WO2016120904A1 - 意図推定装置及び意図推定方法 - Google Patents
意図推定装置及び意図推定方法 Download PDFInfo
- Publication number
- WO2016120904A1 WO2016120904A1 PCT/JP2015/000367 JP2015000367W WO2016120904A1 WO 2016120904 A1 WO2016120904 A1 WO 2016120904A1 JP 2015000367 W JP2015000367 W JP 2015000367W WO 2016120904 A1 WO2016120904 A1 WO 2016120904A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- intention
- execution order
- unit
- sentence
- feature quantity
- Prior art date
Links
- 238000000034 method Methods 0.000 title description 32
- 238000004458 analytical method Methods 0.000 claims abstract description 73
- 238000000605 extraction Methods 0.000 claims abstract description 70
- 230000000877 morphologic effect Effects 0.000 claims abstract description 24
- 239000000284 extract Substances 0.000 claims abstract description 18
- 150000001875 compounds Chemical class 0.000 claims description 20
- 238000010586 diagram Methods 0.000 description 24
- 230000004044 response Effects 0.000 description 12
- 230000006870 function Effects 0.000 description 9
- 238000013500 data storage Methods 0.000 description 6
- 239000002245 particle Substances 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 4
- 238000007619 statistical method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/268—Morphological analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
- G10L2015/025—Phonemes, fenemes or fenones being the recognition units
Definitions
- the present invention relates to an intention estimation apparatus and an intention estimation method for recognizing text input using a voice or a keyboard to estimate a user's intention and executing an operation intended by the user.
- Such technology is effective when the intention included in the utterance content is one.
- an utterance including a plurality of intentions such as compound sentences is input by a speaker, it is difficult to correctly estimate the plurality of intentions.
- the utterance “I want to stop by Tokyo Tower, but stop by Sky Tree first.” There are two intentions: setting the transit point for Tokyo Tower, which is a facility, and setting the transit point for Sky Tree, which is a facility.
- There is an intention and it is difficult to estimate these two intentions by using the above intention estimation model.
- Patent Document 1 proposes a method for estimating the position of an appropriate division point of an input text based on intention estimation and the probability of compound sentence division for an utterance including a plurality of intentions.
- Patent Document 1 simply outputs the result of estimating a plurality of intentions based on the division points, and cannot estimate the execution order of machine commands corresponding to the estimated plurality of intentions. There was a problem that the user's intention could not be accurately estimated.
- the present invention has been made to solve the above-described problems, and an object thereof is to provide an intention estimation apparatus and an intention estimation method that can accurately estimate a user's intention.
- the intention estimation apparatus includes a morpheme analyzer that performs morpheme analysis on a compound sentence including a plurality of intentions, a syntax analysis of the compound sentence that has been analyzed by the morpheme analyzer, and a first simple sentence and a second sentence.
- a syntax analysis part that divides the sentence into simple sentences, a first intention included in the first simple sentence, and an intention estimator that estimates the second intention included in the second simple sentence
- a feature quantity extraction unit that extracts a morpheme indicating the execution order of operations as a first feature quantity, and extracts a morpheme indicating an execution order of operations included in the second simple sentence as a second feature quantity; and feature quantity extraction Execution for estimating the execution order of the first operation corresponding to the first intention and the second operation corresponding to the second intention based on the first feature amount and the second feature amount extracted by the unit And an order estimation unit.
- the intention estimation method includes a step of morphological analysis of a compound sentence including a plurality of intentions, a step of parsing the compound sentence subjected to morphological analysis and dividing it into a plurality of single sentences, and a plurality of single sentences respectively. Included in each of a plurality of simple sentences, a step of extracting a morpheme indicating an operation execution order included in each of a plurality of single sentences as a feature quantity, and included in each of a plurality of single sentences based on a feature quantity included in each of the plurality of single sentences. And estimating the execution order of each operation corresponding to the intended intention.
- the intention estimation apparatus and the intention estimation method of the present invention since the execution order of operations corresponding to the intention is estimated based on the feature amount extracted from the single sentence, it is possible to accurately estimate the user's intention.
- FIG. 5 is a diagram illustrating an example of an intention estimation model according to Embodiment 1.
- FIG. 6 is a diagram illustrating an example of a feature amount extraction rule according to Embodiment 1.
- FIG. An example of the kind of execution order which concerns on Embodiment 1 is shown.
- 6 is a diagram illustrating an example of an execution order estimation model according to Embodiment 1.
- FIG. 2 is a diagram illustrating a hardware configuration example of an intention estimation apparatus 1 according to Embodiment 1.
- FIG. 6 is a diagram illustrating an example of learning data according to Embodiment 1.
- FIG. 6 is a flowchart for explaining execution order estimation model generation processing according to the first embodiment; 6 is a diagram showing an example of interaction according to Embodiment 1.
- FIG. 5 is a flowchart for explaining intention estimation processing according to Embodiment 1; It is a figure which shows the score of each feature-value with respect to each execution order which concerns on Embodiment 1.
- FIG. 6 is a diagram illustrating a calculation formula for obtaining a product of scores according to Embodiment 1.
- FIG. It is a figure which shows the final score with respect to each execution order which concerns on Embodiment 1.
- FIG. It is a figure which shows the structural example of the intention estimation apparatus 1B which concerns on Embodiment 2.
- FIG. 10 is a flowchart for explaining an execution order estimation model generation process according to the second embodiment.
- FIG. 10 is a diagram showing an example of interaction according to the second embodiment.
- 10 is a flowchart for explaining intention estimation processing according to Embodiment 2; It is a figure which shows the final score with respect to each execution order which concerns on Embodiment 2.
- Embodiment 1 FIG. Embodiment 1 of the present invention will be described below with reference to the drawings.
- FIG. 1 is a diagram illustrating a configuration example of an intention estimation apparatus 1 according to the first embodiment.
- the intention estimation apparatus 1 includes a voice input unit 101, a voice recognition unit 102, a morpheme analysis unit 103, a syntax analysis unit 104, an intention estimation model storage unit 105, an intention estimation unit 106, and a feature amount extraction rule storage unit.
- the voice input unit 101 accepts voice input.
- the speech recognition unit 102 recognizes speech data corresponding to the speech input input to the speech input unit 101, converts the speech data into text data, and outputs the text data to the morpheme analysis unit 103.
- the text data is a compound sentence including a plurality of intentions.
- a compound sentence is composed of a plurality of simple sentences, and one simple sentence includes one intention.
- the morpheme analysis unit 103 performs morpheme analysis on the text data converted by the speech recognition unit 102 and outputs the result to the syntax analysis unit 104.
- the morpheme analysis is a natural language processing technique in which a text is divided into morphemes (the smallest unit having meaning in a language) and parts of speech are identified using a dictionary. For example, a simple sentence “go to Tokyo Tower” is divided into morphemes such as “Tokyo Tower / proprietary nouns, go / case particles, go / verbs”. A morpheme may be expressed as an independent word or a feature.
- the syntax analysis unit 104 analyzes the structure of a sentence in units of phrases or clauses (syntactic analysis) based on the grammatical rules for the text data analyzed by the morpheme analysis unit 103.
- the syntax analysis unit 103 divides the text into a plurality of single sentences and outputs the morphological analysis results of each single sentence to the intention estimation unit 106 and the feature amount extraction unit 108.
- a syntax analysis method for example, a CYK (Cocke-Younger-Kasami) method or the like can be used.
- the text (compound sentence) is described as including two simple sentences 1 and 2; however, the text is not limited to this, and may be composed of three or more simple sentences.
- the parsing unit 103 does not need to output data corresponding to all the divided single sentences to the intention estimation unit 106 and the feature amount extraction unit 108.
- the single sentence 1 and the single sentence 2 are input to the input text (compound sentence). Even if single sentence 3 is included, only single sentence 1 and single sentence 2 may be output.
- the intention estimation model storage unit 105 stores an intention estimation model for performing intention estimation using morphemes as features.
- the main intention indicates the classification or function of the intention.
- the main intention corresponds to an upper layer machine command (destination setting, listening to music, etc.) that the user first operates.
- the slot name and slot value indicate information necessary for executing the main intention.
- FIG. 2 is a diagram illustrating an example of the intention estimation model according to the first embodiment.
- the intention estimation unit 106 estimates the intentions included in the plurality of single sentences using the intention estimation model based on the morphological analysis results of the plurality of single sentences input from the syntax analysis unit 104, and sends the results to the command execution unit 111. Output.
- a maximum entropy method can be used as the intention estimation method. That is, the intention estimation unit 106 uses a statistical method to estimate how likely the intention corresponding to the input morpheme is from a large amount of collected morpheme and intention sets.
- the feature quantity extraction rule storage unit 107 is used to estimate the execution order, and stores feature quantity extraction rules (feature quantity extraction conditions) that are rules for extracting morphemes in simple sentences as feature quantities.
- FIG. 3 is a diagram showing an example of the feature amount extraction rule according to the first embodiment.
- the appearance position condition means the appearance position (start of sentence, sentence end, etc.) of a morpheme in a single sentence
- the part of speech condition means the part of speech of the morpheme existing at the appearance position.
- the part of speech of a morpheme that appears at the beginning of a sentence in a simple sentence is a time noun and a case particle
- the morpheme is extracted as a feature quantity.
- the part of speech of a morpheme that appears at the end of a sentence in a simple sentence is a connected particle, the morpheme is extracted as a feature quantity.
- the feature quantity extraction unit 108 uses a morpheme indicating the execution order of operations included in each single sentence as a feature quantity using a feature quantity extraction rule based on the morpheme analysis results of a plurality of single sentences input from the syntax analysis unit 104. Extract. That is, the feature quantity extraction unit 108 extracts each feature quantity from the position of the morpheme and the part of speech of the morpheme in a plurality of simple sentences. For example, for a simple sentence “go to Tokyo Tower first”, the morpheme analysis unit 103 reads “first / hour noun + case particle, Tokyo Tower / proprietary noun, to / case particle, go / verb” and so on. Morphological analysis is performed.
- the feature quantity extraction unit 108 extracts the “first” morpheme as a feature quantity according to the feature quantity extraction rule shown in FIG. To do. Then, the feature quantity extraction unit 108 outputs the extracted feature quantities to the execution order estimation unit 110.
- the execution order estimation model storage unit 109 stores an execution order estimation model (execution order estimation information) used to estimate the execution order of a plurality of single sentences included in the text.
- FIG. 4 shows an example of the types of execution order according to the first embodiment.
- the execution order of a plurality of simple sentences means in which order the operations corresponding to the intention of each simple sentence are executed.
- simple sentence 1 is executed first (single sentence 1 priority: No. 1) or simple sentence 2 is executed first ( Single sentence 2 priority: No. 2), whether single sentence 1 and single sentence 2 are executed simultaneously (execution simultaneously: No. 3), whether only single sentence 1 is executed (only single sentence 1 is executed: No. 4), or only single sentence 2 is executed (Simple sentence 2 only: No. 5).
- FIG. 5 is a diagram illustrating an example of the execution order estimation model according to the first embodiment.
- the execution order estimation model includes information (execution information) in which an execution order type of an operation corresponding to each intention of a plurality of single sentences is associated with a score of a feature amount included in each of the plurality of single sentences.
- Executid information For example, if there is a feature amount “so” in the single sentence 1, the score of “single sentence 1 priority” is 0.07, the score of “single sentence 2 priority” is 0.25, and the execution score is 0.03, and the single sentence 1 The score of only execution is 0.05, and the score of execution of only single sentence 2 is 0.6.
- software may calculate according to the content of the feature-value, or a user may set arbitrarily.
- a feature amount “So” in the single sentence 1 it is highly likely that the intention included in the single sentence 2 connected to the single sentence 1 is executed.
- the single sentence 2 priority score is the second highest. If there is a feature amount “te” in the single sentence 2, the score of the “single sentence 1 priority” is 0.2, the score of the single sentence 2 priority is 0.2, the execution score is 0.2, and only the single sentence 1 is executed. Is 0.2, and the execution score for single sentence 2 is 0.2.
- the execution order estimation unit 110 estimates the execution order of operations corresponding to each intention included in a plurality of single sentences based on the feature quantities in each single sentence extracted by the feature quantity extraction unit 108, and the command execution unit 111 Output to. For example, if the text “I want to go to Tokyo Tower, but go to Sky Tree first” is entered, the feature amount of simple sentence 1 “I want to go to Tokyo Tower” and simple sentence 2 “Go to Sky Tree first” The order of execution with the priority of the single sentence 2 is estimated by using the feature amount of “by the side”. Details of the execution order estimation algorithm will be described later.
- the command execution unit 111 includes the intentions included in each of the plurality of single sentences based on the intentions included in the plurality of single sentences estimated by the intention estimation unit 106 and the execution order of the operations estimated by the execution order estimation unit 110.
- the response generation unit 112 generates a response corresponding to the machine command executed by the command execution unit 111.
- the response may be generated in the form of text data, or a synthesized sound may be generated as voice data.
- voice data for example, a synthesized sound such as “Set Skytree as a transit location. Set Tokyo Tower as a transit location” may be used.
- the notification unit 113 notifies a user such as a driver of the response generated by the response generation unit 112. That is, the notification unit 113 notifies the user that the command execution unit 111 has executed a plurality of machine commands.
- the notification mode may be anything as long as the user can recognize the notification, such as notification by display, notification by voice, or notification by vibration.
- FIG. 6 is a diagram illustrating a hardware configuration example of the intention estimation apparatus 1 according to the first embodiment.
- the intention estimation device 1 includes a processing device 150 such as a CPU (Central Processing Unit), a storage device 160 such as a ROM (Read Only Memory) and a hard disk device, and an input device 170 such as a keyboard and a microphone.
- the output device 180 such as a speaker or a display is connected via a bus.
- the CPU may have its own memory.
- the voice input unit 101 shown in FIG. 1 is realized by the input device 170, and the notification unit 113 is realized by the output device 180.
- the data stored in the intention estimation model storage unit 105, the feature quantity extraction rule storage unit 107, the execution order estimation model storage unit 109, and the learning data storage unit 114 described later are stored in the storage device 160, respectively.
- the “ ⁇ unit” such as the generation unit 115 is also stored in the storage device 160 as a program.
- the processing device 150 implements the functions in the “ ⁇ units” by appropriately reading and executing the program stored in the storage device 160. That is, by combining the hardware that is the processing device 150 and the software that is the above-described program, the above-described “ ⁇ unit” functions are realized.
- the function is realized by one processing device 150, but a plurality of processing devices are used by, for example, performing a part of the function by a processing device in an external server. A function may be realized. Therefore, “a processor” as the processing device 150 is a concept including not only one processing device but also a plurality of processing devices.
- the operation of the intention estimation apparatus 1 according to Embodiment 1 will be described. First, the operation related to the generation process of the execution order estimation model will be described.
- FIG. 7 is a diagram showing a configuration example of the intention estimation apparatus 1 for explaining the generation process of the execution order estimation model according to the first embodiment.
- the learning data storage unit 114 stores learning data in which an execution order is assigned to a plurality of sentence examples.
- FIG. 8 is a diagram showing an example of learning data according to the first embodiment.
- the learning data is data in which an execution order is given to a plurality of sentence examples (No. 1, No. 2, No. 3,). For example, sentence No. 1 “Go on the highway because time is strict” is divided into simple sentence 1 “Because time is severe” and simple sentence 2 “Go on the highway”, and the execution order is “Execute only single sentence 2” Is done. This execution order is given in advance by the learning data creator.
- the execution order estimation model generation unit 115 learns the correspondence of the execution order stored in the learning data storage unit 114 using a statistical method.
- the execution order estimation model generation unit 115 generates an execution order estimation model using the feature amount extracted by the feature amount extraction unit 108 and the execution order stored in the learning data storage unit 114.
- FIG. 9 is a flowchart for explaining execution order estimation model generation processing according to the first embodiment.
- the morpheme analysis unit 103 performs morpheme analysis on the simple sentence 1 and the simple sentence 2 in each sentence example of the learning data stored in the learning data storage unit 114 (step ST1). For example, sentence No. 1, the morphological analysis unit 103 performs morphological analysis on the simple sentence 1 “Because time is severe” and simple sentence 2 “Go on the expressway”.
- the morpheme analysis unit 103 outputs the result of the morpheme analysis to the feature amount extraction unit 108 via the syntax analysis unit 104.
- the feature quantity extraction unit 108 performs a feature quantity extraction process on the single sentence 1 and the single sentence 2 subjected to the morphological analysis based on the feature quantity extraction rules stored in the feature quantity extraction rule storage unit 107 (step ST2). For example, sentence No. In other words, the feature amount extraction unit 108 extracts feature amounts “so” and “te”. Further, the feature quantity extraction unit 108 adds single sentence position information to the extracted feature quantities (step ST3). For example, the feature quantity extraction unit 108 adds the single sentence position information “single sentence 1” to the feature quantity “no” extracted from the single sentence 1 so that it becomes “single sentence 1_no”.
- the feature quantity extraction unit 108 adds the single sentence position information “single sentence 2” to the feature quantity “te” extracted from the single sentence 2 so as to be “single sentence 2_te”.
- the feature quantity to which the single sentence position information is added may be expressed as a feature quantity with single sentence position information.
- the feature quantity extraction unit 108 outputs information on the extracted feature quantity to the execution order estimation model generation unit 115.
- the execution order estimation model generation unit 115 generates an execution order estimation model based on the feature amount and the execution order included in the learning data (step ST4). For example, in the case of the feature quantities “single sentence 1_no” and “single sentence 2_te”, the execution order estimation model generation unit 115 executes the execution order included in the learning data “execute only the single sentence 2” as shown in FIG. Therefore, it is determined that the feature quantity “single sentence 2_te” has a higher score than the feature quantity “single sentence 1_de”. The execution order estimation model generation unit 115 performs the same processing as described above on all sentence examples included in the learning data, and finally generates an execution order estimation model as shown in FIG.
- FIG. 10 is a diagram showing an example of dialogue according to the first embodiment.
- FIG. 11 is a flowchart for explaining intention estimation processing according to the first embodiment.
- the speech recognition apparatus 1 utters “Please tell me when it beeps” (S1).
- the user speaks “I want to go to XX” (U1).
- the utterance from the speech recognition apparatus 1 is represented as “S”
- the utterance from the user is represented as “U”. The same applies to the following.
- the speech recognition unit 102 performs speech recognition processing on the user input (step ST01) and converts it into text data.
- the morpheme analysis unit 103 performs a morpheme analysis process on the converted text data (step ST02).
- the syntax analysis unit 104 performs a syntax analysis process on the text data subjected to morphological analysis (step ST03). If the text data is a compound sentence, the compound sentence is divided into a plurality of simple sentences. When the text data is not a compound sentence (step ST04-No), the process proceeds to the process after step ST05, and when the text data is a compound sentence (step ST04-Yes), the process proceeds to a process after step ST07.
- the syntax analysis unit 104 outputs simple text data subjected to morphological analysis to the intention estimation unit 106.
- the intention estimation unit 106 uses the intention estimation model to execute intention estimation processing on the input single sentence (step ST05).
- the command execution unit 111 executes a machine command corresponding to the intention estimation result by the intention estimation unit 106 (step ST06). For example, the command execution unit 111 executes an operation of setting the facility xx as the destination. Then, the response generation unit 112 generates a synthesized sound corresponding to the machine command executed by the command execution unit 111. As the synthesized sound, for example, “XX has been set as the destination”. The notification unit 113 notifies the user of the synthesized sound generated by the response generation unit 112 through a speaker or the like. As a result, as shown in “S2” in FIG. 10, a notification such as “XX is the destination” is given to the user.
- step ST01 when the user speaks as indicated by “U2”, the speech recognition unit 102 performs speech recognition processing on the user input and converts it into text data, and the morpheme analysis unit 103 performs morpheme analysis on the text data. Processing is performed (steps ST01 and ST02).
- the syntax analysis unit 104 performs a syntax analysis process on the text data (step ST03).
- text data related to user input is as follows: simple sentence 1 (first simple sentence) “I want to go to XX first” and simple sentence 2 (second simple sentence) , Divided into multiple single sentences. Therefore, the syntax analysis unit 104 outputs the text data of each single sentence subjected to the morphological analysis to the intention estimation unit 106 and the feature amount extraction unit 108, and the processes after step ST07 are performed.
- the intention estimation unit 106 performs intention estimation processing for each of the single sentence 1 and the single sentence 2 using the intention estimation model (step ST07).
- the feature quantity extraction unit 108 performs feature quantity extraction processing for each of the single sentence 1 and the single sentence 2 using the feature quantity extraction rule (step ST08).
- the feature quantity extraction unit 108 extracts the first feature quantity for the single sentence 1, that is, the feature quantities “single sentence 1_first” and “single sentence 1_but”, and for the single sentence 2.
- the feature quantity extraction unit 108 outputs information on the feature quantity extracted for each simple sentence to the execution order estimation unit 110.
- the execution order estimation unit 110 estimates the execution order of each single sentence based on the information related to the feature amount output from the feature amount extraction unit 108 and the execution order estimation model stored in the execution order estimation model storage unit 109. (Step ST09). Hereinafter, details of the estimation of the execution order will be described.
- the execution order estimation unit 110 obtains a score of each feature quantity for each execution order by comparing the feature quantity output from the feature quantity extraction unit 108 with the execution order estimation model.
- FIG. 12 is a diagram showing the score of each feature quantity with respect to each execution order according to the first embodiment.
- the score of the feature quantity “single sentence 1_first” is 0.45
- the score of the feature quantity “single sentence 1_but” is 0.2
- the score of the feature quantity “single sentence 2 — Yappari” is 0.1
- the score of the feature quantity “single sentence 2 — first” is 0.2.
- the score of each feature amount can be obtained for other execution orders.
- the execution order estimation unit 110 obtains a product of scores of each feature amount in each execution order.
- FIG. 13 is a diagram illustrating a calculation formula for obtaining a product of scores according to the first embodiment.
- Si is a score of the i-th feature amount with respect to the execution order of the estimation target.
- S is a final score representing the product of Si with respect to the execution order of the estimation target.
- FIG. 14 is a diagram showing a final score for each execution order according to the first embodiment.
- the execution order estimation unit 110 calculates the final score shown in FIG. 14 using the calculation formula shown in FIG. In this example, for the execution order “single sentence 1 priority”, the score of the feature quantity “single sentence 1_first” is 0.45, the score of the feature quantity “single sentence 1_but” is 0.2, and the feature quantity “ Since the score of “single sentence 2 — Yappa” is 0.1 and the score of feature quantity “single sentence 2 — first” is 0.2, the final score S, which is the product of these, is calculated as 1.8e-3. Similarly, final scores are also calculated for the other execution orders.
- the execution order estimation unit 110 estimates the execution order “single sentence 2 priority” having the highest score among the calculated final scores of each execution order as an appropriate execution order. In other words, the execution order estimation unit 110 estimates the execution order of each operation based on the scores of the plurality of feature amounts included in the execution order estimation model, and more specifically, by the product of the scores of the plurality of feature amounts. The execution order of each operation is estimated based on the obtained final score.
- the command execution unit 111 is based on each intention included in the plurality of single sentences estimated by the intention estimation unit 106 and the execution order of the plurality of single sentences estimated by the execution order estimation unit 110.
- a machine command corresponding to the intention included in each of the plurality of single sentences is executed (step ST10).
- An operation corresponding to the first intention may be expressed as a first operation
- an operation corresponding to the second intention may be expressed as a second operation.
- the response generation unit 112 sets the synthesized sound “ ⁇ corresponding to the machine command executed by the command execution unit 111 as a stopover. Is generated and the notification unit 113 notifies the user of the synthesized sound (step ST11).
- the syntax analysis unit 104 divides the input compound sentence into a plurality of single sentences, and the feature amount extraction unit 108 extracts the feature amounts included in each of the plurality of single sentences, and determines the execution order. Since the execution order of processing corresponding to the intention included in each of a plurality of single sentences is determined by the unit 110 based on the extracted feature quantity, the intention can be estimated based on the execution order of the intention, and the user's intention can be accurately determined. Can be estimated.
- command execution unit 111 executes machine commands corresponding to the intentions included in each of the plurality of single sentences based on the execution order of the plurality of single sentences estimated by the execution order estimation unit 110, the operation burden on the user is reduced. It becomes possible to do.
- the execution order estimation unit 110 uses the main intention feature amount (main intention feature amount) obtained from the intention estimation unit 106 in addition to the feature amount obtained from the feature amount extraction unit 108 to determine the execution order. presume. The description of the main intention feature amount will be described later.
- FIG. 15 is a diagram illustrating a configuration example of the intention estimation apparatus 1B according to the second embodiment.
- the intention estimation apparatus 1B according to the second embodiment is intended in that it includes an execution order estimation model storage unit 109B and an execution order estimation unit 110B instead of the execution order estimation model storage unit 109 and the execution order estimation unit 110 according to the first embodiment. Different from the estimation device 1. Since other configurations are the same as those in the first embodiment, the same reference numerals as those in FIG.
- the execution order estimation model storage unit 109B stores an execution order estimation model (execution order estimation information) used to estimate the execution order of a plurality of single sentences included in the text.
- FIG. 16 is a diagram illustrating an example of an execution order estimation model according to the second embodiment.
- each execution order type is associated with a feature amount score included in each of a plurality of single sentences, and each execution order type and each of a plurality of single sentences are associated with each other. Is the information in which the score of the main intention feature amount included in the is associated.
- “single sentence 1_route addition” and “single sentence 2_destination setting” are main intention feature quantities.
- the main intention feature amount represents the main intention extracted from the intention estimation result by the intention estimation unit 106 to which single sentence position information is added. For example, “single sentence 1_addition of transit point” indicates that the intention included in the single sentence 1 is “addition of transit point”. “Simple sentence 2_destination setting” indicates that the intention included in simple sentence 2 is “destination setting”.
- the score in the execution order “single sentence 1 priority” is 0.2, and the score in the execution order “single sentence 2 priority” is 0.25.
- the score in the execution order “simultaneous execution” is 0.2
- the score in the execution order “single sentence 1 only” is 0.25
- the score in the execution order “single sentence 1 only” is 0.1.
- the score in the execution order “single sentence 2 priority” is 0.45.
- the score for “execution at the same time” is 0.05, the score for the execution order “execute only single statement 1” is 0.1, and the score for the execution order “execute only single statement 1” is 0.3.
- the score of the execution order “single sentence 2 priority” increases.
- FIG. 17 is a diagram illustrating a configuration example of the intention estimation apparatus 1B for explaining the generation process of the execution order estimation model according to the second embodiment.
- FIG. 18 is a flowchart for explaining an execution order estimation model generation process according to the second embodiment.
- the morpheme analysis unit 103 performs morpheme analysis on the simple sentence 1 and the simple sentence 2 in each sentence example of the learning data shown in FIG. 8 (step ST001). For example, sentence No. 1, the morphological analysis unit 103 performs morphological analysis on the simple sentence 1 “Because time is severe” and simple sentence 2 “Go on the expressway”. The morpheme analysis unit 103 outputs the result of the morpheme analysis to the feature amount extraction unit 108 and the intention estimation unit 106 via the syntax analysis unit 104.
- the feature quantity extraction unit 108 performs feature quantity extraction processing on the single sentence 1 and the single sentence 2 subjected to morphological analysis based on the feature quantity extraction rules stored in the feature quantity extraction rule storage unit 107 (step ST002). For example, sentence No. In other words, the feature amount extraction unit 108 extracts feature amounts “so” and “te”. Further, the feature quantity extraction unit 108 adds single sentence position information to the extracted feature quantities (step ST003). For example, the feature quantity extraction unit 108 adds the single sentence position information “single sentence 1” to the feature quantity “no” extracted from the single sentence 1 so that it becomes “single sentence 1_no”.
- the feature quantity extraction unit 108 adds the single sentence position information “single sentence 2” to the feature quantity “te” extracted from the single sentence 2 so as to be “single sentence 2_te”.
- the feature amount extraction unit 108 outputs information on the extracted feature amount to the execution order estimation model generation unit 115B.
- the intention estimation unit 106 estimates the intentions included in the single sentence 1 and the single sentence 2 subjected to the morphological analysis, and extracts them as main intention feature quantities (step ST004). For example, sentence No. In other words, the intention estimation unit 106 extracts the main intention feature quantity “route setting” from the single sentence 1, and extracts the main intention feature quantity “destination setting” from the single sentence 2. In addition, the intention estimation unit 106 adds the single sentence position information “single sentence 1” to the main intention feature amount extracted from the single sentence 1 so that “single sentence 1_route location setting” is obtained. Moreover, the intention estimation part 106 adds single sentence position information to the extracted main intention feature-value (step ST005).
- the intention estimation unit 106 adds the single sentence position information “single sentence 1” to the main intention feature amount extracted from the single sentence 1 to make “single sentence 1_route location setting”. In addition, the intention estimation unit 106 adds the single sentence position information “single sentence 2” to the main intention feature amount extracted from the single sentence 2 to make “single sentence 2_destination setting”. The intention estimation unit 106 outputs information regarding the extracted feature amount to the execution order estimation model generation unit 115B.
- the execution order estimation model generation unit 115B generates an execution order estimation model based on the feature amount, the main intention feature amount, and the execution order included in the learning data (step ST006).
- the execution order estimation model generation unit 115B performs the same processing as described above on all sentence examples included in the learning data, and finally generates an execution order estimation model as shown in FIG.
- FIG. 19 is a diagram showing an example of dialogue according to the second embodiment.
- FIG. 20 is a flowchart for explaining intention estimation processing according to the second embodiment.
- the speech recognition apparatus 1 utters “Please tell me when it beeps” (S11). On the other hand, the user utters “Stop first and go to ⁇ ⁇ ” (U22).
- the speech recognition unit 102 performs speech recognition processing on user input to convert it into text data
- the morpheme analysis unit 103 performs morpheme analysis processing on the text data (steps ST0001 and ST0002).
- syntax analysis section 104 performs syntax analysis processing on the text data (step ST0003).
- the text data related to the user input includes a plurality of single sentences such as single sentence 1 (first simple sentence) “go to XX first” and simple sentence 2 (second simple sentence) “go to ⁇ ”. It is divided into.
- the syntax analysis unit 104 outputs the text data of each single sentence subjected to the morphological analysis to the intention estimation unit 106 and the feature amount extraction unit 108, and the processes after step ST07 are performed.
- the processing in step ST0005 and step ST0006 is the same as that described in FIG.
- the intention estimation unit 106 performs intention estimation processing on the single sentence 1 and the single sentence 2 using the intention estimation model (step ST0007).
- the feature quantity extraction unit 108 performs feature quantity extraction processing for each of the single sentence 1 and the single sentence 2 using the feature quantity extraction rule (step ST0008).
- the feature quantity extraction unit 108 extracts feature quantities “single sentence 1_first” and “single sentence 1_te” for the single sentence 1. Note that there is no feature quantity extracted for the single sentence 2.
- Single sentence position information is also given to the feature amount.
- the feature quantity extraction unit 108 outputs information on the feature quantity extracted for each simple sentence to the execution order estimation unit 110B.
- the execution order estimation unit 110B extracts a main intention feature quantity from the intention of each single sentence estimated by the intention estimation unit 106 (step ST0009).
- Single sentence position information is also given to the main intention feature.
- the execution order estimation unit 110B has been described as performing the main intended feature quantity extraction process, but the present invention is not limited thereto.
- the intention estimation unit 106 may perform main intention feature amount extraction processing and output the extracted information to the execution order estimation unit 110B.
- the execution order estimation unit 110B estimates the execution order of operations corresponding to the intention of each single sentence based on the information on the feature quantity output from the feature quantity extraction unit 108 and the extracted information on the main intention feature quantity ( Step ST0010). Similar to the first embodiment, the execution order estimation unit 110B compares the feature quantity and the main intention feature quantity with the execution order estimation model to obtain the score of the feature quantity and the main intention feature quantity for each execution order. Then, the execution order estimation unit 110B obtains the product of the feature amount and the main intention feature amount score in each execution order using the calculation formula shown in FIG. That is, the execution order estimation unit 110B estimates the execution order of operations based on the final score obtained by multiplying the scores of a plurality of feature amounts included in each single sentence and the score of the main intention feature amount.
- FIG. 21 is a diagram showing a final score for each execution order according to the second embodiment.
- the score of the feature quantity “single sentence 1_first” is 0.45
- the score of the feature quantity “single sentence 1_te” is 0.2
- the main intention feature Since the score of the quantity “single sentence 1_route setting” is 0.2 and the score of the main intention feature quantity “single sentence 2_destination setting” is 0.1, the final score S that is the product of these is 1.8e ⁇ 3 is calculated.
- final scores are also calculated for the other execution orders.
- the execution order estimation unit 110 estimates the execution order “single sentence 2 priority” having the highest score among the calculated final scores of each execution order as an appropriate execution order.
- the command execution unit 111 is based on each intention included in the plurality of single sentences estimated by the intention estimation unit 106 and the execution order of the plurality of single sentences estimated by the execution order estimation unit 110.
- a machine command corresponding to the intention included in each of the plurality of single sentences is executed (step ST0011).
- the response generation unit 112 sets the synthesized sound “ ⁇ corresponding to the machine command executed by the command execution unit 111 as the destination. ”Is generated, and the notification unit 113 notifies the user of the synthesized sound (step ST0012).
- the execution order estimation unit 110B uses the main intention feature amount obtained from the intention estimation unit 106 in addition to the feature amount extracted by the feature amount extraction unit 108. Since the execution order of the operations corresponding to the intentions included in is estimated, it is possible to estimate the user's intentions more accurately than in the first embodiment.
- the navigation device can usually perform the route setting operation only after performing the destination setting operation, in the first embodiment, “go first to OO and go to ⁇ ⁇ . ", The machine command cannot be properly executed, and as a result, the user's intention cannot be estimated accurately.
- the execution order estimation unit 110B considers the execution order “single sentence 2” in consideration of the main intention feature amount. Since “priority” is determined, the machine command can be appropriately executed, and the user's intention can be accurately estimated.
- 1, 1B intention estimation device 101 speech input unit, 102 speech recognition unit, 103 morphological analysis unit, 104 syntax analysis unit, 105 intention estimation model storage unit, 106 intention estimation unit, 107 feature quantity extraction rule storage unit, 108 feature quantity Extraction unit, 109, 109B execution order estimation model storage unit, 110, 110B execution order estimation unit, 111 command execution unit, 112 response generation unit, 113 notification unit, 114 learning data storage unit, 115, 115B execution order estimation model generation Unit, 150 processing device, 160 storage device, 170 input device, 180 output device
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Machine Translation (AREA)
Abstract
Description
以下図面を用いて本発明の実施の形態1を説明する。
以下図面を用いて本発明の実施の形態2について説明する。実施の形態2では、実行順序推定部110は、特徴量抽出部108から得た特徴量に加え、意図推定部106から得た主意図の特徴量(主意図特徴量)を用いて実行順序を推定する。主意図特徴量の説明については後述する。
Claims (9)
- 複数の意図が含まれる複文に対して形態素解析を行う形態素解析部と、
前記形態素解析部による形態素解析された前記複文を構文解析して第1の単文と第2の単文とに分割する構文解析部と、
前記第1の単文に含まれる第1の意図と、前記第2の単文に含まれる第2の意図とを推定する意図推定部と、
前記第1の単文に含まれる操作の実行順序を示す形態素を第1の特徴量として抽出するとともに、前記第2の単文に含まれる操作の実行順序を示す形態素を第2の特徴量として抽出する特徴量抽出部と、
前記特徴量抽出部により抽出された前記第1の特徴量と前記第2の特徴量とに基づき、前記第1の意図に対応する第1の操作と前記第2の意図に対応する第2の操作の実行順序を推定する実行順序推定部と、を備えることを特徴とする意図推定装置。 - 前記第1の操作と前記第2の操作の実行順序の種別と、前記第1の特徴量と前記第2の特徴量を含む複数の特徴量のスコアとが対応付けて実行順序情報として記憶される実行順序情報記憶部を備え、
前記実行順序推定部は、前記実行順序情報に含まれる前記複数の特徴量のスコアに基づき前記第1の操作と前記第2の操作の実行順序を推定することを特徴とする請求項1に記載の意図推定装置。 - 前記実行順序推定部は、前記実行順序情報を用いて前記第1の特徴量のスコアと前記第2の特徴量のスコアとを求め、前記第1の特徴量のスコアと前記第2の特徴量のスコアとの積により得られた最終スコアに基づき前記第1の操作と前記第2の操作の実行順序を推定することを特徴とする請求項2に記載の意図推定装置。
- 前記特徴量抽出部は、前記第1の単文中における形態素の位置及び前記形態素の品詞から前記第1の特徴量を抽出するとともに、前記第2の単文中における形態素の位置及び前記形態素の品詞から前記第2の特徴量を抽出することを特徴とする請求項1から3のいずれか1項に記載の意図推定装置。
- 前記実行順序情報記憶部において、前記第1の操作と前記第2の操作の実行順序の種別には、前記第1の意図の分類または機能を示す第1の主意図特徴量と前記第2の意図の分類または機能を示す第2の主意図特徴量とを含む複数の主意図特徴量のスコアがさらに対応付けられており、
前記実行順序推定部は、前記複数の特徴量のスコアと前記複数の主意図特徴量のスコアとに基づき、前記第1の操作と前記第2の操作の実行順序を推定することを特徴とする請求項2から4のいずれか1項に記載の意図推定装置。 - 前記実行順序推定部は、前記第1の特徴量のスコア、前記第2の特徴量のスコア、前記第1の主意図特徴量のスコア、及び前記第2の主意図特徴量のスコアの積により得られた最終スコアに基づき、前記第1の操作と前記第2の操作の実行順序を推定することを特徴とする請求項5に記載の意図推定装置。
- 複数の意図が含まれる音声の入力を受け付ける音声入力部と、
前記音声入力部に入力された前記音声に対応する音声データを認識し、前記複数の意図が含まれる複文のテキストデータに変換する音声認識部を備えることを特徴とする請求項1から請求項6のいずれか1項に記載の意図推定装置。 - 前記実行順序推定部により推定された実行順序に基づき、前記第1の操作と前記第2の操作を実行する操作実行部と、
前記操作実行部により前記第1の操作と前記第2の操作とが実行されたことをユーザーに通知する通知部とを備えることを特徴とする請求項1から7のいずれか1項に記載の意図推定装置。 - 複数の意図が含まれる複文を形態素解析するステップと、
前記形態素解析された前記複文を構文解析して複数の単文に分割するステップと、
前記複数の単文それぞれに含まれる意図を推定するステップと、
前記複数の単文それぞれに含まれる、操作の実行順序を示す形態素を特徴量として抽出するステップと、
前記複数の単文それぞれに含まれる特徴量に基づき、前記複数の単文それぞれに含まれる意図に対応する各操作の実行順序を推定するステップと、を有することを特徴とする意図推定方法。
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201580074585.4A CN107209758A (zh) | 2015-01-28 | 2015-01-28 | 意图估计装置以及意图估计方法 |
US15/546,801 US10460034B2 (en) | 2015-01-28 | 2015-01-28 | Intention inference system and intention inference method |
DE112015006069.3T DE112015006069T5 (de) | 2015-01-28 | 2015-01-28 | Absichtsinferenzsystem und absichtsinferenzverfahren |
PCT/JP2015/000367 WO2016120904A1 (ja) | 2015-01-28 | 2015-01-28 | 意図推定装置及び意図推定方法 |
JP2016554689A JP6328260B2 (ja) | 2015-01-28 | 2015-01-28 | 意図推定装置及び意図推定方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2015/000367 WO2016120904A1 (ja) | 2015-01-28 | 2015-01-28 | 意図推定装置及び意図推定方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2016120904A1 true WO2016120904A1 (ja) | 2016-08-04 |
Family
ID=56542573
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2015/000367 WO2016120904A1 (ja) | 2015-01-28 | 2015-01-28 | 意図推定装置及び意図推定方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US10460034B2 (ja) |
JP (1) | JP6328260B2 (ja) |
CN (1) | CN107209758A (ja) |
DE (1) | DE112015006069T5 (ja) |
WO (1) | WO2016120904A1 (ja) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6275354B1 (ja) * | 2016-03-30 | 2018-02-07 | 三菱電機株式会社 | 意図推定装置及び意図推定方法 |
JP2018185561A (ja) * | 2017-04-24 | 2018-11-22 | 株式会社日立製作所 | 対話支援システム、対話支援方法、及び対話支援プログラム |
WO2018229937A1 (ja) * | 2017-06-15 | 2018-12-20 | 三菱電機株式会社 | 意図推定装置及び意図推定方法 |
WO2019087811A1 (ja) * | 2017-11-02 | 2019-05-09 | ソニー株式会社 | 情報処理装置、及び情報処理方法 |
JP2019204515A (ja) * | 2016-12-30 | 2019-11-28 | グーグル エルエルシー | 音声起動されたコンピュータネットワーク環境におけるシーケンス依存データメッセージ統合 |
JP2020047101A (ja) * | 2018-09-20 | 2020-03-26 | Zホールディングス株式会社 | コミュニケーション支援装置、ユーザデバイス、コミュニケーション支援方法、およびプログラム |
KR20200114079A (ko) * | 2019-03-27 | 2020-10-07 | 주식회사 포시에스 | 다수의 의도가 포함된 문장으로부터 의도 및 의도에 대응되는 내용을 검출하는 전자문서 시스템 |
JP2021089360A (ja) * | 2019-12-04 | 2021-06-10 | 本田技研工業株式会社 | エージェント装置、エージェント方法、及びプログラム |
JP7231171B1 (ja) | 2022-07-21 | 2023-03-01 | ソプラ株式会社 | 処理動作支援装置及びプログラム |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2017058804A (ja) * | 2015-09-15 | 2017-03-23 | 株式会社東芝 | 検出装置、方法およびプログラム |
CN108563790B (zh) * | 2018-04-28 | 2021-10-08 | 科大讯飞股份有限公司 | 一种语义理解方法及装置、设备、计算机可读介质 |
JP6516938B1 (ja) * | 2018-06-15 | 2019-05-22 | 三菱電機株式会社 | 機器制御装置、機器制御システム、機器制御方法、および、機器制御プログラム |
CN109522472A (zh) * | 2018-09-30 | 2019-03-26 | 中国农业大学烟台研究院 | 一种用户意图估计方法 |
CN109977211A (zh) * | 2019-03-28 | 2019-07-05 | 联想(北京)有限公司 | 一种数据处理方法及电子设备 |
CN110096595A (zh) * | 2019-05-06 | 2019-08-06 | 上海互问信息科技有限公司 | 一种基于混合策略的复句用户查询语句的意图识别方法 |
CN110297544B (zh) * | 2019-06-28 | 2021-08-17 | 联想(北京)有限公司 | 输入信息响应方法及装置、计算机***和可读存储介质 |
CN111737962A (zh) * | 2020-06-24 | 2020-10-02 | 平安科技(深圳)有限公司 | 一种实体修订方法、装置、计算机设备和可读存储介质 |
CN111933127B (zh) * | 2020-07-31 | 2024-06-25 | 升智信息科技(南京)有限公司 | 一种具备自学习能力的意图识别方法及意图识别*** |
CN112163086B (zh) * | 2020-10-30 | 2023-02-24 | 海信视像科技股份有限公司 | 多意图的识别方法、显示设备 |
CN114818644B (zh) * | 2022-06-27 | 2022-10-04 | 北京云迹科技股份有限公司 | 文本模板生成方法、装置、设备及存储介质 |
WO2024011097A1 (en) * | 2022-07-05 | 2024-01-11 | Staircase Ai Inc | Intention reader |
Family Cites Families (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3275840B2 (ja) | 1998-08-12 | 2002-04-22 | 日本電気株式会社 | 機器操作システム及び方法、並びに記録媒体 |
JP2000200273A (ja) | 1998-11-04 | 2000-07-18 | Atr Interpreting Telecommunications Res Lab | 発話意図認識装置 |
KR100446627B1 (ko) * | 2002-03-29 | 2004-09-04 | 삼성전자주식회사 | 음성대화 인터페이스를 이용한 정보제공장치 및 그 방법 |
JP4539149B2 (ja) | 2004-04-14 | 2010-09-08 | ソニー株式会社 | 情報処理装置および情報処理方法、並びに、プログラム |
CN101131689B (zh) * | 2006-08-22 | 2010-08-18 | 苗玉水 | 汉语外语句型转换双向机器翻译方法 |
JP2009020423A (ja) * | 2007-07-13 | 2009-01-29 | Fujitsu Ten Ltd | 音声認識装置および音声認識方法 |
US8380489B1 (en) * | 2009-02-11 | 2013-02-19 | Guangsheng Zhang | System, methods, and data structure for quantitative assessment of symbolic associations in natural language |
JP2010224194A (ja) | 2009-03-23 | 2010-10-07 | Sony Corp | 音声認識装置及び音声認識方法、言語モデル生成装置及び言語モデル生成方法、並びにコンピューター・プログラム |
KR101295642B1 (ko) * | 2009-12-07 | 2013-08-13 | 한국전자통신연구원 | 음성인식결과 문장에 대한 문형분류장치 및 방법 |
KR20110070057A (ko) * | 2009-12-18 | 2011-06-24 | 한국전자통신연구원 | 자연어 기반 여행 추천 장치 및 방법 |
JP2012047924A (ja) | 2010-08-26 | 2012-03-08 | Sony Corp | 情報処理装置、および情報処理方法、並びにプログラム |
CN104020840B (zh) * | 2013-03-03 | 2019-01-11 | 上海能感物联网有限公司 | 外语文本遥控计算机程序运行的方法 |
CN104133812B (zh) * | 2014-07-17 | 2017-03-08 | 北京信息科技大学 | 一种面向用户查询意图的汉语句子相似度分层计算方法及装置 |
-
2015
- 2015-01-28 JP JP2016554689A patent/JP6328260B2/ja active Active
- 2015-01-28 CN CN201580074585.4A patent/CN107209758A/zh active Pending
- 2015-01-28 WO PCT/JP2015/000367 patent/WO2016120904A1/ja active Application Filing
- 2015-01-28 US US15/546,801 patent/US10460034B2/en active Active
- 2015-01-28 DE DE112015006069.3T patent/DE112015006069T5/de active Pending
Non-Patent Citations (3)
Title |
---|
"Kakushu Media Shori no Yoso Gijutsu to Engine Kaihatsu ni Torikumu NTT Media Intelligence Kenkyusho no R&D Tenkai Zatsudan Taiwa Gijutsu Onsei Agent no Seino Kojo ni wa Zatsudan ga Hitsuyo Haba Hiroi Wadai ni Taio dekiru Taiwa Gijutsu o Kaihatsu", BUSINESS COMMUNICATION, vol. 51, no. 2, 1 February 2014 (2014-02-01), pages 20 - 21, ISSN: 0385-695X * |
HIROSHI FUJIMOTO ET AL.: "Development of Car Navigation System Operated by Naturally Speaking", THE TRANSACTIONS OF THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS (J96-D), vol. J96-D, no. 11, 1 November 2013 (2013-11-01), pages 2815 - 2824, ISSN: 1880-4535 * |
MIZUKI FUJISAWA ET AL.: "A control method for the multiple actions of animated agents by natural language", IEICE TECHNICAL REPORT, vol. 101, no. 243, 24 July 2001 (2001-07-24), pages 31 - 38, ISSN: 0913-5685 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6275354B1 (ja) * | 2016-03-30 | 2018-02-07 | 三菱電機株式会社 | 意図推定装置及び意図推定方法 |
JP2019204515A (ja) * | 2016-12-30 | 2019-11-28 | グーグル エルエルシー | 音声起動されたコンピュータネットワーク環境におけるシーケンス依存データメッセージ統合 |
JP2018185561A (ja) * | 2017-04-24 | 2018-11-22 | 株式会社日立製作所 | 対話支援システム、対話支援方法、及び対話支援プログラム |
WO2018229937A1 (ja) * | 2017-06-15 | 2018-12-20 | 三菱電機株式会社 | 意図推定装置及び意図推定方法 |
JPWO2018229937A1 (ja) * | 2017-06-15 | 2019-07-11 | 三菱電機株式会社 | 意図推定装置及び意図推定方法 |
JPWO2019087811A1 (ja) * | 2017-11-02 | 2020-09-24 | ソニー株式会社 | 情報処理装置、及び情報処理方法 |
WO2019087811A1 (ja) * | 2017-11-02 | 2019-05-09 | ソニー株式会社 | 情報処理装置、及び情報処理方法 |
JP2020047101A (ja) * | 2018-09-20 | 2020-03-26 | Zホールディングス株式会社 | コミュニケーション支援装置、ユーザデバイス、コミュニケーション支援方法、およびプログラム |
JP7182969B2 (ja) | 2018-09-20 | 2022-12-05 | ヤフー株式会社 | コミュニケーション支援装置、ユーザデバイス、コミュニケーション支援方法、およびプログラム |
KR20200114079A (ko) * | 2019-03-27 | 2020-10-07 | 주식회사 포시에스 | 다수의 의도가 포함된 문장으로부터 의도 및 의도에 대응되는 내용을 검출하는 전자문서 시스템 |
KR102189567B1 (ko) | 2019-03-27 | 2020-12-11 | 주식회사 포시에스 | 다수의 의도가 포함된 문장으로부터 의도 및 의도에 대응되는 내용을 검출하는 전자문서 시스템 |
JP2021089360A (ja) * | 2019-12-04 | 2021-06-10 | 本田技研工業株式会社 | エージェント装置、エージェント方法、及びプログラム |
JP7178983B2 (ja) | 2019-12-04 | 2022-11-28 | 本田技研工業株式会社 | エージェント装置、エージェント方法、及びプログラム |
JP7231171B1 (ja) | 2022-07-21 | 2023-03-01 | ソプラ株式会社 | 処理動作支援装置及びプログラム |
JP2024014130A (ja) * | 2022-07-21 | 2024-02-01 | ソプラ株式会社 | 処理動作支援装置及びプログラム |
Also Published As
Publication number | Publication date |
---|---|
JP6328260B2 (ja) | 2018-05-23 |
JPWO2016120904A1 (ja) | 2017-04-27 |
US10460034B2 (en) | 2019-10-29 |
DE112015006069T5 (de) | 2017-10-12 |
US20170371863A1 (en) | 2017-12-28 |
CN107209758A (zh) | 2017-09-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6328260B2 (ja) | 意図推定装置及び意図推定方法 | |
KR102413692B1 (ko) | 음성 인식을 위한 음향 점수 계산 장치 및 방법, 음성 인식 장치 및 방법, 전자 장치 | |
US11450313B2 (en) | Determining phonetic relationships | |
EP2700071B1 (en) | Speech recognition using multiple language models | |
KR102386854B1 (ko) | 통합 모델 기반의 음성 인식 장치 및 방법 | |
CN110675855B (zh) | 一种语音识别方法、电子设备及计算机可读存储介质 | |
EP2869298A1 (en) | Information identification method and apparatus | |
US9589563B2 (en) | Speech recognition of partial proper names by natural language processing | |
US9734821B2 (en) | Testing words in a pronunciation lexicon | |
KR102117082B1 (ko) | 음성 인식 방법 및 음성 인식 장치 | |
US20160314116A1 (en) | Interpretation apparatus and method | |
EP3667660A1 (en) | Information processing device and information processing method | |
US20190005950A1 (en) | Intention estimation device and intention estimation method | |
JP6875819B2 (ja) | 音響モデル入力データの正規化装置及び方法と、音声認識装置 | |
KR20160058531A (ko) | 딥 러닝을 이용하는 구문 분석 모델 구축 방법 및 이를 수행하는 장치 | |
JP4700522B2 (ja) | 音声認識装置及び音声認識プログラム | |
JP2015045689A (ja) | 音声認識システムについての音声認識結果を評価する方法、並びに、そのコンピュータ及びコンピュータ・プログラム | |
CN116361316A (zh) | 一种语义引擎适配方法、装置、设备及存储介质 | |
JPWO2018229937A1 (ja) | 意図推定装置及び意図推定方法 | |
JP2018077698A (ja) | 発話文抽出装置、不適切話題語リスト生成装置、方法、及びプログラム | |
CN112329484A (zh) | 自然语言的翻译方法及装置 | |
CN112951204B (zh) | 语音合成方法和装置 | |
JP2009129405A (ja) | 感情推定装置、事例感情情報生成装置、及び感情推定プログラム | |
Araújo et al. | Context-sensitive ASR for controlling the navigation of mobile robots |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2016554689 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15879805 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15546801 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 112015006069 Country of ref document: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15879805 Country of ref document: EP Kind code of ref document: A1 |