JP2016164707A - Automatic translation device and translation model learning device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To increase accuracy of automatic translation of long sentences that can be divided into structural components.SOLUTION: A claim translation part 328 includes: a pattern search part 500 which identifies patterns of English claims 334 after syntactic analysis and divides them into structural components; a Japanese pattern generation part 504 which identifies Japanese patterns associated with English patterns; an input sentence division part 502 and a structural component kind determination part 510 which identify correspondence and grammatical characteristics of the structural components between English and Japanese patterns; a statistical automatic translation machine 524 which uses either a noun phrase translation model 320 or a verb phrase translation model 322 which are learned to use word strings of the structural components of the English claims 334 for translation of word strings of the grammatical characteristics of the structural components and generates Japanese translation; and a Japanese pattern substitution part 526 which substitutes respective Japanese translations of the structural components of the English claims 334 for structural components of the Japanese pattern according to the correspondence of the structural components and generates Japanese claims 326.SELECTED DRAWING: Figure 9

Description

  TECHNICAL FIELD The present invention relates to automatic translation technology, and in particular, technology that handles long and difficult-to-translate sentences while following certain rules, such as the claims of a patent application claim (hereinafter referred to as “claims”). About.

  It may be necessary to know the meaning of a claim written in a foreign language in order to confirm the technical content of the patent application written in a foreign language or to obtain knowledge about the scope of rights. If you have a small number of subjects and knowledge of the language of the target document, you may not have much trouble reading the original. However, when there are a large number of target documents or knowledge of the language of the document is poor, it is unrealistic to read all the claims of the target document. In such a case, it is often performed to try to grasp the contents by translating each claim into its own language by automatic translation.

  However, it is well known that when a claim is automatically translated into another language, its quality is low. In the claims, it is customary to describe many limitations relating to the invention in one sentence. Therefore, the claim text tends to be long. In addition, in technical fields where many inventors have difficulty in development, the difference from the prior art may not be clear unless relatively many limitations (constituent requirements) are added to the claims. As a result, claims are often sentences of length that rarely occur in ordinary documents. Although automatic translation technology has been greatly developed and improved in recent years, high-precision translation is not yet expected, especially when the original translation is long, as in the case of claims.

  A proposal for solving such a problem is made in Patent Document 1 described later. The technique described in Patent Document 1 is basically based on the idea that a long claim is divided into its constituent elements (hereinafter referred to as “structural parts”) and translated for each structural part. . This takes advantage of the fact that, particularly recently, claims are generally divided into a plurality of structural parts and the structural parts are separated by a specific separation pattern. ing. The structural parts of the claims can be the subject matter of the invention, one or more components and a transitional phrase inserted between the subject matter and the components of the invention. The order of these descriptions varies depending on the language. For example, in the case of Japanese, one or more constituent elements are located at the top, followed by a transition phrase, followed by the transition phrase followed by the subject matter of the invention. The reverse is true for English, typically in the form of an inventive subject at the beginning, followed by a transition phrase, followed by one or more components.

  In the case of Japanese, “contains (characteristic)”, “comprises (characteristic)”, “consists of (characteristically)”, etc. are mainly used as transition phrases. In English, “comprising:”, “including:”, “consisting of:”, etc. are used as transition phrases.

  Between one or a plurality of constituent elements, “,” + line feed is mainly used in Japanese, and “;” is used in English.

  By detecting such a delimiter pattern in a claim, the structure of the claim can be determined.

  FIG. 1 shows a rough process flow of the technique disclosed in Patent Document 1. Referring to FIG. 1, the claim translation procedure 30 in Patent Document 1 uses a computer, and performs a morphological analysis and a syntax analysis on a claim text to generate syntax information, and a syntax In the result of the analysis, a delimiter pattern of the structural part is detected, a step 42 for dividing the text of the claim into components, a character string indicating the structural part name is specified in the text for each structural part, and the character And a step 44 of attaching a tag indicating that the column is a structural part name to the column. As a result of the parsing, a tree-like syntactic structure is formed between the structural parts. In Patent Document 1, this syntax structure is expressed in the form of a tag attached to a structural component, a word inside the structural component, and the like.

  In the first embodiment of Patent Document 1, it is assumed that the target claim is English. In the case of English, a word corresponding to the name of the structural part is often described at the beginning of the text relating to each structural part, and in most cases, an explanation regarding the structural part is described thereafter. Therefore, the process as in step 44 can be easily performed. In addition, descriptions of structural parts are often patterned. In Patent Document 1, an explanatory text that is patterned as described above is called an explanatory pattern text.

  The translation procedure 30 of Patent Document 1 further specifies an explanation pattern sentence included in the structural part to be processed, divides the structural part to be processed for each pattern sentence, and the divided part is an explanation pattern sentence. Step 46 for attaching a tag indicating the presence, Step 48 for translating the structural part name of the structural part to be processed into a target language by automatic translation, and changing the sentence structure to one suitable for translation for each explanation pattern sentence Step 50, translating each explanatory sentence whose sentence structure has been changed in step 50 into the target language 52 by automatic translation, and using the structural part name and each pattern sentence thus translated into the target language as tags. Therefore, it includes a step 54 of displaying a tree.

  The change of the sentence structure in step 50 is a change to a format in which the explanatory text includes the subject and the predicate. For example, an English related pronoun is omitted and a structural part name is inserted instead, or a noun or noun phrase that is modified by an explanatory pattern sentence is inserted. In Patent Document 1, including this, a conversion pattern is set in advance for sentence conversion. This conversion of the sentence structure seems to be to match the explanation pattern sentence to the word order of general sentences.

  Referring to FIG. 2, the automatic translation method according to Patent Document 1 performs the following processing more specifically. Assume that the English claim 70 is to be translated. In Patent Document 1, an English claim is decomposed into a plurality of structural parts by finding an English delimiter pattern, and a tree structure is formed between them (process 72). As described above, it is possible to determine the structure of the English claim by paying attention to the delimiter pattern in the English claim. According to the result, the English claim can be disassembled into a plurality of structural parts.

  In the procedure according to Patent Document 1, as a result of the processing 72, an English pattern 74 having a tree structure is formed. In the example shown in FIG. 2, the English pattern 74 includes a subject 90, a transition phrase 92, and a description 94. A Japanese pattern 78 having a tree structure is generated by automatically translating the subject 90, the transition phrase 92, and the description 94 separately (process 76). The Japanese pattern 78 has the same structure as the English pattern 74, and the text is converted into Japanese. That is, the Japanese pattern 78 includes a Japanese translation 100 of the subject, a Japanese translation 102 of the transition phrase, and a Japanese translation 104 of the explanation. In Patent Document 1, instead of generating a one sentence Japanese claim from the Japanese pattern 78, the Japanese pattern 78 is displayed as it is in a tree format.

  According to Patent Document 1, by performing such processing, it is only necessary to automatically translate each structural component, so that the translation accuracy of the explanation pattern sentence to be configured is improved from the conventional one. Furthermore, the translated text is displayed in a tree structure, which makes it easier to understand the content of the complaint.

JP2014-199476 (particularly FIGS. 5, 7, 8 and paragraphs 31, 41, 51-77)

  Certainly, the technique of Patent Document 1 seems to improve the accuracy of translation rather than translating the entire claim together. However, when automatic translation is used, there is still a problem to be solved by the technique of Patent Document 1.

  The biggest problem is the change in the sentence structure that constitutes the explanation. The change of the sentence structure is a process that is necessary because the explanatory text is not a normal sentence but a format that modifies each structural component. If a sentence structure conversion pattern is prepared in advance and the explanation pattern sentence is completely changed, there may be no problem even if each explanation sentence is translated using the existing automatic translation technology. However, sentence variations that can occur as claims are virtually infinite, and it is impossible to prepare sentence structure conversion rules in advance. Moreover, there is no clear disclosure in Patent Document 1 regarding what kind of conversion pattern should be prepared. Therefore, there is no high possibility that the explanatory text is converted into a sentence structure suitable for automatic translation. Even when such conversion is not performed correctly and the explanation is not in the form of a sentence, the automatic translation apparatus interprets each explanation as a normal sentence and performs translation. As a result, the translation results obtained are unlikely to be understandable. In particular, when the structural part is a noun phrase, if the automatic translation apparatus regards it as a sentence and translates it, the adverse effect on the translated sentence increases.

  For example, referring to FIG. 3, consider an English claim 120 having two noun phrases 1 and 2 corresponding to explanatory text. According to Japanese claim 122 obtained by automatic translation by the present inventor according to Patent Document 1, translation for noun phrase 1 and noun phrase 2 is hardly understood. As shown in FIG. 3, in the case where each noun phrase or the like is translated as a sentence even though the sentence structure is not converted successfully when processed according to Patent Document 1, the obtained translated sentence Is difficult to understand and the purpose of automatic translation cannot be fulfilled.

  In addition, such problems arise not only with respect to the translation of claims of patent applications, but also with the translation of sentences composed of a plurality of structural parts, which are similarly long and described according to specific rules. It is. For example, such problems may occur in laws, regulations, treaties, treaties, and instructions for using various machines, electronic devices, and software.

  Accordingly, an object of the present invention is to provide an automatic translation apparatus capable of improving the accuracy of automatic translation for a long sentence that can be divided into a plurality of structural parts described according to a specific format.

  The automatic translation apparatus according to the first aspect of the present invention translates a sentence in a first language into a sentence in a second language different from the first language. The automatic translation apparatus specifies a sentence pattern of a first language, divides a sentence of the first language into structural parts, a sentence pattern of the first language identified by the dividing means, Pattern specification means for specifying the second language sentence pattern, the first language sentence pattern divided by the dividing means, and the second language pattern specified by the pattern specification means The correspondence specifying means for specifying the correspondence between the structural parts and the grammatical characteristics of each structural part, and for each of the structural parts in the first language, the first A translation unit that performs automatic translation using a model for translation from a language to a second language to generate a translation of the second language, and each of the structural parts of the first language is obtained by the translation unit. Translation of the second language By substituting one of the structural parts of the second language sentence pattern specified by the pattern specifying means according to the correspondence relationship of the structural parts specified by the correspondence specifying means, the sentence of the first language Substitution means for generating a sentence in a second language that is a translation.

  The sentence in the first language may be, for example, a patent application claim, a law, a contract, a treaty, and instructions for using various machines, electronic devices, and software.

  Preferably, the automatic translation apparatus is provided between the correspondence specifying means and the translation means, receives the structural component of the first language in which the correspondence relation and the grammatical characteristics are specified by the correspondence specifying means, A word rearrangement unit that rearranges the order of the word strings constituting the structural part in accordance with the order in which each translated word of the word string appears in the second language, and supplies the word translation unit to the translation unit. Including.

  More preferably, the word rearrangement means performs syntax analysis on each of the structural parts of the first language to generate a syntax analysis tree of the first language, and the first word generated by the syntax analysis means. In the parse tree of the second language obtained by conversion by the conversion means for converting the parse tree of one language into the parse tree of the second language in accordance with a conversion rule prepared in advance, Rearrangement means for rearranging and outputting the words of the structural parts of the first language in accordance with the appearance order of the words.

  For each of the structural parts of the first language, the translation means is preliminarily optimized for the translation of the word strings having the grammatical characteristics of the structural parts with respect to the word strings constituting the structural parts. A grammatical property-specific translation unit that performs automatic translation using a model for translation from a language into a second language and generates a translation in the second language may be included.

  More preferably, the pattern specifying means divides the sentence in the first language into a plurality of structural parts by a predetermined delimiter pattern included in the sentence in the first language, and divides by the dividing means. The grammatical characteristic determining means for determining the grammatical characteristics of each structural part, the appearance order of the structural parts divided by the dividing means, and the grammatical characteristics determined by the grammatical characteristic determining means for each structural part. Means for specifying the sentence pattern.

  The sentence in the first language may be a claim of a patent application written in the first language.

  The translation model learning device according to the second aspect of the present invention translates a sentence having a specific grammatical characteristic of a first language into a sentence of a second language different from the first language by statistical translation. Learn the translation model used. The apparatus includes: a collecting unit for collecting a plurality of parallel translations composed of a sentence having a specific grammatical characteristic of a first language and a translation of the sentence in a second language different from the first language; For learning a statistical model necessary for statistical translation from a sentence of a specific grammatical characteristic of a first language into a sentence of a second language, using a plurality of parallel translations collected by And statistical learning means.

It is a flowchart which shows the procedure of the automatic translation process of the claim of the conventional patent application. It is a figure which shows typically the translation process of the text in the automatic translation process of the claim of the conventional patent application. It is a schematic diagram for demonstrating the subject in the automatic translation process of the claim of the conventional patent application. It is a schematic diagram for demonstrating the learning process required for the process of automatic translation in the 1st Embodiment of this invention. It is a schematic diagram for demonstrating the translation process of the text in the claim automatic translation process based on the 1st Embodiment of this invention. It is a block diagram which shows the whole structure of the automatic translation system which concerns on the 1st Embodiment of this invention. It is a block diagram which shows schematic structure of the model learning part for structural component translation which learns the various models for automatic translation in the 1st Embodiment of this invention. FIG. 3 is a block diagram illustrating a schematic configuration of a syntax analysis model learning unit that learns a model for parsing a noun phrase and a verb phrase in the first embodiment of the present invention. 1 is a block diagram showing a schematic configuration of a claim translation unit that automatically translates claims of an English patent application into Japanese in the first embodiment of the present invention. FIG. It is a block diagram which shows the whole structure of the automatic translation system which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows schematic structure of the model learning part for structural component translation which learns the various models for automatic translation in the 2nd Embodiment of this invention. In the 2nd Embodiment of this invention, it is a block diagram which shows schematic structure of the claim translation part which translates the claim of an English patent application into Japanese automatically. It is a figure which shows the external appearance of the computer system which is a hardware flat form which implement | achieves the automatic translation system which concerns on each embodiment of this invention. It is a block diagram which shows the hardware constitutions inside the computer system shown in FIG.

  In the following description and drawings, the same parts are denoted by the same reference numerals. Therefore, detailed description thereof will not be repeated. The following description relates to the case of translating an English claim of a patent application into Japanese. However, the translation object of the following embodiment is not limited to such a case. For example, the present invention can be applied to a case where a Japanese claim is translated into English. Furthermore, it can be applied to laws and regulations, covenants, and treaties that are difficult to translate as well as claims of patent applications, and instructions for using various machines, electronic devices, and software.

[First Embodiment]
<Basic concept>
The automatic machine translation system for claims according to the first embodiment described below employs statistical machine translation. Referring to FIG. 4, regardless of whether in English or Japanese, the claims of the patent application are arranged and described in accordance with any of certain patterns, although there are various variations. In the present embodiment, such an arrangement pattern of structural parts of a claim is referred to as a claim pattern. This is similar to the conventional English pattern 74 shown in FIG. 2, and can be represented by a tree structure. An English claim pattern is called an English pattern, and a Japanese claim pattern is called a Japanese pattern. Moreover, an English pattern and a Japanese pattern can be matched by comparing an actual English claim with a corresponding Japanese claim. That is, a claim written according to a certain English pattern can be translated into a Japanese claim that matches a Japanese pattern corresponding to the English pattern.

  These claim patterns are obtained by classifying a plurality of claims. In each claim pattern, each structural part is given a grammatical characteristic of the structural part. For example, if a character string or word string forming a structural part is a noun phrase, the structural part is marked as a noun phrase, and if it is a verb phrase, it is marked as a verb phrase. Two claim patterns with the same appearance of the tree structure are different claim patterns if the grammatical characteristics of a certain structural component are different. It should be noted that such a claim pattern is different from the “separation pattern” used in Patent Document 1.

  By comparing the claim pattern of the English claim with the claim pattern of the Japanese claim corresponding to the English claim, the English pattern and the Japanese pattern can be associated with each other. Further, the corresponding claim patterns can be associated with each other. Such a pair of claim patterns is stored in advance as pattern classification data, and is used when learning a model for translation as described below. In the following embodiment, it is assumed that an English pattern and a Japanese pattern are associated one-to-one.

  In the following embodiment, it is assumed that an English claim is translated into a Japanese claim. An English pattern is determined by decomposing a word string of an English claim as a source language into a plurality of structural parts and determining grammatical characteristics of each structural part. As a result, the contents of the text constituting each structural part can also be obtained. It should be noted that, as in the case of Patent Document 1, the division pattern in the claims can be used for disassembling the structural parts. In particular, the transitional phrase is specific to the claim expression and has few variations, so it can be identified relatively easily. After the transition phrase is specified, the structural part corresponding to the subject of the invention is located either before or after the transition phrase according to the language, and one or more structural parts corresponding to the description are located on the other side. Separation between these structural parts can be performed by finding a specific separation pattern, or a statistical model for separation is created in advance to identify a portion that is likely to be a separation position. Can also be done.

  In the present embodiment, learning is performed for a separate translation model for each grammatical characteristic of a structural component (for example, the type of a noun phrase or a verb phrase), and a model corresponding to the grammatical characteristic of each structural component is used. Automatic translation for each structural part. That is, when translating a structural part composed of noun phrases, a translation model learned in advance using only noun phrases is used. When translating structural parts composed of verb phrases, a translation model learned in advance using only verb phrases is used. As a result, it is possible to obtain a more easily understood Japanese translation of each structural component.

  The original English pattern is related to a certain Japanese pattern, and which structural component of the English pattern corresponds to which structural component of the Japanese pattern is recorded in advance. Using this correspondence, the Japanese translation of each structural part is substituted into the position of the structural part corresponding to the original structural part of the Japanese pattern corresponding to the original English pattern, so that the Japanese translation of the claim Can be obtained.

  The term “translation model” used here means a translation model from a source language to a target language, a language model of the target language, and a phrase table generated in the course of learning thereof, which are used in a statistical translation apparatus. Refers to a set of models. In the following embodiment, the source language is English and the target language is Japanese. In the following embodiments, noun phrases and verb phrases are adopted as grammatical characteristics.

  FIG. 4 schematically shows the learning process of the noun phrase translation model 210 and the verb phrase translation model 214. Here, the English pattern and the Japanese pattern are collected in advance, and the correspondence between them is also established. This correspondence is one-to-one as described above. That is, when an English pattern is determined, a Japanese pattern corresponding to the English pattern is also determined, and the correspondence between the structural parts is similarly determined one-to-one.

  Referring to FIG. 4, a plurality of complaint translations 170 are prepared for learning. Each parallel translation 170 includes an English claim 172 and a Japanese claim 174 which is a translation of the English claim. For example, there are many published application documents that have a parallel translation relationship, such as a Japanese application translated into English and filed in the United States. The bilingual 170 may be selected from the set of published applications.

  The English pattern of the English claim 172 is determined, the claim is decomposed into structural parts according to the English pattern, and the grammatical characteristics of each structural part are classified according to the English pattern (process 176). Similarly, the Japanese claim 174 is decomposed into structural parts according to the Japanese pattern corresponding to the English pattern, and the grammatical characteristics of each structural part are classified according to the Japanese pattern (process 180). This process can be executed manually or by a machine by preparing a machine-readable pattern collection.

  As a result of this pattern classification, for example, it is determined that the English claim 172 matches the English pattern 178, and the Japanese claim 174 is determined to match the Japanese pattern 182. The structural parts of the English pattern and the Japanese pattern are associated in advance, and each structural part can be paired (process 184). For example, in the example shown in FIG. 4, the subject of the English claim is associated with the subject of the Japanese claim. The English phrase transition phrase is associated with the Japanese complaint transition phrase. The description of the English claim is associated with the description of the Japanese claim.

  Of the structural parts combined in this way, in the present embodiment, focusing on the grammatical characteristics of the structural parts, the pairs of the structural parts are classified into separate sets for each grammatical characteristic (grammatical type). That is, if the text of the pair of structural parts is a noun phrase, the pair of structural parts is classified into a noun phrase pair set 198 and if it is a verb phrase, it is classified into a verb phrase pair set 200 (process 196). The set 198 of noun phrase pairs thus obtained includes a large number of pairs of noun phrases of English claims and noun phrases of Japanese claims, and the set 200 of verb phrase pairs includes verb phrases and Japanese of English claims. Includes many pairs of claim verb phrases.

  Finally, the noun phrase translation model 210 is learned using the noun phrase pair set 198 as learning data, and the verb phrase translation model 214 is learned using the verb phrase pair set 200 as learning data.

  Referring to FIG. 5, the following processing is performed at the time of translation from English to Japanese. By comparing the English claim 70 with the English pattern, the claim pattern of the English claim 70 is specified. In this example, it is assumed that the English pattern 74 is specified. The English pattern 74 in this example includes three structural parts: a subject, a transitional phrase, and a description. Each of these has an area for storing text. In these areas, the English text corresponding to the subject, the English text corresponding to the transition phrase, and the English text corresponding to the explanation obtained by disassembling the English claim 70 into structural parts are stored (process 72).

  Next, a Japanese pattern associated with the identified English pattern is identified and generated (process 76). Specifically, information specifying the structure of the Japanese pattern 78 paired with the specified English pattern 74 is read from the claim pattern storage unit, and a new area for storing the new Japanese pattern is newly created. Secured in the storage area. In this example, the retrieved Japanese pattern 78 includes a description, a transition phrase, and a subject. These are arranged in this order, and each has an area for storing the corresponding text in the storage device. When generating Japanese patterns, these areas are empty.

  Subsequently, the subject of the English pattern, the transitional phrase, and the explanatory text are automatically translated using a model selected according to their grammatical characteristics (process 240). For example, since the subject is a noun phrase, it is translated using a noun phrase translation model. Among the descriptions, those that are verb phrases are translated using a verb phrase translation model, and those that are noun phrases are translated using a noun phrase translation model. Since the transition phrase is a verb phrase, it is translated using a verb phrase translation model. The result of translating the subject is the subject area of the Japanese pattern 78, the result of translating the transition phrase is the transition phrase area of the Japanese pattern 78, and the result of translating the description is the explanation area of the Japanese pattern. Each is stored. As a result, a translated Japanese pattern 242 is obtained. By connecting the text of each area of the obtained Japanese pattern 242 in the order of each area of the Japanese pattern 242, a Japanese claim that is a translation of the English claim 70 is obtained.

<Configuration>
With reference to FIG. 6, the claim translation system 300 which implement | achieves the automatic translation process mentioned above which concerns on this Embodiment is for translating an English claim into a Japanese claim. The claim translation system 300 includes a noun phrase translation model 320 used for English-Japanese translation of a noun phrase, a verb phrase translation model 322 used for English-Japanese translation of a verb phrase, and learning of these. Using the structural part translation model learning unit 318, the verb phrase syntax analysis model 312 and the noun phrase syntax analysis model 314, the English claim 324 is parsed and the parsed English claim 334 is output. For parsing the input English claim 334 after parsing and outputting the Japanese claim 326 using the parsing unit 332, the noun phrase translation model 320, and the verb phrase translation model 322 A translation unit 328.

  In order to learn the noun phrase translation model 320 and the verb phrase translation model 322, the structural part translation model learning unit 318 stores a complaint parallel translation corpus 310 that stores a large number of parallel translations of claims in English and Japanese, Verb phrase parsing model 312 for parsing Japanese verb phrases, Noun phrase parsing model 314 for parsing English noun phrases, English patterns related to claims and corresponding Japanese patterns Pattern classification data 316 in which a plurality of pairs are stored, and a parse tree conversion rule 330 for performing rearrangement on English structural parts, which will be described later, are used. The verb phrase syntax analysis model 312 is a syntax analysis model in which only English verb phrases are learned as learning data. The noun phrase syntax analysis model 314 is a syntax analysis model in which only English noun phrases are learned as learning data.

  With reference to FIG. 7, the structural part translation model learning unit 318 sequentially reads the claim parallel translations from the claim parallel translation corpus 310 and collates the English claims with the pattern classification data 316, thereby matching the English patterns. A pattern classification unit 350 for reading a Japanese pattern paired with the English pattern, a claim read out from the claim bilingual corpus 310, and an English claim based on the English pattern read out by the pattern classification unit 350 A structural part classifying unit 352 that classifies the structural part into a noun phrase and a verb phrase and outputs it, a noun phrase data storage unit 354 that stores an English noun phrase output by the structural part classifying unit 352, and a structural part A verb phrase data storage unit 356 that stores verb phrases output from the classification unit 352.

  The model learning unit for structural component translation 318 further performs a noun phrase pre-arrangement that performs an English claim word rearrangement process called pre-arrangement for each of the noun phrase data stored in the noun phrase data storage unit 354. Stored in the noun phrase learning data storage unit 362 and the noun phrase learning data storage unit 362 for storing the pre-sorted noun phrase data output from the noun phrase pre-ordering unit 358. A noun phrase model learning unit 364 for learning the noun phrase translation model 320 using the pair of the English claim and the Japanese claim after the rearrangement as learning data.

  The structural part translation model learning unit 318 further includes a verb phrase pre-ordering unit 360 that performs pre-ordering of English claim words for each of the verb phrase data stored in the verb phrase data storage unit 356, and a verb. The verb phrase learning data storage unit 366 for storing the pre-sorted verb phrase data output from the phrase pre-sorting unit 360, and the pre-sorted pre-sorted data stored in the verb phrase learning data storage unit 366 And a verb phrase model learning unit 368 for learning the verb phrase translation model 322 by using pairs of English claims and Japanese claims as learning data.

  The noun phrase pre-arrangement unit 358 performs syntax analysis for each noun phrase stored in the noun phrase data storage unit 354 using the noun phrase syntax analysis model 314 and outputs a syntax analysis tree. 380 and the parsing tree output by the parsing unit 380 are transformed into a parsing tree having the same structure as the corresponding Japanese parsing tree using the conversion rules stored in the parsing tree conversion rule 330. For rearranging English words in the order defined by the tree structure conversion unit 382 and the post-conversion parse tree output by the tree structure conversion unit 382 and storing them in the noun phrase learning data storage unit 362 A word rearrangement unit 384. In this embodiment, pre-ordering is performed in which the original (English) words are rearranged before translation. However, this is merely an example, and other rearrangement methods may be used. . For example, it is possible to employ a so-called post-sorting method in which words are translated according to the order of the words in the original text, and then the translated words are rearranged so as to be in the word order of the target language.

  The verb phrase pre-ordering unit 360 performs parsing for each verb phrase stored in the verb phrase data storage unit 356 using the verb phrase parsing model 312 and outputs a parsing tree. 400 and the parsing tree output by the parsing unit 400 are transformed into a parsing tree having the same structure as the corresponding Japanese parsing tree by using the conversion rule stored in the parsing tree conversion rule 330. The words in the English verb phrase are rearranged in the order specified by the tree structure conversion unit 402 and the post-conversion parse tree output by the tree structure conversion unit 402, and stored in the verb phrase learning data storage unit 366 A word rearrangement unit 404 for storage.

  In the present embodiment, the verb phrase syntax analysis model 312 and the noun phrase syntax analysis model 314 are models that have been learned to perform syntax analysis of an English verb phrase and an English noun phrase, respectively. Such learning is performed in advance by the model learning unit 442 for syntax analysis having a configuration as shown in FIG.

  Referring to FIG. 8, the parsing model learning unit 442 extracts a noun phrase from the source language corpus 440 storing a large amount of English documents, and decomposes the noun phrase and the various noun phrases. A noun phrase extraction / decomposition unit 460 for extracting words and phrases (hereinafter collectively referred to as “noun phrases”) and a noun phrase extracted by the noun phrase extraction / decomposition unit 460 are stored. Noun phrase syntax analysis learning data storage unit 462 and a noun phrase syntax analysis model 314 for learning the noun phrase syntax analysis model 314 using the noun phrase syntax analysis learning data storage unit 462. And a parsing model learning unit 464. The noun phrase parsing model 314 is a statistical model. When an English noun phrase is given, learning is performed so as to output a parse tree having the highest likelihood as a parse tree of the noun phrase.

  The parsing model learning unit 442 further includes a verb phrase extraction unit 470 for extracting a verb phrase from the source language corpus 440, and a verb phrase syntax analysis learning for storing the verb phrase extracted by the verb phrase extraction unit 470. Data storage unit 472 and verb phrase syntax analysis model learning unit for learning verb phrase syntax analysis model 312 using verb phrase syntax analysis learning data stored in verb phrase syntax analysis learning data storage unit 472 474. The verb phrase parsing model 312 is a statistical model. When an English verb phrase is given, learning is performed so as to output a parse tree having the highest likelihood as a parse tree of the verb phrase.

  Referring to FIG. 9, the claim translation unit 328 shown in FIG. 6 receives the input English claim 334 after parsing, searches the pattern classification data 316, and matches the English claim 334 after parsing. A pattern search unit 500 that reads a pattern pair having a pattern, a Japanese pattern generation unit 504 that extracts a Japanese pattern from the pattern pair read by the pattern search unit 500 and generates an empty Japanese pattern, and Japanese And a Japanese pattern storage unit 506 for storing the Japanese pattern generated by the pattern generation unit 504.

  The claim translation unit 328 further receives the post-parsing English claim 334 with an English pattern from the pattern search unit 500, and divides the post-parsing English claim 334 into structural parts in accordance with the structure of the English pattern. An input sentence division unit 502 to be output, a structural part storage unit 508 for storing a structural part output by the input sentence division unit 502 in association with a corresponding portion of an English pattern, and a structure stored in the structural part storage unit 508 The part is read in order, the structural part type determination unit 510 that outputs a determination signal by specifying whether the read out structural part is a noun phrase or a verb phrase from the English pattern, and the value of the determination signal output from the structural part type determination unit 510 In response to the selector 512 for selecting either the verb phrase parsing model 312 or the noun phrase parsing model 314; In response to the value of the determination signal output from the component type determination unit 510, the type is determined by the selector 522 that selects either the verb phrase translation model 322 or the noun phrase translation model 320, and the structural component type determination unit 510. And a syntactic analysis unit 514 that reads the structural component from the structural component storage unit 508, performs parsing using the parsing model selected by the selector 512, and outputs a parsing tree.

  The claim translation unit 328 further converts the syntax analysis tree output from the syntax analysis unit 514 into a corresponding Japanese syntax analysis tree according to the syntax analysis tree conversion rule stored in the syntax analysis tree conversion rule 330. A tree conversion unit 518.

  The claim translation unit 328 further includes a word rearrangement unit 520 that outputs an English word string in which English words are rearranged according to the order in which each word appears in the Japanese syntax analysis output from the syntax analysis tree conversion unit 518; A translation model selected by the selector 522 upon receipt of an English word string output from the word rearrangement unit 520 (a verb phrase translation model 322 when the target input sentence is a verb phrase, and a noun phrase translation model when the target input sentence is a noun phrase) Statistical automatic translator 524 that performs automatic statistical translation from English to Japanese using model 320), and Japanese sentences output by statistical automatic translator 524 are stored in Japanese pattern storage unit 506. And a Japanese pattern substitution unit 526 that substitutes the pattern into an area corresponding to the structural component to be translated.

<Operation>
The claim translation system 300 whose configuration has been described above operates as follows. The operation of the claim translation system 300 is roughly divided into three stages. The first stage is learning of the verb phrase syntax analysis model 312 and the noun phrase syntax analysis model 314 shown in FIG. The second stage is learning of the noun phrase translation model 320 and the verb phrase translation model 322 shown in FIG. When these two processes are completed, the syntax analysis unit 332 and the claim translation unit 328 can translate the English claim 324 into Japanese and output the Japanese claim 326 as a third stage. Prior to these processes, it is assumed that the pattern classification data 316 and the parse tree conversion rule 330 are prepared in advance by some means. Hereinafter, these will be described in order.

<Stage 1: Learning a model for parsing>
Referring to FIG. 8, an English source language corpus 440 is prepared in advance. For each sentence stored in the source language corpus 440, morphological analysis and syntax analysis are performed in advance, and information of the syntax analysis tree is attached.

  The noun phrase extraction / decomposition unit 460 extracts, from the source language corpus 440, a noun phrase, a word included in the noun phrase, and a partial parse tree indicating the syntax structure of the noun phrase, and stores noun phrase parsing learning data storage Stored in the unit 462. The noun phrase syntax analysis model learning unit 464 reads information stored in the noun phrase syntax analysis learning data storage unit 462 and learns the noun phrase syntax analysis model 314. As a result of this learning, the noun phrase parsing model 314 can be used for calculations that output the most likely parsing tree when an English noun phrase is given.

  On the other hand, the verb phrase extraction unit 470 extracts a verb phrase and a partial parse tree indicating the syntax structure of the verb phrase from the source language corpus 440 and stores them in the verb phrase syntax analysis learning data storage unit 472. The verb phrase syntax analysis model learning unit 474 reads the verb phrases stored in the verb phrase syntax analysis learning data storage unit 472 and learns the verb phrase syntax analysis model 312. As a result of this learning, the verb phrase parsing model 312 can be used for calculation to output the most likely parsing tree when an English verb phrase is given.

<Second stage: Learning translation model>
With reference to FIG. 7, the pattern classification unit 350 reads the Japanese-English parallel translation of the complaint from the parallel translation corpus 310 and collates it with the pattern included in the pattern classification data 316 to identify the English complaint pattern. The pattern classification unit 350 reads the specified English pattern from the pattern classification data 316. The pattern classification unit 350 divides the English claim into structural parts according to the English pattern, and substitutes it into the text area of each structural part of the English pattern. The structural part classification unit 352 stores each structural part of the English pattern to which the text is assigned in the noun phrase data storage unit 354 if it is a noun phrase, and in the verb phrase data storage unit 356 if it is a verb phrase. When this process is completed for all the complaint parallel translations stored in the complaint parallel translation corpus 310, or in parallel with this process, the noun phrase pre-ordering unit 358 and the verb phrase pre-sorting unit 360 can operate.

  The syntax analysis unit 380 of the noun phrase pre-arrangement unit 358 reads the noun phrase data stored in the noun phrase data storage unit 354 in order. The syntax analysis unit 380 analyzes the syntax of the read noun phrase using the noun phrase syntax analysis model 314, and outputs the syntax analysis tree to the tree structure conversion unit 382. When the parse tree is given, the tree structure conversion unit 382 converts the noun phrase parse tree of the English claim into the corresponding Japanese language using one of the tree structure conversion rules stored in the parse tree conversion rule 330. The data is converted into a parse tree structure and given to the word rearrangement unit 384. The word rearrangement unit 384 rearranges English words in the order determined by the Japanese parsing tree and accumulates them in the noun phrase learning data storage unit 362. The noun phrase model learning unit 364 learns the noun phrase translation model 320 using the noun phrase learning data stored in the noun phrase learning data storage unit 362 in this manner.

  The syntax analysis unit 400 of the verb phrase pre-ordering unit 360 sequentially reads the verb phrase data stored in the verb phrase data storage unit 356. The syntax analysis unit 400 parses the read verb phrase using the verb phrase syntax analysis model 312 to output a parse tree to the tree structure conversion unit 402. When the parse tree is given, the tree structure conversion unit 402 converts the parse tree of the verb phrase of the English claim into the corresponding Japanese language using one of the tree structure conversion rules stored in the parse tree conversion rule 330. The data is converted into a parse tree structure and given to the word rearrangement unit 404. The word rearrangement unit 404 rearranges English words in the order determined by the Japanese parsing tree and stores them in the verb phrase learning data storage unit 366. The verb phrase model learning unit 368 learns the verb phrase translation model 322 using the verb phrase learning data stored in the verb phrase learning data storage unit 366 in this way.

  As described above, when the processing for all of the claim parallel translations stored in the claim parallel translation corpus 310 is completed, the noun phrase and verb phrase translation model 322 is used to translate the English and Japanese noun phrases and verb phrases. Can be translated.

<Stage 3: Claim translation>
When translating a claim, the claim translation unit 328 operates as follows. Prior to the processing of the claim translation unit 328, the syntax analysis unit 332 shown in FIG. As a result, the English claim 334 after parsing is input to the claim translation unit 328.

  Referring to FIG. 9, when an English claim 334 after parsing is given to the claim translation unit 328, the pattern search unit 500 compares the pattern stored in the pattern classification data 316 with the English claim 334 after parsing. And search for matching English patterns. The pattern search unit 500 reads the English pattern that best matches the English of the English claim 334 after parsing from the pattern classification data 316 together with the corresponding Japanese pattern, and sends the Japanese pattern to the Japanese pattern generation unit 504. The pattern and the parsed English claim 334 are given to the input sentence dividing unit 502, respectively.

  Based on the given Japanese pattern, the Japanese pattern generation unit 504 creates a Japanese pattern text storage area in the Japanese pattern storage unit 506 for each structural component of the Japanese pattern. Each of these areas is empty at this point.

  The input sentence dividing unit 502 divides the English claim 334 after the parsing according to the English pattern read from the pattern classification data 316, substitutes it into the text storage area of each structural component, and stores it in the structural component storage unit 508. . When these structural components are stored in the structural component storage unit 508, the structural component type determination unit 510 reads one of them, determines whether the structural component is a noun phrase or a verb phrase, and sends a determination signal to the selector 512 and This is given to the selector 522. In response to this determination signal, the selector 512 selects the verb phrase parsing model 312 when the structural component is a verb phrase, and the noun phrase parsing model 314 when it is a noun phrase. To the syntax analysis unit 514. The selector 522 selects the verb phrase translation model 322 when the structural component is a verb phrase, and selects the noun phrase translation model 320 when the structural component is a noun phrase, and connects to the statistical automatic translator 524. .

  The syntax analysis unit 514 receives the structural component from the structural component type determination unit 510 and uses the verb phrase syntax analysis model 312 or the noun phrase syntax analysis model 314 connected via the selector 512 to analyze the structural component. To do. As a result, the syntax analysis tree is given from the syntax analysis unit 514 to the syntax analysis tree conversion unit 518. In this syntax analysis, the syntax analysis unit 514 uses the noun phrase syntax analysis model 314 if the target structural component is a noun phrase, and uses the verb phrase syntax analysis model 312 if it is a verb phrase. , And parse each target structural part. The verb phrase syntax analysis model 312 is learned by the syntax analysis model learning unit 442 shown in FIG. 8 as verb phrases as learning data, and the noun phrase syntax analysis model 314 is used for syntax analysis using noun phrases as learning data. This is learned by the model learning unit 442. Therefore, the syntax analysis performed by the noun phrase syntax analysis model 314 is optimal according to the type of the structural component.

  When the syntax analysis tree conversion unit 518 receives the syntax analysis tree from the syntax analysis unit 514, the syntax analysis tree conversion unit 518 refers to the syntax analysis tree conversion rule 330, and among the conversion rules, the syntax analysis rule for the syntax analysis tree received from the syntax analysis unit 514. Choose the most appropriate one. The parse tree conversion unit 518 uses the selected conversion rule to convert the parse tree given from the parse tree 514 to the English that is to be the translation for the English represented by the parse tree. Convert to a parse tree that shows the syntax structure. In the converted parse tree, the tree structure is a Japanese sentence structure, but the words assigned to each leaf are English words. The parse tree conversion unit 518 gives the converted parse tree to the word rearrangement unit 520.

  The word rearrangement unit 520 uses the converted parse tree given from the parse tree conversion unit 518 to rearrange the English words appearing in the parse tree in the order determined by the structure of the parse tree. The obtained English word string is given to the statistical automatic translator 524 as an input.

  The statistical automatic translator 524 performs statistical translation from English to Japanese using the translation model connected by the selector 522 for the English word string, and the resulting Japanese word string Is given to the Japanese pattern substitution unit 526. The translation model used at this time is a verb phrase translation model 322 when the structural component to be translated is a verb phrase, and is a noun phrase translation model 320 when it is a noun phrase. Therefore, in the English-Japanese translation of each structural component, there is a high possibility that noun phrases are correctly translated as noun phrases and verb phrases as verb phrases.

  The Japanese pattern substitution unit 526 stores the Japanese pattern for the Japanese word string output from the statistical automatic translator 524 using the correspondence information between the original English structural component and the Japanese pattern structural component. The translation result is substituted into the text storage area of the corresponding structural part of the Japanese pattern stored in the part 506.

  When the above processing is performed on all the structural parts stored in the structural part storage unit 508, each structural part of the Japanese pattern stored in the Japanese pattern storage unit 506 has a corresponding English structure. A word string obtained by translating parts into Japanese is stored. These translation results were performed using a verb phrase translation model 322 optimized for a verb phrase if the original English structural component was a verb phrase, and optimized for a noun phrase if it was a noun phrase. This was performed using the noun phrase translation model 320. Therefore, it is expected that the individual structural parts constituting the Japanese claims will be accurate reflecting the type of the original English structural parts. The structural parts of the Japanese pattern storage unit 506 are arranged according to the appearance order of the structural parts in the Japanese claim. Therefore, by reading each structural part stored in the Japanese pattern storage unit 506 in order from the top and displaying it appropriately, the Japanese claim 326 which is an accurate translation for the English claim 324 can be obtained.

[Second Embodiment]
In the first embodiment, in the translation model learning by the structural part translation model learning unit 318 shown in FIG. 7, the noun phrase pre-rearrangement unit 358 and the verb phrase pre-rearrangement unit 360 perform noun phrases and verbs. Pre-ordered phrases. In addition, when translating an English claim by the claim translation unit 328 shown in FIG. Thus, by rearranging the words in advance during learning and translation, the efficiency of learning increases and the accuracy of translation also increases. Therefore, it is desirable to perform word rearrangement in this way. However, the present invention is not limited to such an embodiment. Depending on the grammatical characteristics of the structural parts of the English pattern (types such as verb phrases or noun phrases), different translation models are pre-learned. Anything can be used as long as automatic translation is performed using an appropriate translation model. In the second embodiment, prior rearrangement is omitted.

  FIG. 10 is a block diagram showing the overall configuration of the claim translation system 550 according to the second embodiment. The claim translation system 550 is different from the claim translation system 300 shown in FIG. 6 in that the verb phrase parsing model 312, the noun phrase parsing model 314, and the parsing tree conversion rule 330 are not used. 6 includes a structural part translation model learning unit 570 and a claim translation unit 572 that do not use pre-ordering, instead of the structural part translation model learning unit 318 and the claim translation unit 328, respectively. Note that the verb phrase syntax analysis model 312 and the noun phrase syntax analysis model 314 are used in the syntax analysis by the syntax analysis unit 332 as in the first embodiment.

  FIG. 11 shows a schematic block diagram of the structural part translation model learning unit 570. The structural part translation model learning unit 570 is different from the structural part translation model learning unit 318 shown in FIG. 7 in that the noun phrase data storage unit 354, the verb phrase data storage unit 356, and the noun phrase pre-arrangement for pre-ordering. The noun phrase output from the structural part classification unit 352 is directly stored in the noun phrase learning data storage unit 362 without including the replacement unit 358 and the verb phrase pre-ordering unit 360, and the verb phrase is directly learned for the verb phrase. This is a point accumulated in the data storage unit 366. In other respects, each part of the structural part translation model learning unit 570 is the same as the corresponding part of the structural part translation model learning unit 318.

  FIG. 12 shows a schematic block diagram of the claim translation unit 572 shown in FIG. Compared with FIG. 9, the claim translation unit 572 does not include the selector 512, the syntax analysis unit 514, the parse tree conversion rule 330, the parse tree conversion unit 518, and the word rearrangement unit 520 necessary for pre-ordering, 9 is different from the claim translation unit 328 of FIG. 9 in that the output of the structural part storage unit 508 is directly given to the statistical automatic translator 524. In other respects, each part of the claim translation unit 572 has the same configuration as each part of the claim translation unit 328.

  This second embodiment is different from the first embodiment in that pre-sorting is not performed during learning of a translation model or during translation. Even with such a configuration, the translation of the claims can be performed with higher accuracy than the conventional one. In particular, in the translation between languages in which the word order of the two languages is relatively close, it is possible to expect a sufficient improvement in translation accuracy without using pre-ordering.

<Modification>
In the first embodiment, pre-ordering is used, and in the second embodiment, pre-ordering is not used. The present invention is not limited to such an embodiment, and for example, it may be possible to select whether or not to perform rearrangement. By doing in this way, it becomes possible to perform both translation with pre-sorting and translation without translation according to the target language.

  In the above embodiment, a verb phrase translation model and a noun phrase translation model are used as translation models. However, the present invention is not limited to such an embodiment, and other models such as an adverb phrase translation model may be used.

[Realization by computer]
The claim translation system 300 according to the first embodiment, the claim translation system 550 according to the second embodiment, and other modified examples include computer hardware and a computer program executed on the computer hardware. And can be realized. FIG. 13 shows the external appearance of the computer system 930, and FIG. 14 shows the internal configuration of the computer system 930.

  Referring to FIG. 13, the computer system 930 includes a computer 940 having a memory port 952 and a DVD (Digital Versatile Disc) drive 950, a keyboard 946, a mouse 948, and a monitor 942.

  Referring to FIG. 14, in addition to memory port 952 and DVD drive 950, computer 940 includes a CPU (Central Processing Unit) 956, a bus 966 connected to CPU 956, memory port 952, and DVD drive 950, and a boot program. A read-only memory (ROM) 958 for storing etc., a random access memory (RAM) 960 connected to the bus 966 for storing program instructions, system programs, work data, etc. The computer system 930 further includes a network interface (I / F) 944 that provides a connection to a network 968 that allows communication with other terminals.

  A computer program for causing the computer system 930 to function as each functional unit of the claim translation system 300 and the claim translation system 550 according to the above-described embodiment is stored in the DVD 962 or the removable memory 964 installed in the DVD drive 950 or the memory port 952. It is stored and further transferred to the hard disk 954. Alternatively, the program may be transmitted to the computer 940 through the network 968 and stored in the hard disk 954. The program is loaded into the RAM 960 when executed. The program may be loaded directly from the DVD 962 into the RAM 960 from the removable memory 964 or via the network 968.

  This program includes an instruction sequence including a plurality of instructions for causing the computer 940 to function as each functional unit of the claim translation system 300 and the claim translation system 550 according to the above-described embodiment. Some of the basic functions required to cause computer 940 to perform this operation are an operating system or third party program that runs on computer 940 or various dynamically linkable programming toolkits or programs installed on computer 940. Provided by the library. Therefore, this program itself does not necessarily include all of the object code for realizing the functions necessary for realizing the system and method of this embodiment. This program can be used as a system as described above by dynamically calling the appropriate program in the appropriate function or programming toolkit or program library at run time in a controlled manner to achieve the desired result. It is only necessary to include an instruction for realizing the function. Of course, all necessary functions may be provided only by the program.

  In addition, the functional units of the claim translation system 300 and the claim translation system 550 are processed by being distributed to different computers, or are distributed and processed by different computers in different regions via a network. May be.

  The embodiment disclosed herein is merely an example, and the present invention is not limited to the above-described embodiment. The scope of the present invention is indicated by each claim of the claims after taking into account the description of the detailed description of the invention, and all modifications within the meaning and scope equivalent to the wording described therein are included. Including.

300, 550 Claim translation system 310 Claim translation corpus 312 Verb phrase parsing model 314 Noun phrase parsing model 316 Pattern classification data 318, 570 Structural part translation model learning unit 320 Noun phrase translation model 322 Verb phrase translation model 328, 572 Claim translation unit 330 Parse tree conversion rule 358 Noun phrase pre-ordering unit 360 Verb phrase pre-ordering unit 442 Parsing model learning unit 524 Statistical automatic translator

Claims (6)

An automatic translation device that translates a sentence in a first language into a sentence in a second language different from the first language,
A dividing unit that identifies a sentence pattern of the first language and divides the sentence of the first language into structural parts;
Pattern specifying means for specifying a pattern of the sentence in the second language, which is associated in advance with the pattern of the sentence in the first language specified by the dividing means;
Between the sentence pattern of the first language divided by the dividing means and the pattern of the second language specified by the pattern specifying means, the correspondence between the structural parts and the grammatical characteristics of each structural part A correspondence identifying means for identifying
For each of the structural parts of the first language, automatic translation using a model for translation from the first language to the second language is performed on a word string constituting the structural part, A translation means for generating a second language translation;
For each of the structural parts in the first language, the pattern specifying means determines the translation in the second language obtained by the translation means in accordance with the correspondence relationship of the structural parts specified by the correspondence specifying means. Substituting means for generating a sentence of the second language, which is a translation of the sentence of the first language, by substituting into any of the structural components of the pattern of the sentence of the second language specified by Including automatic translation device. Further, the first language structural component, which is provided between the correspondence specifying unit and the translation unit and has the correspondence relationship and the grammatical characteristics specified by the correspondence specifying unit, is received. A word rearrangement unit that rearranges the order of the word strings constituting the structural component in accordance with the order in which the respective translations of the word strings appear in the second language, and gives them to the translation unit. The automatic translation apparatus according to claim 1. The word rearranging means is:
A syntax analysis means for generating a syntax analysis tree of the first language by performing syntax analysis according to a grammatical characteristic of the structure component for each of the structural components of the first language;
Conversion means for converting the parse tree of the first language generated by the parse analysis means into the parse tree of the second language according to a conversion rule prepared in advance;
Rearrangement means for rearranging and outputting the words of the structural parts of the first language according to the appearance order of the words in the parsing tree of the second language obtained by the conversion by the conversion means. Item 3. The automatic translation device according to Item 2.   For each of the structural parts in the first language, the translation means is pre-optimized for translation of a word string having a grammatical characteristic of the structural part with respect to a word string constituting the structural part. 4. A grammatical characteristic translation unit that performs automatic translation using a model for translation from a first language to the second language to generate a translation of the second language. The automatic translation apparatus in any one of. The pattern specifying means includes
Dividing means for dividing the sentence in the first language into a plurality of structural parts according to a predetermined delimiter pattern included in the sentence in the first language;
A grammatical characteristic determining means for determining a grammatical characteristic of each structural part divided by the dividing means;
Means for specifying a sentence pattern of the first language according to the appearance order of the structural parts divided by the dividing means and the grammatical characteristics determined by the grammatical characteristic determining means for each structural part. The automatic translation apparatus in any one of Claims 1-4. Translation model learning apparatus for learning a translation model used when a sentence having a specific grammatical characteristic of a first language is translated into a sentence of a second language different from the first language by statistical translation Because
A collecting means for collecting a plurality of parallel translations of the sentence having the specific grammatical characteristic of the first language and a translation of the sentence in a second language different from the first language;
Statistical data necessary for statistical translation from the sentence of the specific grammatical characteristic of the first language to the sentence of the second language using the plurality of parallel translations collected by the collecting means as learning data A model learning device for translation, including statistical learning means for learning a model.

Images