This interpreter includes a question sentence translation database 101, a response sentence database 1101, an input operation unit 103, a question sentence selection unit 104, a display unit 105, a voice output unit 1102, and a response sentence extraction unit 1103. , Voice recognition dictionary 108, voice input unit 109, voice recognition unit 11
0.

The same components as those of the first embodiment are denoted by the same reference numerals, and the description thereof will be omitted. Only the components unique to the present embodiment will be described.

The response sentence database 1101 stores, in addition to the standard response sentence in the second language (English) and the translation response sentence in the first language (Japanese) described in the first embodiment, the response sentence The correspondence between the word and the word of the translation response sentence is stored in association with the same identification number. Furthermore, whether the word or word string constituting the response sentence is an essential requirement as a response to the question sentence, and if it is an essential requirement, whether or not there is a possibility of replacement, speech recognition was not correctly performed The re-question rule at the time is stored.

Here, the replaceability means that a word constituting a standard response sentence is expected to be replaced with another word, and includes the attribute of a word serving as a replacement condition.

The re-question rule defines a rule for asking a re-question when the speech recognizing unit 110 cannot correctly recognize a word constituting the response sentence.

FIG. 12 shows the response sentence database 110
FIG. 3 is a diagram showing an example of the content of No. 1; The identification number "T100
The words “15” and “15” constituting the response sentence “15 minutes” identified by “1” and the translated response sentence “15 minutes” have identification numbers “1” and “minutes”. The identification number “2” is added to “minute”, which indicates that the word is the corresponding word. Similarly, words corresponding to the response sentence identified by the identification number “T1005” and the translated response sentence are also given the same identification number.

Each of the lists 1201 and 1202 has a “required requirement” column 1203, a “replaceability” column 1204, and a “requestion rule” column 1205 for the words constituting the response sentence. The list 1201 contains the identification number “T100
The words “15” and “mi” that make up the response sentence identified by “1”
nutes "is an" essential requirement "or not. "YES" 12 because both words are "required"
06 is described in the “essential requirement” column 1203.

Since both words are expected to be replaced with other words, the “replaceability” column 1204 shows YE
S1207 is described, and the word in which the word "15" is replaced is "number", and the word in which the word "minutes" is replaced is "hour unit", for example, "second", "hour". ing.

A rule corresponding to the case where the word “15” or a word that satisfies the replacement condition is not correctly recognized by speech is described as “KEYBORD” 1208 in the re-question rule column 1205. The rule “KEYBORD” 1208 means that an input of a numeral for replacing the word “15” is accepted from a conversation partner by operating the keyboard.

The corresponding rule when the word “minutes” or the word that matches the replacement condition is not correctly recognized by speech is described as “MENU” 1209 in the re-question rule column 1205.
This rule “MENU” 1209 indicates that “minute”
"s", "second", and "hour" are displayed, which means that the conversation partner can select one of them.

Similarly, list 1202 includes identification number “T”
1005 ”describes whether the word constituting the response sentence identified as“ 1005 ”is“ essential requirement ”.

The word string “by taxi” is an essential requirement, the attribute of the word used as the replacement condition is “transportation”, and “STR“ By what? ”” 1210 is described as a re-question rule. I have. Here, the re-question rule "STR" By what
"" 1210 is a question sentence to the conversation partner by voice "By what
? "Is output.

The voice output unit 1102 stores a voice pattern corresponding to the re-question rule.

The voice pattern of “Input numerals with keyboard.” Is stored in correspondence with the re-question rule “KEYBOARD” 1208, and “Indica” is stored in correspondence with “MENU” 1209.
te appropriate menu item. ", and corresponding to" STR "By what?"
What? "

The voice output unit 1102 is connected to the response sentence extraction unit 11
When the notification of the re-question rule is received from 03, the corresponding voice pattern is output as voice. The response sentence extraction unit 11
When the instruction for re-question is received from 03, the English-translated question sentence is output again.

The response sentence extracting unit 1103 performs the following processing after selecting a similar response sentence by receiving a character response sentence from the voice recognition unit 110 in addition to the configuration described in the first embodiment. .

The response sentence extracting unit 1103 determines whether or not the character response sentence and the similar response sentence completely match by comparing the words constituting both sentences. If they completely match, it is assumed that a response to the answer to the question sentence has been established.

If they do not completely match, it is determined by referring to the list 1201 of the response sentence database 1101 whether or not a different part (a word of a similar response sentence) is an essential requirement. If it is not an essential requirement, it is considered to be satisfied as an answer to the question sentence.

When a different part is an essential requirement, it is similarly determined whether or not the word has a substitutability.
Judge by looking at 1 etc. If there is no possibility of replacement, it will not hold as the answer to the question. At this time, the voice output unit 1102 is instructed to re-question that the question sentence has not been transmitted to the conversation partner.

When there is a possibility of replacement, it is determined whether or not words corresponding to different parts of the character response sentence match the attribute of the word as the replacement condition. When it is determined that they match, the character response sentence is established as an answer to the question sentence.

When it is determined that they do not match, the essential requirement is missing as the answer to the question sentence, so the re-question rule is obtained from the list 1201 or the like.

When the re-question rule is “KEYBOARD” 1208, the response sentence extracting unit 1103
02 is notified of the re-question rule "KEYBOARD", and the display unit 105 displays "input request by keyboard". As a result, the user can use the input operation unit 10 as a conversation partner.
It is understood that the input operation is received by using No.3.

When the re-question rule is “MENU” 1209, the voice output unit 1102 outputs the re-question rule “MENU”
And the display unit 105 displays “second” and “mi”
nutes "and" hour ". In this case, since the attribute of the word of the replacement condition is “time unit”, the menu item is “second” or the like, but the content of the menu item is changed depending on the attribute of the word.

The re-question rule is “STR“ By what? ”” 1210
, The re-question rule “ST
R "By what?"

The response sentence extracting unit 1103 is connected to the voice output unit 11
After the notification of “STR“ By what? ”” Is received in 02 and the notification of the character response sentence (word string) is received from the voice recognition unit 110, the word string of the similar response sentence or the “by“ t
If it is "axi", "by bus", "by train" or "on foot", the character response sentence is corrected.

Also, the re-question rules “KEYBOARD”, “MENU”
Is notified to the audio output unit 1102, and the input operation unit 10
In response to the notification that a conversation partner has input a number or an instruction for a menu item from 3, the corresponding word of the character response sentence is changed to a number or a menu item and displayed on the display unit 105.

The response sentence extracting unit 103 is the same as that of the first embodiment.
In the present embodiment, words constituting a character response sentence and a similar response sentence are displayed by changing the display attribute of the similar response sentence, but in the present embodiment, The correspondence between words constituting a similar response sentence and words constituting the translation response sentence in the sentence database 1101 is checked from the identification number, and the display attribute of the translation response sentence that matches the character response sentence is changed.

When the similarity of all the candidate characters is equal to or smaller than a predetermined threshold, the speech recognition unit 110 determines that the candidate character cannot be recognized and notifies the response sentence extraction unit 1103 of the part. In this case, the response sentence extraction unit 1103 treats the words constituting the response sentence in the same manner as when the words could not be correctly recognized.

Now, when the English question sentence “How long does it take to the airport?” Similar to that of the first embodiment is output from the voice output unit 106, the voice response unit “xxx” is output by the voice recognition unit 110. When "minutes by taxi." is recognized ("xxx" indicates that speech recognition is not possible), the response sentence extraction unit 1103 indicates that "xxx" is an essential requirement, the replacement condition is a number, and the re-question rule is It is determined that it is “KEYBOARD” 1208. From the audio output unit 1102, "Input nume
rals with keyboard. " The device is handed over to the conversation partner, and the input of the number “20” is received from the input operation unit 103. As a result, “xxx” for which speech recognition was not possible is replaced with the numeral “20”.

FIG. 13 shows an example of the contents displayed on the display unit 105. In response example 1301, the corresponding part of translation response sentence that matches character response sentence 1302 is highlighted. Although the change of the display attribute is reverse display, another attribute, such as a change of the display color, may be used. This allows the user to further easily understand the answer from the conversation partner. Next, the operation of the present embodiment will be described with reference to the flowchart shown in FIG. Since the operations up to S920 shown in FIG. 9 of the first embodiment are the same, only the operations unique to the present embodiment will be described.

At S 920, response sentence extracting unit 1103
Selects a response sentence similar to a character response sentence from a list of response sentences.

Next, the response sentence extraction unit 1103 determines whether or not the response sentence similar to the character response sentence completely matches (S1402).
6, when it is determined that they do not match, it is determined whether the different part (word or word string) is an essential requirement by referring to the list 1201 of the response sentence database 1101 (S1404). If it is not an essential requirement, the process moves to S1416.
It is determined whether or not there is a possibility of replacement by referring to (S140)
6).

When it is determined that replacement is not possible, it is determined that the question message has not been transmitted to the conversation partner, and the voice output unit 110
2 is instructed to ask a question again, and the process returns to S914 of the first embodiment.

When it is determined that replacement is possible, it is determined whether or not the word of the character response sentence matches the replacement condition (S1).
408) If it matches, the process moves to S1416. If not, the voice output unit 1102 is notified to output a predetermined voice according to the re-question rule.

The voice output unit 1102 outputs a voice according to the re-question rule and transmits it to the conversation partner (S1410). A re-response from the conversation partner is received from the voice input unit 109 or the input operation unit 103 (S1412).

The voice recognition unit 110 recognizes voice input from the voice input unit and converts a word or a word string into the response sentence extraction unit 1.
Then, the input operation unit 103 notifies the response sentence extraction unit 1103 of the input word (S1414).

The response sentence extracting unit 1103 displays a response sentence similar to the display unit 105, its translated response sentence, and a character response sentence as a recognition result. At this time, the display attribute of the translation response sentence that matches the character response sentence is changed (S141).
6).

In this way, even if the voice response from the conversation partner cannot be recognized once, a re-question according to the content is prepared, and an appropriate input is received from the conversation partner, and The accuracy of response sentence recognition can be improved.

In the above embodiment, Japanese is used as the first language and English is used as the second language. However, in the interpreter according to the present invention, the first language and the second language are reversed. Or another language is the first or second language
It may be a language. In this case, the question sentence translation database 101, the voice output units 106 and 1102, the response sentence databases 102 and 1101, and the speech recognition dictionary 10
8 is the content corresponding to each language.

Further, in the above embodiment, FIGS.
Although each unit shown in the configuration diagram shown in FIG. 1 has performed its function, a computer-readable floppy disk or a CD-ROM can be used to execute a program that performs the function of each unit.
It is recorded on a recording medium such as a ROM. Attach this recording medium to a portable terminal device or the like that does not have the function unique to this interpreter,
An interpreting device having the same function as the present device can be provided.

[0092]

As described above, according to the present invention, a question sentence database in which a question sentence in a first language and a question sentence in a second language obtained by translating the same are stored in pairs, A response sentence database storing a plurality of standard second language response sentences responding to the question sentence, and a first language response sentence translated therefrom; an input operation means; a voice input means;
A question sentence extracting means for searching the question sentence database for a question sentence in a first language inputted by a user with the input operation means and extracting a paired question sentence in a second language; Voice recognition means for recognizing the input voice as a text response in a second language when a conversation partner of the user inputs a voice in a second language in response to the output of the question text in two languages; A response sentence extracting unit that extracts a response sentence in a second language similar to the character response sentence recognized by the voice recognition unit and a corresponding response sentence in the first language from the response sentence database; The second
Display means for displaying a response sentence in a language, a response sentence in a first language, and a character response sentence recognized by the voice recognition means are provided. With such a configuration, since the user's conversation partner simply responds by voice so as to have a normal conversation, the user is presented with a response sentence that answers the question in a first language that the user can understand. You can have a smooth conversation.

The speech recognition means analyzes a speech input from the speech input means by analyzing a speech recognition dictionary storing a word standard pattern and a word corresponding to the word standard pattern. A division unit that divides the pattern into segment patterns; and calculates a similarity between the pattern of the segment divided by the division unit and a word standard pattern stored in the speech recognition dictionary. A candidate word extraction unit for extracting a word corresponding to the pattern as a candidate word; and a character response sentence generation unit for generating a character response sentence by arranging candidate words having the highest similarity of each segment extracted by the candidate word extraction unit. I have it. With such a configuration, the response sentence of the conversation partner can be recognized as a character response sentence.

When the extracted candidate word matches the word constituting the response sentence extracted by the response sentence extraction means, the candidate word extraction unit increases the similarity at a predetermined magnification. . With such a configuration, the recognition accuracy of a voice response sentence can be improved by increasing the probability of recognizing a word predicted as a response sentence even when the voice of an unspecified conversation partner is used. Further, the response sentence database includes information indicating whether a word constituting the response sentence is an essential element as a response sentence, and whether the word is expected to be replaced with another word when the word is an essential element. When expected, the semantic attribute of the word expected to be replaced is stored, and the words constituting the character response sentence recognized by the voice recognition means are compared with the words constituting the similar response sentence. A character response sentence determination unit that determines that a character response sentence does not hold as an answer to a question when a word lacking an essential element and having a semantic attribute different from the meaning attribute of the word to be replaced is set as a word corresponding to the response sentence Is to be provided. With such a configuration, it is determined whether or not the character response sentence recognized by the voice recognition unit is established as an answer to the question sentence.

Further, in the response sentence database, when a word constituting the response sentence is an essential element, a re-question rule when the word cannot be recognized is stored. When the means determines that the character response sentence does not hold as the answer to the question, it includes an output means for outputting a re-question in accordance with the re-question rule. With such a configuration, a character response sentence that does not hold the answer to the question sentence can be a correctly recognized character response sentence by asking the conversation partner again.

Further, the words constituting the character response sentence displayed on the display means are compared with the words constituting the response sentence in the second language extracted by the response sentence extraction means, and the words of the matching words are compared. Control means for changing the display attributes of the bilingual response sentence is provided. With such a configuration,
It is easy to understand the response sentence from the conversation partner.

[0097] Corresponding words constituting the response sentence in the first language and the response sentence in the second language stored in the response sentence database are given the same identifier and displayed on the display means. The words constituting the character response sentence are compared with the words constituting the response sentence in the second language extracted by the response sentence extracting means, and the response sentence in the first language to which the same identifier as the matching word is assigned And a display control means for changing the display attribute. With such a configuration, it is possible to clarify the difference between the response sentence of the second language from the conversation partner and the content of the first language that the user can understand.

Further, a computer-readable recording medium, which stores a question sentence in a first language and a question sentence in a second language translated from the first sentence, A response sentence database storing a plurality of standard second language answer sentences responding to the question sentence of the above and a translated answer sentence of the first language, and input operation means; Input means, and a question sentence extracting means for searching the question sentence database for a question sentence in a first language input by a user with the input operation means, and extracting a question sentence in a second language paired with the query sentence database; Speech recognition for recognizing the input voice as a text response sentence in the second language when the conversation partner of the user inputs the voice response in the second language in response to the output of the extracted question text in the second language Means and recognition by the voice recognition means Response sentence extracting means for extracting a response sentence of a second language similar to the extracted character response sentence and a corresponding response sentence of the first language from the response sentence database, and a second language extracted by the response sentence extracting means And a display control unit for displaying the response sentence of the first language, the response sentence of the first language, and the character response sentence recognized by the voice recognition unit. Thereby, a portable terminal device without an interpreting function can be used as an efficient interpreting device.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a translator according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of the contents of a question sentence translation database of the embodiment.

FIG. 3 is a diagram showing an example of the contents of a response sentence database according to the embodiment.

FIG. 4 is a diagram showing an example of contents displayed on a display unit of the embodiment.

FIG. 5 is a diagram showing another example of the content displayed on the display unit of the embodiment.

FIG. 6 is a diagram showing an example of the contents of a speech recognition dictionary according to the embodiment.

FIG. 7A shows an example of a character response sentence of a voice response input to the voice input unit of the embodiment. (B)
FIG. 4 is an explanatory diagram illustrating a state in which the speech recognition unit of the embodiment recognizes word candidates by dividing the speech signal into speech patterns of respective segments in word units.

FIG. 8 is an explanatory diagram showing how the speech recognition unit of the embodiment corrects the similarity of word candidates and recognizes a character response sentence.

FIG. 9 is a flowchart illustrating the operation of the embodiment.

FIG. 10 is a flowchart illustrating a detailed operation of S918 in FIG. 9;

FIG. 11 is a configuration diagram of a translator according to a second embodiment of the present invention.

FIG. 12 is a diagram showing an example of the contents of a response sentence database according to the embodiment.

FIG. 13 is a diagram illustrating an example of content displayed on the display unit of the embodiment.

FIG. 14 is a flowchart illustrating an operation of the embodiment.

[Explanation of symbols]

 101 Question sentence translation database 102, 1101 Response sentence database 103 Input operation unit 104 Question sentence selection unit 105 Display unit 106, 1102 Speech output unit 107, 1103 Response sentence extraction unit 108 Speech recognition dictionary 109 Speech input unit 110 Speech recognition unit

Claims (8)

[Claims] 1. A question sentence in a first language and a second translated sentence
A question sentence database storing a pair of language question sentences, a plurality of standard second language answer sentences responding to the second language question sentence, and a first language answer sentence translated therefrom , An input operation means, a voice input means, and a first sentence input by a user in the input operation means. A question sentence extracting means for extracting a question sentence in the second language, and when the conversation partner of the user receives the output of the extracted question sentence in the second language and inputs a response sentence in the second language by voice, Voice recognition means for recognizing an input voice as a character response sentence in a second language; and a response sentence in a first language corresponding to a response sentence in a second language similar to the character response sentence recognized by the voice recognition means; From the response sentence database Response sentence extraction means; and display means for displaying a response sentence in a second language, a response sentence in a first language, and a character response sentence recognized by the voice recognition means, extracted by the response sentence extraction means. Interpreting device characterized by the above-mentioned. 2. The speech recognition means, comprising: a speech recognition dictionary that stores a word standard pattern and a word corresponding to the word standard pattern; A division unit that divides the segment pattern into segment patterns; and calculates a similarity between the segment pattern divided by the division unit and a word standard pattern stored in the speech recognition dictionary. A candidate word extraction unit for extracting a word corresponding to the pattern as a candidate word; and a character response sentence generation unit for generating a character response sentence by arranging candidate words having the highest similarity of each segment extracted by the candidate word extraction unit. The interpreting device according to claim 1, further comprising: 3. The candidate word extracting section increases the similarity at a predetermined magnification when the extracted candidate word matches a word constituting the response sentence extracted by the response sentence extracting means. 3. The translator according to claim 2, wherein: 4. In the response sentence database, it is expected that a word constituting a response sentence is an essential element as a response sentence, and that when the word is an essential element, the word is replaced with another word. And the semantic attribute of the word that is expected to be replaced when it is expected, and the word constituting the character response sentence recognized by the voice recognition means and the word constituting the similar response sentence are stored. A character response that determines that a character response sentence does not hold as the answer to a question when a word that lacks an essential element and has a semantic attribute different from the semantic attribute of the word that is expected to be replaced is the word corresponding to the response sentence 2. The interpreter according to claim 1, further comprising a sentence judging unit. 5. The response sentence database stores, when a word constituting the response sentence is an essential element, a re-question rule when the word cannot be recognized. The interpreting apparatus according to claim 4, further comprising an output unit that outputs a re-question in accordance with the re-question rule when the means determines that the character response sentence is not satisfied as an answer to the question. 6. A word constituting a character response sentence displayed on the display means is compared with a word constituting a response sentence in a second language extracted by the response sentence extraction means, and the second word of the matching word is compared. 6. The interpreting apparatus according to claim 1, further comprising control means for changing a display attribute of a bilingual response sentence. 7. The same identifier is assigned to a corresponding word constituting a response sentence in a first language and a response sentence in a second language stored in the response sentence database and displayed on the display means. The words constituting the character response sentence are compared with the words constituting the response sentence in the second language extracted by the response sentence extracting means, and the response sentence in the first language to which the same identifier as the matching word is assigned The interpreting apparatus according to claim 1, further comprising a display control unit configured to change a display attribute of the interpreter. 8. A computer-readable recording medium, comprising: a question sentence database storing a pair of a first language question sentence and a translated second language question sentence; A response sentence database that stores a plurality of standard second-language response sentences that respond to the question sentence in question and a first-language response sentence translated from them in advance, Input means, a question sentence extracting means for searching the question sentence database for a question sentence in a first language input by a user with the input operation means, and extracting a paired question sentence in a second language; Speech recognition for recognizing the input voice as a text response sentence in the second language when the conversation partner of the user inputs the voice response in the second language in response to the output of the extracted question text in the second language Means and recognition by the voice recognition means Response sentence extracting means for extracting from the response sentence database a response sentence in a second language similar to the extracted character response sentence and a response sentence in the first language, and a second language extracted by the response sentence extracting means And a display control unit for displaying the response sentence of the first language, the response sentence of the first language, and the character response sentence recognized by the voice recognition unit.