JPH06243162A - Machine translation system - Google Patents

Abstract

PURPOSE:To facilitate reading and understanding of object language by selecting and presenting only information on target language of a part of speech and a translated word, etc. CONSTITUTION:An input document 31 is inputted by an OCR 32, etc. In that case, not only the contents of the document, but also information (font, multiple column, etc.) related to a format of the document are stored simultaneously. The inputted document is divided into morphemes by a morpheme analysis processing, and allowed to consult a dictionary. To a part of speech of the divided morpheme, a part-of-speech estimation processing is performed, based on part-of-speech connection probability, and an optimal part of speech is determined with respect to each morpheme. In accordance with the determined optimal part of speech, information (translated word, etc.) of second language is retrieved. A first language sentence and information of second language are allowed to correspond to each other, and in accordance with format information stored previously, an output form in shaped. In the end, a parallel translation document is outputted by a printer, etc.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to a machine translation device, and more particularly to a machine translation device having a function of displaying or printing target language information corresponding to a target language sentence according to a format of a target language document.

[0002]

2. Description of the Related Art Conventionally, as means for obtaining second language (target language) information corresponding to a first language (target language) sentence, for example, "Electronic Dictionary with Image Scanner" is disclosed in Japanese Patent Laid-Open No. 1-234977. Proposed. The method of this publication is a method in which an image scanner is provided as an input unit and the read first language is converted into a second language by an electronic dictionary (hereinafter, electronic dictionary). In this method, the first language read is transformed into the original form (standard form),
The second language information is output by searching the electronic dictionary.

However, according to this method, the first language is retrieved from the read electronic dictionary and the information described in the relevant part is simply output. In general, one linguistic unit (sentence, phrase, word) can have a plurality of interpretations, and an electronic dictionary often has a plurality of interpretations in one heading item. Therefore, the above-mentioned JP-A-1
In the method based on Japanese Patent No. 234977, the user must select a plurality of output second language information at any time. This work becomes a great burden when understanding the contents described in the first language in the second language. As another conventional technique, there is a "translation display device" in Japanese Patent Laid-Open No. 138660/1990.
As in Japanese Patent No. 4977, the input first language item is searched from the electronic dictionary, and the described second language information is simply output, which is suitable for understanding the first language sentence. It does not select language information.

An unknown word may be included in the first language original sentence due to an OCR error or insufficient entry of a dictionary entry in the dictionary. Since the part of speech is estimated based on the statistics of the word sequence,
When the number of unknown words in the first language sentence increases, the statistics of the word arrangement often cannot be used. Assuming all part-of-speech as candidates for unknown words, the statistics of word sequences will be less accurate if adjacent words are unknown words. Therefore, the number of cases in which correction is made by mistake is increased, and the accuracy of part-of-speech estimation is reduced. Further, when collectively translating a document that is a set of a plurality of sentences, it is usual to divide the sentence into sentence units and then perform linguistic processing. However, since it is difficult to automatically divide the sentence by sentence, in Japanese Patent Application No. 4-121277 proposed earlier,
It is only divided into physical lines. The Japanese Patent Application No. 4-1
In No. 21277, since the part-of-speech is estimated by locally applying the syntax constraint, it is not a fatal problem even if the processing unit is not the whole sentence. However, as for words at both ends of the physical line, the concatenated information cannot be used because the adjacent word is separated into the previous line or the next line. As a result, the accuracy of part-of-speech estimation for words near both ends of a physical line may decrease.

[0005]

[Object] The present invention has been made in view of the above situation, stores the concatenation probability of the part of speech in the first language, and determines and determines the part of speech of each word in the first language sentence based on the concatenation probability. To provide a machine translation device having a function of retrieving second language information (translated word) corresponding to a first language from an electronic dictionary according to a generated part-of-speech, and displaying or printing out in accordance with a format of a document in the first language. It was made for the purpose.

[0006]

In order to achieve the above object, the present invention provides (1)
At least an input means for inputting a target language sentence,
A storage unit that stores the target language sentence input by the input unit, an analysis dictionary that stores knowledge necessary for language analysis, and a target language sentence that is the translation result of the target language sentence. In a language processing system including a display means, part-of-speech estimation means for estimating the part-of-speech of the target language, and information on the target language corresponding to the target language from the analysis dictionary according to the part-of-speech estimated by the part-of-speech estimation means )
And (2) the part-of-speech estimating means stores the concatenation probability of the part-of-speech of the target language, and the part of the target language sentence based on the concatenation probability. And a determining means for determining the part of speech of each word,
Further, (3) the input means uses an optical character reading device means, and (4) the display means displays or prints the target language sentence and the target language information in association with each other. Having output means for outputting,
(5) In the above (4), the output means stores the target language sentence and the target language information based on the second storage means for storing the format of the document in the target language and the format of the document in the target language. Or (6) at least input means for inputting a target language sentence, storage means for storing the target language sentence input by the input means, and language analysis An analysis dictionary that stores necessary knowledge, a display unit for displaying the target language sentence and a target language sentence that is a translation result thereof, a part-of-speech estimation unit that estimates the part-of-speech of the target language, and the part-of-speech estimation unit. According to the part-of-speech estimated by the above, in a language processing system equipped with an extracting means for extracting information of a target language corresponding to a target language from the analysis dictionary, a multi-part-of-speech word is obtained by using information on connection between words. Part-of-speech estimating means for estimating a part-of-speech, storage means for storing the immediately preceding row and the immediately following row in addition to the current row, connecting means for connecting the immediately preceding row and the immediately following row stored in the storage means to the current row, and the connection A part of speech estimating means for estimating a part of speech for the line after connection created by the means, and (7) in (6), a determining means for determining an unknown word included in the target language sentence. And counting means for counting the number of unknown words judged by the judging means, and judging means for judging whether or not to perform the part-of-speech estimation processing based on the counting result of the counting means. ) In the above (6), it comprises a judging means for judging a basic word in the target language sentence, and a suppressing means for suppressing the output of the target language information based on the judgment result of the judging means, and further, ( 9) In (6) above, part of speech The determination unit for determining the priority of the determined part of speech in the dictionary, and the suppression unit for suppressing the output of the target language information based on the determination unit; The method further comprises: an authenticating unit that obtains the rank of the possibility of part-of-speech by estimation, and an output unit that outputs the second language information corresponding to a plurality of high-ranking part-of-speech candidates according to the rank by the authenticating unit. ) In the above (6), an authenticating means for obtaining a plurality of possibilities of a part of speech by part-of-speech estimation, an evaluating means for evaluating the degree of possibility of the authenticating means, and a plurality of degrees of possibility of the evaluation are used. It comprises a judging means for judging the accuracy of one solution and a changing means for changing the number of solutions to be output based on the judgment result of the judging means, or (12) at least for inputting a target language sentence Typing hands A column, a storage unit for storing the target language sentence input by the input unit, an analysis dictionary storing knowledge necessary for language analysis, the target language sentence and a target language sentence as a translation result thereof. A display unit for displaying, a part-of-speech estimating unit that estimates the part-of-speech of the target language, and an extracting unit that extracts information of a target language corresponding to the target language from the analysis dictionary according to the part-of-speech estimated by the part-of-speech estimating unit. In a language processing system equipped with, a part-of-speech estimation means for estimating a part-of-speech of a multi-part-of-speech word using information about concatenation of words and a compound word template for describing a syntactic type of a compound word of a target language Storage means for also storing information (translation word, part of speech, etc.), collation means for collating the stored template with the target language sentence, and elements in the phrase template in the collation means And a display unit for displaying the target language information corresponding to the idiom template based on the execution result of the collating unit and the extracting unit, and further including an extracting unit for extracting a word or a phrase in the corresponding target language sentence. (13) In (12), output means for separating and outputting the target language information of the idiom template and the target language information corresponding to the word (phrase) in the target language sentence that matches the element in the template. And (14) in (12), the target language information of the phrase template and the target language information corresponding to the word or phrase in the target language sentence that matches the element in the template are modified. · Having output means for synthesizing and outputting as one target language phrase, and further,
(15) In the above (12), a detection unit that detects an unknown word in the target language sentence, a storage unit that stores a transformation pattern of a word in the target language and a syntactic category (part of speech etc.) after the transformation, and the detection unit. For the unknown word detected in, by applying the deformation pattern, an authenticating means for obtaining the word before the deformation,
And a search means for searching the first language analysis dictionary for the untransformed word obtained by the recognizing means, and a determining means for determining the syntactic category of the unknown word according to the result of the search means. (16) In (15) above, according to the result of the search means, there is provided a display means for displaying the target language information of the target language word before transformation and the transformation information together, further, (17) In (16) above, there is provided display means for displaying the target language information of the target language word before transformation after transforming it with the transformation information of the target language sentence according to the result of the search means. . Hereinafter, description will be given based on examples of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of a machine translation device having a proper word selection function according to the present invention. In the figure, 1 is a display device, 2 is a display control unit, and 3 is language processing. The unit, 4 is an input control unit, and 5 is an input device. The first language (target language) input by the input device 5 is translated into the second language (target language) by the language processing unit 3 via the input control unit 4. The translated second language is displayed on the display device 1 via the display control unit 2.

FIG. 2 is a block diagram of a bilingual document creating process in a machine translation apparatus having a bilingual document creating function according to the present invention. In the figure, 21 is a keyboard and OCR (Optical Ch
aracter Reader; an optical character reader, etc., a first language input means, 22 a morphological analysis means for dividing the first language sentence into morphemes, 23 an analysis dictionary in which language information used in morphological analysis is described, Reference numeral 24 is a part-of-speech estimation means for obtaining an optimum part-of-speech for a first language sentence, 25 is a part-of-speech concatenation probability table used by the part-of-speech estimation means, and 26 is an output form for producing an output in which information of the first language and second language is associated with each other. Deciding means, 27
Is a means for outputting to the output device according to the contents created by the output form determining means.

Input means (O) for inputting the first language sentence
The first language sentence (original sentence) input by (CR etc.) 21 is stored by the storage means. Morphological analysis means 22 using an analysis dictionary 23 that stores knowledge necessary for language analysis.
Morphological analysis is performed by. The first language sentence (original sentence) and the second language sentence which is the translation result thereof are displayed by the display means. The connection probability of the part of speech in the first language is stored as a connection probability table 25. The part-of-speech of each word of the first language sentence is estimated by the part-of-speech estimating means 24 based on the above-mentioned concatenation probability, and the second part corresponding to the first language from the analysis dictionary 23 is estimated according to the part-of-speech estimated by the connecting means. The information (translated word) of the language is extracted, and the output form is determined by the determination means 26 by associating the information of the first language sentence with the information of the second language.
Displayed or printed out according to 7. Based on the format of the document in the first language, the output unit 27 displays or prints out the information in the first language and the information in the second language. As a result, the first language sentence can be smoothly understood in the second language without presenting unnecessary information.

FIG. 3 is a block diagram of the bilingual document creation processing section.
In the figure, 31 is an example of an input document described in the first language, 32
Is OCR, 33 is parallel translation creation processing, 34 is a printing device, and 35 is
Is an example of an output document to which the information of the second language is added. The input document 31 is input by the OCR device 32 or the like. At that time, not only the content of the document but also information about the format of the document (font, column, etc.) is stored at the same time. The input document is divided into morphemes by the morphological analysis process,
Look up in the dictionary. Part-of-speech estimation processing is performed on the part-of-speech of the divided morphemes based on the part-of-speech concatenation probability, and the optimum part-of-speech for each morpheme is determined. The second language information (translated word, etc.) is searched according to the determined optimum part-of-speech. Then, the first language sentence and the information of the second language are made to correspond to each other, and the output form is shaped according to the previously stored format information. Finally,
Output the bilingual document with a printer. In the figure, the parallel translation for the first language sentence is indicated by a broken line below the first language sentence. When assigning a translated word, the second word is added below the word in the first language.
If a language translation is output, correspondence between the two languages can be easily taken visually.

Each means constituting the present invention will be described below. Hereinafter, unless otherwise specified, the English-Japanese machine translation will be described as an example, but the same applies to other two languages. (1) Input means A document in the first language is input to the machine translation system by type input using a keyboard, OCR input using a scanner, or the like. The format information of the document in the first language includes the size of the document, the font used, the position of each word in the document, the number of columns, etc., and is stored together with the document content.

(2) Morphological analysis means The first language sentence is divided into morphemes (corresponding to words). In languages such as English, which have spaces between words and are written separately, it is returned to the original form (standard form) by the utilization change table,
Look up the dictionary and ask for usage, parts of speech, translations, etc. In the case of a solid language such as Japanese, the optimum word range is obtained based on the rules such as the longest match method and the minimum phrase method.

(3) Part-of-Speech Estimating Means The part-of-speech concatenation probability is estimated from the first-language corpus tagged with the part-of-speech tag, and the probability of occurrence of a part-of-speech string that can be taken by the first language sentence to be translated is estimated based on the concatenation probability. Is calculated, and the part-of-speech sequence having the highest probability is selected. That is, the following is assumed as a probabilistic model of language. Part-of-speech sequence T = t ₁ ,
The probability P (T) that t ₂ , t ₃ , ..., tn appears is

[0014]

[Equation 1]

It is indicated by Where P (ti | t ₁ , t ₂ , t ₃ ,
, T _i-1 ) is the probability that ti is connected after the part _-of- speech sequence t ₁ , t ₂ , t ₃ , ..., T _i-1 appears. However, P (ti
It is difficult to obtain | t ₁ , t ₂ , t ₃ , ..., T _i-1 ).
For example, if the part-of-speech type is V, t ₁ , t ₂ , t ₃ , ..., T _i-1
The number of different part-of-speech sequences of V is V to the power of (i-1), and it is impossible to obtain the probability for each of these (V =
If it is 100, V ¹⁰ = 10 ²⁰ ways). The connection probability P (ti | t ₁ , t ₂ , t ₃ , ..., T _i-1 ) has to be obtained approximately.

Therefore, the part of speech that influences the concatenation of parts of speech is immediately preceding 2
Trigram model P (ti | t ₁ , t ₂ , t ₃ , ..., T _i-1 ) = P (ti | t _i-2 , t
_i-1 ) used. Assuming the trigram model, the appearance probability of the part-of-speech sequence T is

[0017]

[Equation 2]

Can be approximated by Concatenation probability P (ti | t _i-2 , t
_i-1 ) is the appearance frequency C (t _i-2 , t _i-1 ), C (t _i-2 , t _i-1 ) and C (t _i-2 , t _i-1 ) from the English corpus with the part _-of- speech tag as shown in the following equation. It can be estimated by the ratio of t _i-2 , t _i-1 , t _i ).

[0019]

[Equation 3]

In the multi-part-of-speech word, each part-of-speech is not used with equal probability, and there is a bias in the probability of using each noun. If this information is used together, the part-of-speech sequence appearance probability can be estimated more accurately. Expression (3.3) is the appearance probability of the part-of-speech sequence, not the appearance probability of the word sequence. In machine translation, a word string is already given as an input, so (3.
3) The expression is expanded to “the word string W = w ₁ , w ₂ , ..., Wn and the probability that the part-of-speech string is T”. Therefore, the relative probability of word part-of-speech occurrence P (ti · wi): The joint probability that the word wi appears and the part-of-speech is ti is calculated. P (ti · wi) can be easily obtained from the tagged corpus by the following equation.

[0021]

[Equation 4]

If the logarithmic probability is used, the multiplication part becomes an addition, so that the probability calculation can be speeded up.

In order to assign an optimum part-of-speech to each word in the first language, the part-of-speech sequence T having the maximum probability in the expression (3.3) or (3.5) is set as a part-of-speech candidate (result of morphological analysis). You have to choose from. The result after morphological analysis is a set of morphemes in which a range of words (in some cases multiple words are connected), a prototypical word, and a part-of-speech penalty in that range are set. From this set
Select morphemes for Left-to-Right. It is basically a beam search algorithm. At that time, since one morpheme may be composed of a plurality of words, the process should be advanced in synchronization with the terminal point of the morpheme, not in synchronization with the morpheme number.

4 (a) to 4 (c) are diagrams showing a procedure for selecting the maximum appearance probability part-of-speech sequence. FIG. 4 (a) is a morphological analysis result, FIG. 4 (b) is the selection process, and FIG. Each method of solution management is shown. The selection algorithms are summarized below. The morpheme starting from the beginning of the sentence is stored in the memory table.
When the beam width is exceeded, the upper beam width pieces are stored. The morpheme terminal points i = 1, 2, 3, ..., I are executed. The beam width w = 1, 2, 3, ..., W is executed. Search for a morpheme that can be connected to the terminal point i, and use the part-of-speech sequence path Path
The appearance probability after connection with (i, w) is calculated.

The storage table of the end position after connection is updated in the following manner. If the storage table has room (less than the beam width), the history and appearance probability of the path after connection obtained in step 3 are stored. ○ If there is no room in the storage table (if the beam width is satisfied), the path with the smallest appearance probability is selected from the part-of-speech sequence paths stored in the storage table. If it is lower than the appearance probability obtained in, store the path history after connection and the appearance probability obtained in. ★ If it is too high, reject the pass you asked for. The part-of-speech sequence paths stored in the storage table at the end position i = I, which is the end of the sentence, are sorted by the appearance probability.

FIG. 5 is a flowchart of the maximum appearance probability part-of-speech sequence selection procedure.
It is a figure which shows a row chart. Follow the steps below
Will be explained in order.step1 ; A memory table whose morphemes starting from the beginning have been initialized
Store the beam width pieces in the bull. When beam width is exceeded
The upper beam widths of the appearance probability are stored in. Go to step 2
Mu.step2 The position counter is the morpheme stored in step 1.
End position closest to the beginning of the sentence (that is, the shortest form
Set to the end position of the element. Proceed to step 3.step3 If the position counter is at the tail position, step13
Go to. If the tail end has not been reached, proceed to step 4.step4 ; Clear the beam width counter. Go to step 5
Mu.step5 ; Beam width exceeded or at current beam position
If there is no solution candidate, go to step 12. Less than the beam width,
If there is a solution candidate at the current beam position, go to step 6. step6 ; Extract connectable morphemes from the current solution candidates. st
Go to ep7.

[0027]step7; Step if there is no connectable morpheme
Proceed to 11. If there is, go to step 8. step8 ; Take out one connectable morpheme. Go to step 9
Mu.step9 ; Current candidate and extracted morpheme are connected, and after connection
The probability of occurrence of the solution candidate of (3.3) or (3.5)
Calculate by Go to step 10.step10 ; Input the combined solution candidates to the solution candidate update process.
It Go to step 7.step11 The beam width counter is incremented. st
Go to ep5.step12 The position counter is incremented. step3
Go to.step13 ; Probability of appearance of solution candidates stored at the tail position
Sort from highest to lowest. The first solution candidate is the optimal part-of-speech sequence
is there.

FIG. 6 is a flowchart showing a solution candidate updating process. Hereinafter, each step will be described in order. step 10-1 ; If the storage table of the end position of the solution candidate after connection is less than the beam width, the process proceeds to step 10-2. If the beam width is satisfied, proceed to step 10-4. step10-2 ; Store the solution candidates after connection in the storage table at the end position. Go to step 10-3. step10-3 ; increment the beam width counter of the storage table of the end position of the solution candidate after connection. Go to step 11. step10-4 ; Among the solution candidates stored in the storage table of the end positions after connection, the solution candidate having the lowest appearance probability is obtained. Go to step 10-5. step10-5 ; If the appearance probability obtained in step10-4 is lower than the appearance probability of the solution candidate after connection, the solution candidate after connection is set to st
Store in place of the solution candidate found in ep10-4, step11
Go to. If it is higher, the updating process is not performed and the process proceeds to step 11.

(4) Output Form Determining Means The part of speech of the word in the first language is estimated by the part of speech estimation processing, and the information in the second language corresponding to the part of speech is obtained from the dictionary. In this processing, the output content is shaped so that the information in the second language corresponding to the first language sentence is displayed or printed out based on the format of the document in the first language. Since the format information of the document in the first language is stored in the input means (1), the first language sentence may be output in accordance with it.
It is convenient to output the corresponding information in the second language like ruby under the first language sentence, because the correspondence between the two languages can be visually obtained. To output like ruby, calculate the blank distance between lines from the format information of the first language sentence, select the font that fits in that distance, and select the font size from the position of each corresponding word to the line blank direction ( The information of the second language may be output at a position shifted to the left for vertical writing and to the bottom for horizontal writing.

As described above, the part-of-speech estimating means estimates the part-of-speech in the first language and presents only the corresponding second-language information to the user. In addition to the method described in the embodiment, the part-of-speech estimating means parses a part-of-speech candidate and estimates the part-of-speech collected in an appropriate linguistic unit (sentence, phrase, etc.) by the syntactic analysis as the part-of-speech of each word, or as a language model HMM. (Hidden
Markov Model: Hidden Markov Model).

The above is a description of the first and second embodiments (claims 1 and 2). Next, a third embodiment (claim 3) will be described. In particular, when the OCR processing is used for the input means, the input speed is remarkably improved and the user can concentrate on understanding the output result because the user does not have to bother with the first language sentence. The burden of reading is significantly reduced.

Next, a fourth embodiment (claim 4) will be described. Since natural language is ambiguous, it is difficult to always select the correct part of speech by the part-of-speech estimating means in (3). Therefore, it is necessary to compare the obtained information in the second language with the first language sentence as the source of the information. If the output form determining means outputs the information of the second language as if it were the ruby of the first language so that the correspondence between the two languages can be visually obtained, the user can easily compare the two languages. Therefore, the burden of reading comprehension is significantly reduced.

Next, a fifth embodiment (claim 5) will be described. By saving the format of the first language sentence and outputting the result after adding the information of the second language as an image of the document of the first language, the second language can be displayed without losing the impression of the document of the first language. Can understand the information. For example, when a document in the first language including a chart or a document in the first language having columns is output without saving the format, the user obtains the document in the first language and the present invention. It is necessary to make a reading comprehension while collating two documents with a bilingual sentence. If the bilingual document according to the present invention is output according to the format information such as the chart or column of the document in the first language, the user can concentrate only on the document according to the present invention, and the burden of reading comprehension is significant. It will be reduced.

When a document, which is a set of sentences, is processed in a batch, it is usual to divide it into sentence units before performing language processing such as part-of-speech estimation processing. However, it is difficult to automatically divide each sentence, and it is necessary for the user to confirm and correct it. On the other hand, trigr described in the embodiment
The part-of-speech estimation method using am is convenient because the part-of-speech estimation processing syntax constraint is locally applied to estimate the part-of-speech, so that it is not a fatal problem even if the processing unit is not a whole sentence. However, as for words at both ends of the physical line, the concatenated information cannot be used because the adjacent word is separated into the previous line or the next line. As a result, the accuracy of part-of-speech estimation for words near both ends of a physical line may decrease. Therefore, if the previous line and the immediately following line are stored and connected to the current line and the above-described part-of-speech estimation processing is performed on the connected line, the words near both ends of the line are in the middle of the line, so the part-of-speech estimation accuracy is It never falls.

FIG. 7 is a flowchart of a part-of-speech estimation process for connecting the immediately preceding line and the immediately following line to the current line. Hereinafter, each step will be described in order. step 1 : Clear the contents of the immediately preceding line storage buffer, the current line storage buffer, and the immediately following line storage buffer. step2 : If the line is completed, go to step6. If not, go to step 3. step3 : Read one line into the current line storage buffer. step4: If the line is ended, the process advances to the step6. If not, go to step 5. step5 : Immediately after, one line is read into the line storage buffer. step 6 : The input line is created by concatenating the contents of the immediately preceding line storage buffer, the current line storage buffer, the next line storage buffer, and each buffer.

[0036]step7: Part-of-speech estimation processing for the combined line
Carry out.step8 : Based on the result of part-of-speech estimation processing,
Create a parallel translation.step9 : The contents of the current line storage buffer are stored in the previous line storage buffer
Copy it to the user. Immediately after the contents of the line storage buffer are
Copy to fa.step10 : Immediately after, read one line in the line storage buffer.step11 : If the line ends, go to step 12. At the end
If so, return to step 6. step12 : Previous line storage buffer contents and current line storage buffer
Create an input line by concatenating with the contents of the key.step13 : Perform part-of-speech estimation processing on the line after connection
Itstep14 : Based on the result of part-of-speech estimation processing,
Create a parallel translation.

Although the immediately preceding line and the immediately following line are all connected to the current line in the embodiment, the same effect can be obtained by connecting only a few adjacent words to the current line. Further, there is also an advantage that the processing time is shortened because the input line is shortened. More than,
It is an explanation of claim 6.

Next, claim 7 will be described. OCR
An unknown word may be included in the first language original sentence due to a mistake or a lack of entry of the entry word in the dictionary. Since the part-of-speech is estimated based on the word sequence statistics, when the number of unknown words in the first language sentence increases, the word sequence statistics often cannot be used. Assuming all part-of-speech as candidates for unknown words, the statistics of word sequences will be less accurate if adjacent words are unknown words. Therefore, the number of cases in which correction is made by mistake is increased, and the accuracy of part-of-speech estimation is reduced.

Therefore, when the ratio of unknown words contained in the input line exceeds the set value, the part-of-speech estimation processing is not executed,
Switch to a method that unconditionally selects the first part of speech in the dictionary. Whether or not it is an unknown word can be easily determined as an unknown word if the word is not an entry word in the dictionary. To calculate the ratio of unknown words in one line, a word counter in one line and an unknown word counter may be prepared, and the number of words in one line and the number of unknown words may be counted. The unknown word rate is defined by the following equation. Unknown word rate = unknown word count / total word count If this is larger than the set value, the part-of-speech estimation process is not performed. This can prevent erroneous part-of-speech estimation for lines with many unknown words.

FIG. 8 is a flow chart for judging whether to execute the part-of-speech estimation processing based on the ratio of unknown words. Hereinafter, each step will be described in order. step1 : If the word ends, proceed to step9. If not finished, proceed to step 2. step2 : Read one word. step3 : increment the word counter. step4 : Look up the read word in the dictionary. step5 : If the entry is found as a result of looking up the dictionary, go to step6. If there is no entry word, proceed to step 7. step6 : Store dictionary information. step7 : Store default information for unknown words. step8 : increment the unknown word counter. step9 : Since the input of the line is completed, the unknown word rate of the line is calculated. step10 : If the unknown word rate is higher than the set value, the part-of-speech estimation process is not performed, and the process proceeds to step12. If the unknown word rate is smaller than the set value, part of speech estimation processing is performed.
Go to p11.

Next, claim 8 will be described. Of the second language information corresponding to the first language sentence, it is troublesome that well-known contents are frequently output. Therefore, a list of words whose output is suppressed is set in advance, and the output of the second language information is suppressed for the words corresponding to this list. The word list whose output is suppressed can be easily realized by providing an output suppression column in the dictionary and marking the word whose output is to be suppressed in the output suppression column. FIG. 9 shows an example of a dictionary provided with an output suppression field.

In FIG. 9, the output of the entry word with a circle in the "output suppression" column is suppressed when the second language information is displayed. As a result, the output is suppressed for the first language in which the second language information is well known, and the user can concentrate on the information necessary for the user, so the usability is significantly improved.

Next, claim 9 will be described. Of the headwords with multiple usages, one usage is very well known and the second language information about that usage does not need to be output, but it is used in another usage. Outputs the second language information. For example, the English word "book" is almost always used as a noun and relatively rarely as a verb. Therefore, it is complicated to output the second language information when it is used as a noun, but it is very effective to output the second language information when it is used as a verb. The user does not lose sight of the unfamiliar second language information because the usage is generally rare.

As an example of determining a rare usage, there is a means for entering the priority by a numerical value for each usage as shown in FIG. The larger the priority number, the more preferred the usage. A threshold value may be set in advance for the priority, and the usage having the priority exceeding the threshold may be determined as well-known information, and the output of the second language information may be suppressed. . Alternatively, since it takes a lot of labor to enter the numerical values of the priority, it is possible to arrange the usages in order of priority as shown in FIG. In this case, if there is a circle in the "Output suppression" column, only the highest priority usage (first),
The output of the second language information is suppressed. A display example according to the present invention is shown in FIG. If the English word "book" has a noun usage, the second language information is not displayed, but if it has a verb usage, the second language information is displayed.

Next, claim 10 will be described. Machine processing of natural language has not been completed, and its accuracy is not sufficiently high. The part-of-speech estimation process is not always correct in all cases. Therefore, it is often the case that a plurality of solutions are obtained with the accuracy as a correct answer. In the example of the part-of-speech estimation processing described in the present embodiment, the probability of appearance of the part-of-speech estimation result is obtained as the accuracy of the solution. When outputting only the second language information corresponding to the first solution of the part-of-speech estimation process, if this solution is incorrect, the user cannot obtain correct second language information. Therefore, if the second language information is output in the order of the accuracy of the part-of-speech estimation result, the user can obtain a plurality of solutions in the order of high accuracy. Therefore, even if the part-of-speech estimation result is incorrect, the second language information is not output at all. You can avoid getting lost. In addition to outputting all the solutions in order, how many higher order solutions may be output may be set in advance. 12
FIG. 3 is a diagram showing an example of outputting a plurality of solutions.

Next, claim 11 will be described. As described above, in natural language processing, it is often the case that multiple solutions are obtained with certainty, instead of obtaining only one solution. Therefore, by outputting only the highly accurate solution and outputting multiple ambiguous solutions, it is possible to avoid excessive output of unnecessary solutions, which significantly improves the usability. To be done. In order to determine a highly accurate solution, it is only necessary to compare the obtained accuracy of the plurality of solutions (often quantified). For example, the previous English word "book"
In the case of, it is assumed that the probability of being a book noun: 100 and the probability of being a book verb: 50 are required. The larger the numerical value, the higher the probability that the correct answer is. If the accuracy is higher than a preset value, it is determined that the answer may be correct, and the corresponding second language information is output. If the set value is 30, booked book book will output the translated words of noun and verb.

In addition to the method of setting the threshold value for accuracy,
If the ratio of the accuracy of the solution and the accuracy of the second solution is large, the difference between the accuracy of the first solution and the accuracy of the second solution is large. Therefore, there is a method of determining that the first solution is highly likely to be the correct solution. . For example, if the ratio of the first solution to the second solution is 10: 6 or more, and only the first solution is output, in the above example, the probability of being a book noun: the probability of being a book verb = 100: 50. = 10: 5, only the first solution is output. In addition, there is also a method of obtaining an average value of the accuracies of the top multiple solutions and outputting only the solutions exceeding the average value.

Next, claim 12 will be described. First
The original language may contain unknown words. Claim 1
In the inventions described in to 5, since the part-of-speech is estimated based on the word sequence statistics, when the number of unknown words in the first language sentence increases, the word sequence statistics often cannot be used. Assuming that all parts of speech are candidates for unknown words, the statistics of word sequences will be less accurate if adjacent words are unknown words. Therefore, the number of cases in which correction is made by mistake is increased, and the accuracy of part-of-speech estimation is reduced.
Also, correlative phrases in English (eg "would rather… t
han ~ ":"It's better if you do ... "), etc., and when all words are not concatenated, the information of the word sequences that are separated is used. It is difficult to obtain the second language information corresponding to this in the inventions described in claims 6 to 11.

In order to handle a compound phrase / correlation phrase composed of a plurality of non-connected words (hereinafter, correlation phrase), it is necessary to judge whether or not the correlation phrase is included in the English sentence. For that purpose, the correlation terms may be patterned and stored, and may be collated with the English sentence. Specifically, the matching process is performed according to the following procedure. Of the multiple words that make up the correlation phrase, for the words whose surface is not defined (eg, the words that correspond to "~", "…" in "would rather… than ～"), define their syntactic categories (previously In the above example, a template is created as shown in Table 1 below (a verb or a verb phrase) (hereinafter, an idiom template). In Table 1, what is written between the surface words is an identifier and its syntactic category. For example, A (noun phrase): identifier A, whose syntactic category represents a noun phrase. The syntactic categories used here are defined by the context-free grammar, also attached to Table 1.

[0050]

[Table 1]

[0051]

Morphological analysis is performed on the input sentence as described above. The input sentence is collated with the idiom template in Table 1. For example, the input sentence is "I would rather die than disgrace myself." (4). The “phrase pattern in the first language” column of Table 1 and the input sentence are compared and collated. By the string matching of the surface words, the idiom pattern: would rather A (verb) than B (verb phrase) (5) is matched. At this time, there is a correspondence with A (verb phrase): die ... (6) B (verb phrase): disgrace myself ... (7). By the morphological analysis already performed, I know that. Therefore, it can be easily confirmed from the definition (in context-free grammar) of the "verb phrase" that the syntactic category "verb phrase" satisfies the correspondences of the above (6) and (7). The matching process between the definition by the context-free grammar and the morpheme result can be realized by a general analysis method of the context-free grammar.

By the matching process, the idiom template that matches the input sentence is rather rather A (verb phrase) than B (verb phrase).

[Phase] It is better to perform A than B. (8), so add this to the morpheme candidate. For example, it is summarized in the following format. (Start position, end position, first language surface layer, part of speech, second language information) = (2,7, would rather die than disgrace myself.

[Phase] B is better than B) As a result of the above processing, the correlation word candidates can be handled in the conventional morpheme candidate format. Therefore, the part-of-speech disambiguation can be resolved without changing the part-of-speech disambiguation processing of the above-described embodiment.

Next, claim 13 will be described. According to claim 12, it becomes possible to obtain the second language information of the phrase / correlated phrase composed of a plurality of words that are not concatenated in the first language. In the above method, the second language information of the correlated phrase is obtained. However, the second language information regarding the indefinite element in the correlation phrase cannot be obtained. In the previous example, only A is better than B is displayed, so A (verb phrase): die ... (6) B (verb phrase): disgrace myself ... (7 ) Information is not displayed. In the second language information of only idioms,
It is not enough to read the English sentence.

Therefore, if the information about the indefinite element of the idiom template is also displayed in addition to the idiom information, the second language information necessary for reading the English sentence will be provided. For example, the following is displayed. A is better than B (A: Die, B: Shame on myself) I leave the identifiers A and B in the idiom template, and follow the second language information corresponding to A and B together with the identifier. A, B
The second language information corresponding to is easily obtained by the correspondence relationship of (3) and (4) and the morphological analysis result.
In this embodiment, the identifiers are given in the order of appearance of the phrase patterns in the first language, and the second language information corresponding to that order is displayed. However, the corresponding second language information is displayed in the order of appearance of the identifiers in the second language information. You can display it. According to this, it is better to do A than to B (it becomes a shame of B:
I myself will die: A). Therefore, high-level linguistic processing is not required because each piece of information is simply connected.

Next, claim 14 will be described. According to the thirteenth aspect, although the advanced language processing is not required, since the identifier remains, the naturalness of the second language is impaired. Therefore, by performing language processing in the second language, natural second language information is created over the entire idiom. As a result, the unnaturalness due to the second language is significantly improved, and the easiness of understanding is improved. Examples will be shown below.
In order to transform / synthesize in the second language, in the second language, utilization information is added to the dictionary for words that have utilization. die

[Verb] die (na ・ 5dan) disgrace

[Name] Disgrace

[Verb] becomes a shame (Na ・ 5dan) myself

[Name] I myself add the word order in the second language in the definition of syntactic categories. For example, in English, the order is verb noun: eg disgrace myself, but in Japanese, it is a noun verb: eg my own shame. An example for the syntax categories in Table 1 is shown. The word order of the first language is on the left of the ":", and the word order of the second language is on the right.

As described above, by preparing the word order in the second language and the second language utilization information of morpheme candidates,
The second language information of the morpheme candidate "would rather die than disgrace myself" is "It is better to die than to be ashamed of myself." This transformation / synthesis processing is a general sentence generation method such as machine translation. As described above, by performing the language processing on the second language, the identifier and the like are unnecessary, so that the second language can be displayed in a natural format, and the readability is remarkably improved.

Next, claim 15 will be described. OC
An unknown word may be included in the first language original sentence due to an R mistake, insufficient entry of a dictionary entry in the dictionary, or the like. In the above-described embodiment, since the part-of-speech is estimated based on the word sequence statistics, when the number of unknown words in the first language sentence increases, the word sequence statistics often cannot be used. Assuming all parts of speech as candidates for unknown words, if adjacent words are unknown words, the word sequence statistics will be less accurate. Therefore, the number of cases in which correction is made by mistake is increased, and the accuracy of part-of-speech estimation is reduced.

By the way, a characteristic surface pattern is generally found in words belonging to the same syntactic category. For example, in English, the verb (rarely adjective) is "
The word "ment" is connected to the end of the word to make it a noun. Other words that end in "able" and "ish" are adjectives. Thus, by paying attention to the characteristic surface pattern, the syntax of the word As described above, in this embodiment, there are two types of unknown words: A. Caused by an OCR miss and B. Not found in a dictionary entry word. Even in such a case, if the surface layer pattern can be detected, the syntactic category of the unknown word can be estimated.
The surface patterns are summarized in Table 2 below.

[0060]

[Table 2]

As a result of the dictionary lookup, when there is no corresponding entry word, an attempt is made to estimate the syntactic category of the unknown word. Apply the surface pattern of Table 2 to unknown words,
If they match, the syntactic category to which the pattern belongs is determined as the syntactic category of the unknown word. If there is no matching surface pattern, all syntax categories are candidates. By this processing, the syntactic category of the unknown word can be estimated, so that it is possible to suppress the deterioration of the accuracy of part-of-speech estimation due to the unknown word.

FIG. 13 uses the surface layer pattern of claim 15.
In the flowchart of the process of estimating the syntactic category of an unknown word
is there.step1 : First, read the word and go to step 2.step2 : Next, search the dictionary and proceed to step 3.step3 : If the word exists in the dictionary headword,
Proceed to step 5. If it does not exist, go to step 4.step4 : The word is used as a surface pattern for syntax category estimation.
Collate and proceed to step 6. step5 : Determine the syntactic category of the word based on the dictionary
Set.step6 : St if there is a surface pattern that matches the word
Go to ep8. If not, proceed to step 7.step7 : Determine that the word is an unknown word.step8 : Syntax category corresponding to the matching surface pattern
To the syntactic category of the word.

Next, claim 16 will be described. Among unknown words, there is a word that is determined as an unknown word because there is no entry word in the dictionary. It is natural that a completely new word is not registered as a headword, but there are many cases where a derivative word from which a basic word can be easily derived is not registered. Such a derivative word is a modification of the basic word based on the surface layer pattern of Table 2. Therefore, the surface pattern of Table 2 is applied to the unknown word, the word is returned to the basic word, and the dictionary is searched again. If the basic word exists, the second language information and the surface pattern information are combined. To display. A display example is shown in FIG. The English word "embellishment" is a word that is used infrequently and is often not listed in a simple dictionary. Therefore, unless a dictionary with a large number of headwords is used, it is likely to be determined as an unknown word. Even with such difficult words, the second language information can be obtained incompletely. In the example of FIG. 14, "embellish" (

[Verb] Make it beautiful), and "embellishment" does not exist. As a result, not only the syntactic category of the unknown word can be determined, but also the second language information of the unknown word can be displayed incompletely, and reading of an English sentence including the unknown word is significantly facilitated. Also, apparent unknown words (words excluding completely new words) can be reduced without increasing the number of index words in the dictionary.

FIG. 15 uses the surface layer pattern of claim 16.
Flow chart for synthesizing unknown language second language information
It isstep1 : First, read the word and go to step 2.step2 : Next, search the dictionary and proceed to step 3.step3 : St if the word is present in the dictionary entry
Proceed to ep5. If it does not exist, go to step 4.step4 : The word is used as a surface pattern for syntax category estimation.
Collate and proceed to step 6. step5 : Determine the second language information of the word based on the dictionary
Set.step6 : St if there is a surface pattern that matches the word
Go to ep8. If not, proceed to step 7.step7 : The word is judged as an unknown word and proceed to step 12.
Mu.step8 : Return the word to the basic form based on the surface pattern
Then, proceed to step 9.step9 : Dictionary search for the basic form found in step 8 above
And proceed to step 10.step10 : There is a basic form of the word in the dictionary entry
If so, proceed to step 11. If it does not exist, proceed to step 13.
Mu.step11 : Second obtained from surface pattern for syntax category estimation
The language information and the second language information obtained by searching the basic form in a dictionary
To synthesize.step12 : The second language information is empty.step13 : The second language information is empty.

According to claim 16, the second word concerning the unknown word
Although the linguistic information can be obtained, the syntactic category name in the first language, which is unnatural for the second language (in FIG. 14,

[Name]) is displayed. The user selects the second language information (“beautify”) and the syntax category (

Since it is necessary to understand after combining two kinds of information different from (name)), smoothness is impaired.
Therefore, if the second language information is typically transformed to indicate the syntax category of the first language, it is not necessary to display the syntax category of the first language together.

The present invention will be described below with reference to examples.
The second, which expresses the syntax category of the first language as shown in Table 2,
Here is a summary of typical language expressions. For example, the ending is "
The syntactic category of a first language (English) word that is "ish" is an adjective, and a typical expression in the second language (Japanese) that indicates it is "like". This expression and
It is combined with the second language information other than the part that is the source of this expression. In the previous example, "embellishment" is "embellish"
+ "Ment", so the second language information about embellish: beautify ment: ~ is required, and by combining these, embellishment: beautification is required.

In this embodiment, since language processing is performed in the second language, modification information (type of utilization etc.) must be stored together with the second language information. The language processing in the second language is a general sentence generation method such as machine translation. In order to distinguish the second language information obtained by the normal method from the second language information obtained by the present invention, when displaying the second language information according to the present invention, a hyphen is displayed as follows. The range of deformation may be specified using symbols such as ("-") and parentheses ("(", ")"). embellishment beautify-thing beautify (thing) beautify +

Furthermore, if the OCR device and the printing device are merged, the bilingual document creation processing is built in, and the devices are put together in a copy machine-like device, the user can obtain the bilingual document as if copying the document. Therefore, the usability is remarkably improved as compared with the case where the individual devices are combined as shown in FIG. In addition, the scanning process of the OCR device and the printing process of the printing device are very similar to the copying process of a commercially available copy machine, and there is only a slight cost increase when implementing the present invention on the copy machine. FIG. 16 shows a device image of the parallel translation copying machine.

The present invention is not limited to the above embodiments, and Means for estimating the part of speech in the first language 1. Means for displaying or printing out the first language sentence and the second language information in association with each other. 3. Means for storing the format of the first language document 4. Means for displaying or printing out the first language sentence and second language information based on the format of the first language document. By combining OCR (optical character reading device) and printing device and putting them together in a copy machine-like device to imitate the feeling of use of the copy machine, the first language document of It is essential to provide a machine translation device having a function of helping reading in a second language.

Also, 6. Input means for inputting first language sentence 7. First language sentence (original sentence) storage means 8. 8. An analysis dictionary that stores the knowledge necessary for linguistic analysis. The first language sentence (original sentence) and its translation result, the second sentence
Display means for displaying language sentence 10. Means for estimating the part of speech of the first language 11. 12. A language processing system comprising: means for extracting second language information (translated word) corresponding to a first language from an analysis dictionary in accordance with the part of speech estimated by the estimating means of 10. 13. A means for estimating the part-of-speech of a multi-part-of-speech word using information about the connection between words Means for storing the immediately preceding line and the immediately following line in addition to the current line 14. 14. A means for connecting the immediately preceding row and the immediately following row stored in 13 to the current row. 15. A means for estimating a part of speech for the line after connection created in the above 14. 17. Means for determining unknown word included in first language sentence 17. A means for counting the number of unknown words determined in the above 16. 18. A means for judging whether or not to perform the part-of-speech estimation processing based on the counting result of the above 17. 20. Means for determining a basic word in the first language sentence 20. A means for suppressing the output of the second language information based on the judgment result of 19 above. Part-of-speech means for determining priority of estimated part-of-speech in dictionary 22. 23. A means for suppressing the output of the second language information based on the judging means of 21. 24. A means for obtaining the rank of the possibility of part-of-speech by part-of-speech estimation A machine translation having a function of outputting only the second language information required by the user with high accuracy by providing means for outputting the second language information corresponding to the plurality of high-ranking part-of-speech candidates according to the order of 23 above. There is an essence in providing the device.

25. Input means for inputting first language sentence 26. First language sentence (original sentence) storage means 27. 28. An analysis dictionary that stores the knowledge necessary for language analysis 28. First language sentence (original sentence) and its translation result, second sentence
Display means for displaying language sentence 29. A means for estimating the part of speech of the first language 30. 31. In a language processing system including means for extracting second language information (translated word) corresponding to a first language from an analysis dictionary according to a part of speech estimated by the estimating means of 29. 32. A means for estimating the part of speech of a multi-part-of-speech word using information about the connection between words. Second language information (translation word, part of speech, etc.) corresponding to the idiom template that describes the syntactic type of the idiom in the first language
33. Means for storing together 34. A means for collating the template stored in 32 with the first language sentence. 34. A means for extracting the word (phrase) in the first language sentence corresponding to the element in the idiom template in the matching means of 33. A machine translation device comprising: means for displaying the second language information corresponding to the idiom template based on the execution results of the matching means of 33 and the extracting means of 34. 36. The second language information of the phrase template and the words (phrases) in the first language sentence that match the elements in the template
A machine translation device having means for separating and outputting the second language information corresponding to. 37. The second language information of the phrase template and the second language information corresponding to the word (phrase) in the first language sentence that matches the element in the template are transformed / combined and output as one second language phrase. Means 38. 39. Means for detecting an unknown word in the first language sentence A means for storing the transformation pattern of the word of the first language and the syntactic category (part of speech etc.) after transformation 41. Means for applying 14 transformation patterns to the unknown word detected by the 37 detection means to obtain a word before transformation 41. 42. A unit for retrieving the untransformed word obtained in 40 in the first language analysis dictionary. A machine translation apparatus having means for displaying together the second language information and the transformation information of the first language word before transformation according to the result of the search means of 41. 43. According to the result of the search means of 41, the second language information of the first language word before the transformation is transformed according to the transformation information in the first language sentence, and then the means for displaying the transformed second language information is necessary for the user. It is essential to provide a machine translation device having a function of outputting only the second language information with high accuracy.

[0072]

As is apparent from the above description, the present invention has the following effects. (1) In claims 1 and 2, only the information in the second language (part of speech, translation, etc.) necessary for understanding the first language sentence can be selected and presented to the user. The burden of reading and comprehending sentences can be significantly reduced. (2) In claim 3, the input speed is remarkably improved,
Since the user can concentrate on understanding the output result because it does not bother humans, the burden of reading the first language sentence is significantly reduced. (3) In claim 4, if the information of the second language is output as if it is the ruby of the first language so that the correspondence between the two languages can be visually obtained, the user can easily collate the two languages. Therefore, the burden of reading comprehension is significantly reduced. (4) In claim 5, if the bilingual document according to the present invention is output according to the format information such as the chart or column of the document in the first language, the user does not need to refer to the original document. So we can concentrate exclusively on the document according to the invention,
The reading burden is significantly reduced. (5) According to the sixth aspect, the concatenation constraint can be used near both ends of the line, so that it is possible to avoid a decrease in part-of-speech estimation accuracy. (6) In claim 7, since it is possible to avoid performing the part-of-speech estimation process on a line with many unknown words,
It is possible to avoid lowering the accuracy of part-of-speech estimation. (7) According to claim 8, the output of known information is suppressed, so that the user can concentrate on understanding unknown information, and the usability is significantly improved. (8) According to the ninth aspect, the second language information regarding the rare usage is output, which can help the user to understand. (9) In claim 10, since a plurality of solutions are output in order of priority, even if the part-of-speech estimation process is incorrect,
The user can select an appropriate solution and continue reading. (10) In claim 11, since a plurality of solutions are output only for ambiguous solutions, the user does not see unnecessary information about solutions that are likely to be correct, and there are a plurality of ambiguous solutions. An appropriate solution can be selected from, and the burden on reading comprehension can be significantly reduced. (11) According to claim 12, the second language information of the idiom / correlated phrase of the first language, which is composed of a plurality of words that are not connected, can be obtained, so that the reading of the first language sentence is significantly facilitated. . (12) In claim 13, the second language information of the phrase correlation phrase of the first language, which is composed of a plurality of words that are not connected, is output without performing language processing in the second language.
It is not necessary to have language processing for the second language. (13) In claim 14, the second language information of the idiom / correlated phrase in the first language, which is composed of a plurality of words that are not connected, can be transformed / combined by performing language processing in the second language. Improves readability in two languages significantly. (14) In claim 15, since the syntactic category candidates for the unknown word can be limited, it is possible to avoid lowering the accuracy of part-of-speech estimation of the adjacent word. (15) In claim 16, the syntactic category candidates for the unknown word can be limited, and the second language information of the unknown word can be output although it is unnatural as the second language. The readability in the second language is significantly improved without providing the language processing for the two languages. (16) In claim 17, the syntactic category candidates for the unknown word can be limited, and the second language information of the unknown word can be output as the second information in a natural form. Significantly improve readability in language.

[Brief description of drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of a machine translation device according to the present invention.

FIG. 2 is a block diagram of bilingual document creation processing according to the present invention.

FIG. 3 is a configuration diagram of a bilingual document creation processing unit according to the present invention.

FIG. 4 is a diagram showing a maximum appearance probability part-of-speech string selection procedure according to the present invention.

FIG. 5 is a diagram showing a flowchart of a maximum appearance probability part-of-speech string selection procedure according to the present invention.

FIG. 6 is a diagram showing a flowchart of a solution candidate updating process according to the present invention.

FIG. 7 is a diagram showing a flowchart of part-of-speech estimation processing for connecting the immediately preceding line and the immediately following line to the current line according to the present invention.

FIG. 8 is a diagram showing a flowchart of a judgment of execution of part-of-speech estimation processing based on an unknown word rate according to the present invention.

FIG. 9 is a diagram showing an example of a dictionary having an output suppression field according to the present invention.

FIG. 10 is a diagram showing an example of a dictionary having a part-of-speech priority according to the present invention.

FIG. 11 is a diagram showing an example of output suppression for frequent usage according to the present invention.

FIG. 12 is a diagram showing an example of outputting a plurality of solutions according to the present invention.

FIG. 13 is a flowchart of an unknown word syntax category estimation process using a surface pattern according to the present invention.

FIG. 14 is a display example of second language information of an unknown word according to the present invention.

FIG. 15 is a flowchart of a synthesis process of second language information of an unknown word using a surface layer pattern according to the present invention.

FIG. 16 is a diagram showing an image of a device of a parallel translation copying machine according to the present invention.

[Explanation of symbols]

1 ... Display device, 2 ... Display control unit, 3 ... Language processing unit, 4 ...
Input control section, 5 ... Input device, 21 ... First language input means, 22 ... Morphological analysis means, 23 ... Analysis dictionary, 24 ... Part of speech estimation means, 25 ... Part of speech concatenation probability table, 26 ... Output form determination means, 27 … Output means.

Claims (17)

[Claims] 1. An input unit for inputting at least a target language sentence, a storage unit for storing the target language sentence input by the input unit, and an analysis dictionary for storing knowledge necessary for language analysis. In a language processing system including the target language sentence and a display unit for displaying a target language sentence that is a translation result of the target language sentence, a part-of-speech estimating unit that estimates the part-of-speech of the target language, and the part-of-speech estimating unit estimates the part of speech. A machine translation device comprising: an extracting unit that extracts information of a target language corresponding to a target language from the analysis dictionary according to a part of speech. 2. The part-of-speech estimating unit includes a first storage unit that stores a concatenation probability of a part-of-speech of a target language, and a determining unit that determines a part-of-speech of each word of the target language sentence based on the concatenation probability. The machine translation device according to claim 1, characterized by comprising. 3. The machine translation device according to claim 1, wherein the input means uses an optical character reader means. 4. The machine translation device according to claim 1, wherein the display unit has an output unit that displays or prints out the target language sentence and the target language information in association with each other. 5. The output means displays or prints out the target language sentence and the target language information based on the second storage means for storing the format of the target language document and the format of the target language document. The machine translation device according to claim 4, further comprising output means. 6. At least an input means for inputting a target language sentence, a storage means for storing the target language sentence input by the input means, and an analysis dictionary storing knowledge necessary for language analysis. A display unit for displaying the target language sentence and a target language sentence that is a translation result thereof, a part-of-speech estimation unit for estimating a part-of-speech of the target language, and the analysis dictionary according to the part-of-speech estimated by the part-of-speech estimation unit. In a language processing system having an extracting means for extracting information of a target language corresponding to a target language from, a part-of-speech estimating means for estimating a part-of-speech of a multi-part-of-speech word using information about connection between words,
In addition to the current row, a storage means for storing the immediately preceding row and the immediately following row, a connecting means for connecting the immediately preceding row and the immediately following row stored in the storage means to the current row, and a connected row created by the connecting means A machine translation device comprising: a part-of-speech estimating means for estimating a part-of-speech. 7. A determination unit that determines an unknown word included in the target language sentence, a counting unit that counts the number of unknown words determined by the determination unit, and a part-of-speech estimation process based on the counting result of the counting unit. 7. The machine translation device according to claim 6, further comprising a determination unit that determines whether or not to execute. 8. A determination means for determining a basic word in a target language sentence, and a suppression means for suppressing output of target language information based on a determination result of the determination means. The machine translation device described. 9. The method according to claim 6, further comprising: a determination unit that determines the priority of the part-of-speech estimated part-of-speech in the dictionary, and a suppression unit that suppresses the output of the target language information based on the determination unit. Machine translation device. 10. An authenticating means for obtaining a rank of likelihood of part of speech by part-of-speech estimation, and an output means for outputting second language information corresponding to a plurality of high-ranking part-of-speech candidates in accordance with the ranking by the acknowledging means. 7. The machine translation device according to claim 6, wherein: 11. A first solution using a qualifying means for obtaining a plurality of possibilities of a part of speech by part-of-speech estimation, an evaluation means for evaluating the degree of possibility of the recognizing means, and a plurality of the degree of possibility of the evaluation. 7. The machine translation apparatus according to claim 6, further comprising: a determining unit that determines the accuracy of the above, and a changing unit that changes the number of solutions to be output based on the determination result of the determining unit. 12. At least an input means for inputting a target language sentence, a storage means for storing the target language sentence input by the input means, and an analysis dictionary for storing knowledge necessary for language analysis. A display unit for displaying the target language sentence and a target language sentence that is a translation result thereof, a part-of-speech estimation unit for estimating a part-of-speech of the target language, and the analysis dictionary according to the part-of-speech estimated by the part-of-speech estimation unit. In a language processing system equipped with an extraction means for extracting information of a target language corresponding to a target language, a part-of-speech estimation means for estimating a part-of-speech of a multi-part-of-speech word using information about concatenation of words A storage unit for storing the compound phrase template for describing the syntactic type of, together with corresponding target language information, and a collation unit for collating the stored template with the target language sentence; Extraction means for extracting the word or phrase in the target language sentence corresponding to the element in the phrase template in the matching means, and the target language information corresponding to the phrase template based on the execution results of the matching means and the extraction means A machine translation device comprising: 13. The target language information of the idiom template,
13. An output unit that separates and outputs target language information corresponding to a word or a phrase in a target language sentence that matches an element in the template.
The machine translation device described. 14. The target language information of the idiom template,
Transformation based on the target language information corresponding to the word or phrase in the target language sentence that matches the element in the template
13. The machine translation device according to claim 12, further comprising an output unit that synthesizes and outputs as one target language phrase. 15. A detection unit for detecting an unknown word in a target language sentence, a storage unit for storing a transformation pattern of a word in the target language and a syntactic category after transformation, and an unknown word detected by the detection unit. An applying means for applying the transformation pattern to obtain a pre-transformation word, and a searching means for searching the first language analysis dictionary for the pre-transformation word found by the identification means.
13. The machine translation device according to claim 12, further comprising: determining means for determining a syntactic category of an unknown word according to a result of the searching means. 16. The machine translation according to claim 15, further comprising display means for displaying together target language information of the target language word before transformation and transformation information according to the result of the search means. apparatus. 17. The display means for displaying the target language information of the target language word before transformation after transforming the target language information of the target language word with the transformation information of the target language sentence according to the result of the search means. Machine translation device.