JPH03137771A - Syntax analyzer - Google Patents

Abstract

PURPOSE:To improve the efficiency of processing and to economize a memory by controlling the progression of grammatical construction based on the depth of stacks. CONSTITUTION:One word is read in from the start of a sentence by a word reading-in part 1 with respect to a word string, and a dictionary part 5 is used to perform the dictionary retrieval by a dictionary retrieving part 2. Thereafter, a phrase structure rule part 6 is used to apply phrase structure rules to results of the dictionary retrieval by a rule applying part 3, and results are preserved in a stack area part 8. At this time, each stack of the stack area part is checked by a stack check part 7, and stacks which do not accord with conditions are erased.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a parsing device for natural language processing.

Conventional technology Conventionally, in a syntactic analysis device, for an input character string, one word is read from the beginning of the sentence, a dictionary search is performed using the words; The bottom-up syntax analysis method is well known.

In such a syntax analysis method, there are generally two methods for finding a solution.

■) Output all syntax analysis results (Parallel Pa
rsing) 2) Output only one syntax analysis result (Seq
Problems to be Solved by the Invention There are problems when using method 1 of the above two methods. The problem is that in order to come up with all possible solutions, it is necessary to have (y) intermediate analysis results that are considered possible no matter how unnatural.For this reason, if the sentence becomes long, if there is,
This results in an explosive increase in the number of intermediate analysis results that must be retained for that sentence (this is referred to as a "combination explosion").

Means for Solving the Problems Therefore, in order to solve these problems, the present invention reads each word from the beginning of a sentence, performs a dictionary search using a dictionary, and uses phrase structure rules to find the beginning of the sentence. In a syntax analyzer that performs syntax analysis from the bottom up from the beginning to the end of the sentence,
A stack area section for storing intermediate results of the syntax analysis in a stack format is provided, and a stack inspection section is provided for controlling the progress of the syntax based on the depth of the stack.

Therefore, it is possible to use the stack area section to express the intermediate results of syntax analysis as a stack consisting of subtrees, and use the stack inspection section to erase unnatural intermediate results of the analysis based on the depth of the stack. This makes it possible to improve processing efficiency by discarding useless or unnatural stacks based on stack depth.

Example An embodiment of the present invention will be described based on the drawings.

FIG. 1 shows the basic configuration of this device. In this case, the word reading unit 1 into which the word string is input, the dictionary search unit 2, the rule application unit 3, and the final judgment unit 4 perform bottom-up syntax analysis using the dictionary unit 5 and the phrase structure rule unit 6. This is a well-known configuration, and this embodiment is characterized by the provision of a stack area section 7 and a stack inspection section 8. The stack area section 7 has the function of storing intermediate results of syntax analysis in a stack format, and the stack inspection section 8 has the function of controlling the progress of the syntax based on the depth of the stack. Note that # indicates the flow of control, and → indicates the flow of data.

Furthermore, FIG. 2 shows the basic processing flow of this apparatus, and the overall flow of FIG. 1 will now be described based on this diagram.

A word reading unit 1 reads one word from the beginning of a sentence into a word string, and a dictionary search unit 2 uses a dictionary unit 5 to perform a dictionary search for the word. Thereafter, the phrase structure rule is applied to the result of the dictionary search by the rule applying unit 3 using the phrase structure rule unit 6, and the result is transferred to the stack area unit 8.
Save to. At this time, the stack inspection section 7 inspects each stack in the stack area section 7, and stacks that do not meet the conditions are erased. When all stacks have been inspected, the process returns to the initial state and starts processing the next word. When a word is finally found in this way, a final judgment process is performed, that is, a check is made to see if it is a sentence, and the series of operations ends.

FIG. 5 shows an example of phrase structure rules, which are held in the phrase structure rule section 6. In addition, Figure 6 shows examples of words, including at least the ``notation'' and ``part of speech.''
These are held in the dictionary section 5.

First, the processing in the stack inspection section 8 will be described.

3 and 4 show the state of the algorithm in the star check section 8 at t2.

In this case, Figure 3 shows that the stack depth
This shows an example of erasing a stack larger than max). In addition, FIG. 4 shows a stano stack that is deeper than the shallowest stack in the stack area by a certain number of points (min 10),
This shows an example of erasing the data.

Here, we will examine stack processing when a "sentence" as shown below is input.

Heading home.” Suppose there is a sentence.

At this time, the stack state during the analysis is shown in FIG. Note that → indicates dictionary lookup, and - indicates rule application. Further, the symbol above → corresponds to the rule number shown in the phrase structure rules of FIG. 5 mentioned above. Therefore, as can be seen from FIG. 7, it can be seen that in this embodiment, even a deep stack can have only six stages.

Let us now compare the sentence analysis process described above with the conventional example shown in FIG. If no limit is placed on the stack depth, a stack of as many as 10 levels as shown on the left side of FIG. 8 may occur. By applying phrase structure rules, this stack degenerates to nine stages as shown on the right side, but ultimately does not reach S, resulting in a so-called useless stack. This can be seen from the human cognitive process, where it becomes inconvenient to record and peel subtrees when the number of subtrees exceeds a certain level, and a stack that is too deep can result in an unnatural state. become. Therefore, for this reason, it is meaningful to impose restrictions on the depth of the stack as in this embodiment.

Effects of the Invention The present invention provides a syntax analysis device that reads each word from the beginning of a sentence, performs a dictionary search using a dictionary, and performs syntax analysis from the beginning to the end of the sentence from the bottom up using phrase structure rules. A stack area section is provided to store intermediate results of syntax analysis in a stack format, and a stack inspection section is provided to control the progress of the syntax based on the depth of the stack. It is now possible to express the intermediate results of analysis as a stack consisting of subtrees, and use the stack inspection unit to erase unnatural intermediate results based on the depth of the stack. By discarding unnecessary or unnatural stacks based on this, it is possible to improve processing efficiency and save memory.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing an embodiment of the present invention, Figure 2 is a flowchart showing the basic overall flow, Figures 3 and 4 are stack inspection algorithms, and Figure 5 is phrase structure rules. FIG. 6 is an explanatory diagram showing an example of words in a dictionary. FIG. 7 is an explanatory diagram showing a state in the middle of stack analysis. FIG. 8 is an explanatory diagram showing a conventional stack state. It is. 1. Word reading section, 2. Dictionary search section, 3. Rule application section, 4. Final judgment section, 5. Dictionary section, 6. Phrase structure rule section, 7. Stack area section, 8.・Stack Inspection Department 3'' Figure 12, Applicant Rico Co., Ltd. Figure 7 Heading to Professor's House

Claims (1)

[Claims] In a syntactic analysis device that reads each word from the beginning of a sentence, performs a dictionary search using a dictionary, and performs syntactic analysis from the beginning to the end of the sentence from the bottom up using phrase structure rules, the intermediate results of the syntactic analysis are A syntax analysis device, comprising: a stack area section for storing data in a stack format; and a stack inspection section for controlling the progress of the syntax based on the depth of the stack.