JP2870259B2 - Japanese sentence analysis method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention analyzes input Japanese sentences such as a machine translation system, a Japanese text editing support system, and a man-machine interface in Japanese.
The present invention relates to a Japanese sentence analysis method for outputting a semantic expression such as an intermediate language.

[0002]

2. Description of the Related Art Current Japanese processing systems such as a machine translation system can analyze a single sentence to a considerable extent. Due to features that are not found in a single sentence, such as "the number of certain combinations becomes enormous", "frequent omissions, anatomical expressions, sharing, etc.", sufficient analysis cannot be performed.

In order to completely analyze these, knowledge such as context information and common sense is required. However, as can be seen from the fact that a human can also estimate the structure of a sentence containing many unknown words, it is considered that clues for performing structural analysis of a compound sentence or a compound sentence exist in the surface sentence in some form. Therefore, by focusing on the vocabulary appearing on the surface that has the function of defining the in-sentence context structure, the vocabulary is used as a method for estimating the structure of compound sentences and multiple sentences prior to parsing and semantic analysis. Contextual grammar (Shinichiro Kamei,
Kazushi Muraki: “Lexical Discourse G
Proposal of Rammar ", IEICE Technical Committee on Language Understanding and Communication, 456-7, pp. 1-5
(1986)) and a context estimating method using the same (Japanese Patent Application No. 61-75820) and a context analyzing method (Japanese Patent Application No. 61-2)
16510) has been proposed. FIG. 8 shows an embodiment of the context analysis system based on the vocabulary context grammar.

[0004] This context analysis method is characterized in that, as the context analysis means, information on vocabulary defining the in-sentence context structure is collected in advance, and the context format holding means for formalizing and holding the relationship between the expressions, Context structure matching means for detecting a candidate of a context structure in an input sentence prior to syntax analysis and semantic analysis by matching a dictionary content string obtained as a result of morphological analysis of a sentence with context information held in the context format holding means And a context structure estimating means for estimating the most probable context structure candidate from a plurality of context structure candidates in the input sentence obtained by the context structure matching means. In particular, focus on functional words such as connecting particles that show the relationship between elementary sentences in compound sentences and compound sentences, and hold information about the functional words such as "strength as a break in sentences" in the context format holding means. , Estimate the sentence structure. FIG. 9 is a table summarizing information obtained by dividing connection forms between words such as connection particles based on what elements can be included in a clause in a clause accompanying each connection form.

[0005]

According to the prior art described above, the dependency relationship between elementary sentences in a compound sentence or a compound sentence is analyzed prior to syntactic analysis and semantic analysis using functional words such as connecting particles as an information source. Estimation can improve the quality and efficiency of parsing and semantic analysis.

[0006] However, the relationship between elementary single sentences in compound sentences and compound sentences also shows that "elements common to a plurality of single sentences appear only in one place and are omitted in other parts" in addition to connecting particles, that is, shared. It is also presented in the form. Of particular importance among them is the sharing of the subject indicated by the particle "ha" and the nominative case indicated by the case particle "ga". In this prior art, no mechanism is provided for estimating the sharing of these subjects and nominatives.

[0007] The conditions for sharing the subject and the nominative in compound sentences and compound sentences are conventionally described in various forms, but all of them are regarded as problems of parsing, semantic analysis or discourse analysis. There has been no technique for estimating the sharing of the subject / nominative when the morphological analysis is completed.

[0008] The object of the present invention is to provide a subject
By collecting information on vocabulary that prescribes sharing of the nominative in advance, it is possible to accurately identify the subject and nominative in Japanese compound sentences and multiple sentences,
An object of the present invention is to provide a Japanese sentence analysis method that can improve the accuracy and speed of Japanese sentence analysis by reducing the ambiguity of syntactic analysis and semantic analysis by estimating the sharing of the subject and the nominative prior to the semantic analysis. .

[0009]

According to the present invention, there is provided a Japanese sentence analyzing method comprising: input reading means for reading an input of a Japanese sentence;
In a Japanese sentence analysis system comprising a morphological analysis means, a syntax analysis means, and a semantic analysis means for analyzing an input sentence using a dictionary lookup function for the input sentence and dictionary information after the dictionary lookup,
In Japanese, the subject indicated by the particle "ha" or the subject which collects and retains information about the vocabulary that prescribes the sharing of multiple nouns of the nominative case indicated by the case particle "ga" in advance. Nominal nominative vocabulary information holding means, and a subject / nominative share estimating means for estimating whether or not there is a subject / nominative sharing using the vocabulary information held in the subject / nominative shared vocabulary information holding means, Accurate recognition of the sharing of the subject and nominative in Japanese compound sentences and multiple sentences based on the estimation by the subject and nominative sharing estimating means, and by presuming the existence of the sharing of the subject and nominative prior to parsing and semantic analysis. It is characterized by improving the accuracy and speed of parsing Japanese sentences by reducing the ambiguity of syntactic analysis and semantic analysis.

[0010]

According to the Japanese sentence analysis method of the present invention, when an input of a Japanese sentence is read by an input reading means, the morphological analysis means analyzes the input sentence using a dictionary lookup function for the input sentence and dictionary information after the dictionary lookup. And outputs the morphological analysis result.

The subject / nominative shared vocabulary information holding means includes a vocabulary defining sharing between a plurality of words of the subject indicated by the particle "ha" or the nominative indicated by the case particle "ga" in Japanese. Information is collected and stored in advance.

The subject / nominative shared estimating means receives as input the morphological analysis result output from the morphological analyzing means, and determines whether the vocabulary held in the subject / nominative shared vocabulary information holding means exists in the morphological analysis result. A search is performed to estimate whether or not the subject / nominative sharing exists using information on the vocabulary, and the estimation result is added to the morphological analysis result and output.

The syntactic analysis means / semantic analysis means receives the output of the subject / nominative share estimating means as input, and uses the morphological analysis result and the estimated subject / nominative shared information to perform parsing / semantic analysis.
Performs semantic analysis and outputs the analysis result.

[0014]

Next, an embodiment of the present invention will be described with reference to FIGS.

FIG. 1 is a block diagram showing an embodiment of the present invention described in claim 1.

When the input of the Japanese sentence is read by the input reading means 1, the morphological analysis means 3 analyzes the input sentence using the dictionary lookup function for the input sentence and the dictionary information after the dictionary lookup while referring to the dictionary 2. And output the morphological analysis result.

The subject / nominal shared vocabulary information holding means 7 includes:
In Japanese, information on vocabulary that prescribes sharing between a plurality of verbs of the subject indicated by the particle "ha" or the nominative case indicated by the "ga" is stored in advance.

The subject / nominative shared estimating means 8 receives the morphological analysis result output from the morphological analyzing means 3 as input, and the vocabulary held in the subject / nominative shared vocabulary information holding means 7 is present in the morphological analysis result. Whether or not the subject / nominative sharing is present is estimated using the information on the vocabulary, and the estimation result is added to the morphological analysis result and output.

The parsing means 4 receives the output of the subject / nominative sharing estimating means 8 as input, performs a parsing using the morphological analysis result and the estimated subject / nominative shared information, and outputs the parsing result. .

The semantic analysis means 5 receives the output of the syntactic analysis means 4 as input and performs semantic analysis using the morphological analysis result / syntax analysis result and the shared information of the estimated subject / nominative.
An analysis result 6 such as an intermediate language or a semantic expression is output.

FIG. 2 is a block diagram showing an embodiment of the present invention described in claim 2.

The context analyzing means 9 comprising the context format holding means 10, the context structure collating means 11 and the context structure estimating means 12 is a prior art context analysis system (FIG. 61-216510), which is the same as the context analysis means 9 including the context format holding means 10, the context structure matching means 11, and the context structure estimation means 12.

The context format holding means 10 collects and describes in advance information on vocabulary defining the in-sentence context structure. For example, in terms of the conjunctive forms between verbs (continuous discontinuous forms, connective particles, connective nouns, etc.), in the clauses with each connective form, what elements can be included inside the clauses or the relative internodes Is information obtained by classifying each connection type into levels according to the inclusion relation. FIG. 9 shows an example in which levels are classified based on what elements can be included in a node in a node accompanying each connection type. By using the information on the level division, the context structure estimating means 12 can estimate the in-sentence context structure, that is, the dependency structure between words, prior to syntax analysis and semantic analysis.

In the embodiment of the present invention, in addition to the vocabulary information held in the subject / nominative shared vocabulary information holding means 7, information on the vocabulary described in the context format holding means 10, such as information on connection-type level division, is used. Using the context structure estimated by the context structure estimating means 12, the subject and nominative sharing estimating means 8
Presumes the existence of shared subject and nominative in. Information on the relationship between the context structure and the sharing of the subject / nominative used in the subject / nominative sharing estimating means 8 is stored in the subject / nominative shared vocabulary information holding means 7 in advance. As an example, FIG. 3 shows a part of the information indicating the relation between the connection form between verbs and the sharing condition of the subject and the nominative between the main sentence and the subordinate clause.

The operation of this embodiment will be described below using a specific example of sentence analysis.

1) He returned while talking.

2) He returned while talking.

The sentence of 1) or 2) is input reading means 1
Is read, the morphological analysis unit 3 performs morphological analysis. The conjunctive particle "whisper" has the meaning of parallel operation and reverse connection, and when connected to an operation predicate, it indicates parallel operation, and when connected to a state predicate, it indicates the reverse connection.
Here, "speak" is a behavioral predicate, so that "while" here is determined to represent a parallel action at this morphological analysis stage. Subsequently, the context analysis unit 9 performs a context analysis. First, based on the information shown in FIG. 9 held in the context format holding means 10, the context structure collating means 11 determines the level of the connecting particle “while” representing the parallel action as 5,
This information is added to the morphological analysis result of “while” by the context structure estimation means 12 and becomes the output of the context analysis means 9. Next, according to the vocabulary information shown in FIG. 3 held in the subject / nominative shared vocabulary information holding means 7, the clause at this level cannot have its own ga as well as its own ga, ie, the main sentence and the subordinate It turns out that nominatives are always shared. This information is added to the result of the context analysis by the subject / nominative share estimating means 8 and output. By using this information, in the syntactic analysis means 4 and the semantic analysis means 5, "he" or "he" is related to "talking" or "returned". It is easy to recognize that the nominative of "returned" is also "his". Further, by this estimation, the ambiguity of the syntax analysis and the semantic analysis can be reduced in advance, and the accuracy and speed of the analysis can be improved.

3) He said he was drunk.

When the sentence 3) is read by the input reading means 1, morphological analysis is performed by the morphological analyzing means 3, and the result is passed to the context analyzing means 9. First, based on the information shown in FIG. 9 held in the context format holding means 10,
The context structure matching unit 11 determines that the level of “to” representing the citation is 0, and this information is added to the result of the morphological analysis of “and” by the context structure estimation unit 12 and becomes the output of the context analysis unit 9. Next, from the vocabulary information shown in FIG. 3 held in the subject / nominative shared vocabulary information holding means 7, it can be seen that the nominative case of this level is independent of the nominative case of the main sentence. This information is added to the result of the context analysis by the subject / nominative share estimating means 8 and output. By using this information, in the syntactic analysis means 4 and the semantic analysis means 5, there are cases where "he" is related to "drunk" and cases where "he is" is said.

When "he" is related to "said", "who is drunk" is completely independent of "he".

That is, the sentence 3) has three interpretations and cannot be determined without other information. It can be certified. In fact, when the contexts of 4), 6) and 8) are set, the parts corresponding to 3) are 5), 7) and 9)
It means.

4) When I asked her what happened, he said he was drunk.

5) (She) said, "He is drunk."

6) When I asked him what happened, he said he was drunk.

7) He said, "I am drunk."

8) When I asked him what happened, she said he was drunk.

9) He said, "(She is) drunk."

FIG. 4 is a block diagram showing an embodiment of the present invention described in claim 3.

Here, the nominative personal restricted vocabulary information holding means 13
Holds information about a vocabulary having a nominative personal limitation, such as an emotion predicate, as shown in FIG. 5 by taking the emotional adjective "pain" as an example. First, this information will be described.

Sentences are always described according to some psychological viewpoint of the speaker. The emotion predicate is one of the expressions that are deeply related to this viewpoint together with the exchange expression, etc., and the emotion verbs such as “hate / fear / fright” and “hot / wanted / lonely”
(Takashi Masuoka, Yukinori Takubo: "Basic Japanese Grammar", Kuroshio Publishing (1989)). In particular, emotional adjectives "need a viewpoint to be placed on the subject" (Muratake Masatake: "Viewpoint and Semantic Structure", Lecture Modern Language 1 "Basic Structure of Japanese", Sanseido (1983)), It is known that there are personal restrictions on, for example, "The question"? "Can only be used in the second person."
(Yoshifumi Akira: “Comparative Japanese Studies”, Japan Translator Training Center (1979)). Based on this, FIG. 5 summarizes the restriction on the personality of the nominative of the emotional adjective, taking "pain" as an example. Here, the expression with a circle can be used, but the expression with a cross cannot be used.
* 1 is also used as a metaphor such as "He has a headache (for me)." * 2 It is possible to use "he is in pain".

Next, the subject / nominative shared vocabulary information holding means 7 holds information on the sharing of the subject / nominative case when this emotion predicate appears in the quoted section. In the above, in the embodiment of the second aspect, taking the sentence of 3) as an example, it has been shown that the nominative case of the quoted phrase is independent of the subject and nominative case of the main sentence. Verbs that take the "to" in Japanese include "say" (speak, speak, answer, etc.) and "think" (think, feel, doubt, etc.) (Hideo Teramura: "Japanese grammar ( (Below) ”, National Institute for Japanese Language (1981)), as a feature of Japanese,“ It appears in “... to” while the main word “U” is uttered ”(Hideo Teramura:“ Japanese Syntax and meaning I
I ", Kuroshio Publishing (1984)).
That is, even if "" (brackets) does not exist, it is directly spoken, and there is no phenomenon such as consistency of the tense between the main sentence and the quoted section appearing in English or the like. Therefore, the nominative personal restriction shown in FIG. 5 can be applied to the emotion predicate in the quote section as it is. By using this emotion predicate restriction,
The subject sharing between the main text and the quotation which should be independent can be estimated by the subject / nominative sharing estimating means 8. The estimation method here will be described using a specific example.

10) He said it was cold.

When this 10) is analyzed in the same manner as the sentence 3), the following three interpretations corresponding to 5), 7) and 9) are obtained.

11) "He is cold."

12) He said, "I am cold."

13) He said, "(Someone) is cold." However, when looking at FIG. 5, the subject of "cold" is limited to the first person, so that the utterance directly becomes non-sentence.
The interpretation of 1) and 13) is impossible, ie the correct interpretation is 12). This is the same in the case of 14).

14) He said it was cold. Therefore, "If a predicate whose nominative is limited to the first person appears in a quotation clause,
Nominatives are always shared between the main sentence and the quotation. "
This kind of information is stored in the subject / nominative shared vocabulary information holding means 7, and the subject and nominative shared estimating means 8 estimates the sharing of the subject between the main sentence and the quoted passage.

Hereinafter, the operation of this embodiment will be described in detail with reference to the above example.

When the sentence (10) is read by the input reading means 1, morphological analysis is performed by the morphological analyzing means 3. Next, the subject / nominative share estimating means 8 performs the subject / nominative share estimation. First, the information relating to the "cold" nominative limitation as shown in FIG. 5 held in the nominative personal limitation vocabulary information holding means 13 provides information that "the nominative of" cold "is limited to the first person". Next, the nominative case is shared between the main sentence and the quotation section according to the information on the sharing of the subject and nominative case when this emotion predicate held in the subject / nominative shared vocabulary information holding means 7 appears in the quotation section. It is estimated that This estimation result is added to the morphological analysis result and output, and the syntactic analysis means 4 and the semantic analysis means 5 correct among 11), 12) and 13) 12)
Only the interpretation of In addition, 1
By eliminating the possibility of the interpretation of 1) and 13), the accuracy and speed of analysis can be improved.

FIG. 6 is a block diagram showing an embodiment of the present invention described in claim 4.

Here, the supplementary nominative restricted vocabulary information holding means 14
Holds information relating to vocabulary for controlling the nominative of the complement, such as a word expressing a strong will of the speaker and a word requesting the act of another person as shown in FIG. Based on this information, the subject / nominative share estimating means 8 can estimate the presence of the subject / nominative sharing.

The operation of this embodiment will be described below using a specific example of sentence analysis.

15) The patients have decided to seek compensation from the State.

16) Patients called on the State to compensate.

When the sentence (15) is read by the input reading means 1, morphological analysis is performed by the morphological analyzing means 3. The vocabulary information shown in FIG. 7 held in the supplementary nominative restricted vocabulary information holding means 14 indicates that "in the case of" decide to do ", the nominative matches in the complement and the main sentence." The subject / nominative share estimating means 9 adds and outputs the result to the morphological analysis. By using this information, the syntactic analysis means 4 and the semantic analysis means 5 can determine that the nominative of "seek" and the nominative of "determined" are also "patients".

On the other hand, in the case 16), if the vocabulary information shown in FIG. It is expressed as two cases. " By using this information, in the syntactic analysis means 4 and the semantic analysis means 5, the subject of “determined” is “compensate” with “patients”.
Can be recognized as "country".

The context format holding means 10, the context structure matching means 11, and the context structure estimating means 12 according to the second aspect.
The context analysis means 9 comprising: the nominative personal limited vocabulary information holding means 13 described in claim 3; and the supplementary nominative personality limited vocabulary information holding means 14 described in claim 4 are all independent of each other and can be used in any combination. be able to. Also, by using them in combination, it is possible to more accurately perform the subject / nominative sharing recognition, and it is possible to further improve the accuracy and speed of analysis of Japanese sentences.

Further, here, as an embodiment, an example in which the subject / nominative shared estimation is performed immediately after the morphological analysis has been described, but this can be performed at an arbitrary stage after performing the morphological analysis. it can.

[0060]

As described above, according to the present invention, the subject / nominative sharing recognition can be performed more accurately when analyzing Japanese multiple / multiple sentences. Further, by presuming the sharing of the subject and the nominative prior to the parsing / semantic analysis, the ambiguity of the parsing / semantic analysis can be reduced, and the accuracy and speed of parsing the Japanese sentence can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention described in claim 1;

FIG. 2 is a block diagram showing one embodiment of the present invention described in claim 2;

FIG. 3 shows the relation between the connection form between words and the sharing condition of the subject and the nominative between the main sentence and the subordinate clause, which is stored in the subject and nominative shared vocabulary information holding means 8 of the present invention described in claim 2; Is shown in the form of a table.

FIG. 4 is a block diagram showing one embodiment of the present invention described in claim 3;

FIG. 5 is an example of an emotional adjective “painful” for information on a vocabulary having a nominative personality such as an emotion predicate, which is held in the subject / nominative vocabulary information holding means 8 according to the present invention. This is the table shown in FIG.

FIG. 6 is a block diagram showing an embodiment of the present invention described in claim 4.

FIG. 7 is a block diagram showing an embodiment of the present invention, wherein the subject / nominative shared vocabulary information holding unit 8 holds supplementary words such as words expressing strong intention of the speaker and words requiring the speaker to act on another person. Information about the vocabulary that controls the nominative status of a sentence is shown in the form of a table.

FIG. 8 is a block diagram showing one embodiment of a conventional context analysis method (Japanese Patent Application No. 61-216510) based on a vocabulary context grammar.

FIG. 9 is a diagram showing information, which is stored in the context format holding unit 11 and is obtained by classifying connection forms between declinable words based on what elements can be included in a clause in a clause accompanying each connection form. This is a summarized table.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input reading means 2 Dictionary 3 Morphological analysis means 4 Parsing means 5 Semantic analysis means 6 Analysis result 7 Subject / Nominative shared vocabulary information holding means 8 Subject / Nominative shared estimating means 9 Context analysis means 10 Context format holding means 11 Context structure collation Means 12 Context structure estimating means 13 Nominative personal limited vocabulary information holding means 14 Complementary nominative personalized vocabulary information holding means

Claims (4)

(57) [Claims] 1. An input reading means for reading an input of a Japanese sentence, a morphological analysis means for analyzing an input sentence using a dictionary lookup function for the input sentence and dictionary information after the dictionary lookup, a syntax analysis means, a semantic analysis means, and a memory. Japanese processing with
In the Japanese sentence analysis method used in the system, information on the vocabulary that prescribes the sharing of the subject indicated by the particle "ha" or the nominative case indicated by the case particle "ga" between a plurality of declinable words in Japanese. Collect and said
Estimating whether there is a subject / nominative sharing using the subject / nominative shared vocabulary information holding means held in the memory and the vocabulary information held in the subject / nominative shared vocabulary information holding means. Equipped with subject / nominative sharing estimation means, parsing
Estimate the existence of shared subject and nominative prior to semantic analysis
Japanese sentence analysis method characterized by that . 2. An input reading means for reading an input of a Japanese sentence, a morphological analysis means for analyzing an input sentence using a dictionary lookup function for the input sentence and dictionary information after the dictionary lookup, a syntax analysis means, a semantic analysis means, and a memory. Japanese processing with
In the Japanese sentence analysis system used in the system, information on vocabulary defining the in-sentence context structure is collected in advance, and the relationship between the expressions is formalized to form the aforementioned sentence.
The context format holding means held in the memory and the dictionary content sequence obtained as a result of morphological analysis of the input sentence are compared with the context information held in the context format holding means, thereby enabling syntax analysis and
Context structure matching means for detecting a context structure candidate in the input sentence prior to semantic analysis, and a context structure for estimating the most probable context structure candidate from a plurality of context structure candidates in the input sentence obtained by the context structure matching means Estimating means and information on vocabulary that prescribes sharing between a plurality of words of the subject indicated by the particle "ha" or the nominative case indicated by the case particle "ga" in Japanese, and collecting the memo in advance
The subject / nominative shared vocabulary information holding means held in the library, the vocabulary information held in the subject / nominative shared vocabulary information holding means, and the context structure estimated by the context structure estimating means. - with the subject-nominative shared estimating means for estimating whether share exists nominative, parsing Yi
Prior to taste analysis, the existence of shared
Japanese sentence analysis method characterized by having: 3. An input reading means for reading an input of a Japanese sentence, a morphological analysis means for analyzing an input sentence using a dictionary lookup function for the input sentence and dictionary information after the dictionary lookup, a syntax analysis means, a semantic analysis means, and a memory. Japanese processing with
In the Japanese sentence analysis method used in the system, information on the vocabulary that prescribes the sharing of the subject indicated by the particle "ha" or the nominative case indicated by the case particle "ga" between a plurality of declinable words in Japanese. Collect and said
And subject-nominative shared vocabulary information holding unit operable to hold in memory, and Primary person restriction vocabulary information holding unit that holds information about the vocabulary is nominative the person limitations such as emotional predicates in the memory, the nominative When a vocabulary with personal restrictions appears in the quotation section, using the vocabulary information held by the subject / nominative vocabulary information holding means and the vocabulary information held by the nominative personal restriction vocabulary information holding means, Equipped with a subject / nominative share estimating means for estimating whether or not there is subject / nominative sharing .
Japanese sentence analysis method characterized by estimating existence . 4. An input reading means for reading an input of a Japanese sentence, a morphological analysis means for analyzing an input sentence using a dictionary lookup function for the input sentence and dictionary information after the dictionary lookup, a syntax analysis means, a semantic analysis means, and a memory. Japanese processing with
In the Japanese sentence analysis method used in the system, information on the vocabulary that prescribes the sharing of the subject indicated by the particle "ha" or the nominative case indicated by the case particle "ga" between a plurality of declinable words in Japanese. Collect and said
Means for holding vocabulary information sharing the subject and nominative in memory, and information about vocabulary that controls the nominative noun in the complement, such as words that express strong intentions of the speaker or words that require the speaker to do something else In the memory, the vocabulary information held in the subject / nominative shared vocabulary information holding means, and the vocabulary information held in the complementary nominative limited vocabulary information holding means. using, with the subject-nominative shared estimating means for estimating whether the sharing of the subject-nominative present subject matter, mainly prior to the syntax analysis and semantic analysis
A Japanese sentence analysis method characterized by estimating the existence of case sharing .