JPH11328178A - Machine translating device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the precision of syntax analysis and to shorten the time required for the syntax analysis by processing a unit of expression form, represented by a collocation, in an original text as one morpheme. SOLUTION: An English-Japanese collocation dictionary is prepared on a hard disk device and this English-Japanese collocation dictionary contains units of expression form represented by collocations. In a syntax analyzing process, an expression form composed of words combined by a coordinate conjunction is retrieved from an English text (S200) and then recognized as one morpheme when it is stored in the English-Japanese collocation dictionary (S210: YES) or when the words constituting the retrieved expression form have the same prefix or suffix (S240: YES), so that the syntax is analyzed (S250) without separation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device for converting an input original text into a target language and outputting a translated sentence.

[0002]

2. Description of the Related Art Conventionally, in a machine translation system, an analysis unit performs an analysis process for an original text to be translated in units of one sentence. For example, morphological analysis (Morephore Analysis) → syntactic analysis (also called syntax analysis. Syntax Analysis, Parsing) → semantic analysis (Mean
ing Analysis) is known. First, words are extracted as morphemes by morphological analysis, and then syntactic analysis is performed. The syntactic analysis is a process of analyzing the regular arrangement of the cut-out words. As an example of the syntactic analysis processing, there is a syntactic analysis employing a top-down vertical search method. This syntactic analysis process uses a context-free grammar (ContextFree Grammar,
The syntactic structure is analyzed as a parse tree (Parsing Tree) by repeating the analysis process of branching from root → node → leaf according to CFG) until it reaches the terminal which is the minimum unit of analysis. In the semantic analysis process, for example, the semantic attribute of the main part is determined by referring to and comparing the noun semantic attributes described in the system dictionary, and information on the sentence pattern that can be formed by the detected predicate is referred to. The collation determines the sentence pattern and the meaning.

[0003]

By the way, in the original text, there are collocations of the same part of speech connected by, for example, coordination conjunctions. For example, increase and decrease,
It takes the form of A and B or A, B and C, such as an editor, a translator and a detranslator (editing means, converting means, reprinting means). These collocations are common forms of expression in English sentences. In particular, the forms A, B and C are triads.
called d). Such collocations are collectively meaningful and can be said to be the original morpheme.
Therefore, it is desirable to grasp an English sentence as a chain of expression forms such as collocations when performing English sentence analysis. However, in the morphological analysis for extracting words as described above, such collocations are mechanically separated.

[0004] Therefore, in the conventional syntactic analysis, even though the word is a collocation, the analysis is performed for each individual word constituting the collocation, and in particular, an English sentence is delimited immediately before the coordination conjunction. There were many. Therefore, there has been a problem that errors in syntactic analysis are likely to occur. Also, to separate and analyze collocations that do not need to be separated on the way,
There was a problem that parsing took a lot of time. In syntactic analysis, the analysis time increases exponentially with longer sentences and more complex syntax. For this reason, especially in syntactic analysis of an English sentence having an extremely long and complicated syntax such as an English patent specification, the analysis time is increased.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and processes a group of expression forms in a source text, such as collocations, as one morpheme, thereby improving the accuracy of syntactic analysis. And to shorten the time required for syntactic analysis.

[0006]

Means for Solving the Problems and Effects of the Invention According to a first aspect of the present invention, there is provided a machine translation apparatus which converts an input original text into a target language and outputs a translated sentence. In a machine translation device, a word having a part common to an array of character data is composed of at least an expression form storage unit that stores an expression form formed by being combined by a coordinating conjunction, Search means for searching the expression form to be searched in the original text, and when the expression form searched by the search means is stored in the expression form storage means, the expression form is recognized as one morpheme, and the syntax And a syntax analysis means for analyzing the data.

Here, character data means, for example, the word "increase", i, n, c,
... means each alphabet. The word having a portion common to the character data includes, for example, increase and decrease. That is, in this case, the character data c, r, e, a, s,
The arrangement of e is common.

The expression form storage means stores an expression form in which words having a common part in character data are connected by coordination conjunctions. Such an expression form is not limited to one composed of only two words, but also includes one composed of three or more words. For example, an editor, a translator and a detranslator as described above
(Editing means, converting means, reprinting means). In particular, in many cases, the expression form composed of words having a common character data array is composed of words having a common prefix (suffix) or suffix (suffix). Therefore, an expression form composed of words having the same prefix or suffix may be stored.

The search means searches the original text for an expression form composed of words connected by coordination conjunctions. In other words, it searches the original text for expressions A and B, expressions A, B and C called triade, or expressions A or B. When the expression form searched by the search means is stored in the expression form storage means, the syntax analysis means recognizes the expression form as one morpheme and analyzes the syntax.

Conventionally, in machine translation, morphological analysis is performed to cut out the original text into words as morphemes, and then syntactic analysis is performed. Therefore, even a so-called collocation is automatically cut out into each word, and in syntactic analysis, an English sentence is often divided just before the coordination conjunction in the collocation. For this reason, errors in syntactic analysis are likely to occur, and the time required for syntactic analysis has been long.

Therefore, in the present invention, an expression form in which words having a part common to the arrangement of character data are combined by coordinating conjunctions is stored in advance, and those expression forms are stored as one morpheme. Perform parsing. Here, the premise of the technical idea of the present invention is the fact that "in a group of expression forms represented by collocation, the arrangement of character data is very often a specific arrangement." . Therefore, by recognizing an expression form in which such an arrangement of character data is a specific arrangement as one morpheme, it is possible to analyze a syntax without separating a group of expression forms represented by so-called collocations. Can be. As a result, the occurrence of errors in the syntactic analysis can be suppressed, and the time required for the syntactic analysis can be reduced.

Furthermore, as a feature of English sentences, in a unitary expression form such as a collocation, not only does the array of character data constituting a word have a commonality, but also the sound (pronunciation) of the word has a commonality. Often have. Therefore, as described in claim 2, the expression form stored in the expression form storage means may be constituted by words having common alliteration or rhyme in addition to the character data array. In this case, not only the array of character data,
) and a rhyme (End-Rhyme) are stored. As a result, it is possible to eliminate an exceptional expression form that does not form a united expression form even if the arrangement of the word character data has commonality by determining the commonality of alliteration or rhyme. The occurrence of errors in the analysis can be effectively suppressed.

It is conceivable that the expression form storing means further stores an expression form constituted by a preposition pair or an adverb pair. This is because the preposition pair or the adverb pair may form a united expression form. For example, a pair of prepositions that make up a verb phrase
Switch on and off and power on and off are such.

Further, as described in claim 4, the expression form storage means may further store an expression form composed of a pair of synonyms or antonyms. This is because pairs of synonyms or antonyms may form a united expression form. For example, add and delete
That is it.

[0015] By storing various expression forms in advance in this way, it is possible to analyze the syntax without separating a group of expression forms in the original text as one morpheme, and to obtain an error in syntactic analysis. And the time required for syntactic analysis can be reduced as much as possible.

By the way, when the syntax is analyzed using the above-mentioned expression form as one morpheme, in order to determine what syntactic element of the expression form is a syntactic element of the sentence, the corresponding expression form is used. It is necessary to specify the part of speech to do. At this time, for example, in the case of addition and deletion, it is considered that the part of speech corresponding to this expression form is specified as a noun based on the part of speech of each word constituting the expression form with reference to a dictionary stored in advance. Can be However, considering that the expression form recognized as one morpheme as described above is stored in the expression form storage means, it is conceivable to adopt the configuration described in claim 5. That is, in addition to the configuration shown in claims 1 to 4, the configuration is such that the expression form storage means stores the part of speech of the expression form when the expression form is regarded as a unit in association with the expression form. Stored, the parsing means refers to the part of speech of the expression form stored in the expression form storage means,
It is characterized in that it is configured to perform syntax analysis.

In this case, the expression form storage means stores the part of speech of each of the expression forms in association with each of the expression forms. For example, drag and drop is a verb, addition and deletion is a noun, positive and n
egative is an invariant adjective and directly and immediately is an invariant adverb. Also, for example, switch on
An expression form composed of a specific verb such as and off and a preposition pair or an adverb pair may be stored as a phrasal verb. Then, the syntax analysis means analyzes the syntax with reference to the part of speech corresponding to the expression form stored in advance. As a result, the time required for syntactic analysis can be reduced.

The machine translation apparatus described above searches for an expression form composed of words connected by coordinating conjunctions in the original text, and when the searched expression form is stored in the expression form storage means. Was used to analyze the syntax without separating and recognizing the expression form as a morpheme, but even if the expression form was not stored in the expression form storage means, the expression form to be handled as a group May be present in the original text.

Therefore, as set forth in claim 6, the search means further searches for an expression form composed of words having the same prefix or suffix, and the syntax analysis means searches the expression form. It is conceivable that an expression form having the same prefix or suffix is recognized as one morpheme, and the syntax is analyzed. In this case, the search means searches for an expression form composed of words joined by coordinating conjunctions, and also retrieves an expression form composed of words having the same prefix or suffix. Then, the syntactic analysis means recognizes the expression form composed of words having the same prefix or suffix as one morpheme, even if the expression form is not the expression form stored in the expression form storage means. The analysis is performed.

That is, as described above, since a group of expression forms represented by collocations are often composed of words having a common prefix or suffix, such an identical prefix or suffix is used. By searching for an expression form composed of words having words, and performing syntactic analysis using the searched expression form as one morpheme, even if the expression form is not stored in advance in the expression form storage means, it is not separated. Parsing can be performed. As a result, errors in syntactic analysis can be further prevented. In addition, the time required for syntactic analysis can be further reduced.

As a feature of English sentences, in a unitary expression form such as collocation, not only does the array of character data constituting a word have a commonality but also the sound (pronunciation) of the word. Paying attention to the fact that the words are often searched, as shown in claim 7, an expression form in which the search means is composed of words having the same suffix or prefix and having a common alliteration or rhyme And the parsing means recognizes the expression form having the same prefix or suffix and the same acronym or rhyme as one morpheme searched by the search means as one morpheme, and performs syntax analysis. Is also effective.

It should be noted that an expression form composed of words having the same prefix or suffix retrieved by the retrieval means as described above, or a word composed of words having the same prefix or suffix together with the same initial or rhyme. If the expression form to be performed is not stored in the expression form storage means, the storage control means may additionally store the expression form in the expression form storage means.

In such a machine translation apparatus,
For example, the result of the syntax analysis by the syntax analysis means may be edited so that it is easy for the user to understand, and may be output as it is on the screen in its original text. Alternatively, after performing the syntax analysis, the pre-stored dictionary may be referred to. The translation into the target language may be output as a translation. However, when considering translation, there may be cases where it is desirable to perform translation specific to the expression form. For example, increase and dec
rease means increase / decrease and translate.

Therefore, as shown in claim 9, claim 1
In addition to the configurations shown in FIGS. 1 to 8, the expression form storage means stores a target language translation of the expression form corresponding to the expression form, and refers to the target language translation stored in the expression form storage means. By doing so, it is conceivable to adopt a configuration including a translation unit that converts the original text analyzed by the syntax analysis unit into a target language and outputs a translated sentence.

In this case, the expression form storage means stores a target language translation of each of the expression forms in association with each of the expression forms. For example, drag and d
rop is "drag and drop", addition and dele
Option is “addition or deletion”, positive and negative is “positive and negative”, and directly and immediately is “directly and immediately”. The translation unit refers to the target language translation corresponding to the expression form stored in advance, converts the original text analyzed by the syntax analysis unit into the target language, and outputs a translated sentence. As a result, since each expression form is translated into an appropriate expression, it is possible to output a translated sentence with less discomfort.

Further, as set forth in claim 10,
It is conceivable to provide a translation sentence editing unit for editing the translation sentence output by the translation unit. The translated sentence editing means converts, for example, the passive expression in the original sentence into the active expression. by this,
Further, it is possible to output a translated sentence with less discomfort.

Considering that there are always exceptions to the rules of any language, for example, even if the expression forms are composed of words having the same prefix or suffix, they are grouped together. It may not be possible to handle it Therefore, the present invention further comprises editing means for the user to edit the input original text or the original text analyzed by the parsing means, and the translation means is edited by the editing means. It is preferable to convert the original text into the target language and output the translated text. In this case, the user can edit, for example, the original text input on the screen by the editing unit or the original text analyzed by the parsing unit. Then, the translation means can convert the output from the editing means into a target language and output a translated sentence. As described above, by allowing the user to edit the original text, the machine translator becomes easy to use, and the spread of the machine translator can be achieved.

From this point of view, it is also effective to adopt the structure described in claim 12. That is, in addition to the configuration described in claim 11, the configuration includes a mode for outputting an analysis result by the syntax analysis unit, a mode for converting the output from the syntax analysis unit to a target language, and a mode for outputting the output from the editing unit. A mode switching means is provided for the user to selectively switch between a mode for conversion to a language and a mode for conversion to a language.

For example, it is conceivable to provide a configuration in which each mode can be selected from a menu. In the mode in which the analysis result by the syntax analysis means is output, the analysis result is output as it is without executing the translation process. For example, it is conceivable that a sentence is edited and output on a screen so that the modification relationship can be understood. This mode has a flow of English sentence reading processing → syntax analysis. This mode is advantageous in that, when a long sentence or a complex compound sentence such as an English patent specification is increased, the original text can be output as it is, so that the modification relation in the original text can be easily grasped. . In the mode in which the output from the parsing means is converted into the target language, morphological analysis, syntactic analysis, and semantic analysis are collectively performed. That is, the flow is English reading processing → syntax analysis → translation editing processing. Further, in the mode in which the output from the editing means is converted into the target language, the original text edited into a predetermined format by the editing means is converted into the target language. In other words, the flow is English reading processing → translation editing processing.

[0030]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a machine translation apparatus according to the present embodiment. The machine translation device according to the present embodiment includes an information processing device 10, a hard disk device (hereinafter, referred to as "HD device") 13, a CD-ROM device (hereinafter, referred to as "CD device") 15, and a printer 17.
, An optical disk device 19, a CRT display device 21, an input device 23, and a flexible disk device (hereinafter “FD”).
Device. " ) 25.

The information processing apparatus 10 includes a central processing unit 31 configured as a logical operation circuit and an external interface 3.
3, the internal interface 35, and the memory device 39
And a communication interface 37 as main parts. The external interface 33 includes the HD device 1
3, a CD device 15, a printer 17, and an optical disk device 19 are detachably connected. The internal interface 35 includes a CRT display device 21 and an input device 23.
And the FD device 25 are connected.

In the CD device 15, a patent publication CD-ROM (not shown) issued by the United States Patent and Trademark Office is set.
A control code is stored in the CD-ROM together with a text file containing technical information such as a patent specification. According to the control code, for example, only a claim portion can be extracted.

The printer 17 and the optical disk device 19
Starting with the text file on the CD-ROM mentioned above,
It performs input / output of the result of parsing the English patent specification and the translation of the English patent specification in the CD-ROM. CRT display device 2
1 also starts with the data file in the CD-ROM,
The results of parsing the English patent specification in the D-ROM and the translated translation are displayed. Further, the CRT display device 21 displays a menu related to the English text translation process. The input device 23 includes a keyboard (not shown) and a mouse (not shown), and a user can select various menus displayed on the CRT display device 21 with the keyboard or the mouse. This menu will be described later.

The FD device 25 reads and writes simple and small-volume data such as system settings for each user. The central processing unit 31 includes a well-known CPU, ROM, and RAM. A work area for translating English text is assigned to the memory device 39.

The communication interface 37 transmits and receives data to and from an external device. Communication interface 37
Has a demodulation function and can be connected to a general line. Therefore, the English text file is received from the other party via the communication line, and the translated text is transmitted to the other party via the communication line.

FIG. 2 shows the internal configuration of the HD device 13 described above. The HD device 13 includes an English-Japanese dictionary 41, a technical term dictionary 43, an English-Japanese collocation dictionary 45, a program storage area 47, an analysis result storage area 49, and a translation sentence storage area 5.
1 and a relevance information model 53.

As will be described later, the program storage area 47 stores a program for analyzing the input English text and executing a translation process. This program is executed by the central processing unit 3 of the information processing apparatus 10 shown in FIG.
1 and developed on the memory device 39 via the external interface 33 of the information processing device 10. Then, the central processing unit 31 executes a predetermined process based on the developed program. This program is composed of four English sentence analysis modules: a lexical analysis module, a syntax analysis module, a semantic analysis module, and a document synthesis module.

The lexical analysis module includes rules for recognizing (cutting out) individual words from text data input as character strings, attribute data of words to be cut out, and the like. The syntactic analysis module analyzes a sentence structure by associating a predetermined syntactic analysis rule with a sequence of cut-out words and checking which rule the word sequence matches. Parsing rules are, for example, if the sentence pattern of an English sentence is
It is described in a meta-language such as Bacchus notation (BNF), and is performed by a driver routine (not shown) that operates an analysis process for analyzing a syntactic structure based on a parse tree employing a predetermined parsing algorithm. The semantic analysis module is described as a translation rule based on, for example, the Tokieda-Miura grammar, and performs semantic analysis along a parse tree output from the syntax analysis module. The document synthesis module generates a Japanese sentence from a parse tree output from the syntax analysis module and a Japanese translation of each English word obtained by the semantic analysis by the semantic analysis module.

The relevance information model 53 includes a patent specification, a legal book, an academic paper, a technical literature, a newspaper, a novel,
Optimal models for various target texts such as columns are prepared, and the relevance information model is selected according to the target text read from a CD-ROM or the like.

The English-Japanese dictionary 41 and the technical term dictionary 43 include:
For each word, phonetic symbols are stored together with the part of speech, sentence pattern, meaning, and the like. The English-Japanese collocation dictionary 45 contains a set of expressions that should not be separated in syntactic analysis, such as collocations composed of words of the same part of speech connected by coordinating conjunctions or collocations similar to group prepositions. Many are stored. Further, the English-Japanese collocation dictionary 45 stores a part of speech and a Japanese translation corresponding to such an expression form.
Since the configuration provided with the English-Japanese collocation dictionary 45 is a characteristic part of the machine translation apparatus of the present embodiment, the expression form stored in the English-Japanese collocation dictionary 45 is specifically illustrated here. Deepen understanding.

"Collocation" is, for example, a BBI collocation dictionary (Th
e BBI Dictionary of English WordCombinations, M.Ben
son, E. BEnson, R. Ilson, 1997, John Benjamins). The BBI collocation dictionary includes grammatical collocations and lexical collocations (Lexica).
The collocations are arranged as (l Collocations), and these collocations are stored in the English-Japanese collocation dictionary 45. further,
In addition to these collocations, in the present embodiment, rhetorical collocations (Rhetorical Collocations) and affixing collocations (Affixa
The collocational expression, which can be referred to as "action collocations", is also stored in the English-Japanese collocation dictionary 45. Specific examples are shown below.

First, a specific example of rhetorical collocation is shown. α Parallel words with opposite words increase and decrease (increase / decrease) present or absent (attendance) a client and a server (client / server) These are idioms of two-letter kanji in Japanese. , Strength, good and evil. β Parallelism of synonyms immorality and criminality describe and illustrate γ Parallelism of related words receive and store plug and play (plug and play) ) Δ Repetition of the same word Compound deveces and devices drivers The key and key cylinder Next, specific examples of affix type collocations are shown for each part of speech. 1. Noun 1-1-a Form of -ty and -ty connectivity and reliability quantity and quality 1-1-b Form of -ty or -ty equality and inequality (equation and Inequality) 1-2-a-ion and -ion format allocation and deallocation addition and deletion 1-2-b or-ion format acceleration or decelerarion 1-3-a Form of -or and -or exterior and interior (noun) (noun) (adjective) 1-4-a Opposite pair a client and a server 1 -5 pairs of synonyms and related words direction and distance files and folders information and instructions 1-6 repetition of the same words devices and devices drivers (device and device drivers) Ba) Thus, it is the prefix or suffix stored as complex word pairs of words having the same. Here, V indicates a verb, ad indicates an adjective, adv indicates an adverb, and prep indicates a preposition. In addition, up and
Verbs that take both down are likely to be used in collocational expressions. For example, scale up or down. Since the following verbs take both up and down, these verbs are likely to be collocations in the form of V up and down or V up or down.
It is stored in the English-Japanese collocation dictionary 45.

Ask bang batter bid bear beat belt bid blind blaze bring blow brush back boil buckle bolt buff build bundlebear bunk burn belt break button buy call close crack cut go hang hold jump knock lay let lie lock look make mark paste push rub scale screw 2. scrub set shacke show sit slip splash stand step tear tie touch tumble turn wash write 3. Adjectives positive and negative (input terminals) a syntax and a semantic (data type) a first and a second (region) both active and passive (dispatching) Note that () is a specific example of a noun following an adjective. Adverbs individually and arbitrarily Group Preposition and Group Preposition in Collision with in Comparsion With, etc. As described above, the English-Japanese collocation dictionary 45 stores the parts of speech corresponding to each of the stored collocations. In this way, by referring to the English-Japanese collocation dictionary 45 in the syntax analysis, the part of speech of the collocation can be specified, and a well-known syntax analysis described later is realized.

The English-Japanese collocation dictionary 45 stores Japanese translations corresponding to the stored collocations. A specific example of this Japanese translation is shown below. 1. Noun collocation (1) Collocation composed of antonyms In this case, “X and Y” is basically used as the translation of “X and Y”.

Connection and disconnection "connection and disconnection, connection and disconnection" addition and deletion "addition and deletion" creation and destruction "creation and destruction""creation and destruction" Since there is no problem, the translation of "ya" is appropriate, but for devices and elements, the translation "to" is assigned in the case of a pair of opposite functions. It is like the following example.

Encoder and decoder (2) -1 Collocations composed of synonyms illustration and description “Examples and explanations including drawings” the invention and its practical application “The present invention and its embodiments” '' The organization and operation of the invention
"Structure and operation of the invention" section and subsection "Sections and subsections" If the word is polysemous, prepare a plurality of translations.

(2) -2 A collocation composed of synonyms and indicating chain, linkage, and continuation of actions and processes. In this case, it is basically necessary to prepare a translated word using a middle point (•). activation and initialization `` start / initialization '' start location and destination location `` start location / destination '' disconnection and reconnection `` disconnection / reconnection '' error detection and correction `` error detection / error correction '' distribution and redemption `` distribution / exchange '' extraction and purification “Extraction and purification” Verb collocations (1) Collocations indicating chaining, interlocking, and continuation of actions and processing In this case, it is basic to prepare a translation using a middle point (•).

Select and activate “select and activate” Adjective collocation In this case, if there is a sense of incongruity in expression, a translated word "and" is provisionally prepared. For example, as shown in the following example.

Active and Passive Next, processing executed in the information processing apparatus 10 of the machine translation apparatus according to the present embodiment will be described. In the present embodiment, an example in which an English patent specification stored in a patent publication CD-ROM set in the CD device 15 described above is translated will be described.

First, the user displays a menu on the CRT display device 21 by performing a predetermined instruction from the input device 23, for example, clicking a file menu with a mouse. According to this menu, an English text file to be translated is specified from the input device 23. Here, when the English text file of the English patent specification stored in the patent publication CD-ROM is designated, the English text file is read into the memory device 39. At this time, the user selects patent translation from the menu. As a result, the optimal model for the patent specification is read from the relevance information model 53 described above.

After that, the information processing device 10 executes a morphological analysis process. This morphological analysis processing will be described based on the flowchart of FIG. FIG. 3 shows a subroutine incorporated in the lexical analysis module described above.
First, in the first step S100, words are cut out. In this process, for each sentence of the English text in the English patent specification text file read into the memory device 39, the lexical analysis, the punctuation mark identification, and the sentence segmentation identification are sequentially executed from the top. By executing this processing, individual sentences are identified from the entire text, and words / punctuation marks are extracted as tokens. In addition,
This analysis / identification processing is known as a morphological analysis rule for English sentences.

At S110, array configuration data is formed. In this process, a serial number and an attribute, and a serial number of a sentence to which the word belongs are assigned to words cut out from the entire text. Label a word if it is at the beginning of a sentence. That is, it is stored as concordance data having an array structure. In this process, a paragraph is identified. For example, if a sentence has a line feed code at the end of the preceding sentence and there are two or more blank codes between the preceding sentence and the sentence, the sentence following the blank code is recognized as a paragraph start sentence and a paragraph serial number is assigned. It is.

Then, they are arranged as a data table having a hierarchical structure of the above-described paragraph serial number, sentence serial number, and word serial number. Analysis of phrase composition is performed only at comma positions. Through the processing of S110, the structure of the English text is extracted as a hierarchical array data structure, and the number of paragraphs is detected.

In S120, an editing process is performed. This processing is for extracting relevance information and editing text based on the selected relevance information model. It is determined whether or not the adverb / preposition phrase is located at the beginning of the sentence, and whether or not the first word serial number of the adverb (phrase) / preposition phrase has a head prefix. Then, it is determined whether or not the adverb (phrase) / preposition phrase at the beginning of the sentence is a predetermined phrase. This determination is a process for detecting whether there is a phrase presenting information related to the context at the beginning of the paragraph start sentence.

In an actual English sentence, in particular, a compound sentence is often used to express the interrelationship between two events and ideas, but the main clause and the subordinate clause are subordinate, as the grammar. It does not necessarily have a structure connected via a conjunction. However, a compound sentence which is an English sentence having a subordinate relation (Hypotaxis) is reduced to a simple sentence combination from the viewpoint that all of the compound sentences can be regarded as a single sentence combination (Parataxis).

For example, in the case of a continuation usage related clause, the connecting sentence is deleted from the original sentence, the original sentence is divided into two sentences, and a connecting adverb is added to the subsequent sentence. In general formula, Clause A $
conjClause B. → Sentence A: $ Adv, Sentence B. or $ Conj Clause Y, Clause X. → Sentence Y: $ Adv, Se
ntence X. (Clause is a clause, $ conj is a conjunction,
$ Adv is a connective adverb, Sentence is a sentence).

In S130, a special rewriting process is performed.
In this process, a rewriting process specific to a text is performed based on the selected relevance information model. For example, in the claims (hereinafter referred to as "claims"), there are many complicated syntaxes, and when translated, the relations such as modifiers are in Japanese. Therefore, here, the rewriting process of the claim is performed based on the selected relevance information, that is, the optimal model for the patent specification. Hereinafter, the flow will be briefly described.

First, a claim portion is extracted according to the control code in the English patent CD-ROM. This claim portion often has a characteristic structure including a “transition phrase”. Therefore, the above-mentioned relevance information model includes comprisi
ng:, comprising in combination, the improvement com
Transition phrases such as prising are stored in the form of a table, and the claims are divided before and after the transition phrase with reference to this table. The relevance information model stores typical patterns of English language claims. Therefore, next, the claims are shaped according to this pattern, and the claims are hierarchized. Thereafter, the morphological analysis processing ends. By this morphological analysis processing, the unique hard-to-translate expression of the U.S. patent specification is rewritten into a standard English expression which is equivalent in information.

Next, a syntactic analysis process which is a characteristic process of the present embodiment will be described with reference to a flowchart shown in FIG. First, in the first step S200, an English text search process is executed. This process searches for an expression form composed of words connected by coordination conjunctions. For example, the expressions A and B, the expressions A, B and C called triade,
Alternatively, search for the expression form A or B in the original text.

In S210, it is determined whether or not the expression form searched in S200 is stored in the English-Japanese collocation dictionary 45 shown in FIG. When the searched expression form is stored in the English-Japanese collocation dictionary 45 (S210: Y
ES), in S220, the part of speech corresponding to the expression form is read out, and in S230, the expression form is recognized as one morpheme, and the process proceeds to S250. On the other hand, when the searched expression form is not stored in the English-Japanese collocation dictionary 45 (S210: NO), the process proceeds to S240.

In S240, it is determined whether or not the prefix or suffix of each word constituting the searched expression form is the same. Here, if it is determined that the prefix or suffix of each word is the same (S240: YES), the expression form is recognized as one morpheme in S230, and S250
Move to. On the other hand, when it is determined that the prefix or the suffix of each word is not the same (S240: NO), the process proceeds to S250. In the present embodiment, a prefix / suffix data table for determining the same prefix or suffix is prepared. This data table is prepared, for example, in the English-Japanese collocation dictionary 45 and summarizes what affixes are. Also, phonetic symbols may be stored in this data table in association with the affixes.

In S250, the syntax of the English text is analyzed. Here, expression forms recognized as one morpheme in the above-described processing are not separated as a unit. If the expression form recognized as one morpheme is stored in the English-Japanese collocation dictionary 45, S2
Reference is made to the part-of-speech read out in step 20 to analyze the syntax. Since the syntax analysis method is well known, a detailed description thereof will be omitted. Then, the main syntactic analysis processing ends.

Thereafter, the information processing apparatus 10 executes a semantic analysis process. In the semantic analysis processing, when an expression form recognized as one morpheme is stored in the English-Japanese collocation dictionary 45, a Japanese translation corresponding to the expression form is referred to. When the expression form recognized as one morpheme is not stored in the English-Japanese collocation dictionary 45, the meaning of each word constituting the expression form is referred to the English-Japanese dictionary 41 or the technical term dictionary 43, and the expression form is referred to. Translate. Subsequently, the information processing device 10 executes a text synthesizing process. In this process,
A document composition module generates a Japanese sentence from the parse tree that is the result of the syntax analysis described above and individual English words or collocation Japanese translations obtained by the semantic analysis processing. Also, in the sentence synthesis processing, if the generated Japanese sentence has a passive expression, it is converted into an active expression to shape the Japanese sentence.

In the present embodiment, as a Japanese sentence generation rule, a transformation rule for directly transforming an English syntax into a Japanese syntax or a translation method of a structure reduction transformation is adopted. Then, a translation rule for translating English sentences in the order of description is adopted. There is no so-called back-to-front translation method. For example, in the morphological analysis process, S
In the text edited in the process of 120, the sentence is divided at the connection phrase. For this purpose, the preceding sentence (the preceding section before editing) is forcibly translated first, and the succeeding sentence (the subsequent section before editing) is subsequently translated. As a result, since the translation process is executed in the order of the description of the original text, a Japanese translation according to the description order of the original text is generated and output. In the output Japanese text, the translated sentence of the following clause is
It is never written before the translation in the previous section. In other words, it is not a Japanese sentence that translates an English sentence from back to front.

As described above, in the machine translation apparatus according to the present embodiment, the user has designated a patent translation in accordance with the menu displayed on the CRT display device 21, and the translation is further displayed on the CRT display device 21. The menu “Patent Translation” has a plurality of options. For example, "patent translation (batch processing)", "patent translation (pre-processing and reading)", and "patent translation (no pre-processing)". Here, "preprocessing" means to edit a complicated English sentence into a simple English sentence, or to add a predetermined label to a specific expression. Therefore, when "patent translation (batch processing)" is selected, English text editing → morphological analysis processing (see FIG. 3) → syntactic analysis processing (see FIG. 4) → semantic analysis processing → translation generation → translation editing processing It is done. If "patent translation (preprocessing and reading)" is selected, English text editing → morphological analysis processing → syntactic analysis processing → semantic analysis processing → translation generation is performed, and then the processing is temporarily terminated. If "patent translation (no pre-processing)" is selected, pre-processing is not performed, and morphological analysis processing → syntactic analysis processing → semantic analysis processing → translation generation is performed on the specified file. Then, in the machine translation apparatus of the present embodiment, the English text file that has been executed up to the syntactic analysis processing is manually edited via the input device 23, and the edited text is again subjected to the morphological analysis processing → syntactic analysis processing. → Semantic analysis processing → translation sentence generation is possible.

Next, the effects of the machine translation apparatus according to the present embodiment will be described. In order to facilitate understanding of the description, first, problems in a conventional machine translation apparatus will be briefly described. 2. Description of the Related Art Conventionally, in machine translation, morphological analysis is performed to cut out an original text into words as morphemes, and then syntactic analysis is performed. Therefore, even if it is a so-called collocation, it is cut out mechanically into each word,
In syntactic analysis, English sentences are often separated and analyzed just before the coordination conjunction in the collocation. For this reason, errors in syntactic analysis are likely to occur, and the time required for syntactic analysis has been long.

On the other hand, in the machine translation apparatus of the present embodiment, an English-Japanese collocation dictionary 45 is prepared in the HD device 13, and the English-Japanese collocation dictionary 45 has a set of expression forms represented by so-called collocations. Is stored. Then, in the syntactic analysis processing, an expression form composed of words connected by coordination conjunctions is searched in the English text (S200 in FIG. 4), and is stored in the English-Japanese collocation dictionary 45 (FIG. 4). S210: YES) or, if the prefixes or suffixes of the words constituting the searched expression form are the same (S240: YES in FIG. 4), the searched expression form is regarded as one morpheme. Recognize and analyze the syntax without separation (S250 in FIG. 4). As a result, the occurrence of errors in the syntactic analysis can be suppressed, and the time required for the syntactic analysis can be reduced.

The English-Japanese collocation dictionary 45 contains B
In addition to the grammatical collocations and lexical collocations described in the BI collocation dictionary, the rhetorical collocations and the affixing collocations described above are also stored with specific examples, and these collocational expressions are also recognized as one morpheme . Furthermore, even if the expression form is not stored in the English-Japanese collocation dictionary 45, if the words constituting the searched expression form have the same prefix or suffix,
The searched expression form is recognized as one morpheme.
As a result, a group of expression forms existing in the English text can be surely recognized as one morpheme, and the syntax can be analyzed without separation.

The English-Japanese collocation dictionary 45 stores the part of speech and the Japanese translation of the collocation in correspondence with the stored collocation. As described above, by storing the part of speech corresponding to the collocation, accurate syntax analysis can be quickly performed in S220 in FIG. Also, by storing the Japanese corresponding to the collocations, it is possible to generate a Japanese sentence without a sense of incongruity in the document synthesizing process.

Furthermore, in the machine translation apparatus of the present embodiment, when selecting an English text file to be translated,
Since the most suitable model for the English text file to be translated is selected from the relevance information model 53, it is possible to execute analysis processing and translation processing that are optimal for each English text. For example, by selecting “Patent Translation” from the menu, the optimal model for the patent specification is selected from the relevance information model 53 and read. At this time,
As described above, “Patent Translation (batch processing)”, “Patent Translation (Pre-processing and reading)”, “Patent Translation (No pre-processing)”, etc. ing. With this option, it is possible to separately execute English sentence editing which is “preprocessing”, morphological analysis processing / syntactic analysis processing / semantic analysis processing, and translated sentence editing processing. In addition, the user can manually edit the English text and perform a morphological analysis process, a syntactic analysis process, a semantic analysis process, and a translated text editing process on the edited text.

A specific example of the translation process using the English-Japanese collocation dictionary 45 will be described below in comparison with a specific example of the conventional translation process. The following English sentences were translated. In one preferred embodim
ent of the present invention, five icons, eachrepr
esenting one of the subject matter categories, are
displayed and may be selected and activated by th
In a conventional machine translator, five icons (one representing each of the content categories) are displayed from this English sentence, "In one preferred embodiment of the present invention, You could choose to activate the viewer through the remote control. "

1) The analysis of the back and failed, and the auxiliary verb
It does not control activated. 2) In addition, by the viewer manipulating the remote
Control could not be analyzed as a semantic subject of the passive voice, so the viewer was used as the object of activated and the active tense was analyzed as a past tense.

3) Therefore, by was analyzed as an adverb and it was translated as "through". If, on the other hand, a collocation dictionary is used, from this English sentence, "in one preferred embodiment of the invention, five icons (one representing each of the content categories) are displayed and the remote control is displayed. It may be selected and activated by the operating viewer. "

1) Selected and activated are passive,
Analyzed as being dominated by may. The translation of may is also correct. 2) By is correctly analyzed as a preposition that takes the passive subject as an object.

The English-Japanese collocation dictionary 45 in the HD device 15 described above corresponds to “expression form storage means”, and the central processing unit 31 of the information processing apparatus 10 includes “search means”, “syntax analysis means”, and “translation”. Means "and" translation sentence editing means ".
The CRT display device 21 and the input device 23 correspond to “editing means” and “mode switching means”. And FIG.
The process of S200 in FIG. 4 corresponds to a process as a search unit, and the process of S250 in FIG. 4 corresponds to a process as a syntax analysis unit.

As described above, the present invention is not limited to such embodiments at all, and can be implemented in various forms without departing from the gist of the present invention. For example, it is conceivable to add the following expression form to the English-Japanese collocation dictionary of the above embodiment.

A. As phr, a simile of emphasis
ase is an idiomatic expression busy as bees "busy like a bee" cold as ice "cold like ice" cool as cucumber "calm down" white as snow "white as snow" Even if the expression is not stored in the English-Japanese collocation dictionary 45, if the adjective ad and the noun N have the same prefix or suffix in the form of ad as N, An expression form is also conceivable. It is like the following example.

Dry as dust “dry as dust” still as a stone “does not move as stone” b. What is in the form of Z prep Y of X Z (V / N) prep X (unit) of Y "Y of X Z" attack by a large swarm of bees of whales "Observation of a group of whales" V indicates a verb and N indicates a noun.

The unit is an English-specific expression representing a set of specific substances, animals, plants, symbols, and the like. a colony (swarm) of bees a herd of buffaloes a herd of buffaloes a herd of buffaloes a pack of dogs a pack of dogs a bounquet of flowers a bunch of A pride of lions "a group of lions, a group of lions" a school of whales "a group of whales, a group of whales"

A bit (piece, word) of advice an article of clothing an act of violence a set of instructions a pair of ~ a couple of ~ In the above-described embodiment, in the syntactic analysis processing shown in FIG. In S240, it is determined whether the prefix or the suffix is the same. Alternatively, it may be determined whether the prefix or the suffix is the same as the acronym or the rhyme together with the prefix or the suffix. . In this case, based on the phonetic symbols stored in the English-Japanese dictionary 41 and the technical term dictionary 43, it may be possible to determine the identity of the adjectives and rhymes, or as described above, the data on the prefix / suffix. If phonetic symbols are stored in the table, the data table may be referred to determine the identity of alliteration and rhyme. In this way, exceptional forms that do not form a united form even if the words have a common prefix or suffix can be eliminated by determining the commonality of alliteration or rhyme. And the occurrence of errors in syntactic analysis can be effectively suppressed.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of a machine translation device according to an embodiment.

FIG. 2 is an explanatory diagram illustrating an HD device according to the embodiment.

FIG. 3 is a flowchart illustrating a morphological analysis process performed by the information processing apparatus.

FIG. 4 is a flowchart illustrating a syntactic analysis process performed by the information processing apparatus.

[Explanation of symbols]

Reference Signs List 10 information processing device 13 hard disk device 15 CD-ROM device 17 printer 19 optical disk device 21 CRT display device 23 input device 25 flexible disk device 39 memory device 31 central processing unit 41 English-Japanese dictionary 43 ... Technical term dictionary 45 ... English-Japanese collocation dictionary 47 ... Program storage area 49 ... Analysis result storage area 51 ... Translation sentence storage area 53 ... Relevance information model

Claims (12)

[Claims] 1. A machine translation device that converts an input original text into a target language and outputs a translated sentence, wherein a word having a part common to an array of character data is connected by a coordination conjunction. An expression form storing means for storing at least a form; a search means for searching in the original text for an expression form composed of words connected by coordinating conjunctions; and an expression form searched by the search means, A machine translation device comprising: a syntax analysis unit that, when stored in a form storage unit, recognizes the expression form as a morpheme and analyzes the syntax. 2. The machine translation device according to claim 1, wherein the expression form stored in the expression form storage means is composed of words having a common initial or rhyme in addition to the character data array. A machine translation device, comprising: 3. The machine translation apparatus according to claim 1, wherein the expression form storage unit further stores an expression form composed of a preposition pair or an adverb pair. apparatus. 4. The machine translation apparatus according to claim 1, wherein the expression form storage unit further stores an expression form composed of a pair of synonyms or antonyms. Machine translation equipment. 5. The machine translation device according to claim 1, wherein said expression form storage means is associated with said expression form,
The part-of-speech of the expression form when the expression form is regarded as a group is stored, and the parsing means analyzes the syntax by referring to the part-of-speech of the expression form stored in the expression form storage means. A machine translator characterized by being configured as described above. 6. The machine translation apparatus according to claim 1, wherein the search unit further searches for an expression form composed of words having the same prefix or suffix, The machine translation device is characterized in that the analysis means is configured to recognize the expression form having the same prefix or suffix searched by the search means as one morpheme and to analyze the syntax. 7. The machine translation device according to claim 1, wherein said search means includes a word having the same suffix or prefix, and further having a common acronym or rhyme. A form is searched, and the parsing unit recognizes an expression form having the same prefix or suffix and the same algebra or rhyme as one morpheme searched by the searching unit, and performs syntax analysis. A machine translator characterized by being configured as described above. 8. The machine translation apparatus according to claim 6, wherein the expression form searched by the search unit is:
A machine translation apparatus comprising: a storage control unit that, when not stored in the expression form storage unit, additionally stores the expression form in the expression form storage unit. 9. The machine translation apparatus according to claim 1, further comprising: an expression form storage unit, wherein the expression form storage means corresponds to the expression form;
A target language translation of the expression form is stored. By referring to the target language translation stored in the expression form storage means, the original text analyzed by the syntax analysis means is converted into a target language, and the translated text is translated. Machine translation device, comprising: a translation unit that outputs a translation. 10. The machine translation apparatus according to claim 9, further comprising a translation editing unit that edits the translation output by the translation unit. 11. The machine translation apparatus according to claim 9, further comprising an editing unit for allowing a user to edit the input original text or the original text analyzed by the syntax analysis unit. The machine translation device is characterized in that the translation means is configured to convert the original text edited by the editing means into a target language and to output a translation. 12. The machine translation device according to claim 11, further comprising: a mode for outputting an analysis result by the syntax analysis unit; a mode for converting an output from the syntax analysis unit into a target language; A machine translation device comprising a mode switching unit for allowing a user to selectively switch between a mode for converting an output from a target language into a target language.