TW202422535A - Language data processing system and method and computer program product - Google Patents

Abstract

A language data processing method implemented by a language data processing system comprises: (A) using a language processing model to identify multiple designated intents expressed by a speech data, and determining each designated intent as one of a clear designated intent and a fuzzy designated intent according to the vocabulary in the speech data, wherein each designated intent corresponds to one of a plurality of intent labels; (B) determining N target designated intents from the clear designated intents, and executing a control program corresponding to the intent label corresponding to each target designated intent, wherein N is an integer greater than or equal to 1.

Description

Language data processing system and method and computer program product

The present invention relates to a data processing system, and more particularly to a language data processing system suitable for processing oral contents of a user. The present invention also relates to a language data processing method suitable for processing oral contents of a user, and a computer program product for enabling an electronic device to implement the language data processing method.

With the development of natural language processing technology, the voice control functions of more and more electronic devices in the existing technology are no longer limited to preset specific voice commands, but allow users to describe their needs in a more colloquial way.

In terms of the speaking habits of ordinary people, it is obviously a more convenient way of expression to express multiple requirements at once in one sentence. However, this way of expression will make the sentence longer, and some requirements may not be described clearly enough, which will cause obstacles to the existing technology when performing intent analysis. Therefore, how to better process multiple requirements expressed by users at one time based on natural language processing technology has become an issue to be explored by the present invention.

Therefore, one of the objects of the present invention is to provide a language data processing system that can provide improvements to the prior art.

The language data processing system of the present invention comprises a processing unit and a storage unit electrically connected to the processing unit. The storage unit stores a language processing model implemented by machine learning technology. The language processing model comprises a plurality of intention labels, and each intention label corresponds to a control program related to the operation mode of the processing unit. The processing unit is used to: use the language processing model to identify multiple designated intentions expressed by a speech data, and determine each designated intention as one of a clear designated intention and a fuzzy designated intention according to the vocabulary in the speech data, wherein each designated intention corresponds to one of the intention labels; determine N target designated intentions from the clear designated intentions, and execute the control program corresponding to the intention label corresponding to each target designated intention, wherein N is an integer greater than or equal to 1.

In some implementations of the language data processing system of the present invention, for each designated intent, the processing unit determines the designated intent as the clear designated intent or the fuzzy designated intent in the manner that includes: the processing unit determines whether there are one or more key words in the speech and text data that can be used by the processing unit to execute the control program related to the designated intent; if the processing unit determines that the speech and text data contains the key words, the processing unit determines the designated intent as the clear designated intent; if the processing unit determines that the key words do not exist in the speech and text data, the processing unit determines the designated intent as the fuzzy designated intent.

In some implementations of the language data processing system of the present invention, after the processing unit determines each designated intent as the clear designated intent or the fuzzy designated intent, the processing unit is further used to: when one or more of the designated intents are determined by the processing unit as the fuzzy designated intent, for at least one of the fuzzy designated intents, use the language processing model to generate a query message corresponding to the fuzzy designated intent, and make the query The message is output by an output module; after the query message is output by the output module, when another voice-text data corresponding to another voice input is obtained, it is determined whether the semantics of the other voice-text data matches the fuzzy specified intent, so that the processing unit can execute the control program related to the fuzzy specified intent, and, when it is determined that the semantics of the other voice-text data matches the fuzzy specified intent, the fuzzy specified intent is changed to a clear specified intent.

In some implementation aspects of the language data processing system of the present invention, the processing unit determines the N target specified intentions from the explicit specified intentions in a manner that includes: the processing unit determines whether there are multiple explicit specified intentions among the explicit specified intentions that conflict with each other and are collectively regarded as a group of conflicting intentions, and, when it is determined that the group of conflicting intentions exists among the explicit specified intentions, the processing unit only uses a single explicit specified intention among the group of conflicting intentions as one of the target specified intentions of the N target specified intentions.

In some implementations of the language data processing system of the present invention, one of the intent labels is set as an exclusive intent label, and the processing unit determines the N target specified intentions from the explicitly specified intentions in a manner that includes: the processing unit determines whether the intent label corresponding to one of the explicitly specified intentions is the exclusive intent label; if the determination result is yes, the processing unit uses the explicitly specified intent corresponding to the exclusive intent label as the only target specified intention.

In some implementations of the language data processing system of the present invention, there is an order between the intent labels, and when the number of target designated intentions determined by the processing unit is multiple, the processing unit executes the control programs corresponding to the intent labels corresponding to the target designated intentions one by one in sequence according to the order between the intent labels respectively corresponding to the target designated intentions.

Another object of the present invention is to provide a language data processing method that can provide improvement over the prior art.

The language data processing method of the present invention is implemented by a language data processing system. The language data processing system includes a processing unit and a storage unit electrically connected to the processing unit. The storage unit stores a language processing model implemented by machine learning technology. The language processing model includes a plurality of intention labels, and each intention label corresponds to a control program related to the operation mode of the processing unit. The language data processing method includes: (A) the processing unit uses the language processing model to identify a plurality of intention labels from a speech data; (A) the processing unit determines N target designated intentions from the clear designated intentions, and executes the control program corresponding to the intention label corresponding to each target designated intention, wherein N is an integer greater than or equal to 1.

In some implementations of the language data processing method of the present invention, in step (A), for each designated intent, the processing unit determines the designated intent as the clear designated intent or the fuzzy designated intent in a manner that includes: the processing unit determines whether there are one or more key words in the speech and text data that can be used by the processing unit to execute the control program related to the designated intent; if the processing unit determines that the speech and text data contains the key words, the processing unit determines the designated intent as the clear designated intent; if the processing unit determines that the speech and text data does not contain the key words, the processing unit determines the designated intent as the fuzzy designated intent.

In some embodiments of the language data processing method of the present invention, the language data processing method further includes between steps (A) and (B): (C) when one or more of the specified intents are determined by the processing unit to be the fuzzy specified intent, for at least one of the fuzzy specified intents, the processing unit generates a query message corresponding to the fuzzy specified intent using the language processing model, and causes the query message to be output by an output module; (D) in the processing After the query message is output by the output module, when the processing unit obtains another voice-text data corresponding to another voice input, the processing unit determines whether the semantics of the other voice-text data matches the fuzzy designated intent, so that the processing unit can execute the control program related to the fuzzy designated intent. Moreover, when the processing unit determines that the semantics of the other voice-text data matches the fuzzy designated intent, the processing unit changes the fuzzy designated intent into a clear designated intent.

In some implementation aspects of the language data processing method of the present invention, in step (B), the processing unit determines the N target specified intentions from the explicit specified intentions in a manner that includes: the processing unit determines whether there are multiple explicit specified intentions among the explicit specified intentions that conflict with each other and are collectively regarded as a group of conflicting intentions, and, when it is determined that the group of conflicting intentions exists among the explicit specified intentions, the processing unit only uses a single explicit specified intention of the group of conflicting intentions as one of the target specified intentions of the N target specified intentions.

In some implementations of the language data processing method of the present invention, one of the intent labels is set as an exclusive intent label; wherein, in step (B), the processing unit determines the N target specified intentions from the explicitly specified intentions in a manner that includes: the processing unit determines whether the intent label corresponding to one of the explicitly specified intentions is the exclusive intent label; if the determination result is yes, the processing unit uses the explicitly specified intent corresponding to the exclusive intent label as the only target specified intention.

In some implementations of the language data processing method of the present invention, there is an order between the intent labels; wherein, in step (B), when the number of target designated intents determined by the processing unit is multiple, the processing unit executes the control programs corresponding to the intent labels corresponding to the target designated intents one by one in sequence according to the order between the intent labels respectively corresponding to the target designated intents.

Another object of the present invention is to provide a computer program product that can provide improvements to the prior art.

The computer program product of the present invention includes a language processing model implemented using machine learning technology, wherein the language processing model includes multiple intention labels, and each intention label corresponds to a control program related to the operation mode of an electronic device. The computer program product is used to be loaded and run by the electronic device so that the electronic device implements the language data processing method described in any of the aforementioned implementations.

The effect of the present invention is that after the language data processing system obtains the speech and text data and identifies the designated intents therefrom, it can classify each designated intent as a clear designated intent or a fuzzy designated intent according to the vocabulary of the speech and text data, and execute the relevant control program only according to the (such) clear designated intent. Therefore, even if some designated intents expressed by the speech and text data are not clear enough (i.e., there are fuzzy designated intents), the language data processing system can still process the designated intents that are expressed clearly enough (i.e., the clear designated intents) to meet the needs of the user.

Before the invention is described in detail, it should be noted that, if not specifically defined, the "electrical connection" described in this patent specification generally refers to "wired electrical connection" achieved by connecting multiple electronic devices/devices/components to each other through conductive materials, and "radio connection" for unidirectional/bidirectional wireless signal transmission through wireless communication technology. In addition, the "electrical connection" described in this patent specification also generally refers to "direct electrical connection" formed by direct connection between multiple electronic devices/devices/components, and "indirect electrical connection" formed by indirect connection between multiple electronic devices/devices/components through other electronic devices/devices/components.

Referring to FIG. 1 , an embodiment of the language data processing system 1 of the present invention is applicable to multiple end devices 5 (only one of which is shown in FIG. 1 ) being electrically connected via a network, so that the end devices 5 can communicate with the language data processing system 1 via the network. However, for the convenience of description, the operation of the present embodiment will be explained below with reference to the end device 5 shown in FIG. 1 .

In the application of this embodiment, the end device 5 includes a processing module 51, an input module 52 electrically connected to the processing module 51, and an output module 53 electrically connected to the processing module 51. The processing module 51 is a central processor, the input module 52 has at least a microphone electrically connected to the processing module 51, and the output module 53 has at least a speaker and a display electrically connected to the processing module 51. More specifically, in the application of this embodiment, the user device 5 is a voice control service device suitable for being installed in a guest room of an accommodation facility for use by the guest, and the user device 5 can use the microphone of the input module 52 to receive the voice of the guest, use the speaker of the output module 53 to play the voice used to respond to the guest, and use the display of the output module 53 to display information for the guest's reference. In addition, the processing module 51 of the user device 5 can also output control instructions in a wireless manner (such as Wi-Fi or Bluetooth) to control the operation of electronic equipment such as air conditioners, telephones, televisions and lamps in the guest room.

It should be noted that in different application modes of the present embodiment, the user-end device 5 can also be implemented as a mobile electronic device (such as a mobile phone, a tablet computer or a laptop), a desktop computer or an Internet-connected television. Therefore, the language data processing system 1 is not limited to being used as a voice-controlled service device in guest rooms of accommodation facilities.

In the present embodiment, the language data processing system 1 is a server device, and the language data processing system 1 includes a processing unit 11 and a storage unit 12 electrically connected to the processing unit 11. The processing unit 11 is a central processing unit with data calculation and processing functions in the present embodiment, and the storage unit 12 is a data storage device (such as a hard disk) for storing digital data. However, in other embodiments, the processing unit 11 can also be implemented as a combination of multiple central processing units, and the storage unit 12 can also be implemented as different types of computer-readable recording media (such as flash memory), or a combination of multiple computer-readable recording media of the same/different types, and is not limited to the present embodiment. It should be noted that in different embodiments, the language data processing system 1 can also be implemented as different types of electronic devices such as mobile phones, tablet computers, laptop computers, desktop computers, etc., or the language data processing system 1 can also be implemented as multiple server devices electrically connected to each other. Therefore, it should be understood that the actual implementation of the language data processing system 1 in terms of hardware is not limited to this embodiment.

In this embodiment, the storage unit 12 of the language data processing system 1 stores a language processing model M implemented using machine learning technology and capable of being run by the processing unit 11, as well as a plurality of control programs P that are pre-set and used for the processing unit 11 to run to control the user device 5.

The language processing model M itself is a neural network capable of realizing natural language understanding and generation, wherein the language processing model M includes a plurality of pre-set intention labels L, and each intention label L is used to present the semantic features of a user's intention, that is, to represent a user's intention. In more detail, the language processing model M is trained by machine learning using at least a plurality of sentence data as training data after the intention labels L are set. Each sentence data is a sentence expressing an intention in natural language form, such as "help me call room 303", "set the alarm for 8 o'clock tomorrow morning", "help me turn on the TV", and "turn off the light in the bathroom". Preferably, some of the sentence data may be sentences expressing multiple different intentions at once in natural language, such as "turn on the TV and headlights, and turn on the air conditioner", "turn off the music, and call room 501", and "play a Jay Chou song, and then help me check the weather in Taipei tomorrow", etc., but not limited to this. By using the sentence data for machine learning, the language processing model M can be used to perform semantic analysis on the text data after being trained, and identify one or more user intentions expressed by the text data from the text data according to the intention labels L.

For the intention labels L included in the language processing model M, the semantic features represented by each intention label L represent a user intention. In this embodiment, since the language data processing system 1 is a voice-controlled service device (i.e., the user-end device 5) applied in the guest rooms of accommodation facilities, the intention labels L are mainly related to the intentions that the user may have in the guest rooms, such as "making a call", "setting an alarm", "turning on home appliances", "turning off home appliances", "querying data" and "playing music". However, the language data processing system 1 is not limited to the voice-controlled service device applied in the guest rooms of accommodation facilities, and therefore, the intentions represented by the intention labels L are certainly not limited to the aforementioned examples.

In this embodiment, there is a pre-set order between the intent labels L. For example, for the two intent labels L representing "turn on home appliances" and "query data", the order of the two intent labels L is higher than the intent label L representing "play music". The meaning of the setting of the above order is that if the user expresses three requirements of "turn on home appliances", "query data" and "play music" at the same time, the requirements of "turn on home appliances" and "query data" should be met first, and the requirement of "play music" should be met last. It should be understood that the order of the intent labels L can be freely designed and adjusted based on user experience or other empirical factors, so the actual ordering method is not a technical focus and will not be described in detail here.

Furthermore, in the present embodiment, one of the intent labels L is set as an exclusive intent label L' (shown in FIG. 1), and the user intent represented by the exclusive intent label L' is regarded as a top priority intent that has the highest priority and must be satisfied first. Taking the present embodiment as an example, the exclusive intent label L' is, for example, the intent label L representing the user intent of "making an outgoing call", however, which user intent intent label L is set as the exclusive intent label L' can be freely set according to different considerations and needs, and, in different embodiments, multiple of the intent labels L can be set as multiple exclusive intent labels L' with priorities among each other, and the example is not limited to the present embodiment.

In this embodiment, the control programs P stored in the storage unit 12 are used for the processing unit 11 to communicate with the processing module 51 of the user device 5 through the network, so as to control the user device 5 in different ways through the processing module 51. More specifically, each control program P in this embodiment is used to make the user device 5 output specific data in a specific way under the control of the processing unit 11, such as outputting sound data in a playing way, outputting text and/or image data in a display way, or outputting control command data to the air conditioner, telephone, television, lamp and other electronic equipment in the same guest room in a wired/wireless communication way to control their operation. Furthermore, in the present embodiment, there is a correspondence relationship between the control programs P and the intention labels L included in the language processing model M. More specifically, each intention label L in the language processing model M corresponds to at least one of the control programs P. For example, the intention label L representing "set alarm" corresponds to one of the control programs P used to enable the user device 5 to set its own built-in alarm function. For example, the intention label L representing "make an outgoing call" corresponds to another control program P used to enable the user device 5 to output a voice call request to a call device.

With reference to FIG. 2 , the following exemplarily describes in detail how the language data processing system 1 of this embodiment cooperates with the user device 5 to implement a language data processing method.

First, in step S1, the processing unit 11 obtains a speech text data corresponding to a speech input.

More specifically, in this embodiment, the voice input is a series of user voice signals received by the processing module 51 of the user device 5 via the microphone of the input module 52. In other words, the voice input is a sentence spoken by the user. Furthermore, when the processing module 51 of the user device 5 receives the voice input, it will immediately process the voice input using voice-to-text technology to generate the voice-to-text data that matches the content of the voice input, and transmit the voice-to-text data to the processing unit 11 of the language data processing system 1. However, in different embodiments, the processing module 51 of the user device 5 may also transmit the voice input to the processing unit 11 immediately upon receiving the voice input, and the processing unit 11 performs voice-to-text processing on the voice input to generate the voice-to-text data. Therefore, the method by which the processing unit 11 obtains the voice-to-text data is not limited to this embodiment.

After the processing unit 11 obtains the voice and text data, the process proceeds to step S2.

In step S2, the processing unit 11 loads and runs the language processing model M, and uses the language processing model M to perform an intention analysis process on the speech and text data. In this embodiment, the processing unit 11 performs the intention analysis process on the speech and text data in a manner that includes identifying multiple designated intentions expressed by the speech and text data from the speech and text data, and determining each designated intention as one of a clear designated intention and a fuzzy designated intention based on multiple vocabularies possessed by the speech and text data. The "vocabulary" generally refers to a single character and a word composed of multiple characters. On the other hand, the processing unit 11 performs intent recognition on the speech data according to the intent labels L in the language processing model M, so each specified intent recognized by the processing unit 11 corresponds to a matching intent label L among the intent labels L.

To be more specific, for each designated intent identified by the processing unit 11, the processing unit 11 determines whether the designated intent is the explicit designated intent or the fuzzy designated intent by determining whether there are one or more key words in the vocabularies of the speech and text data that semantically match the designated intent and enable the processing unit 11 to execute the control program P corresponding to the intent label L corresponding to the designated intent. If the processing unit 11 determines that there are one or more key words in the speech data that semantically match the designated intent, so that the processing unit 11 can execute the control program P corresponding to the designated intent according to the key words, the processing unit 11 will determine the designated intent as the explicit designated intent. On the contrary, if the processing unit 11 determines that there are no key words in the speech data that semantically match the designated intent, so that the processing unit 11 cannot execute the control program P corresponding to the designated intent, the processing unit 11 will determine the designated intent as the ambiguous designated intent.

For example, assuming that one of the designated intents identified by the processing unit 11 corresponds to the intent label L representing "making an outgoing call", the processing unit 11 will determine whether any of the vocabularies in the voice-text data indicates a specific calling party (such as other room numbers in the same accommodation facility, a specific store, institution or telephone number). If so, the processing unit 11 will use the vocabulary indicating the calling party as a key vocabulary matching one of the designated intents, and determine the key vocabulary of one of the designated intents to be a clear designated intent. If not, the processing unit 11 will determine that there is no vocabulary matching one of the designated intents in the voice-text data, and determine the key vocabulary of one of the designated intents to be an ambiguous designated intent.

For another example, assuming that one of the other specified intents identified by the processing unit 11 corresponds to the intent label L representing "set an alarm", the processing unit 11 will determine whether any of the words in the speech data indicates a specific alarm time (such as "8:00 a.m." or "one hour later"). If so, the processing unit 11 The vocabulary indicating the alarm time will be used as the key vocabulary matching the other specified intention, and the key vocabulary of the other specified intention will be determined as a clear specified intention. If not, the processing unit 11 will determine that there is no vocabulary matching the other specified intention in the voice and text data, and the key vocabulary of the other specified intention will be determined as a fuzzy specified intention.

It is to be noted that, in the present embodiment, each intention label L includes a necessary information feature, and the necessary information feature is a semantic feature used to indicate the necessary information of the user intention represented by the intention label L. For example, the intention label L represents the user intention of "making an outgoing call". For example, "a specific person to call" is the necessary information of the user intention of "making an outgoing call", so the necessary information feature of the intention label L is a semantic feature used to indicate the "specific person to call". Thus, by running the language processing model M, the processing unit 11 can determine whether there is a key word matching the necessary information feature in the speech data according to the necessary information feature of the intention label L corresponding to each designated intention, and further determine whether the designated intention is a clear designated intention or a fuzzy designated intention. However, in different embodiments, the function of determining each designated intention as a clear designated intention or a fuzzy designated intention can also be achieved by allowing the language processing model M to perform machine learning, and is not limited to this embodiment.

After the processing unit 11 determines each designated intention as a clear designated intention or a fuzzy designated intention, the process proceeds to step S3.

In step S3, the processing unit 11 determines whether any of the specified intentions belongs to the fuzzy specified intention, that is, whether any fuzzy specified intention exists among the specified intentions. If the processing unit 11 determines that K of the specified intentions belong to the fuzzy specified intention (K is an integer greater than or equal to 1), that is, there is a fuzzy specified intention, the process proceeds to step S4. On the other hand, if the processing unit 11 determines that none of the specified intentions belongs to the fuzzy specified intention, that is, there is no fuzzy specified intention, the process proceeds to step S6.

In step S4 following step S3, when there are K fuzzy specified intentions among the specified intentions, for at least one of the K fuzzy specified intentions (hereinafter referred to as "the fuzzy specified intention"), the processing unit 11 uses the language processing model M to generate a query message corresponding to the fuzzy specified intention and used to guide the user to supplement the explanation in spoken language, and transmits the query message to the processing module 51 of the user device 5, so that the processing module 51 outputs the query message through the speaker of the output module 53. For example, if the fuzzy specified intent corresponds to the intent label L representing "making a call", the inquiry message may be, for example, "Who do you want to call?" For another example, if the fuzzy specified intent corresponds to the intent label L representing "setting an alarm", the inquiry message may be, for example, "What time do you want to set the alarm?", but the present invention is not limited thereto.

After the processing unit 11 causes the query message to be output by the output module 53 of the user device 5, the process proceeds to step S5.

In step S5, when the processing unit 11 obtains another voice text data corresponding to another voice input within a waiting response period after the query message is output by the output module 53 (for example, within five seconds after the query message is output), the processing unit 11 determines whether the semantics of the other voice text data matches the fuzzy specified intent, so that the processing unit 11 can execute the control program P corresponding to the intent label L corresponding to the fuzzy specified intent. More specifically, the processing unit 11 determines whether any of the other one or more vocabularies of the other voice text data has a semantic match with the fuzzy specified intent and can be used as a key vocabulary corresponding to the fuzzy specified intent.

If the processing unit 11 determines that the semantics of the other voice and text data matches the fuzzy designated intent, the processing unit 11 changes the fuzzy designated intent into another clear designated intent, that is, converts the fuzzy designated intent into a clear designated intent based on the key words in the other voice and text data. On the other hand, if the processing unit 11 determines that the semantics of the other voice and text data does not match the fuzzy designated intent, the processing unit 11 can, for example, generate and transmit another inquiry message to the user device 5 for guiding the user to supplement the explanation in oral language, so that it is output by the speaker of the user device 5. However, if the number of times the processing unit 11 outputs inquiry messages based on the same fuzzy specified intent has reached a preset upper threshold of the number of inquiries, but the fuzzy specified intent is still unable to be converted into a clear specified intent, the processing unit 11 will not continue to process the fuzzy specified intent, that is, it will abandon (ignore) the fuzzy specified intent.

For example, assuming that the fuzzy specified intent corresponds to the intent label L representing "making an outgoing call", and the inquiry message is "Who do you want to call?", if the other voice text data is "Call Room 602 for me", the processing unit 11 will determine that "Room 602" in the other voice text data semantically matches the fuzzy specified intent, thereby changing the fuzzy specified intent to a clear specified intent. For another example, assuming that the fuzzy specified intent corresponds to the intent label L representing "set an alarm", and the inquiry message is "What time do you want to set the alarm for?", if the other voice-text data is "Set it to 5:30 p.m.", the processing unit 11 will determine that "5:30 p.m." in the other voice-text data semantically matches the fuzzy specified intent, and thus change the fuzzy specified intent to a clear specified intent.

After the processing unit 11 determines whether to change the fuzzy designation intention to a clear designation intention, the process proceeds to step S6.

In step S6, the processing unit 11 uses the language processing model M to perform a conflict elimination process and a screening process on all the explicit designation intentions to determine N target designation intentions from the explicit designation intentions. In this embodiment, the processing unit 11 may, for example, first perform the conflict elimination process and then perform the screening process, but is not limited thereto. In addition, N is an integer greater than or equal to 1, that is, the target designation intention determined by the processing unit 11 may be a single one or multiple ones.

In this embodiment, the processing unit 11 uses the language processing model M to perform the conflict elimination processing on the explicitly specified intentions, which is to determine whether there are multiple explicitly specified intentions among the explicitly specified intentions that conflict with each other, and if it is determined that there are indeed multiple explicitly specified intentions among the explicitly specified intentions that conflict with each other, the processing unit 11 will treat the explicitly specified intentions that conflict with each other as a group of conflicting intentions. Next, when it is determined that there are one or more groups of conflicting intents in the explicit designated intents, for each group of conflicting intents, the processing unit 11 selects a single explicit designated intent from the group of conflicting intents as a candidate designated intent, which is equivalent to discarding the other explicit designated intents in the same group of conflicting intents except the candidate designated intent. Moreover, in an implementation mode of the present embodiment, the processing unit 11 selects the explicit designated intent in the group of conflicting intents that is first expressed by the voice and text data as the candidate designated intent, which is equivalent to discarding the other explicit designated intents in the group of conflicting intents that are expressed by the voice and text data more slowly, but the present invention is not limited thereto.

It is particularly noted that the aforementioned "multiple conflicting explicit intents" means that the explicit intents are semantically contradictory, or the control programs P corresponding to the explicit intents are not suitable for simultaneous execution by the processing unit 11. Whether the explicit intents are semantically contradictory can be realized by using the natural language understanding function of the language processing model M, and which control programs P are not suitable for simultaneous execution are pre-set. For example, multiple of the control programs P used to control the user end device 5 to play sound can be collectively set as control programs P that are not suitable for simultaneous execution by the processing unit 11.

For example, assuming that the voice text data is "play the music louder, then lower the volume, and stop playing songs", the processing unit 11 will identify three semantically contradictory explicit intents, namely "increase the volume of music", "lower the volume of music", and "stop playing music" from the voice text data. Moreover, since the voice text data first expresses the meaning of "increase the volume of music", the processing unit 11 will use the explicit intent representing "increase the volume of music" as a candidate intent, and discard the other two explicit intents, namely "lower the volume of music" and "stop playing music", which are expressed more slowly.

For another example, assuming that the voice text data is "Help me play news and music", the processing unit 11 will identify two explicit designated intents, namely "play news" and "play music", from the voice text data. The two explicit designated intents are used to instruct the user end device 5 to play sound, but based on the consideration of listening experience, the user end device 5 is not suitable to operate in the mode of "playing news and music at the same time". In other words, the two control programs P corresponding to the two explicit designated intents are not suitable to be executed by the processing unit 11 at the same time. Therefore, in this example, the processing unit 11 will determine that the two explicitly specified intentions of "play news" and "play music" conflict with each other, and treat the two together as a group of conflicting intentions. In addition, the processing unit 11 will, for example, use the explicitly specified intention of "play news" expressed first as a candidate specified intention, and discard the explicitly specified intention of "play music" expressed more slowly.

After the processing unit 11 selects a candidate designated intent from each group of conflicting intents, the conflict elimination process is completed.

In this embodiment, the processing unit 11 performs the screening process by determining whether any of the candidate designated intents has an intent label L corresponding to the exclusive intent label L’. If the determination result is yes, the processing unit 11 will use the candidate designated intent corresponding to the exclusive intent label L’ as the only target designated intent, which is equivalent to discarding other candidate designated intents that do not correspond to the exclusive intent label L’. If the determination result is no (i.e., no candidate designated intent corresponds to the exclusive intent label L’), the processing unit 11 will use all candidate designated intents as target designated intents. For example, assuming that the exclusive intent label L' represents the user intent of "making an outgoing call", and assuming that the voice text data is "help me call the counter and turn on the TV and air conditioner", in this case, the processing unit 11 will identify three explicitly specified intents, namely "making an outgoing call", "turning on the TV" and "turning on the air conditioner" from the voice text data, and, by performing the filtering process on the three explicitly specified intents, the processing unit 11 will only determine the explicitly specified intent representing "making an outgoing call" as the only target specified intent, and discard the other two explicitly specified intents representing "turning on the TV" and "turning on the air conditioner".

It is to be noted that, in another embodiment, the processing unit 11 may only perform the conflict elimination process on the explicitly specified intents in step S6, without performing the screening process. Furthermore, in the embodiment, if there are one or more groups of conflicting intents among the explicitly specified intents, the processing unit 11 directly selects a single explicitly specified intent from each group of conflicting intents as the target specified intent, and if the explicitly specified intents do not conflict with each other, the processing unit 11 directly uses the explicitly specified intents as the N target specified intents.

After the processing unit 11 determines the N target designated intentions from the explicit designated intentions, the process proceeds to step S7.

In step S7, the processing unit 11 executes the control program P corresponding to the intent label L corresponding to each target designation intent. It is worth mentioning that each target designation intent must be a clear designation intent, so if there is still a fuzzy designation intent after step S5 is executed, the processing unit 11 is equivalent to abandoning (ignoring) the fuzzy designation intent, and it does not affect the processing of the clear designation intent.

In addition, in the present embodiment, since the intention labels L in the language processing model M are sequential, when the number of target designation intentions determined by the processing unit 11 is multiple (i.e., N is greater than 1), the processing unit 11 in the present embodiment executes the control programs P corresponding to the intention labels L corresponding to the target designation intentions one by one in sequence according to the order between the intention labels L respectively corresponding to the target designation intentions, but is not limited to this.

For example, assuming that the voice text data is "Play the theme song of Titanic, then help me check tomorrow's weather, and turn on the headlights", the processing unit 11 will, for example, determine three target-specified intents, namely "play music", "check weather data", and "turn on bedroom lights" based on the voice text data, and, based on the order of the three intent labels L corresponding to the three target-specified intents, the processing unit 11 For example, the control program P corresponding to the target designation intent corresponding to "turn on the bedroom light" is first executed to control the lighting of the guest room lamp through the processing module 51 of the user end device 5. Then, the processing unit 11 executes the control program P corresponding to the target designation intent corresponding to "query weather data" so that the processing module 51 of the user end device 5 controls the display of the output module 53 to display weather data for the user's reference. Finally, the processing unit 11 executes the control program P corresponding to the target designation intent corresponding to "play music" so that the processing module 51 of the user end device 5 controls the speaker of the output module 53 to play a music media file.

After the processing unit 11 executes the control program P corresponding to the intent label L corresponding to each target designated intent, the language data processing method of this embodiment ends.

The above is an example explanation of how the language data processing system 1 of this embodiment implements the language data processing method.

It is to be noted that if the number of fuzzy designated intentions determined by the processing unit 11 in step S2 is multiple, the processing unit 11 may, for example, generate and output the query message for the fuzzy designated intention that is first expressed by the voice and text data among the fuzzy designated intentions in step S4, and then determine in step S5 whether to change it to a clear designated intention. However, in different implementations, the processing unit 11 may also generate and output the query message for the fuzzy designated intention whose corresponding intention label L has the highest order according to the order between the intention labels L corresponding to the fuzzy designated intentions in step S4. On the other hand, if the number of fuzzy designated intentions determined by the processing unit 11 in step S2 is multiple, the processing unit 11 may generate and output a query message for only a single fuzzy designated intention as shown in FIG. 2 to determine whether to change it to a clear designated intention. However, in different embodiments, the processing unit 11 may also repeatedly execute steps S4 and S5 multiple times to generate and output a query message for all fuzzy designated intentions, or a part of multiple fuzzy designated intentions, and determine whether to change it to a clear designated intention.

It should be understood that steps S1 to S7 of this embodiment and the flowchart of FIG2 are only used to illustrate one of the practicable ways of the language data processing method of the present invention. It should be understood that even if steps S1 to S7 are merged, split or adjusted in sequence, if the process after merging, splitting or adjusting in sequence achieves substantially the same effect as that of the present embodiment in substantially the same way, it still belongs to the practicable aspects of the language data processing method of the present invention. Therefore, steps S1 to S7 of this embodiment and the flowchart of FIG2 are not used to limit the practicable scope of the present invention.

The present invention also provides an embodiment of a computer program product, which is an application program and can be stored in a computer-readable recording medium and loaded and run by an electronic device (such as a mobile phone, a tablet computer, a laptop computer, and a desktop computer, etc.), and the computer program product includes the language processing model M shown in FIG1. Furthermore, when the computer program product is loaded and run by the electronic device, the computer program product can enable the electronic device to be used as a language data processing system provided by the present invention (such as the language data processing system 1 shown in FIG1), thereby implementing the media file selection method provided by the present invention.

In summary, by implementing the language data processing method, after the language data processing system 1 obtains the voice and text data and identifies the designated intents therefrom, it can classify each designated intent as a clear designated intent or a fuzzy designated intent according to the vocabulary of the voice and text data, and execute the relevant control program P only according to the (etc.) clear designated intents. Therefore, even if some designated intents expressed by the voice and text data are not clear enough (i.e., there are fuzzy designated intents), the language data processing system 1 can still process the designated intents that are expressed clearly enough (i.e., the clear designated intents) to meet the needs of the user. On the other hand, for the fuzzy designated intentions that are not expressed clearly enough, the language data processing system 1 of the present embodiment can generate and output corresponding inquiry messages for the fuzzy designated intentions to guide the user to supplement his or her needs in spoken language, and judge whether the fuzzy designated intentions can be converted into clear designated intentions based on the subsequently obtained voice and text data. In this way, even if the user does not describe certain needs clearly at the beginning, the language data processing system 1 can achieve the effect of clarifying the user's needs by actively outputting inquiry messages. On the other hand, by executing the conflict elimination process, the language data processing system 1 can actively eliminate conflicting designated intentions. In this way, even if the multiple requirements expressed by the user are contradictory, the language data processing system 1 can still process at least one of the requirements, and will not be unable to operate due to the semantic contradictions in the voice and text data. Based on the above, the language data processing system 1 can perform more complete and detailed processing on multiple requirements expressed by the user at one time, thereby providing a significant improvement on the existing technology, and can indeed achieve the purpose of the present invention.

However, the above is only an example of the implementation of the present invention, and it cannot be used to limit the scope of the implementation of the present invention. All simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the content of the patent specification are still within the scope of the patent of the present invention.

1: Language data processing system 11: Processing unit 12: Storage unit M: Language processing model L: Intent label L’: Exclusive intent label P: Control program 5: User end device 51: Processing module 52: Input module 53: Output module S1~S7: Steps

Other features and functions of the present invention will be clearly presented in the implementation method with reference to the drawings, wherein: FIG. 1 is a block diagram, exemplarily showing an implementation of the language data processing system of the present invention, and a user end device suitable for cooperating with the implementation; and FIG. 2 is a flow chart, used to exemplarily illustrate how the implementation and the user end device cooperate to implement a language data processing method.

S1~S7: Steps

Claims (13)

A language data processing system comprises: a processing unit; and a storage unit electrically connected to the processing unit and storing a language processing model implemented by machine learning technology, wherein the language processing model comprises a plurality of intention labels, and each intention label corresponds to a control program related to the operation mode of the processing unit; wherein the processing unit is used to: identify a plurality of designated intentions expressed by a speech data using the language processing model, and determine each designated intention as one of a clear designated intention and a fuzzy designated intention according to the vocabulary in the speech data, wherein each designated intention corresponds to one of the intention labels; and Determine N target specified intentions from the explicit specified intentions, and execute the control program corresponding to the intention label corresponding to each target specified intention, where N is an integer greater than or equal to 1. A language data processing system as described in claim 1, wherein, for each designated intent, the processing unit determines the designated intent as the clear designated intent or the fuzzy designated intent in a manner that includes: the processing unit determines whether there are one or more key words in the speech and text data that can be used by the processing unit to execute the control program related to the designated intent; if the processing unit determines that the speech and text data contains the key words, the processing unit determines the designated intent as the clear designated intent; if the processing unit determines that the key words do not exist in the speech and text data, the processing unit determines the designated intent as the fuzzy designated intent. A language data processing system as described in claim 1, wherein, after the processing unit determines each designated intent as the explicit designated intent or the fuzzy designated intent, the processing unit is further used to: When one or more designated intents of the designated intents are determined by the processing unit as the fuzzy designated intent, for at least one of the fuzzy designated intents, use the language processing model to generate a query message corresponding to the fuzzy designated intent, and output the query message by an output module; and After the query message is output by the output module, when another voice-text data corresponding to another voice input is obtained, it is determined whether the semantics of the other voice-text data matches the fuzzy specified intent, so that the processing unit can execute the control program related to the fuzzy specified intent, and, if it is determined that the semantics of the other voice-text data matches the fuzzy specified intent, the fuzzy specified intent is changed to a clear specified intent. A language data processing system as described in claim 1, wherein the processing unit determines the N target specified intentions from the explicit specified intentions in a manner that includes: the processing unit determines whether there are multiple explicit specified intentions among the explicit specified intentions that conflict with each other and are collectively regarded as a group of conflicting intentions, and, when it is determined that the group of conflicting intentions exists among the explicit specified intentions, the processing unit only uses a single explicit specified intention of the group of conflicting intentions as one of the target specified intentions of the N target specified intentions. A language data processing system as described in claim 1, wherein one of the intent labels is set as an exclusive intent label, and the processing unit determines the N target specified intentions from the explicitly specified intentions in a manner comprising: the processing unit determines whether the intent label corresponding to one of the explicitly specified intentions is the exclusive intent label, and if the determination result is yes, the processing unit uses the explicitly specified intention corresponding to the exclusive intent label as the only target specified intention. A language data processing system as described in claim 1, wherein there is an order between the intent labels, and when the number of target designated intentions determined by the processing unit is multiple, the processing unit executes the control programs corresponding to the intent labels corresponding to the target designated intentions one by one in sequence according to the order between the intent labels respectively corresponding to the target designated intentions. A language data processing method is implemented by a language data processing system, the language data processing system includes a processing unit and a storage unit electrically connected to the processing unit, the storage unit stores a language processing model implemented using machine learning technology, the language processing model includes a plurality of intention labels, and each intention label corresponds to a control program related to the operation mode of the processing unit; the language data processing method includes: (A) The processing unit uses the language processing model to identify multiple designated intents expressed by a speech data, and determines each designated intent as one of a clear designated intent and a fuzzy designated intent according to the vocabulary in the speech data, wherein each designated intent corresponds to one of the intent labels; and (B) The processing unit determines N target designated intents from the clear designated intents, and executes the control program corresponding to the intent label corresponding to each target designated intent, wherein N is an integer greater than or equal to 1. A language data processing method as described in claim 7, wherein, in step (A), for each designated intent, the processing unit determines the designated intent as the clear designated intent or the fuzzy designated intent in a manner that includes: the processing unit determines whether there are one or more key words in the speech and text data that can be used by the processing unit to execute the control program related to the designated intent; if the processing unit determines that the speech and text data contains the key words, the processing unit determines the designated intent as the clear designated intent; if the processing unit determines that the speech and text data does not contain the key words, the processing unit determines the designated intent as the fuzzy designated intent. The language data processing method as described in claim 7 further includes between steps (A) and (B): (C) when one or more of the specified intents are determined by the processing unit to be the fuzzy specified intent, for at least one of the fuzzy specified intents, the processing unit generates a query message corresponding to the fuzzy specified intent using the language processing model, and causes the query message to be output by an output module; and (D) After the processing unit causes the query message to be output by the output module, when the processing unit obtains another voice-text data corresponding to another voice input, the processing unit determines whether the semantics of the other voice-text data matches the fuzzy designated intent, so that the processing unit can execute the control program related to the fuzzy designated intent, and when the processing unit determines that the semantics of the other voice-text data matches the fuzzy designated intent, the processing unit changes the fuzzy designated intent into a clear designated intent. A language data processing method as described in claim 7, wherein in step (B), the processing unit determines the N target specified intentions from the explicit specified intentions in a manner that includes: the processing unit determines whether there are multiple explicit specified intentions among the explicit specified intentions that conflict with each other and are collectively regarded as a group of conflicting intentions, and, when it is determined that the group of conflicting intentions exists among the explicit specified intentions, the processing unit only uses a single explicit specified intention of the group of conflicting intentions as one of the target specified intentions of the N target specified intentions. As described in claim 7, the language data processing method, one of the intent labels is set as an exclusive intent label; wherein, in step (B), the processing unit determines the N target specified intentions from the explicitly specified intentions in a manner that includes: the processing unit determines whether the intent label corresponding to one of the explicitly specified intentions is the exclusive intent label, and if the determination result is yes, the processing unit uses the explicitly specified intent corresponding to the exclusive intent label as the only target specified intention. As described in claim 7, the language data processing method has an order between the intent labels; wherein, in step (B), when the number of target designated intents determined by the processing unit is multiple, the processing unit executes the control programs corresponding to the intent labels corresponding to the target designated intents one by one in sequence according to the order between the intent labels respectively corresponding to the target designated intents. A computer program product includes a language processing model implemented using machine learning technology, wherein the language processing model includes multiple intention labels, and each intention label corresponds to a control program related to the operation mode of an electronic device. The computer program product is used to be loaded and run by the electronic device so that the electronic device implements the language data processing method described in any one of claims 7 to 12.

Images