JPH07334506A - Machine translation device - Google Patents

Abstract

PURPOSE:To provide a machine translation device which can automatically select an appropriate synonym considering a content that the original is to express, at the time of selecting translated word of a word constituting the original. CONSTITUTION:A input word concept factor analysis means 21 retrieves an input language concept factor dictionary 42 with the word of the original received from an original analys means 20, and stores the result in an input word concept factor evaluation table 43. A synonym extraction means 25 retrieves a synonym classification disctionary 47 with the word received from a translated sentence generation means 26 and transfers the synonym concerned to an output word concept factor analysis means 24. The output word concept factor analysis means 24 retrieves an output language concept factor dictionary 46 by the received synonym and stores the result in an output word concept factor evaluation table 45. A concept factor evaluation means 22 evaluates the input word concept factor evaluation table 43 and the output word concept factor evaluation table 45 and a translated word substitution means 23 instructs the translated sentence generation means 26 to carry out the substitution of the translated word.

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, and more particularly to a translation editing system of a machine translation device for translating a sentence in an arbitrary language into another language.

[0002]

2. Description of the Related Art A conventional machine translation device is configured to output only one translated sentence for an input sentence to be translated, so that the contents of the input sentence can be read without human intervention. It is difficult to obtain a translation of an appropriate and natural expression as a target language.

Therefore, prior to the machine translation process, the input sentence is modified in advance so that it can be easily translated, or the translated output sentence is modified to make it easier to understand. The former correction is called pre-editing, and the latter correction is called post-editing.

Since these pre-editing and post-editing are for supplementing a portion exceeding the processing capability of the machine translation apparatus, they are mainly performed by humans. A high-quality translated sentence can be obtained by combining the translation processing in such a machine translation device and the pre-editing / post-editing performed by a human.

By the way, many machine translation apparatuses have a support function that allows a person to easily perform pre-editing and post-editing on this apparatus. One of the post-editing functions supported by such a machine translation device is selection of translated words.

[0006] This is done by using a machine translation device to perform an editing operation in which a human verifies the translated text output as the execution result of the translation process and replaces the phrase in the translated text with another appropriate synonym or synonym. It is to be implemented.

For example, according to the translation device disclosed in Japanese Patent Laid-Open No. 61-260362, / This rose
smells sweet. When the English sentence "/" is entered, (a) "This rose has a sweet smell." (B) "This rose has a sweet scent." If "smelling" comes first, the machine translation device outputs the translated sentence of (a). For this reason, in order to obtain a more appropriate translation of (b), the "smell" portion is indicated on the display device (CRT), and the content of synonyms related to that translation "smell" is displayed. Further, an editing operation such as replacing "smell" with "scent" is performed by instructing "scent" from the displayed synonyms.

[0008]

However, which synonym should be selected depends on whether the corresponding word in the input sentence is used to represent a concept such as and. It depends on various situations in which the word is used, such as in what context the sentence it contains is used. It was necessary to give a replacement instruction with an appropriate translation.

On the other hand, in order to appropriately make such a judgment, the operator of the machine translation device needs to be familiar with the language system forming the input sentence and the output sentence, and further Some fields require literary insight into the use of words.

As described above, in order to obtain a high-quality translated sentence by means of selecting a synonym provided by the conventional machine translation device, there is more room for human intervention, and the final translated text quality depends on the translation skill of the operator. There was a problem that it was easily affected by.

An object of the present invention is to provide a machine translation device which can obtain a natural expression from the initial stage of generation of a translated sentence and can greatly improve the translated sentence quality.

[0012]

A machine translation apparatus of the first invention uses an input means for inputting an original sentence to be translated, a translation dictionary storing knowledge information used for translation processing, and contents of the translation dictionary. Is stored in the intermediate language storage means using the contents of the translation dictionary and the intermediate language storage means for storing the intermediate language obtained as a result of the original text analysis means. In a machine translation device including a translated sentence generation unit that converts the intermediate language into a translated sentence and an output unit that outputs a translated sentence obtained as a result of the translated sentence generation unit, one or more concepts that can be recalled from words in an input language are defined. Input language concept factor storage means prestored as input language concept factor information and whether or not information corresponding to the words forming the original sentence is stored in the input language concept factor storage means Input word concept factor analysis means for reading the information of the corresponding word when stored in memory, and input word concept factor evaluation storage means for storing the input language concept factor information read by the input word concept factor analysis means. An output language concept factor storage means for storing in advance one or more concepts that can be recalled from the words of the output language as output language concept factor information, and a synonym storage means for storing a synonym relation of the output language words in advance. The synonym extraction which reads out the information of the corresponding synonym when the synonym is stored in the synonym storage means by checking whether or not the synonym of the translation obtained by the translation generation means is stored in the synonym storage means Means and whether or not information corresponding to the synonyms read by the synonym extraction means is stored in the output language concept factor storage means. Output word concept factor analysis means for reading the information of the corresponding word if stored, and output word concept factor evaluation storage means for storing the output language concept factor information read by the output word concept factor analysis means. , Reading out the input language concept factor information stored in the input word concept factor evaluation storage means, and checking whether the output language concept factor information about the corresponding synonym is stored in the output word concept factor evaluation storage means If stored, the concept factor evaluation means for reading out the output language concept factor information of the corresponding synonym, comparing the similarity with the input language concept factor information, and selecting the most similar synonym, and the concept. The translation word replacement means replaces the synonym selected by the factor evaluation means with the translation word in the translation.

In the machine translation apparatus of the second invention, in the first invention, the concept that the input language concept factor storage means can recall from the word of the input language is relative to one or more modifier pairs having an opposite concept. It is characterized in that it stores information quantified as the ratio of the physical bias.

The machine translation apparatus of the third invention is the machine translation apparatus of the first invention, wherein the output language concept factor storage means associates a concept that can be recalled from a word of the output language with respect to one or more modifier pairs having an opposite concept. It is characterized in that it stores information quantified as the ratio of the physical bias.

[0015]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a machine translation apparatus of the present invention.

As shown in FIG. 1, the machine translation apparatus according to the present embodiment includes an input device 1 such as a keyboard for inputting an input sentence to be translated and control information for setting an operating environment of the device, and an input. Source language analysis means 20 for analyzing a sentence to generate an intermediate language, intermediate language structure file 44 for storing the generated intermediate language, and input language syntax conversion in which knowledge information used by the source text analysis means 20 is registered in advance. A rule 40 and an input language word conversion dictionary 41, and an input language concept factor dictionary 42 in which a concept that can be recalled from each word or idiom allowed in the system of the input language is registered in advance as numerical information. , An input word concept factor analysis means 21 for applying concept information to each word constituting an input sentence using this, and an input word concept factor evaluation table 43 for storing the result, A translation generation unit 26 for generating a translation of a target language from an intermediate language stored in the language structure file 44, an output language syntax conversion rule 48 and an output language word in which knowledge information used by the translation generation unit 26 is registered in advance. The conversion dictionary 49, the synonym classification dictionary 47 in which the synonym relationships of words and phrases allowed in the output language system are registered in advance, and by referring to this, the translation generation unit 26 adopts them as translations for intermediate languages. A synonym extraction means 25 for extracting a synonym for each word / idiom, and an output in which a concept conceivable from the word / idiom that is allowed in the system of the output language is registered in advance as numerical information Language concept factor dictionary 46
And an output word concept factor analysis means 24 for assigning concept information to each synonym extracted by the synonym extraction means 25 using this, an output word concept factor evaluation table 45 for storing the result, and this output word Conceptual factor evaluation table 4
The concept factor evaluating means 22 for comparing the contents of No. 5 with the contents of the input word concept factor evaluation table 43 based on the control information given in advance from the input device 1 and selecting the word judged to be optimum from the synonyms 22 And a word replacement unit 23 that replaces the translated word initially selected by the translated sentence generation unit 26 using the word selected by the concept factor evaluation unit 22, and an output device such as a display device that outputs the finally obtained translated sentence. Three
It consists of and.

The above-mentioned rules, dictionaries, tables and files are constituted by a storage device 4 for storing information, and the above-mentioned means are constituted by a data processing device 2 which operates under program control.

FIG. 5 is a memory content diagram of an example of the input language concept factor dictionary 42. However, this example assumes that the input language is English, and only the components necessary for the explanation are described.

Referring to FIG. 5, the input language concept factor dictionary 42 is a conceptual symbol 5a in which individual words allowed in the language system of the input language are expressed in an intermediate language, and actual words corresponding thereto. 5b and a concept factor value 5c that defines a concept that can be recalled from this word as a matrix of numerical elements.

Here, referring to FIG. 4, the concept factor value 5c
Will be described, that is, a method of evaluating a concept that can be recalled from the word 5b and defining the result as a concept factor value 5c.

FIG. 4 shows the Japanese word "spy"
1 is an example showing a method for graded evaluation using a pair of modifiers having a confrontation concept of "masculine or feminine";
A numerical value from 1 to 7 is assigned to each stage as a concept factor value.

This concept factor value is "masculine" as it is closer to 1 with a numerical value of 4 representing "to chira chi nara" as an intermediate point and "feminine" as it is closer to 7. It is meant to be included in the word "spy". Here, since the concept factor value is defined as 2, it is shown that "quite masculine" is evaluated as one of the concepts recalled from the word "spy".

The input language concept factor dictionary 42 is obtained by preparing a plurality of modifier pairs having an opposite concept based on such an evaluation method and evaluating the concept factor value for each modifier pair. A matrix whose elements are the conceptual factor values is defined for the word to be evaluated.

FIG. 6 is a storage content diagram of an example of the output language concept factor dictionary 46. However, this embodiment assumes that the output language is Japanese, and only the components necessary for the explanation are described.

Referring to FIG. 6, the output language concept factor dictionary 42 represents the concept symbols 6a in which individual words allowed in the language system of the output language are expressed in an intermediate language, and the corresponding actual words. 6b and a concept factor value 6c that defines a concept that can be recalled from this word as a matrix of numerical elements.

Regarding the method of defining the concept factor value 6c of the output language, that is, the method of evaluating the concept that can be recalled from the word 6b of the output language and defining the result as the concept factor value 6c, refer to FIG. Since it is the same as that described above, the description is omitted here.

Next, an example of the synonym classification dictionary 47 will be described with reference to the drawings.

FIG. 7 is a storage content diagram of an example of the synonym classification dictionary 47.

The conceptual symbol 7a is a conceptual symbol that is defined in advance for a word permitted in the system of the output language and is used as a heading for referring to the table. Further, the synonym group 7b is a synonym having a meaning of a word expressed by the concept symbol 7a as a superordinate concept defined by using the concept symbol. In this example, the concept symbol of the Japanese word "spy" is "SPY", and the concept symbol of the word "piggy" is "PIG",
Both words are shown to be usable as translations of the conceptual symbol "SPY".

Next, the operation of this embodiment will be described with reference to the drawings.

9 to 12 are flow charts showing the process steps 1 to 4 of the present embodiment separately.

When an original sentence to be translated is input from the input device 1, the original sentence analysis means 20 analyzes the original sentence (step S40 in FIG. 9).

The source text analysis means 20, as shown in FIG.
It is composed of a morpheme analysis unit 20a, a syntax analysis unit 20b, and a semantic analysis unit 20c.

The morphological analysis unit 20a decomposes the input sentence into words and recognizes grammatical attributes such as the part of speech of each word.
At this time, the input language syntax conversion rule 40 and the input language word conversion dictionary 41 are used.

The syntax analysis unit 20b is a morphological analysis unit 20a.
Based on the part-of-speech information and the like obtained by, the relationship between words and the dependency of words is recognized. At this time, the grammar rules and the like regarding the input language stored in the input language syntax conversion rule 40 are used.

The semantic analysis unit 20c uses the input language syntax conversion rules 40 and the input language word conversion dictionary 41 to check the semantic relationship between words and recognizes the semantic expression of the input sentence.
Here, the meaning of a polysemous word in which one word has a plurality of meanings is also limited. The result analyzed in this way is stored in the intermediate language structure file 44 as a concept symbol (generally called an intermediate language) that does not depend on each language targeted by the machine translation device.

The source sentence analysis means 50 for generating an intermediate language from an input sentence, and the input language syntax conversion rules 40 and the input language word conversion dictionary 41 referred to by the source language analysis means 50 are all known methods, for example, JP-A-4-321
Since each analysis unit and word dictionary described in Japanese Patent No. 178 can be used, description thereof will be omitted.

When the analysis of the original sentence is completed, the translated sentence generating means 2
A translation of the target language is generated by 6 (step S50 in FIG. 9).

As shown in FIG. 3, the translated sentence generation means 26 is composed of a meaning generation unit 26a, a syntax generation unit 26b, and a morpheme generation unit 26c.

The meaning generator 26a sequentially inputs the analysis results of the source sentence analysis means 20 stored in the intermediate language structure file 44 and outputs them to the synonym extraction means 25. After this, for the intermediate language output to the synonym extraction means 25,
Since the replacement information of the intermediate language is output from the translation replacement unit 23, the replacement information is replaced with a word or phrase of the target language to be output as a translation. At this time, the output language word conversion dictionary 49 is used which has a correspondence relationship between the conceptual symbols of the intermediate language and the respective words constituting the target language.

The syntax generation unit 26b recognizes the word-to-word relationship and determines the word order while using the output language word conversion dictionary 49. At this time, the output language syntax conversion rule 48 for the target language is also used.

The morpheme generation unit 26c uses the output language word conversion dictionary 49 to perform a utilization process such as a change in the ending of each word to generate a target translation.

The translated sentence generation means 26 for generating translated sentences from the intermediate language stored in the intermediate language structure file 44.
Or the output language syntax conversion rule 48 referenced by this
As for the output language word conversion dictionary 49, a known method, for example, each generation unit described in Japanese Patent Laid-Open No. 4-321178 and a bilingual dictionary can be used, and therefore description thereof will be omitted.

The input word concept factor analysis means 21 inputs the words constituting the original sentence from the original sentence analysis means 20 (step S1 in FIG. 9), and determines whether the same word as this word is stored in the input language concept factor dictionary 42. (Step S
2).

When the same word is stored in the input language concept factor dictionary 42, the input word concept factor analysis means 21 reads the contents of the corresponding input language concept factor dictionary 42,
It is output to the input word concept factor evaluation table 43 (step S3).

On the other hand, the same word is the input language concept factor dictionary 4
If not stored in 2, the input word concept factor analysis means 21 does not execute any processing. The input word concept factor analysis means 21 repeats the processes of steps S1 to S3 until the end of the original sentence is detected (steps S4 and S5).

The synonym extraction means 25 is a translation generation means 26.
The concept symbol (translated word candidate) of the translated word that constitutes the translated sentence is input from (step S6 in FIG. 10), and it is checked whether or not a record having the same conceptual symbol as the conceptual symbol of this translated word is stored in the synonym classification dictionary 47, and the same. When the concept symbol is stored in the synonym classification dictionary 47, the synonym extraction means 25 reads the corresponding information from the synonym classification dictionary 47 and outputs it to the output word concept factor analysis means 24 (step S7).

On the other hand, if the same symbol as the translated word is not stored in the synonym classification dictionary 47, the synonym extraction means 25 does not perform any processing.

The output word concept factor analysis means 24 inputs the information about the synonyms of the translated words constituting the translated sentence from the synonym extraction means 25, and outputs the same concept symbols as the individual concept symbols classified in the synonyms. It is checked whether or not the possession is stored in the output language concept factor dictionary 46 (step S8).

When the same concept symbol as the synonym of the translated word is stored in the output language concept factor dictionary 46, the output word concept factor analysis means 24 reads the contents of the corresponding output language concept factor dictionary 46 and outputs the output word concept. The data is output to the factor evaluation table 45 (step S9).

On the other hand, when the same concept symbol as the synonym of the translated word is not stored in the output language concept factor dictionary 46, the output word concept factor analysis means 24 does not execute any processing.

The output word concept factor analysis means 24 executes the processing of steps S8 to S9 for all the synonyms which are input from the synonym extraction means 25 and which are classified into the specific translated words. In addition, the processing of steps S7 to S9
Repeat until the end of the translated sentence is detected (step S10).
And step S11).

The concept factor evaluation means 22 sequentially reads the information stored in the input word concept factor evaluation table 43 (step S12 in FIG. 11), and this information notifies the end of one sentence forming the original sentence. It is determined whether or not (step S13).

If the result of this judgment does not notify the end of one sentence, the concept factor value is added for each concept factor (step S14), and if the end of one sentence is notified, that sentence is terminated. The average of the concept factor values of the words constituting the is calculated, and the result is stored in the input word concept factor evaluation table 43 (step S15).

Next, the concept factor evaluation means 22 determines whether the information read out in step S12 is the end of one sentence forming the original sentence and at the same time the end of one paragraph forming the original sentence. The determination is made (step S16).

If the result of this determination is the notification of the end of a paragraph, the average of the concept factor values of the words forming the paragraph is calculated, and the result is stored in the input word concept factor evaluation table 43. If not stored (step S17) and notifying the end of one paragraph, the concept factor evaluation means 22 does not execute any processing.

The concept factor evaluation means 22 starts from step S12.
The process of S18 is repeated until the end of the input word concept factor evaluation table is detected (step S18).

Next, the concept factor evaluation means 22 sequentially reads out the information stored in the input word concept factor evaluation table 43 (step S19 in FIG. 12), and associates it with the individual words described therein. Information on the synonyms that have been read is read from the output word concept factor evaluation table 45 (step S20).

Here, the concept factor value of the input word defined in the word read out in step S19 and the step S
The similarity with the concept factor value of the output word defined in the word read in 20 is calculated, and this is set as the word concept similarity rate (steps S21 and S22).

Next, the average value of the concept factor values defined for the word read in step S19 is calculated in step S1.
5, the similarity between this value and the concept factor value of the output word defined in the word read in step S20 is calculated, and this is set as the sentence concept similarity rate (step S23 and S24).

Further, with respect to one paragraph of the input word containing the word read in step S19, the average value of the concept factor values defined in the individual words forming the paragraph is calculated in step S17. Therefore, the similarity between this value and the concept factor value of the output word defined in the word read in step S20 is calculated, and this is set as the paragraph concept similarity rate (steps S25 and S26). .

Here, three sums of the word concept similarity rate, the sentence concept similarity rate, and the paragraph concept similarity rate are obtained and used as the total concept similarity rate of the synonyms selected in step S20 (step S27).

The concept factor evaluating means 22 carries out step S19.
The steps S20 to S27 are performed for each synonym in the output word concept factor evaluation table 45 that is associated with the word read in step S2 (step S2).
8) A synonym that maximizes the value of the overall concept similarity rate is output to the translation word replacement unit 23 (step S29).

The concept factor evaluating means 22 carries out step S19.
The process from S29 to S29 is repeated until the end of the input word concept factor evaluation table 43 is detected (step S30).

Details of the numerical calculation performed in steps S12 to S30 will be described with reference to the drawings.

FIG. 8 shows the input word concept factor evaluation table 43.
Is a mathematical model of.

_Let P _r be an arbitrary paragraph that constitutes the input source text.
(R = 1, 2, 3, ...) _Let P _{r be} a finite set G of sentences S _i (i = 1, 2, ..., d) that compose this paragraph.
And the elements S _i are arranged without duplication.

[0069] Now, the number of statements that make up any paragraph P _r and d pieces, the following equation function for obtaining the number of statements contained in paragraph P _r as F _L (P _r) (1) and ( 2) is defined.

[0070]

[0071]

[0072]

The sentence S _i is an arrangement of the elements W _ij of the finite set H of the constituent words without duplication.
Further, let m be the number of words that compose S _i , and let F _W (S _i ) be a function that finds the number of words included in the sentence S _i , and define the following equations (3) and (4).

[0074]

[0075]

When the word W _ij forming the sentence S _i has the concept factor value a _ijk for each of the n concept evaluation items C _K (K = 1, 2, ..., N), the word W _ij is individually defined. An x-row matrix having the word of as an element is set in the input word concept factor evaluation table. However, x is the total number of words forming the paragraph P _r and is defined by the following equation.

[0077]

The contents of the input word concept factor evaluation table 43 can be determined by the above equations (1) to (5). Here, the element a _ijk of the concept factor value of the word W _ij is defined by the following equation.

[0079]

Next, the column average of the elements a _ijk of all the words forming the sentence S _i containing the word W _ij is calculated by the formula (7) as S _i bar.

[0081]

Here, for example, _in indicates that the second suffix, j in this case, is averaged.

Further, the column average of the elements a _ijk of all the words forming the paragraph P _r including the sentence S _i is calculated by the formula (8) as the P _r bar.

[0084]

Here, for example, _..n means that the first and second suffixes, in this case, i and j are averaged.

When one of the translation word candidates for the word W _{ij of the} input word is T _ij and this T _ij has the concept factor value b _ijk for each of the n concept evaluation items C _k , T _ij and W _ij. Also, for S _i bar and P _r bar, the function values defined in the equations (9) to (11) are calculated.

[0087]

[0088]

[0089]

Here, the values obtained by the equations (9) to (11) and three constants V _w given in advance from the input device 1,
Using V _s and V _p , the overall concept similarity g _t is calculated by the following equation.

[0091]

V _w , V _s , and V _p are respectively expressed by equation (9) and equation (1
0) and a constant for weighting the values calculated by Expression (11), which are a word concept weight coefficient, a sentence concept weight coefficient, and a paragraph concept weight coefficient, respectively.

The user of the machine translation device gives these three coefficient values from the input device 1 such as a keyboard in advance. Thus, when there are a plurality of synonyms as translation word candidates, it is possible to change the criterion of which to adopt and specify the style of the translated text to be output.

For example, when the value of V _w is set to be larger than the values of V _s and V _p , the most commonly used first meaningful word is translated as the translated word for the corresponding word on the original sentence side. As it tends to be adopted as, it becomes a literal translation, and it is effective when there is a tendency to dislike the ambiguity of translated texts, such as natural science texts.

On the contrary, when the values of V _s and V _p are made larger than that of V _w , each of them has a sentence containing the corresponding word on the original sentence side or a concept similar to the concept contained in the paragraph. It becomes more free as the tendency to adopt words as translated words becomes stronger, and it is effective when the meaning of a word has a large proportion depending on the usage of the word, such as a sentence in a literary field such as a novel or essay. .

The overall concept similarity g _t is calculated for all translation word candidates (synonyms) for the word W _{ij of the} input word,
The word that maximizes this value is determined as the translation of the input word W _ij .

[0097]

As described above, in the machine translation apparatus of the present invention, the input word concept factor analysis means tries to express what concept the sentence or paragraph constituting the original sentence to be machine translated is intended to express. Whether or not it is converted to numerical information, and the output word concept factor analysis means compares the concepts of each of the multiple translation word candidates to numerical information by the concept factor evaluation means, and Since it is configured to automatically select a conceptually similar translated word, the effect that a natural expression can be obtained at the initial generation stage of the translated text, and as in the case of natural science documents mainly targeted by conventional translation machines, Not only in areas where you can clearly determine what important information to translate, but also in novels and essays,
This has the effect of improving the translation quality in a field in which the meaning of words is highly dependent on the language usage.

Further, since the evaluation criterion of the concept factor evaluating means is made adjustable by the control information given from the input device, the style of the translated sentence can be defined in advance.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a machine translation device of the present invention.

FIG. 2 is a block diagram of an original sentence analysis unit in FIG.

FIG. 3 is a block diagram of a translated sentence generation unit in FIG.

FIG. 4 is an evaluation diagram illustrating an example of conceptual evaluation of a word by a modifier pair.

5 is a storage content diagram showing an example of an input language concept factor dictionary used in FIG. 1. FIG.

6 is a storage content diagram showing an example of an output language concept factor dictionary used in FIG. 1. FIG.

7 is a storage content diagram showing an example of a synonym classification dictionary used in FIG. 1. FIG.

8 is a storage content diagram showing an example of an input language concept factor evaluation table used in FIG. 1. FIG.

9 is a first flowchart showing the operation of FIG. 1. FIG.

10 is a second flowchart showing the operation of FIG. 1. FIG.

11 is a third flowchart showing the operation of FIG. 1. FIG.

12 is a fourth flowchart showing the operation of FIG. 1. FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 input device 2 data processing device 3 output device 4 storage device 20 original sentence analysis means 21 input word concept factor analysis means 22 concept factor evaluation means 23 translation word replacement means 24 output word concept factor analysis means 25 synonym extraction means 26 translation generation means 40 input Language syntax conversion rules 41 Input language word conversion dictionary 42 Input language concept factor dictionary 43 Input word concept factor evaluation table 44 Intermediate language structure file 45 Output word concept factor evaluation table 46 Output language concept factor dictionary 47 Synonym classification dictionary 48 Output language syntax conversion Rule 49 Output language word conversion dictionary

Claims (3)

[Claims] 1. An input unit for inputting an original sentence to be translated, a translation dictionary storing knowledge information used for translation processing, an original sentence analyzing unit for converting the original sentence into an intermediate language by using the contents of the translation dictionary, and An intermediate language storage unit for storing the intermediate language obtained as a result of the original sentence analysis unit, and a translated sentence generation unit for converting the intermediate language stored in the intermediate language storage unit into a translated sentence using the contents of the translation dictionary and the translated sentence. An input language concept factor memory in which one or more concepts that can be recalled from a word in the input language are stored in advance as input language concept factor information in a machine translation device including an output unit that outputs a translated sentence obtained as a result of the generation unit. Means and information on the corresponding word if it is stored by the input language concept factor storage means by checking whether or not the information corresponding to the word constituting the original sentence is stored Input word concept factor analysis means, an input word concept factor evaluation storage means for storing the input language concept factor information read by the input word concept factor analysis means, and one or more that can be recalled from words in the output language. The output language concept factor storage means pre-stores the concept of as the output language concept factor information, the synonym storage means pre-stores the synonym relation of the words of the output language, and the translated sentence obtained by the translated sentence generation means. For the translated word, it is checked whether the synonym is stored in the synonym storage means, and if the synonym is stored, the synonym extraction means for reading the information of the corresponding synonym, and the synonym read by the synonym extraction means It is checked whether or not the information corresponding to is stored in the output language concept factor storage means, and if stored, the corresponding word information is stored. The output word concept factor analysis means, the output word concept factor evaluation storage means for storing the output language concept factor information read by the output word concept factor analysis means, and the input word concept factor evaluation storage means The input language concept factor information is read out and it is checked whether or not the output language concept factor information for the corresponding synonym is stored in the output word concept factor evaluation storage means, and if it is stored, the output of the corresponding synonym is output. Conceptual factor evaluation means for reading out the language conceptual factor information and comparing the similarities with the input language conceptual factor information and selecting a synonym with the highest similarity, and a synonym selected by the conceptual factor evaluation means and a translation A machine translation device comprising: a translation word replacement unit that replaces a translation word. 2. The input language concept factor storage means stores information quantifying a concept that can be recalled from a word in the input language as a relative bias ratio for one or more modifier pairs having an opposite concept. The machine translation device according to claim 1, wherein 3. The output language concept factor storage means stores information that quantifies a concept that can be recalled from a word of the output language as a relative bias ratio with respect to one or more modifier pairs having an opposite concept. The machine translation device according to claim 1, wherein