JPH05120330A - Syntax analysis control system - Google Patents

Abstract

PURPOSE:To exactly and efficiently execute analysis by selecting an optimum candidate by simultaneously comparing and evaluating plural candidates, for which a 'to' infinitive phrase can be modified, in the middle of the analysis. CONSTITUTION:A syntax analysis part 1 inputs English word sequences to which a morpheme analysis processing is executed, successively pays attention from the phrase on the side of a sentence head, judges the modifying relation and decides English phrase structure or depending structure. In this case, when the phrase under the attention is the 'to' infinitive phrase, a 'to' infinitive phrase modifying destination deciding means composed of a depending relation decision part 20 and a semantic relation judgement rule volume 3 extracts and compares all the modifying relations between the word, which exists in the phrase adjacent to the sentence head side of the phrase under the attention and can be modified by the 'to' infinitive phrase, and the 'to' infinitive phrase. Among the extracted modifying relations, the most suitable relation is roughly selected while considering the entire relations in terms of syntax and meaning of the English sentences, and the modifying destination and use of the phrase in terms of meaning under the attention are decided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a syntactic analysis control system for a syntactic analysis system that mechanically analyzes a phrase structure or dependency structure of an English sentence, and particularly to a method of determining a modification destination etc. when a sentence includes a to infinitive phrase. The present invention relates to a parsing control method characterized by.

[0002]

2. Description of the Related Art In the case of mechanically analyzing a phrase structure or a dependency structure of a sentence, a morphological analysis process is performed in which an input sentence is first decomposed into words and the dictionary is referred to with its syntactical information and semantic information I do. Then, using the syntactic information and the semantic information added to the word by the morphological analysis process, the syntactic analysis process of analyzing the phrase structure or the dependency structure according to a predetermined rule is performed.

By the way, in the case of analyzing a phrase structure or dependency structure of an English sentence, when the above-mentioned English sentence includes a to-infinitive phrase, the above-to-infinitive phrase is modified because of its variety of usages and semantic relationships. There may be more than one word in the English sentence. In a parsing process using a conventional parsing system for a sentence including such an infinitive phrase, there is no direct interaction between a plurality of rules for determining a phrase structure, etc. It has only the ability to determine local modification relations and phrase structure, and it is not always possible to determine the proper modification destination of the to infinitive phrase at the time of applying each rule.

Further, in the conventional syntax analysis system, the rule for judging the dependency structure and the phrase structure is originally for judging only the top node of the parse tree expressing the already created phrase for the above relation. It was a thing. That is, in the case of phrase structure analysis, it is a rule for determining whether or not to construct a higher-level phrase having a phrase category or a vocabulary category located at the top of the already created phrase indicator as a dependency structure. In the case of analysis, it was a rule for judging whether or not the dependency relationship between the words located at the top of the dependency structure tree already created was established.

Therefore, conventionally, the analysis is proceeded while validating all the application results of each rule, and at the end of the syntax analysis,
Alternatively, it is necessary to forcibly select one and a half legitimate ones out of a plurality of output analysis results by referring to a case at a certain point during parsing that has been specially set or by a person making a judgment. there were.

[0006]

However, the above-described conventional parsing system determines an appropriate modification destination of a component of an English sentence containing a component having a plurality of words that can be modified as described above. However, there are disadvantages in that the analysis efficiency is poor and it is difficult to accurately determine the modification destination, because a plurality of extra analysis processes are involved. In addition, since the conventional rule for determining the dependency structure or phrase structure is targeted only at the top node of the parse tree, the conventional control method is used when there are multiple candidates for which the modifying component in the sentence can be modified. However, it was difficult to compare and evaluate these candidates at the same time during the analysis and select the optimal candidate.

The present invention solves the above-mentioned conventional drawbacks, and for an English sentence containing a to-infinitive phrase that locally has a plurality of modifiers, it is possible to efficiently determine the proper modifier of the to-infinitive from a global point of view. The purpose is to make a positive decision.

[0008]

In order to achieve the above object, the present invention inputs a word string of an input English sentence that has been subjected to morphological analysis processing, and sequentially focuses on words from the beginning of the word string, When it is determined that the word under consideration and the word adjacent to the sentence head side of the word under consideration are not combined as one phrase, the word under consideration and the word adjacent to the sentence head side of the word under attention are stacked. Storing and moving one point toward the end of the sentence to make it a new phrase under consideration, and it was determined that the phrase under consideration and the phrase adjacent to the beginning of the phrase under consideration are collected as one phrase. In this case, the word under consideration and the word adjacent to the head side of the word under consideration are combined into a new word under consideration, and the beginning of the word under consideration recursively extracted from the stack. Side by side words and above In a parsing control method of a parsing system that performs a parsing process for determining a phrase structure or a dependency structure of the input English sentence by determining a modification relation with the phrase in question, the phrase under attention is a to infinitive phrase. In this case, the word that exists in the word adjacent to the beginning of the word of interest and can be modified by the to infinitive phrase and the word to
All modifier relationships with infinitive phrases are extracted and compared, and among the extracted modifier relationships, the most appropriate one is globally selected in consideration of the overall syntactic and semantic relationship of the English sentence. A parsing control method, comprising: to infinitive phrase modification destination determining means for determining a modification destination and semantic usage of the phrase.

[0009]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the syntax of a syntax analysis system using a syntax analysis control method according to an embodiment of the present invention.

As shown in the figure, the syntax analysis system of the present embodiment inputs an English word string subjected to morpheme analysis processing, sequentially pays attention to words at the beginning of the sentence, and determines their modification relationships to determine the English phrase structure. A syntactic analysis unit 1 for determining a dependency structure and a completeness determination rule set 2 and a semantic relation determination rule set 3 as a rule base.

The syntactic analysis unit 1 and the completeness determination unit 10 (hereinafter, referred to as the above-mentioned, will be referred to as "completeness determination unit 10" for determining whether or not the word under consideration and the word adjacent to the sentence head side of the word under consideration are combined into one phrase. It is referred to as “completed” when two words and phrases are collected as one phrase), and the dependency relationship determining unit 20 that determines whether or not there is a modifying relationship between the two words and phrases.
And an analytic tree creating unit 30 that creates an analytic tree based on the judgment of the dependency relationship determining unit 20, and an analytic control unit 40 that controls the above units.

The completeness determination unit 10 has a completeness determination table 11 as shown in FIG. Then, the word under consideration and the word adjacent to the sentence head side of the word under attention are collated with the completeness determination table 11. Here, A in FIG. 3 indicates a word adjacent to the beginning of the word under consideration (hereinafter, this word is referred to as word A), and B indicates the word under attention (hereinafter, This phrase is called phrase B).

The completeness determination table 11 includes words A and B.
Is defined as to whether or not the phrase can be modified based on the completeness of the phrase. Based on the completeness determination table 11, the completeness determination unit 10 determines whether or not the phrase A and B can be modified.
That is, when the phrases A and B correspond to the combination indicated by the ⊚ mark in FIG. 3, the phrase B is a necessary component waiting for the phrase A to be modified to complete one phrase. , Or a component that can complete the phrase A and modify the phrase A arbitrarily, and the completeness determination unit 10 determines that the phrase B may modify the phrase A.

On the other hand, when the terms A and B correspond to an unmarked combination, the term B is arbitrarily modified even if it is a component that the term A is modified to complete one phrase. It is not a component that can be modified, and the completeness determination unit 10 uses the phrase B.
Judges that the phrase A is not modified.

Further, when the terms A and B correspond to the combination indicated by a star, the adequacy of the qualification differs depending on the types of the terms A and B, the presence / absence of the qualification relation with other terms, etc. The determination unit 10 individually determines whether or not there is a possibility that the phrase B modifies the phrase A based on the completeness determination rule stored in the completeness determination rule collection 2.

Here, in the completeness judgment rule collection 2, whether or not the words A and B can be modified is described based on the types of the words A and B and the presence / absence of a modification relationship with other words. A completeness determination rule is stored corresponding to each name of the combinations shown by the above-mentioned asterisks. The following table shows an example of a rule for determining completeness.

[0017]

[Table 1]

In the case where the word A modifies the word B, the completeness judging section 10 also judges the possibility.

When the completeness judging unit 10 judges the words A and B based on the completeness judging table 11 and then judges that there is a possibility that a modifying relationship is established between the words A and B, the words and phrases are judged. A and B are passed to the dependency relationship determining unit 20, and when it is determined that they are not satisfied, they are passed to the analysis control unit 40.

The dependency relationship determining section 20 has a dependency relationship determination table 21 as shown in FIG. Then, the phrases A and B passed from the completeness determination unit 10 are collated with the dependency relation determination table 21. Here, A and B in FIG. 4 are the same as A and B in FIG.

The dependency relation table 21 includes the words A,
The presence / absence of a modification relationship regarding B is defined, and the dependency relationship determining unit 20 determines the modification relationship between the words A and B based on the dependency relationship determination table 21. That is, when the terms A and B correspond to the combination indicated by ⊚ in FIG. 4, the term B necessarily modifies the term A, and accordingly, the dependency determining unit 20 has the terms A and B in a predetermined relationship. Determine the modifying relationship between.

On the other hand, when the words A and B correspond to the unmarked combination, the word B never modifies the word A, and the dependency determining section 20 determines that the modifier relationship between the words A and B is not established. To do.

When the terms A and B correspond to the combination indicated by a star, it is necessary to determine whether the term B modifies the term A or whether the term A modifies the term B. Needs to consider the overall syntactic / semantic relationship including other phrases in addition to the combination of the phrases A and B.
Depending on the combination of B, the modifying relationship between them is individually determined.

The analytic tree creating section 30 includes a dependency determining section 20.
When it is determined that the modification relation between the phrases A and B is established, the parse tree is created based on the result. Then, this analysis tree is passed to the analysis control unit 40.

The analysis control unit 40 defines the parse tree passed from the parse tree creating unit 30 as a new word B, and the completeness judging unit 10
Pass to. Further, when the completeness determination unit 10 or the dependency relationship determination unit 20 determines that the modification relation is not established between the phrases A and B, the phrase A is put in the analysis stack, and the completeness determination is performed with the phrase B as a new phrase A. Hand over to part 10.

Here, a method of determining the modification destination of the to infinitive phrase when the input English sentence includes the to infinitive phrase will be described in detail. The to-infinitive phrase has various uses as a modifier for other words, such as a verb modifier, a noun modifier, an adjective modifier, an adverb modifier, and a preposition modifier. Also, in the semantic relationship in each usage, when modifying a verb, for example, a noun is modified as an object, an adverb is modified, or a complement is modified. In the case of modification, there are various relations such as modification so as to define the content attribute of the noun, modification when the noun has a target attribute, and the like.

As described above, since the to infinitive phrase has various usages and semantic relationships with the modified word, the word that can be modified by the to infinitive phrase (hereinafter, simply referred to as “modified word”) is in English. There may be a plurality of cases. In such a case, the to-infinitive phrase modification destination determining unit in the syntax analysis control method of the present embodiment determines a valid modification destination of the to-infinitive phrase.

The to infinitive phrase modification destination determining means of this embodiment comprises a dependency relationship determining section 20 of the syntactic analysis section 1 and a semantic relationship determination rule collection 3.

In the above-mentioned semantic relation determination rule set 4, there are rules for determining a modifying relation between words, such as how the infinitive phrase modifies a given word and with what semantic relation. It is stored. As this rule, the rule used in the conventional parsing system can be used.

The dependency relationship determining section 20 first sets in advance for each modified relationship corresponding to the modified relationship between the infinitive phrase and the modified word obtained from the semantic relationship determination rule set 3 for each modified word. The added points are added. The table below shows an example of setting points for each modification relationship. It should be noted that "C" in the table is "adjective of type""means that adjectives are classified into several types based on their relationship with other words in the sentence.
Indicates that the score has been set for each type.

[0031]

[Table 2]

The scores shown in the above table are for determining the relative priority of each modified word having each modification relation with the to infinitive phrase. Therefore, the numerical value of the score itself need not be limited to that shown in Table 2 as long as the same priority can be obtained.

For example, in determining the modification destination of a to-infinitive phrase when a noun is between a phrase that modifies a verb and a to-infinitive phrase, the verb is modified with the target to-infinitive, and the noun is associated with the objective case relation. When the result of qualifying with is obtained, comparing with Table 2, it is 4 points if the verb is modified with the objective to infinitive, and 20 points if the noun is modified with the objective case relation, and the objective case relation is scored. Is high, the result is that “the objective case relation is given priority over the objective to infinitive”. Of course, a rule base storing a large number of such rules may be constructed, and when there are a plurality of to-infinitive qualifiers, the rules in the above rule base may be applied to determine the proper modification destination.

Further, the order of priority among the modified words is empirically determined, and can be changed according to the field containing the content of the English sentence to be parsed. In this case, in particular, if the priority order is set to rule base, it is necessary to change the rule as a whole, but by making it as shown in Table 2, partial changes can be made easily. In addition, when the scores of the corresponding qualification relationships are the same for multiple modified words, that is, when the priority of the modified words cannot be determined, information based on the structure of other English sentences is used. Based on, etc., the modification destination of the infinitive is determined. Such information includes 1. 1. When determining the priority order between verbs, prioritize the farthest.
2. When determining the priority order between nouns and adjectives, prioritize the closest ones. When determining the priority order between different parts of speech, there is a verb priority (default).

Next, the dependency determining section 20 selects the modified word with the highest added score as the legal modified target of the to infinitive phrase. FIG. 2 is a flow chart showing the modification destination determining operation by the above infinitive phrase modification destination determining means.

The to-infinitive phrase modification destination determining means described above is used when the modified word of the to-infinitive phrase is on the sentence head side of the to-infinitive phrase (that is, the modified word is in the phrase A,
to infinitive phrase B)), t
Even if it is on the tail end side of the infinitive phrase (that is, the modified word is in the phrase B and the to infinitive phrase is the phrase A), it can be used without any difference.

When there is only one word that can be modified by the to infinitive phrase in the English sentence, the modification destination of the to infinitive phrase is uniquely determined without selecting by the above-described to infinitive phrase modification destination determining means. Needless to say.

Next, the analysis control method of this embodiment will be described in detail with reference to specific analysis examples. FIG. 5 shows the syntax analysis system of the present embodiment, "The company withd
rew a possible toincrease
the number of authenticated
shares. "A" and "B" in the case of analyzing the English sentence "," and the words in the analysis stack 31. The above example sentence is input to the syntactic analysis unit 1 after morphological analysis, and is input word by word from the beginning of the sentence. The modification relationship is determined.

In the present embodiment, the description of the determination of the modification relation for each word is omitted, and as shown in FIG.
e company withdraw a prop
The determination of the modification relation between the complete phrase "phrase A" and the infinitive phrase "phrase B" will be described.

Here, the phrase A is a verb phrase and the phrase B is
Is a to infinitive phrase, the completeness determination unit 10 determines that a modification relation may be established in light of the completeness determination table 11, and the dependency relation determination unit 20 determines that the phrase A and the phrase B are satisfied. Pass to.

The dependency relationship determining unit 20 compares the terms A and B with the dependency determination table 21 and individually determines them. First, all the modification relationships between all the words in the phrase A that can be modified by the to infinitive which is the phrase B and the phrase B are extracted.
Here, the infinitive phrase of phrase B is “wit” in phrase A.
Since "hdrew" or "proposal" is modified, "to infinitive phrase is" w
There is a possibility that it will be modified by the adverbial usage (purpose). "and" to infinitive phrase is "prop
osal "may be modified by adjective usage (derivative from verb)."

Next, the dependency relationship determining section 20 selects an optimum one among the extracted modification relationships. Here, Table 2
"To modify with adverb usage (purpose)", which is more related to "withdraw", has 4 points, "proposa"
There is a possibility of modifying "withdrew" with an adverbial usage (purpose) for "to infinitive phrase" since the "adjective usage (derivative from verb)" relationship with l "is 10 points. . ”Rather than“ to infinitive phrase is “proposal” adjective usage (derived from verb)
May be modified with. Therefore, it is determined that the phrase B modifies “proposal” in the phrase A by the adjective usage. Then, these judgment results are passed to the analysis tree creating unit 30. As above
The proper modification destination of the infinitive phrase will be determined.

[0043]

As described above, according to the syntactic analysis control method of the syntactic analysis system of the present invention, the determination of the local word-to-word modification relation is always performed under the global comprehensive determination. Therefore, even if there are multiple candidates in the input English sentence that include a to infinitive phrase and the to infinitive phrase can be modified, the candidates are compared and evaluated at the same time during the analysis. Therefore, it is possible to easily perform the control of selecting the optimum candidate, and it is possible to perform the accurate and efficient analysis.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a syntax analysis system using a syntax analysis control system of the present invention.

FIG. 2 is a flowchart showing a to infinitive phrase modification destination determining means by the syntax analysis control system of FIG.

FIG. 3 is a diagram showing a completeness determination table of FIG.

FIG. 4 is a diagram showing a dependency relationship determination table of FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Syntax analysis part 2 Completeness judgment rule collection 3 Semantic relationship judgment rule collection 10 Completeness judgment part 20 Dependency determination part 30 Analysis tree creation part 40 Analysis control part

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] October 19, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Brief description of the drawing

[Correction method] Change

[Correction content]

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a syntax analysis system using a syntax analysis control system of the present invention.

FIG. 2 is a flowchart showing a to infinitive phrase modification destination determining means according to the syntax analysis control method of FIG.

FIG. 3 is a diagram showing a completeness determination table of FIG.

FIG. 4 is a diagram showing a dependency relationship determination table of FIG. 1.

FIG. 5 is a diagram showing an analysis example according to the present embodiment.

[Explanation of Codes] 1 syntax analysis unit 2 completeness determination rule collection 3 semantic relation determination rule collection 10 completeness determination unit 20 dependency relationship determination unit 30 parse tree creation unit 40 analysis control unit

Claims (1)

[Claims] 1. A word string of an input English sentence that has been subjected to morphological analysis processing is input, and attention is sequentially paid to words beginning with a sentence beginning side of the word string. When it is determined that the words and phrases are not combined as one phrase, the word of interest and the word adjacent to the sentence beginning side of the word of interest are stored in the stack, and the point of interest is moved one by one toward the end of the sentence. As a new phrase under attention, and when it is determined that the phrase under consideration and the phrase adjacent to the sentence beginning side of the phrase under consideration are collected as one phrase, the phrase under consideration and the sentence beginning side of the phrase under consideration The adjacent words and phrases are combined into a new word of interest, and the modified relation between the adjacent words and the word of interest is recursively extracted from the stack at the beginning of the word of interest. Of the above input English sentence In a parsing control method of a parsing system that performs a parsing process for determining a structure or a dependency structure, when the word of interest is a to infinitive phrase, among words adjacent to the beginning of the word of interest Existing in the above infinitive phrase and all the modifying relations of the above infinitive phrase and the above infinitive phrase are extracted and compared, and the most appropriate one of the extracted modifying relations is extracted as a whole syntactical sentence of the English sentence. A syntactic analysis control method comprising: to infinitive phrase modification destination determining means for globally selecting in consideration of a semantic relationship and determining a modification destination and a semantic usage of the phrase of interest.