JPH02226364A - Syntax analyzing device - Google Patents

Abstract

PURPOSE:To smoothly and speedily execute exact analysis by including information concerning an erroneous sentence, which includes an erroneous phrase syntax rule, in a grammatical rule table besides a correct phrase syntax rule and allocating these information concerning the erroneous sentence, which includes the erroneous syntax rule, to a sentence including an error. CONSTITUTION:In a syntax analysis part 2, the information including words are inputted from a vocabulary analysis part 1 and the information including the phrase syntax rule and the information including the kind of the word are respectively read from first and second storing parts 3 and 4. Then, the syntax of the input sentence is analyzed by these information and the information, which includes the words, inputted from the vocabulary analysis part 1. In this case, since the information concerning the erroneous sentence including the erroneous phrase syntax rule are included in the grammatical rule table stored in the storing means 3 besides the correct phrase syntax rule, the syntax analysis of the erroneous part in this input sentence can be executed by using the information concerning the erroneous sentence, which includes this erroneous phrase syntax rule, even when there is the erroneous part in the input sentence. Thus, the exact analysis of the input sentence can be smoothly and speedily executed.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a syntactic analysis device, and for example, in parsing natural language, a sentence containing a grammatical error is converted into a correct sentence containing no errors. It also concerns something that can be parsed.

[Prior Art] The principle of conventional syntactic analysis is to create phrase structure rules according to the type of sentence to be analyzed (for example, English, Japanese, etc.), and then to analyze the input sentence. By applying each combination of words to the phrase structure rules, the input sentence is expressed as a combination of the phrase structure rules, and the structure of the input sentence is known.

By the way, if the input sentence is completely correct, it can be analyzed relatively smoothly by applying the above-described structural analysis principle as is.

However, if there is an error in the input sentence, it often becomes impossible to analyze the structure simply by applying the above principle. Therefore, various methods have been proposed in the past that have made it possible to analyze the structure of sentences containing errors.

Many of these methods are good in theory, but require highly specialized grammar knowledge when designing the system to actually perform the parsing, or require highly specialized grammar knowledge in the finished system. When structural analysis is performed, it often takes a considerable amount of time, which is a drawback.

Therefore, as a method that allows easy system design and efficient analysis, for example, Japanese Patent Application Laid-Open No. 60-3366
A method described in Publication No. 3 is proposed.

With this method, for example, if an article appears in the input sentence, it is the beginning of a noun phrase, and it can be predicted that either an adjective or a noun will follow, so you can cut it there. If the expected part of speech does not appear, it is determined that the word contains an error, and the erroneous part is skipped in the subsequent analysis using the above regulation condition. This is what I decided to do.

In other words, when a part of speech other than a noun or adjective comes after an article, it is obvious that it is an error, so there is no need to go back to the article and redo the analysis from there (backtracking). The aim is to improve the efficiency of analysis.

[Problem to be solved by the invention] However, the above method can only detect errors in sentences where cuts can be made.
There is, and the part where you can put in this Kak1~ is,
This method was limited to parts where specific conditions were met to determine the subsequent part of speech, and there was a limitation in that errors could not be detected in other cases.

The present invention was made against the above-mentioned background, and aims to provide a syntax analysis device that can detect errors in various aspects and perform efficient analysis. .

[Means for Solving the Problems] The present invention solves the above problems by having a configuration including the following elements.

■ A word yellow analysis unit that decomposes the input sentence into words and outputs information including the decomposed words.

■ A first storage means for storing a grammar rule table having information including phrase structure rules created in advance according to the type of input sentence.

(2) A second storage means for storing a word dictionary including information including the types of each word depending on the type of input sentence.

(2) Input information including decomposed words from the vocabulary analysis unit, read out information including phrase structure rules and information including word types from the first storage means and second storage means, respectively, and combine these information with each other. a syntactic analysis unit that analyzes the syntax of the input sentence using information including words input from the word phrasing analysis unit;

(2) The grammar rule table stored in the first storage means includes information regarding erroneous sentences that include incorrect phrase structure rules in addition to correct phrase structure rules.

[Operation] According to the above configuration, the input sentence is broken down into words by the vocabulary analysis section, and analyzed by the syntax analysis section.

The syntactic analysis unit inputs information including words from the vocabulary analysis unit, reads out information including phrase structure rules and information including word types from the first storage means and second storage means, respectively, and processes them. The syntax of the input sentence is analyzed using the information and the information including the group words input from the vocabulary analysis section.

At this time, since the grammar rule table stored in the first storage means includes information regarding the erroneous sentence including the erroneous phrase structure rule in addition to the correct phrase structure rule, the erroneous part of the input sentence is Even in such a case, the erroneous part of the input sentence can be quickly parsed using information about the erroneous sentence including the erroneous phrase structure rule.

That is, for example, in an input sentence, what should be the correct phrase structure rABJ is changed to rAb
, the correct phrase structure rAB is included in the grammar rule table.
Since rAbJ is included as an erroneous phrase structure as well as J, it is possible to continue parsing by applying this erroneous phrase structure rule rAb to the input sentence, and also to clarify that this part is an error. can be shown.

[Embodiment] (First Embodiment) FIG. 1 is a functional block diagram showing a schematic configuration of a parsing device according to a first embodiment of the present invention, and FIG. 2 shows an example of the actual configuration of the device in FIG. 1. FIG. 3 is an explanatory diagram of the grammar rule table in FIG. 1, FIG. 4 is an explanatory diagram of the word dictionary in FIG. 1, and FIG. 5 is an explanatory diagram of the syntax analysis device in FIGS. FIG. 3 is an explanatory diagram of a syntax parsing tree during analysis. Hereinafter, a first embodiment of the present invention will be described in detail with reference to these drawings.

In FIG. 1, reference numeral 1 is a vocabulary analysis section, reference numeral 2 is a syntactic analysis section, reference numeral 3 is a first storage means for storing a grammar rule table, and reference numeral 4 is a second storage means for storing a word dictionary.

The vocabulary analysis section 1 decomposes an input sentence into words and sends information including the decomposed words to the syntax analysis section 2.

The syntactic analysis unit 2 inputs information including decomposed words from the vocabulary analysis unit 1, and inputs the information including the decomposed words from the vocabulary analysis unit 1, and inputs the information including the grammar rule table stored in the first storage unit 3 and the grammar rule table stored in the second storage unit 4. Information including phrase structure rules and information including word types are read from each word dictionary, and the syntax of the input sentence is analyzed using these information and the information including words input from the vocabulary analysis unit 1, and the analysis result is obtained. and sends the message to another information processing section (not shown), such as a natural language processing section.

The first storage means 3 stores a grammar rule table as shown in FIG. 3, and the second storage means 4 stores a word dictionary as shown in FIG. 4.

The grammar rule table is composed of phrase structure rules and messages, and the word dictionary is composed of words and the parts of speech of the words.

The phrase structure rules indicate rules regarding the correct phrase structure and incorrect phrase structure of a word depending on the type of input sentence, such as English or Japanese. In addition, in the grammar table,
In English, all the phrase structure rules of English are listed, but the third
In the figure, only a portion is shown to simplify the explanation.

The phrase structure rules in FIG. 3 represent the following contents.

That is, for example, the phrase structure rule ■ in FIG. 3 indicates the correct phrase structure rule, and np --> art
l nl, which means 'np' (noun phrase) is 'art
It means that it is composed of "l" (article) and rnL+ (noun). Here, according to the word dictionary shown in Figure 4, "rartl" is the article "a",
Also, since "nl" is the noun "boOk",
The combination of words ``ra book'' can be applied to the rule shown in phrase structure rule ■ in the grammar rule table, and it can be seen that it has a correct phrase structure.

In addition, phrase structure rule ■ represents an incorrect phrase structure rule, *np-->artl n2, which means that *"np" (noun phrase) is rar
It is composed of tL+ (article) and rn2'' (noun), which indicates that it is an incorrect phrase structure.

Here, rartlJ is the article "a" and "n2" is the noun rorangeJ, so
The word combination "a Orange" can be applied to the rule shown in the phrase structure rule ■ of the grammar rule table, and it can be seen that it has an incorrect phrase structure. In this case, a message ``The article a is used for singular nouns starting with a consonant.'' is added to indicate the content of the error in accordance with the phrase structure rule of this error.

Furthermore, for example, the phrase structure rule ■ is vp-->vnp, which means that "VpJ (verb phrase) is "■" (
(verb) and "np" (noun phrase). In this case, "np" is a phrase structure shown by rules ■ to ■ (note that rules ■ and ■ are incorrect phrase structures).

FIG. 5 is a syntactic analysis tree showing how the input sentence "II have an book" is analyzed by the syntax analysis device configured as described above.

First, the input sentence rI have an book"
is determined by the vocabulary analysis unit 1 to be “■”, rhaVeJ
, ranJ, and "bOOk", and information such as the decomposed words and their arrangement relationships is sent to the syntax analysis unit 2.

The syntactic analysis unit 2 applies information including phrase structure rules and information including word types read from the first storage means 3 and second storage means 4 to this input information. That is, from the word dictionary, rIJ =
rprom”, 川’1aVe”=”■”, ran”=
rart2”, rb.

It can be seen that the result is "Ok" - "nl". Also, from the above phrase structure rules, 'promJ-'npJ, r'art2
nlJ = '*npJ, and furthermore, 'V*
npJ: "vp", "np vp", "S", and the input sentence is analyzed as having the syntax shown in the parse tree of FIG. Note that in this case, this syntax is 'an b
The part "ook" is incorrect, and the message "The article a is used for singular nouns starting with a consonant" is added.

The syntax analysis device having the above configuration actually has the following features:
This can be obtained by the hardware configuration shown in FIG.

In Fig. 2, reference numeral 5 indicates a central processing unit (CPU).
, 6 is the main memory, 7 is the magnetic disk, 8 is the magnetic disk.
9 is a keyboard, and 9 is a display.

The CPU controls the main memory 6, magnetic disk 7, keyboard 8, and display 9 to execute programs to be described later, and is composed of an LSI or the like used in a central processing unit of a normal computer.

The main memory 6 temporarily stores programs and the like to be executed by the CPU, and is composed of a random access memory and the like.

The magnetic disk 7 is a storage means for storing necessary information, and is a program containing the contents of the first storage means 3, the second storage means 4, the vocabulary analysis section 1, and the syntax analysis section 2 in this embodiment. etc. are stored.

Further, the keyboard 8 is an input means for inputting sentences and other necessary input processing, and the display is a display means for displaying the contents inputted by the keyboard 8 and other necessary information.

As explained above, according to the syntax analysis device of this embodiment, it is possible to analyze an input sentence containing an error extremely smoothly, and also to identify in which part and how the error occurs. Can be clearly pointed out.

(Second Embodiment) In this embodiment, in addition to the "phrase structure rules" and "messages" in the grammar rule table stored in the first storage means 3 in the first embodiment, "error detection rules" and "Information to be passed to the top of the parse tree" is added, and "word information" is added in addition to "word" and "part of speech" to the word dictionary stored in the second storage means 4. The basic configuration is the same as that of the first embodiment, except that a part of the syntax analysis section 2 is changed. In the following explanation, only the parts characteristic of the second embodiment will be explained,
Description of parts common to the first embodiment will be omitted.

Figure 6 is an explanatory diagram of the grammar rule table of the second embodiment;
The figure is an explanatory diagram of a word dictionary according to the second embodiment.

In FIG. 6, the phrase structure rules and the meanings of the messages are the same as in the first embodiment.

The meaning of "information passed to the top of the parse tree" is as follows.

That is, for example, "information passed to the top of the parse tree" in FIG. 6, vp, n1nsyou=v, n1nsyouvp, te
nse=v,tense is 'VP
The personal name of J (verb phrase) is the same as the personal name of "■" (verb).

The tense of 'vp' (verb phrase) is 'V' (
It is the same as the tense of verb).

It means that.

Also, in the "condition part" of the "error detection rule", n, 5uu=s & n, gotou=v& a
rt, hyousou=a is the phrase structure 'nl) --> art n',
suu (number; singular or plural) of n (noun)
is S (singular), gotou (first letter) of n (noun) is V (vowel), and hyousou (superficial; distinction between a or an, etc.) of art (article) is a.
If , the phrase structure is incorrect.

It means that. Then, corresponding to the conditional part, the message ``The article a is used for singular nouns starting with a consonant.'' is added.

In addition, the same phrase structure rnp-->art n'' further includes other conditional parts, namely, n, 5uu=s & n, gotou=c& a
rt, hyousou=an is provided, but this indicates another form of error, such as suu C number of n (noun); distinction between singular or plural form>
is S (singular), gotou (first letter) of n (noun) is c (consonant), and h of art (article)
If yousou (distinction between surface layer a, an, etc.) is an, it is an incorrect seasonal structure.

It means that. In this case, the message is ``The article an is used for singular nouns that start with a vowel.

" is added.

In addition to the part of speech of each word, the word dictionary shown in FIG.
sou (surface layer; distinction of a or an, etc.), n1nsy
ou (person), tense (tense), 5uu
(singular or plural; S is singular) and goto
u (distinguishing whether the first letter is a vowel or a consonant;
V is a vowel and C is a consonant).

The syntactic analysis unit in the second embodiment (corresponding to the syntactic analysis unit 2 in the first embodiment) has a function of performing syntactic analysis according to the following procedure using the above-mentioned grammar rules and word dictionary.

FIG. 8 is a flow diagram of syntax analysis in the second embodiment. Hereinafter, the syntax analysis procedure of the second embodiment will be explained with reference to FIG. In FIG. 8, after an input sentence is decomposed into words by a vocabulary analysis unit (corresponding to vocabulary analysis unit 1 of the first embodiment), information including the decomposed words is analyzed by a syntactic analysis unit. The subsequent processing is shown below. Syntax analysis is started by the 5TART command, and the following processing is executed. In addition, each step of the following ■～■
This corresponds to the steps indicated by ■ to ■ in FIG.

■ Variable MS that holds information to be passed to the top of the analysis tree
Initialize G and proceed to step ■.

■ Check whether there is a rule applicable to the phrase structure of the input sentence among the phrase structure rules in the grammar rule table.

If YES, proceed to step (2); if NO, proceed to step FAIL, where the analysis fails and ends.

■ From this step, error detection processing begins.

A variable indicating the rule described in each condition part of the error detection rule in the grammar rule table corresponding to the applied phrase structure is set to 1, and the process proceeds to step (2).

■ Check whether there is an Nth rule in the condition part of the error detection rule. If YES, proceed to the next step (2); if NO, return to step (2) and repeat the process from (2) onwards.

■ Check whether the condition part of the Nth rule is true. If YES, proceed to the next step (2); if NO, proceed to step (2), then return to step (2), and repeat the subsequent processing.

■ Output the message of the Nth rule and proceed to the next step ■.

■ Increment N by 1, proceed to step ■, and repeat the subsequent processing.

■ Information passed to the top of the parse tree of the grammar rule table is MS
Set it to G and proceed to the next step (■).

■ Check whether the left side of the rule is rsuccess, 1. If YES, proceed to the next step of 5 success and end the analysis; if NO, return to step (2) and repeat the subsequent processing.

FIG. 9 shows an input sentence II containing an error.
This figure shows a syntactic analysis tree when ``book'' is analyzed using the above-described procedure. In FIG. 9, rhas
J and ranJ are grammatically incorrect, but by following the steps above, the phrase structure 'np-->
The first condition part rule rn, 5uu==s & n, got of the error discovery rule corresponding to "art n"
ou=v &art, hyousou=a" is applied, and the corresponding message "The article a is used for singular nouns starting with a consonant." is output.

On the other hand, for rhasJ, the phrase structure “S>np V
l) The first condition part rule rvp of the error detection rule corresponding to J, tense=present & n
p, n1nsyou: #vp, n1nsyou" is applied, and the corresponding message "The subject and verb persons do not match." is output. In addition, the phrase structure r
When applying the error discovery rule corresponding to 'VpJ 'nPJ', 'information passed to the top of the parse tree' corresponding to the phrase structure of 'VpJ 'nPJ' is taken into consideration.

In this way, the analysis is carried out smoothly, and at the same time, the erroneous part and the content of the error are clearly pointed out (as a message).

According to the second embodiment described above, in addition to the same advantages as the first embodiment, the following advantages are also obtained. That is, the first embodiment is basically an example in which errors are discovered simply by applying a pre-created erroneous phrase structure to the erroneous part of the input sentence, but all possible errors are When expressed in terms of phrase structure rules, the number of phrase structure rules becomes enormous. For example, for articles a, an, t
It is necessary to create a phrase structure for each distinction such as he, or for each inflected form of a verb or adjective, or for each singular or plural form of a noun.

Therefore, in order to store these phrase structure rules, a storage device with a large storage capacity is required, and the analysis time becomes long. Furthermore, using only phrase structure rules, e.g.
It is also difficult to detect mismatches between the subject and verb.

On the other hand, according to the second embodiment described above, "information to be passed to the top of the parse tree" and "error detection rule" are added corresponding to each phrase structure rule, and "word information"
As a result, the number of phrase structure rules itself can be significantly reduced, and mismatches in the subject and verb person can also be easily detected, making it possible to smoothly and accurately analyze input sentences containing errors. And it can be done quickly.

Although the grammar rules of the second embodiment described above are examples in which there are no incorrect phrase structure rules, it goes without saying that incorrect phrase structure rules may be included if necessary.

[Effects of the Invention] As detailed above, the present invention has the following advantages:
By including information about erroneous sentences that include incorrect phrase structure rules in addition to correct phrase structure rules, accurate analysis can be performed by applying information about erroneous sentences that include these erroneous phrase structure rules to sentences containing errors. This allows the process to be carried out smoothly and quickly.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram showing a schematic configuration of a parsing device according to a first embodiment of the present invention, FIG. 2 is a block diagram showing an example of an actual configuration of the device in FIG. 1, and FIG. FIG. 4 is an explanatory diagram of the word dictionary in FIG. 1; FIG. 5 is an explanatory diagram of the syntactic parsing tree during syntax analysis by the parsing device of FIGS. 1 and 2; FIG. 6 is an explanatory diagram of the grammar rule table of the second embodiment, FIG. 7 is an explanatory diagram of the word dictionary of the second embodiment, FIG. 8 is a flowchart of syntax analysis of the second embodiment, and FIG. FIG. 7 is an explanatory diagram of a parsing tree of a second embodiment. DESCRIPTION OF SYMBOLS 1... Vocabulary analysis unit, 2... Syntactic analysis unit, 3... First storage means, 4... Second storage means, 5... Central processing unit, 6... Main Memory, 7...Magnetic disk, 8...Keyboard, 9...Display.

Claims (1)

[Claims] A grammar comprising: a lexical analysis unit that decomposes an input sentence into words and outputs information including the decomposed words; and information including phrase structure rules created in advance according to the type of input sentence. a first storage means for storing a rule table; a second storage means for storing a word dictionary including information including types of each word according to the type of input sentence; and words decomposed from the vocabulary analysis section. , and reads out information including phrase structure rules and information including word types from the first storage means and second storage means, respectively, and includes these information and the words input from the vocabulary analysis section. a syntactic analysis unit that analyzes the syntax of the input sentence based on the information;
In the grammar rule table stored in the first storage means,
A syntactic analysis device characterized by including information regarding erroneous sentences including incorrect phrase structure rules in addition to correct phrase structure rules.