JPH01189767A - Parallel sentence recognition system - Google Patents

Abstract

PURPOSE:To reduce comparison circuits and to efficiently recognize a sentence form by setting the finite number of positions, analyzing a syntax and checking only one position to a candidate throughout plural sentence forms in parallel. CONSTITUTION:An input sentence is initially converted to a slot table represented by the finite number of the positions through a sentence form extracting means and a slot table forming means or the like in the analysis of a morphological element. Then, in the analysis of the syntax, a candidate obtained through a sentence form constituting element candidate detecting means and a slot table operating means or the like is inserted into one position in the slot table and compared with the slot condition of all the probable sentence forms. Consequently, they coincide, it is recognized as the sentence form constituting element and the coinciding slot is buried. Subsequently, these processes are repeated to produce no sentence form constituting element, and then, the sentence from having no idle slot but having the largest number of buried slots is recognized as the sentence form of the declinable parts of speech of the input sentence. Thereby, the number of times comparison is comparatively reduced to efficiently recognize the sentence form.

Description

[Detailed description of the invention] (Industrial application field) The present invention relates to a sentence pattern recognition method.

(Prior Art) The following two methods are known as conventional sentence pattern recognition methods. In one method, one of the possible sentence patterns of a predicate in an input sentence is assumed to be the sentence pattern of the predicate, and the input sentence is analyzed based on that assumption. When a parsing failure is detected due to the failure of a sentence structure component that should appear, backtrack and select one of the other sentence patterns, and repeat this process until a sentence pattern in which no failure is detected is detected. The sentence pattern that is not recognized is recognized as the sentence pattern of the predicate of the input sentence.

In another method, a table is prepared in which all possible sentence patterns of the predicate in the input sentence are expressed in parallel as rows, and their constituent grammatical cases are expressed as conditions for thrower 1. Then, the features of the candidate sentence pattern components in the input sentence are
Compare all slots 1 to conditions of a certain sentence pattern in order until they match, fill in the matching slots, repeat this operation in order from the first to the last sentence pattern, and fill in the empty slots I.
The sentence pattern that matches the most without ... is recognized as the sentence pattern of the input sentence.

(Problem to be solved by the invention) The former sentence pattern recognition method described above depends on the sentence pattern recognized from among possible sentence patterns or on the assumed order, so it gives priority to the sentence pattern with the largest number of elements. If you try to analyze it after that, you will end up backtracking a lot because the elements with the most number of elements are not necessarily the most frequent ones. Furthermore, if analysis is attempted in order of frequency, there is a problem in that elements that should originally be recognized as sentence pattern constituents are not recognized and the analysis is successful.

Although the latter method does not have the same problems as the former, there is a problem in that the candidate's background and the conditions for thrower 1 are compared extremely many times. - In table operations, the number of comparisons is roughly several times the number of sentence patterns,
There was a problem that the processing time and man-hours were extremely long.

An object of the present invention is to provide a method for eliminating such drawbacks, reducing the number of comparisons, and efficiently recognizing sentence patterns.

(Means for Solving the Problems) In the sentence pattern recognition method of the present invention, grammatical cases constituting a sentence pattern are expressed as conditions for thrower 1 in a finite number of positions organized by positional relationships with predicates and surface grammatical cases. death,
It is characterized by checking only one position of candidates in parallel across multiple sentence patterns.

(Operation) In the present invention, by setting a finite number of positions, comparison with the candidate's background is unnecessary for positions other than specific positions, and the number of comparisons is reduced to a fraction of that of the prior art.
This leads to a reduction in processing man-hours and an improvement in processing speed.

(Example) Next, an example of the present invention will be described with reference to FIGS. 1 to 3.

FIG. 1 is a flowchart 1 of a natural language analysis method using the present invention. The input sentence is divided into words, which are the smallest units of meaning, using a dictionary through morphological analysis.
Possible parts of speech are assigned to each word. For a word that may be a predicate, all possible sentence patterns of the predicate are extracted. The extracted sentence pattern information is converted into a slot table expressed by a finite number of positions using a slot table template. The morphological analysis results are summarized in the syntactic analysis to a size that can be used as the constituent elements of a sentence pattern, and the features are inserted into one position in the slot table using the slot table manipulation function, and all possible It is compared with the sentence pattern slot condition. If there is a match, the candidate is recognized as a sentence pattern component, and only sentence patterns with matching slots are left as eligible, and the matching slots are filled. This process is repeated as long as sentence structure component candidates can appear, or if they cannot appear, the sentence pattern with the largest number of filled slots without empty slots is determined to be the sentence pattern of the predicate in the input sentence. Certified.

FIG. 2 shows an example of conversion from sentence pattern information of predicates to a slot table. Figure (a) is an example of English, and figure (a) is an example of English.
b) shows an example in Japanese. In the English example, the position is set by the relative position to the predicate. The subject (SUB) is set in position 1, and the complement (COMP), direct object (DO'B), and indirect object ('JOB) immediately after the predicate are set in position 2. The elements further back are in position 3 and position 4 in order.
is set to

In the Japanese example, positions are determined by surface grammatical cases. Grammatical cases in which the particle becomes ``ga'' after the sentence pattern undergoes a certain shift are set in position GA, and the particle becomes 1.
” is set to position wo.

As shown in Figure 3 <a->, in English the positions are selected from position 1 to position 2.
, position 3, and so on. In Japanese, as shown in FIG. 3(b), a position is selected by looking at the spelling of the particle of a candidate sentence structure element.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing an example of a sentence analysis method using the sentence parallel recognition method of the present invention, and FIG. 2 (a>, (b))
FIGS. 3(a) and 3(b), an example of converting a slot table using positions from sentence pattern information of predicates, are examples of how to select positions on the slot table.

Claims (1)

[Claims] In a natural language processing system that recognizes sentence patterns, for a predicate in an input sentence, all possible sentence patterns of the predicate are expressed in parallel as lines, and the constituent grammatical cases are expressed in terms of their positional relationship with the predicate and surface layer. Prepare a table expressed as slot conditions in a finite number of positions organized by grammatical case,
For candidates of sentence pattern constituents in the input sentence, select one position in the table and compare the features of the candidate with the slot conditions at the position of all possible sentence patterns, and if there is a slot that matches the condition. A sentence pattern recognition method characterized in that a plurality of sentence patterns are processed in parallel by making only sentence patterns having the slot eligible.