Contextual sentence determination system, machine translation device, and learning device

Abstract

Provided are a context sentence determination system that is capable of determining an appropriate context, and a machine translation device and a learning device in which the context sentence determination system is used.　In a context sentence determination system 10, an acquisition unit 11 acquires a sentence to be translated and a plurality of context candidate sentences. A context sentence determination unit 13 determines one or several context sentences from the plurality of context candidate sentences by using the meanings of the sentence to be translated and each of the plurality of context candidate sentences. For example, the context sentence determination system 10 comprises a context evaluation value calculation unit 12 as a degree-of-similarity derivation unit. The context evaluation value calculation unit 12 derives a context evaluation value considering the sentence to be translated, the plurality of context candidate sentences, and the meanings of each of these. The context evaluation value indicates the degree of similarity between sentences. The context sentence determination unit 13 determines one or several context sentences on the basis of the context evaluation value.

Description

Technical Field

[0001] The present invention relates to a context sentence determination system, a machine translation device, and a learning device for determining a context used in machine translation.

Background Art

[0002] In machine translation, by considering the context at the time of execution, it may be possible to appropriately distinguish translations for ambiguous words. For example, when translating "He is famous scholar in this area." written in English into Japanese, considering "He teaches politics." as the context, "area" is translated as "field". Here, the context is a sentence with a strong semantic connection that has information from the surrounding sentences, such as a document or a conversation. The number of sentences before the context exists from the translation target sentence varies depending on the translation target sentence.

[0003] In the following Non-Patent Document 1, a context sentence determination model (neural network) for determining an appropriate context for a translation target sentence is described. This context sentence determination model is executed as a preprocess of translation execution. When given the translation target sentence and a plurality of context candidate sentences as inputs, an appropriate context for the translation target sentence is output. This context sentence determination model is generated by learning with the translation target sentence and the context candidate sentence as inputs and a previously determined context as the output.

[0004] In Non-Patent Document 1, as a context sentence determination method during model learning, the translation evaluation value for the translation result when each context candidate sentence is used as the context is used. The translation evaluation value is an index representing translation accuracy, and the sentence considered when outputting a translation result with a high value is determined as the context. The background for using this index for context determination is the assumption that when considering an appropriate context for the translation target sentence, a translation result with highly accurate translation discrimination can be output.

Prior Art Documents

Non-Patent Documents

[0005] [Non-Patent Document 1] Xiaomian Kang, Yang Zhao, Jiajun Zhang, Chengqing Zong, "Dynamic Context Selection for Document-level Neural Machine Translation via Reinforcement Learning," [online], November, 2020, Association for Computational Linguistics, [Retrieved April 19, 2023], Internet<URL:https: / / aclanthology.org / 2020.emnlp-main.175 / > [Overview of the project] [Problems that the invention aims to solve]

[0006] In machine translation, even if a sentence that is not actually in context is treated as context, it may still be treated as context if it results in a correct translation and a high translation evaluation score. For example, when translating the English sentence "He is a famous scholar in this area." into Japanese, even if "Good morning everyone." is considered as context, the English word "area" may be translated as "field" in Japanese, and in this case, the translation evaluation score will be high. As a result, an inappropriate sentence may be determined as context. Therefore, there is a possibility that the machine translation may not be correct.

[0007] Therefore, the present invention aims to provide a contextual sentence determination system capable of determining appropriate contextual sentences during machine translation, a machine translation device utilizing this system, and a learning device. [Means for solving the problem]

[0008] The contextual sentence determination system of the present invention comprises an acquisition unit that acquires a sentence to be translated and a plurality of contextual candidate sentences, and a determination unit that uses the meaning of the sentence to be translated and each of the plurality of contextual candidate sentences to determine one or more contextual sentences from the plurality of contextual candidate sentences. [Effects of the Invention]

[0009] According to the present invention, it is possible to determine the appropriate contextual sentence. [Brief explanation of the drawing]

[0010] [Figure 1] This is an overview diagram illustrating the processing concept of the contextual sentence determination method. [Figure 2] This is a functional block diagram showing the functional configuration of the contextual sentence determination system 10 of this disclosure. [Figure 3] This is a flowchart showing the processing flow performed by the contextual sentence determination system 10 of Embodiment 1. [Figure 4] This is a schematic diagram of a contextual sentence determination method using a word semantic similarity calculation model 12a as the contextual evaluation value calculation unit 12. [Figure 5] This figure shows how BERT extracts semantic vectors from each word vector in the word semantic similarity calculation model 12a. [Figure 6] This figure shows how the cosine similarity of each semantic vector is calculated in the word semantic similarity calculation model 12a. [Figure 7] This figure shows how word semantic similarity is calculated in the word semantic similarity calculation model 12a, based on the cosine similarity and IDF value of each semantic vector. [Figure 8] This figure shows the similarity between each semantic vector output from BERT in the word semantic similarity calculation model 12a. [Figure 9] This figure shows how word semantic similarity is calculated in the word semantic similarity calculation model 12a, based on the cosine similarity of each semantic vector. [Figure 10]It is a schematic diagram showing the processing concept of a context sentence determination method using the semantic similarity of the entire sentences of the sentence to be translated and the context candidate sentences. [Figure 11] It is a schematic diagram showing the processing concept of a context sentence determination method using coreference resolution. [Figure 12] It is a functional block diagram showing the functional configuration of the context sentence determination system 10c of Embodiment 4. [Figure 13] It is a diagram showing a coreference relationship. [Figure 14] It is a diagram showing an overview of the entire machine translation system. [Figure 15] It is a functional block diagram showing the functional configuration of the machine translation device 20 of Embodiment 5. [Figure 16] It is a flowchart showing the flow of processing performed by the machine translation device 20 and the context sentence determination system 10 according to Embodiment 5. [Figure 17] It is a diagram showing an overview of the entire context sentence determination model generation method. [Figure 18] It is a schematic diagram showing translation execution processing. [Figure 19] It is a functional block diagram showing the functional configuration of the context sentence determination model learning device 30. [Figure 20] It is a flowchart showing the flow of learning processing of the context sentence determination model learning device 30 according to Embodiment 6. [Figure 21] It is a block diagram showing the functional configuration of the context sentence determination system 10d. [Figure 22] It is a diagram showing an example of the hardware configuration of the context sentence determination system 10, the machine translation device 20, and the context sentence determination model learning device 30 according to an embodiment of the present disclosure.

Embodiments for Carrying Out the Invention

[0011] Embodiments of the present disclosure will be described with reference to the accompanying drawings. Where possible, the same parts are denoted by the same reference numerals and duplicate descriptions are omitted.

[0012] [Embodiment 1] In one embodiment of this disclosure, a contextual sentence determination device will be described with reference to Figures 1 to 3. Figure 1 is a schematic diagram showing the processing concept of the contextual sentence determination method. Figure 1(a) is a diagram showing the processing overview of the contextual evaluation value calculation unit 12. In this disclosure (including embodiments 2 to 6), the sentence to be translated is in English and is assumed to be translated into Japanese, but it is not limited to these languages.

[0013] Figure 1(a) shows that when "He is famous scholar in this area." is the target sentence for translation and "He teaches politics in his school." is the candidate contextual sentence, both sentences are input to the contextual evaluation value calculation unit 12, and a contextual evaluation value of 37.88 is output. This contextual evaluation value is shown as a percentage, and the word similarity described later is also expressed in the same way.

[0014] Figure 1(b) is a correspondence table showing the relationship between the sentence to be translated, the candidate contextual sentence, the contextual evaluation value, and the decision result indicating whether or not it is contextual. As shown in the figure, the sentence to be translated is the sentence to be translated. The candidate contextual sentence is a candidate sentence that will be referenced when the sentence to be translated is translated.

[0015] The contextual evaluation values ​​in Figure 1(b) are values ​​output from the contextual evaluation value calculation unit 12. Each contextual candidate sentence is associated with its corresponding contextual evaluation value.

[0016] Figure 1(c) shows how five candidate contextual sentences for the sentence to be translated and the sentence to be translated are input to the contextual sentence determination system 10 to determine one or more contextual sentences. One or more contextual sentences are sentences that will be referenced during machine translation. The contextual sentence determination system 10 (including the contextual evaluation value calculation unit 12) generates the above correspondence table 12x and determines one contextual sentence based on the contextual evaluation value of each candidate contextual sentence. This correspondence table 12x is generated by the contextual evaluation value calculation unit 12. In this disclosure, "He teaches politics in his school." which has the highest contextual evaluation value, is determined as the contextual sentence to be translated. This makes it possible to determine an appropriate contextual sentence for one sentence to be translated. Note that it is not limited to determining one contextual sentence with the highest contextual evaluation value, but multiple contextual sentences with a value above a threshold may also be determined.

[0017] Figure 2 is a functional block diagram showing the functional configuration of the contextual sentence determination system 10 of this disclosure. The contextual sentence determination system 10 is a system for executing the processes shown in Figure 1. This contextual sentence determination system 10 is composed of an acquisition unit 11, a contextual evaluation value calculation unit 12, a contextual sentence determination unit 13, and an output unit 14.

[0018] The acquisition unit 11 is responsible for acquiring the sentence to be translated and the contextual candidate sentences. The sentence to be translated and the contextual candidate sentences are stored in advance in a machine translation target sentence DB (not shown), and a predetermined number of sentences are acquired as contextual candidate sentences by working backward from the sentence to be translated. In this disclosure, five sentences are used, but this is not the only option; there may be six or more sentences, or four or fewer sentences.

[0019] The context evaluation value calculation unit 12 is the part that calculates evaluation values ​​between the sentence to be translated and the candidate context sentences. It calculates the strength of the semantic connection between the two sentences (hereinafter referred to as the context evaluation value) and outputs the context evaluation value to the correspondence table 12x between the sentence to be translated and the candidate context sentences. This embodiment 1 shows the conceptual processing. The specific processing of the context evaluation value calculation unit 12 is shown in embodiments 2 to 4.

[0020] The contextual sentence determination unit 13 is the part that determines the contextual sentence based on the contextual evaluation values ​​described in the correspondence table 12x.

[0021] The output unit 14 is the part that outputs the determined contextual sentence as the determination result. For example, the output unit 14 outputs the sentence to be translated and the contextual sentence in association to a training database or to the machine translation unit.

[0022] As shown in the figure, the contextual sentence determination system 10 acquires the sentence to be translated in language A and candidate contextual sentences in language A, and outputs the context of the sentence to be translated. In this disclosure, language A is English and language B is Japanese, but the system is not limited to these and can be applied to other languages ​​as well.

[0023] Figure 3 is a flowchart showing the processing flow performed by the contextual sentence determination system 10 of Embodiment 1.

[0024] The acquisition unit 11 of the contextual sentence determination system 10 acquires the sentence to be translated (language A) and the candidate contextual sentences (language A) (S101). The sentence to be translated is the sentence to be translated at the time of translation execution, and the candidate contextual sentences represent multiple sentences that can serve as contextual sentences for the sentence to be translated. In Figure 1, the sentence to be translated is "He is famous scholar in this area.", and the candidate contextual sentences are a total of 5 sentences such as "Good evening everyone." and "Welcome to the seminar."

[0025] In the contextual sentence determination system 10, the contextual evaluation value calculation unit 12 takes the sentence to be translated and each of the candidate contextual sentences as input, calculates the strength of the semantic connection between the two sentences (hereinafter referred to as the contextual evaluation value), and outputs it to the contextual sentence determination unit 13 (S102). For example, in Figure 1, the contextual evaluation value calculation unit 12 calculates and outputs a contextual evaluation value of 37.88 for the candidate contextual sentence "He teaches politics in his school." and the sentence to be translated "He is famous scholar in this area.". The specific method for calculating the contextual evaluation value will be described later in Embodiments 2, 3, and 4.

[0026] Then, the context evaluation value calculation unit 12 outputs context evaluation values ​​for each of the candidate context sentences and the sentence to be translated, and generates a correspondence table 12x.

[0027] The contextual sentence determination unit 13 determines a sentence to be used as the context based on the contextual evaluation value (S103). For example, it determines the context with the highest contextual evaluation value. For example, in Figure 1, the contextual sentence determination unit 13 determines "He teaches politics in his school." which has the highest contextual evaluation value, as the contextual sentence. Note that the number of contextual sentences to be determined does not have to be limited to one sentence; it may be the top two sentences with the highest contextual evaluation values, or sentences with contextual evaluation values ​​above a set threshold, etc.

[0028] In the contextual sentence determination system 10, the output unit 14 outputs the determined contextual sentence as the determination result.

[0029] The operation and effects of the contextual sentence determination system 10 in Embodiment 1, configured in this manner, will now be explained. According to Embodiment 1, the possibility that inappropriate sentences will be determined as contextual sentences, which was a problem with conventional contextual sentence determination methods, is eliminated, and sentences that truly resemble contextual sentences can be provided as contextual sentences. Therefore, by applying the contextual sentence determination system 10 before translation of contextual data or during training of the contextual sentence determination model 30b described later, appropriate contextual sentences can be presented to the user.

[0030] [Embodiment 2] Next, Embodiment 2 of the present disclosure will be described. Embodiment 2 of the present disclosure differs from Embodiment 1 in that, in the context evaluation value calculation unit 12 in Figure 2 and step S102 in Figure 3, the similarity of the meanings of the words contained in the sentence to be translated and the candidate context sentence are specifically used as the context evaluation value.

[0031] The following will mainly describe the differences in configuration and operation of Embodiment 1 described above. Components identical to those in Embodiment 1 will be denoted by the same reference numerals, and their descriptions will be omitted as appropriate.

[0032] In step S103, the contextual sentence determination system 10 functions as a word semantic similarity calculation model 12a in the contextual evaluation value calculation unit 12, and calculates the semantic similarity of words between sentences as a contextual evaluation value. The similarity is high when there are many words with similar meanings between the two sentences.

[0033] Figure 4 is a schematic diagram of a contextual sentence determination method using a word-semantic similarity calculation model 12a as the contextual evaluation value calculation unit 12. In Figure 4(a), the sentence to be translated, "He is famous scholar in this area." and the candidate contextual sentence, "He teaches politics in his school." are input to the word-semantic similarity calculation model 12a, and a contextual evaluation value of 37.88 is output. This word-semantic similarity calculation model 12a is a calculation model that calculates the semantic similarity of words contained in a sentence.

[0034] For example, the target sentence "He is famous scholar in this area." and the candidate context sentence "He teaches politics in his school." both contain words that can be used in an academic context, such as "scholar," "teaches," and "school," as well as the word "He," which is common to both sentences. The word-semantic similarity calculation model 12a outputs a context evaluation value of 37.88 due to the presence of these words. In other words, the word-semantic similarity calculation model 12a outputs a high similarity between two sentences that contain many words used in the same field.

[0035] The word-meaning similarity calculation model 12a employs the BERTScore, a representative method, to output evaluation values. The reason for using this index as a contextual evaluation value is the assumption that in sentences with strong contextual connections, the meanings of the words included are similar.

[0036] Figure 4(b) is a correspondence table 12x that associates the target sentence for translation, candidate context sentences, word similarity scores output by the word semantic similarity calculation model 12a, and the determination result of whether or not a sentence is a context sentence. The context sentence determination unit 13 determines whether or not each candidate context sentence is a context for the target sentence based on its word similarity score (for example, the similarity score with the highest value).

[0037] Figure 4(c) shows that the contextual sentence determination system 10 takes the sentence to be translated and the contextual syntax as input and outputs one contextual sentence. As mentioned above, multiple contextual sentences may be determined and output.

[0038] The operation and effects of the contextual sentence determination system 10 in this second embodiment will be described. According to the second embodiment, if a sentence that truly resembles a context contains many words that have similar meanings to the sentence to be translated, the system can correctly determine it as a context.

[0039] Figure 5 shows the details of the word similarity calculation procedure. The word semantic similarity calculation model 12a uses BERT (Bidirectional Encoder Representations from Transformers). BERT is a natural language processing technique that is primarily used to extract the meaning of text. The BERTSocre mentioned above is an index obtained using the semantic vectors of words extracted by BERT, and is an index used to measure the similarity between two documents.

[0040] As shown in the figure, the target sentence "He is famous scholar in this area." and the candidate context sentence "He teaches politics in his school." are input to BERT. BERT extracts word vectors x1, x2, etc. from the target sentence and candidate context sentence on a word-by-word basis, and converts them into semantic vectors x'1, x'2, etc., and outputs them. In Figure 5, when BERT receives "He" (feature x1) as input, it converts it into a context-aware semantic vector x'1 and outputs it. Similarly, BERT extracts features x and y from the target sentence and candidate context sentence, and outputs the semantic vectors x' and y' for each word.

[0041] Figure 6 shows how the word semantic similarity calculation model 12a calculates the cosine similarity between each semantic vector output from BERT. Figure 6(a) shows that the cosine similarity is calculated for all combinations by cross-multiplying the feature vectors. Figure 6(b) shows the cosine similarity of each word combination between the target sentence and the candidate context sentence in a matrix. The word semantic similarity calculation model 12a extracts the combination with the highest cosine similarity between the semantic vectors of each word (horizontally) in the target sentence and each word in the candidate context sentence.

[0042] In Figure 6(b), the cosine similarity between "He" in the target sentence and "He" in the candidate context sentence is 0.87, and this value is extracted as the combination with the highest score. A similar extraction of combinations is performed for all words in the target sentence.

[0043] Figure 7 is an explanatory diagram for calculating evaluation values ​​using IDF. Figure 7(a) shows the cosine similarity between each word (semantic vector) in the target sentence and the candidate context sentence, and the IDF values ​​of the words in the target sentence. Figure 7(b) is an explanatory diagram showing the calculation of evaluation values ​​using IDF values.

[0044] Here, the word semantic similarity calculation model 12a calculates the BERTScore according to Equation 1 using the similarity to the word and the IDF value determined above. i , y i x' indicates the words of the sentence to be translated and the candidate sentence for context, respectively. i , y' i This shows the meaning vector of maxx'. i T y' i This formula shows the maximum similarity. In this formula, we calculate the similarity when j (the index of the candidate context sentence) is changed one by one for i (the index of the sentence to be translated), and then find the maximum value.

number

[0045] Instead of using the IDF value shown in Equation 1 for evaluation, the average similarity score may be used as the basis.

[0046] Figure 8 shows the similarity between each semantic vector output from BERT in the word semantic similarity calculation model 12a. As shown in Figure 8(a), the word semantic similarity calculation model 12a takes one word vector from the sentence to be translated and calculates the similarity (cosine similarity) with each word in the candidate context sentence. For each word in the sentence to be translated, the word semantic similarity calculation model 12a determines the word with the highest similarity in the candidate context sentence. For example, in Figure 8(b), for "scholar" in the sentence to be translated, "school" in the candidate context sentence is determined to be the word with the highest similarity.

[0047] As shown in Figure 5, the word semantic similarity calculation model 12a sequentially extracts the semantic vectors x'2, etc., of the words in the target sentence and calculates the similarity between each of these semantic vectors x'2 and the semantic vectors y'1, etc., of the candidate context sentence. In the example in Figure 8(b), the similarity between semantic vector x'4, "scholar," and semantic vector y'6, "school," is output as a high value of 0.82.

[0048] The word-semantic similarity calculation model 12a calculates the average of the maximum similarity values ​​obtained from each word, and uses this average value as the similarity between the target sentence and the candidate contextual sentence. The contextual sentence determination unit 13 determines the contextual sentence based on this average value. Note that other statistical indicators may be used instead of the average value.

[0049] Figure 9 shows the detailed calculation procedure for obtaining the evaluation value using the average value explained in Figure 8 above. Figure 9(a) shows the cosine similarity between each word (semantic vector) in the target sentence and the candidate context sentence. Figure 9(b) is an explanatory diagram showing the calculation of the evaluation value using the average value. Here, the IDF value as in Figure 7 is not used, and as shown in Equation 2, only the average of the maximum similarity is calculated.

number

[0050] The word semantic similarity calculation model 12a calculates the average of the maximum similarity values ​​obtained from each word. The calculation formula is shown in Figure 8.

[0051] Using IDF values ​​allows for a more accurate calculation of similarity, capturing the characteristics of the text better. For example, the importance of "scholar" and "school" can be increased, while the importance of words like "he," "in," and "is" can be decreased.

[0052] The contextual sentence determination unit 13 determines the context of the sentence to be translated based on the similarity calculated by the above method.

[0053] Next, the operation and effects of Embodiment 2 will be described. According to Embodiment 2, when a truly contextual sentence contains many words with similar meanings to the sentence to be translated, it can be correctly determined as a context.

[0054] [Embodiment 3] Next, Embodiment 3 of the present disclosure will be described. Embodiment 3 of the present disclosure differs from Embodiments 1 and 2 in that, in the context evaluation value calculation unit 12 and step S102, the semantic similarity of the entire sentence between the sentence to be translated and the candidate context sentence is specifically used as the context evaluation value.

[0055] The following will mainly describe the differences between the configuration and operation of Embodiments 1 and 2 described above. Components identical to those in Embodiments 1 and 2 will be denoted by the same reference numerals, and their descriptions will be omitted as appropriate.

[0056] Figure 10 is a schematic diagram illustrating the processing concept of a contextual sentence determination method that uses the semantic similarity of the entire sentence between the sentence to be translated and the candidate contextual sentence. In this figure, the contextual evaluation value calculation unit 12 of the contextual sentence determination system 10 functions as a sentence semantic similarity calculation model 12b, and calculates the semantic similarity of sentences as a contextual evaluation value. The sentence semantic similarity calculation model 12b has a SentenceBERT that vectorizes sentences, and uses this SentenceBERT to generate sentence vectors of the input sentences and outputs the similarity between sentence vectors. This similarity is output in such a way that it is higher when the meanings of the sentences are similar between the two sentences. For example, as shown in Figure 10(a), the sentence semantic similarity calculation model 12b takes the sentence to be translated "He is famous scholar in this area." and the candidate contextual sentence "He teaches politics in this area." as input, outputs a sentence semantic similarity of 39.34, and generates a correspondence table 12x.

[0057] As shown in Figure 10(b), the text semantic similarity calculation model 12b records this text semantic similarity in correspondence table 12x, associating it with the text to be translated and the candidate context text.

[0058] The contextual sentence determination unit 13 refers to this correspondence table 12x and determines that the contextual candidate sentence corresponding to the sentence with the highest semantic similarity is the contextual sentence. In Figure 10(b), the contextual sentence determination unit 13 determines that both the sentence to be translated, "He is famous scholar in this area," and the contextual candidate sentence, "He teaches politics in his school," have similar meanings as sentences that can be used in an academic context, and calculates the highest value of 39.34, determining that the contextual candidate sentence is the context.

[0059] Figure 10(c) shows that five candidate contextual sentences for the sentence to be translated are input into the contextual sentence determination system 10 to determine one or more contextual sentences. This contextual sentence determination system 10 determines one or more contextual sentences based on the semantic similarity of the sentences in the correspondence table above. In this disclosure, "He teaches politics in his school." which has the highest semantic similarity, is determined as one contextual sentence.

[0060] A representative method for calculating this evaluation value is SentenceBERT. SentenceBERT is a learning model that is trained to produce sentence vectors that are similar to each other from a pair of similar sentences. In this disclosure, this can be used to determine the similarity between two sentences. The reason for using this metric as a contextual evaluation value is the assumption that sentences with strong semantic connections in context will have similar meanings as sentences.

[0061] Next, the operation and effects of the contextual sentence determination system 10b of Embodiment 3 of this disclosure will be described. According to Embodiment 3, if a sentence that is truly contextual has a similar meaning to the sentence to be translated, it can be correctly determined as context. Therefore, an appropriate context can be presented using a more specific calculation method than Embodiment 1. In Embodiment 2, regardless of the meaning of the sentence, if it contained similar words, it could potentially be considered context, but in Embodiment 3, the meaning of the entire sentence can be considered.

[0062] [Embodiment 4] Next, Embodiment 4 of the present disclosure will be described. Embodiment 4 of the present disclosure differs from Embodiments 1, 2, and 3 in that, in step S102, the presence or absence of a word in a coreference relationship with a word in the sentence to be translated is used as the contextual evaluation value. A coreference relationship is the relationship between two noun expressions when they refer to the same discourse object. Coreference analysis, in natural language processing, is the process of grouping words and phrases that represent the same entity into the same cluster when a document is input. The input is a document, and the output is cluster information. The target is not limited to noun phrases only. In this Embodiment 4, the input is a document formed by concatenating the sentence to be translated and the contextual candidate sentence, and the output is cluster information of words in a coreference relationship within the document.

[0063] The following will mainly describe the differences in configuration and operation from the embodiments 1, 2, and 3 described above. Components identical to those in embodiments 1, 2, and 3 will be denoted by the same reference numerals, and their descriptions will be omitted as appropriate.

[0064] Figure 11 is a schematic diagram illustrating the processing concept of a contextual sentence determination method using coreference analysis. In Figure 11(a), the coreference analysis model 12c performs coreference analysis on a series of preceding contextual candidate sentences, including the sentence to be translated, and outputs whether or not there is a coreference relationship between the contextual candidate sentences and words in the sentence to be translated. Coreference analysis can extract groups of words that have a coreference relationship by inputting data from multiple consecutive sentences.

[0065] In Figure 11(a), Good morning, everyone. Welcome to the seminar. Today's speaker is Tom. He is my friend. He teaches politics in his school. He is famous scholar in this area. These six sentences will be treated as a single document and subjected to coreference analysis.

[0066] The coreference analysis model 12c then extracts groups of words that refer to the same discourse phenomenon from these five sentences. Here, the coreference analysis model 12c determines that "politics" and "Tom" are in a coreference relationship with "this area" and "He" in the target sentence, respectively, and that sentences containing them have a coreference relationship.

[0067] Figure 11(b) shows a correspondence table 12x that indicates the sentence to be translated, candidate contextual sentences, presence or absence of coreference information, and the result of determining whether or not it is a contextual sentence. The contextual sentence determination unit 13 determines whether or not it is a contextual sentence based on the presence or absence of coreference information in this correspondence table 12x.

[0068] The reason for using this index as a contextual evaluation value is the assumption that sentences with strong contextual connections will contain words that are in a coreference relationship.

[0069] Figure 11(c) shows that five candidate contextual sentences, including the sentence to be translated, are input into the contextual sentence determination system 10c to determine one contextual sentence. This contextual sentence determination system 10c determines one contextual sentence based on the presence or absence of coreference information in the correspondence table 12x. In this disclosure, among the candidate contextual sentences that are determined to have coreference information, "He teaches politics in his school." is determined to be the contextual sentence.

[0070] Furthermore, the following can be used as criteria for determining a single contextual sentence based on coreference information. For example, if the coreference results are morphologically analyzed and the target sentence contains demonstrative pronouns such as "it" and "they," or determiners such as "the" and "this," a candidate contextual sentence containing a noun phrase corresponding to those demonstrative pronouns or determiners may be determined as a single contextual sentence. In Figure 11(a), the coreference analysis model 12c actually outputs results showing that "He" in the target sentence is coreferenced not only with "Tom" but also with "He" in "He is my friend." and "He teaches politics in his school." However, by applying this method, it becomes possible to determine the contextual sentence only for sentences that include details of the demonstrative pronouns.

[0071] Furthermore, in combination with embodiments 2 and 3, word similarity (or sentence-level similarity) may also be calculated separately from the coreference results, and among the context candidate sentences for which coreference results exist, the sentence with the highest context evaluation value in relation to the sentence to be translated may be determined as a single context sentence.

[0072] Alternatively, instead of determining a single contextual sentence, all of the candidate contextual sentences that are output as having a coreference relationship with the sentence to be translated can be concatenated and combined into a single sentence. For example, taking Figure 11 as an example, "Today's speaker is Tom. He teaches politics in his school." can be determined as a single contextual sentence.

[0073] Figure 12 is a functional block diagram showing the functional configuration of the contextual sentence determination system 10c of Embodiment 4. The contextual sentence determination system 10c is a system for executing the processes shown in the figure. This contextual sentence determination system 10c is composed of an acquisition unit 11, a coreference analysis model 12c, a contextual sentence determination unit 13, and an output unit 14.

[0074] The acquisition unit 11c is responsible for acquiring multiple contextual candidate sentences and sentences to be translated. In this embodiment 4, the acquisition unit 11c acquires a series of sentences. Although not shown in Figure 11(a), an identification flag is attached to each sentence to distinguish between sentences to be translated and contextual candidate sentences. The acquisition unit 11c identifies the identification flag and passes the sentences to be translated and contextual candidate sentences to the coreference analysis model 12c. This identification flag includes information indicating that a sentence is to be translated, and in the case of contextual candidate sentences, it indicates how many sentences before the sentence to be translated (n sentences prior). For example, "Good evening, everyone." is a sentence that is 5 sentences before the sentence to be translated, so it is given an identification flag indicating n=5. "He teaches politics in his school." is a sentence that is 1 sentence before the sentence to be translated, so it is given an identification flag indicating n=1.

[0075] The coreference analysis model 12c is the part that outputs the coreference relationships between the target sentence and each candidate contextual sentence in a series of sentences. Figure 13 shows the coreference relationships between candidate contextual sentences and the target sentence. As shown in Figure 13, the coreference analysis model 12c performs coreference analysis on the sentences from "Good morning, everyone." to "He is famous scholar in this area." as a single document and outputs that "politics" and "Tom" are coreferenced with "this area" and "He" in the target sentence, respectively. The coreference analysis model 12c outputs whether or not these relationships exist in the correspondence table 12x (see Figure 11(b)).

[0076] In this disclosure, the coreference analysis model 12c analyzes the coreference relationship between the sentence to be translated and the candidate contextual sentence, and outputs cluster information of words in a coreference relationship.

[0077] The contextual sentence determination unit 13c determines a single contextual sentence based on the presence or absence of coreference information in the correspondence table 12x obtained by the coreference analysis model 12c. In this disclosure, the contextual sentence determination unit 13c determines "He teaches politics in his school." as the contextual sentence. Note that the presence or absence of coreference information may be "present" for multiple sentences. In this case, as described above, multiple candidate contextual sentences may be treated as a single sentence (for example, "Today's speaker is Tom. He teaches politics in his school."), or they may be combined with word similarity or sentence similarity as in other embodiments.

[0078] Next, the operation and effects of the contextual sentence determination system 10 of Embodiment 4 of this disclosure will be described. According to Embodiment 4, if a sentence that is truly contextual contains words that are in a coreference relationship with the words in the sentence to be translated, the effect is obtained that it can be correctly determined as context. Therefore, it is possible to present an appropriate context using a more specific calculation method than Embodiment 1. In addition, when differentiating between the translation of demonstrative pronouns such as "it" or "they," Embodiment 4 can use sentences containing the words they refer to as context, and can determine the context with higher accuracy compared to Embodiments 2 and 3.

[0079] [Embodiment 5] Next, Embodiment 5 of the present disclosure will be described. Embodiment 5 of the present disclosure relates to a machine translation device 20 that applies one of the contextual sentence determination systems 10 (including 10a, 10b, and 10c) from Embodiments 1 to 4 and determines the contextual sentence each time translation is performed, thereby translating while taking the context into consideration.

[0080] The following will mainly describe the differences between the configuration and operation of Embodiment 1 described above. Components identical to those in Embodiment 1 will be denoted by the same reference numerals, and their descriptions will be omitted as appropriate.

[0081] In this embodiment, a machine translation system applying the present invention will be described using Figures 14 to 17. Figure 14 is a diagram showing an overview of the entire machine translation system.

[0082] As shown in the figure, the contextual sentence determination system 10 takes five candidate contextual sentences as input and determines one contextual sentence that can serve as the context for the sentence to be translated. In this case, for the sentence to be translated, "He is famous scholar in this area," the candidate contextual sentence "He teaches politics in his school" is output as the contextual sentence. Note that although the contextual sentence determination system 10 is described here as a representative example, a contextual sentence determination system 10a equipped with a word semantic similarity calculation model 12a or another contextual sentence determination system 10b, etc., may also be used.

[0083] Figure 15 is a functional block diagram showing the functional configuration of the machine translation device 20 of Embodiment 5. The machine translation device 20 is composed of an acquisition unit 21, a machine translation unit 22, and an output unit 23.

[0084] The acquisition unit 21 is responsible for acquiring the sentence to be translated (language A) and the contextual sentence (language A) determined by the contextual sentence determination system 10b. The sentence to be translated is stored in the sentence to be translated storage unit (not shown) and is acquired in accordance with a contextual sentence determined by the contextual sentence determination system 10.

[0085] The machine translation unit 22 is the part that translates the target sentence by referring to a determined contextual sentence.

[0086] The output unit 23 is the part that outputs the translated result sentence (language B).

[0087] Figure 16 is a flowchart showing the processing flow performed by the machine translation device 20 and contextual sentence determination system 10 according to Embodiment 5.

[0088] The acquisition unit 11 of the context sentence determination system 10 acquires the sentence to be translated (language A) and the candidate context sentences (language A) (S201). The context sentence determination unit 13 of the context sentence determination system 10 determines a single context sentence based on the sentence to be translated (language A) and the candidate context sentences (language A) (S202). Then, the output unit 14 outputs the determined context sentence (language A) to the machine translation device 20.

[0089] The acquisition unit 21 of the machine translation device 20 acquires the sentence to be translated and a contextual sentence (language A) for the sentence to be translated from the contextual sentence determination system 10, and the machine translation unit 22 performs machine translation of the sentence to be translated, taking the contextual sentence into consideration (S203). Then, the output unit 23 outputs the translation result (language B) (S204).

[0090] Next, the operation and effects of Embodiment 5 will be described. According to Embodiment 5, the effect is obtained that machine translation can take appropriate context into consideration. Therefore, the user can be presented with a correct translation result that takes appropriate context into consideration and differentiates between words with ambiguous meanings, compared to the results of a normal machine translation.

[0091] [Embodiment 6] Next, Embodiment 6 of the present disclosure will be described. Embodiment 6 of the present disclosure differs from Embodiment 5 in that it uses a contextual sentence determination model (neural network) that has been trained using data on which contextual sentences were determined in step S202 of Figure 16 above. The contextual sentence determination model is a classification model that takes the sentence to be translated and the candidate contextual sentences as inputs and outputs the context determined from the contextual evaluation value.

[0092] In Embodiment 6, this contextual sentence determination model is used to determine a single contextual sentence from the sentence to be translated and multiple candidate contextual sentences, which differs from Embodiment 5 in that a contextual evaluation value is calculated and the contextual sentence is determined each time translation is performed.

[0093] The following will mainly describe the differences in configuration and operation from Embodiments 1 and 5 described above. Components identical to those in Embodiments 1 and 5 will be denoted by the same reference numerals, and their descriptions will be omitted as appropriate.

[0094] In this embodiment, the method for generating a contextual sentence determination model will be explained using Figures 17 to 19.

[0095] Figure 17 shows an overview of the entire method for generating a contextual sentence determination model. Figure 17(a) shows the collection of training data for generating the contextual sentence determination model, and Figure 17(b) shows the training method using the training database.

[0096] As shown in Figure 17(a), the contextual sentence is determined using the contextual sentence determination system 10 shown in Embodiments 1 to 4, and the sentence to be translated at that time, multiple contextual candidate sentences, and the contextual sentence determined therefrom are stored in the learning DB 30a.

[0097] As shown in Figure 17(b), the contextual sentence determination model learning device 30 uses the sentences to be translated and the candidate contextual sentences stored in the learning DB 30a as explanatory variables, and the determined contextual sentence as the target variable, and learns using known machine learning methods to generate a contextual sentence determination model 30b.

[0098] Figure 18 is a schematic diagram illustrating the translation execution process. As shown in the figure, the contextual sentence determination model 30b, upon input of multiple contextual candidate sentences and a sentence to be translated, outputs a single contextual sentence determined from the contextual candidate sentences. For the sake of explanation in this disclosure, the same sentences used during the training process are used, but it is naturally applicable to different sentences as well.

[0099] The machine translation device 20 acquires the contextual sentence output from the contextual sentence determination model 30b and translates the sentence to be translated by referring to it. In this disclosure, the machine translation device 20 acquires the contextual sentence "He teaches politics in his school." and the sentence to be translated "He is famous scholar in this area." and performs machine translation. In this disclosure, the Japanese translation result can be "He is a famous scholar in this field." That is, the English "in this area" can be translated as "in this field" in Japanese, rather than "in this region". When translating the English word "area" into Japanese, it can mean either "region" or "field," and it is necessary to determine which meaning is intended depending on the context.

[0100] Figure 19 is a functional block diagram showing the functional configuration of the contextual sentence determination model learning device 30. The contextual sentence determination model learning device 30 consists of an acquisition unit 31, a learning unit 32, and a contextual sentence determination model 30b.

[0101] The acquisition unit 31 retrieves training data from the training DB 30a. In this disclosure, the training target sentences and multiple training context candidate sentences are used as explanatory variables, and the determined training context sentence is acquired as the target variable. These training target sentences, training context candidate sentences, and training context sentences (collectively referred to as training data) are based, for example, on the training target sentences, context candidate sentences, and context sentences obtained by embodiments 1 to 4.

[0102] The learning unit 32 is the part that generates a contextual sentence decision model 30b by performing machine learning based on the acquired training data. Specifically, the learning unit 33 generates a contextual sentence decision model 30b by performing machine learning with the target sentence (language A) and the candidate contextual sentence (language A) as explanatory variables and the corresponding contextual sentence as the target variable.

[0103] The contextual sentence determination model 30b is a machine learning model that, when given a sentence to be translated and multiple candidate contextual sentences as input, outputs a single contextual sentence.

[0104] Figure 20 is a flowchart showing the learning process flow of the contextual sentence determination model learning device 30 according to Embodiment 6.

[0105] The acquisition unit 31 acquires the sentence to be translated (language A), the candidate context sentences (language A), and a corresponding context sentence from the learning DB 30a (S301). The learning unit 33 uses the sentence to be translated (language A) and the candidate context sentences (language A) as explanatory variables and the corresponding context sentence as the target variable, and performs machine learning to generate a context sentence decision model 30b (S302).

[0106] By using this contextual sentence determination model 30b in the contextual sentence determination unit 13 of S202 in Figure 2, it is possible to obtain machine translation results that take appropriate context into consideration. Figure 21 is a block diagram showing the functional configuration of the contextual sentence determination system 10d. This contextual sentence determination system 10d consists of an acquisition unit 11d, a contextual sentence determination model 30b, and an output unit 14. The acquisition unit 11d is the part that acquires the sentence to be translated and multiple contextual candidate sentences. When the contextual sentence determination model 30b receives these sentences to be translated and multiple contextual candidate sentences as input, it outputs one or more appropriate contextual sentences that take meaning into consideration. The output unit 14 outputs one or more contextual sentences.

[0107] Next, the operation and effects of Embodiment 6 will be described. According to Embodiment 6, by using a pre-trained model, context can be acquired at a faster speed compared to context acquisition during translation execution in Embodiment 5. Therefore, in situations where high processing speed is required, such as real-time translation methods, the delay during translation execution can be reduced, and the user can be presented with a correct translation result that takes the appropriate context into consideration.

[0108] [Regarding the effects of the contextual sentence determination system, machine translation device, and learning device described herein] In the contextual sentence determination system 10 of the present disclosure as shown in Embodiment 1, the acquisition unit 11 acquires the sentence to be translated and a plurality of contextual candidate sentences. The contextual sentence determination unit 13 then uses the meaning of the sentence to be translated and each of the plurality of contextual candidate sentences to determine one or more contextual sentences from the plurality of contextual candidate sentences.

[0109] For example, in Embodiment 1, the contextual sentence determination system 10 includes a contextual evaluation value calculation unit 12 as a similarity derivation unit. This contextual evaluation value calculation unit 12 derives a contextual evaluation value that takes into account the sentence to be translated, a plurality of contextual candidate sentences, and their respective meanings. This contextual evaluation value indicates the similarity between sentences. The contextual sentence determination unit 13 determines one or more contextual sentences based on this contextual evaluation value.

[0110] In Embodiment 2, the context evaluation value calculation unit 12 functions as a word semantic similarity calculation model 12a. This word semantic similarity calculation model 12a outputs multiple word similarities between a word contained in the sentence to be translated and a word contained in each of the multiple candidate context sentences.

[0111] More specifically, the word semantic similarity calculation model 12a derives the inter-word similarity for all combinations of words contained in the sentence to be translated and words contained in each of the multiple candidate context sentences, and outputs the average value of these similarities as the aforementioned similarity.

[0112] Furthermore, the similarity of the entire text may be determined without examining the similarity at the word level. For example, as shown in Embodiment 3, the similarity derivation unit may function as a text-semantic similarity calculation model 12b. This text-semantic similarity calculation model 12b outputs the inter-sentence similarity of the sentence-level semantics in the target sentence and the multiple candidate context sentences.

[0113] These similarity calculations are performed using vectors. Specifically, the word-semantic similarity calculation model 12a or the sentence-semantic similarity calculation model 12b derives semantic vectors of the sentences to be translated and the candidate context sentences, or of the words contained in those sentences, and derives similarity based on these semantic vectors.

[0114] In this way, the contextual sentences to be referenced when machine translating the target text can be determined based on the similarity between sentences or words. The contextual sentences determined based on the similarity of sentences or words are contextual sentences that take the meaning of the sentence into consideration. This is natural when determining context and allows for the determination of appropriate contextual sentences.

[0115] Another method involves processing based on coreference relationships, as described in Embodiment 4. In Embodiment 4, the contextual sentence determination system 10c includes a coreference analysis model 12c that acquires coreference relationships between the sentence to be translated and a plurality of candidate contextual sentences. The contextual sentence determination unit 13c determines one or more contextual sentences based on these coreference relationships.

[0116] In this way, if a truly contextual sentence contains words that are in a coreference relationship with the words in the sentence to be translated, the effect is obtained that it can be correctly determined as the context. Therefore, an appropriate context can be presented using a more specific calculation method than in Embodiment 1.

[0117] Furthermore, by using a contextual determination model that has been pre-trained on the contextual sentences determined as described above, it is possible to determine the appropriate contextual sentence in the same manner as described above.

[0118] In other words, the contextual sentence determination system 10d in Embodiment 6 includes a contextual sentence determination model 30b, and when a sentence to be translated and a plurality of candidate contextual sentences are input, it outputs one or more contextual sentences. This contextual sentence determination model 30b is trained using one or more training contextual sentences determined using the meaning of the training sentence to be translated and each of the plurality of candidate training contextual sentences. In this case, the training contextual sentences determined are those obtained by the contextual sentence determination system 10 described in Embodiments 1 to 4.

[0119] This contextual sentence determination model 30b is trained using the sentence to be translated and several candidate contextual sentences as explanatory variables, and the training contextual sentence as the target variable.

[0120] In other words, the contextual sentence determination model learning device 30 includes a learning DB 30a (learning data storage unit) that stores the learning contextual sentences, learning context candidate sentences, and learning translation target sentences acquired in embodiments 1 to 4. The learning unit 32 then generates a contextual sentence determination model 30b using machine learning, with the learning translation target sentences and learning context candidate sentences in the learning data as explanatory variables and the learning context as the target variable.

[0121] In this way, the contextual sentence determination model 30b is generated, enabling appropriate determination of contextual sentences while considering their meaning.

[0122] Embodiment 5 shows a machine translation device 20 that performs machine translation using the context and the sentence to be translated determined by the contextual sentence determination systems 10, 10a, and 10d in Embodiments 1 to 4.

[0123] In this machine translation device 20, the acquisition unit 21 acquires the text to be translated and the determined contextual text. Then, the machine translation unit 22 translates the text based on the contextual text using machine translation.

[0124] This allows for translation based on appropriate contextual sentences. By avoiding the use of meaningless or inappropriate contextual sentences, more accurate translations become possible.

[0125] The contextual sentence determination system, machine translation device, and learning device of this disclosure comprise the following configurations.

[0126] [1] A unit that retrieves the sentence to be translated and multiple contextual candidate sentences, A determination unit that determines one or more contextual sentences from the multiple contextual candidate sentences using the meaning of the sentence to be translated and each of the multiple contextual candidate sentences, A contextual sentence determination system equipped with the following features.

[0127] [2] The system further comprises a similarity derivation unit that derives a similarity score considering the meaning of the sentence to be translated, the plurality of candidate contextual sentences, and each of them. The determination unit determines one or more contextual sentences based on the similarity. [1] The contextual sentence determination system described.

[0128] [3] The similarity derivation unit is, Multiple word similarity scores are derived between the words contained in the sentence to be translated and the words contained in each of the multiple candidate context sentences. [2] The contextual sentence determination system described.

[0129] [4] The similarity derivation unit is, The system derives the word-to-word similarity for all combinations of words in the sentence to be translated and words in each of the multiple candidate context sentences, and derives the average value of these similarities as the overall similarity. [3] The contextual sentence determination system described.

[0130] [5] The similarity derivation unit is, The system derives semantic vectors from the sentences to be translated and the candidate context sentences, or from the words contained in those sentences, and derives similarity based on those semantic vectors. A contextual sentence determination system described in any one of [2] to [4].

[0131] [6] The system further includes a coreference analysis unit that acquires coreference relationships between the sentence to be translated and the plurality of candidate contextual sentences. The determination unit determines one or more contextual sentences based on the coreference relationship. [1] The contextual sentence determination system described.

[0132] [7] The determination unit is a contextual sentence determination model that takes the sentence to be translated and the plurality of candidate contextual sentences as input and outputs one or more contextual sentences, The aforementioned contextual sentence determination model is, The learning process uses one or more learning context sentences determined using the meanings of the target sentence for translation and each of several candidate learning context sentences. [1] The contextual sentence determination system described.

[0133] [8] The aforementioned contextual sentence determination model is, The following is learned with the aforementioned sentence to be translated and the aforementioned multiple candidate context sentences as explanatory variables, and the aforementioned training context sentence as the target variable. [7] The contextual sentence determination system described.

[0134] [9] A translation acquisition unit that acquires the contextual sentence and the sentence to be translated determined in the contextual sentence determination system described in any one of [1] to [8], A machine translation unit that translates the aforementioned translated text by machine translation based on the aforementioned context, A machine translation device equipped with the following features.

[0135]

[10] A learning data storage unit that stores the learning context sentence, learning context candidate sentence, and learning translation target sentence determined in any one of the context sentence determination systems described in [1] to [6], A learning unit generates a contextual sentence determination model by machine learning, using the training translation target sentence and the training context candidate sentence in the training data as explanatory variables and the training context sentence as the target variable. A learning device equipped with the following features.

[0136] The block diagram used in the description of the above embodiment shows functional units. These functional blocks (components) are realized by any combination of at least one of hardware and software. Furthermore, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one device that is physically or logically coupled, or it may be realized using two or more physically or logically separated devices that are directly or indirectly connected (for example, using wired or wireless connections). A functional block may be realized by combining the above one device or the above multiple devices with software.

[0137] Functions include, but are not limited to, judgment, decision, judgment, calculation, calculation, processing, derivation, investigation, exploration, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, assumption, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating (mapping), and assigning. For example, a functional block (configuration part) that enables transmission is called a transmitting unit or transmitter. As mentioned above, the method of implementation is not particularly limited.

[0138] For example, the contextual sentence determination system 10, machine translation device 20, and contextual sentence determination model learning device 30 in one embodiment of the present disclosure may function as a computer that processes the contextual sentence determination method, machine translation method, or learning method of the present disclosure. Figure 22 is a diagram showing an example of the hardware configuration of the contextual sentence determination system 10, machine translation device 20, and contextual sentence determination model learning device 30 according to one embodiment of the present disclosure. The above-described contextual sentence determination system 10, machine translation device 20, and contextual sentence determination model learning device 30 may be physically configured as a computer device including a processor 1001, memory 1002, storage 1003, communication device 1004, input device 1005, output device 1006, bus 1007, etc.

[0139] In the following explanation, the term "device" can be replaced with "circuit," "device," "unit," etc. The hardware configuration of the contextual sentence determination system 10, the machine translation device 20, and the contextual sentence determination model learning device 30 may include one or more of the devices shown in the figure, or it may be configured to omit some of the devices.

[0140] Each function in the contextual sentence determination system 10, the machine translation device 20, and the contextual sentence determination model learning device 30 is realized by loading predetermined software (programs) onto hardware such as the processor 1001 and memory 1002, which allows the processor 1001 to perform calculations, control communication by the communication device 1004, and control at least one of data reading and writing in the memory 1002 and storage 1003.

[0141] The processor 1001 controls the entire computer, for example, by running the operating system. The processor 1001 may be composed of a central processing unit (CPU) that includes interfaces with peripheral devices, control devices, arithmetic units, registers, etc. For example, the context evaluation value calculation unit 12 and the context statement determination unit 13 described above may be implemented by the processor 1001.

[0142] Furthermore, the processor 1001 reads programs (program code), software modules, data, etc., from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes accordingly. The program used is one that causes the computer to execute at least a part of the operations described in the above embodiment. For example, the context evaluation value calculation unit 12 and the context statement determination unit 13 may be stored in the memory 1002 and implemented by a control program that operates on the processor 1001, and other functional blocks may be implemented similarly. The above-described various processes have been explained as being executed by one processor 1001, but they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may also be transmitted from a network via a telecommunications line.

[0143] Memory 1002 is a computer-readable recording medium and may consist of at least one of the following: ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), etc. Memory 1002 may also be called a register, cache, main memory, etc. Memory 1002 can store executable programs (program code), software modules, etc., for implementing a context determination method, machine translation method, or learning method according to one embodiment of the present disclosure.

[0144] Storage 1003 is a computer-readable recording medium and may consist of at least one of the following: an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disc, a digital multipurpose disc, a Blu-ray® disc), a smart card, flash memory (e.g., a card, a stick, a key drive), a floppy® disk, a magnetic strip, etc. Storage 1003 may also be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, server, or other suitable medium including at least one of memory 1002 and storage 1003.

[0145] The communication device 1004 is hardware (transceiver / receiver device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as a network device, network controller, network card, communication module, etc. The communication device 1004 may be configured to include, for example, a high-frequency switch, duplexer, filter, frequency synthesizer, etc., in order to implement at least one of frequency division duplex (FDD) and time division duplex (TDD).

[0146] The input device 1005 is an input device that accepts input from an external source (e.g., a keyboard, mouse, microphone, switch, button, sensor, etc.). The output device 1006 is an output device that outputs to an external source (e.g., a display, speaker, LED lamp, etc.). The input device 1005 and the output device 1006 may be configured as an integrated unit (e.g., a touch panel).

[0147] Furthermore, each device, such as the processor 1001 and memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or different buses may be configured for each device.

[0148] Furthermore, the contextual sentence determination system 10, the machine translation device 20, and the contextual sentence determination model learning device 30 may include hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array), and some or all of each functional block may be realized by such hardware. For example, the processor 1001 may be implemented using at least one of these hardware components.

[0149] Information notification is not limited to the embodiments described herein and may be carried out by other means. For example, information notification may be carried out by physical layer signaling (e.g., DCI (Downlink Control Information), UCI (Uplink Control Information)), upper layer signaling (e.g., RRC (Radio Resource Control) signaling, MAC (Medium Access Control) signaling, broadcast information (MIB (Master Information Block), SIB (System Information Block))), other signals, or combinations thereof. RRC signaling may also be called RRC messages, and may be, for example, RRC Connection Setup messages, RRC Connection Reconfiguration messages, etc.

[0150] The processing procedures, sequences, flowcharts, etc., of each aspect / embodiment described herein may be reordered, provided they are consistent with each other. For example, the methods described herein present various step elements in an exemplary order and are not limited to that specific order.

[0151] Input and output information may be stored in a specific location (e.g., memory) or managed using a management table. Input and output information may be overwritten, updated, or appended to. Output information may be deleted. Input information may be transmitted to other devices.

[0152] The determination may be made by a value represented by 1 bit (0 or 1), by a boolean value (true or false), or by a numerical comparison (for example, a comparison with a predetermined value).

[0153] Each aspect / embodiment described herein may be used individually, in combination, or switched between as needed during implementation. Furthermore, notification of specific information (e.g., notification that "X is") is not limited to explicit notification, but may also be implicit (e.g., by not providing such notification).

[0154] Although the present disclosure has been described in detail above, it will be clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the intent and scope of the present disclosure as defined by the claims. Accordingly, the descriptions in the present disclosure are illustrative and not intended to be restrictive in any way.

[0155] Software should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, and so on, whether they are called software, firmware, middleware, microcode, hardware description languages, or by any other name.

[0156] Furthermore, software, instructions, information, etc., may be transmitted and received via a transmission medium. For example, if software is transmitted from a website, server, or other remote source using at least one of wired technologies (such as coaxial cable, fiber optic cable, twisted pair, or digital subscriber line (DSL)) and wireless technologies (such as infrared or microwave), then at least one of these wired and wireless technologies is included in the definition of a transmission medium.

[0157] The information, signals, etc. described in this disclosure may be represented using any of the various different techniques. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0158] In addition, terms used in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings. For example, at least one of the channel and symbol may be a signal (signaling). Also, a signal may be a message. Furthermore, a component carrier (CC) may be called a carrier frequency, cell, frequency carrier, etc.

[0159] Furthermore, the information, parameters, etc., described in this disclosure may be expressed using absolute values, relative values ​​from a given value, or other corresponding information. For example, wireless resources may be indicated by an index.

[0160] The names used for the parameters described above are not restrictive in any way. Furthermore, the formulas and other expressions using these parameters may differ from those expressly disclosed in this disclosure. Various channels (e.g., PUCCH, PDCCH, etc.) and information elements can be identified by any suitable name, and therefore, the various names assigned to these various channels and information elements are not restrictive in any way.

[0161] In this disclosure, terms such as "Mobile Station (MS)," "user terminal," "User Equipment (UE)," and "terminal" may be used interchangeably.

[0162] A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other appropriate term.

[0163] As used in this disclosure, the terms “determining” and “determining” may encompass a wide variety of actions. “Determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, or inquiring (e.g., searching in a table, database, or other data structure), or ascertaining. “Determining” may also include, for example, receiving (e.g., receiving information), transmitting (e.g., sending information), inputting, outputting, or accessing (e.g., accessing data in memory). Furthermore, "judgment" and "decision" can include considering something as having been "judged" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. In other words, "judgment" and "decision" can include considering something as having been "judged" or "decided" after some action. Also, "judgment (decision)" can be reinterpreted as "assuming," "expecting," or "considering."

[0164] The terms “connected,” “coupled,” or any variation thereof, mean any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” with each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, “connection” may be reinterpreted as “access.” As used in this disclosure, two elements may be considered to be “connected” or “coupled” with each other using at least one of one or more wires, cables, and printed electrical connections, and, in some non-limiting and non-exclusive examples, electromagnetic energy having wavelengths in the radio frequency domain, microwave domain, and optical (both visible and invisible) domain.

[0165] In this disclosure, the phrase "based on" does not mean "based solely on" unless otherwise specified. In other words, the phrase "based on" means both "based solely on" and "based at least on."

[0166] Any reference to elements using designations such as “first,” “second,” etc., as used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Accordingly, references to first and second elements do not imply that only two elements may be adopted, or that the first element must precede the second element in any way.

[0167] Where the terms “include,” “including,” and their variations are used in this disclosure, these terms are intended to be inclusive, as is the term “comprising.” Furthermore, the term “or” as used in this disclosure is not intended to mean exclusive OR.

[0168] In this disclosure, if articles are added by translation, such as a, an, and the in English, this disclosure may include the fact that the noun following these articles is plural.

[0169] In this disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "combine" may be interpreted similarly to "different." [Explanation of Symbols]

[0170] 10, 10a, 10b, 10c, 10d... Contextual sentence determination system, 11, 11c, 11d... Acquisition unit, 12... Contextual evaluation value calculation unit, 12a... Word semantic similarity calculation model, 12b... Sentence semantic similarity calculation model, 12c... Coreference analysis model, 13, 13c... Contextual sentence determination unit, 14... Output unit, 20... Machine translation device, 21... Acquisition unit, 22... Machine translation unit, 23... Output unit, 30... Contextual sentence determination model learning device, 31... Acquisition unit, 32... Learning unit, 33... Learning unit, 30b... Contextual sentence determination model.

Claims

1. A contextual sentence determination system comprising a processor and memory, The aforementioned processor, A unit that retrieves the sentence to be translated and multiple contextual candidate sentences, It functions as a decision unit that determines one or more contextual sentences from the multiple contextual candidate sentences, using the meaning of the sentence to be translated and each of the multiple contextual candidate sentences. The aforementioned determination unit, The similarity between the semantic vector of each word in the aforementioned sentence to be translated and the semantic vector of each word in the aforementioned candidate context sentence is calculated. For each word in the aforementioned sentence to be translated, the maximum similarity to the word with the highest similarity among the words in the aforementioned candidate context sentence is identified. The average value of the identified maximum similarity is calculated as the similarity between the sentence to be translated and the candidate context sentences, and one or more context sentences are determined based on this similarity. Contextual sentence determination system.

2. The processor further, It functions as a coreference analysis unit that performs coreference analysis on a series of sentences including the sentence to be translated and the plurality of candidate context sentences, and determines whether or not there is a coreference relationship where a word that refers to the same object as a word included in the sentence to be translated is included in the candidate context sentences. The aforementioned determination unit, Based on the similarity and the presence or absence of the coreference relationship, determine one or more contextual sentences. The contextual sentence determination system according to claim 1.

3. The determination unit is a contextual sentence determination model that takes the sentence to be translated and the plurality of candidate contextual sentences as input and outputs one or more contextual sentences, The aforementioned contextual sentence determination model is, The following is learned using the aforementioned sentence to be translated and the aforementioned multiple candidate context sentences as explanatory variables, and the training context sentence as the target variable. The contextual sentence determination system according to claim 1.

4. A translation acquisition unit that acquires the contextual sentence and the sentence to be translated determined in the contextual sentence determination system described in claim 1, A machine translation unit that performs machine translation of the target sentence by referring to the contextual sentence and determining the meaning of the words contained in the target sentence, A machine translation device equipped with the following features.

5. A learning data storage unit that stores the contextual sentence, the candidate contextual sentence, and the sentence to be translated determined in the contextual sentence determination system according to claim 1 as learning data, A learning unit generates a contextual sentence determination model by machine learning, using the sentence to be translated and the candidate contextual sentence in the training data as explanatory variables and the contextual sentence as the target variable. A learning device equipped with the following features.

Images