JPH11102196A - Speech interactive system, method of speech interaction, and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech interactive system, a method of speech interaction, and a storage medium therefor inducing user's utterance. SOLUTION: When interaction is started, a language processing part 3 generates an output candidate sentence, a recognition candidate sentence, from expressions and sentence patterns referring to a sentence template dictionary 4, task knowledge data 5, and topic data 6. A voice synthesis part 2 outputs one of the generated output candidate sentences. When a speech is inputted in response to this, the speech recognition part 1 recognizes the recognition candidate sentence with priority generated from the expressions and sentence patterns, and sends the recognized word of the recognition result to the language processing part 3. The language processing part 3 generates the following output candidate sentence and the recognition candidate sentence based on this recognition result. When no speech is inputted or no recognition result is obtained, the following output candidate sentence, etc., are generated accordingly. Interaction is carried out by repeating these.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a spoken dialogue system, a spoken dialogue method, and a recording medium.
The present invention relates to a voice dialogue system for guiding a user's utterance, a voice dialogue method, and a recording medium.

[0002]

2. Description of the Related Art In recent years, research and trial production of a voice dialogue system for exchanging information stored in a machine by voice have been active. Since voice is the most natural input / output means of humans, it is also attracting attention as information input / output means.
However, when a human exchanges information with a machine by voice, the speed and timing of the human utterance and the words used affect a smooth dialogue. Therefore, a technology for changing the response of the machine side according to the speed of human speech has been developed. For example,
The technique disclosed in Japanese Patent Application Laid-Open No. Hei 5-173589 is to prepare a plurality of response sentences having the same meaning but different numbers of characters, and change the response sentences according to the speed of the input utterance.

[0003]

However, this technique is
Since it is only a strategy to change the response according to the user's utterance speed in response to the user, if the utterance speed of the user is very fast or slow, the system side may exceed the limit that can be adjusted.

[0004] It is known that when a human talks with each other, the opponent works according to the tempo and language of the opponent. In order to carry out the dialogue smoothly, it is only necessary to match the user from the machine side. Instead, it is necessary to guide the user's utterance in a natural manner. In this way, even if the capability of the machine is limited, the conversation can be smoothly performed.

[0005] In addition, in the case of speech recognition, which is a component of a dialog system, recognizing not only keywords related to a task but also words such as particles and auxiliary verbs before and after the task leads to a higher recognition rate. . From this, it is effective to guide and predict the utterance of the user.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described circumstances, and performs voice output in consideration of ease of utterance and word order to a user, and outputs utterances from the user by the system. A voice dialogue system, a voice dialogue method, and a recording medium for performing recognition based on the expressed expression.

[0007]

According to a first aspect of the present invention, an output candidate sentence and a recognition candidate sentence of a predetermined expression / sentence pattern are generated in a speech dialogue system for outputting a sentence or word and recognizing an input voice. Means, a means for outputting one of the output candidate sentences by voice, and a means for setting a recognition candidate word or a recognition candidate expression when recognizing a user's input voice, based on the expression / sentence pattern of the output sentence output by voice. Characterized by having
Thus, the user's utterance is guided, and the same expression / sentence pattern as the speech-outputted expression / sentence pattern is preferentially recognized during the user's utterance so that the speech dialogue can be smoothly performed.

According to a second aspect of the present invention, in the first aspect, the preset expression / sentence type output candidate sentence and recognition candidate sentence are set to a predetermined specific condition expression / sentence type output candidate sentence and recognition candidate sentence. The feature is that the sentence is a sentence, so that the expression / sentence pattern that is particularly easy to speak is induced more strongly, and the recognition rate is improved.

[0009] The invention of claim 3 is characterized in that, in the invention of claim 2, a word representing a current topic is positioned at the beginning of the sentence as the specific condition, and the topics that have become topics are arranged in word order. By outputting the above sentence, the user is strongly aware of the topic, and more efficient dialogue is enabled.

A fourth aspect of the present invention is characterized in that, in the second aspect of the present invention, an assertive auxiliary verb is used as the specific condition, thereby increasing the possibility that the user uses the same expression / sentence pattern. This enables efficient dialogue.

[0011] The invention according to claim 5 outputs a sentence or a word,
A means for generating an expression / sentence-type output candidate sentence and a recognition candidate sentence in which the latter half or the end of the sentence is omitted, a means for outputting one of the output candidate sentences by voice, According to the expression / sentence pattern of the output sentence in which the latter half of the output sentence or the end of the sentence is omitted, a word that can be connected to the latter half or the end of the omitted sentence is used as a recognition candidate word or recognition candidate expression when recognizing the input speech of the user. It is characterized by having means for setting, thereby strongly inducing the user's utterance,
It has enabled more efficient dialogue.

[0012] The invention according to claim 6 outputs a sentence or a word,
In a voice interaction method for recognizing an input voice and performing a voice dialogue, an output candidate sentence and a recognition candidate sentence of a preset expression / sentence are generated, one of the generated output candidate sentences is output as a voice, and the voice output is output. According to the sentence expression / sentence pattern, preferentially recognize a recognition candidate word or a recognition candidate expression when recognizing a user's input voice, and generate a next output candidate sentence and a recognition candidate sentence based on the recognition result. An object of the present invention is to provide a voice interaction method in which voice interaction is repeatedly performed.

According to a seventh aspect of the present invention, there is provided a computer-readable recording medium in which relevant data and a program for implementing the voice interactive system or the interactive method according to any one of the first to sixth aspects are recorded.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1A is a block diagram showing the overall configuration of a speech dialogue system according to the present invention, wherein 1 is a speech recognition unit, 2 is a speech synthesis unit, and 3 is a language processing unit. 4 is a sentence template dictionary, 5 is task knowledge data, and 6 is topic data. The speech recognition unit 1 analyzes the characteristics of the input speech, collates the recognition candidates created by the language processing unit 3 sequentially during speech input, and passes the recognition result to the language processing unit 3. The voice synthesizer 2 outputs the utterance sentence created by the language processor 3 as voice.

The language processing unit 3 converts the task knowledge data from the sentence template dictionary 4 into a template (sentence pattern) corresponding to the current dialogue scene based on the utterance type based on the utterance type, topic, and next utterance type written in the topic data 6. 5 are embedded to create a recognition candidate sentence (recognition candidate word string) and an output candidate sentence (output candidate word), which are passed to the speech recognition unit 1 and the speech synthesis unit 2.

FIG. 1B is a processing flow for explaining the outline of the processing of the voice interaction system of the present invention.

When the conversation is started, the language processing unit 3 generates an output candidate sentence and a recognition candidate sentence (S1). here,
The speech synthesis unit 2 outputs one of the output candidate sentences to the user (S2). Depending on whether or not there is a voice input from the user with respect to this output sentence (S3), if there is no voice input, the process returns to the generation of output candidate sentences (NO in S3), and the next output candidate sentence is generated, If there is a voice input (YES in S3), a recognition candidate sentence having the same template number as the output candidate sentence is preferentially recognized during the voice input (S4). If recognized (S5
YES), the speech recognition unit 1 collates the recognition candidate sentence during speech input, and passes a recognition word as a recognition result to the language processing unit 3. The language processing unit 3 determines whether or not the task has been completed based on the recognition result (S7), and determines whether or not the task has been completed (S7).
7) generates the next output candidate sentence and recognition candidate sentence. If no recognition result is obtained (N in S5)
O) recognizes another recognition candidate sentence (S6) and generates the next output candidate sentence and recognition candidate sentence or ends the task depending on whether the task has been completed (S7). Generally, these are repeated, and the dialog ends when the task ends. The speech dialogue system of the present invention is characterized in that a recognition candidate word for speech recognition is set using a template having the same template number as the template used for speech output by the system. Here, the unit of recognition and synthesis may be a word string or a sentence.

FIG. 2A is an example of a sentence template dictionary for explaining the invention of claim 1, which is a template describing a speech type and a sentence pattern representing a scene of a dialogue. Consists of numbers. The utterance type is like a task on which the dialogue is performed or a scene on the dialogue. It describes the meaning of words spoken in a certain scene and their word order in association with each other. The content changed by the scene is expressed as a variable by (), and a semantic category is described.

FIG. 2B shows an example of task knowledge data for explaining the first aspect of the present invention, which comprises semantic categories related to tasks and words representing the categories. In the example of the above sentence template dictionary, in the question scene, the word corresponding to the semantic category "event" of the task knowledge data is located in the "(event)" part, followed by "ha""when"" Is connected. The template number is a number assigned to each utterance type divided according to expression or word order. Those having the same number indicate that they have the same word order or expression. The template at the top of the sentence template dictionary has a template number of “1”. For example, a template having the same template number of another utterance type is used for creating a recognition candidate. The template number is
If there are a plurality of corresponding templates, a plurality of templates can be described.

FIG. 2C shows an example of topic data for explaining the invention of claim 1, which is composed of a current utterance type, a topic, and a next utterance type. A template is selected from the sentence template dictionary according to the utterance type, and the topic is embedded in the template. The next utterance type represents a scene next to the current utterance type, and is used in the present invention to select a template for setting a recognition candidate word when recognizing a user's utterance after system output. The topic describes zero or more pairs of items and words that represent a topic in the current utterance scene. Embed this topic in the template,
Generate a statement.

The flow of processing will be described with reference to a voice dialogue system for performing schedule management and registration. First, the language processing unit 3 generates an output candidate sentence and a recognition candidate sentence. Since the utterance type of the topic data is “question”, output candidates are generated based on the template of the utterance type “question” in the sentence template dictionary. The "conference" of the topic data is embedded in the "(event)" part of the template having the item "(event)" in the topic column of the topic data, and a sentence such as "when is there a meeting" is generated.

At the same time, since the next utterance type of the topic data is “response”, the value of (date) of the task knowledge data is embedded in the “response” template of the sentence template dictionary, and “the meeting is on July 24” A word string or a sentence such as “It is on July 24”, “The meeting is on next Monday” and “It is on next Monday” is generated as a recognition candidate.

Here, one of the output candidate sentences, for example, "When is there a meeting?" Thereafter, when there is a user's voice input, the voice recognition unit 1 checks the recognition candidate sequence created by the language processing unit 3 in the input voice. The recognition candidate sequence preferentially matches a recognition candidate sentence or a recognition candidate sequence created with the same template as the template number of the sentence output.

For example, the system can output a sentence such as "When is a meeting?" And handle a user's utterance such as "I am on next Monday." When the user speaks a different phrase and the word is not recognized during the voice output, another recognition candidate sentence is collated. The above is repeated until the task is completed.

(Invention of Claim 2) FIG. 3 is an example of a sentence template dictionary for explaining the invention of claim 2. The format is the same as the sentence template dictionary of the first aspect of the present invention.

According to a second aspect of the present invention, the user's utterance is guided by outputting an expression that is easy to speak or an expression that easily induces the utterance, and is set as a recognition candidate. For example, the lower the template number, the easier the utterance or the utterance guidance. In the example of the present invention, a word corresponding to the semantic category “event” of the task knowledge data is located in the part where “utterance type” is “question” and “(event)” as template number “1” data. After that, "ha" is connected, followed by "when", and finally "is" is connected. Here, the template number is “1” as an expression that makes it easy to speak a sentence pattern such as
And The task knowledge data and topic data used in the second aspect of the invention are the same as those of the first aspect.

Here, the flow of processing will be described with reference to a voice dialogue system for performing schedule management and registration. The language processing unit 3 generates an output candidate sentence and a recognition candidate sentence. Since the utterance type of the topic data is "question", among the templates whose utterance type in the sentence template dictionary is "question", the template number 1 is used, and the information written in the task knowledge data is embedded. When is it? " At the same time, since the next utterance type of the topic data is "response", the task knowledge data is embedded in the template whose utterance type of the sentence template dictionary is "response", and "the meeting is July 24""7
A word string or a sentence, such as “Month 24”, “Meeting is next Monday”, “Next Monday”, or a sentence is generated as a recognition candidate. Next, the generated sentence "When is the meeting?" After this, if there is a user's voice input,
The speech recognition unit 1 checks the recognition candidate sentence created by the language processing unit 3 in the input speech. The recognition candidate sentence preferentially matches a recognition candidate sentence or a recognition candidate sequence created with one template having the same utterance type as the template number of the sentence output as speech.

For example, utterances such as the system output "When is the meeting?" And the user's "Next Monday is coming" or "Meeting is next Monday" can be handled. If the user speaks a different phrase and the word is not recognized during the voice output, another recognition candidate sentence is checked. The above is repeated until the task is completed.

(Invention of Claim 3) FIGS. 4A and 4
FIGS. 4B and 4C are diagrams showing examples of a sentence template dictionary, task knowledge data, and topic data, respectively, for explaining the invention of claim 3. Claim 1
Is the same as In the topic data of FIG. 4C, when there are a plurality of descriptions in the topic column, it is assumed that the left-most matter is important as the topic.

The invention according to claim 3 is characterized in that the system is designed to make the user speak about a specific topic by giving the output sentence on the system side the current topic first in word order. Also in the setting of the recognition candidate word, a sentence template in which the current topic is arranged in the word order first is preferentially set as a matching target.

The flow of processing will be described with reference to a voice dialogue system for performing schedule management and registration. The language processing unit 3 generates an output candidate sentence and a recognition candidate sentence. Since the utterance type of the topic data is "question", templates having an utterance type of "question" in the sentence template dictionary that have "(event)" at the beginning of the subject at the left end of the topic column of the topic data Select The contents of the topic data are embedded in the template of template number 1 in which "(event)" precedes the word order, and a sentence such as "Is the meeting tomorrow?" At the same time, since the next utterance type of the topic data is "response", a template whose utterance type is "response" is selected in the sentence template dictionary. Embed the value of task knowledge data in the template to create a sentence.
"The meeting is July 24.""The meeting is next Monday."
"July 24 is a meeting.""Next Monday is a meeting."
Is generated as a recognition candidate.

Next, the output sentence "Is the meeting tomorrow?" Thereafter, when there is a user's voice input, the voice recognition unit 1 checks the recognition candidate sequence created by the language processing unit 3 in the input voice. The recognition candidate string preferentially matches the recognition candidate sentences “meeting is on July 24” and “meeting is next Monday” created with the same template number 1 as the template number of the sentence output.

For example, if the system is "Is the meeting tomorrow?"
Is output, the user can handle an utterance such as "Meeting is next Monday." When the user speaks a different phrase and the word is not recognized during the voice output, another recognition candidate sentence is collated. The above is repeated until the task is completed.

The phrase "meeting is tomorrow" and another word order "tomorrow is meeting" is a very similar expression, but it can inform the user that the word located at the beginning of the sentence is a topic, The likelihood of diverting to other topics is reduced.

(Invention of Claim 4) FIGS. 5A and 5
FIGS. 5B and 5C show examples of a sentence template dictionary, task knowledge data, and topic data, respectively, for explaining the invention of claim 4. The data format of the sentence template dictionary is claim 1. The data formats of the task knowledge data and topic data are the same as those of the third aspect.

The invention of claim 4 is characterized in that a sentence pattern using the auxiliary verb "is" is used as an output sentence of the system, so that the user is made to speak in the same sentence pattern. The template number of the template using the auxiliary verb “is” is “1”.

The flow of processing will be described by taking a voice dialogue system for schedule management and registration as an example. The language processing unit 3 generates an output candidate sentence and a recognition candidate sentence. Since the utterance type of the topic data is “question”, a template whose utterance type in the sentence template dictionary is “question” is selected. In particular, the template of template number 1 which is a sentence pattern using the auxiliary verb "is" is used. Using the template of template number 1, embedding topic data items,
Create an output statement. Generate a sentence such as "When is the meeting?" At the same time, in the topic data, since the next utterance type is “response”, the task knowledge data is embedded in a template whose utterance type is “response” to create a recognition candidate sentence. A word string or a sentence such as “Is the meeting tomorrow?” Or “Is there a meeting tomorrow?” Is generated as a recognition candidate.

Next, the output sentence "Is the meeting tomorrow?" Thereafter, when there is a user's voice input, the voice recognition unit 1 checks the recognition candidate sequence created by the language processing unit 3 in the input voice. The recognition candidate sequence preferentially matches a recognition candidate sentence or a recognition candidate sequence created with one template, which is the same as the template number of the sentence output as speech. For example, when the user speaks a different phrase and the word is not recognized during the voice output, another recognition candidate sentence is collated.
The above describes the utterance type, topic data,
Repeat until the task is completed based on the task knowledge data.

Sentence patterns such as "A is B" and "B is" are sentence patterns that can be used regardless of the semantic consistency of A and B, and are easily uttered.
By using this sentence pattern in the system output, there is a high possibility that the user will respond with the same sentence pattern. Further, by using this sentence pattern, it is possible to guide and predict the user's utterance and wording.

(Invention of claim 5) FIGS. 6A and 6
FIGS. 6B and 6C are diagrams showing examples of a sentence template dictionary, task knowledge data, and topic data, respectively, for explaining the invention of claim 5, and the data format of the sentence template dictionary is as follows. The data formats of the task knowledge data and the topic data are the same as those of the third aspect.

According to a fifth aspect of the present invention, with respect to a pair of the utterance type and the next utterance type in the same line of the topic data with the same template number, for example, "question" and "response", the utterance type "question" indicates the contents to be asked. In the utterance type "response", an item to be the answer is set at the beginning of the sentence. In other words, the setting is made such that connecting the templates of the same utterance type with the same template number results in one sentence.

The flow of processing will be described using a voice dialogue system for managing and registering a schedule as an example. The language processing unit 3 generates an output candidate sentence and a recognition candidate sentence. Since the utterance type of the topic data is “question”, a template having the utterance type of “question” is selected from the template dictionary, and the sentence is generated by embedding the word “go out”. Also, a recognition candidate sentence is generated. Since the utterance type of the topic data is “response”, the utterance type of the template dictionary is “response”, and
The value of the task knowledge data is generated by embedding the value of the task knowledge data in the template having the same template number “1”. A sentence such as "It is tomorrow", "It is the day after tomorrow", "It is from 10:00" is generated as a recognition candidate sentence.

Next, the output sentence "going out" is output as voice.
Thereafter, when there is a user input, the voice recognition unit 1 checks the recognition candidate sentence created by the language processing unit 3 in the input voice. For example, the user can recognize an utterance such as "Tomorrow" for a sentence such as "Going out" in the system. The above is repeated until the task is completed.

In the present embodiment, using the nature of supplementing the continuation of the utterance of the other party in daily human conversation, the present embodiment is designed to cause the user to respond to what he wants to respond to at the beginning of the sentence. Utterance and wording can be induced and accurately predicted. In particular, by uttering a key word on the task at the beginning of the sentence, dialog management and task achievement can be performed more reliably.

(Invention of claim 6) A sentence or word is output,
In a spoken dialogue method for recognizing an input speech and performing a spoken dialogue, first, an output candidate sentence and a recognition candidate sentence are generated according to an expression / sentence pattern set in advance by the system, and one of the generated output candidate sentences is output as speech. . Voice input is performed in response to the voice output. The system side preferentially recognizes a recognition candidate word or a recognition candidate expression when recognizing the input voice based on the expression / sentence pattern of the output sentence output by the previous voice, and based on the recognized word or expression, the next output candidate. A sentence and a recognition candidate sentence are generated, and one of the generated output candidate sentences is output as speech.
This is an interactive method that repeats the above until the task is completed.

(Invention of claim 7) In the above-described embodiment of the voice dialogue system and the dialogue method according to the present invention, the expression / sentence dictionaries as information to be stored in each part of the voice dialogue system to perform voice dialogue. The control function for executing the processing operation is recorded (held) as data on a recording medium, the recorded data is read, the device is operated, and the method is performed. Used to execute. As described above, the data for executing the voice dialogue system and the dialogue method of the present invention is stored in the recording medium, and the functions of the voice dialogue system and the method can be realized by a general-purpose computer on which the recording medium is installed.

[0047]

According to the first aspect of the present invention, in a voice interactive system for outputting a sentence or word and recognizing an input voice, means for generating an output candidate sentence and a recognition candidate sentence of a predetermined expression / sentence type are provided. A means for outputting one of the output candidate sentences by voice, and a means for setting a recognition candidate word or a recognition candidate expression when recognizing a user's input voice based on the expression / sentence pattern of the output sentence output by voice. In addition to guiding the user's utterance by the output of the system, efficient and accurate speech recognition can be achieved by preferentially recognizing the expression and sentence pattern that is the same as the sentence output during the user's utterance. Even if the system has limited capabilities, the dialogue can proceed smoothly.

According to the second aspect of the present invention, in addition to the effects of the first aspect, the preset expression / sentence type output candidate sentence and the recognition candidate sentence are set to a predetermined specific condition expression / sentence.
Since the sentence-type output candidate sentence and the recognition candidate sentence are used, an utterance expression can be more strongly induced by making the expression / sentence type particularly easy to speak, and the recognition rate can be improved.

According to the third aspect of the present invention, in addition to the effect of the second aspect of the present invention, the phrase representing the current topic is located at the beginning of the sentence as the specific condition, so that the topical matters are arranged in word order. By outputting the sentence as above, the user can be strongly aware of the topic and can utter the utterance sentence of the topic, and the recognition candidate word is set based on it, so that more efficient dialogue Can do it.

According to the fourth aspect of the present invention, in addition to the effect of the second aspect of the present invention, since the assertive auxiliary verb is used as the specific condition, the sentence pattern by the auxiliary verbs "da" and "da" is presented. Thus, the user is more likely to use the same expression / sentence pattern, the recognition candidate words can be set with high accuracy, and more efficient dialogue can be performed.

According to the fifth aspect of the present invention, in a voice dialogue system for outputting a sentence or a word and recognizing an input voice, an output candidate sentence and a recognition candidate sentence of an expression / sentence type in which the latter half or the end of the sentence is omitted are generated. Means for performing voice output of one of the output candidate sentences, and a recognition candidate word for recognizing a user's input voice by using an expression / sentence pattern of an output sentence in which the latter half or the end of the sentence is omitted. Alternatively, the recognition candidate expression has means for setting a phrase that can be connected to the second half of the omitted sentence or the end of the sentence, so that the user's utterance can be guided, and the setting of the recognition candidate word can be performed with high accuracy, and more efficient Can interact.

According to the sixth aspect of the present invention, since the system predicts and guides the utterance of the user, it is possible to perform a smooth dialogue even when the system has a limited capacity.

According to the invention of claim 7, claims 1 to 6 are provided.
A computer-readable recording medium on which is recorded the relevant data and the program for implementing the voice interaction system or the interaction method according to any one of the above.

[Brief description of the drawings]

FIG. 1 is a diagram showing an overall configuration and a processing flow of a voice interaction system of the present invention.

FIG. 2 is a diagram showing an example of a sentence template dictionary and the like for explaining one embodiment of the present invention.

FIG. 3 is a diagram showing an example of a sentence template dictionary for explaining another embodiment of the present invention.

FIG. 4 is a diagram showing an example of a sentence template dictionary and the like for explaining still another embodiment of the present invention.

FIG. 5 is a diagram showing an example of a sentence template dictionary and the like for explaining still another embodiment of the present invention.

FIG. 6 is a diagram showing an example of a sentence template dictionary and the like for explaining still another embodiment of the present invention.

[Description of Signs] 1 ... Speech recognition unit, 2 ... Speech synthesis unit, 3 ... Language processing unit, 4
... sentence template dictionary, 5 ... task knowledge data, 6 ... topic data.

Claims (7)

[Claims] In a voice interactive system for outputting a sentence or a word and recognizing an input voice, a means for generating an output candidate sentence and a recognition candidate sentence of a preset expression / sentence type, and one of the output candidate sentences is provided. A speech dialogue system comprising: means for outputting a speech; and means for setting a recognition candidate word or a recognition candidate expression for recognizing a user's input speech based on the expression / sentence pattern of the output sentence output as a voice. 2. The method according to claim 1, wherein the preset expression / sentence type output candidate sentence and the recognition candidate sentence are a predetermined specific condition expression / sentence type output candidate sentence and a recognition candidate sentence. A spoken dialogue system as described. 3. The spoken dialogue system according to claim 2, wherein a phrase representing a current topic is located at the beginning of the sentence as the specific condition. 4. The spoken dialogue system according to claim 2, wherein an assertive auxiliary verb is used as the specific condition. 5. A means for outputting a sentence or word and recognizing an input voice, generating an output candidate sentence and a recognition candidate sentence of an expression / sentence type in which the latter half or the end of the sentence is omitted, and said output candidate Omitting the recognition candidate word or recognition candidate expression when recognizing the user's input voice by means of outputting one of the sentences by voice and the expression / sentence pattern of the output sentence in which the latter half or the end of the sentence is omitted. A speech dialogue system comprising means for setting a connectable phrase at the second half or at the end of a sentence. 6. A spoken dialogue method for outputting a sentence or word, recognizing an input speech and performing a spoken dialogue, generating an output candidate sentence and a recognition candidate sentence of a predetermined expression / sentence type, and generating the generated output candidate sentence. One of the voice output, by the expression and sentence pattern of the output sentence that is voice output, preferentially recognize the recognition candidate word or recognition candidate expression when recognizing the input voice of the user,
A voice interaction method for performing a voice interaction by repeatedly generating a next output candidate sentence and a recognition candidate sentence based on a recognition result. 7. A computer-readable recording medium on which relevant data and a program for implementing the voice dialogue system or dialogue method according to claim 1 are recorded.