Character conversion assistance system and character conversion assistance method for character conversion assistance system

Abstract

To acquire and output a conversion character string candidate for input character information from a server device.SOLUTION: The character conversion support system includes client devices CL1 to CLN each having a front end processor that performs character-string processing, and a server device 1 that acquires character strings from the front end processors, wherein the client devices CL1 to CLN are configured to transfer character information input to the front end processors to the server device 1, acquire character-string candidates that are sequentially converted by an inference engine that uses a large-scale language model, and determine the character-string candidates as character strings to be outputted by an outputting unit.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a character conversion support system in which a client device capable of connecting to a specific domain via a specified communication medium, a client application running on the client device, and an application server running on a server device cooperate with each other, and a character conversion support method for the character conversion support system. [Background technology]

[0002] Traditionally, character information processing in computer systems has been controlled by an OS, and a front-end processor (FEP) has been built in to convert input key codes into the intended character information while referring to a dictionary library. In particular, the FEP (front-end processor) in character processing plays an important role in Japanese input systems.

[0003] The FEP is responsible for the process of receiving input from the keyboard and converting it into Japanese characters such as hiragana and kanji. This process includes converting the input romaji into hiragana and then presenting candidate kanji and words for the user to select. In addition to the above, FEP features include the ability to suggest and select candidate words, and support for multiple input methods (e.g., romaji input, kana input).

[0004] On the other hand, FEP has the problem of converting characters unintentionally, especially in the case of technical terms, new words, and foreign words.

[0005] Patent Document 1 listed below discloses that "In order to enable users to input and display a variety of characters even when they have limited computer resources at hand, the web browser of client 30 accesses the web server of dictionary center 20 to receive a character input service. The web browser inputs the pronunciation of the characters, and the web server of the dictionary center converts this into a character code and then converts it into a URL for an image file on the web server of font center 10. As a result, an HTML document containing the URL of the image file of the characters as a source is sent to the browser, and the browser retrieves the image file from the web server of font center 10 based on the URL and displays the characters using the retrieved image."

[0006] Furthermore, the following Patent Document 2 discloses that "in order to convert an audio file into text and identify the speaker using a learning server via the Internet, the speech recognition system 1 is composed of a sound collection unit 2 that inputs the audio, a processing unit 3 that generates an audio file F from the audio, transmits this audio file F to a character conversion server and a speaker identification server, and receives a text file W and the identification result of the speaker H, and a monitor unit 4 that displays the processing result of the processing unit 3. The audio data V separated by silence between utterances by speaker H is transmitted as the audio file F to a character conversion server 5 and a speaker identification server 6, which are cloud services that perform data analysis by self-learning based on data collected from a large number of users via the Internet. The received text file W and the identification result of speaker H are then associated with the audio file F and displayed in chronological order on the monitor unit 4." [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2001-350747 [Patent Document 2] Japanese Patent Publication No. 2020-060735 Summary of the Invention [Problem to be solved by the invention]

[0008] In character conversion processing systems that process Japanese, when converting input hiragana into kanji, it is difficult to obtain a perfect sentence from the input string of characters, voice, and gestures alone, and it is widely known that appropriate expressions cannot be generated without taking into account a wide range of context and the situation of the conversation.

[0009] In particular, in a language like Japanese, where there are many homophones based on ideographic expressions, the input sentence "The warrior fixed the warrior" can be interpreted as either "The bow fixed the bow" or "The player fixed the player" depending on the context. In addition, in actual use, there are often typos and other input errors mixed in with the input in the first place.

[0010] For example, "The captain corrected the captain" may be converted to "The bow corrected the bow," but if there is a typo in "The captain corrected the captain," it may be converted to "Last week corrected the captain." In this way, even a slight difference in input can greatly change the interpretation (the converted string).

[0011] In particular, when various users connect to a specific domain from a client device and input text information on the provided UI screen, there are issues that the text may not be converted into the string of characters that the user intended, or when a long sentence is input, the character conversion process may take a considerable amount of time.

[0012] The present invention solves the above-mentioned problems and provides a character conversion support system and a character conversion support method for a character conversion support system that can obtain and output highly accurate conversion string candidates for input character information even when the same user operating the client device operates in various character processing environments, even if the attributes of the input device, the attributes of the application being launched, the attributes of the service being launched, or the input method and input style of character information change, by employing conversion candidate character strings obtained from an inference engine that infers conversion character strings provided in a server device connected to a network for the functional processing of the front-end processor of the client device. [Means for solving the problem]

[0013] The character conversion support system of the present invention for achieving the above object comprises the following configuration.

[0014] The character conversion support system of the present invention is a character conversion support system in which a client device having a front-end processor that performs character string processing and a server device that acquires character strings from the front-end processor communicate via a predetermined communication medium to coordinate character string conversion processing, and is characterized in that the client device is configured to transfer character strings generated by the front-end processor to the server device, and sequentially acquire and confirm conversion character string candidates from the server device to be sequentially converted by one of a plurality of inference engines that use a large-scale language model. [Effects of the Invention]

[0015] According to the present invention, by employing a conversion candidate character string obtained from an inference engine that infers conversion character strings provided in a server device connected to a network for the functional processing of the front-end processor of the client device, it is possible for the same user operating the client device to obtain and output highly accurate conversion character string candidates for the input character information even when the attributes of the input device, the attributes of the application being launched, the attributes of the service being launched, and the input method and input style of the character information change in various character processing environments. [Brief explanation of the drawings]

[0016] The drawings illustrate particular embodiments of the present invention, including essential features of the invention as well as alternative and preferred embodiments. [Figure 1] 1 is a block diagram illustrating the configuration of a character conversion support system according to an embodiment of the present invention; [Figure 2] FIG. 2 is a block diagram illustrating software functions that are based on the hardware resources of the character conversion support system shown in FIG. 1. [Figure 3] FIG. 2 is a block diagram illustrating the hardware configuration of the server device shown in FIG. [Figure 4] FIG. 2 is a block diagram illustrating the hardware resources and software functions of the client device shown in FIG. 1. [Figure 5] 3 is a flowchart showing an example of a character conversion support method of the character conversion support system according to the present embodiment. [Figure 6] 1 is a flowchart showing an example of a character conversion support method of the character conversion support system according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0017] Next, the best mode for carrying out the present invention will be described with reference to the drawings.

[0018] <System configuration description> [First embodiment] 1 is a block diagram illustrating the configuration of a character conversion support system according to this embodiment. This example illustrates a character conversion support system in which client devices CL1 to CLN that can connect to a specific domain via a predetermined communication medium, a client application 100 running on the client devices CL1 to CLN, and an application server 11 running on a server device 1 cooperate to support character conversion processing.

[0019] In FIG. 1, reference numeral 1 denotes a server device that communicates with client devices CL1 to CLN via a network 21 using a predetermined protocol.

[0020] In the server device 1, reference numeral 11 denotes an application server that communicates with the client applications of the client devices CL1 to CLN via a network 21. Reference numeral 12 denotes a domain knowledge server selection unit (DKS) that receives user information, service information (uniquely assigned service number (Universal Unique IDentifier)), application information, document properties, etc. from a front-end processor (FEP) 102 that performs character conversion processing on information input by an input unit 103 of the client devices CL1 to CLN based on a predetermined language. Then, the DKS 12 executes processing to select one candidate from the knowledge server group 13 that will perform the second inference processing in the character conversion processing.

[0021] A domain name server (DNS) 50 provides an IP address corresponding to a domain name input from the input unit 103 of the client devices CL1 to CLN.

[0022] Client devices CL1 to CLN connect to specific URL1 to URLN via the network 22, which are routed according to the IP addresses (xxx.xxx.xxx) specified by the DNS 50. The specific URL1 to URLN cover a wide range of genres, and when engaging in conversations or chats appropriate to each URL, there is a probability that specific terms will be selected, and it is known that there are certain trends in the character strings used in mutual conversations.

[0023] In this embodiment, the FEP 102 requests the server device 1 in the background to perform a second inference process for inferring character information in accordance with the domain name input from the input unit 103 by the user of the client device CL1 to CLN.

[0024] FIG. 2 is a block diagram explaining the software functions that are based on the hardware resources of the character conversion support system shown in FIG. 1, and the same components as those in FIG. 1 are given the same reference numerals.

[0025] In FIG. 2, the plurality of second inference units 13-1 to 13-N included in the server device 1 include a plurality of knowledge servers 13-1-1 to 13-1-N and profiles 13-2-1 to 13-2-N.

[0026] In the client devices CL1 to CLN, 101 is a switching unit that performs a process of switching between performing the first inference process locally during character conversion or performing the first inference process on the server device 1 side during character conversion based on information (including a domain name) input by the input unit 103.

[0027] The FEP 102 includes a first inference unit 102-1 and a storage unit 102-2 that stores user profile information (including a user dictionary). The storage unit 102-2 is configured to store user information to be transferred to the server device 1. In this embodiment, the user information includes, as the character processing mode used by the user, the properties of the document used by the user, the attributes of the service used by the user, and the attributes of the application used by the user. The user information stored in the storage unit 102-2 is stored so as to be updatable in accordance with the usage environment of the character string entered by the user.

[0028] Fig. 3 is a block diagram for explaining the hardware configuration of the server device 1 shown in Fig. 1. Note that the same components as those in Fig. 1 or 2 are denoted by the same reference numerals.

[0029] In FIG. 3, reference numeral 31 denotes a communication unit that executes functional processing for communicating with client devices CL1 to CLN connected to the network 21 using a predetermined protocol.

[0030] A CPU 32 loads the application server 11 (various application programs including an operating system) stored in an external memory 33 into a RAM 36 and executes it.

[0031] Reference numeral 34 denotes an inference engine unit in which second inference units 13-1 to 13-N are deployed. The second inference units 13-1 to 13-N constantly collect and update a large amount of natural language sentences and usage (corpus) in the social domain in which the user is involved, and maintain the matching degree of character conversion candidates so that they evolve.

[0032] Reference numeral 36-1 denotes a DKS unit, which executes a functional process of delivering a conversion string candidate inferred by one of the knowledge servers 13-1-1 to 13-1-N selected by the CPU 32 from the inference engine unit 34 to the FEP 102 of one of the client devices CL1 to CLN via the network 21. Note that this functional process continues at all times while the user operating the client device CL1 to CLN is inputting a string via the input unit 310.

[0033] 39 is the AI ​​support unit, which performs processing to support innovative character conversion processing in the field of natural language processing (NLP: Neural Language Processing), which is built using a large-scale language model (LLM) built using massive amounts of text data and advanced deep learning technology. The large-scale language model uses T5 (Text-to-Text Transfer Transformer). Numerous derivative models based on the architecture have been proposed. T5 is a descendant of the Transformer released in 2017, and is trained to input hiragana text and output the text converted into kanji.

[0034] The inference engine of the knowledge server group 13 included in the server device 1 cooperates with the AI ​​support unit 39 to execute a process for improving the accuracy of conversion character string candidates for kana characters sequentially acquired from the client devices CL1 to CLN.

[0035] Furthermore, the functional processing of the AI ​​support unit 39 may be incorporated into the inference engine.

[0036] LLM training uses "supervised learning," in which hiragana sentences are used as input and a collection of corresponding kanji sentences (corpus) is used as training material. This means that the hiragana sentences are used as input, the output of T5 is calculated, and the loss is calculated by comparing it with the training kanji sentences. Based on this loss, the parameters of the model are updated according to their gradients.

[0037] The inference engine has a corpus for each language community. For example, for formal situations, it uses texts from newspapers, magazines, and books, while for casual situations, it uses conversational texts and texts from social media.

[0038] The corpus is constantly updated. In other words, the inference engine does not learn from the user's history, but from the entire society to which the user belongs. To do this, the inference engine constantly collects a wide range of information, automatically obtaining corresponding hiragana sentences from mixed kana and kanji sentences through morphological analysis, and then combining these with the original sentences to continuously create "kana and kanji sentence pairs." Learning is deepened by collecting, adding, and updating a large number of these pairs.

[0039] Figure 4 is a block diagram explaining the hardware resources and software functions of the client devices CL1 to CLN shown in Figure 1. Note that the same reference numerals are used for the same components as in Figure 1. Note that the client devices CL1 to CLN are assumed to be various data terminals, including PAD devices, personal computers (PCs), and smartphones.

[0040] Also, in this embodiment, when the kana character conversion mode is selected in the FEP 102, the CPU 301 transfers the key code input from the input unit 310 via the communication unit 304 to the DKS 12 of the server device 1 connected to the network 21, along with key code information and user information (details will be described later).

[0041] DKS12 selects a second inference unit 13-1 to 13-N to determine a conversion candidate character string according to the key code information and user information handed over from one of client devices CL1 to CLN, and the response unit 36-2 responds to FEP102 of the requesting client device CL1 to CLN via the communication unit 31 with the Japanese character string data mixed with kanji characters converted by one of the second inference units.

[0042] At this time, the FEP 102 is controlled so that the browser outputs the character string candidate converted on the server device 1 side to the character input line on the display screen of the display 311 as if it had been converted by itself.

[0043] At this time, the server device 1 freely selects one of the second inference units 13-1 to 13-N according to the TPO of the user operating the client devices CL1 to CLN (the "T" in "TPO" stands for "Time," "P" for "Place," and "O" for "Occasion"), and even during conversion, the server device 1 flexibly switches to a different second inference unit to perform character string conversion processing.

[0044] Furthermore, DKS12 is configured to recognize, for example, whether the Japanese entered by the user operating the client devices CL1 to CLN is polite language, or special words used on SNS (character strings not used in normal formal documents), such as "buzz", and to dynamically switch to character string conversion processing by another second inference unit even during conversion.

[0045] In FIG. 4, 301 denotes a CPU, which activates a BIOS stored in a ROM 302 to control the I / O of input / output devices, and also executes various applications developed in a RAM 303 to perform various data processing.

[0046] A communication unit 304 communicates with the server device 1 via the network 21 using a predetermined protocol, and performs various data processing in a network environment.

[0047] 311 is a display, on which the kana character information input by the input unit 310 is handed over to the server device 1, and converted character string candidates generated by the second inference units 13-1 to 13-N of the server device 1 are continuously received and displayed so that the character information is arranged at the position specified by the cursor.

[0048] While the user is operating the input unit 310, the user continues to receive conversion candidate character strings from the server device 1, but if the user performs an action such as correcting the displayed conversion character string, the server device 1 selects one of the knowledge servers 13-1-1 to 13-1-N to perform the most appropriate second conversion process.

[0049] [Data processing on the server device side] Fig. 5 is a flowchart showing an example of a character conversion support method of the character conversion support system according to this embodiment. Note that (1) to (9) are data processes on the server device 1 side shown in Fig. 1, and are realized by the CPU 32 expanding a control program stored in the external memory 33 onto the RAM 36 and processing it.

[0050] First, the CPU 32 acquires user information (including domain information) from the FEP 102 of any one of the client devices CL1 to CLN shown in FIG. 1 and kana character information input from the input unit 310 (1).

[0051] The CPU 32 delivers the kana character information and user information acquired from the client devices CL1 to CLN shown in FIG. 1 to the DKS 12 (2), and the DKS 12 selects one of the knowledge servers 13-1-1 to 13-1-N based on the acquired domain information (3).

[0052] Next, the CPU 32 delivers the character string candidate converted by the selected knowledge server via the network 21 to the FEP 102 of one of the client devices CL1 to CLN (4).

[0053] Next, CPU 32 determines whether it has received a correction instruction or a reconversion instruction for the conversion character string candidate delivered from FEP 102 of any of client devices CL1 to CLN (5). If CPU 32 determines that it has received a correction instruction or a reconversion instruction, CPU 32 further determines whether it will continue to select one of the knowledge servers selected in step (3) (6). If CPU 32 determines that it will not select the same knowledge server, CPU 32 further controls DKS 12 to select one of knowledge servers 13-1-1 to 13-1-N (7). Then, CPU 32 returns the process to step (4).

[0054] On the other hand, in step (6), if the CPU 32 determines that it will continue to select one of the same knowledge servers 13-1-1 to 13-1-N selected in step (3), the process returns to step (4).

[0055] Meanwhile, in step (5), after a certain time has elapsed since the CPU 32 received a correction instruction or a reconversion instruction for the conversion string candidate handed over from one of the FEPs 102 of the client devices CL1 to CLN (8), the CPU 32 determines whether the state in which it has not acquired user information (including domain information) from one of the FEPs 102 of the client devices CL1 to CLN shown in FIG. 1 and text information input from the input unit 310 continues, i.e., whether it has accepted kana character information from one of the FEPs 102 of the client devices CL1 to CLN (9).

[0056] If the CPU 32 determines that it has received kana character information from the FEP 102 of any of the client devices CL1 to CLN, it returns to step (2) and repeats the same process. On the other hand, if the CPU 32 determines in step (9) that it has not received kana character information from the FEP 102 of any of the client devices CL1 to CLN, it ends this process.

[0057] [Data processing on the client device CL1 to CLN side] Fig. 6 is a flowchart showing an example of a character conversion support method of the character conversion support system according to this embodiment. Note that (11) to (19) are data processes on the client devices CL1 to CLN side shown in Fig. 1, and are realized by the CPU 301 loading a control program stored in an external memory (not shown) onto the RAM 303 and processing it.

[0058] First, when a user operating one of the client devices CL1 to CLN instructs access to one of the domains, the CPU 301 starts a browser from an external memory (not shown) (11). At that time, the user operates the keyboard of the input unit 310 to input a kana character string, for example, using a romaji conversion method (12), and the FEP 102 outputs the input kana character string to the server device 1 (13).

[0059] At this time, the FEP 102 detects the user information (document properties used by the user, services used by the user, and applications used by the user) stored in the storage unit 102-2 (14) and transfers the detected user information to the server device 1 (15).

[0060] Next, when the CPU 301 acquires a candidate conversion string via the network 21 from the knowledge server selected by the server device 1 (one of the second inference units 13-1 to 13-N) (16), it displays the conversion string on the string input line (where the cursor is pointed) specified by the browser on the display 311 (17).

[0061] Next, CPU 301 determines (18) whether each user has issued an instruction to edit the converted character string displayed on the browser on display 311, such as to correct, amend, or delete the converted character string, and if it determines that an instruction to edit has been issued, returns to step (13).

[0062] On the other hand, if the CPU 301 determines in step (18) that an editing instruction has not been issued, it further determines whether kana characters have been input from the input unit 310 (19). If the CPU 301 determines that a character code has not been input, it terminates the process of obtaining character string candidates by the server device 1, and if it determines that a character code has been input, it returns to step (13) and repeats the same process.

[0063] [Effects of the first embodiment] According to this embodiment, the functional processing of the front-end processor 102 of the client devices CL1 to CLN employs conversion candidate character strings obtained from a plurality of inference engines (inference engines that use LLM) that infer conversion character strings provided in the server device 1 connected to the network 21. This allows the same user operating the client devices CL1 to CLN to obtain and output highly accurate conversion character string candidates for the input character information in various character processing environments, even if the attributes of the input device, the attributes of the application to be launched, the attributes of the service to be launched, or the input method and input style of the character information change.

[0064] It should be noted that the system operator can freely choose whether to offer the free service that presents character string conversion candidates provided by the server device 1 to general users, or to offer it as a paid service with customization functions added for specific users.

[0065] Furthermore, in the above system, Japanese is used as an example of a language model, but the language model may also be any major foreign language model. Furthermore, in an environment where users of different languages ​​exchange text information with each other, such as through chat, it is possible to freely create an environment in which similar character string conversion processing can be performed by using a translation model support system.

[0066] Second Embodiment In the above embodiment, an example has been described in which a short sentence is input from the client devices CL1 to CLN, but the present invention can also be applied to a long context.

[0067] In the above system, when the server device 1 interprets a long context, it is expected that the response speed until the conversion becomes visible will be delayed in exchange for conversion accuracy.

[0068] Therefore, every time a user operating a client device CL1 to CLN reads one character (one token) input from the input unit 310, the FEP 102 immediately sends the input character code and a string conversion request to the second inference units 13-1 to 13-N of the server device 1 via the network 21.

[0069] In this way, when the server device 1 receives a character code and a string conversion request, the second inference units 13-1 to 13-N respond to the client devices CL1 to CLN with the character code indicating the character string conversion process and the generated string.

[0070] At this time, the server device 1 controls the character string conversion process so that it starts after rewinding to the beginning of the paragraph (translation unit) of the document each time.

[0071] For this reason, the beginning of a paragraph will be repeatedly converted into a string (rewinded), but since the input characters become longer each time the conversion is performed, it is expected that the conversion accuracy will improve.

[0072] As a result, the client devices CL1 to CLN automatically rewind and display converted character strings so that words that were initially converted incorrectly are properly corrected as input proceeds.

[0073] According to this type of conversion process, if the character input process on the client device CL1 to CLN side is fast, the burden of the character string inference and conversion process on the server device 1 side increases. Therefore, the server device 1 controls the memory unit to temporarily hold subsequent requests for each user until the ongoing character string conversion process is completed.

[0074] When the next process is started, the requests to be held are not executed one by one, but only the last request is executed and the requests that were previously held are deleted.

[0075] [Effects of the second embodiment] As a result, although it appears to the client devices CL1 to CLN that the server device 1 is converting characters one by one in parallel, the server device 1 only needs to have one thread per user.

[0076] In addition, since the client devices CL1 to CLN acquire the results of the character string conversion process for long sentences from the server device 1 in a discontinuous manner, the display on the client devices CL1 to CLN is controlled so that the newly added hiragana character string from the character string last sent to the server device 1 is concatenated to the end of the last converted sentence.

[0077] In this way, even if there is conversion latency in the server device 1 during the inference process of a long chunk, the CPU 301 controls the display on the screen of the display 311 on the client device CL1 to CLN side so that the hiragana input is automatically converted into an appropriate sentence without delay.

[0078] Although rewind processing is generally performed in paragraph units, it is adjusted based on the addition state on the server device 1 side, from a minimum of a sentence unit to a maximum of the entire document being processed.

[0079] Third Embodiment In the above embodiment, an example was described in which a short sentence is input from the client devices CL1 to CLN, but even if the conversion process has already been completed for a converted character string or the like on the display 311 of the client devices CL1 to CLN, the re-conversion process may be controlled later.

[0080] As mentioned above, even if various character string conversion processes are performed, it is expected that there will be cases where misinterpretations will occur due to context.

[0081] In this case, the system may incorporate a configuration that controls the manner of the string conversion process so that each user operating the client devices CL1 to CLN can utilize the intermediate results of the inference, thereby making maximum use of the intermediate results of the inference and reducing the burden of the string input process on the user.

[0082] Specifically, in the server device 1, the second inference units 13-1 to 13-N reserve a conversion buffer in a memory resource not shown, and buffer all plain text before the current paragraph conversion (raw hiragana characters entered by each user and flick trajectories) in this buffer, so that it is possible to rewind (so-called undo) even after the conversion has been confirmed.

[0083] Therefore, in the server device 1, the second inference units 13-1 to 13-N hold the reliability (0.0 to 1.0 (probability notation)) of the inference result for each phrase, and instruct the FEP102 of the client devices CL1 to CLN to highlight part of the generated document according to this reliability.

[0084] [Effects of the third embodiment] This allows each user of client devices CL1 to CLN to quickly determine the parts to be corrected on the screen of display 311, and then return to that screen to selectively restore the sentences to their pre-conversion state, and then proceed to the appropriate correction work.

[0085] Therefore, unlike simple kana-kanji conversion, this embodiment allows users to rewind and perform editing work even after the character conversion process has been finalized, thereby reducing the psychological burden on each user when rewinding.

[0086] Note that the above-mentioned reliability-based highlighting function is not essential, and when each user operates the cursor on the screen, control may be executed to rewind only the relevant phrase and selectively display only the pre-conversion character string.

[0087] Furthermore, in the inference process of character string conversion by the second inference units 13-1 to 13-N in the server device 1, there may be cases where there is a simple input error or the original sentence or conversation expression itself is insufficient.

[0088] Therefore, even if only the particles in the incorrect part are corrected, by controlling the system so that re-evaluation can be carried out on a paragraph-by-paragraph basis, the improved accuracy of the estimation from one correction will spread to other parts, making it possible to automatically proofread the surrounding text as well.

[0089] [Fourth embodiment] In the above embodiment, an example was described in which short sentences are input from client devices CL1 to CLN. However, when each user types in input from the input unit 310, a lightweight first inference unit 102-1 may be provided on the client devices CL1 to CLN to cleanse the user input, such as removing typos and noise, in the early stages of the input, thereby enabling a more flexible response.

[0090] In this way, without relying solely on the inference processing of string conversion by the server device 1, the first inference unit 102-1 on the client devices CL1 to CLN cooperates with the inference processing on the server device 1 side, thereby eliminating erroneous key inputs specific to each user and restricting the transfer of user-specific strings that would hinder inference by the second inference units 13-1 to 13-N to the server device 1, thereby preventing the inference processing in the second inference units 13-1 to 13-N from being contaminated, and making it possible to obtain highly accurate string conversion processing results from the server device 1.

[0091] [Effects of the fourth embodiment] In this case, if the input unit 310 is assumed to be a keyboard, the local first inference unit 102-1 on the client device CL1 to CLN side does not target the character string converted into roman characters, but directly obtains information about the keystrokes themselves and applies a predetermined inference process specific to each user, thereby making it possible to extract a more accurate character string and reflect it in the inference process on the server device 1 side.

[0092] In this way, when the character input mode can be selectively switched between direct input via keyboard and flick input, by selecting the second inference units 13-1 to 13-N according to each input mode, it is possible to expect improvement in the accuracy of character string conversion for each input mode.

[0093] [Effects of the fourth embodiment] In the above embodiment, an example was described in which short sentences are input from client devices CL1 to CLN. However, when each user types in input from the input unit 310, a lightweight first inference unit 102-1 may be provided on the client devices CL1 to CLN to cleanse the user input, such as removing typos and noise, in the early stages of the input, thereby enabling a more flexible response.

[0094] It should be noted that the first inference unit 102-1 is secured for each user on the client device CL1 to CLN side and is provided with a dedicated personal dictionary memory.

[0095] Fifth Embodiment In the above embodiment, an example was described assuming that short sentences are input from the keyboard of the input unit 310 from the client devices CL1 to CLN, but there are individual differences in keystrokes, particularly in the fingertip movements and typing speed, and by measuring and learning the input fluctuations and frequency of typos that are specific to romaji input, for example, and executing control to reflect this in character string conversion candidates, it is possible to expect improved conversion accuracy.

[0096] In particular, since the characteristics of the key input type differ between a user who inputs "xi" and a user who inputs "li" when inputting "i", the system controls so that both "sailor" and "sailor n" are output as the conversion string candidate "sailor".

[0097] [Effects of the fifth embodiment] According to this embodiment, the second inference units 13-1 to 13-N on the server device 1 side that use LLM can correct in advance input character strings that may cause significant erroneous generation.

[0098] Furthermore, in the client devices CL1 to CLN that employ the input unit 310 for flick-inputting character strings, it is also possible to correct errors inherent in mis-touching kana characters to be placed.

[0099] Sixth Embodiment In this embodiment, when a keyboard is used in the client devices CL1 to CLN, the correlation between the time interval between keystrokes and the division of phrases (tokens) may be utilized to control the client devices to execute the following processing when the length of the interval between keystrokes exceeds a certain threshold.

[0100] Specifically, in the server device 1, when each user of the client devices CL1 to CLN uses a keyboard, if the length of the interval between keystrokes exceeds a certain threshold, control is executed to insert a hidden deminator.

[0101] [Effects of the Sixth Embodiment] According to this embodiment, user input is not echoed, but by inserting the above-mentioned delimiters when the string is internally broken down into phrases, it is possible to prevent a chain of domino-like misinterpretations that arise from mistakes in dividing phrases across conjunctions such as "wa" and "e," which are likely to occur when converting long sentences all at once. It should be noted that the same phenomenon is expected to occur when the above character input is replaced by voice input, and similar effects can be expected by taking similar measures.

[0102] Seventh Embodiment In this embodiment, when there are multiple application services used by the same user in client devices CL1 to CLN, it is expected that input tendencies will change even for the same user, so the system may be configured to perform control to switch the selection order so that the higher-ranking second inference unit among the second inference units 13-1 to 13-N that use LLM on the server device 1 side is used.

[0103] Specifically, for example, when the same user performs character input for business purposes, the selection order is controlled to switch to use the higher-ranking second inference unit among the second inference units 13-1 to 13-N to select the target document name and the character string in the process of converting the paragraph from the beginning.

[0104] [Effects of the Seventh Embodiment] According to this embodiment, from the perspective of each user operating the client devices CL1 to CLN, it is expected that natural sentences will be generated according to the type and status of the application being executed, while creating documents using the same interface from the same word processing software or spreadsheet software. In addition, when RAG, which combines external information search, is adopted, there is no need to go through AI chat once during the document creation process, improving convenience. Here, RAG is an abbreviation for Retrieval-Augmented Generation, which means retrieval-augmented generation.

[0105] On the other hand, for conversation and SNS applications, it becomes possible to perform inference processing of long conversion strings across conversations, and at the same time, by taking a life log, it becomes possible to improve the accuracy of proper noun conversion.

[0106] Furthermore, by having the client application 100 provide the sentences of the conversation partner, it becomes possible to include the sentences of the previous conversation partner or group in the inference of the character string conversion. This is expected to efficiently suppress the misinterpretation that frequently occurs in conversation and SNS applications, especially when each sentence is short.

[0107] In addition, in the character conversion support system, the functions of the second inference units 13-1 to 13-N in the server device 1 and the first inference unit 102-1 in the client devices CL1 to CLN are clearly separated, and the server device 1 does not need to acquire user-specific user information (information derived from typing habits) and manage history, and is configured to prevent information linked to personal attributes (personal profile) from leaking out of the server device 1.

[0108] Furthermore, if it is necessary to refer to the behavior history of each user of the client devices CL1 to CLN for business purposes, it is also possible to incorporate a strict account management function into the system independently of this. Note that information linked to personal attributes is limited to low-level statistical information such as typing habits, and it is possible to control this information so that it is only stored on the personal client devices CL1 to CLN side and the server device 1 does not need to know it.

[0109] Furthermore, in this embodiment, the output unit that outputs the candidate strings to be converted by the server device 1 is not limited to a display unit, but may be a device (speaker, earphones with a microphone, goggle device, etc.) that outputs the candidate strings to be converted as audio information in each language.

[0110] The present invention can also be realized by a process in which a program that realizes one or more functions of the above-described embodiments is supplied to a system or device via a network or a storage medium, and one or more processors in a computer of the system or device read and execute the program (a program conforming to the flowcharts shown in FIGS. 5 and 6 above).The present invention can also be realized by a circuit (for example, an ASIC) that realizes one or more functions.

[0111] The disclosure of the present invention described above can be summarized at least as follows.

[0112] (1) A character conversion support system in which a client device having a front-end processor that performs character string processing and a server device that acquires character strings from the front-end processor communicate via a predetermined communication medium to coordinate character string conversion processing, characterized in that the client device is configured to transfer character strings generated by the front-end processor to the server device and sequentially acquire and confirm conversion character string candidates from the server device that are sequentially converted by one of a plurality of inference engines that utilize a large-scale language model.

[0113] (2) A character conversion support system in which a client device having a front-end processor that performs character string processing and a server device that acquires character strings from the front-end processor communicate with each other via a predetermined communication medium to coordinate character string conversion processing, wherein the server device comprises: an acquisition means for acquiring character strings from the front-end processor; a plurality of inference engines that infer candidate conversion character strings for the character strings acquired by the acquisition means using a large-scale language model; and a response means for responding to the front-end processor with candidate conversion character strings generated by any of the inference engines.

[0114] (3) A character conversion support system in which a client device having a front-end processor that performs character string processing and a server device that acquires character strings from the front-end processor communicate via a specified communication medium to coordinate character string conversion processing, wherein the client device is equipped with a transfer means that sequentially transfers character strings and user information entered by a user to the server device via the front-end processor, and the front-end processor sequentially determines conversion character string candidates responded to by the server device and outputs them to an output device.

[0115] (4) A character conversion support system in which a client device having a front-end processor that converts a character string entered by a user into a predetermined character information string communicates with a server device that acquires the character string entered by the user via the front-end processor via a predetermined communication medium to coordinate character string conversion processing, wherein the server device comprises: an acquisition means that continuously acquires user information that identifies the character processing mode used by the user and character strings from the front-end processor; a plurality of inference engines that infer conversion character string candidates for the character string acquired by the acquisition means using a large-scale language model; and a response means that responds to the front-end processor with conversion character string candidates generated by any of the inference engines identified by the user information; the client device comprises transfer means that sequentially transfers the character string entered by the user and the user information to the server device via the front-end processor; and the front-end processor sequentially determines the conversion character string candidates responded to from the server device and outputs them to an output device.

[0116] (5) The character string input by the user is characterized in that it is input by keyboard or flick input.

[0117] (6) The server device is characterized by having a configuration for selecting one of a plurality of inference engines based on the character processing mode included in the user information.

[0118] (7) The character processing mode included in the user information is characterized by including the properties of the document used by the user, the attributes of the service used by the user, and the attributes of the application used by the user.

[0119] (8) A character conversion support method for a character conversion support system in which a client device having a front-end processor that processes character strings and a server device that acquires character strings from the front-end processor communicate via a predetermined communication medium to coordinate character string conversion processing, wherein the server device comprises: an acquisition step of acquiring a character string entered by a user and user information from the front-end processor provided in the client device; an inference step of inferring a conversion character string candidate using one of a plurality of inference engines that utilize a large-scale language model based on the character string and the user information acquired in the acquisition step; and a response step of responding to the front-end processor with the conversion character string candidate generated by one of the inference engines; wherein the client device comprises a transfer step of sequentially transferring the character string entered by the user and the user information to the server device via the front-end processor; and the front-end processor sequentially confirms the conversion character string candidates responded to from the server device and outputs them to an output device. [Industrial Applicability]

[0120] In the above system, an example has been described in which the language conversion process is started by passing a key code to FEP102 installed in client devices CL1 to CLN. However, this is a method that is mainly used as pre-processing in the field of natural language processing, and it is also possible to construct a system that incorporates an analysis means, such as mecab, that breaks down sentences into morphemes (the smallest units in which a word has meaning) based on the grammar of the target language and part-of-speech information of the words.

[0121] Furthermore, the multiple second inference units 13-1 to 13-N on the server device 1 side learn the converted strings and the degree of correction, either independently or in cooperation with each other, and thereby convert various users into specific strings that are specifically requested by each user, and therefore can be evolved to present conversion string candidates that are adapted to the user's unique expressions and versatile conversion candidates in a timely manner.

[0122] Furthermore, since the language types valid for the client devices CL1 to CLN are set up, by arranging them in the multiple second inference units 13-1 to 13-N on the server device 1 according to the language used, it is possible to build a system incorporating worldwide translation function processing even when the languages ​​used by the client devices differ.

[0123] In addition, it is possible to further improve user convenience by expanding functions such as automatic correction of typos, OS-specific language processing functions, automatic conversion trigger functions, functions to customize to individual usage environments, context recognition functions, modeless functions, and functions to independently evolve DNNs at appropriate times. [Explanation of symbols]

[0124] 1. Server device CL1~CLN client device 21 Network 50 DNS

Claims

1. A character conversion support system in which a client device having a front-end processor that processes character strings and a server device that acquires character strings from the front-end processor communicate with each other via a predetermined communication medium to cooperate in character string conversion processing, The client device A character conversion support system characterized by having a configuration in which character strings generated by the front-end processor are transferred to a server device, and converted character string candidates that are successively converted by one of a plurality of inference engines that utilize large-scale language models are sequentially acquired from the server device and confirmed.

2. A character conversion support system in which a client device having a front-end processor that processes character strings and a server device that acquires character strings from the front-end processor communicate with each other via a predetermined communication medium to cooperate in character string conversion processing, The server device an acquisition means for acquiring a character string from the front-end processor; a plurality of inference engines that infer candidates for conversion strings from the strings acquired by the acquisition means by utilizing a large-scale language model; a response means for responding to the front-end processor with a candidate conversion string generated by any one of the inference engines; A character conversion support system comprising:

3. A character conversion support system in which a client device having a front-end processor that processes character strings and a server device that acquires character strings from the front-end processor communicate with each other via a predetermined communication medium to cooperate in character string conversion processing, The client device a transfer means for sequentially transferring the character string and user information input by the user to the server device via the front-end processor; The character conversion support system is characterized in that the front-end processor sequentially determines the conversion character string candidates returned from the server device and outputs them to an output device.

4. A character conversion support system in which a client device having a front-end processor that converts a character string input by a user into a predetermined character information string communicates with a server device that acquires the character string input by the user via the front-end processor via a predetermined communication medium to cooperate in character string conversion processing, The server device an acquisition means for continuously acquiring user information specifying a character processing mode used by the user and character strings from the front-end processor; a plurality of inference engines that infer candidates for conversion strings from the strings acquired by the acquisition means by utilizing a large-scale language model; a response means for responding to the front-end processor with a candidate converted character string generated by any one of the inference engines specified by the user information, The client device a transfer means for sequentially transferring the character string being input by the user and the user information to the server device via the front-end processor; The character conversion support system is characterized in that the front-end processor sequentially determines the conversion character string candidates returned from the server device and outputs them to an output device.

5. 5. The character conversion support system according to claim 4, wherein the character string input by the user is input by keyboard or flick input.

6. The server device 5. The character conversion support system according to claim 4, further comprising a configuration for selecting one of a plurality of inference engines based on the character processing mode included in the user information.

7. The character conversion support system according to claim 4, characterized in that the character processing mode included in the user information includes properties of documents used by the user, attributes of services used by the user, and attributes of applications used by the user.

8. A character conversion support method for a character conversion support system in which a client device having a front-end processor that processes character strings and a server device that acquires character strings from the front-end processor communicate with each other via a predetermined communication medium to cooperate in character string conversion processing, comprising: The server device an acquisition step of acquiring a character string input by a user and user information from the front-end processor included in the client device; an inference step of inferring a candidate conversion string using one of a plurality of inference engines that use a large-scale language model, based on the character string acquired in the acquisition step and the user information; a response step of responding to the front-end processor with the conversion string candidates generated by any one of the inference engines, The client device a transfer step of sequentially transferring the character string and user information input by the user to the server device via the front-end processor; The character conversion support method for the character conversion support system is characterized in that the front-end processor sequentially determines the converted character string candidates returned from the server device and outputs them to an output device.

Images