JPH01189768A - Sentence analyzer - Google Patents

Abstract

PURPOSE:To prevent a translation from being not executed according to the discordance of numbers generated from the careless mistake of the producer of an original text or the mistake of a key inputting operator or the like by disposing a cancelling means for cancelling the decision of the coincidence of a numerical representation in an analyzing means. CONSTITUTION:Original text information from an original sentence input part 100 is inputted to a syntax analyzing part 105 through a storing part 101 and a morphological element analyzing part 102 and separated for a word unit to analyze. In this case, when the discordance of the number is detected in a sentence, it is stored in a discordance information holding part 107. Then, a syntax analyzing result is checked by an analysis evaluating part 108 and it is not fed, and then, the holding part 107 is checked and when there is the discordance information, a mechanism for deciding the coincidence of the number is cancelled to analyze the syntax again. Thereby, a syntactic structure is formed without excluding as a grammatical error, fed as the syntax analyzing result and a meaning is extracted in a meaning analyzing part 109 and outputted through an output part 114. Consequently, according to the discordance of the number generated from the mistake or the like of the producer of the original text or the key inputting operator, no execution of the translation is prevented.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention extracts individual words from a human-generated original text, recognizes the grammatical connections between each word contained in the original text, and generates syntax. The present invention relates to a sentence analysis device that analyzes a structure and further extracts a semantic structure from this syntactic structure.

[Background Art] Research on natural language processing technology has become active in recent years, and machine translation systems are one of the applied technologies using this technology. In translation using this machine translation system, the human-powered original text is first morphologically analyzed and divided into predetermined processing units such as words (phrases), and then syntactic analysis is performed to recognize the grammatical connections between these processing units and create a syntactic structure. Creating. Furthermore, this syntactic structure is analyzed to extract semantic information, a translation dictionary is searched to find corresponding translation words (translation word direction), and the translated text is obtained by combining these according to predetermined translation generation rules.

In particular, in the syntactic analysis unit that recognizes the grammatical connections between the processing units mentioned above, the number match between the subject and the predicate and the number match between the modifier (adjective) and the modified word (noun) are strict When input is an original text written by the morphological analyzer, it creates a syntactic structure while checking the grammatical connections between processing units based on the information about the number of each word obtained by the morphological analysis section. There is.

[Problem to be solved by the invention] However, in such a system, it is difficult to distinguish between the original text of a language in which there is strict numerical agreement between a subject and a predicate, and between a modifier (adjective) and a modified word (noun). Even when obtaining a translated text in a less precise language, if there are discrepancies in the original text due to careless mistakes on the part of the original text creator or mistakes made by key personnel, it may be necessary to The problem was that the translation was not processed at all because it was not accepted.

The present invention was made in view of the above-mentioned conventional example, and it is possible to solve problems when the parsing process of an input sentence fails, and when the number of subjects and predicates matches or the number of modifiers (adjectives) and modified words (nouns) When translating into a language where number matching is not strict, cancel the number matching check, go to mode before the parsing step, and perform the processing after parsing again to avoid careless errors by the original author. It is an object of the present invention to provide a sentence analysis device that prevents translation from not being executed due to mismatches in numbers caused by mistakes or mistakes by key personnel.

[Means for Solving the Problems] In order to achieve the above object, the sentence analysis device of the present invention has the following configuration. That is, an input means for inputting the original text information, a means for decomposing the input original text information into each word, and a means for decomposing the input original text information into each word, and a means for inputting the original text information, and a means for decomposing the input original text information into words, an analysis means for determining whether or not the original text information exists and performing syntactic analysis of the original text information, a semantic analysis means for extracting the meaning of the parsed original text information, and a canceling means for canceling the match determination of the numerical expression in the analysis means. Equipped with.

[Operation] In the above configuration, original text information is input from the input means,
Breaks down the input source text information into each word. Then, syntactic analysis is performed based on each word in the original text information. In numeric matching determination, which is one process during this syntax analysis process, if a numeric mismatch is detected in the input sentence, information to that effect is held. At this time, when the source text information is translated into a language that does not have the concept of matching number expressions, and syntactic analysis fails due to a mismatch in numbers, the syntactic analysis process is performed again without performing the match determination of number expressions. It works like this.

[Embodiments] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[Explanation of translation device (Figure 1)] FIG. 1 is a functional block diagram of the translation device of this embodiment.

In the figure, reference numeral 100 denotes a human resource section such as a keyboard for inputting original text, and character codes input from the human resource section 100 are stored in a storage section 101. Reference numeral 102 denotes a morphological analysis unit that analyzes the input original text in the storage unit 101 and divides it into word units by referring to the dictionary 103. Each word divided into word units is stored in the analysis result holding unit 104. Reference numeral 105 denotes a syntactic analysis unit that checks the connections between each word with reference to the parsing grammar dictionary file 106 and analyzes the syntax of the sentence. Modifier (
It also detects discrepancies in numbers (such as when an adjective immediately follows a plural noun but the noun being described is singular, or vice versa), and if there is discrepant information,
The information is stored in the mismatch information holding unit 107.

Reference numeral 108 denotes an analysis evaluation unit that checks whether the syntax analysis unit 105 has obtained a syntactic analysis result. That is, if some syntax analysis result is sent from the syntax analysis unit 105, it is passed to the semantic analysis unit 109 to continue the subsequent processing. On the other hand, if no result is sent from the parsing section 105, it is determined that the parsing process has failed, and the mismatch information holding section 107 is checked. At this time, if the mismatch information holding unit 107 holds information indicating that a “mismatch” portion has been detected, the syntax analysis unit 1
05 is instructed not to perform a match determination of numerical expressions, and the syntax analysis of the original text is executed again.

109 is a semantic analysis unit that extracts a semantic structure based on the syntactic structure created by parsing. Reference numeral 110 refers to a conversion unit that converts the semantic information extracted from the original text into the semantic structure of the translated text by referring to the conversion dictionary file 111, and 112 refers to the generating grammar file 113 from the semantic structure of the translated text to convert the translated text into a translated text. The generation unit 114 is an output unit that prints or displays the translation result and, if necessary, the original text.

[Description of Translation Device (FIG. 2)] FIG. 2 is a block diagram showing a schematic configuration of the translation device of the embodiment.

In the figure, l is an input section such as a keyboard for inputting the original text, and document information (code data) input from input section 1.
is stored in the RAM of the control unit 2. 2 is a control unit that controls the entire device and performs translation processing such as syntactic analysis and semantic analysis of the input original text, and includes, for example, a CPU such as a microprocessor, or a control program for the CPU as shown in the flowchart in Figure 3. It is equipped with a ROM that stores various data such as text and character pattern data, and a RAM that is used as a work area for the CPU and temporarily stores document information and various data.

Reference numeral 3 denotes a dictionary/grammar file used for semantic conversion in the morphological analysis and syntactic analysis section, which collectively refers to each dictionary, grammar, etc. shown in FIG. 4 is a CRT for displaying original text information, translated text, etc., and 5 is a printer for printing out the original text, translated text, etc.

Below, as an example, we will consider the case of a machine translation system that translates from English to Japanese.

In the input English sentence, there is strict numerical agreement between the subject and the predicate, and between the modifier (adjective) and the modified word (noun), but in the output sentence Japanese In a sentence, the numerical agreement between a subject and a predicate and between a modifier (adjective) and a modified word (noun) is not strict.

Now, as an example of a sentence input into the machine translation system,
The recent rapid progress
in medical promise well f
or the future, ”. This input sentence is input from the input unit 1 to the control unit 2. This sentence is first processed into predetermined processing units such as words (phrases) by the morphological analysis unit 102 and the syntactic analysis unit 105. At the same time, an attempt is made to recognize the grammatical connections between these processing units and create a syntactic structure.

However, in the case of the example sentence above, the subject is third person singular (present tense)
Therefore, 'promise' is changed to 'promise'.
s", but due to a mistake when creating the original text or a mistake when manually inputting the key to the machine translation system, the "promis"
e”, there is a numerical discrepancy between the subject and predicate.

With conventional English-Japanese translation technology, there is a numerical mismatch between the subject and the predicate, which is rejected as violating English grammar.
The process ended without performing the translation process.

On the other hand, the translation process from English to Japanese in the translation apparatus of this embodiment will be explained with reference to the flowchart of FIG.

In step S1, original text information is inputted from the input section 1 and stored in the RAM storage section of the control section 2. In step S2, morphological analysis is performed to divide the input sentence into words. In step S3, the syntax is analyzed based on the grammar. At this time,
If a numerical discrepancy is detected in the sentence being parsed,
Information indicating that there is a mismatch is stored in the mismatch information holding unit 107. In step S4, it is checked whether the syntax analysis result has been sent from step S3. For example, in the example sentence mentioned above, even though the subject "progress" is in the third person singular (present tense), the corresponding predicate verb "promises'" is used instead of "promise".
ise'', there is no grammatical agreement between the subject and the predicate. Therefore, no syntactic analysis results are sent from step S3.

In this case, the process advances to step S5, and the mismatch information holding unit 10
Check 7 to see if there is a discrepancy. If there is non-coincidence information, the process advances to step S6, where the numerical coincidence determination mechanism is canceled, and the process returns to step S3 to perform syntax analysis.

Then, in the case of this embodiment, the above-mentioned ' promi
se” is not rejected as a grammatical error, a syntactic structure is created in step S3, and sent as a syntactic analysis result to step S4.Then, the following step S7
It is semantically analyzed and the meaning is extracted.

In step S8, when the analysis of the entire sentence is completed and the translation is completed, the translation result is output to the CRT 4, printer 5, etc., and the process ends.

By doing this, in the Japanese sentence obtained as a translated sentence, there is no concept of agreement in the number of subjects and predicates, so in the example sentence above, ``...promises well...
・°° is ゛...promise we
Even if it is ll...°°, the translation of that part is "to inspire hope," and there is no problem as a whole translation (Japanese sentence).

Furthermore, this embodiment is not limited to the above-mentioned embodiment, and is applicable to other applications besides the English-Japanese machine translation system. It may also be applied to a translation system that obtains a translated text in a language where numerical matching is not strict.

In addition, in this example, the case where the subject is third person singular (present tense) and the verb of the predicate is not appended with "S"' is explained, but in addition to this, for example, when the subject is plural, the predicate is a singular verb or a verb
It can also be applied to cases where S゛° is given.

In addition, as an application system other than a machine translation system, for example, in intellectual information retrieval, we use languages that require strict numerical agreement between a subject and a predicate, or between a modifier (adjective) and a modified word (noun). It may also be applied to input checks of input sentences and utterances, etc., to allow for errors when inputting information.

As explained above, according to this embodiment, the translation system,
As input for so-called natural language such as intelligent information retrieval systems, sentences etc. are used in languages where the number of subjects and predicates matches strictly, and the numbers of modifiers (adjectives) and modified words (nouns) match strictly. Even if a mismatch in numbers occurs due to a mistake or the like, processing can be continued, which has the effect of resolving the problem of sentences not being translated due to input errors or the like.

[Effects of the Invention] As explained above, according to the present invention, translation is executed even if a grammatical error occurs in the original text due to a key operation when manually writing the original text or an error when creating the original text. This has the effect of increasing the tolerance for input errors and making operations easier for the operator.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram of the translation device of this embodiment, FIG. 2 is a block diagram showing a schematic configuration of the translation device of this embodiment, and FIG. 3 is a flowchart showing the translation process of original text information of the embodiment. . In the figure, 1... input section, 2... control section, 3... dictionary/grammar, 4... CRT (display section), 5... printer, 100... original human resources department, 101 ...Memory part, 1
02...Morphological analysis unit, 103...Dictionary, 104.
...Analysis result holding unit, 105...Syntax analysis unit, 108
...Analysis and evaluation section, 109...Semantic analysis section, 110.
. . . conversion section, 112 . . . generation section, 114 . . . output section.

Claims (1)

[Claims] An input means for inputting original text information, a means for decomposing the input original text information into each word, and whether or not the singular or plural numerical expressions of corresponding words in the original text information match. and a semantic analysis means for extracting the meaning of the syntactically analyzed original text information, and a canceling means for canceling the matching judgment of the numerical expression in the analysis means. ,
A sentence analysis device characterized in that when the original text information is translated into a language that does not have a concept of number matching, the matching judgment of the number expressions is not performed.