JPH08115217A - Grammatical rule extension system - Google Patents

Abstract

PURPOSE: To provide a syntax rule extension system for automatically generating grammatical rules required for making a syntax unanalyzable sentence syntax analyzable. CONSTITUTION: The syntax analysis of a rephrased sentence having the same meaning as the syntax unanalyzable sentence is performed in a syntax analysis process generation device 2 by using the grammatical rules stored in a grammatical rule storage device 1 and a syntax analysis process is generated. The syntax analysis process for the rephrased sentence is converted to the syntax analysis process of the syntax unanalyzable sentence in a syntax analysis process conversion device 3 and the grammatical rules not included in the grammatical rules among the converted syntax analysis process are extracted in an extension grammatical rule extraction device 4.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a grammar rule extension system which is used for a natural language interface such as an expert system and a knowledge base system and which automatically extends grammar rules described by grammar rules.

[0002]

2. Description of the Related Art As a system close to this type of grammar rule expansion system, there is a conventional system, for example, Japanese Patent Laid-Open No. 5-135100.
There is a word learning method and its system described in Japanese Patent Publication. This system automatically learns the meaning of undefined words based on human operation instructions for the interpretation of the system.

[0003]

In the conventional natural language learning system, as described above, the undefined word is simply registered based on the operation instruction, and the grammatical rule used for parsing is expanded. Was not. However, in the natural language interface, it is necessary to parse input sentences with various syntaxes input by the user,
Not only the operation instruction information of undefined words must be registered, but also the grammar rules themselves need to be updated. Since updating such grammar rules is a very complicated and specialized task, it is performed by an expert with knowledge of natural language grammar,
It required cost and time.

The present invention has been made in view of the above,
It is an object of the present invention to provide a grammar rule extension system that automatically generates the grammar rules necessary for making a parsable sentence parseable.

[0005]

In order to achieve the above object, the grammar rule expansion system of the present invention has the same meaning as a sentence that cannot be parsed even if the current grammar rule is used, and a sentence that cannot be parsed. (EN) A grammar rule extension system for newly generating a grammar rule necessary for enabling parsing of a non-parsable sentence when a paraphrase sentence that can be parsed by a current grammar rule is input. A grammar rule storage device that stores rules, and a grammar rule stored in the grammar rule storage device are applied, and the paraphrase sentence is parsed,
A syntactic analysis process generation device that generates a syntactic analysis process from a sequence of words to which a part of speech is given, and a correspondence relationship for each word that constitutes the non-parsizable sentence with respect to the syntactic analysis process of the parsable paraphrase sentence. A parsing process conversion device for converting a parsing process of the paraphrase sentence into a parsing process of the unparsable sentence by checking excess or deficiency, and the unparsable sentence converted by the parsing process conversion device. And a new grammar rule extracting device for extracting grammar rules not included in the grammar rules in the syntactic analysis process.

[0006]

In the grammar rule expansion system of the present invention, the parsing process of a paraphrase sentence having the same meaning as a sentence that cannot be parsed is performed by using grammar rules to generate a parsing process, and the parsing process of the paraphrase sentence is performed. It is converted into a parsing process of an unparseable sentence, and grammar rules not included in the grammar rules are extracted from the converted parsing process.

[0007]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a grammar rule extension system according to an embodiment of the present invention. The grammar rule expansion system shown in FIG. 1 applies a grammar rule storage device 1 that stores a current grammar rule and a grammar rule stored in the grammar rule storage device 1 to parse a paraphrase sentence that can be parsed. A parsing process generation device 2 for generating a parsing process by inputting an unparseable sentence to the parsing process of the paraphrase sentence and converting the parsing process of the paraphrase sentence into the parsing process of the parsing sentence. A parsing process conversion device 3 and an extended grammar rule extraction device 4 for extracting grammatical rules not included in the grammatical rules in the parsing process of the unparseable sentence converted by the parsing process conversion device 3. Has been done.

Next, the operation of the grammar rule expansion system shown in FIG. 1 will be described with reference to the flowchart shown in FIG. In FIG. 2, when the paraphrase text subjected to morphological analysis is input to the syntactic analysis process generation device 2 (step S2
1), the syntactic analysis process generation device 2 uses the grammatical rules stored in the grammar rule storage device 1 to generate and output the parsing process of the paraphrase sentence (step S22). The morphologically unanalyzable sentence is input to the syntactic analysis process conversion device 3 as an input sentence (step S23). The syntactic analysis process conversion device 3 generates and outputs a syntactic analysis process regarding the input sentence from the syntactic analysis process of the input sentence and the paraphrase sentence generated by the syntactic analysis process generation device 2 (step S2).
4). The extended grammar rule extraction device 4 compares the grammar rules stored in the grammar rule storage device 1 with the converted syntactic analysis process to extract and output new grammar rules (step S25,
S26).

As a specific example of the present embodiment, a case will be described in which an input sentence "He has children" is given as a paraphrase sentence "He has children". In addition, since there is no difference between the sentences of "there is a child" in both sentences, in order to simplify the description, only the "he is" and "he is" parts having a difference will be described below.

First, the data description format will be described. X,
Y, Z and W represent variables. [X | Y] indicates that the head of the list is X and the rest is Y. [X
\ Y] represents a difference list between X and Y. For example,
"[A, b, c] \ [b, c]" is the list [a, b,
The difference [a] between [c] and [b, c] is shown. "Postposition ([de | X] \ X)" is treated as a factual grammar in which "de" is a postposition. Also,
"Postpositional phrase (X \ Y):-Noun phrase (X \ Xs), postpositional (Xs \ Y)" means "postpositional phrase is a noun phrase followed by it.
It is treated as a representation of a regular grammar "consisting of postpositions". The left side of ":-" in the regular grammar is called a head, and the right side of ":-" is called a body. Hereinafter, description will be made using this notation. For details, refer to the literature (Leon Sterling and
Ehud Shapiro, THE ART OF PROLOG, MIT Press, 1986, translated by Toshio Matsuda, "Techniques of Prolog", Kyoritsu Shuppan, 1988). For the sake of explanation, grammar rules may be expressed in a tree structure. For example, the rule of "A: -B, C" is expressed as a tree structure in which nodes B and C are connected under the node A, respectively. The nodes B and C are arranged in the order from left to right according to the order of the original rule.

According to the above notation, "postposition phrase ([he,
, Ha] \ []) ”and the result of morphological analysis are given as input sentences, and“ postposition phrase ([he, ha] \ []) ”is given as a paraphrase sentence, the procedure of FIG. 2 is followed. An example of extending the grammar rule will be described. FIG. 3 shows the contents registered in the grammar rule storage device 1 in this embodiment.

The syntactic analysis process generation device 2 receives "postpositional phrase ([he, ha] \ [])" as an input sentence, and from the grammar rules registered in the grammar rule storage device 1, parses the input sentence. Generate a process. The syntax analysis process of the "postpositional phrase ([he, ha] \ [])" for explaining the grammatical rule and the rule in which a specific value is assigned to the grammatical rule is as shown in FIG. This is a postpositional phrase ([he, ha] \
[]) ”Is the result of applying the grammatical rule of FIG. Grammar rules are applied in the following steps. First, a grammatical rule with which the head can match "postpositional phrase ([he, ha] \ [])" is searched for in FIG. Since only R1 is a candidate, collation is performed and a value is assigned to the corresponding variable. Figure 3
Substitute [he, ha] for X and [] for Y of R1. As a result, "postpositional phrase ([he, ha] \ []):-noun phrase ([he, ha] \ Xs), postposition (Xs \ [])" is obtained. Next, this body is checked against the grammar rules in order from the left. This expansion continues until the factual grammar rules are matched. "Noun phrase ([he, ha] \ Xs)" can be matched only with R7. The list of [he, ha] excluding “he”, that is, [ha] is substituted for Xs as a result of the collation. Next, if it is checked whether "postposition ([wa] \ [])" can be matched with the grammar rule, it can be seen that it can be matched with R2 and R3. In the result of matching with R2, the case particle ([wa] \ []) of the body is not selected because there is no grammatical rule that can be matched in FIG. On the other hand, in the result of matching with R3, the body “sub particle ([wa]
\ []) "Can be matched with the factual grammar rule R6 and is therefore selected. As a result, the postpositional phrase ([he, ha]
\ []):-Noun phrase ([he, ha] \ [ha]), postposition ([wa] \ []) "and" postposition ([wa] \ []):-adverb ([wa ] \ []) "Is generated. FIG. 4 shows this in a tree structure as a parsing process. The syntax analysis process generation device 2 outputs this syntax analysis process.

The syntactic analysis process conversion device 3 converts the syntactic analysis process output by the syntactic analysis process generation device 2 into a syntactic analysis process for "postposition phrase ([he, ni, ha] \ [])". . First, find the correspondence with the least difference between the words in the unparseable sentence and the paraphrase sentence, then change the parsing process to compensate for the excess or deficiency of the word based on the correspondence, and finally generalize. To do.

The end of the parsing process in FIG. 4 is a noun ([he, ha] \ [wa]), an adverb ([wa] \ []), and a "postpositional phrase ([he, ni, ha] \ The part of speech of each word appearing in []) is a noun ([he | X] \ X) a case particle ([ni | X] \ X) and a sub particle ([wa | X] \ X).

The correspondence with the smallest number of pair differences is {<noun ([he, ha] \ [wa]), noun ([he | X] \ X)><none, case particle ([ni | X] \ X)><sub particle ([wa] \ []), subparticle ([wa | X] \ X)>}.

FIG. 5 shows how the parsing process is converted based on this correspondence. Change from the correspondence of the first word of the sentence in order. Correspondence <noun ([he,
Ha] \ [wa]) and noun ([he | X] \ X)> are the same, so do not change. <None, case particle ([ni | X] \
X)>, the case particle ([ni |
[X] \ X) is required. The only node that can be added is the postpositional phrase, so add it there. The correspondence <sub particle ([is | X] \ X) and sub particle ([is | X] \ X)> is the same for both, and therefore is not changed. In this way, the parsing process of FIG. 6 is obtained.

Next, the parsing process is generalized. First,
All the arguments of each node are replaced with independent variables as shown in FIG. 6 to obtain FIG. Then, as shown in FIG. 8, the variables of each node are replaced so as to satisfy the grammar rules. Specifically, the variable of the first argument of the parent node is made the same as the first argument of the leftmost child node, and the second argument of the child node is made the same as the first argument of the child node on the right side. . Similarly,
Make the second argument and first argument of adjacent child nodes the same. Finally, the second argument of the rightmost child node is the same as the second argument of the parent node.

As described above, the converted syntax analysis process as shown in FIG. 8 is generated and output from the syntax analysis process conversion device 3.

The extended grammar rule extraction device 4 receives the converted syntax analysis process as an input, extracts a new grammar rule, and outputs it. The procedure for extracting a new grammar rule is to decompose the parsing process into grammar rules, compare the grammar rules stored in the grammar rule storage device 1 and the decomposed grammar rules, and extract the grammar rules that do not exist in the grammar rule storage device 1. Output the rule as a new grammar rule.

The parsing process of FIG. 8 represents the following two grammar rules R * and R #. (R *) Postpositional phrase (X \ W):-Noun (X \ Y), Case particle (Y \ Z) Postpositional (Z \ W) (R #) Postpositional (Z \ W):-Adverbial (Z \ W) Of these, R # is not the new grammar rule because it is the same as R3. On the other hand, there is no grammatical rule corresponding to R *, so R *
Is output as a new grammar rule.

In this way, the postpositional phrase ([he, ni,
When [] \ []) ”is given as an input sentence and“ postpositional phrase ([he, ha] \ []) ”is given as a paraphrase sentence, a new grammar rule R * is extracted and It can be expanded.

[0023]

As described above, according to the present invention,
Parsing a paraphrase sentence having the same meaning as a non-parsing sentence by using grammar rules to generate a parsing process, converting the parsing process for a paraphrase sentence into a parsing process of a non-parsing sentence, Since the grammar rules that are not included in the grammar rules are extracted from the converted parsing process, new grammar rules can be automatically created in a short time without the need for an expert with knowledge of natural language grammar as in the past. It is generated and can be made economical.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a grammar rule extension system according to an exemplary embodiment of the present invention.

FIG. 2 is a flowchart showing an operation of the grammar rule extension system shown in FIG.

FIG. 3 is a diagram showing an example of contents stored in a grammar rule storage device used in the grammar rule expansion system of FIG.

FIG. 4 is a diagram showing a syntax analysis process for a paraphrase sentence in a tree structure.

FIG. 5 is a diagram showing how a parsing process is converted.

FIG. 6 is a diagram showing an obtained syntax analysis process.

FIG. 7 is a diagram showing how the arguments of each node shown in FIG. 6 are replaced with independent variables.

FIG. 8 is a diagram showing a converted parsing process.

[Explanation of symbols]

 1 grammar rule storage device 2 syntactic analysis process generation device 3 syntactic analysis process conversion device 4 extended grammar rule extraction device

Claims (1)

[Claims] 1. When a sentence that cannot be parsed using the current grammar rule and a paraphrase sentence that has the same meaning as the sentence that cannot be parsed and that can be parsed by the current grammar rule are input, the parsing is performed. A grammar rule expansion system for newly generating a grammar rule necessary to enable parsing of an impossible sentence, the grammar rule storage device storing a current grammar rule, and a grammar stored in the grammar rule storage device. Applying a rule, parsing the paraphrase sentence, and a parsing process generation device for generating a parsing process from a sequence of words with parts of speech added, and for the parsing process of the parsable paraphrase sentence Parsing process conversion for converting the parsing process of the paraphrase sentence into the parsing process of the unparsable sentence by checking the correspondence between each word constituting the unparseable sentence And a new grammar rule extracting device for extracting grammatical rules not included in the grammatical rules in the parsing process of the unparseable sentence converted by the parsing process converting device. Grammar rule extension system.