JP3197110B2 - Natural language analyzer and machine translator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language analyzer used for a device that requires analysis of a natural language, such as a sentence comprehension device, and a machine translator equipped with the natural language analyzer.

[0002]

2. Description of the Related Art Conventionally, in a device such as a machine translation device which needs to analyze a natural language, the following method is used to deal with the ambiguity of the dependency of a natural language.

[0003] 1. Performs semantic analysis using semantic information on multiple syntactic structure candidates obtained as a result of syntactic analysis, and selects the most probable syntactic structure candidate from among the multiple syntactic structure candidates based on the semantic analysis result. To be the syntax structure of the input sentence. 2. A plurality of syntax structure candidates obtained as a result of the syntax analysis are displayed in order of likelihood using the next candidate display function of the apparatus, and the operator manually selects one syntax structure candidate. 3. The dependency relationship interpreted by the device is displayed, and a syntax analysis is performed using the dependency relationship changed / selected by the operator to obtain a syntax structure of the input sentence. 4. The operator is instructed in advance by using the pre-processing function of the apparatus, and understands the instruction and performs syntax analysis to obtain the syntax structure of the input sentence.

[0004]

However, the conventional method for dealing with the ambiguity of dependency has the following problems. In other words, in the case of a coping method that selects the most probable syntactic structure based on the result of semantic analysis, ambiguity remains in the dependency unless knowledge such as general common sense or semantic analysis using context is used. The method of removing dependency ambiguity by semantic analysis using context and context has not yet reached the practical stage.

Further, a coping method for displaying a syntax structure candidate using the next candidate display function, a coping method for displaying a dependency relationship interpreted by the apparatus, and a coping method for instructing a pre-modification using a preprocessing function In the case of (1), the work is performed interactively with the apparatus, so that the efficiency is extremely reduced in the case of a large amount of processing.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a natural language analysis device capable of performing correct dependency analysis on an input sentence in a natural language and capable of processing a large number of input sentences.
It is an object of the present invention to provide a machine translation device equipped with the natural language analysis device.

[0007]

According to an aspect of the present invention, a morphological analysis of an input sentence in a natural language input from an input unit is performed by a morphological analysis unit, and the morphological analysis result is obtained. In a natural language analysis device that analyzes the syntax of an input sentence based on a syntax analysis unit based on a morphological analysis performed by the morphological analysis unit, it is determined that a sentence that is added to a dependency relationship in the input sentence has a dependency relationship. A dependency symbol detecting unit for detecting the dependency symbol that is being represented, and a dependency symbol detected by the dependency symbol detecting unit being stored as an attribute of the character of the input sentence. A dependency storage unit for storing the relationship; and a dependency storage unit stored in the dependency storage unit among the syntax analysis results of the input sentence obtained by the syntax analysis unit. And a dependency processing unit that outputs a syntax analysis result satisfying the relationship as a syntax structure of the input sentence, and the morphological analysis unit is stored in the dependency symbol storage unit when the input sentence is divided into a plurality of clauses. It is characterized by having a dividing means which preferentially sets a portion indicated by a dependency symbol as a dividing position.

According to a second aspect of the present invention, in the natural language analysis device according to the first aspect of the present invention, the dependency symbol detection unit is added to each of the two phrases having a dependency relationship, and The present invention is characterized in that a detection means is provided for detecting a specific dependency symbol indicating a range and a dependency relationship.

According to a third aspect of the present invention, in the natural language analysis device according to the first or second aspect, the dependency symbol detecting section has a dependency relationship.
It is characterized by comprising a detecting means for detecting a specific dependency symbol which is added immediately after each of the two phrases and indicates a dependency relationship.

According to a fourth aspect of the present invention, an internal structure of an input sentence in a natural language is converted into an internal structure in a target language by a conversion unit, and is translated by a generation unit based on the obtained internal structure in the target language. A machine translation device for generating a sentence, comprising the natural language analysis device according to any one of claims 1 to 3, wherein the conversion unit and the generation unit are a dependency processing unit in the natural language analysis device. And generating a translated sentence of the input sentence based on the syntax structure of the input sentence in natural language output from the input sentence.

[0011]

According to the first aspect of the present invention, prior to the morphological analysis of the input sentence in the natural language input from the input unit by the morphological analysis unit, the dependency symbol detecting unit detects the phrases having a dependency relationship in the input sentence. Is detected. The dependency symbol thus detected is stored by the dependency storage unit as the attribute of the character of the input sentence, so that the dependency relationship of the input sentence is stored. When the input sentence is divided into a plurality of clauses by the dividing means at the time of the morphological analysis by the morphological analysis unit, the portion indicated by the dependency symbol stored in the dependency storage unit is preferentially divided into the division positions. Is divided as

The syntax of the input sentence is analyzed by the syntax analysis unit based on the result of the morphological analysis by the morphological analysis unit based on the phrase thus divided. Then, the dependency processing unit outputs a syntax analysis result satisfying the dependency relationship stored in the dependency storage unit among the syntax analysis results from the syntax analysis unit as the syntax structure of the input sentence.

In this way, the input sentence is divided into phrases reflecting the portion indicated by the dependency symbol by the operator in advance, and a syntax structure that satisfies the dependency relationship designated by the operator is obtained.

Further, in the invention according to claim 2, prior to performing the morphological analysis, the detecting means of the dependency symbol detecting unit adds each of the two clauses having a dependency relationship to each other in the input sentence to form the phrase. A particular dependency symbol is detected that indicates a range and dependency relationship. In this way, a syntax structure that best reflects the dependency relationship intended by the operator can be obtained according to the clauses in the dependency relationship in the input sentence that is properly instructed by the operator.

Further, in the invention according to claim 3, before the morphological analysis, the detecting means of the dependency symbol detecting unit adds the two clauses in the input sentence immediately after each of the two clauses having a dependency relationship. A particular dependency symbol indicating a dependency relationship is detected. In this way, a syntax structure that best reflects the dependency relationship intended by the operator is obtained according to the end position of the dependency relationship portion in the input sentence specified by the operator.

Further, in the invention according to claim 4, when an input sentence in a natural language is input to the input unit of the natural language analysis device, the dependency processing unit outputs a syntax structure of the input sentence. Then, based on the syntax structure of the output input sentence, the conversion unit converts the internal structure of the input sentence into an internal structure in the target language. Furthermore, a translation is generated by the generation unit based on the obtained internal structure in the target language. Thus,
The translation process is performed by understanding the dependency relationship in the input sentence specified by the operator.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments. In this embodiment, a case where the natural language analysis device of the present invention is mounted on a machine translation device will be described as an example. FIG. 1 is a diagram showing an example of the configuration of a machine translation device equipped with the natural language analysis device of the present embodiment.

In FIG. 1, 1 is a CPU (Central Processing Unit), 2 is a main memory, and 3 is a CRT (Cathode Ray Device).
Tube), 4 for keyboard, 5 for translation module, 6
Is an external storage unit storing a dictionary for translation, grammar rules, tree structure conversion rules, etc., and 7 is a data bus. The translation module 5 translates a given source language into a target language as described in detail later. The keyboard 4 is provided with a start key for driving the machine translation device, a numeric keypad, a character key, and other keys for executing translation.

Under the control of the CPU 1, the input sentence in the source language input from the keyboard 4 is sent to the translation module 5 and uses the dictionary, grammar rules, tree structure conversion rules and the like stored in the external storage unit 6. To the target language. The translation thus obtained is stored in the main memory 2
And displayed on the CRT 3.

FIG. 2 is a block diagram showing an example of the configuration of the translation module 5. The input sentence in the source language input as described above is divided into morphemes in the source language by morphological analysis processing by the lexical analysis unit 8 to obtain information such as part of speech. The syntax analysis unit 9 analyzes the syntax of the input sentence based on the morphological analysis result from the dictionary lookup morphological analysis unit 8 using the grammar rules stored in the external storage unit 6 to obtain a syntax analysis result. Then, the dependency processing unit 10 performs a dependency process, which will be described in detail later, on the syntax analysis result obtained by the syntax analysis unit 9 to specify the syntax structure of the input sentence. The conversion unit 11 uses the tree structure conversion rules stored in the external storage unit 6 to convert the syntax structure of the source language into the syntax structure of the target language. Then, the generation unit 12 generates a sentence of the target language from the syntax structure of the target language obtained as described above.

That is, in the present embodiment, the input unit is constituted by the keyboard 4, and the storage unit is replaced by the external storage unit 6.
It consists of. Then, the input unit, storage unit, dictionary lookup morphological analysis unit 8, syntax analysis unit 9, dependency processing unit 10, and C
PU1 constitutes the natural language analysis device.

By the way, in the Japanese sentence "I saw a student running with binoculars.""I saw a student running on the playground." The syntax analysis unit 9 cannot determine whether the word belongs to “is” or “seen”, and the syntax has ambiguity. Therefore, in the present embodiment, the symbols “*” and “* #” are used as escape characters, and a dependency symbol represented by the escape characters is set.
Then, as exemplified below, a dependency symbol “*” is added before each of two clauses having a dependency relationship in the input sentence, while a dependency symbol “*” is added after the two clauses.
The number added immediately after the dependency symbol "*" indicates that the phrases are in a dependency relationship, and the phrases with the same number are in a dependency relationship. is there.

Therefore, when a dependency symbol is added as follows, (a) "I am a student running * 1 with * 1 binoculars .
I saw 1 * # . (B) "I saw the ** student running ** on the * 1 playground ." In the example sentence (a), the phrase between the dependency symbols "* 1" and "* #" The phrase “with binoculars” and the phrase “saw” are in a dependency relationship, meaning that the phrase “with binoculars” relates to the phrase “saw”.
Similarly, in the example sentence (b), the dependency symbols “* 1” and “* #”
The phrase "at the playground" and the phrase "running" are in a dependency relationship, and the phrase "at the playground" is phrase "running".
Means that

[0024] In addition, the sentence (c) (c) "* 1 yesterday * # * 2 in the playground * # * 2 not running
* # I saw the student * 1 * # . In the case of "", the phrase "Yesterday" and the phrase "Mita" sandwiched between the dependency symbols "* 1" and "* #" have a dependency relationship, and the dependency symbol "* 2" The phrase “at the playground” sandwiched between “* #” and the phrase “running” have a dependency relationship. The phrase “yesterday” relates to the phrase “saw”, and the phrase “at the playground” relates to the phrase “running”.

The dependency symbol is manually inserted into the translation target sentence before the morphological analysis is performed by the dictionary lookup morphological analyzer 8 in the natural language analyzer. Therefore, when adding a dependency symbol to the translation target sentence,
The operator needs to know the definition of “phrase” in the natural language analysis system.

There are various theories on how to classify parts of speech and how to define phrases. Further, in each device that requires natural language analysis processing, the method of setting parts of speech and phrases is defined individually and conveniently. I have. Therefore, when instructing the dependency relationship using the dependency symbol as described above, the operator must understand the definition of the part of speech or phrase in the natural language analysis system, and the dependency symbol must be located at an appropriate position. Cannot be added.

For example, consider the following example sentence (d). (D) "I saw a student running with binoculars." At that time, in the case of a natural language analysis system that treats "using" like a particle, "using binoculars" Treat as one phrase, as in the example sentence (a), using the phrase "using binoculars"
Is related to the phrase "I saw" and analyzes the dependency relationship. On the other hand, there is also a natural language analysis system that treats "using binoculars" as two phrases, the phrase "using binoculars" and the phrase "using".

However, the operator does not always understand the definition of a clause in the natural language analysis system used. Therefore, in the present embodiment, it is possible to add the following dependency symbol so that such an operator can easily designate a portion having a dependency relationship in the input sentence.

That is, in the case of the above example (d), (e) "I saw a student running with * 1 using binoculars .
* 1 . ", The end position of a part having a dependency relationship is indicated by a dependency symbol" * ". As in the above-described case, a number is added immediately after the dependency symbol "*", and the same number indicates the end positions of the phrases in the dependency relationship. By instructing the dependency relation in this manner, the operator does not understand the definition of the clause in the natural language analysis system to be used, and can immediately determine the dependency symbol (the preceding dependency symbol “* 1”). The phrase ending with "te" (the range of this phrase does not matter in particular) is the "ta" immediately before the receiving symbol (the dependency symbol "* 1" located behind).
Can be indicated for a clause ending with (the range of this clause does not matter). Hereinafter, the process of adding a dependency symbol that specifies a phrase range, such as the former, is called “normal addition processing”, and the process of adding a dependency symbol, which specifies only the end position of a phrase, such as the latter, is called “simple addition”. Processing ".

Note that the processing of adding the dependency symbols "*" and "* #" to the input sentence in the present embodiment is not dependent on the machine translation apparatus, but may be performed, for example, on the dependency symbol "*" of a document processing apparatus or the like. "And" * # "can also be implemented by a device capable of inputting. Further, the dependency symbol addition processing in the present embodiment can be performed non-interactively with the machine translation apparatus. Therefore, if the dependency symbol adding process is performed by the machine translation device or the word processor before the translation process by the machine translation device, the translation process can be performed efficiently.

Hereinafter, the processing operation of the machine translation apparatus using the natural language analyzer of the present invention will be described in detail, taking as an example the case where Japanese-English translation is performed by the machine translation apparatus having the above configuration. FIG. 3 is a flowchart of a Japanese-English translation process performed by the machine translation device under the control of the CPU 1. Hereinafter, the Japanese-English translation processing operation will be described with reference to FIG.

The sentence to be translated in Japanese to which the dependency symbol has been added is input from the keyboard 4 or the like, is temporarily stored in the main memory 2, and the Japanese-English translation processing operation starts. In step S1, one sentence is read from the Japanese sentence stored in the main memory 2 and stored in an original sentence buffer (not shown) in the translation module 5 as shown in FIG. However, FIG. 4A shows an example of an input sentence to which a pair of dependency symbols “*”-“* #” indicating the start position and the end position of a phrase has been added by the normal addition processing. On the other hand, FIG. 6B shows an example of an input sentence to which a dependency symbol “*” indicating only the end position of a phrase has been added by the simple addition process. In step S2, it is determined whether or not the dependency symbol exists in one sentence stored in the original sentence buffer. As a result, if it exists, the process proceeds to step S3, and if not, the process proceeds to step S12.

At step S3, the type of the dependency symbol present in the one sentence is a pair of dependency symbols "*"-"* #" indicating the start position and the end position of the phrase, or the end of the phrase. It is determined whether it is a dependency symbol “*” indicating only the position. Then, a flag (not shown) indicating the determination result is set. Here, the type of the dependency symbol is determined by:
For example, it is performed as follows. That is, a first mode in which the normal addition processing is performed and a second mode in which the simple addition processing is performed are set, and the first mode or the second mode is previously set at the time of the dependency symbol addition processing. A code to represent is attached to the input sentence. In this step, the type of the dependency symbol is determined by determining the mode based on the code. Alternatively, the type of the dependency symbol may be determined based on the presence or absence of the dependency symbol “* #” indicating the end position of the phrase in the input sentence.

In step S4, the dependency symbol mixed in the input sentence is extracted from the input sentence, and as shown in FIG. 5, the attribute of the character is added to the position of the character to which the dependency symbol is added in the buffer. Is set as a part of the dependency symbol. At this time, in the case of an input sentence to which a pair of dependency symbols “*”-“* #” is added according to the contents of the flag set in step S3, as shown in FIG. Set. On the other hand, in the case of an input sentence to which a dependency symbol “*” indicating only the end position is added, the input sentence is set as shown in FIG.

In step S5, a morphological analysis process is performed on the input sentence as follows. That is, first, as a pre-process, as shown in FIG. 5, the character string of the input sentence stored in the buffer is delimited at the position indicated by the dependency symbol. At this time, in the case of an input sentence to which a pair of dependency symbols "*"-"* #" is added as shown in FIG. The position is separated from the phrase end position indicated by the symbol "#". The results are shown in FIG. 6 [1] (a). In the following processing, since the information on the phrase end position is not required, the symbol "#" indicating the end position is deleted.
On the other hand, in the case of an input sentence to which a dependency symbol "*" indicating only the end position of a phrase as shown in FIG. 5B is added, the phrase end position indicated by the dependency symbol "number" is used. punctuate. The result is shown in FIG. 6 [1] (b).

When such preprocessing is completed, the dictionary lookup morphological analysis unit 8 performs a morphological analysis process by referring to the dictionary stored in the external storage unit 6. At this time, when dividing into morphemes, the location of the break obtained in the preprocessing is reflected as follows. That is, a conventional morphological analysis is performed on the input sentence by the dictionary lookup morphological analysis unit 8 to obtain a phrase candidate. Thereafter, it is checked whether or not the content of the instruction by the operator is satisfied for all the obtained phrase candidates.

The above check is performed as follows. FIG.
As in the example sentence shown in (a), the set of dependency symbols “*”-“*
In an input sentence where "#" is added to indicate both the start position and end position of a phrase, the character strings "Yesterday", "At the playground", "Running", and " , Those that can become phrases according to the definition of the natural language analysis system are preferentially adopted as phrases. In addition, as in the example sentence shown in FIG. 4B, in an input sentence in which only the end position of a phrase is indicated by adding a dependency symbol “*”, the designated positions “day” and “ , And between "de" and "run", and between "ru" and "raw", which are defined by the natural language analysis system, are used as the clauses. I do. The clause of the input sentence thus obtained is shown in FIG. 6 [2].
As shown in FIG. 4 (a) or FIG. 4 (b), this phrase does not conflict with the phrase position indicated by the dependency symbol by the operator.

In step S6, it is determined whether or not there is a position in the range of the phrase or the end position of the dependency point designated by the operator that does not conform to the break position of the phrase obtained in step S5. You.
As a result, if it exists, the process proceeds to step S7; otherwise, step S7 is skipped. In step S7, since there is a conflicting phrase (that is, not corresponding to the dependency symbol) at the phrase break position specified by the operator, the fact is displayed on the CRT 3 and the processing is shifted to the normal processing, or the error is treated. Or a prompt to change the input sentence or the modification symbol addition location.

In step S8, the syntax analysis unit 9 executes a syntax analysis process using the grammar rules stored in the external storage unit 6. As a result, all of the plurality of obtained parsing results are output. FIG. 7 shows the structure of a parsing tree as a result of the parsing obtained above and the dependency structure by the parsing tree. Depending on the input sentence of this embodiment, three types of dependency structures can be obtained. However, the algorithm of the syntax analysis processing in that case is not particularly limited. Note that, in the syntax analysis tree shown in FIG. 7, among the clauses having a dependency relationship, the dependency clause is described one row below the receiving clause.

In step S9, the dependency processing unit 10
Thus, it is determined whether or not there is a modification structure that matches the modification structure specified by the operator among the modification structure candidates obtained in step S8, as shown in the lower part of FIG. 6 [2]. It is determined based on the receiving instruction information. As a result, if it does not exist, step S10
And if it exists, skip step S10. In step S10, since a dependency structure that does not contradict the dependency instruction information specified by the operator (that is, corresponds to the dependency symbol) cannot be obtained, the fact is displayed on the CRT 3 and the process proceeds to normal processing. , An error, or a prompt to change an input sentence or a portion where a dependency symbol is added. Step S11
Then, a parse tree having a dependency structure suitable for the dependency instruction information is sent to the conversion unit 11.

In step S12, since it is determined in step S2 that no dependency symbol has been added to the input sentence, the process proceeds to normal processing. Then, first, the morphological analysis processing is performed by the dictionary lookup morphological analysis unit 8 using the dictionary stored in the external storage unit 6. Step S13
Then, the parsing unit 9 executes a parsing process using the grammatical rules stored in the external storage unit 6 based on the result of the morphological analysis performed by the dictionary lookup morphological analyzing unit 8.
Then, the obtained parse tree is sent to the conversion unit 11.

In step S14, the parsing tree from the dependency processing unit 10 or the parsing tree from the parsing unit 9 is converted by the conversion unit 11 using the tree structure conversion rules stored in the external storage unit 6. Converted to the internal structure of the target language (English). Then, the generating unit 12 generates an English sentence based on the converted internal structure of the English language. In this case, the algorithm of the conversion process by the conversion unit 11 and the algorithm of the generation process by the generation unit 12 are not particularly limited.

In step S15, a translated sentence corresponding to the input sentence generated in step S14 is displayed on the CRT 3 and stored in a buffer (not shown) of the main memory 2. In step S16, it is determined whether or not the next sentence of the untranslated process exists in the main memory 2. As a result, if there is, the process returns to step S1 to read the next sentence of the untranslated process and start the translation process of the next sentence. On the other hand, if not, the Japanese-English translation processing operation is terminated.

As described above, in the present embodiment, the dependency processing unit 10 is provided in the natural language analysis device, and the dependency symbol "*" indicating the start position of the phrase and the phrase "*" indicating the phrase start position are added to the phrases in the dependency relationship in the input sentence. A dependency symbol “* #” indicating the end position is added. Alternatively, a dependency symbol “*” indicating the end position is added to the end of a portion having a dependency relationship in the input sentence. When the morphological analysis is performed by the dictionary lookup morphological analysis unit 8, the dependency symbols “*”, “*
The input sentence is divided into phrases by reflecting the position where "#" is added. Therefore, the input sentence can be divided into phrases by reflecting the dependency phrase and the receiving phrase desired by the operator. The parsing tree that satisfies the dependency relation desired by the operator is output as the syntax structure of the input sentence from the parsing tree candidates obtained as a result of the parsing processing by the parsing unit 9. Therefore, the parsing of the input sentence In this case, the dependency analysis can be correctly performed in accordance with the instruction of the operator.

The addition of the dependency symbols "*" and "* #" to the input sentence can be carried out non-interactively by the machine translation apparatus or by a separate document processing apparatus or the like. Therefore, it can be added to a large number of sentences without interfering with the translation processing operation. That is, according to the present embodiment, it is possible to execute machine translation processing of a large number of documents.

The algorithm of the Japanese-English translation processing operation in the above embodiment is not limited to the flowchart shown in FIG. In addition, the dependency symbol is a symbol “*”.
And "* #".

In the above embodiment, the natural language analysis device of the present invention is mounted on a machine translation device. That is, the parse tree specified by the dependency processing unit 10 is input to the conversion unit 11 to obtain an internal structure in English, and the generation unit 12 generates an English translation.
However, the present invention is not limited to this. In other words, if the parse tree from the dependency processing unit 10 is sent to another external device, it can be applied to a system such as a text comprehension system that gives instructions to a computer using a natural language.

[0048]

As is apparent from the above description, the natural language analysis apparatus according to the first aspect of the present invention uses the dependency symbol detection unit when the input sentence is divided into a plurality of phrases by the division unit of the morphological analysis unit. A portion indicated by the dependency symbol in the input sentence detected and stored in the dependency storage unit is preferentially set as a division position, and the dependency processing unit outputs the result of the syntax analysis of the input sentence from the syntax analysis unit. Since the syntax analysis result that satisfies the dependency relationship stored in the dependency storage unit is output as the syntax structure of the input sentence, a syntax structure that satisfies the dependency relationship specified in advance by the dependency symbol by the operator is output. Obtainable. Therefore, according to the present invention, correct dependency analysis can be performed on an input sentence in a natural language.

Further, the addition of the dependency symbol can be carried out non-interactively with the natural language analyzer.
According to the present invention, it is possible to analyze a large number of input sentences.

Further, the dependency symbol detecting section of the natural language analysis device according to claim 2 is added to each of the two clauses in the dependency relationship to indicate that they are in a dependency relationship with the range of the clause. Since the specific dependency symbol is detected by the detecting means, when the morphological analysis unit divides the input sentence into phrases, the phrase having the dependency relationship specified by the operator can be correctly reflected and divided. Therefore, according to the present invention, a dependency analysis can be more correctly performed on an input sentence in a natural language in accordance with an instruction from an operator.

In addition, the dependency symbol detecting section of the natural language analysis device according to claim 3 is a special dependency symbol which is added immediately after each of the two clauses in the dependency relationship to indicate that they are in a dependency relationship. Since the receiving symbol is detected by the detecting means, the end position is reflected on the basis of the end position of the dependency-related portion even if the operator does not indicate the range of the dependency-related phrase in the input sentence. Therefore, it can be correctly divided into phrases having a dependency relationship. Therefore, according to the present invention, even an operator who does not know the definition of a phrase automatically correctly associates a natural language input sentence with an input sentence as long as the operator indicates the end position of a dependency relationship in the input sentence. Receive analysis can be performed.

According to a fourth aspect of the present invention, there is provided a machine translation apparatus which converts a conversion unit based on a syntax structure that satisfies a relation of dependency specified by an operator output from a dependency processing unit in the natural language analysis device. Since the translated sentence of the input sentence is generated by the generating unit, a natural translated sentence can be generated by correctly understanding the dependency relationship specified by the operator. Since the instruction of the dependency relationship by the operator at that time can be implemented non-interactively with the machine translation apparatus using the dependency symbol,
According to the present invention, a large number of input sentences can be translated.

[Brief description of the drawings]

FIG. 1 is a block diagram of a machine translator equipped with a natural language analyzer of the present invention.

FIG. 2 is a block diagram of a translation module in FIG. 1;

FIG. 3 is a flowchart of a Japanese-English translation processing operation.

FIG. 4 is a diagram illustrating an example of an input sentence into which a dependency symbol is inserted.

FIG. 5 is a diagram illustrating a state in which a part of a dependency symbol extracted from an input sentence is stored in a buffer.

FIG. 6 is a diagram showing a state in which an input sentence is divided into phrases based on dependency symbols.

7 is a diagram illustrating a parse tree obtained as a result of a parse process performed by the parse unit in FIG. 2 and a dependency structure thereof.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... CPU, 2 ... Main memory, 3 ... CRT, 4 ... Keyboard, 5 ... Translation module, 6 ... External storage device, 8 ... Dictionary morphological analysis unit, 9 ...
Syntax analysis unit, 10 ... dependency processing unit, 1
1 ... Conversion unit, 12 ... Generation unit.

Continuation of front page (56) References JP-A-4-235673 (JP, A) JP-A-1-280868 (JP, A) JP-A-1-8479 (JP, A) JP-A-1-35661 (JP) , A) JP-A-63-317879 (JP, A) Akihiro Hirai, Hiroyuki Kaji, Minoru Ashizawa, "A Method for Detecting Ambiguity of Japanese Dependency Structure for Preediting for Machine Translation", IPSJ Theory Bibliography, Vol. 31, No. 10, p. 1425-p. 1437 (1990.10)

Claims (4)

(57) [Claims] 1. A natural language analysis device for performing a morphological analysis of an input sentence in a natural language input from an input unit by a morphological analysis unit and analyzing a syntax of the input sentence by a syntax analysis unit based on a result of the morphological analysis. Prior to the morphological analysis by the morphological analysis unit, a dependency symbol detection unit that detects a dependency symbol that is added to a phrase having a dependency relationship in the input sentence and indicates a dependency relationship, and the dependency symbol A dependency storage unit that stores a dependency relation of the input sentence by storing the dependency symbol detected by the detection unit as a character attribute of the input sentence; and the input obtained by the syntax analysis unit. Among the syntax analysis results of the sentence, a syntax analysis result that satisfies the dependency relationship stored in the dependency storage unit is output as the syntax structure of the input sentence. The morphological analysis unit, when dividing the input sentence into a plurality of clauses, preferentially sets a location indicated by a dependency symbol stored in the dependency storage unit as a division position. A natural language analyzer having a dividing unit. 2. The natural language analysis device according to claim 1, wherein the dependency symbol detection unit is added to each of the two phrases having a dependency relationship, and has a relationship between the range of the phrase and the dependency. A natural language analysis device comprising a detecting means for detecting a specific dependency symbol indicating the following. 3. The natural language analysis device according to claim 1, wherein the dependency symbol detection unit is added immediately after each of the two clauses in the dependency relationship to determine the dependency relationship. A natural language analysis processing device comprising a detection means for detecting a specific dependency symbol indicating that there is a certain dependency symbol. 4. A machine translation apparatus that converts an internal structure of an input sentence in a natural language into an internal structure in a target language by a conversion unit, and generates a translated sentence by a generation unit based on the obtained internal structure in the target language. The natural language analysis device according to any one of claims 1 to 3, wherein the conversion unit and the generation unit are configured to input the natural language output from the dependency processing unit of the natural language analysis device. A machine translator for generating a translation of the input sentence based on a syntax structure of the sentence.