JPH0728821A - Text processor - Google Patents

Abstract

PURPOSE:To generate a proper translation in accordance with a pattern match definition and an action definition by presenting the vocabulary information of the original by utilizing a machine translation engine and correcting the information by these definitions by looking at the information by a user. CONSTITUTION:This processor is provided with a machine translation engine 21, a morpheme analysis calling part 2 calling a morpheme analysis for an inputted original, a vocabulary information output processing part 3 taking out parts of speech and representative attributes for each word for which the morpheme analysis is performed by referring to a vocabulary information table 8 and a processing execution part 6 generating a translation by performing a processing in accordance with the action definition 14 when a matching is obtained by performing a pattern collation with the pattern match definition 13 in a preliminarily prepared tuning definition 12 based on the word of the inputted original, the taken out parts of speech and a grammar attribute.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text processing device which uses a machine translation engine to perform text processing.

[0002]

2. Description of the Related Art In the current machine translation, an input original sentence is morphologically and syntactically analyzed into a structured representation, and then the translated sentence is formed into a structured representation of a language to be translated.
At this time, since a grammar and a dictionary are used, these are maintained so that a more complete translation can be created.

Usually, when a document is edited, a word processor is used, but this is done by the user in consideration of the structure and contents of the document.

[0004]

It is difficult for the former machine translation described above to generate a complete translation. Even if the grammar and dictionaries are improved, it may not be possible to obtain the perfect translation for the user. Even if the translation result is valid, there is a problem that if the translation is not as desired by the user, it will be treated in the same way as the failed sentence.

Further, the editing of the document by the latter word processor or the like described above is not sufficient for operating the created document, and is merely pattern matching or change of the letter surface. There is a problem that text processing is insufficient because it is not possible to operate a more desirable document by utilizing valid information from a document which is an asset.

The present invention solves these problems.
Present the original vocabulary information using a machine translation engine,
The user sees this and corrects it by the pattern matching definition and the action definition, and the purpose is to generate an appropriate translated sentence according to these definitions.

[0007]

FIG. 1 is a block diagram showing the principle of the present invention. In FIG. 1, the morphological analysis calling unit 2
Is to call the morphological analysis for the input original sentence to perform morphological analysis.

The vocabulary information output processing unit 3 refers to the vocabulary information table 8 for each word subjected to morphological analysis,
For each word whose representative attribute is extracted or morphologically analyzed, the dictionary 9 is searched and the narrowed table 10 is searched based on the encrypted grammatical attribute extracted and the encrypted representative attribute is extracted. The part of speech and the representative attribute of the extracted encrypted representative attribute are referred to by referring to the corresponding code table 11.

The processing execution unit 6 performs pattern matching with the pattern match definition 13 in the tuning definition 12 created in advance based on the input original sentence word, the extracted part-of-speech, and grammatical attribute, and when there is a match, it According to the action definition 14 of No. 1, processing is performed to generate a translated sentence.

The machine translation engine 21 is for morphological analysis, syntactic analysis, and translation generation.

[0011]

According to the present invention, as shown in FIG. 1, the morpheme analysis calling unit 2 calls the morpheme analysis of the machine translation engine 21 to perform morpheme analysis on the input original sentence, and the vocabulary information about the resulting word is used. The output processing unit 3 refers to the vocabulary information table 8 to extract a part of speech and a representative attribute, and the process execution unit 6 extracts a part of speech and a grammatical attribute from the original sentence input word and the extracted part of speech and grammatical attribute. When the pattern is matched with the pattern match definition 13 of No. 1 and is matched, the processing is performed according to the action definition 14 to generate a translated sentence.

The morphological analysis calling unit 2 calls the morphological analysis of the machine translation engine 21 to perform morphological analysis on the input original sentence, and the lexical information output processing unit 3 searches the dictionary 9 for the resulting word. Then, based on the encrypted grammatical attribute extracted, the narrowing table 10 is searched to extract the encrypted representative attribute,
With respect to the extracted representative attribute, the part-of-speech and the representative attribute are extracted with reference to the correspondence code table 11, and the processing execution unit 6 preliminarily uses the input original word and the extracted part-of-speech and grammar attribute. When the pattern matching definition 13 in the created tuning definition 12 is matched with the pattern, the action definition 14 is processed to generate a translated sentence.

In these cases, with respect to the input original sentence word, the extracted part-of-speech, and grammatical attribute, the pattern match definition 13 in which the word notation is removed or expressed by a regular expression, and the pattern matching the pattern match definition 13 are applied. The action definition 14 defining the process is created in advance and saved.

Therefore, the vocabulary information of the original sentence is presented by using the machine translation engine, the user sees this and corrects it by the pattern matching definition 13 and the action definition 14 and saves it. It can be automatically generated.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the construction and operation of an embodiment of the present invention will be described in detail with reference to FIGS.

FIG. 1 is a block diagram showing the principle of the present invention. Figure 1
In the text processing device 1 with a machine translation engine,
A text processing apparatus that uses the machine translation engine 21 to generate a translated sentence, and includes a morphological analysis calling unit 2, a vocabulary information output processing unit 3, a tuning control unit 4, a result output unit 7, and the like. .

The morphological analysis calling unit 2 calls the morphological analysis 22 in the machine translation engine 21 to request morphological analysis of the original sentence in order to perform morphological analysis of the input original sentence.

The vocabulary information output processing unit 3 generates vocabulary information of the input original word and stores it in the vocabulary information table 8 (see FIG. 2). The tuning control unit 4 determines which of the pattern matching definitions 13 created and stored in advance matches the input original sentence, and when matching, executes the process according to the action definition 14 from the process execution unit 6 or the like. It is configured (see FIG. 4).

The matching processing unit 5 performs morphological analysis on the input original sentence to obtain vocabulary information (part of speech, grammatical attribute, etc.).
And the pattern matching definition 13 in the tuning definition 12 which is defined and saved in advance, and the matching one is found.

When the matching processing unit 5 finds a matching pattern matching definition 13, the processing executing unit 6 executes processing according to the action definition 14.

The result output unit 7 outputs the result of generating a translated sentence. The vocabulary information table 8 stores vocabulary information (part of speech, grammatical attributes, etc.) generated by the vocabulary information output processing unit 3 (see (d) of FIG. 3).

The dictionary 9 is a dictionary in which grammatical attributes encrypted in association with the notation are registered in advance. The narrowing table 10 is a table for extracting a representative attribute from the encrypted grammatical attributes retrieved from the dictionary 9 (see (b) of FIG. 3).

The correspondence table 11 is the narrowing table 1
Corresponding to the encrypted representative attribute extracted from 0,
The grammatical attributes and parts of speech that are not encrypted are registered in advance.

The tuning definition 12 is a definition for tuning a translated sentence, and is a pattern match definition 13
And action definition 14.
The pattern match definition 13 defines in advance a pattern for finding an action definition 14 to be applied to the input original sentence word and vocabulary information pattern. For example,-, noun, TOOL; *; in; error; Here, −− represents notation removal, * represents a regular expression and represents an arbitrary character string.

The action definition 14 defines the processing content to be executed when the pattern of the original sentence and the vocabulary information match the pattern match definition 13.
For example (F: DICT $ 1); contain (F: CON $ 1); an; error ;. Here, (F: DICT $ 1) means to look up the first ($ 1) word in the dictionary (F: DICT). contain (F: CO
(N $ 1) indicates that the first word depends on the conjugation (F: CON) and is made into cotains or cotain. Including these functions, the following functions are used in the action definition.

DICT: Dictionary lookup CON: Utilization TRANS: Translation file name: User tool (data is piped) PUT: Output character string following standard output Next, in accordance with the order shown in the flowchart of FIG. 2, vocabulary information and actions The operation when creating a definition will be described in detail.

In FIG. 2, S1 reads the original sentence.
This reads, for example, (1) in which there is an error in the original text copier processing described on the right side.

At S2, morphological analysis is performed. As a result of the morphological analysis, as described on the right side, Copiler; Treatment; Here ,; represents a word delimiter.

In step S3, it is determined whether or not there is morpheme information. This is to extract morpheme information (words) one by one from the beginning of the result of the morpheme analysis of (2) and determine whether there is still morpheme information. In the case of YES, it progresses to S4. In the case of NO, since there is no morpheme information, the process proceeds to S10.

In step S4, vocabulary information is extracted from one piece of morpheme information. This is one morpheme information, for example, the lexical information "noun, TOOL" from "compiler", S5 to S
It is taken out by 9 and set in the buffer. This will be described below.

At S5, the representative information is narrowed down at S6 and S7. In S6, the route attribute is searched. This is shown in FIG. 3 for the morpheme information “compiler”, for example.
The encrypted grammatical attribute “grammar attribute 1, 2, 3, m of the“ compiler ”extracted by referring to the dictionary 9 in (a) of FIG.
For “1, m22”, a route attribute search is performed by referring to the narrowing-down table 10 in FIG.
For, the root attribute m-m2-m22 is found and taken out.

In step S7, the representative attribute is extracted. This is S
Of the root attributes found in 6, the m22 at the end is taken out as the representative attribute. S8 is the corresponding code table 11
, And convert the representative attributes into unencrypted grammar attributes and parts of speech. For example, regarding the representative attribute m22 extracted in S7, the corresponding code table 11 in FIG. 3C is looked at, and the part of speech and grammatical attribute of m22 are taken and converted into "noun, TOOL".

S9 is set in the buffer. This is 1
For one morpheme information (eg "compiler"), S
Part of speech and grammatical attributes converted in 8 (for example, “noun, TOO
L ”) together word / part of speech grammatical attribute compiler, noun, and TOOL (3) are set in the buffer as thesaurus information (information in which words are classified and arranged according to meaning).

In step S10, the vocabulary information A is output. As described above, it is possible to obtain vocabulary information and grammatical attributes for each morpheme information obtained by morpheme analysis of the original sentence, and generate lexical information in which words are classified and arranged according to meaning. For example, the following vocabulary information (4) can be generated and output with respect to the original sentence (1).

Copier, noun, TOOL; processing; error; there is; a (4) Next, the creation of the tuning definition 12 will be described in detail. In FIG. 2, S11 is edited by the user. In this case, the user removes the notation or uses a regular expression to create the vocabulary information A ′ (pattern match definition). For example, the vocabulary information on the right side: compiler, noun, TOOL; process; to; error; to; a; (5) Here, as described on the right side, the "compiler" notation is removed and "process" The lexical information A'in (6) below, in which "" is replaced with the regular expression "*" (arbitrary character string)
(Pattern match definition 13) is created.

---, noun, TOOL; * ;; error; there is; there; (6) In S12, the user creates the action definition 14. This creates (F: DICT $ 1); contain (F: CON $ 1); an; error ;. (7), for example, as described on the right. Here, (F: DICT $ 1) represents the dictionary lookup of the first word. contain (F: CON $ 1) indicates that the use of contain depends on the first word.

As described above, when the pattern matching the pattern matching definition 13 (6) appears in the vocabulary information of the original sentence, the definition (tuning) to execute the process defined by the action definition 14 of (7). The definition 12) is created and saved.

FIG. 3 shows an example dictionary / table of the present invention. FIG. 3A shows an example of the dictionary 9. This dictionary 9
As shown in the figure, the notation and the encrypted grammar attribute are registered in advance. Here, for example, the following notations are registered in advance as grammar attributes (encrypted) compiler grammar attributes 1, 2, 3, m1, m22 converter grammar attributes 7, 8, 9, m22, x11.

FIG. 3B shows an example of the narrowing table 10. This narrowing table 10 is shown in FIG.
This is for searching the root attribute and retrieving the representative attribute in the encrypted grammar attribute retrieved and retrieved from the dictionary 9. For example, from the dictionary 9, the grammar attribute of the notation "compiler""grammar attribute 1, 2, 3, m1, m22"
When m is extracted, m22 of the grammatical attributes finds a root attribute called grammatical attribute m-m2-m22. The last m22 is extracted as a representative attribute. Then, from the corresponding code table 11 of FIG.
The part of speech of 2 "and the unencrypted grammatical attribute are extracted and used as vocabulary information.

FIG. 3C shows the corresponding code table 11
For example: The correspondence code table 11 is for extracting unencrypted grammatical attributes and parts of speech from the representative attributes found by referring to the narrowing-down table 10 of FIG.

FIG. 3D shows an example of the vocabulary information table 8. This vocabulary information table 8 includes S1 to S in FIG.
The vocabulary information (notation, part-of-speech, grammatical attribute) generated by 10 is stored.

Next, according to the order shown in the flowchart of FIG. 4A, the tuning definition 12 including the pattern match definition 13 and the action definition 14 created and defined according to the flowchart of FIG. The operation when generating a translated sentence will be described in detail.

In FIG. 4A, in S21, initial processing and rule expansion are performed. This is to read the contents of the tuning definition file 8 (pattern match definition 13, action definition 14) as various initial processes and rules to make it available, and at the same time, perform core expansion to make the translation after S22 possible. .

A step S22 decides whether or not there is an input sentence. YE
If S, go to S23. If NO, the process ends (END). S23 performs morphological analysis. In this, the morphological analysis calling unit 2 calls the morphological analysis of the machine translation engine 21 and requests the morphological analysis of the input sentence. Here, morphological analysis is performed as shown in (8) below.

Converter; function; to; error; to; yes; (8) S24 extracts vocabulary information from one morpheme information.
For this, one morpheme information, for example, the lexical information “noun, TOOL” from “converter” is taken out from S25 to S29 and set in the buffer.

In S25, the representative information is narrowed down in S26 and S27. In S26, the route attribute is searched. This is, for example, for the morpheme information “converter”, the encrypted grammatical attribute “grammar attribute 7,” which is extracted by referring to the dictionary 9 in FIG.
For 8, 9, m22, x11 ", the route attribute search is performed by referring to the narrowing table 10 in FIG. 3B, and here, for m22, the route attribute m-m2-m22 is found and taken out.

In step S27, the representative attribute is extracted. this is,
Of the root attributes found in S26, as a representative attribute,
Take out the last m22. In step S28, the correspondence code table 11 is looked up and the representative attribute is converted into the unencrypted grammar attribute and part of speech. For example, regarding the representative attribute m22 extracted in S27, the corresponding code table 11 in FIG. 3C is looked at, and the part of speech and grammatical attribute of m22 are set and converted into "noun, TOOL".

S29 is set in the buffer. this is,
For one morpheme information (eg "converter"),
Part of speech and grammatical attributes converted in S28 (for example, "noun, TO
OL ”) together words / parts of speech / grammar attribute converter Noun, TOOL (9) is set in the buffer as thesaurus information (information in which words are classified and arranged by meaning).

By repeating the above processing, the following vocabulary information (10) is generated. Converter, noun, TOOL; function; to; error; but; A; Yes (10) S30 performs matching processing. For this, the vocabulary information (10) created from the input sentence is subjected to pattern matching with the pattern match definition 13 in the tuning definition 12 registered in advance, and a matching one is found.
In this case, it is found that matching is performed as described on the right-hand side; “error” is present in the noun, TOOL; *; defined in S11 and S12 of FIG.

In step S31, it is determined whether a certain rule (pattern match definition 13) is matched. Here, since the answer is YES, the process proceeds to S32. In the case of NO, since the matching rule (pattern match definition 13) was not found, the process ends.

At S31, action processing is performed. Since this is found to match with YES in S31, the action definition 14 of the matched pattern match definition 13,
For example, the process according to the following action definition 14 (F: DICT $ 1); contain (F: CON $ 1); an; error ;. Here, as described on the right side, (F: DICT $ 1) looks up the first word (converter) in the dictionary (CASE of S33: Look up in dictionary). here,
You get a converter.

As described on the right side, contain (F: CON $ 1) makes the utilization depend on the first word (converter) (CASE of S34: utilization processing). Here, converte
rs.

FIG. 4B shows the translation result. Here, as a result of the above-described processing of S21 to S34, the translation result has an error in the compiler processing → The mistake was foun
d in the compiler. → The compiler con
tains an error. Also, there was an error in the converter function → The converter contai
ns an error.

As described above, the user can predefine the tuning definition 12 (the pattern match definition 13 and the action definition 14 that defines the process when the pattern match definition 13 is matched), so that It is possible to generate a translated sentence of.

Next, the text processing other than machine translation will be described in the order shown in the flowchart of FIG.
In FIG. 5A, S40 performs matching processing.

In step S41, it is determined whether a certain rule matches. These S40 and S41 are the same as S30 and S31 of FIG. 3, and as described on the right side, as the vocabulary information of the input sentence, there are compiler, noun, TOOL; ;
When pattern is generated, pattern match definition 1 defined in advance
3 *; to; err; to; As the action definition 14 of this pattern match definition 13, S43, S44,
When the definition of S45 is described, the action process described on the right side is executed.

At S43, CASE: translation is performed. This translates the entire original text ($ 0) corresponding to the action definition of (F: TRAN $ 0) (). In S44, the CASE: sentence is rewritten. this is,
Rewrite the original text, "$ 1; contains an error."

In step S45, the processing is performed by CASE and the user tool. This is (F: / root / atlas / mytool $ b)
Corresponding to the action definition of
The output (the second one here is incorrect in the compiler process) is passed to mytool and processed ().

The above items and can be summarized as follows, which is described in the item of (b) of FIG. Action process output result → The mistake was found in The compiler. Action process output result → The compiler contains an error.

Action processing of (here, / root / atl
The output result of (As / mytool program is a machine translation program) → The compiler contains the mistake.

[0061]

As described above, according to the present invention,
Present the original vocabulary information using a machine translation engine,
Since the user sees this, the pattern matching definition 13 and the action definition 14 are modified and saved, and the structure is used for translation, so that the user defined pattern matching definition 13 and action definition 14 are followed. An appropriate translated text unique to the user can be automatically generated. As a result, the range of tuning by the user is widened, and it becomes possible to generate an appropriate translated sentence.

By effectively utilizing the machine translation engine, the machine translation can be utilized not only for machine translation but also for text processing.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a vocabulary information output process / tuning definition flowchart of the present invention.

FIG. 3 is an example dictionary / table of the present invention.

FIG. 4 is a flowchart for explaining the operation of the present invention (during translation).

FIG. 5 is an explanatory diagram of text processing other than machine translation of the present invention.

[Explanation of symbols]

 1: Text processing device with machine translation engine 2: Morphological analysis calling unit 3: Vocabulary information output processing unit 4: Tuning control unit 5: Matching processing unit 6: Process execution unit 7: Result output unit 8: Vocabulary information table 9: Dictionary 10: Narrowing table 11: Correspondence table 12: Tuning definition 13: Pattern match definition 14: Action definition 21: Machine translation engine 22: Morphological analysis 23: Syntax analysis 24: Translated sentence generation 25: Dictionary

Claims (3)

[Claims] 1. A text processing apparatus for performing text processing using a machine translation engine, comprising a machine translation engine (21) for morphological analysis, syntactic analysis, and translation generation, and calling the above morphological analysis for an input original sentence. A morphological analysis calling unit (2) for morphological analysis, a vocabulary information output processing unit (3) for extracting a part of speech and a representative attribute by referring to the vocabulary information table (8) for each morphologically analyzed word, and the input original text. Tuning definition created in advance based on the above words, extracted parts of speech, and grammatical attributes (12)
When the pattern matching definition (13) in the pattern is matched and matched, according to the action definition (14),
A text processing device comprising: a processing execution unit (6) for performing processing to generate a translated sentence. 2. A text processing device for performing text processing using a machine translation engine, a machine translation engine (21) for morphological analysis, syntactic analysis, and translation generation, and calling the above morphological analysis for an input original sentence. A morphological analysis calling unit (2) for morphological analysis and a narrowing-down table (10) based on the encrypted grammatical attributes retrieved by searching the dictionary (9) for each morphologically analyzed word. The encrypted representative attribute is extracted, and the vocabulary information output processing unit (3) for extracting the part of speech and the representative attribute with reference to the corresponding code table (11) for the extracted encrypted representative attribute is input. Tuning definition created in advance based on the above words in the original sentence, extracted parts of speech, and grammatical attributes (12)
When the pattern matching definition (13) in the pattern is matched and matched, according to the action definition (14),
A text processing device comprising: a processing execution unit (6) for performing processing to generate a translated sentence. 3. A pattern match definition (13) in which the word notation is removed or a regular expression is expressed with respect to the above-mentioned word of the input original sentence and the extracted part-of-speech and grammatical attribute,
The text processing device according to claim 1 or 2, wherein an action definition (14) defining a process for a pattern matching the pattern matching definition (13) is created and stored in advance. .