A method and system for constructing a chinese character-universal braille contrast corpus

Abstract

The application provides a Chinese character-general Braille contrast corpus construction method and system, which fully utilizes the word segmentation information in the existing Chinese character-current Braille contrast corpus, and obtains the current Braille corresponding to the pinyin with tones in a semi-automatic way, and then obtains the general Braille. The application adds the tones to the pinyin without the marked tones by the computer, and then automatically judges which pinyin tones may have errors, displays the pinyin with the possible errors and the corresponding Chinese characters on the screen for manual correction and modification. Through the way, the ratio of the pinyin needing manual correction and modification is controlled in a small range, and the corpus construction can be realized quickly.

Description

Technical Field

[0001] The present invention relates to the technical field of semantic recognition and the technical field of braille processing, particularly to the field of constructing a Chinese-Universal Braille parallel corpus. Background Art

[0002] Braille is an important way for blind people to read and obtain information. It is a tactile symbol system, printed on paper or displayed on a dot display, and read by touch. The basic unit of braille is called a "cell", and one cell contains 6 dots. By setting whether each dot has a dot or not, a total of 64 combinations can be formed (including the empty cell where none of the 6 dots has a dot), and these combinations constitute the most basic braille symbols.

[0003] Since it is impossible to uniquely map Chinese characters to 64 braille symbols, Chinese braille is defined as a phonetic script. In Chinese braille, in most cases, 2-3 cells are used to represent one Chinese character, where one cell represents the initial consonant, one cell represents the final vowel, and in some cases, an additional cell is needed to represent the tone. The biggest difference between Chinese braille and Chinese text lies in its "word segmentation and continuous writing" rule, that is, it is required to separate words with an empty cell. Braille word segmentation is different from Chinese word segmentation, and many phrases in Chinese need to be continuously written in braille. For word segmentation and continuous writing, more than 100 rules based on morphology, grammar, and semantics are given in the Chinese Braille Standard, such as "‘不’ should be continuously written with verbs, modal verbs, adjectives, prepositions, and monosyllabic degree adverbs".

[0004] In order to further reduce ambiguity, braille has also formulated a tone marking rule, that is, it stipulates which syllables need to explicitly mark the tone. Currently, there are mainly two braille schemes in use, namely "Current Braille" and "Universal Braille". The word segmentation rules of the two braille schemes are the same, and the main difference lies in the tone marking rule. In Current Braille, in order to save reading time and printing costs, it is stipulated that only easily confused words, unfamiliar words, ancient Chinese content words, very used monosyllabic words, etc. are marked with tones. Generally, it is considered that the tone marking rate of Current Braille is about 5%. However, a large number of unmarked syllables are prone to cause confusion and affect reading and understanding. Therefore, the National Language Commission issued the "Universal Braille" scheme in 2018. Although Universal Braille adopts a certain default setting for tones, after conversion by the default rule, the tones of the vast majority of syllables can be determined. Since the release date in 2018, the country has vigorously promoted Universal Braille. However, due to the lack of tools and systems for automatic conversion from Chinese characters to Universal Braille, and the low efficiency of manual conversion, the current number of Universal Braille books and materials is small, and the text corpus of Chinese-Universal Braille parallel is still blank.

[0005] The corpus of Chinese characters paired with universal Braille plays a crucial role in both theoretical research and the development of digital processing systems for Braille. In theoretical research, the corpus allows for the practical analysis of various linguistic phenomena in real Braille texts, providing data support for Braille teaching and the improvement of Braille schemes. In Braille digital processing, current mainstream Chinese-Braille bidirectional translation and automatic Braille proofreading systems are generally based on machine learning techniques, i.e., using relevant corpora to train machine learning models. Currently, the lack of Chinese-Braille bidirectional translation text corpora is one of the most significant factors hindering the development of technologies such as Chinese-Braille bidirectional translation.

[0006] Chinese-Braille parallel corpora suitable for theoretical research and machine learning model training generally require a one-to-one correspondence between Chinese characters and Braille at the sentence and word levels. This poses a significant challenge to corpus construction because there is currently very little text content that directly possesses this characteristic. Currently, there are two methods for constructing sentence- and word-level parallel Chinese-Braille corpora: one is manual organization, annotation, and alignment of the corpus, which is extremely time-consuming, labor-intensive, and inefficient, potentially taking several years to build a large-scale corpus; the other is automatic extraction and alignment, which automatically extracts and aligns sentences and words from corresponding but not precisely matched Chinese and Braille texts to form a corpus that meets the requirements. This method is more efficient but relies on corresponding Chinese and Braille texts. For general Braille, due to its relatively recent adoption and the scarcity of general Braille books and materials, it is difficult to support the construction of a large-scale corpus, therefore this method is currently unsuitable. Summary of the Invention

[0007] To address the current lack of a large-scale universal Braille corpus and the high cost and low efficiency of obtaining a universal Braille corpus through manual conversion or annotation, this paper proposes a highly efficient method for constructing a Chinese character-universal Braille corpus by utilizing existing Chinese character-current Braille parallel corpora, combined with machine learning, automated computer processing, and a small amount of manual proofreading.

[0008] To address the shortcomings of existing technologies, this invention proposes a method for constructing a Chinese character-universal Braille parallel corpus, which includes:

[0009] Step 1: Obtain Chinese character-current Braille corpus with Chinese character text-current Braille correspondence. Using the mapping table between current Braille syllables and pinyin syllables, convert the current Braille in the Chinese character-current Braille corpus into its corresponding Chinese pinyin. Then, using a dictionary with Chinese character-pinyin correspondence, add tone information to the pinyin strings that are missing tones in the converted pinyin strings to obtain the dictionary-matched pinyin strings corresponding to the Chinese character text in the Chinese character-current Braille corpus.

[0010] Step 2: Input the Chinese text in the current Braille corpus into the machine learning model. The machine learning model outputs the model-predicted pinyin string corresponding to the Chinese text, and the confidence level of the pinyin with tone in the model-predicted pinyin string.

[0011] Step 3: Based on the dictionary-matched pinyin string, the model-predicted pinyin string, and the confidence level, determine whether the pinyin string corresponding to each current Braille word in the current Braille corpus needs to be manually proofread. If so, manually proofread the dictionary-matched pinyin string of the current Braille word to obtain the general Braille pinyin string of the current Braille word. Otherwise, combine the dictionary-matched pinyin string with the model-predicted pinyin string to generate the general Braille pinyin string of the current Braille word.

[0012] Step 4: Using a mapping table between general Braille syllables and toned Pinyin syllables, a Chinese character-general Braille parallel corpus is constructed by converting the general Braille Pinyin string into the general Braille of the current Braille words.

[0013] The method for constructing the Chinese character-universal Braille parallel corpus, wherein step 1 includes:

[0014] Step 11: Determine if the current Braille word is a single-character word. If not, proceed to Step 12. If yes, search for the Chinese character in the Chinese-Pinyin dictionary.

[0015] If the search finds a match, and the Chinese character corresponds to a unique pinyin with a tone, then this pinyin with a tone will be used as the pinyin corresponding to the current Braille word.

[0016] If a match is found and the Chinese character corresponds to multiple different pinyin with tones, then the pinyin strings converted from the current Braille word are compared. If a pinyin with a tone can be uniquely identified in the dictionary based on the pinyin string, then the pinyin string corresponding to the current Braille word is set as the unique pinyin with a tone; if a unique pinyin with a tone cannot be identified, then the pinyin corresponding to the current Braille word is marked as a polyphonic character.

[0017] Step 12: Determine whether the current Braille word is a string of Chinese characters consisting of multiple characters, and whether the string can be found in a Chinese character-pinyin dictionary. If not, proceed to step 13; otherwise, continue the determination:

[0018] If a Chinese character word in a dictionary that corresponds to the current Braille word has the same Chinese character string as the Chinese character string, and the Chinese character word has a unique pinyin string with tone in the dictionary, then the pinyin string with tone is taken as the pinyin string corresponding to the current Braille word.

[0019] If a Chinese character in a Chinese-Pinyin dictionary has a word with the same Chinese character string as the corresponding existing Braille word, and the word corresponds to multiple different Pinyin strings with tones in the dictionary, then the Pinyin strings converted from the existing Braille word are compared. If a Pinyin string with tones can be uniquely identified in the dictionary based on the Pinyin string, then the uniquely identified Pinyin string with tones is taken as the Pinyin string corresponding to the existing Braille word. If a unique Pinyin string with tones still cannot be identified, then the identical parts of the multiple Pinyin strings corresponding to the existing Braille word in the Chinese-Pinyin dictionary are retained, and the different parts are marked as polyphonic characters.

[0020] Step 13: The current Braille word is a string of Chinese characters consisting of multiple characters, and there is no Chinese word in the Chinese-Pinyin dictionary that has the same string of Chinese characters as the corresponding Braille word. The processing is as follows:

[0021] If a Chinese character in a Chinese-Pinyin dictionary has a substring that is identical to the Chinese character string corresponding to the current Braille word, and the matching method is unique, then the corresponding Pinyin string is obtained by matching the Chinese character substring of the Chinese character in the Chinese-Pinyin dictionary to the current Braille string. The corresponding Pinyin string is obtained by processing the remaining Chinese character in the Chinese character string corresponding to the current Braille word. All the Pinyin strings obtained by the above processing are then connected in order to obtain the Pinyin string with tone corresponding to the current Braille word.

[0022] If a Chinese character word in a Chinese character-pinyin dictionary has a partial substring that is the same as the Chinese character string corresponding to the current Braille word, and the matching method is not unique, then the entire Pinyin string corresponding to the current Braille word will be temporarily marked as a multiple matching word.

[0023] If there is no Chinese character word in the Chinese character-pinyin dictionary that is identical to any substring of the Chinese character string corresponding to the current Braille word, then each Chinese character in the current Braille Chinese character string is matched with the Chinese character-pinyin dictionary to obtain its corresponding pinyin string.

[0024] The method for constructing the Chinese character-universal Braille parallel corpus, wherein step 2 includes:

[0025] The Chinese text in the current Braille corpus with word-level correspondence is segmented into sentences according to punctuation. Then, the Chinese text of each sentence is input into the machine learning model to obtain the pinyin string corresponding to the Chinese text and the probability of each pinyin.

[0026] The method for constructing the Chinese character-general Braille parallel corpus includes step 3, which involves determining whether one of the following two conditions is met. If the condition is met, the pinyin string needs to be manually proofread.

[0027] Condition 1: The current Braille word dictionary matches multiple words in the Pinyin string;

[0028] Condition 2: The current Braille word-to-phonetic dictionary matching pinyin string contains polyphonic characters, and the confidence of the pinyin corresponding to the Chinese character in the model prediction pinyin string is less than a preset threshold.

[0029] This invention also proposes a system for constructing a Chinese character-universal Braille parallel corpus, which includes:

[0030] The first matching module is used to obtain Chinese character-current Braille corpus with Chinese character text-current Braille correspondence, convert the current Braille in the Chinese character-current Braille corpus into its corresponding Chinese Pinyin through a mapping table between current Braille syllables and Pinyin syllables, and add tone information to the Pinyin string that is missing tone in the converted Pinyin string through a dictionary with Chinese character-Pinyin correspondence, so as to obtain the dictionary matching Pinyin string corresponding to the Chinese character text in the Chinese character-current Braille corpus;

[0031] The second matching module is used to input the Chinese character text in the current Braille corpus into the machine learning model. The machine learning model outputs the model-predicted pinyin string corresponding to the Chinese character text, as well as the confidence level of the pinyin with tone in the model-predicted pinyin string.

[0032] The first judgment module is used to determine whether the pinyin string corresponding to each current Braille word in the current Braille corpus needs to be manually proofread based on the dictionary matching pinyin string, the model predicted pinyin string and the confidence level. If so, the dictionary matching pinyin string of the current Braille word is manually proofread to obtain the general Braille pinyin string of the current Braille word. Otherwise, the dictionary matching pinyin string and the model predicted pinyin string are combined to generate the general Braille pinyin string of the current Braille word.

[0033] The conversion module is used to construct a Chinese character-general Braille parallel corpus by using a mapping table between general Braille syllables and tone-indicated pinyin syllables and converting the general Braille pinyin string into the general Braille of the current Braille words.

[0034] The system for constructing the Chinese character-universal Braille parallel corpus, wherein the first matching module includes:

[0035] The second judgment module is used to determine whether the current Braille word is a single-character word. If not, the third judgment module is called. If it is, the Chinese character is searched in the Chinese character-pinyin dictionary.

[0036] If the search finds a match, and the Chinese character corresponds to a unique pinyin with a tone, then this pinyin with a tone will be used as the pinyin corresponding to the current Braille word.

[0037] If a match is found and the Chinese character corresponds to multiple different pinyin with tones, then the pinyin strings converted from the current Braille word are compared. If a pinyin with a tone can be uniquely identified in the dictionary based on the pinyin string, then the pinyin string corresponding to the current Braille word is set as the unique pinyin with a tone; if a unique pinyin with a tone cannot be identified, then the pinyin corresponding to the current Braille word is marked as a polyphonic character.

[0038] The third judgment module is used to determine whether the current Braille word is a string of Chinese characters consisting of multiple characters, and whether the string can be found in a Chinese character-pinyin dictionary. If not, the judgment module is called; if so, the judgment continues.

[0039] If a Chinese character word in a dictionary that corresponds to the current Braille word has the same Chinese character string as the Chinese character string, and the Chinese character word has a unique pinyin string with tone in the dictionary, then the pinyin string with tone is taken as the pinyin string corresponding to the current Braille word.

[0040] If a Chinese character in a Chinese-Pinyin dictionary has a word with the same Chinese character string as the corresponding existing Braille word, and the word corresponds to multiple different Pinyin strings with tones in the dictionary, then the Pinyin strings converted from the existing Braille word are compared. If a Pinyin string with tones can be uniquely identified in the dictionary based on the Pinyin string, then the uniquely identified Pinyin string with tones is taken as the Pinyin string corresponding to the existing Braille word. If a unique Pinyin string with tones still cannot be identified, then the identical parts of the multiple Pinyin strings corresponding to the existing Braille word in the Chinese-Pinyin dictionary are retained, and the different parts are marked as polyphonic characters.

[0041] The determination module is used to determine if the current Braille word is a string of Chinese characters consisting of multiple characters, and if there is no Chinese word in the Chinese-Pinyin dictionary that has the same string of Chinese characters as the corresponding current Braille word. The processing is divided into the following parts:

[0042] If a Chinese character in a Chinese-Pinyin dictionary has a substring that is identical to the Chinese character string corresponding to the current Braille word, and the matching method is unique, then the corresponding Pinyin string is obtained by matching the Chinese character substring of the Chinese character in the Chinese-Pinyin dictionary to the current Braille string. The corresponding Pinyin string is obtained by processing the remaining Chinese character in the Chinese character string corresponding to the current Braille word. All the Pinyin strings obtained by the above processing are then connected in order to obtain the Pinyin string with tone corresponding to the current Braille word.

[0043] If a Chinese character word in a Chinese character-pinyin dictionary has a partial substring that is the same as the Chinese character string corresponding to the current Braille word, and the matching method is not unique, then the entire Pinyin string corresponding to the current Braille word will be temporarily marked as a multiple matching word.

[0044] If there is no Chinese character word in the Chinese character-pinyin dictionary that is identical to any substring of the Chinese character string corresponding to the current Braille word, then each Chinese character in the current Braille Chinese character string is matched with the Chinese character-pinyin dictionary to obtain its corresponding pinyin string.

[0045] The system for constructing the Chinese character-general Braille corpus includes module 2, which is used to segment the Chinese character text in the word-level Chinese character-current Braille corpus into sentences according to punctuation, and then input the Chinese character text of each sentence into the machine learning model in turn to obtain the pinyin string corresponding to the Chinese character text and the probability corresponding to each pinyin.

[0046] The system for constructing the Chinese character-general Braille parallel corpus, wherein module 3 is used to determine whether one of the following two conditions is met; if it is met, the pinyin string needs to be manually proofread.

[0047] Condition 1: The current Braille word dictionary matches multiple words in the Pinyin string;

[0048] Condition 2: The current Braille word-to-phonetic dictionary matching pinyin string contains polyphonic characters, and the confidence of the pinyin corresponding to the Chinese character in the model prediction pinyin string is less than a preset threshold.

[0049] The present invention also proposes a storage medium for storing a program that executes the method for constructing any of the Chinese character-general Braille parallel corpora.

[0050] This invention also proposes a client application for constructing any of the aforementioned Chinese character-general Braille parallel corpora.

[0051] As can be seen from the above solutions, the advantages of the present invention are:

[0052] By employing the Chinese character-general Braille parallel corpus construction method proposed in this invention, the drawbacks of current manual construction of Chinese character-general Braille parallel corpora, such as low efficiency and high cost, can be overcome. This method fully utilizes word segmentation information from existing Chinese character-current Braille parallel corpora, and employs a semi-automatic approach to obtain the pinyin with tones corresponding to current Braille, thereby obtaining general Braille. In this approach, the computer first automatically adds tones to pinyin that are not marked with tones, then automatically identifies which pinyin may have incorrect tones, and displays these potentially incorrect pinyin and their corresponding Chinese characters on the screen for manual proofreading and correction. This method keeps the proportion of pinyin requiring manual proofreading and correction to a very small range, enabling rapid corpus construction. Attached Figure Description

[0053] Figure 1 A flowchart of the main methods for constructing a Chinese character-general Braille parallel corpus;

[0054] Figure 2 Example diagram of a Chinese character-current Braille parallel corpus;

[0055] Figure 3 This is an example diagram of converting current Braille into Pinyin from a Chinese character-current Braille comparative corpus;

[0056] Figure 4 Example image showing the addition of dictionary-matched pinyin strings to a Chinese character-current Braille corpus;

[0057] Figure 5 Example image showing the addition of model-predicted pinyin strings and confidence levels to a Chinese character-current Braille parallel corpus;

[0058] Figure 6 Example diagram for determining whether existing Braille words need manual proofreading;

[0059] Figure 7 This is a sample diagram of the final universal Braille text. Detailed Implementation

[0060] The method of this invention is based on a word-level comparative Chinese character-current Braille corpus to further construct a word-level comparative Chinese character-general Braille corpus. The word-level comparative Chinese character-current Braille corpus used refers to comparing Chinese characters and current Braille according to Braille word segmentation, with each current Braille word corresponding to its corresponding Chinese character string.

[0061] The main steps of the method include:

[0062] 1. Obtain the corresponding pinyin string from the existing Braille.

[0063] A mapping table between existing Braille syllables and Pinyin syllables can be used to convert existing Braille into its corresponding Pinyin. Whether the resulting Pinyin syllable has a tone depends on whether the corresponding existing Braille syllable has a tone.

[0064] 2. Based on the Chinese characters in the Chinese-current Braille parallel corpus and the pinyin strings converted from current Braille, a Chinese-pinyin parallel dictionary is used to add tones to the pinyin strings that lack tones in the current Braille-converted pinyin strings. Pinyin characters that are confusing or whose tones cannot be determined are marked as "polyphonic characters." After this step, a corresponding pinyin string, called a "dictionary-matched pinyin string," can be obtained for each Braille word. The specific steps are as follows:

[0065] 2.1 If a current Braille word is a single-character word, that is, it corresponds to a single Chinese character, then look up the Chinese character in a dictionary with both Chinese character and pinyin:

[0066] (1) If in a Chinese character-pinyin dictionary, the Chinese character corresponds to a unique pinyin with a tone, then the pinyin with the tone shall be used as the pinyin corresponding to the current Braille word.

[0067] (2) If the Chinese character corresponds to multiple different pinyin with tones in the Chinese character-pinyin dictionary, then compare the pinyin strings converted from the current Braille word. If a pinyin with a tone can be uniquely determined in the dictionary based on the pinyin string, then the pinyin string corresponding to the current Braille word is set as the uniquely determined pinyin with a tone. If a unique pinyin with a tone still cannot be determined, then the pinyin corresponding to the current Braille word is temporarily marked as a "polyphonic character" and left for subsequent steps.

[0068] 2.2 If a current Braille word is a string of Chinese characters, and this string can be found in a Chinese-Pinyin dictionary, then it is handled in two ways:

[0069] (1) If there is a Chinese character word in the Chinese character-pinyin dictionary that is the same as the Chinese character string corresponding to the current Braille word, and the Chinese character word corresponds to a unique pinyin string with tone in the dictionary, then the pinyin string with tone shall be used as the pinyin corresponding to the current Braille.

[0070] (2) If there is a Chinese character word in the Chinese-Pinyin dictionary that is the same as the Chinese character string corresponding to the current Braille word, and the Chinese character word corresponds to multiple different Pinyin strings with tones in the dictionary, then compare the Pinyin strings converted from the current Braille word. If a Pinyin string with tones in the dictionary can be uniquely determined based on the Pinyin string, then the uniquely determined Pinyin string with tones is taken as the Pinyin string corresponding to the current Braille word; if a unique Pinyin string with tones still cannot be determined, then the same part in the multiple Pinyin strings corresponding to the current Braille word in the Chinese-Pinyin dictionary is retained, and the different parts are temporarily marked as "polyphonic characters". This marking is taken as the Pinyin string corresponding to the current Braille word and left for subsequent steps.

[0071] 2.3 If a current Braille word is a string of Chinese characters consisting of multiple characters, and there is no corresponding Chinese character string in a Chinese-Pinyin dictionary, it is handled in three ways:

[0072] (1) If a Chinese character word in the Chinese-Pinyin dictionary has a substring that is identical to the corresponding Chinese character string of the current Braille word, and the matching method is unique, then process the current Braille string corresponding to the substring of the Chinese character word in the Chinese-Pinyin dictionary according to the processing method in step 2.2 above to obtain its corresponding Pinyin string; then process the current Braille string corresponding to each Chinese character in the remaining part of the Chinese character string corresponding to the current Braille word according to the processing method in step 2.1 above to obtain its corresponding Pinyin string. After connecting all the Pinyin strings obtained by the above processing in order, the Pinyin string with tone corresponding to the current Braille word can be obtained;

[0073] (2) If there are Chinese characters in the Chinese-Pinyin dictionary that are identical to a substring of the Chinese character string corresponding to the current Braille word, and the matching method is not unique, then the entire Pinyin string corresponding to the current Braille word will be temporarily marked as "uncertain, matching multiple words" and left for subsequent steps to process;

[0074] (3) If there is no Chinese character word in the Chinese character-pinyin dictionary that is the same as any substring of the Chinese character string corresponding to the current Braille word, then the current Braille corresponding to each Chinese character of the current Braille string is processed in the manner described in step 2.1, and the results are then connected in sequence. The connected result is used as the pinyin string corresponding to the current Braille.

[0075] 3. Using a pre-trained machine learning model, the corresponding pinyin with tones is automatically labeled based on the Chinese text in the Chinese-current Braille corpus. This is called the "model-predicted pinyin string". The probability of each pinyin is output as its confidence level.

[0076] Using a Chinese character-Pinyin parallel corpus, a BERT pre-trained model was fine-tuned to obtain a model that automatically annotates Chinese text with Pinyin. The Chinese text in the word-level parallel corpus of Chinese characters and current Braille was segmented into sentences according to punctuation. Then, the Chinese text of each sentence was sequentially input into the model to obtain the corresponding Pinyin string and the probability of each Pinyin syllable. By establishing a correspondence between the Pinyin syllables and their corresponding probabilities and the Chinese characters and words in the Chinese text, a Pinyin string with tone marks can be obtained for each current Braille word in the corpus. This is called the "model-predicted Pinyin string," and the probability of each Pinyin syllable in this string can also be obtained as the confidence level of that Pinyin.

[0077] 4. For each existing Braille word, combine the dictionary-matched pinyin string, the model-predicted pinyin string, and the pinyin confidence score to determine whether the corresponding pinyin string needs manual proofreading. If manual proofreading is required, the model-predicted pinyin string is used as the "pinyin string to be proofread" for the existing Braille word; if manual proofreading is not required, the dictionary-matched pinyin string and the model-predicted pinyin string are combined to generate the "general Braille pinyin string" for the existing Braille word.

[0078] For each existing Braille word, manual verification is required if one of the following two conditions is met: (1) The dictionary matching pinyin string corresponding to the existing Braille word contains the substring "uncertain, matching multiple words"; (2) The dictionary matching pinyin string corresponding to the existing Braille word contains the substring "polyphonic character", and the confidence of the pinyin corresponding to the Chinese character corresponding to the "polyphonic character" substring in the model prediction pinyin string is less than a preset threshold.

[0079] When an existing Braille word needs to be manually proofread, the model-predicted pinyin string is used as the "pinyin string to be proofread" corresponding to the existing Braille word for subsequent manual proofreading.

[0080] No manual verification is required when an existing Braille word meets one of the following two conditions: (1) The dictionary matching pinyin string corresponding to the existing Braille word does not contain the substring "polyphonic character" or "uncertain, matching multiple words", that is, the dictionary matching pinyin string only contains definite pinyin; (2) The dictionary matching pinyin string corresponding to the existing Braille word contains the substring "polyphonic character", but the confidence of the pinyin corresponding to the Chinese character of the "polyphonic character" substring in the model prediction pinyin string is greater than the preset threshold.

[0081] When an existing Braille word does not require manual proofreading, a "general Braille pinyin string" is generated by combining the dictionary-matched pinyin string and the model-predicted pinyin string. The generation method is as follows: if the dictionary-matched pinyin string corresponding to the existing Braille word does not contain the substring "polyphonic character" or "uncertain, matching multiple words", that is, the dictionary-matched pinyin string only contains definite pinyin, then the dictionary-matched pinyin string is directly used as the "general Braille pinyin string"; if the dictionary-matched pinyin string corresponding to the existing Braille word contains the substring "polyphonic character", and the confidence of the pinyin corresponding to the Chinese character of the "polyphonic character" substring in the model-predicted pinyin string is greater than a preset threshold, then based on the dictionary-matched pinyin string, the substring "polyphonic character" is replaced with the pinyin corresponding to the Chinese character of the "polyphonic character" substring in the model-predicted pinyin string.

[0082] 5. Present the existing Braille words that require manual proofreading, along with their corresponding Chinese character strings, pinyin strings, and pinyin strings to be proofreading, to the proofreaders in a format suitable for inspection and modification. The proofreaders will then confirm or modify the information.

[0083] 6. After manual modification is completed, the manually confirmed or modified "Pinyin string to be proofread" is automatically used as the universal Braille Pinyin string for the current Braille word. In this way, each current Braille word has a corresponding "universal Braille Pinyin string".

[0084] 7. Convert the general Braille pinyin string to general Braille. Use a mapping table between general Braille syllables and pinyin syllables with tones to convert the general Braille pinyin string generated in the previous steps into general Braille. The mapping table between general Braille syllables and pinyin syllables with tones can be obtained by organizing the general Braille spelling rules.

[0085] To make the above features and effects of the present invention clearer and easier to understand, specific embodiments are described below, and detailed descriptions are provided in conjunction with the accompanying drawings.

[0086] The main process of the Chinese character-universal Braille parallel corpus construction method proposed in this invention is as follows: Figure 1 As shown, this method, based on an existing Chinese character-current Braille parallel corpus, primarily aims to add tones to syllables in current Braille that are not marked with tones, referencing the corresponding Chinese characters. To ensure tones accuracy, the method employs a semi-automatic approach: the computer first automatically adds tones to the unmarked syllables, then automatically identifies which syllables may have incorrect tones, displaying these potentially erroneous syllables and their corresponding Chinese characters on the screen for manual correction. In identifying potentially erroneous syllables, the goal is to ensure that as many erroneous syllables as possible are missed, while minimizing the number of syllables obtained to reduce the workload of manual correction. Therefore, this invention uses a method that integrates the confidence scores of a dictionary and a machine learning model to determine which syllables have automatically added tones that are reliable and require no manual review, and which syllables may have incorrect tones that require manual review.

[0087] The method proposed in this invention further constructs a word-level parallel Chinese character-current Braille corpus based on a word-level parallel Chinese character-current Braille corpus. The word-level parallel Chinese character-current Braille corpus used refers to a parallel structure of Chinese characters and current Braille according to Braille word segmentation, where each current Braille word corresponds to its corresponding Chinese character string. In one embodiment, a segment of the corpus is as follows: Figure 2 As shown. Each line of the corpus contains a current Braille word and its corresponding Chinese character string, or Braille punctuation marks and symbols and their corresponding Chinese punctuation marks and symbols.

[0088] The method for constructing a Chinese character-universal Braille parallel corpus proposed in this invention mainly consists of the following steps:

[0089] 1. Obtain the corresponding pinyin string from the existing Braille.

[0090] Chinese Braille is essentially a phonetic script that represents the pronunciation of Chinese characters. Therefore, there is an almost one-to-one correspondence between the Braille syllables and the syllables of Chinese Pinyin. Through a mapping table between Braille syllables and Pinyin syllables, the current Braille can be converted into its corresponding Chinese Pinyin. Whether the obtained Pinyin syllable has a tone depends on whether the corresponding current Braille syllable has a tone, as shown in the last column of Figure 3 . In addition, for the punctuation marks and symbols in the current Braille, the corresponding Chinese punctuation marks and symbols can be filled in the Pinyin string. Figure 3 As shown in the last column of Figure 3 . In addition, for the punctuation marks and symbols in the current Braille, the corresponding Chinese punctuation marks and symbols can be filled in the Pinyin string.

[0091] 2. Based on the Chinese characters in the Chinese character-current Braille parallel corpus and the Pinyin strings obtained by converting the current Braille, using the Chinese character-Pinyin parallel dictionaries and glossaries, add tones to the Pinyin strings without tones obtained by converting the current Braille. For the Pinyin with confusion and unable to determine the tone, mark it as "polysyllabic character".

[0092] In this step, a Chinese character-Pinyin parallel dictionary and a Chinese character-Pinyin parallel glossary are needed. The former lists the Pinyin with tones corresponding to all known Chinese characters, and lists multiple Pinyin for polysyllabic characters at the same time; the latter lists the Pinyin with tones corresponding to multi-character words (including two-character words) in Chinese, and lists multiple pronunciations for words with multiple pronunciations at the same time. The multiple pronunciations of the said characters or words include the situation where one of the pronunciations is (or includes) a light tone. For example, the word "dìdao" has two corresponding Pinyins: ① "di4 / dao4", which means "underground road or tunnel"; ② "di4 / dao5" (the tone "5" indicates a light tone), which means "authentic". The said Chinese character-Pinyin parallel dictionary and Chinese character-Pinyin parallel glossary can be compiled by referring to authoritative dictionaries such as the Modern Chinese Dictionary and the Contemporary Chinese Dictionary.

[0093] For a current Braille word, when adding tones to the Pinyin without tones obtained by converting the said current Braille word using the Chinese character-Pinyin parallel dictionaries and glossaries, the following steps are carried out:

[0094] 1) If a certain current Braille word is a single-character word, that is, it corresponds to a single Chinese character, then look up this Chinese character in the Chinese character-Pinyin parallel dictionary:

[0095] (1) If in the Chinese character-Pinyin parallel dictionary, this Chinese character corresponds to a unique Pinyin with a tone, such as "山" corresponding to the Pinyin "shan1", then use this Pinyin with a tone as the Pinyin corresponding to this current Braille word;

[0096] (2) If in a Chinese character - pinyin dictionary, a Chinese character corresponds to multiple different pinyins with tones, such as "差" corresponding to "cha1", "cha4", and "chai1", then compare the pinyin string converted from the current Braille word. If a pinyin with tone in the dictionary can be uniquely determined based on this pinyin string. For example, when the pinyin string converted from the current Braille word is "chai", the pinyin with tone can only be "chai1", and when the pinyin string converted from the current Braille word is the toned "cha1", the pinyin with tone of course can only be "cha1". In this case, use the uniquely determined pinyin with tone as the corresponding pinyin of the current Braille word; if the uniquely determined pinyin with tone still cannot be determined, such as when the pinyin string converted from the current Braille word is "cha" and it is impossible to judge whether it is "cha1" or "cha4", then temporarily mark the corresponding pinyin of the current Braille word as "polysyllabic character", to be processed in subsequent steps;

[0097] 2) If a certain current Braille word corresponds to a Chinese character string composed of multiple characters, and the Chinese character string can be found in a Chinese character - pinyin dictionary, it is processed in two cases:

[0098] (1) If there is a Chinese character word in the Chinese character - pinyin dictionary that is the same as the Chinese character string corresponding to the current Braille word, and this Chinese character word corresponds to a unique pinyin string with tone in the dictionary, such as "北京" corresponding to "bei3 / jing1", then use this pinyin string with tone as the corresponding pinyin of the current Braille word;

[0099] (2) If there is a Chinese character word in the Chinese character - pinyin dictionary that is the same as the Chinese character string corresponding to the current Braille word, and this Chinese character word corresponds to multiple different pinyin strings with tone in the dictionary, such as "地道" corresponding to "di4 / dao4" and "di4 / dao5", then compare the pinyin string converted from the current Braille word. If a pinyin string with tone in the dictionary can be uniquely determined based on this pinyin string, then use the uniquely determined pinyin string with tone as the corresponding pinyin string of the current Braille word. For example, if the pinyin string converted from the current Braille word is "di / dao4", then "di4 / dao4" can be uniquely determined; if the uniquely determined pinyin string with tone still cannot be determined, then retain the same part in the multiple pinyin strings, and temporarily mark the different part as "polysyllabic character". In this example, mark it as "di4 / polysyllabic character", and use this mark as the corresponding pinyin of the current Braille, to be processed in subsequent steps;

[0100] 3) If a certain current Braille word is a Chinese character string composed of multiple characters, and there is no Chinese character word in the Chinese character - pinyin dictionary that is the same as the Chinese character string corresponding to the current Braille word, it is processed in 3 cases:

[0101] (1) If there is a partial substring of the Chinese character string corresponding to the current Braille word that is the same as a Chinese character word in the Chinese-character to pinyin dictionary, and the matching method is unique. For example, "憧憬" in "憧憬着" can be matched to "憧憬" in the Chinese-character to pinyin dictionary, and there is no other substring in "憧憬着" that matches a Chinese character word in the Chinese-character to pinyin dictionary. Then, process the current Braille corresponding to the substring of the Chinese character word matched to the Chinese-character to pinyin dictionary according to the processing method in step 2) above to obtain its corresponding pinyin string. For example, the current Braille corresponding to "憧憬" in this example, after processing, can obtain the pinyin string with tones "chong1 / jing3". Then, process the current Braille corresponding to each Chinese character in the remaining part of the Chinese character string corresponding to the current Braille word according to the processing method in step 1) above, and respectively obtain its corresponding pinyin string. For example, the current Braille corresponding to "着" in this example, after processing, can obtain the pinyin with tones "zhe5". After connecting all the pinyin strings obtained by the above processing in order, the pinyin string with tones corresponding to the current Braille word can be obtained. For example, in this example, it is "chong1 / jing3 / zhe5";

[0102] (2) If there is a partial substring of the Chinese character string corresponding to the current Braille word that is the same as a Chinese character word in the Chinese-character to pinyin dictionary, and the matching method is not unique. For example, both "证明" and "明了" in "证明了" can be matched to Chinese character words in the Chinese-character to pinyin dictionary, and their corresponding pinyins are "zheng4 / ming2" and "ming2 / liao3" respectively. At this time, since it is impossible to determine which match is correct, therefore, temporarily mark the pinyin string corresponding to the entire current Braille word as "uncertain, multiple words matched" and leave it for subsequent steps to process;

[0103] (3) If there is no Chinese character word in the Chinese-character to pinyin dictionary that is the same as any substring of the Chinese character string corresponding to the current Braille word, then process the current Braille corresponding to each Chinese character in the Chinese character string corresponding to the current Braille word in turn according to the processing method in step 1), and connect the results obtained in turn. The connected result is used as the pinyin string corresponding to the current Braille word. For example, the Chinese character string corresponding to a certain current Braille word is "尽收于", and neither this Chinese character string nor any of its substrings can be matched to the same Chinese character word in the Chinese-character to pinyin dictionary. Therefore, divide it into 3 Chinese characters "尽", "收", and "于" in order. For the current Braille corresponding to these 3 Chinese characters, obtain their corresponding pinyins according to step 1) respectively, which are "polysyllabic word", "shou1", and "yu2". Therefore, after connecting these 3 pinyins, the pinyin string corresponding to the current Braille word can be obtained as "polysyllabic word / shou1 / yu2".

[0104] After step 2, for each Braille word, a corresponding pinyin string can be obtained, which is called the "dictionary matching pinyin string". For example Figure 4 As shown.

[0105] 3. Using a pre-trained machine learning model, the corresponding pinyin with tones is automatically labeled based on the Chinese text in the Chinese-current Braille corpus. This is called the "model-predicted pinyin string". The probability of each pinyin is output as its confidence level.

[0106] Currently, in fields such as speech synthesis, there exist corpora with Chinese characters and pinyin annotations. Machine learning models can be trained based on these corpora to automatically annotate Chinese text with pinyin.

[0107] In one embodiment, a BERT pre-trained model is used. BERT is a deep neural network model based on self-attention. This model can be pre-trained using a large amount of Chinese text, and then fine-tuned using a Chinese character-pinyin parallel corpus to obtain a model that can automatically annotate Chinese text with pinyin.

[0108] After obtaining the trained model, the Chinese text in the word-level Chinese-Braille parallel corpus is segmented into sentences according to punctuation marks such as periods, question marks, and exclamation marks. Then, the Chinese text of each sentence is sequentially input into the model to obtain the corresponding pinyin string and the probability of each pinyin. The pinyin strings and their corresponding probabilities are then mapped to the Chinese characters and words in the Chinese text, thus also mapping them to existing Braille words. In this way, for each Braille word in the word-level Chinese-Braille parallel corpus, a pinyin string with tones can be obtained, called the "model-predicted pinyin string," and the probability of each pinyin in this string can also be obtained as the confidence level of that pinyin. Figure 5 Provide an example sentence.

[0109] 4. For each existing Braille word, combine the dictionary-matched pinyin string, the model-predicted pinyin string, and the pinyin confidence score to determine whether the pinyin corresponding to the existing Braille word needs manual proofreading. If manual proofreading is required, the model-predicted pinyin string is used as the "pinyin string to be proofread" for the existing Braille word; if manual proofreading is not required, the dictionary-matched pinyin string and the model-predicted pinyin string are combined to generate the "general Braille pinyin string" corresponding to the existing Braille word.

[0110] For each existing Braille word, manual verification is required if one of the following two conditions is met: (1) the dictionary matching pinyin string corresponding to the existing Braille word contains the substring "uncertain, multiple matching words"; (2) the dictionary matching pinyin string corresponding to the existing Braille word contains the substring "polyphonic character", and the confidence of the pinyin corresponding to the Chinese character corresponding to the "polyphonic character" substring in the model prediction pinyin string is less than a preset threshold. In one embodiment, a preferred value of the threshold is 0.95.

[0111] When an existing Braille word needs to be manually proofread, the model-predicted pinyin string is used as the "pinyin string to be proofread" corresponding to the existing Braille word for subsequent manual proofreading.

[0112] No manual verification is required when an existing Braille word meets one of the following two conditions: (1) The dictionary matching pinyin string corresponding to the existing Braille word does not contain the substring "polyphonic character" or "uncertain, matching multiple words", that is, the dictionary matching pinyin string only contains definite pinyin; (2) The dictionary matching pinyin string corresponding to the existing Braille word contains the substring "polyphonic character", but the confidence of the pinyin corresponding to the Chinese character of the "polyphonic character" substring in the model prediction pinyin string is greater than or equal to a preset threshold. In one embodiment, a preferred value of the threshold is 0.95.

[0113] When an existing Braille word does not require manual proofreading, a "general Braille pinyin string" is generated by combining the dictionary-matched pinyin string and the model-predicted pinyin string. The generation method is as follows: if the dictionary-matched pinyin string corresponding to the existing Braille word does not contain the substring "polyphonic character" or "uncertain, matching multiple words," that is, the dictionary-matched pinyin string only contains definite pinyin, then the dictionary-matched pinyin string is directly used as the "general Braille pinyin string"; if the dictionary-matched pinyin string corresponding to the existing Braille word contains the substring "polyphonic character," and the confidence level of the pinyin corresponding to the Chinese character of the "polyphonic character" substring in the model-predicted pinyin string is greater than or equal to a pre-set threshold, then based on the dictionary-matched pinyin string, the substring "polyphonic character" is replaced with the pinyin corresponding to the Chinese character of the "polyphonic character" substring in the model-predicted pinyin string.

[0114] Figure 6An example in the implementation process is given. In the example, since the Chinese character string "应用于" has a pinyin string "不确定,匹配多词" due to dictionary matching, manual proofreading is required. In this case, because it is impossible to determine whether this Braille word is composed of "应用" (pinyin: ying4 / yong4) and "于", or "应" (pinyin: ying1) and "用于", there is ambiguity, so manual proofreading is needed. For other Braille words, no manual proofreading is required. Among them, although the Chinese characters "并", "和", "分" are polyphonic characters, because the confidence of the model-predicted pinyin string is very high, greater than the preset threshold of 0.95 in this embodiment, it is considered that no manual proofreading is required, and the model-predicted pinyin string is used as its general Braille pinyin string. It can be seen that in this way, most polyphonic characters do not require manual proofreading, greatly reducing the workload of manual proofreading. Moreover, for the pinyin string to be proofread, since the model-predicted pinyin string is provided and this pinyin string also has a high accuracy rate, in most cases, only manual confirmation is required without modification, which can also greatly improve efficiency.

[0115] 5. Present information such as the current Braille words that need to be manually proofread, their corresponding Chinese character strings, pinyin strings, pinyin strings to be proofread, etc. to the proofreader in a format suitable for inspection and modification, and let the proofreader confirm or modify them.

[0116] In one embodiment, the Figure 6 content shown is directly presented to the proofreader on the computer screen. The proofreader checks the pinyin string to be proofread of the current Braille words that need to be manually proofread. If no modification is needed, it remains unchanged. If modification is needed, the pinyin string is directly modified to the correct form.

[0117] 6. After the manual modification is completed, automatically use the "pinyin string to be proofread" that has been manually confirmed or modified as the general Braille pinyin string of this current Braille. In this way, each current Braille word has a corresponding "general Braille pinyin string".

[0118] 7. Convert the general Braille pinyin string into general Braille. Use the mapping table between general Braille syllables and tone-marked pinyin syllables to convert the general Braille pinyin string generated in the previous steps into general Braille. The mapping table between general Braille syllables and tone-marked pinyin syllables can be sorted out according to the general Braille spelling rules.

[0119] The final general Braille obtained is as shown in the last column of Figure 7 the following.

[0120] The following are system embodiments corresponding to the above method embodiments. This embodiment can be implemented in conjunction with the above embodiments. The relevant technical details mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.

[0121] This invention also proposes a system for constructing a Chinese character-universal Braille parallel corpus, which includes:

[0122] The first matching module is used to obtain Chinese character-current Braille corpus with Chinese character text-current Braille correspondence, convert the current Braille in the Chinese character-current Braille corpus into its corresponding Chinese Pinyin through a mapping table between current Braille syllables and Pinyin syllables, and add tone information to the Pinyin string that is missing tone in the converted Pinyin string through a dictionary with Chinese character-Pinyin correspondence, so as to obtain the dictionary matching Pinyin string corresponding to the Chinese character text in the Chinese character-current Braille corpus;

[0123] The second matching module is used to input the Chinese character text in the current Braille corpus into the machine learning model. The machine learning model outputs the model-predicted pinyin string corresponding to the Chinese character text, as well as the confidence level of the pinyin with tone in the model-predicted pinyin string.

[0124] The first judgment module is used to determine whether the pinyin string corresponding to each current Braille word in the current Braille corpus needs to be manually proofread based on the dictionary matching pinyin string, the model predicted pinyin string and the confidence level. If so, the dictionary matching pinyin string of the current Braille word is manually proofread to obtain the general Braille pinyin string of the current Braille word. Otherwise, the dictionary matching pinyin string and the model predicted pinyin string are combined to generate the general Braille pinyin string of the current Braille word.

[0125] The conversion module is used to construct a Chinese character-general Braille parallel corpus by using a mapping table between general Braille syllables and tone-indicated pinyin syllables and converting the general Braille pinyin string into the general Braille of the current Braille words.

[0126] The system for constructing the Chinese character-universal Braille parallel corpus, wherein the first matching module includes:

[0127] The second judgment module is used to determine whether the current Braille word is a single-character word. If not, the third judgment module is called. If it is, the Chinese character is searched in the Chinese character-pinyin dictionary.

[0128] If the search finds a match, and the Chinese character corresponds to a unique pinyin with a tone, then this pinyin with a tone will be used as the pinyin corresponding to the current Braille word.

[0129] If a match is found and the Chinese character corresponds to multiple different pinyin with tones, then the pinyin strings converted from the current Braille word are compared. If a pinyin with a tone can be uniquely identified in the dictionary based on the pinyin string, then the pinyin string corresponding to the current Braille word is set as the unique pinyin with a tone; if a unique pinyin with a tone cannot be identified, then the pinyin corresponding to the current Braille word is marked as a polyphonic character.

[0130] The third judgment module is used to determine whether the current Braille word is a string of Chinese characters consisting of multiple characters, and whether the string can be found in a Chinese character-pinyin dictionary. If not, the judgment module is called; if so, the judgment continues.

[0131] If a Chinese character word in a dictionary that corresponds to the current Braille word has the same Chinese character string as the Chinese character string, and the Chinese character word has a unique pinyin string with tone in the dictionary, then the pinyin string with tone is taken as the pinyin string corresponding to the current Braille word.

[0132] If a Chinese character in a Chinese-Pinyin dictionary has a word with the same Chinese character string as the corresponding existing Braille word, and the word corresponds to multiple different Pinyin strings with tones in the dictionary, then the Pinyin strings converted from the existing Braille word are compared. If a Pinyin string with tones can be uniquely identified in the dictionary based on the Pinyin string, then the uniquely identified Pinyin string with tones is taken as the Pinyin string corresponding to the existing Braille word. If a unique Pinyin string with tones still cannot be identified, then the identical parts of the multiple Pinyin strings corresponding to the existing Braille word in the Chinese-Pinyin dictionary are retained, and the different parts are marked as polyphonic characters.

[0133] The determination module is used to determine if the current Braille word is a string of Chinese characters consisting of multiple characters, and if there is no Chinese word in the Chinese-Pinyin dictionary that has the same string of Chinese characters as the corresponding current Braille word. The processing is divided into the following parts:

[0134] If a Chinese character in a Chinese-Pinyin dictionary has a substring that is identical to the Chinese character string corresponding to the current Braille word, and the matching method is unique, then the corresponding Pinyin string is obtained by matching the Chinese character substring of the Chinese character in the Chinese-Pinyin dictionary to the current Braille string. The corresponding Pinyin string is obtained by processing the remaining Chinese character in the Chinese character string corresponding to the current Braille word. All the Pinyin strings obtained by the above processing are then connected in order to obtain the Pinyin string with tone corresponding to the current Braille word.

[0135] If a Chinese character word in a Chinese character-pinyin dictionary has a partial substring that is the same as the Chinese character string corresponding to the current Braille word, and the matching method is not unique, then the entire Pinyin string corresponding to the current Braille word will be temporarily marked as a multiple matching word.

[0136] If there is no Chinese character word in the Chinese character-pinyin dictionary that is identical to any substring of the Chinese character string corresponding to the current Braille word, then each Chinese character in the current Braille Chinese character string is matched with the Chinese character-pinyin dictionary to obtain its corresponding pinyin string.

[0137] The system for constructing the Chinese character-general Braille corpus includes module 2, which is used to segment the Chinese character text in the word-level Chinese character-current Braille corpus into sentences according to punctuation, and then input the Chinese character text of each sentence into the machine learning model in turn to obtain the pinyin string corresponding to the Chinese character text and the probability corresponding to each pinyin.

[0138] The system for constructing the Chinese character-general Braille parallel corpus, wherein module 3 is used to determine whether one of the following two conditions is met; if it is met, the pinyin string needs to be manually proofread.

[0139] Condition 1: The current Braille word dictionary matches multiple words in the Pinyin string;

[0140] Condition 2: The current Braille word-to-phonetic dictionary matching pinyin string contains polyphonic characters, and the confidence of the pinyin corresponding to the Chinese character in the model prediction pinyin string is less than a preset threshold.

[0141] The present invention also proposes a storage medium for storing a program that executes the method for constructing any of the Chinese character-general Braille parallel corpora.

[0142] This invention also proposes a client application for constructing any of the aforementioned Chinese character-general Braille parallel corpora.

Claims

1. A method for constructing a Chinese character-Universal Braille parallel corpus, characterized in that, The method comprises the following steps: Step 1, obtaining a Chinese character-current Braille corpus with Chinese character text-current Braille pairs, converting the current Braille in the Chinese character-current Braille corpus into corresponding Chinese pinyin by using a mapping table between current Braille syllables and pinyin syllables, and adding tone information to the pinyin lacking tone in the converted pinyin string by using a dictionary with Chinese character-pinyin pairs, to obtain a dictionary-matched pinyin string corresponding to the Chinese character text in the Chinese character-current Braille corpus; Step 2, inputting the Chinese character text in the Chinese character-current Braille corpus into a machine learning model, and outputting a model-predicted pinyin string corresponding to the Chinese character text and a confidence degree of the Chinese pinyin with tone in the model-predicted pinyin string; Step 3, judging whether the pinyin string corresponding to each current Braille word in the Chinese character-current Braille corpus needs to be manually corrected according to the dictionary-matched pinyin string, the model-predicted pinyin string and the confidence degree, if yes, manually correcting the dictionary-matched pinyin string of the current Braille word to obtain a general Braille pinyin string of the current Braille word, otherwise, combining the dictionary-matched pinyin string and the model-predicted pinyin string to generate the general Braille pinyin string of the current Braille word; Step 4, using a mapping table between general Braille syllables and pinyin syllables with tone, converting the general Braille pinyin string into general Braille of the current Braille word to construct a Chinese character-general Braille corpus; The step 1 comprises: Step 11, judging whether the current current Braille word is a single-character word, if no, executing step 12, if yes, searching the Chinese character in a dictionary with Chinese character-pinyin pairs: If the search hits, the Chinese character corresponds to a unique pinyin with tone, and the pinyin with tone is taken as the pinyin corresponding to the current Braille word; If the search hits, the Chinese character corresponds to multiple different pinyins with tone, and the pinyin string converted from the current Braille word is compared, if the pinyin string can uniquely determine a pinyin with tone in the dictionary, the pinyin string of the current Braille word is set as the uniquely determined pinyin with tone, and if the unique pinyin with tone cannot be determined, the pinyin corresponding to the current Braille word is marked as a multi-syllable word; Step 12, judging whether the current current Braille word is a Chinese character string composed of multiple characters and the Chinese character string can be found in the dictionary with Chinese character-pinyin pairs, if no, executing step 13, if yes, continuing to judge: If the dictionary with Chinese character-pinyin pairs has a Chinese character word identical to the Chinese character string corresponding to the current Braille word, and the Chinese character word corresponds to a unique pinyin string with tone in the dictionary, the pinyin string with tone is taken as the pinyin corresponding to the current Braille. If there is a Chinese character word in the Chinese character-pinyin dictionary that has the same Chinese character string as the current existing braille word, and the Chinese character word corresponds to multiple different pinyin strings with tones in the dictionary, compare the pinyin string converted from the current existing braille word, and if a unique pinyin string with tone can be determined from the pinyin string, then the uniquely determined pinyin string with tone is taken as the corresponding pinyin string of the current existing braille word; if it is still impossible to determine a unique pinyin string with tone, then the same part of the multiple pinyin strings corresponding to the current existing braille word in the Chinese character-pinyin dictionary is retained, and the different parts are marked as multi-sound words; Step 13, the current existing braille word is a Chinese character string composed of multiple characters, and there is no Chinese character word in the Chinese character-pinyin dictionary that has the same Chinese character string as the current existing braille word, which is divided into the following processing contents: If there is a Chinese character word in the Chinese character-pinyin dictionary that has a part of the substring of the Chinese character string corresponding to the current existing braille word, and the matching method is unique, then the pinyin string corresponding to the current existing braille word is obtained according to the existing braille corresponding to the Chinese character substring of the Chinese character word matched in the Chinese character-pinyin dictionary; process the existing braille corresponding to each Chinese character in the remaining part of the Chinese character string corresponding to the current existing braille word to obtain the corresponding pinyin string, and connect all the pinyin strings obtained by the above processing in order to obtain the pinyin string with tone corresponding to the current existing braille word; If there is a Chinese character word in the Chinese character-pinyin dictionary that has a part of the substring of the Chinese character string corresponding to the current existing braille word, and the matching method is not unique, then the pinyin string corresponding to the entire current existing braille word is temporarily marked as matching multiple words; If there is no Chinese character word in the Chinese character-pinyin dictionary that has any substring of the Chinese character string corresponding to the current existing braille word, then match the Chinese character-pinyin dictionary according to each Chinese character of the Chinese character string corresponding to the current existing braille word to obtain the corresponding pinyin string; This step 3 includes judging whether one of the following two conditions is met, if met, manual correction of the pinyin string is required; Condition one, the current existing braille word corresponding to the dictionary matching pinyin string contains matching multiple words; Condition two, the current existing braille word corresponding to the dictionary matching pinyin string contains a multi-sound word, and the confidence of the pinyin corresponding to the Chinese character in the model predicted pinyin string is less than a preset threshold.

2. The method for constructing a Chinese character-universal Braille parallel corpus as described in claim 1, characterized in that, This step 2 includes: Divide the Chinese character text in the Chinese character-existing braille corpus with word-level correspondence into sentences according to punctuation, and then input the Chinese character text of each sentence in turn into the machine learning model to obtain the pinyin string corresponding to the Chinese character text and the probability of each pinyin.

3. A system for constructing a Chinese character-Universal Braille parallel corpus, characterized in that, It includes: A first matching module is configured to obtain a Chinese character-existing braille corpus with Chinese character text-existing braille correspondence, convert the existing braille in the Chinese character-existing braille corpus into the corresponding Chinese pinyin through a mapping table between existing braille syllables and pinyin syllables, and add tone information to the pinyin lacking tone in the converted pinyin string through a dictionary with Chinese character-pinyin correspondence to obtain the dictionary matching pinyin string corresponding to the Chinese character text in the Chinese character-existing braille corpus; The second matching module is configured to input the Chinese character text in the Chinese character-current Braille corpus into a machine learning model, and the machine learning model outputs a model predicted pinyin string corresponding to the Chinese character text and a confidence degree of a Chinese pinyin with a tone in the model predicted pinyin string. The first judgment module is configured to judge, according to the dictionary matching pinyin string, the model predicted pinyin string and the confidence degree, whether the pinyin string corresponding to each current Braille word in the Chinese character-current Braille corpus needs to be manually corrected. If yes, the dictionary matching pinyin string of the current Braille word is manually corrected to obtain a general Braille pinyin string of the current Braille word. Otherwise, the general Braille pinyin string of the current Braille word is generated by combining the dictionary matching pinyin string and the model predicted pinyin string. The conversion module is configured to convert the general Braille pinyin string into general Braille of the current Braille word by using a mapping table between the general Braille syllable and the pinyin syllable with a tone, and to construct a Chinese character-general Braille corpus. The first matching module includes: The second judgment module is configured to judge whether the current current Braille word is a single-character word. If no, the third judgment module is called. If yes, the Chinese character is found in the Chinese character-pinyin dictionary. If the search hits, the Chinese character corresponds to a unique pinyin with a tone, and the pinyin with the tone is taken as the pinyin corresponding to the current Braille word. If the search hits, the Chinese character corresponds to multiple different pinyins with a tone, and the pinyin string obtained by converting the current Braille word is compared. If the pinyin string can uniquely determine a pinyin with a tone in the dictionary, the pinyin string of the current Braille word is set as the uniquely determined pinyin with a tone. If the unique pinyin with a tone cannot be determined, the pinyin corresponding to the current Braille word is marked as a multi-syllable word. The third judgment module is configured to judge whether the current current Braille word is a Chinese character string composed of multiple characters and the Chinese character string can be found in the Chinese character-pinyin dictionary. If no, the judgment module is called. If yes, the judgment continues. If the Chinese character-pinyin dictionary has a Chinese character word that is the same as the Chinese character string corresponding to the current Braille word, and the Chinese character word corresponds to a unique pinyin string with a tone in the dictionary, the pinyin string with the tone is taken as the pinyin corresponding to the current Braille. If the Chinese character-pinyin dictionary has a Chinese character word that is the same as the Chinese character string corresponding to the current Braille word, and the Chinese character word corresponds to multiple different pinyin strings with a tone in the dictionary, the pinyin string obtained by converting the current Braille word is compared. If the pinyin string can uniquely determine a pinyin string with a tone in the dictionary, the uniquely determined pinyin string with a tone is taken as the pinyin string corresponding to the current Braille word. If the unique pinyin string with a tone still cannot be determined, the same part of the multiple pinyin strings corresponding to the current Braille word in the Chinese character-pinyin dictionary is kept, and the different part is marked as a multi-syllable word. The judgment module is configured to judge that the current current Braille word is a Chinese character string composed of multiple characters, and there is no Chinese character word in the Chinese character-pinyin dictionary that is the same as the Chinese character string corresponding to the current Braille word. The judgment module is divided into the following processing contents: If there is a Chinese character word in the Chinese character-pinyin dictionary that has a partial substring identical to the Chinese character string corresponding to the current braille word, and the matching mode is unique, then the pinyin string corresponding to the current braille word is obtained by matching the Chinese character string corresponding to the current braille word to the Chinese character-pinyin dictionary, and the pinyin string corresponding to the current braille word is obtained by processing the current braille corresponding to each Chinese character in the remaining part of the Chinese character string corresponding to the current braille word, and then connecting all the pinyin strings obtained by processing in the above manner in order to obtain the pinyin string with tone corresponding to the current braille word; If there is a Chinese character word in the Chinese character-pinyin dictionary that has a partial substring identical to the Chinese character string corresponding to the current braille word, and the matching mode is not unique, then the pinyin string corresponding to the current braille word is temporarily marked as a multi-word matching; If there is no Chinese character word in the Chinese character-pinyin dictionary that has any substring identical to the Chinese character string corresponding to the current braille word, then the pinyin string corresponding to the current braille word is obtained by matching each Chinese character in the Chinese character string corresponding to the current braille word to the Chinese character-pinyin dictionary. The first judging module is configured to judge whether one of the following two conditions is met, and if so, the pinyin string needs to be manually corrected. Condition one: the current braille word dictionary matching the pinyin string contains a multi-word matching. Condition two: the current braille word dictionary matching the pinyin string contains a multi-phonetic word, and the confidence of the pinyin corresponding to the Chinese character in the model predicted pinyin string is less than a preset threshold.

4. The system for constructing a Chinese character-Universal Braille correspondence corpus according to claim 3, wherein, The second matching module is configured to cut the Chinese character text in the Chinese character-current braille corpus of the word-level dictionary into sentences according to punctuation, and then input the Chinese character text of each sentence into the machine learning model in turn to obtain the pinyin string corresponding to the Chinese character text and the probability corresponding to each pinyin.

5. A storage medium for storing a program for executing the method for constructing a Chinese character-universal braille dictionary according to claim 1 or 2.

6. A client for the system for constructing a Chinese character-universal braille dictionary according to claim 3 or 4.

Images