JPH0855123A - Machine translation system with idiom registering function - Google Patents

Abstract

PURPOSE:To provide a machine translation system capable of corresponding also to a modification form of a fixed part in an index word of an idiom and leading a user definition mark into a variable part in respect to a machine translation system with an idiom registering function. CONSTITUTION:The machine translation system is provided with an idiom registering means 2 for registering an entry word and a translated word of an idiom so that the fixed part of the idiom is expressed by a normal word, a normal word string or a modification developing mark representing the modified expression of the word or the word string and the variable part of the idiom is expressed by the composite format of a 1st representative mark representing the set of words or word strings sharing a prescribed attribute, a modification form developing means 10 for generating and developing the fixed part of the index word of an idiom to be translated to all previously set modified expressions and an idiom translation means 4 for identifying an input character string or a part of the character string with the index word or the like obtained by developing the fixed part of the index word of the idiom is developed to a modified expression.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an electronic dictionary, an information retrieval device equipped with the electronic dictionary, or a machine translation device equipped with the electronic dictionary, and in particular, registers and retrieves / translates an idiom including a variable part. The present invention relates to a dictionary search device having a possible idiom registration function.

[0002]

2. Description of the Related Art Language processors currently in practical use include word processors for supporting human document creation activities, and machine translation devices for translating documents written in one language into another language. and so on. These language processing devices are provided with dictionaries that store information according to their respective purposes. The term "dictionary" used here refers to a system that integrates a large number of headwords and a set of various types of information attached to the headwords as one unit so that desired items can be easily searched using the headwords. They are arranged upright.

The dictionary is recorded in a machine-readable format on a machine-readable non-volatile medium in principle. Such a dictionary will be called an electronic dictionary. When the electronic dictionary is used in machine translation, a word string of the original word (source original word) (including one word only) is used as the found word, and various information attached to the word string is used as the found word. Information such as the part of speech of the word, the form attribute, the translated word, and the part of speech of the translated word is used.

If the document to be processed or created by the user using such a language processing device includes a word that is not described as a headword in the dictionary provided in this device, The work efficiency will be significantly reduced. Therefore, it is preferable that the number of headwords included in the dictionary is larger. Also, in the case of machine translation, it is recommended to use not only each word in the original language but also an idiom as a found word, and the corresponding target language wording, etc. as a pair and register as many such pairs as possible. It is desirable in terms of translation efficiency.

Usually, many idioms, such as numbers, possessive pronouns, and reflexive pronouns, change their forms in relation to the subject and other words. For example [do one's bes
One's in t] is a possessive pronoun such as your, my, his, her depending on the subject. For efficiency of translation processing and dictionary development, such an idiom does not enumerate all idioms in which specific words are entered, but any idiom that shares a certain grammatical feature as a part of a headword is optional. It is preferable that the word can be registered in such a way that it can be entered. In the following, if a word or phrase that shares a certain grammatical feature is included, a part that contains an arbitrary word is called a variable part, and a part of a word or word string that is the skeleton of an idiom is called a fixed part.

Some idioms have a plurality of variable parts, and in some cases not only words but also noun phrases and sentences are applied to the variable parts. In order to express such various variable parts, a machine translation device that introduces a symbol following * (this symbol is hereinafter referred to as a representative symbol) has been proposed.

For example, in the case of registering the English word idiom "N timesas ... as ~" having the Japanese translation "... than N times ...", it can be described as follows. English word [* m times as * ad as * CN] Translated word * CN times * m times more than * CN Heading (word string), word with "*" at the beginning
(* m, * ad, * CN), that is, the representative symbol is the variable part,
The other words (times, as) are the fixed part.

The variable part can be matched with any word of the part of speech represented by the representative symbol (m: number, ad: adjective, C: sentence, N: noun phrase, CN: C or N in the above example). , The fixed part cannot match unless it has a word with that notation. The representative symbol that can be designated for the variable portion is defined in advance in the system.

Now, [This apple is three times as big as that orange.
] Is input, the following is an outline of machine translation processing. First, the dictionary is looked up for each word.
As a result, the following word information is obtained. three number (m) big adjective (ad) that article (d) orange noun (n)

Next, an idiom search and analysis process is executed, and matching of the fixed part and the variable part is performed. At this time, it can be seen that part of the input sentence matches the idiom registered as described above. Here, the variable part * m and “three” match, and * ad
And “big” match, * CN and “that orange” match. Also, "times as" and "as" match as a fixed part (# sign part) of the idiom.

Next, a translation of the variable part of the idiom is generated. * CN That orange * m 3 * ad Large Finally, a Japanese sentence for the input sentence is generated by combining with the translation of the fixed part.

[0012]

However, in the machine translation system described above, (1) the utilization change cannot be written in the fixed part of the idiom heading. (2) Symbols other than the system-defined representative symbols cannot be used in the variable part of the idiom heading. There are two restrictions. As a result, the number of headwords registered in the idiom increases, resulting in an increase in storage capacity and search time, and a problem that the user, that is, the person who registers the idiom in the dictionary, has a heavy burden.

Since the matching of the fixed part of the idiom headword consisting of the fixed part and the variable part and the input sentence is compared only by the character strings, nouns that change in multiple ways, verbs that change in utilization,
In the case of input sentences that include auxiliary verbs, adjectives, and words added by auxiliary verbs and negative adverbs (not), simply registering the idiom in the basic form that is often used will cause matching to fail. Therefore, it is necessary to enumerate all the variations shown above as different headings.

The English idiom "There is no ~ ing." Having a Japanese translation "I can't do" will be described as an example. This idiom is registered in the following format. English word [There is no * Ving] I can't translate * Ving. Here, “* Ving” is a representative symbol representing “progressive form of verb”.

However, "There was no going to school
The sentence "yesterday." is a past tense sentence even though it contains this idiom, so "There was no" in the sentence.
Cannot match the fixed part of the heading, "There is no". In other words, in the case of an idiom that includes words that change in inflection such as verbs, it is necessary to expand and register all inflections. In this example, [There was no] and [There will
I had to register another idiom with the heading be no.

As the variable part of the idiom heading consisting of the fixed part and the variable part, only the part defined in advance by the machine translation system is used. It is impossible to predict all the demands of users and prepare all representative symbols in a commercialized system that is targeted by a wide range of users. It is because there was a problem in efficiency to define it.

In such a conventional machine translation system, when the desired representative symbol is not defined, the English word idiom "as-as ... can" having a Japanese translation "as much as possible" is taken as an example. explain. The word that can be put in "~" between two as is an adjective original class or a noun phrase modified by an adjective original class, but if the representative symbol defined in the system is "noun including adjective phrase"
If there is no such symbol, the user has no choice but to use the closest representative symbol defined by the system, for example, the idiom specified as * N (noun phrase) as follows. English words [as * N as * n * 3 can] Translated as much as possible * N where * 3 is the representative symbol for the nominative pronoun.

However, in such an approximate registration, even a sentence that does not actually contain an idiom is erroneously recognized,
Translation accuracy may be reduced. For example, [I regard the person as policeman as you can see.
I know you, so I see him as a policeman. The input sentence is not originally an idiom sentence because the noun phrase between the two as does not include an adjective, but in the above registration So policeman matches * N, and mistakenly recognizes that this statement contains an idiom.

Furthermore, as policem, which is interpreted as an idiom
The erroneous analysis that the antecedent of the relative pronoun in which an as you can is the subject and the see is a verb is the person happens to occur, and as a result, the following mistranslation is generated. "I will consider who the policeman sees as much as possible."

Since only the representative symbols defined by the system can be used in this way, only approximate designation can be made, and there may be a case where an incorrect translation is not intended by the user. Further, if the specification is detailed, the headline is often complicated. This becomes particularly remarkable when a plurality of similar detailed conditions appear. As described above, since the conventional machine translation device has the above-described two restrictions, the number of headwords registered in the idiom increases, resulting in an increase in storage capacity and search time, and a burden on the user. There was a problem that it became large.

Therefore, the present invention has been made in consideration of the above circumstances, and it is possible to cope with a variation of the fixed part of the idiom headword, and to introduce a user-defined symbol into the variable part. An object of the present invention is to provide a machine translation device having an idiom registration function that can perform.

[0022]

FIG. 1 shows a basic block diagram of the present invention. In the figure, the present invention comprises an input means 1 for inputting a character string and a symbol, a fixed part consisting of a predetermined word or word string, and a variable part capable of changing to a word or word string having a common attribute. For the idiom, the fixed part is represented by a normal word, a word string, or a change expansion symbol that represents a modified expression of the word or word string, and the variable part shares a predetermined attribute. An idiom registration means 2 for registering an entry word expressed in a format in which a first representative symbol representing a set of columns and a translation of the idiom are registered, and a dictionary and a processing result necessary for idiom registration and translation processing are stored. Storage means 9, a dictionary lookup / morpheme analysis means 3 for decomposing the input word string into morphemes and performing grammatical analysis, and the idiom headword for translation Variant expansion means 10 for generating and expanding a constant part into all preset variant expressions, an input character string or a part thereof, and a idiom entry word registered or fixed by the variation expansion means 10. An idiom translation means 4 for identifying a headword whose part is expanded into a modified expression and generating a translated word of a character string corresponding to the identified headword of the idiom, a syntax analysis means 5, and a syntax conversion means 6.
And a translated sentence generating means 7 and an output means 8 for outputting the translated sentence.
The present invention provides a machine translation device having an idiom registration function, which comprises:

Further, the variation expansion means 10 uses an expression form in which a fixed part of an idiom headword is changed by utilizing words constituting the fixed part or an expression form in which an auxiliary verb or a negative adverb is connected to the fixed part. It may be generated and expanded.

Further, a representative symbol registration means 11 for registering a second representative symbol representing a word or a word string having a newly defined attribute and the attribute value in the storage means 9, and a heading of an idiom to be translated. A representative symbol expanding means 12 for generating and expanding the second representative symbol included in a variable part of a word into a defined attribute and its attribute value is provided, and the idiom registration means 2 is inputted by the input means 1. The idiom headword and its translation expressed by using the first representative symbol and / or the second representative symbol are registered, and the idiom translating means 4 causes the attribute of the input character string or a part thereof and the attribute value thereof. And the representative symbol expansion means 1
The attribute of the second representative symbol of the entry word generated and expanded by 2 and its attribute value may be identified.

Further, the storage means 9 has a dictionary memory 9a having grammar and translation information for translating an input character string, a buffer memory 9b for storing a result of processing up to translation generation, It is preferably composed of an idiom registration memory 9c for storing the idiom registered by the idiom registration means 2.

The dictionary lookup / morpheme analysis means 3 decomposes the input character string into words to generate part-of-speech information of each word, and a translation word extraction part 3b to generate translation candidates for each word. It is preferable to be composed of

The idiom translation means 4 searches the idiom registration memory, selects an idiom finding word candidate that can match the decomposed word string and expression form, and a representative of the idioms. An idiom identifying unit 4b that identifies the found word of an idiom whose attribute corresponding to the position of the symbol corresponds to the attribute given to the representative symbol from one of the idiom candidates, and the representative symbol An idiom analysis unit 4c that analyzes the syntax of a word or a word string to generate a sentence structure of the entire idiom, and an idiom translation word generation unit 4d that generates a translated word of the idiom part of the input word string based on the sentence structure of the idiom. It is preferable to be composed of

Here, as the input means 1 in FIG. 1, a keyboard, or a pointing device such as a mouse, a pen or a trackball is used, but the input means 1 is not limited to this, and other input devices may be used. . The storage means 9 is usually a ROM, a RAM, a floppy disk or a hard disk,
The storage device is not limited to this, and other storage devices may be used. Especially, the dictionary memory 9a is preferably a ROM, and the buffer memory 9b, the idiom registration memory 9c and the representative storage memory 9d are preferably RAM.

A CPU is usually used as the idiom registration means 2, the representative symbol registration means 11, the dictionary lookup / morpheme analysis means 3, the idiom translation means 4, the syntax analysis means 5, the syntax conversion means 6 and the translation sentence generation means 7. Used, RO
It is preferable that a microcomputer including peripheral circuits such as M, RAM, and I / O interface is used, and that the ROM or RAM contains a program for controlling the operation of the document processing apparatus.

[0030]

When an idiom headword and a translated word are registered, the fixed part of the idiom is represented by a normal word, a word string, or a change expansion symbol that represents a modified expression of the word or words, and the variable part of the idiom is predetermined. The idiom registration means 2 registers in the storage means 9 by being expressed in a composite form of the first representative symbols that represent a word or a set of word strings that share the attribute of.

With this, it is possible to register a plurality of idiom headwords in which the variable portion has the same attribute and the fixed portion may be variously modified and expressed by one headword. That is, according to the present invention, it is not necessary to register all possible patterns for the fixed part and the variable part as idiom headwords, and it is possible to suppress an increase in registered idiom headwords.

The input character string is decomposed into words, and a candidate of an idiom headword that can match a part of the decomposed word-by-word expression form is searched, and further changed into a fixed part of the headword. When there is an expansion symbol, the modified expression generated and expanded by the variation expansion means and the word or word string in charge of the position where the expansion symbol is present are identified to identify the idiom.

As described above, according to the present invention, a word or word string corresponding to a fixed part of an idiom in an input sentence is completely different from a word or word string registered as an entry word of the idiom. An idiom can also be identified if it does not match, but matches its variant.

The variable part of the idiom is expressed in a format including a newly defined attribute and a second representative symbol representing a word or word string having the attribute value, and the storage means 9
Be registered with.

Then, the input character string is decomposed into words, and a candidate for an idiom headword that can match a partial expression form of the decomposed word string is searched for, and further, in the variable part of the headword. If there is a second representative symbol in the attribute, the attribute and its attribute value generated and developed by the representative symbol developing means, and the attribute of the word or word string in charge of the position where the second representative symbol exists and its attribute Identify with the value,
Identify the idiom.

As described above, according to the present invention, the variable part of the idiom headword is expressed by using the second representative symbol representing the word or word string having the newly defined attribute and the attribute value. In addition, since the idiom is identified, the length of the entry word registered as an idiom can be suppressed, and at the same time the user can register the idiom that is necessary for the user himself or has an expression form peculiar to the text of a certain field. Can be identified.

[0037]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below based on the embodiments shown in the drawings. The present invention is not limited to this. Before describing the embodiments, the concept of machine translation will be briefly described. Referring to FIG. 2, the analysis processing performed in machine translation has various analysis levels. The machine translation is a process for sequentially performing the processing of each level and finally obtaining the target language shown on the right side of FIG. 2 when the source language shown on the upper left side of FIG. 2 is input. That is, when the source language is input, first, the level L1 dictionary lookup process, the level L2 morphological analysis process, the level L3 syntactic analysis process, ...
Finally, the morpheme generation process of level L10 is performed to generate the target language.

Machine translation is roughly divided into the following two types depending on the level of analysis processing to be performed. First
Parses up to the intermediate language, which is a concept independent of both the source language and the target language shown at level L6, and from there, generates context at level L7, generates meaning at level L8, generates syntax at level L9, and generates at level L10. This is a pivot method that proceeds to morpheme generation and generates a target language. Second, the above-mentioned level L2 morphological analysis,
The level L3 syntax analysis, the level L4 semantic analysis, and the level L5 context analysis are performed to obtain the internal structure of the source language, and then the target language at the same level as the obtained internal structure of the source language. This is a transfer method in which the target language is generated after conversion into the internal structure of.

The contents of each analysis process shown in FIG. 2 will be described below. (1) Dictionary lookup, morpheme analysis Here, a sentence input while referring to a dictionary in which morphemes are stored is divided into morpheme strings (word strings), and grammatical information such as a part of speech for each word and a translated word are obtained, Further, processing for analyzing tense, person name, number, etc. is performed.

(2) Syntactic analysis Here, a process for determining the structure of a sentence (structure analysis tree) such as dependency between words is performed. (3) Semantic analysis Based on the results of a plurality of structural analyses, processing is performed to discriminate between semantically correct and improper ones. (4) Context Analysis In the context analysis process, a process of understanding the topic of the input sentence and removing abbreviated parts and ambiguities contained in the input sentence is performed.

Next, a block diagram of the machine translation apparatus according to the embodiment of the present invention shown in FIG. 3 will be described. In the figure, 31 is a main CPU (central processing unit), 32 is a main memory, 3 is a display device including a CRT (cathode ray tube) or LCD (liquid crystal display device), 34 is a keyboard,
Reference numeral 35 is a translation module, 36 is a dictionary memory connected to the translation module 35 for storing translation dictionaries, grammar rules, tree transformation structure rules, and the like, and 37 is a bus connecting the above-mentioned components.

The dictionary memory 36 also has a storage area in which idioms and representative symbols uniquely defined by the user can be stored. The CPU 31 controls an idiom registration process, a representative symbol registration process, and a translation module 35 process described below.

When a source language sentence is input, the translation module 35 translates the sentence in a predetermined procedure and outputs the target language. Ie keyboard 3
The source language input from 4 is sent to the translation module 35 under the control of the main CPU 31. The translation module 35 uses the dictionary, grammar rules, tree structure conversion rules, etc. stored in the dictionary memory 36 to translate the input source language into a target language as described in detail later. The result is temporarily stored in the main memory 32 and displayed on the display device 33.

FIG. 4 shows a block diagram of the translation module 35. The translation module 35 is connected to the bus 37, translates a source language input via the bus 37 according to a predetermined translation program, and uses the bus 3 as a target language.
7, a translation CPU 45 for outputting to 7, a translation program memory 46 connected to the bus 37 for storing a translation program executed by the translation CPU 45, and for storing the input source language original text for each word. Buffer A (40) and a buffer B for storing information such as the part of speech and translation of each word obtained by referring to the dictionary included in the dictionary memory 36 for each word stored in the buffer A (40). (41), a buffer C (42) for storing information about a source language structure analysis tree, and a buffer D (43) for storing a target language structure analysis tree converted from the source language structure analysis tree. ) And buffer D
A buffer E for storing a sentence prepared as a target language form by supplementing the structure analysis tree of the target language stored in (43) with an appropriate annex (for example, a particle or auxiliary verb in Japanese). 44) and.

Translation module 3 having the above configuration
5, the analysis is performed up to at least the level of the level L3 syntax analysis shown in FIG. Here, the translation program describing the translation processing procedure is composed of a dictionary lookup / morphological analysis unit, an idiom translation unit, a syntactic analysis unit, a syntactic conversion unit, a translated sentence generation unit, a variation expansion unit and a representative symbol expansion unit. It

The operation of the English-Japanese translation by the machine translation apparatus of this embodiment will be described below with reference to FIGS. Here, an example of an English sentence “This is a pen.” That does not include an idiom is given, and an outline of the operation of translating this English sentence into a Japanese sentence is shown.

First, the read original sentence is decomposed into morphemes by morphological analysis, and as shown in FIG.
(40) (see FIG. 4). Subsequently, the translation C based on the program stored in the translation program memory 46
Under the control of the PU 45, the dictionary look-up / morpheme analysis unit refers to the dictionary stored in the dictionary memory 36 for each word of the original sentence stored in the buffer A (40), and thereby translates each word. Information is obtained. For example, part-of-speech information, which is a part of the information, is stored in the buffer B (41) as shown in FIG.

Here, it is a multi-part-of-speech word of "this", and has two parts-of-speech, a pronoun and a referential adjective. The part of speech of "is" is a verb. Similarly, the parts of speech for "a" and "pen" are stored in the buffer B (41). "Thi
“S” is a multi-part-of-speech word, but what the part-of-speech in a sentence is is uniquely determined by the process corresponding to the parsing unit in the translation program.

In the process corresponding to the syntactic analysis unit of the translation program, a structure analysis tree showing the dependency relation between each word is shown in FIG. 7, for example, according to the dictionary and grammatical rules stored in the dictionary memory 36. Is decided. The result of this syntax analysis is stored in the buffer C (42) in FIG.

The determination of the structure analysis tree is performed as follows. From the grammatical rules stored in the dictionary memory 36, the following grammatical rules regarding English can be obtained. Sentence → Main part, Predicate Main part → Noun phrase Predicate → Verb, Noun phrase Noun phrase → Pronoun Noun phrase → Article, Noun

Of these rules, for example, the first rule is
"Sentence is composed of main part and predicate." Hereinafter, the structural analysis tree is determined according to these rules. It should be noted that such grammatical rules are similarly prepared for Japanese, and correspondence is made between English grammatical rules and Japanese grammatical rules.

In the process corresponding to the syntax conversion unit in the translation program, the structure of the structural analysis tree (see FIG. 7) of the input English sentence is shown in FIG. 8 using the tree structure conversion rule of the dictionary memory 36. Converted to the structure of the parse tree for the Japanese sentence shown. The obtained result is stored in the buffer D (43) shown in FIG. The example sentence "This is a pen." Used in this description is converted into the Japanese character string "this is a pen."

The portion of the translation program that performs the processing corresponding to the translated sentence generation unit is shown in FIG. 9 by adding an appropriate particle "ha" or auxiliary verb to the obtained Japanese character string "this is a pen". It is put in the Japanese form as shown and stored in the buffer E (44) of FIG. The obtained Japanese sentence “This is a pen.” Is output from the translation module 35 shown in FIG. 3, stored in the main memory 32, and displayed on the display device 33.

The above is an outline of the translation process of a sentence that does not include an idiom. In the translation process of a sentence that includes an idiom, in addition to the above process, idiom identification, analysis and translated word generation process in the idiom translation unit are performed. Further, in connection with the processing of the idiom translation unit, the processing of the variation expansion unit and the representative symbol expansion unit is performed.

Here, the variation expansion unit is, as will be described later,
Matching processing is performed on fixed parts of idioms, taking into account variations such as usage variations. The representative symbol expansion unit performs matching processing on the representative symbol uniquely defined by the user. The processing of each unit described above is executed by the translation CPU 45 of the translation module 35 according to the procedure of the translation program.

Next, the registration of an idiom in which a fixed part of a headword is expressed by a modified expansion symbol will be described with reference to FIGS. The modified expansion symbol is the following symbol. * Part-of-speech symbol (word) The "part-of-speech symbol" is for designating which part-of-speech is changed when the change target "word" is a multi-part-of-speech.

For example, the idiom "as to as can be" is described as follows. English word [as * a as * x (can) be.] Translated word * a As above, x is a part-of-speech symbol representing an auxiliary verb, and can as an auxiliary verb.
Even though the past tense could be, it represents this idiom. The part-of-speech symbol x is specified because can is a multi-part-of-speech word.

That is, can has a usage as a noun or a verb, and if no part of speech is specified, it changes as a noun (cans
This is because it is difficult to specify whether or not it should be done or canned as a verb.

Further, here, the last be has no change expansion designation. This could be either can or could,
This is because the be verb can have only its original form. in this way,
When there is a variation but a shape other than the headline cannot be recognized, the variation expansion designation is not performed. If there are more idiom registrations to be changed, the default change specification may be reversed and the change specification may be made.

The translation contents using this idiom will be described below by taking the case where there is an input sentence [I was as happy as could be.] As an example. FIG. 10 shows a flowchart of the translation process of the idiom part.

First, steps S111 to S11 in FIG.
In 2, the dictionary is sequentially searched from the first word of the input sentence. The idiom containing the representative symbol is registered in the basic dictionary like other words, so the third word (when s = 3)
The heading [as * aas * x (can) be] is searched while looking up the as dictionary (steps S113, S117, S11).
8).

Next, in step S114, matching is performed between each word in the idiom heading and the word in the input sentence. Since matching between a fixed part word in an idiom and a word in an input sentence can be processed at high speed only by comparing character strings, first, only the fixed part is matched.

In this example, [as * a as * x (can) be.] And “as happy as could b.
Matching processing of the fixed part between e. ”begins.

FIG. 11 shows a flowchart of the matching process of this fixed part. First, the variable p indicating the word number in the idiom is initialized (step S121). Step in S122, it is determined whether the word W _P word number p in the idiom are fixed part, not the fixed part, that is, if a variable portion, the step S13
2, the process proceeds to S133, and the process is repeated until all words are checked.

If the word W _P is a fixed part,
In step S123, it is determined whether the word W _P is the variation expansion designation. If the word W _P is a variant expansion designation,
The process proceeds to steps S125 and S126, and the matching process in which the utilization is changed is entered, but if the word W _P is not the variation expansion designation, the words W _P and W _S are compared (step S
124), if they match, the matching process for the next word is repeated (steps S130 to 133, S12).
2).

In the case of the above idiom, since the input word matches "as" at the beginning of the idiom heading, the processing moves to the next word "* a" in the idiom heading. (Steps S122-S124, S130-S133). Since “* a” is the representative symbol for the variable part, the next word “as”
Processing shifts to (steps S122, S132, S13).
3). Here, since "as" does not include the variation expansion designation, character string comparison is performed (step S124), and it is found that they match.

Next, the process moves to the matching of the next word of the idiom and the input sentence, that is, * x (can) and "could" (steps S130 to S133). Since * x (can) is a variation expansion designation, the process proceeds to step S125, and the utilization of "can" as an auxiliary verb is changed before the comparison of character strings is started.

In addition to the inflection changes such as tenses and singularities, additions of auxiliary verbs and "not" are also taken into consideration in the inflection change, the inflection table shown in FIG. Change. In this example, "can"
It can be seen that the input "can" is included in the "could" and "can" that have been expanded and expanded (step S126), so the next word is searched for matching (steps S130 to S1).
33).

The processing moves to the matching of the next input word (“be”) and be in each of the idiom and the input sentence. Similarly, it can be seen that they also match, and in the end, matching of the fixed portion is successful (ends in step S133).

Next, the variable part matching process (step S115) in FIG. 10 is started. FIG. 12 shows a flowchart of this variable part matching processing. First, the variable part is looked up in the dictionary (step S141). When matching the fixed parts, it is known that the target word of the representative symbol * a can be happy sandwiched between the fixed parts, so the dictionary of happy is searched.

Next, it is checked whether or not there is a user representative symbol in the representative symbols (step S142), and if there is, the user representative symbol is replaced with the definition body (step S14).
3). Next, it is checked whether or not it can be checked at the morpheme level (step S144). Since "* a" is a system-defined word part-of-speech, it is confirmed that the part-of-speech of "happy" is an adjective (step S145), and the matching of the variable part is completed.

If the target word of the variable portion is composed of a plurality of words, that is, if the representative symbol is a phrase part of speech, the syntactic process is called and the designated attribute is checked (step S146). With the above, it was found that the input sentence contains the idiom English word [as * a as * x (can) be.] Translated by * a, so finally, the translation of the idiom part as follows is generated, Store it in the translation buffer,
The idiom processing is completed (step S116). Idiom part [as happy as could be]

Further processing is repeated (step S11).
7, S118), the dictionary lookup is restarted from the next word of the idiom. In this example, the range of idiom extends to the end of the sentence, so dictionary lookup is completed at this point. After that, with respect to the word string of "I was" other than idiom, normal parsing and syntax conversion are performed, and finally, in the translation generation process, the Japanese translation "I was there" other than idiom The Japanese translation "I'm as happy as possible" is synthesized, and the following translated sentence is obtained. "I was happiest."

Next, an example of an idiom using a representative symbol registered by the user and a sentence translation process using the idiom will be described. The user shall use the following format to register a new representative symbol. <New symbol> “:: =” <Definition body> The representative symbol described in this format is CPU
It is stored in the dictionary memory 36 by 31.

The definition body is described as follows. "{"<Grammarattribute>"/"<value>","<grammarattribute>
"/"<Value>"," ... "}" As a grammar attribute and value (attribute value) that can be specified, the user is open to all parameters internally defined by the translation system. This makes it possible to define representative symbols using detailed grammatical and semantic constraints, which can be easily used for idiom registration. As the attribute and the value, for example, the attribute shown in FIG. 14 can be considered.

As described above, when the user can specify the attributes and values at various levels other than the morphological analysis, it is possible to check whether a certain word string is a representative symbol defined by the user only at the dictionary lookup stage. become unable. For example, whether or not the part of speech of a certain word string is a "noun phrase" cannot be known until the stage of parsing. The value of each attribute in FIG. 14 clearly indicates at which stage the check can be performed.

In the following, description will be made based on the idiom "as adjective a, no singular noun phrase as ... can""this is as good as it is". Since the word order of this idiom is special, it is necessary to specify the grammatical constraint "singular noun phrase without articles".

The registration processing and translation processing of the representative symbol representing the grammatical constraint will be described below. According to the table of FIG. 14, "singular noun phrase without article" is that the part-of-speech (cat) is a noun phrase (n), the inflectional form (inf) is singular (sg), and the usage (use) is no article (detnil). The representative symbol representing can be registered as * Nsg :: = {cat / n, inf / sg, use / detnil}.

By using this representative symbol, the above idiom can be easily registered as follows. English words [as * aa * Nsg as * x (can) be] Translated words * a * Nsg

Next, the translation process using this idiom will be described with reference to FIG. If there is [I bought as large a hat as could be.], First, the processes from idiom search to fixed part matching are performed in the same manner as in the above-described embodiment.

Next, the matching of variable parts will be described. In the example, since this idiom includes the representative symbol defined by the user (step S142), the user representative symbol is replaced with the definition body (step S14).
3).

In the case of the definition body of the representative symbol * Nsg of this example, it is expanded as {cat / n, inf / sg, use / detnil}. Referring to FIG. 14, of these, cat / n and inf / sg can be checked at the morphological analysis level (step S145), whereas use / detnil is an article anonym for the first time until syntactic analysis. It is understood (step S146). That is, in this example, "a hat"
Since the “a” of is included in the idiom heading, the only variable part is “hat”, which is an article noun (* Nsg).

After the matching process has succeeded and the idiom has been recognized as described above, the process proceeds in the same manner as in the above embodiment, and the translation of the idiom portion "a cap that is as large as this" is obtained. The idiom part [as large a hat as could be] Translated The largest hat In addition, the translated sentence of the whole sentence "I bought the largest hat" is obtained.

[0084]

According to the present invention, the fixed part of the idiom headword is registered by the expression including the change expansion symbol, and the fixed part expressed by the change expansion symbol is generated and expanded into the modified expression, and this modified expression is generated. In order to identify a word or a word string corresponding to the position where the change expansion symbol exists, a plurality of idiom headwords whose fixed parts may be variously modified are registered as one headword. be able to.
Therefore, it is not necessary to register all possible patterns for various modified expressions of the fixed part, it is possible to suppress an increase in the entry word of the registered idiom, and further it is possible to suppress the idiom storage capacity and the search time.

Further, according to the present invention, the variable part of the idiom headword is expressed by using the second representative symbol representing the word or word string having the newly defined attribute and the attribute value, and Since the idiom is identified, the length of the entry word registered as an idiom can be suppressed, and the idiom that is necessary for the user himself or has an expression form peculiar to the text of a certain field can be registered and identified. be able to.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of a machine translation device of the present invention.

FIG. 2 is a diagram schematically showing the concept of machine translation.

FIG. 3 is a configuration block diagram of an embodiment of the present invention.

4 is a block diagram of a translation module 35 shown in FIG.

5 is a diagram schematically showing the stored contents of a buffer A. FIG.

FIG. 6 is a diagram schematically showing stored contents of a buffer B.

FIG. 7 is a diagram schematically showing stored contents of a buffer C.

FIG. 8 is a diagram schematically showing stored contents of a buffer D.

FIG. 9 is a diagram schematically showing stored contents of a buffer E.

FIG. 10 is a flowchart showing dictionary lookup / idiom processing.

FIG. 11 is a flowchart showing matching processing of a fixed part.

FIG. 12 is a flowchart showing matching processing of a variable portion.

FIG. 13 is a diagram schematically showing a variation table.

FIG. 14 is a diagram schematically showing attributes that can be designated as a representative symbol.

[Explanation of symbols]

 1 input means 2 idiom registration means 3 dictionary lookup / morphological analysis means 3a part-of-speech extraction section 3b translated word extraction section 4 idiom translation means 4a idiom search section 4b idiom identification section 4c idiom analysis section 4d idiom translation word generation section 5 syntax analysis means 6 syntax conversion Means 7 Translated sentence generation means 8 Output means 9 Storage means 9a Dictionary memory 9b Buffer memory 9c Idiom registration memory 9d Representative symbol registration memory 10 Variant expansion means 11 Representative symbol registration means 12 Representative symbol expansion means 31 Main CPU 32 Main memory 33 Display Device 34 Keyboard 35 Translation module 36 Dictionary memory 37 Bus

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Takeshi Kitsumi 22-22 Nagaike-cho, Abeno-ku, Osaka-shi, Osaka Prefecture

Claims (3)

[Claims] 1. An idiom comprising an input means for inputting a character string and a symbol, a fixed part consisting of a predetermined word or word string, and a variable part capable of changing to a word or word string having a common attribute. On the other hand, the fixed part is expressed by a normal word, a word string, or a change expansion symbol that represents a modified expression of the word or word string, and the variable part is a set of words or word strings that share a predetermined attribute. An idiom registering means for registering a headword expressed in a format in which a first representative symbol representative of and a translation of the idiom, and a storage means for storing a dictionary and a processing result necessary for idiom registration and translation processing , A dictionary lookup / morpheme analysis means for decomposing the input word string into morphemes and performing grammatical analysis, and a fixed part for the idiom headword to be translated is set in advance. The modified form expansion means for generating and expanding all the modified expressions, the input character string or a part thereof, and the registered idiom headword or the modified part of the headword is expanded to the modified expression. Idiom translation means for identifying the identified headword and generating a translated word of the character string corresponding to the identified idiom headword, syntactic analysis means, syntactic conversion means, translation sentence generation means,
A machine translation device having an idiom registration function, comprising: an output unit that outputs a translated sentence. 2. The variation expansion means transforms a fixed part of an idiom headword into an expression form in which a word constituting the fixed part is utilized or changed or an expression form in which an auxiliary verb or a negative adverb is connected to the fixed part. A machine translation device having an idiom registration function according to claim 1, wherein the machine translation device generates and develops. 3. A representative symbol registering means for registering a second representative symbol representing a word or a word string having a newly defined attribute and its attribute value in the storage means, and a headword of an idiom to be translated. A representative symbol expanding means for generating and expanding the second representative symbol included in the variable portion into a defined attribute and its attribute value, wherein the idiom registration means receives the first representative input by the input means. The idiom headword and its translated word expressed by using the symbol and / or the second representative symbol are registered, and the idiom translation means stores the attribute of the input character string or a part thereof and its attribute value, and the representative symbol expansion. The idiom registration function according to claim 1 or 2, characterized in that the attribute of the second representative symbol of the entry word generated and developed by the means and the attribute value thereof are identified. Machine translation device that.