JP5166503B2 - Interactive device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an interaction device capable of keeping interaction natural and controlling interaction content. <P>SOLUTION: An interaction order storage unit stores plural patterns of interaction sentences. An output unit outputs a first output sentence. A first reception unit receives input of an input sentence. A second reception unit receives input of a question sentence. A first probability calculation unit calculates, on the basis of the patterns, a first probability indicating a probability of outputting an interaction sentence identical or similar to the first output sentence and then outputting each of first interaction sentences that is any of the interaction sentences stored in the interaction sentence storage unit. A second probability calculation unit calculates, on the basis of the patterns, a second probability indicating a probability of outputting an interaction sentence identical or similar to the question sentence two steps after each of the first interaction sentences. A selection unit selects a first interaction sentence having the largest value obtained by weighted-adding the first probability and the second probability as a second output sentence. <P>COPYRIGHT: (C)2012,JPO&amp;INPIT

Description

  Embodiments described herein relate generally to an interactive apparatus that interacts with a user.

  2. Description of the Related Art Conventionally, an apparatus that collects information from a system when a user interacts with the system using text sentence input or voice input, and conversely, an apparatus that collects information from a user by a system is known. In a device that collects information from the system, it is necessary for the system to analyze the intention of the user's input and present the information according to the intention, but it is difficult to analyze the intention of the input from natural dialogue .

  Therefore, a general dialogue system employs a method in which the system side presents choices and the user selects one of them to convey the intention to the system. In addition, many methods using options are also adopted for the apparatus for collecting information from the user by the system.

  On the other hand, as a method for collecting information from the system by the user or a method for collecting information from the user, a method using a dialogue chain of dialogue history of prepared natural dialogue sentences is disclosed. For example, a method has been proposed in which the system creates a response sentence using a conversation history of a natural conversation sentence, or guesses a response from a user who has not been heard.

Japanese Patent No. 4246548

  However, in the conventional method, it is difficult to control the content of the dialogue while maintaining the naturalness of the dialogue.

  The dialogue apparatus according to the embodiment includes a dialogue sentence storage unit, a dialogue order storage unit, an output unit, a first reception unit, a second reception unit, a first probability calculation unit, and a second probability calculation unit. And a section. The dialog sentence storage unit stores a dialog sentence. The dialogue order storage unit stores a plurality of dialogue sentence patterns. The output unit outputs the first output sentence. The first receiving unit receives an input sentence. The second reception unit receives an input of a question sentence. The first probability calculation unit outputs a dialogue sentence that is the same as or similar to the first output sentence, and then outputs each of the first dialogue sentences that are one of the dialogue sentences stored in the dialogue sentence storage unit. The first probability representing the probability of being calculated is calculated based on the pattern. The second probability calculation unit calculates, based on the pattern, a second probability that represents a probability that a dialogue sentence that is the same as or similar to the question sentence is output next to the first conversation sentence for each first dialogue sentence. . The selection unit selects, as the second output sentence, the first dialog sentence having the largest value obtained by weighted addition of the first probability and the second probability.

The block diagram which shows the structure of the dialogue apparatus concerning 1st Embodiment. The figure which shows an example of the pattern of the dialog order memorize | stored in the dialog order memory | storage part. The figure which shows an example of the pattern of the dialog order by a user and an output part. The flowchart which shows the procedure of the output sentence output process by a dialogue apparatus. The block diagram which shows the structure of the dialogue apparatus concerning 2nd Embodiment. The flowchart which shows the procedure of the output sentence output process by a dialogue apparatus. Explanatory drawing which shows the hardware constitutions of the dialogue apparatus concerning 1st or 2nd Embodiment.

(First embodiment)
FIG. 1 is a block diagram illustrating the configuration of the interactive apparatus according to the first embodiment. As shown in FIG. 1, the dialogue apparatus 100 includes a first reception unit 101, a second reception unit 102, a next utterance transition probability calculation unit 103, a question guidance transition probability calculation unit 110, a selection unit 115, and a recognition unit. Unit 116, detection unit 117, output unit 118, similarity calculation unit 119, dialogue sentence storage unit 150, dialogue order storage unit 160, and operation display unit 120.

  The dialogue storage unit 150 stores past dialogues between the user and the dialogue device 100 or dialogues acquired from the outside in units of sentences. Specifically, the dialogue sentence includes an input sentence inputted by the user and an output sentence outputted by the dialogue apparatus 100. The sentence unit may be text information input at once from a keyboard (not shown) or the like, or voice information input at once from a speech recognition device (not shown). It is good also considering the attached part as one sentence.

  The dialogue order storage unit 160 stores a plurality of dialogue order patterns of a series of input sentences by the user and output sentences by the dialogue apparatus 100. Here, a series of dialogue orders refers to a sequence of dialogues composed of a plurality of dialogue sentences, and the series of dialogues is such that the user or the dialogue device 100 has predetermined input or output of the next dialogue sentence. It may end when the time is interrupted, or may end when the topic changes. When the dialogue sentence is a grammatical sentence and a point is added as one sentence, the dialogue order in the dialogue order storage unit 160 is not replaced with the user and the output unit 118, but the dialogue sentence order. It becomes.

  FIG. 2 is a diagram illustrating an example of a conversation order pattern stored in the conversation order storage unit 160. As shown in FIG. The dialogues 1 to 9 are stored by adding a sign to each dialogue sentence. For example, in FIG. In the dialogue 1, “good morning” is first output as the first output sentence (S <b> 1) by the interactive device 100, and then the input sentence (U <b> 1) “good morning” is inputted from the user.

  Further, the dialogue order storage unit 160 may store a dialogue state with the user for each pattern. The dialogue status is information indicating the dialogue status when the dialogue with the user is being performed. For example, time (morning, noon, evening), weather, temperature, physical condition of the user (a cold is caught, etc.) ), Etc., that affect the content of the dialogue.

  The detection unit 117 detects the current dialog state. For example, the detection unit 117 detects time, temperature, and the like acquired from the interaction device 100 or from the outside as the interaction state. Moreover, the detection part 117 may detect a dialog condition from the said input sentence, when the specific word which shows a dialog condition is contained in the input sentence which the user input by the 1st reception part 101. FIG.

  The 1st reception part 101 receives the input sentence which the user input by text information or audio | voice information. For example, the first receiving unit 101 receives text information from the operation display unit 120 using a keyboard or the like, or receives audio information from a microphone (not shown).

  Moreover, the 1st reception part 101 memorize | stores an input sentence in the dialogue sentence memory | storage part 150, when an input sentence is received by text information. Further, when the first reception unit 101 receives an input sentence as voice information, the first reception unit 101 converts the input sentence into text information through a voice recognition process and stores the text information in the dialogue sentence storage unit 150.

  The input sentence obtained by the first reception unit 101 is recorded together with the output sentence immediately before the corresponding one, and saved as an answer to the output sentence, so that the system side can collect user information. . When the input sentence is received as voice information, it may be converted and recorded as text information by voice recognition processing, or the voice information may be recorded and recorded as it is to collect information.

  The 2nd reception part 102 receives the input of the question sentence (henceforth a request | requirement question sentence) which requests | requires a system to ask a user a question. Here, the request question sentence is a sentence in which the content requested by the user is in a question format, and the question content may be sequentially entered in a list like a questionnaire.

  The similarity calculation unit 119 calculates the similarity between the dialogue sentence stored in the dialogue sentence storage unit 150 or the dialogue order storage unit 160 and the input sentence input by the user or the output sentence output by the output unit 118. . In addition, the similarity calculation unit 119 calculates the similarity between the dialogue sentence stored in the dialogue sentence storage unit 150 or the dialogue order storage unit 160 and the request question sentence received by the second reception unit 102.

  For example, the similarity calculation unit 119 may perform morphological analysis of the dialogue sentence and calculate the similarity using the number or ratio of the same word included, or the edit distance obtained by the number of key inputs when correcting the sentence The similarity may be calculated using, or the similarity may be calculated by weighting each part of speech of the word. The similarity calculation unit 119 performs normalization so that the similarity of identical dialogue sentences is 1, and the similarity of completely different dialogue sentences is 0. The similarity calculation unit 119 stores the calculated similarity in the dialogue sentence storage unit 150 together with the input sentence that has been input.

  The selection unit 115 selects an output sentence to be output next by the output unit 118 from the output sentence output immediately before by the output unit 118 and the input sentence input by the user. For example, the selection unit 115 weights and adds the next utterance transition probability calculated by the first calculation unit 103 and the question guidance transition probability calculated by the second calculation unit 110, and selects the dialogue sentence having the largest added value. Select as second output sentence.

  Here, an example of a specific pattern will be described. FIG. 3 is a diagram illustrating an example of a dialogue order pattern by the user and the output unit 118. In Figure 3, the dialogue A "Hello" is output as S1 (first output sentence) by first output unit 118. Then, interactive statement B, "Hello" is input as U1 (input sentence) by the user. Then, the dialogue sentence C “???” that is output by the output unit 118 as S2 (second output sentence) is next to the dialogue sentence F “I slept well?” Next to the dialogue sentence C “???”. "Is determined to be a spoken dialogue sentence. It should be noted that the dialogue sentence “Did you sleep well?” Is a dialogue sentence similar to the request question sentence received by the second reception unit 102.

  The next utterance transition probability calculation unit 103 calculates the next utterance transition probability from the patterns stored in the dialogue order storage unit 160. Here, the next utterance transition probability is a probability of a dialogue sentence to be uttered next when a similar dialogue has been conducted in the past. Specifically, an interactive sentence that is the same as or similar to the output sentence (first output sentence) (S1) output by the output unit 118 immediately before is output by the system, and then the same or similar to the input sentence (U1). This is a probability that a dialogue sentence to be inputted is input by the user, and then a certain dialogue sentence (first dialogue sentence) stored in the dialogue sentence storage unit 150 is output.

  First, the next utterance transition probability calculation unit 103 obtains the total number of patterns in which a dialogue sentence that is the same as or similar to the input sentence (U1) follows the dialogue sentence that is the same or similar to the first output sentence (S1). For example, in FIG. 6-No. Focusing on the pattern shown by the dotted line 9, the total number of the dialogue A "Hello" → dialogue B "Hello" occurs as shown in FIG. 3 is four. The following speech transition probability calculation unit 103, determine the total number 4 pattern of dialogue sentence A "Hello" → interactive statement B, "Hello" has occurred.

  Then, next utterance transition probability calculation unit 103, for each conversation sentences stored in the conversation sentence storage unit 150, following the conversation sentence pattern of dialogue A "Hello" → dialogue B "Hello" (first (Dialogue sentence) is output.

  For example, when attention is paid to FIG. 2 regarding the pattern in which the dialog sentence C is generated after the first pattern of the dialog sentence A → dialog sentence B shown in FIG. 6 and no. At 8, dialogue sentence A "Hello" → next to the same dialogue sentence C of the dialogue B, "Hello", "It was hot yesterday" has occurred. So, for the dialogue sentence "It was hot yesterday," is, next to output the number of patterns of the first pattern of dialogue sentence A "Hello" → interactive statement B, "Hello" is two. On the other hand, no. 7 and no. At 9, interactive text C is different dialogue sentence "This time I hot," "felt fatigue not?" Has occurred one by one, interactive sentence A "Hello" → of the first pattern of the dialogue B, "Hello" next The number of patterns output to each is 1.

  In this case, the number of first patterns for the dialogue sentence C “It was hot yesterday” is the number of patterns in which the dialogue sentence is output next to the first pattern, and the dialogue sentences “It is hot these days” and “Daraku” For “No?”, The number of patterns is 1.

The next utterance transition probability calculation unit 103 calculates the next utterance transition probability PN _ABC by the following equation (1) using the obtained total number and the number of patterns.
PN _ABC = number of (A → B → C) / total number of (A → B) (1)

However, equation (1) is a probability when the same dialogue sentence exists in the dialogue sentence A and the dialogue sentence B in the dialogue order storage unit 160. However, in reality, there is a high possibility that the same dialogue sentence does not exist in the dialogue order storage unit 160. Therefore, the next utterance transition probability calculation unit 103 calculates the next utterance transition probability PNS _ABC when considering the similarity of each dialogue sentence by the following equation (2).
PNS _ABC = Σn {number of occurrences of (An → Bn → C) × (similarity of A and An) × (similarity of B and Bn) / total number of occurrences of (An → Bn)}} ( 2)
Here, An indicates a dialog sentence similar to the dialog sentence A, and Bn indicates a dialog sentence similar to the dialog sentence B. Further, Σn means that the sum is calculated for each similar dialogue sentence.

Further, the next utterance transition probability calculation unit 103, when the conversation state is stored in the conversation order storage unit 160, the conversation state stored in the conversation order storage unit 160 and the current conversation detected by the detection unit 117. A higher probability is calculated for a dialogue sentence having a higher degree of similarity to the situation. For example, the next utterance transition probability PNST _ABC considering the conversation state is calculated by the following equation (2.2).
PNST _ABC = Σn {number of occurrences of (An → Bn → C) × (similarity of A and An) × (similarity of B and Bn) × T / (total number of occurrences of An → Bn)}・ (2.2)
Here, T is the similarity between the current dialog status and the dialog status when An → Bn → C occurs, and is a coefficient normalized to be 1 when matching and 0 when completely different. .

  The question guidance transition probability calculation unit 110 calculates a question guidance transition probability for each dialogue sentence (first dialogue sentence) stored in the dialogue sentence storage unit 150. Here, the question-induced transition probability is a probability that a dialog sentence that is the same as or similar to the question sentence is output next to the first dialog sentence.

  The question guidance transition probability calculation unit 110 obtains the total number of patterns including dialogue sentences that are the same as or similar to the requested question sentence among the patterns stored in the dialogue order storage unit 160.

  For example, the question induction transition probability calculation unit 110 obtains the total number of patterns including the dialogue sentence F that is the same as or similar to the request question sentence. In this case, in FIG. 3 and no. 6-No. In FIG. 8, the dialogue sentence F “I slept well?” That is the same as or similar to the request question sentence has occurred. Therefore, the second total number calculation unit 111 calculates the total number 4 of the second patterns including the dialogue sentence F “I slept well?” That is the same as or similar to the request question sentence.

  The question guidance transition probability calculation unit 110 obtains the number of patterns of a pattern in which a dialogue sentence that is the same as or similar to the request question sentence included in the second pattern is output next to the dialogue sentence for each dialogue sentence. For example, the question guidance transition probability calculation unit 110 is the same as or similar to the request question sentence after the dialogue sentence D for each dialogue sentence stored in the dialogue sentence storage unit 150 (hereinafter referred to as the dialogue sentence D). The number of patterns in which the dialogue sentence F occurs is obtained.

  For example, in FIG. 3 and no. In FIG. 8, the dialogue sentence F “I slept well?” Occurs two times after the same dialogue sentence D “It was hot yesterday”. Therefore, for the dialogue sentence D “Yesterday was hot”, the number of patterns in which F “I slept well?” That is the same or similar to the request question sentence after the dialogue sentence D “Yesterday was hot” occurred. 2. On the other hand, no. In FIG. 7, the number of patterns in which the dialogue sentence F “I slept well?” Similar to the request question sentence is generated after the dialogue sentence D “You feel like summer?

  In this case, the question induced transition probability calculation unit 110, for the dialogue sentence D “Yesterday was hot”, the number of patterns in which the dialogue sentence F “I slept well?” Similar to the request question sentence was generated 2 later. For the dialogue sentence “It's hot these days”, ask for 1 pattern.

The question guidance transition probability calculation unit 110 calculates the question guidance transition probability PA _DF by the following equation (3) using the obtained number of patterns and the total number.
PA _DF = number of (D → * → F) / total number of F (3)

Further, the question induced transition probability calculation unit 110 calculates the question induced transition probability PAS _DF considering the similarity according to the following equation (4).
PAS _DF = Σn {number of (Dn → * → Fn) × (similarity between D and Dn) × (similarity between F and Fn) / total number of F} (4)
Here, Dn indicates a dialog sentence similar to the dialog sentence D, and Fn indicates a dialog sentence similar to the dialog sentence F. Further, Σn means that the sum is calculated for each similar dialogue sentence.

Further, the next utterance transition probability calculation unit 103, when the conversation state is stored in the conversation order storage unit 160, the conversation state stored in the conversation order storage unit 160 and the current conversation detected by the detection unit 117. A higher probability is calculated for a dialogue sentence having a higher degree of similarity to the situation.
For example, the question-induced transition probability PAST _DF considering the dialogue situation is calculated by the following equation (4.2): PAS _DF = Σn {number of occurrences of (Dn → * → Fn) × (similarity between D and Dn) X (similarity between F and Fn) x total number of T / F} (4.2)

Here, T is the similarity between the current dialog status and the dialog status when Dn → * → Fn occurs, and is a coefficient normalized to be 1 when matching and 0 when completely different. .

  The selection unit 115 sets the dialogue sentence having the largest value obtained by weighted addition of the next utterance transition probability calculated by the next utterance transition probability calculation unit 103 and the question guidance transition probability calculated by the question guidance transition probability calculation unit 110. 2 Select as output sentence. For example, in FIG. 2, the dialogue sentence with the highest next utterance transition probability calculated by the next utterance transition probability calculating unit 103 is “It was hot yesterday”, and the question guidance calculated by the question guidance transition probability calculating unit 110 The dialogue sentence with the highest transition probability is “It was hot yesterday”. Therefore, in the pattern shown in FIG. 2, the dialogue sentence with the highest next utterance transition probability matches the dialogue sentence with the highest question-induced transition probability. However, the selection unit 115 is particularly effective when the dialogue sentence having the highest next utterance transition probability is different from the dialogue sentence having the highest question guidance transition probability by performing weighted addition.

  The output unit 118 outputs the dialogue sentence selected by the selection unit 115 as text information to a display screen (not shown) of the operation display unit 120, or outputs a voice from a speaker (not shown) by a voice synthesizer (not shown). To do. Hereinafter, the dialogue sentence output by the output unit 118 is referred to as an output sentence.

  The operation display unit 120 displays an operation screen (not shown) such as a touch panel for receiving an input sentence from the user and outputting an output sentence from the output unit 118. In addition, this may be a microphone or a speaker when operating with audio information.

  Next, a procedure of output sentence output processing by the interactive apparatus 100 configured as described above will be described. FIG. 4 is a flowchart showing a procedure of output sentence output processing by the dialogue apparatus 100.

  The 2nd reception part 102 receives the input of a request question sentence (step S1). For example, the second reception unit 102 receives an input of a question content to the user as a request question sentence by text information or voice information.

  The similarity calculation unit 119 calculates the similarity between the request question sentence received by the second reception unit 102 and the dialogue sentence stored in the dialogue sentence storage unit 150 (step S2).

  The output unit 118 outputs the system utterance (S1) selected by the selection unit 115 (step S3). Here, since the system utterance (S1) is the utterance of the first turn, the selection unit 115 selects a greeting as a standard sentence without considering the sequence of conversations so far.

  The similarity calculation unit 119 calculates the similarity between the output sentence output by the output unit 118 and the dialogue sentence stored in the dialogue sentence storage unit 150 (step S4). The similarity calculation unit 119 stores the calculated similarity in the dialogue order storage unit 160.

  The 1st reception part 101 receives the input of the speech (U1) by a user as an input sentence (step S5). Here, the 1st reception part 101 will receive the speech (U1) by a user as a response to the system speech (S1) output in step S3.

  The similarity calculation unit 119 calculates the similarity between the input sentence received by the first reception unit 101 and the dialogue sentence stored in the dialogue sentence storage unit 150 (step S6). The similarity calculation unit 119 stores the calculated similarity in the dialogue order storage unit 160.

  The next utterance transition probability calculation unit 103 calculates the next utterance transition probability (step S8). Here, first, the next utterance transition probability calculation unit 103 obtains the total number of occurrences of the dialogue sentence B as the user utterance (U1) after the dialogue sentence A as the system utterance (S1) from the dialogue order storage unit 160. . Next, the next utterance transition probability calculation unit 103 stores in the dialogue order storage unit 160 the number of patterns of the dialogue sentence that occurs next to the dialogue sentence A as the system utterance (S1) and the dialogue sentence B as the user utterance (U1). Ask for each dialogue sentence. Then, the next utterance transition probability calculation unit 103 calculates the next utterance transition probability by dividing the obtained number of patterns by the total number.

  The question guidance transition probability calculation unit 110 calculates a question guidance transition probability (step S8). First, the question induction transition probability calculation unit 110 obtains the total number of dialogue sentences F that are the same as or similar to the requested question sentence from the dialogue order storage unit 160. Next, the question induction transition probability calculation unit 110 obtains the number of patterns in which a dialogue sentence F that is the same as or similar to the requested question sentence after two dialogue sentences D is generated for each dialogue sentence. Then, the question induction transition probability calculation unit 110 calculates the question induction transition probability by dividing the obtained number of patterns by the total number.

  The selection unit 115 selects the system utterance (S2) as an output sentence (second output sentence) to be output with respect to the input sentence (step S9). Here, the selection unit 115 has the largest value obtained by weighted addition of the next utterance transition probability calculated by the next utterance transition probability calculation unit 103 and the question guidance transition probability calculation unit 110. Select the sentence as the second output sentence. The output unit 118 outputs the system utterance (S2) selected by the selection unit 115 as a second output sentence (step S10).

  Thus, according to the present embodiment, since the dialogue sentence having the largest value obtained by weighted addition of the next utterance transition probability and the question induction transition probability is output as the second output sentence, the natural flow of the dialogue is maintained. The information desired to be obtained from the user can be acquired.

(Second Embodiment)
In the first embodiment, the selection unit selects the second output sentence using the question guidance transition probability calculated from the request question sentence. On the other hand, in the second embodiment, the second output sentence is selected using the probability calculated from the dialogue sentence output next to the first output sentence output by the output unit.

  FIG. 5 is a block diagram illustrating a configuration of the interactive apparatus according to the second embodiment. As illustrated in FIG. 5, the dialogue apparatus 200 according to the second embodiment includes a first reception unit 101, a second reception unit 102, a first calculation unit 103, a similar utterance calculation unit 210, and a selection unit. 215, a recognition unit 116, a detection unit 117, an output unit 118, a similarity calculation unit 119, a dialogue sentence storage unit 150, a dialogue order storage unit 160, and an operation display unit 120. Note that the functions and configurations of the units other than the similar utterance calculation unit 210 and the selection unit 215 are the same as those in the first embodiment.

  The similar utterance calculation unit 210 calculates a similar utterance transition probability. Here, the similar utterance transition probability means that for each dialogue sentence (first dialogue sentence) stored in the dialogue sentence storage unit 150, the dialogue sentence output next to the first dialogue sentence is the first dialogue sentence. Is the probability of a similar dialogue sentence that is the same or similar to the next dialogue sentence following the same or similar dialogue sentence.

  The similar utterance calculation unit 210 obtains the total number of the third patterns including the similar dialogue sentence among the patterns stored in the dialogue order storage unit 160. For example, the similar utterance calculation unit 210 obtains the total number of third patterns including the similar dialogue sentence G.

  For example, in FIG. 1 and No. 2, a similar dialogue sentence G “Hid you sleep well yesterday?” Has occurred. Therefore, the third total number calculation unit 211 obtains the total number 2 of the third patterns including the similar dialogue sentence G “I was able to sleep well yesterday?”.

  The similar utterance calculation unit 210 obtains, for each dialogue sentence, the number of patterns in which the similar dialogue sentence included in the third pattern is output next to the dialogue sentence. For example, for each dialogue sentence, the similar utterance calculation unit 210 sets the dialogue sentence as the dialogue sentence E, and obtains the number of patterns in which the similar dialogue sentence G occurs after the dialogue sentence E.

  For example, in FIG. 1 and No. 2, the dialogue sentence G “Did you sleep well yesterday?” Occurs after the same dialogue sentence E “Good morning”. Therefore, for the dialogue sentence E “Good morning”, the number of patterns in which the similar dialogue sentence G “I slept well yesterday?” Occurs after the dialogue sentence E “Good morning” is two. .

  In this case, the similar utterance calculation unit 210 obtains the pattern number 2 of the patterns in which the similar dialogue sentence G “I was able to sleep well yesterday” occurs two times later for the dialogue sentence E “Good morning”.

The similar utterance transition probability calculation unit 210 calculates the similar utterance transition probability by dividing the number of patterns obtained by the third pattern number calculation unit 212 by the total number of third patterns calculated by the third total number calculation unit 211. Specifically, the similar utterance transition probability calculation unit 210 calculates the similar utterance transition probability PNA _GI using the following equation (5).
PNA _GI = number of (G → * → I) / total number of I (5)

In addition, the similar utterance transition probability calculation unit 210 calculates the similar utterance transition probability PNAS _GI considering the similarity according to the following equation (6).
PNAS _GI = Σn {number of occurrences of (Gn → * → In) × (similarity between G and Gn) × (similarity between I and In)) (6)
Here, Gn indicates a dialog sentence similar to the dialog sentence G, and In indicates a dialog sentence similar to the dialog sentence I. Further, Σn means that the sum is calculated for each similar dialogue sentence.

  Further, the next utterance transition probability calculation unit 103, when the conversation state is stored in the conversation order storage unit 160, the conversation state stored in the conversation order storage unit 160 and the current conversation detected by the detection unit 117. A higher probability is calculated for a dialogue sentence having a higher degree of similarity to the situation.

  The selection unit 215 selects the dialogue sentence having the largest value obtained by weighted addition of the next utterance transition probability calculated by the next utterance transition probability calculation unit 103 and the similar utterance transition probability calculation unit 210. 2 Select as output sentence. For example, in FIG. 2, the dialogue sentence with the highest next utterance transition probability calculated by the next utterance transition probability calculation unit 103 is “Yesterday was hot”, and the similar utterance calculated by the similar utterance transition probability calculation unit 210 The dialogue sentence with the highest transition probability is “Good morning”. Therefore, in the pattern shown in FIG. 2, the dialogue sentence having the highest next utterance transition probability is different from the dialogue sentence having the highest question-induced transition probability. Here, the selection unit 215 selects a dialog sentence having a higher weighted value as a second output sentence by performing weighted addition.

  In addition, when there is a dialogue sentence including dialogue information in the next utterance transition probability or the similar utterance transition probability, the selection unit 215 may increase the weight of the dialogue sentence including the dialogue information and perform weighted addition.

  Next, a procedure of output sentence output processing by the interactive apparatus 200 configured as described above will be described. FIG. 6 is a flowchart showing a procedure of output sentence output processing by the dialogue apparatus 200.

  The processing from step S21 to step S25 is the same as the processing from step S3 to step S7 described in FIG. 4 in the first embodiment.

  In step S26, the similar utterance transition probability calculation unit 210 calculates a similar utterance transition probability (step S26). Here, first, the similar utterance transition probability calculation unit 210 obtains the total number of similar dialogue sentences G from the dialogue order storage unit 160. Next, the similar utterance transition probability calculation unit 210 obtains, for each dialog sentence, the number of patterns in which the similar dialog sentence G occurs after the dialog sentence E. Then, the similar utterance transition probability calculation unit 210 calculates the similar utterance transition probability by dividing the obtained number of patterns by the total number.

  The selection unit 215 selects the system utterance (S2) as the second output sentence (step S27). Here, the selection unit 215 has the largest value obtained by weighting and adding the next utterance transition probability calculated by the next utterance transition probability calculation unit 103 and the similar utterance transition probability calculation unit 210. Select the sentence as the second output sentence. The output unit 118 outputs the system utterance (S2) selected by the selection unit 215 as a second output sentence (step S28).

  As described above, according to the present embodiment, when the first receiving unit receives voice information, even when the voice recognition rate by the recognition unit 116 is low and the voice cannot be recognized, the dialogue is natural. Can continue.

(Modification)
As a modification, the configuration of the interactive device 100 according to the first embodiment and the configuration of the interactive device 200 according to the second embodiment may be combined. For example, it is good also as a structure which replaces the question guidance transition probability calculation part 110 with which the dialogue apparatus 100 concerning 1st Embodiment is equipped with the similar utterance transition probability calculation part 210 with which the dialogue apparatus 200 concerning 2nd Embodiment is equipped.

  Thus, according to the present embodiment, since the dialogue sentence having the largest value obtained by weighted addition of the next utterance transition probability and the similar utterance transition probability is output as the second output sentence, the natural flow of the dialogue is maintained. The information desired to be obtained from the user can be acquired.

  As described above, according to the present embodiment, when the first reception unit receives the voice information, the flow is natural even when the voice recognition rate by the recognition unit 116 is low and the voice cannot be recognized. You can continue the conversation.

  As described above, according to the first and second embodiments, a natural flow of dialogue can be maintained.

  Next, the hardware configuration of the interactive apparatus according to the first or second embodiment will be described with reference to FIG. FIG. 7 is an explanatory diagram illustrating a hardware configuration of the interactive apparatus according to the first or second embodiment.

  The interactive device according to the first or second embodiment includes a control device such as a CPU (Central Processing Unit) 51, a storage device such as a ROM (Read Only Memory) 52, a RAM (RANDOM Access Memory) 53, and a network. A communication I / F 54 that communicates by connecting to each other and a bus 61 that connects each unit are provided.

  The output sentence output program executed by the interactive apparatus according to the first or second embodiment is provided by being incorporated in advance in the ROM 52 or the like.

  The output sentence output program executed by the interactive apparatus according to the first or second embodiment is a file in an installable format or an executable format, and is a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD). ), A CD-R (Compact Disk Recordable), a DVD (Digital Versatile Disk), or the like.

  Further, the output sentence output program executed by the interactive apparatus according to the first or second embodiment is stored on a computer connected to a network such as the Internet and is provided by being downloaded via the network. It may be configured. Further, the output sentence output program executed by the interactive device according to the first or second embodiment may be provided or distributed via a network such as the Internet.

  The output sentence output program executed by the interactive device according to the first or second embodiment is a computer that includes each part of the above-described interactive device (first reception unit, second reception unit, next utterance transition probability calculation unit, question (Guidance transition probability calculation unit, selection unit, recognition unit, detection unit, output unit, similarity calculation unit). In this computer, the CPU 51 can read and execute an output sentence output program from a computer-readable storage medium on a main storage device.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Moreover, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, the constituent elements over different embodiments may be appropriately combined.

  In the present invention, three chain structures of output → input → output or input → output → input are described, but the present invention is not limited to three. It can respond similarly.

DESCRIPTION OF SYMBOLS 100,200 Dialogue device 101 1st reception part 102 2nd reception part 103 Next utterance transition probability calculation part 110 Question guidance transition probability calculation part 115 Selection part 117 Detection part 118 Output part 119 Similarity degree calculation part 120 Operation display part 150 Dialogue sentence Storage unit 160 Dialog sequence storage unit 210 Similar utterance transition probability calculation unit

Claims (5)

A dialogue storage unit for storing dialogue statements;
A dialogue order storage unit for storing a plurality of dialogue order patterns of the dialogue sentence;
An output unit for outputting a first output sentence;
A first accepting unit for accepting an input sentence input by a user with respect to the first output sentence;
A second reception unit for receiving input of a question sentence;
The dialog sentence that is the same as or similar to the first output sentence is output, the dialog sentence that is the same or similar to the input sentence is input, and then the dialog stored in the dialog sentence storage unit A first probability calculating unit that calculates a first probability representing a probability that each of the first dialogue sentences that are any of the sentences is output based on the pattern;
Based on the pattern, for each of the first dialogue sentences, a second probability is calculated that represents a probability that the dialogue sentence that is the same as or similar to the question sentence is output next to the first dialogue sentence. A probability calculator;
A selection unit that selects, as the second output sentence, the first dialogue sentence having the largest value obtained by weighted addition of the calculated first probability and the second probability;
An interactive apparatus characterized by comprising: The dialog order storage unit further stores a dialog status with the user for each pattern,
The first probability calculation unit calculates the first probability that is larger for the dialogue sentence having a higher similarity between the dialogue situation stored for the pattern including the dialogue sentence and the current dialogue situation,
The second probability calculation unit calculates the second probability that is larger for the dialogue sentence having a higher similarity between the dialogue situation stored for the pattern including the dialogue sentence and the current dialogue situation;
The interactive apparatus according to claim 1. Based on the pattern, for each of the first dialogue sentences, the dialogue sentence output next to the first dialogue sentence is next to the dialogue sentence that is the same as or similar to the first dialogue sentence. A third probability calculator for calculating a third probability representing a probability that is the same as or similar to the output dialogue sentence;
Further comprising
The selection unit selects, as the second output sentence, the first dialogue sentence having the largest value obtained by weighted addition of the calculated first probability and the third probability;
The interactive apparatus according to claim 1. A dialogue storage unit for storing dialogue statements;
A dialogue order storage unit for storing a plurality of dialogue order patterns of the dialogue sentence;
An output unit for outputting a first output sentence;
A first accepting unit for accepting an input sentence input by a user with respect to the first output sentence;
The dialog sentence that is the same as or similar to the first output sentence is output, the dialog sentence that is the same or similar to the input sentence is input, and then the dialog stored in the dialog sentence storage unit A first probability calculating unit that calculates a first probability representing a probability that each of the first dialogue sentences that are any of the sentences is output based on the pattern;
Based on the pattern, for each of the first dialogue sentences, the dialogue sentence output next to the first dialogue sentence is next to the dialogue sentence that is the same as or similar to the first dialogue sentence. A third probability calculation unit for calculating a third probability representing a probability that is the same as or similar to the output dialogue sentence;
A selection unit that selects, as the second output sentence, the first dialogue sentence having the largest value obtained by weighted addition of the calculated first probability and the third probability;
An interactive apparatus characterized by comprising: The dialog order storage unit further stores a dialog status with the user for each pattern,
The first probability calculation unit calculates the first probability that is larger for the dialogue sentence having a higher similarity between the dialogue situation stored for the pattern including the dialogue sentence and the current dialogue situation,
The second probability calculation unit calculates the second probability that is larger for the dialogue sentence having a higher similarity between the dialogue situation stored for the pattern including the dialogue sentence and the current dialogue situation;
The interactive apparatus according to claim 4.

Images