JP2004086919A - Mechanical translation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a mechanical translation system equipped with a word dictionary allowing a user to freely customize 'the way of translation'. <P>SOLUTION: This mechanical translation system is provided with the word dictionary 9 in which an entry word every word, the equivalent term in translation when present, the grammatical attribute, and analysis information showing the relation with other words registered therein, a grammar database 10 for storing essential construction grammar, a morpheme analytic function part 3 for inputting a sentence and collating it with the word dictionary to divide it into morphemes, a construction analytic function part 4 for extracting the subject from the word group of the morphemes divided by the morpheme analytic function part 3, trying the generation of a construction tree from the analytic information of the subject, and judging the adaptability with the construction identity of each word, thereby determining an optimum construction tree, an equivalent term determining function part 5 for determining the equivalent term of each word in the construction tree determined by the construction analytic function part 4, and a translated sentence generating function part 6 for applying the equivalent term to each word of the construction tree to generate a translated sentence. <P>COPYRIGHT: (C)2004,JPO

Description

The present invention relates to a machine translation system that automatically translates a sentence written in a certain language into another language, and particularly relates to a word (relationship information) between each word and the other words (analysis information) accompanying each word. The present invention relates to a machine translation system that stores in a dictionary and performs a rational automatic translation using analytic information centering on a head, based on a head-driven phrase structure grammar (HPSG) theory.

機械 Fifteen years have passed since commercial machine translation software appeared in Japan. At present, there are many low-cost machine translation software, and machine translation software has been widely used. In particular, the recent Internet boom has been a tailwind, and translation software is rapidly spreading, and it is one of the most successful business application software.

However, current machine translation software is mostly used as a “speed-reading tool” to capture the general meaning of the entire sentence. As a tool to support, the degree of perfection is still low, and its spread is slow.

In recent years, internationalization and openness have progressed rapidly around the world, and in Japan, the number of situations where large quantities of “practical translations” must be produced has increased. Therefore, instead of a translation that can understand the general meaning of conventional machine translation software, it generates a translation that is faithful to the original and that can be conveyed correctly to third parties, and supports "translation" in the true meaning New machine translation software is expected to appear.

翻 訳 Generally, translation software is composed of a “word dictionary”, a “grammar database”, and a “program”.

(4) The "word dictionary" has, for each word, various information necessary for analyzing the original sentence and generating the target sentence, in addition to the part of speech and the translated word. The "grammar database" describes, as rules, what parts of speech appear in what order and what syntax they have in order for a sequence of words to be formed as a sentence. The "program" inputs the original sentence, determines the sentence structure using the "word dictionary" and "grammar database" for each sentence ("analysis" phase), and generates a target sentence based on that structure ("Generate" phase).

In the word dictionary of conventional machine translation software, word headwords (spellings), translations, parts of speech, and the like are registered. It was registered.

Here, the configuration of a conventional machine translation system and the flow of its processing will be described.

FIG. 12 shows the configuration of a conventional machine translation system and the flow of its processing. The conventional machine translation system 11 includes a standard input unit 12, a morphological analysis function unit 13, a syntax analysis function unit 14, a translated word determination function unit 15, a generated tree generation / transformation function unit 16, and a translated sentence generation function unit 17. And a standard output unit 18.

The machine translation system 11 has a word dictionary 19 and a grammar database 20.

The word dictionary 19 describes headwords (spellings), parts of speech (nouns, verbs, adjectives, etc.) and translations of each word.

The grammar database 20 describes syntax rules (grammar rules) for parts of speech. The rule of the syntax regarding the part of speech is a rule for associating the syntax pattern with the arrangement and appearance pattern of the part of speech. The number of grammar rules ranges, for example, from 2000 to 3000.

According to the conventional machine translation system 11, an English character string is input from the standard input unit 12, and the English character string is passed to the morphological analysis function unit 13.

The morphological analysis function unit 13 extracts words from the English character string with reference to the word dictionary 9 and passes the words to the syntactic analysis function unit 14 together with information on the parts of speech (nouns, verbs, adjectives, adverbs, etc.).

The syntax analysis function unit 14 determines a syntax tree from the part-of-speech array of words in the English character string and the grammar database 20.

For example, in the sentence "This is a pen.", "This" (noun), "is" (verb), "a" (article), and "pen" (noun) are extracted by the morphological analysis function unit 13, A syntax pattern that matches these part-of-speech arrangements illustrated in FIG. As described above, the grammar database 20 stores the syntax patterns of a large number of part-of-speech arrays, and searches for a syntax pattern that matches the part-of-speech array of the input English character string by a pattern matching method to create a syntax tree.

The syntax analysis function unit 14 sends the syntax tree (syntax analysis data) obtained by the above method to the translated word determination function unit 15.

The translated word determination function unit 15 performs semantic processing to determine a translated word corresponding to the syntax tree, and outputs syntactic analysis data and translated word data to the generation tree generation / transformation function unit 16.

The generation tree generation / transformation function unit 16 refers to the grammar database 20, generates a syntax tree for generating Japanese in accordance with a specific translation rule, and outputs it to the translated sentence generation function unit 17.

The translated sentence generation function unit 17 translates each word of the syntax tree for generating Japanese into Japanese and outputs Japanese translation data (Japanese sentence character string) via the standard output unit 18.

機械 In machine translation by the above-mentioned conventional machine translation system, the crucial factor in parsing is the grammar rules registered in the grammar database.

場合 If there were any omissions in the grammar rules, of course, the optimal parse tree could not be found. Further, even if it is described in a grammar rule, there is no means for determining an optimal solution from among a large number of solutions derived by analysis based on the grammar rule. Such sentences would always produce incorrect results if the "right interpretation" was not chosen as the first solution. In this case, no matter how the user specifies words in the word dictionary, the grammar database itself does not generate an appropriate syntax tree, so that an appropriate translation cannot be obtained.

In other words, conventional machine translation systems rely on "grammar categories" and "grammar rules" built into the system for "analysis of English sentences" and "generation of translated sentences". Is governed by the grammar rules, and the user can change (customize) only the word-by-word dictionary description, for example, customization to output a specific translated word, in order to obtain the desired translation. There was a limit.

Here, the limitation of the conventional machine translation will be described with reference to a specific example.

Now, consider the case of translating the English character string “Time flies like an arrow.” Into Japanese.

に は In this English character string, both words “flies” and “like” can be verbs, so these two words can be predicates for the whole sentence.

場合 If “flies” (fly) is a predicate, “Time flies like an arrow” becomes a syntax tree as shown in FIG.

In the parse tree of FIG. 9, the predicate "flies" has the subject of "Time", and the words "like" and below are interpreted as prepositional phrases, and the preposition "like" represents the word of the other party (here, called the object). It is understood that the object has "" an arrow "".

When interpreted as such a parse tree, the input English character string “Time flies like an arrow.” Is translated as “Time flies like an arrow.”

On the other hand, when “like” is used as a predicate, “Time flies like an arrow” becomes a syntax tree as shown in FIG.

に お い て In the syntax tree of FIG. 10, the subject “like” has the subject “Time flies” and the object “an arrow”.

解 Given this syntax tree, the input English character string “Time flies like an arrow.” Is translated as “Time flies like arrows.”

The conventional machine translation system could not select a syntax tree desired by the user from both of the above syntax trees.

A first problem to be solved by the present invention is to provide a machine translation system having a word dictionary in which a user can freely customize “translation method”.

Secondly, a problem to be solved by the present invention is to provide a machine translation system that can determine an optimal syntax tree from a plurality of possible syntax trees from a grammar.

The machine translation system according to claim 1 of the present application
A word dictionary that registers headwords for each word, translations when they exist, grammatical attributes, and analysis information indicating the relationship with other words,
A grammar database that stores the main syntactic grammars,
A morphological analysis function unit for inputting a sentence, collating with the word dictionary and decomposing into morphemes;
A syntax for extracting a head from a group of words among the morphemes decomposed by the morphological analysis function unit and selecting a match that matches the syntactic feature of each word of the preceding and following morphemes from the analysis information of the head to determine a syntax tree Analysis function part,
A translated word determining function unit that determines a translated word corresponding to each word in the syntax tree determined by the syntax analyzing function unit;
A translation generating function unit that generates a translation by applying a translation to each word of the syntax tree.

The machine translation system according to claim 2 of the present application is the system according to claim 1,
In the word dictionary, generation information describing a special translation rule when a predetermined word satisfies a user-specified condition is registered,
It has a generator tree generation / modification function unit that modifies the syntax tree determined by the syntax analysis function unit according to the translation rule.

The machine translation system according to claim 3 of the present application is the machine translation system according to claim 1 or 2,
In the word dictionary, the semantic attribute of the target word related when the word becomes the subject is registered as the analysis information of the word, and the semantic attribute is registered in the analysis information of the word that can be the subject of the subject. It is characterized by having been done.

The machine translation system according to claim 4 of the present application is the system according to any one of claims 1 to 3,
It has a dictionary registration unit for registering and updating at least one of the semantic attributes of words, analysis information, and generation information.

The machine translation system according to claim 5 of the present application is the system according to any one of claims 1 to 4,
When there are a plurality of semantic attributes, analysis information, or generation information of a word, a word dictionary in which the user designates the priority of the semantic attribute to be applied, analysis information, or generation information via a dictionary registration unit is provided. It is characterized by having.

The machine translation system according to claim 6 of the present application is the system according to any one of claims 1 to 4,
When there are a plurality of semantic attributes, analysis information, or generation information of a word, the syntax analysis function unit searches the word dictionary for the most recently applied semantic attribute, analysis information, or generation information. It is to be.

According to the present invention, when translating a sentence that is very similar to, but slightly different from, a previously translated sentence, a translation is generated with reference to the previously translated sentence, which is a pre-stage of a so-called “self-learning type machine translation system”. The conventional dictionary description (word-part-of-speech-translation) has been expanded so that not only a conventional dictionary registration for each word but also analysis information and generation information for each word are described in a word dictionary.

This allows users to write arbitrary rules in the analysis and generation information of word dictionaries, and allows for a wider range of syntax analysis and customization of translation generation, including fixed wording. .

カ ス タ マ イ ズ This customization is limited to the translation of defined words, so the user can customize the translation flexibly and finely for each word.

Also, by giving the semantic attribute of a word to the analysis information, it is possible to correctly select the head and the word related thereto, and thereby to select the correct syntax tree from a plurality of syntax trees that are grammatically possible. Became.

Furthermore, according to the kanji conversion front-end processor having a word dictionary in which semantic attributes are registered along with the words and the semantic attributes of words that can be taken by adjuncts and predicates, according to the relationship between the semantic attributes of the words, Kanji conversion can now be performed correctly from multiple homonyms.

Hereinafter, the machine translation system according to the present invention will be specifically described with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of a machine translation system according to the present invention. Although the present invention is not limited to translation between specific languages, an example in which English is translated into Japanese for ease of understanding is shown. Therefore, the words “English” and “Japanese” in the following description are appropriately read as the translated language and the translated language according to the language to be translated.

機械 The machine translation system according to the present invention has almost the same components as the conventional machine translation system in terms of configuration. However, in the machine translation system according to the present invention, the registered contents of the dictionary are significantly different from those of the conventional dictionary, and accordingly, the method of parsing and generating / transforming the generated tree is significantly different from the conventional machine translation system.

The machine translation system 1 of FIG. 1 includes a standard input unit 2, a morphological analysis function unit 3, a syntax analysis function unit 4, a translated word determination function unit 5, a generated tree generation / transformation function unit 6, and a translated sentence generation function unit. 7 and a standard output unit 8.

The machine translation system 1 has a word dictionary 9 and a grammar database 10.

The standard input unit 2 and the standard output unit 8 are known input means and output means.

The word dictionary 9 of the present invention includes, for each word, a headword, a translated word when present, a grammatical attribute, analysis information indicating a relationship with another word, and a special translation when a predetermined condition is satisfied. This is a dictionary in which generation information describing rules is registered.

“Grammatical attributes” refers to information such as part of speech, numbers, person names, and cases. “Analysis information indicating the relationship with other words” refers to information that describes what word a certain word requires as a subject, what kind of word as a complement is required, and the like. "Generated information" refers to information describing a translation method corresponding to a word of an input sentence that satisfies a certain arrangement condition in order to perform natural translation in Japanese.

Note that the headwords, translations, grammatical attributes, and the like of the words are stored in a part called the content part of the word dictionary 9, the analysis information is stored in the analysis binary part of the word dictionary 9, and the generation information is the binary for the generation of the word dictionary 9. Stored in the department.

The grammar database 10 of the present invention is a dictionary that stores grammar of main syntax such as sentence patterns. In the grammar database of the conventional machine translation system, detailed and huge grammar rules (for example, 2000 to 3000 rules) are stored for each part of speech / word type, whereas the grammar database 10 of the present invention is stored in the grammar database 10 of the present invention. Only dozens of rules, such as basic sentence patterns, are stored.

The morphological analysis function unit 3 is means for decomposing the input character string (sentence) into quotes, parentheses, dashes (these are referred to as block data), and words. The morphological analysis function unit 3 can recognize block data from the input character string, divide words by the block data and spaces, and extract words.

Note that block data and words are collectively referred to as “morpheme” in this specification.

The morphological analysis function unit 3 inputs an English character string from the standard input unit 2, creates a block data list including quotation marks, parentheses, and dashes from the English character string input as described above, and creates a block from the English character string. The word character string divided except for the data list is extracted, each word is searched from the content part of the word dictionary 9, and word data corresponding to each word is created. The morphological analysis function unit 3 collects the word data and the first created block data list and outputs them to the syntax analysis function unit 4 as morphological data.

The syntax analysis function unit 4 is a means for determining an optimal syntax tree (sentence structure, that is, a tree structure expressing the relationship between words) from morphological data.

The syntax analysis function unit 4 receives the morphological data from the morphological analysis function unit 3 and converts all the word data in the morphological data into phrase structure data with reference to the analysis binary unit of the word dictionary 9. Here, the “phrase” is a group of plural words having the same function as a noun, an adjective, an adverb, and the like. At the time of conversion into “phrase”, the syntax analysis function unit 4 follows the block data list included in the morphological data input from the morphological analysis function unit 3 and specifies the designated block portion (enclosed in quotation marks, parentheses, and dashes). Part) are analyzed as one phrase.

Next, the syntax analysis function unit 4 performs a syntax analysis based on the phrase structure data, the analysis binary unit of the word dictionary 9 and the information of the grammar database 10 to obtain syntax analysis data (data representing a syntax tree) of the entire sentence. It is created and output to the translated word determination function unit 5.

This syntactic analysis focuses on the head of the sentence (often the verb of the sentence) and the grammatical relationship between the subject registered in the head word and the predicate or complement (part of speech, number This is a process of determining an optimal syntax using analysis information such as a personal name, a personal attribute, and the like, and a semantic attribute (described later). This syntax analysis will be described again later with a specific example.

The translated word determination function unit 5 is a means for determining a translated word of a word adapted to the syntax analysis data.

The translated word determination function unit 5 receives the syntax analysis data from the syntax analysis function unit 4, determines which translated word is to be adopted based on the entire sentence structure and the translated word selection information of each word, and the sentence has a plurality of interpretations. In this case, the weight of each interpretation is calculated, the one with the smallest weight is selected as a solution, and the syntax analysis data and the translated word data are output to the generation tree generation / deformation function unit 6.

The generation tree generation / deformation function unit 6 generates a generation tree for Japanese translation (a syntax tree for Japanese generation) so that a natural translation can be performed as Japanese when a specific condition is satisfied. It is a means to transform the parse tree.

Specifically, the generation tree generation / deformation function unit 6 receives the syntax analysis data and the translated word data from the translated word determination function unit 5, and describes the data in the binary unit for generation of the word dictionary 9 based on the arrangement and form of the words included in the data. According to the generated generation information, a syntax tree for generating a Japanese language is created, or the original syntax tree is transformed so as to generate a more understandable Japanese language. The Japanese-language generation syntax tree after the transformation is output to the translated sentence generation function unit 7 by the generation tree generation / transformation function unit 6.

The translated sentence generation function unit 7 is a means for applying translated words to the syntax tree for generating Japanese language and outputting Japanese translation data (Japanese sentence character string).

Specifically, the translated sentence generation function unit 7 receives the syntax tree for generating Japanese from the generation tree generation / deformation function unit 6, creates a Japanese sentence character string in accordance with the translated word described in the word dictionary 9, and inputs this to the standard input. Output to section 2.

The translation generation function unit 7 shall satisfy the following requirements.
Information that does not correspond one-to-one with English (modality such as permission and duty) is recorded in the content section of the word dictionary 9 in the form of Japanese additional information (additional generation information).

{Circle around (2)} The utilization table of the vocabulary word is described in the binary part for generation of the word dictionary 9, and how to utilize the vocabulary word follows the utilization table data. The utilization table also describes how the utilization word changes depending on the additional information described above.

The data for words to be added to each word (such as particles for nouns) is also described in the generation binary part of the word dictionary 9 according to the hierarchical relationship of the generation tree, and words to be added to words or phrases according to this data are described. Add.

In addition, among the constituent means of the machine translation system 1, the generated tree generation / deformation function unit 6 can be omitted according to the purpose of the system. For example, in a simple system for the purpose of only direct translation or draft translation, the generation tree generation / deformation function unit 6 can be omitted as appropriate.

Although the machine translation system 1 does not show a means for customizing a dictionary, the present invention also includes a system in which dictionary registration means for customizing the word dictionary 9 is appropriately added.

In addition, the analysis binary part of the word dictionary 9 is a description in which only the grammatical attribute of the word related to the head is described, and a semantic category (semantic attribute) of the word related to the head is added. There can be. The word dictionary 9 and a translation method using the word dictionary 9 will be described below using specific examples.

First, an overview of HPSG theory is explained. HPSG is an abbreviation of Head-Driven Phrase Structure Grammar. At the heart of this theory is the concept of "Head = head (word that is the center of a sentence or phrase)" as the name implies. In HPSG, it is assumed that a phrase or sentence always has a central word = head, and the nature of the phrase is described in the word which is the head of the phrase.

In the following, the concept of the subject will be explained using the example of the sentence "I go." In this example, a case will be described where the information of the semantic attribute is not used in the syntax determination process.

The sentence "I go." Is simply a sentence aligned with the "subject verb," but this is formed as a sentence because the subject word "go" has the property of "having the subject." Only if "I" satisfies the condition of the subject. In the present invention, it is described in the word dictionary 9 that the word "go" is a word that can be a subject and has the property of "having a subject", which accompanies the word "go". The property “having the subject” and the condition of the subject are the analysis information of “go”.

Similarly, for example, "I see you." Is formed as a sentence because the subject "see" has both properties of "having a subject" and "having an object" and its subject "I" satisfies the condition, and "you" satisfies the condition of the object. In this case, the analysis information that the word "see" has "the subject" and "the object" and the condition of the subject and the condition of the object are described along with the word "see" in the word dictionary 9. I have.

In order to specifically describe the condition of the “subject” that “I” of “I go.” Must satisfy, a “syntax feature” is defined for each word in the word dictionary 9. The syntactic feature includes the grammatical attributes of the word, the analysis information, and the like, and the properties of the elements constituting the syntax. The syntactic feature has a format of "set of features and their values", and is defined as shown in FIG. .

In FIG. 2, the "part of speech" on the left side is the feature (name), and the "noun, verb ..." on the right side is the feature value. ("{}" Is one of the elements in {} Take). Based on this definition, the syntactic features of each word can be described as shown in FIG.

FIG. 3 shows a “word dictionary” that defines what syntactic feature each word of “I go.” Has. If there are multiple syntactic features, express them by separating them with slashes.

句 The “subject phrase = <…>” of “go” is the analysis information of “go”, and has one syntactic feature and has a role of indicating the condition of the subject of the word. For example, in the triad of "go" = "goes", the value of "the subject phrase = <...>" changes from "person = first person" of "go" to "person = third person".

In order for the character string “I go” to be a sentence, the condition of “the subject phrase = <…>” of “go” must match the syntactic feature value of “I”. To specify this restriction, a grammar rule as shown in FIG. 4 is defined.

“New phrase → phrase 1 phrase 2” in the expression (1) in FIG. 4 indicates that “a newly created phrase is composed of“ phrase 1 phrase 2 ””, and “phrase 2 [becomes the subject Phrase = <condition of subject>] ”indicates that the value of the feature“ phrase that is the subject ”of phrase 2 is represented by a variable of“ condition of subject ”.

The following explains how these grammatical rules are applied. Applying the string "I go" to this grammar rule yields clause 1 = "I" and clause 2 = "go". The analysis information of "go" is described in the dictionary in the form of syntactic feature as shown in FIG.

Applying FIG. 5 to “Phrase 2 [Phrase that becomes subject = <condition of subject>]” in the grammar rules, the result is as shown in FIG. After (2) if in the grammar rules, the condition of each phrase is shown. “Phrase 1 = subject condition” indicates a condition that “the syntactic feature of phrase 1 and the syntactic feature assigned to the variable“ subject condition ”are consistent”.

In the sentence "I go", the two in FIG. 7 match without any contradiction, and the condition is satisfied, so that this grammatical rule is established.

(After (3) then of the grammar rules, it shows what phrases are created by applying the grammar rules. “New phrase: subject = phrase 1” indicates that “the subject of the new phrase becomes“ phrase 1 ””, and “new phrase: subject = phrase 2” indicates that “the new phrase“ subject ”is“ phrase 2 ”. "]. In other words, the feature value of the "subject" of the newly created phrase (the phrase corresponding to the entire character string "I go") contains the contents of "phrase 1 =" I "" as it is, and the feature value of the "subject" is The content of “phrase 2 =“ go ”” is entered as it is. Therefore, "I go" is created as a phrase having a configuration as shown in FIG.

Thus, in the example sentence of "I go.", "Go" has the subject of the first person singular noun, and "I" is the first person singular noun, so "I" is the subject of "go". As a result, the subject + verb syntax was determined.

で は In the syntax determination process in the above example, the grammatical attribute of the word is the sole criterion. Next, parsing using a semantic attribute as a criterion will be described with reference to "Time flies like an arrow."

“Time flies like an arrow.” Can be headed by “flies” (fly) and “” like ”(preferred) as described above.

場合 When “flies” (fly) is used as the head, “Time flies like an arrow” becomes a syntax tree as shown in FIG. 9 as described above.

に お い て In the syntax tree of FIG. 9, the subject “flies” has the subject “Time”, and the words after “like” are interpreted as prepositional phrases. Also, inside a prepositional phrase, "like" is interpreted as having "an arrow" as its object, in which case "Time flies like an arrow." "(Literally translated).

On the other hand, when “like” is the head, “Time flies like an arrow” becomes a syntax tree as shown in FIG.

10, the subject "like" has the subject "Time flies" and the object "" an arrow. "In this case," Time flies like an arrow. " , "Time flies like arrows. Is interpreted as ".

For such two parse trees, the conventional machine translation system could not select an optimal parse tree because each of these parse trees could be grammatically correct. When the dictionary 9 interprets “like” as the subject of “like”, the fact that the subject has “semantic attribute representing a person” can be registered as the analysis information of “like”. On the other hand, if "flies" is interpreted as a noun (fly), the fact that "flies" is present can be registered as analysis information of "flies".

Therefore, if the subject is “flies” (flies) (having “semantic attributes representing insects”) as in the syntax tree of FIG. 10, the semantic attributes of the subject “flies” and the semantic attributes of the subject “like” match. Will not do. On the other hand, if "flies" is interpreted as the head of "fly" as in the syntax tree of FIG. 9, such a mismatch in semantic attributes does not occur. Therefore, according to the machine translation system 1 of the present invention, the syntax tree in FIG. 10 is excluded and the syntax tree in FIG. 9 is selected.

As described above, the conventional machine translation system has no means for determining an optimal syntax tree, whereas the machine translation system according to the present invention can determine an appropriate syntax tree.

Finally, the generator tree generation / deformation function of the present invention will be described.

HPSG is a theory mainly for performing “syntactic analysis”, but the applicant uses the framework to develop modules other than “syntactic analysis”. By doing so, it is possible to perform “syntax analysis” more finely than a conventional machine translation system, as well as context information (information such as “modified by a specific modifier”) for “translation determination” and “generation”. Correspondence information between the original sentence and the translated sentence for "tree generation / transformation" (information such as "adjective in the original sentence but adverb in the translated sentence") can also be described in the dictionary, allowing for more detailed translation than conventional machine translation systems. Can be set.

The machine translation system 1 of the present embodiment generates / transforms a generated tree for performing natural translation into Japanese under certain conditions.

For example, consider the sentence "He is a good swimmer."

FIG. 11 shows a state of generation / transformation of a generation tree and generation of a translated sentence in the above sentence example.

In a conventional machine translation system, the subject is "he", the verb is "is", and the complement is "a good swimmer", and "subject" is "complement". . "

The machine translation system 1 of the present invention parses the subject as "is", the subject as "he", and the complement as "a good swimmer" based on the HPSG theory.

Next, according to the machine translation system 1 of the present invention, “noun (swimmer) that can be a verb is added to the noun of the complement in the generation binary part of the word dictionary 9 for“ is ”, and it is an adjective ( When modified by "good", the noun is translated by a verb expression, and the generation information that "noga" is added to the nomenclature and the whole is translated by "ga syntax" is described, and "swimmer" If the verb translation “swimming” is registered in the dictionary description, a generator tree (syntax tree for generating Japanese) suitable for the verb expression of the noun and the adverb expression of the adjective is generated as shown in FIG. .

Next, the machine translation system 1 applies a translated word to this generated tree, and generates a translated sentence “He is good at swimming”.

こ と By writing the generated information as described in the above example in the word dictionary 9 precisely, it is possible to overcome the unnaturalness of the conventional direct-translation-like translation and to generate a more natural Japanese translation.

(4) The machine translation system according to the present invention describes headwords, translations, grammatical attributes, analysis information, generation information, and the like (syntax features) in the word dictionary as described above.

Therefore, the machine translation system according to the present invention can easily include “dictionary registration means” for allowing a user to freely customize translation.

That is, if the dictionary registration means registers / updates the semantic attributes of the words in the word dictionary, the analysis information, and the generation information, the words have the semantic attributes, analysis information, and generation information as specified by the user. work.

Here, there is a great advantage that the analysis information registered by the user for the word is applied only to the translation of the word.

That is, in the conventional machine translation system, it is necessary to define a grammar rule in order to define a translation method. However, once a grammar rule is defined, the grammar rule is applied to all words, not just the translation of the target word. In this case, there is a problem that an undesirable translation method is applied unexpectedly. On the other hand, according to the method of defining the analysis information for each word as in the present invention, the analysis information is applied only to the translation of the word, so that the user can be finely customized.

Originally, it seems that a language captures the characteristics of a language by combining a word with a word to create meaning, but translation software based on such vocabulary theory has not yet appeared. The method based on the HPSG theory of the present invention is characterized in that most of the syntax information used to grasp the structure of a sentence can be described in a "vocabulary", that is, a "dictionary" instead of a "grammar rule". In order to further improve the accuracy of “selection of translation” and “generation of translation”, more detailed generation rules can be described on the dictionary. As a result, even if the translation software fails to produce a correct translation, the user can obtain the desired analysis result, translation, and translation if the user corrects the dictionary. As a result, the range in which the user can customize is greatly increased compared to the conventional case, and a machine translation system with a high "learning effect" can be realized.

In other words, in the machine translation system of the present invention developed based on the HPSG theory, most of the information on syntax is defined as a lexical property, and this can be described in a dictionary. This expands existing dictionary descriptions (descriptions of analysis information and generated information), and enables not only conventional dictionary registration in word units, but also broader dictionary registration including fixed wording. By simply changing the registered contents of the dictionary, a machine translation system that can obtain the desired translation can be realized.

Further, the dictionary registering means can specify a semantic attribute to be applied, or analysis information, or a priority order of generation information, when there are a plurality of meaning attributes, analysis information, or generation information of a word. can do.

Also, when there are a plurality of meaning attributes, analysis information, or generation information of a word, it is possible to automatically prioritize the most recently applied meaning attribute, analysis information, or generation information.

The above is the description of the machine translation system of the present invention. The following describes a kanji conversion front-end processor that utilizes the features of the word dictionary.

Generally, it is difficult for kanji conversion front-end processors to convert homonyms.

The conventional kanji conversion front-end processor had a word dictionary in which the parts of speech and headwords of words were registered, a context analysis dictionary for converting homonyms and a program therefor.

On the other hand, the kanji conversion front end processor according to the present invention is characterized in that not only words but also semantic attributes of words, adjuncts, and semantic attributes of words that can be taken by predicates are registered in the word dictionary. .

For example, consider the sentence "Your reporter returned by train." In this case, "Kisha" must be converted to Kanji into four different words.

The words that are used in the word, such as "no" for "your company", "ha" for "reporter", "de" for "by train", and "ha" for "returned", are included Word. An adjunct has the role of indicating the attribute of the word in which it is used. "-" Is attached to a noun indicating the partner to which the user belongs, and the noun has a semantic attribute of a person, an organization, an object,.... "~ Ha" is attached to a noun indicating the subject, and the noun has a semantic attribute of person, organization, object, .... "~" Is attached to a noun indicating a means, and the noun has a semantic attribute of an object. The word "to" is attached to a verb indicating an action.

On the other hand, "you" is a noun with a semantic attribute of organization, "reporter" is a noun with a semantic attribute of human, "train" is a noun with a semantic attribute of vehicle, and "going home" indicates action and has the subject of a semantic attribute of human. .

で は In the above sentence example, “returned to home” is a predicate, and has a subject having a semantic attribute of human. As a result, the subject "Kishaha" becomes "Reporter", the one showing its action means "By train", and the person to whom the reporter belongs becomes "Your company".

As described above, the word dictionary for kanji conversion according to the present invention registers semantic attributes and other generation information for kanji conversion along with words. Therefore, similarly to the word dictionary of the machine translation system described above, user customization for performing a specific kanji conversion can be easily realized.

In other words, by providing a dictionary registration means for registering and updating the semantic attribute of a word or generated information in a specific usage, a word can have an arbitrary semantic attribute, and a specific adjunct or predicate On the other hand, it can be converted to a specific kanji.

Also, as in the case of the machine translation system, when there are a plurality of semantic attributes and generation information of a word, the dictionary registration means can specify the semantic attributes to be applied and the priority of the generation information. Further, when there are a plurality of meaning attributes and generation information of a word, it is also possible to preferentially apply the semantic attribute and generation information applied first.

The applicant of the present invention has one object to provide a machine translation system capable of performing effective "example sentence translation".

例 “Translation of example sentences” was proposed more than ten years ago, but it has only appeared in commercial translation software in the last two years or so. In the "Example Sentence Translation" developed by the applicant, the result of translation is stored in the database as a pair of the original sentence and the translated sentence, and when the exact same sentence appears, this translation is used. Things.

However, since the probability that the same sentence appears is generally low, it is also possible to allow some variable expressions and register the example sentence so that even if the example sentence is partially different, the translated example can be adopted. We are devising. With this technology, you can use a translation technique based on the traditional "grammar rules" as a core, or if you want to produce a more appropriate smooth Japanese expression instead of a direct translation-like expression produced by machine translation. By registering pairs of original sentences and translated sentences in the example sentence database for each sentence, the user can have a means of customizing the translation system other than the dictionary.

A system that stores the example sentences translated by the translator in the past in the example sentence database, searches the example sentence database for a sentence that matches or is similar to the sentence to be translated and displays it, and supports human translation while referring to the translated sentence , "Translation memory".

The present applicant incorporated the "translation memory" into machine translation and fused "example sentence translation" and "conventional rule-based translation" several years ago using the expression "integrated translation software". Incorporated from. In the future, in addition to matching technology and similar sentence search technology, there will be many future improvements such as expansion of example sentences including more flexible variable expressions and user-friendly registration methods. I am convinced that it is one pillar that breaks the wall of the.

The present applicant has developed the latest linguistic theory, HPSG (Head-driven Phrase Structure Grammar) theory (typical of "vocabulary theory"), from the conventional translation technology based on "grammar rules". We have been working on the development of a machine translation system for several years. The purpose is to build "next-generation translation technology" to increase the accuracy of translation and dramatically increase the scope of user customization.

FIG. 1 is a block diagram of an embodiment of a machine translation system according to the present invention. The figure which shows the definition of a syntactic feature. The figure which shows the description example of the syntactic feature of each word. The figure which shows an example of a grammar rule. The figure which shows the description example of the syntactic feature of "go". The figure which shows the condition of the subject in a grammar rule. The figure which shows the condition of the subject of "I go". The figure which shows the subject and the head of "I go". The figure which shows the syntax tree when "flies" of "Time flies like an arrow." Is used as a predicate. The figure which shows the syntax tree when "like" of "Time flies like an arrow." Is used as a predicate. FIG. 6 is a diagram showing a state of generation / conversion of a generation tree. FIG. 2 is a block diagram showing a configuration of a conventional machine translation system. The figure which showed the example of the syntax tree by the conventional machine translation system.

Explanation of reference numerals

DESCRIPTION OF SYMBOLS 1 Machine translation system 2 Standard input unit 3 Morphological analysis function unit 4 Syntax analysis function unit 5 Translated word determination function unit 6 Generated tree generation / transformation function unit 7 Translated sentence generation function unit 8 Standard output unit 9 Word dictionary 10 Grammar database

Claims (6)

A word dictionary that registers headwords for each word, translations when they exist, grammatical attributes, and analysis information indicating the relationship with other words,
A grammar database that stores the main syntactic grammars,
A morphological analysis function unit for inputting a sentence, collating with the word dictionary and decomposing into morphemes;
A parse tree is determined by extracting a head from a group of words among the morphemes decomposed by the morphological analysis function unit and selecting a morphological information that matches the syntactic feature of each word of the preceding and following morphemes from the analysis information of the head. A parsing function unit,
A translated word determining function unit that determines a translated word corresponding to each word in the syntax tree determined by the syntax analyzing function unit;
A translation generation function unit that generates a translation by applying a translation to each word of the syntax tree. In the word dictionary, generation information describing a special translation rule when a predetermined word satisfies a user-specified condition is registered,
The machine translation system according to claim 1, further comprising a generation tree generation / deformation function unit that deforms the syntax tree determined by the syntax analysis function unit according to the translation rule. In the word dictionary, the semantic attribute of the target word related when the word becomes the head is registered as the analysis information of the word, and the semantic attribute is registered in the analysis information of the word that can be the target of the head. The machine translation system according to claim 1, wherein the translation is performed. 4. The machine translation system according to claim 1, further comprising a dictionary registration unit that allows a user to register / update at least one of a semantic attribute of a word, analysis information, and generation information. When there are a plurality of semantic attributes, analysis information, or generation information of a word, a word dictionary in which the user designates the priority of the semantic attribute to be applied, analysis information, or generation information via a dictionary registration unit is provided. The machine translation system according to any one of claims 1 to 4, wherein the system is provided. When there are a plurality of semantic attributes, analysis information, or generation information of a word, the syntax analysis function unit searches the word dictionary for the most recently applied semantic attribute, analysis information, or generation information. The machine translation system according to any one of claims 1 to 4, wherein

Images