JPH04139580A - Keyword searching system - Google Patents

Abstract

PURPOSE:To retrieve the word from significance by executing a collation by containing similarity of each predicate and case of a semantic structure network of an inputted sentence and a semantic structure network of semantics of a keyword of a dictionary and outputting a coincident keyword. CONSTITUTION:A syntax analyzing part 1 generates a semantic structure network by executing a syntax analysis of an inputted input sentence, a semantic structure network 3 is a semantic structure network generated by executing a syntax analysis of semantics of a keyword of a dictionary, and a semantics collating part 4 executes a collation with the semantic structure network 3 generated from the dictionary with regard to the predicate and the case, and also, a similar one in the semantic structure network generated from the input sentence by the syntax analyzing part 1. In this case, the syntax analyzing part 1 executes the syntax analysis of the inputted input sentence and generates the semantic structure network, and with regard to its predicate and case, a collation is executed by containing similarity with the semantic structure network 3 generated from the dictionary, and the keyword of that which is subjected to matching is outputted. In such a way, a word of the dictionary can be retrieved from significance.

Description

[Detailed Description of the Invention] [Summary] Regarding a headword search system that searches for headwords based on their meanings, the semantic structure network of the input sentence and the semantic structure network of the meaning of the headword in a dictionary are used to search for predicates and cases. 1 [41) The purpose of this system is to provide a system that searches for words based on their meaning by performing matching including J and outputting the headwords that failed. A structure network, a syntactic analysis unit that parses the input sentence to create a semantic structure network, and sequentially compares the predicates and cases in the semantic structure network created by this syntactic analysis unit with the above semantic structure network. The configuration is configured to output the headwords of the matches.

[Industrial application field]

The present invention relates to a headword search system that searches for headwords based on their meaning. When creating sentences using a word processor or the like, it is desirable to search a dictionary for words that match the writer's image.

[Conventional technology]

The main purpose of conventional dictionary systems is to input pronunciations and search for orthographic information, semantic information, etc. of the word, and does not have a function to extract the word intended by the user based on its meaning.

However, when creating a sentence, I often couldn't find an appropriate St word that matched the image and had to make do with a very simple list of words.

There is also a similar word dictionary and a similar dictionary for searching for similar words. This is a higher class word (e.g. 1 presentation)
to lower class words (e.g. "show", "show", "
Although I was able to draw the fake J), I was far from being able to find the appropriate word from the above image.

(Problem to be solved by the invention) Conventional dictionary systems search for the kanji and meaning of the word based on the pronunciation of the word, or search for the kanji (orthography) of the word.
Therefore, you can only search for the pronunciation and meaning of the word, and even if you have an image (meaning) of the word you want to express, there is a problem that there is no way to search if you do not know the exact word itself. Ta. Furthermore, although there is the above-mentioned similar word dictionary, this dictionary divides words into certain categories and describes their superior/subordinate relationships, and there is a problem in that it is impossible to search for the word that the writer intends.

The present invention compares the semantic structure network of the input sentence with the semantic structure network of the meaning of the headword in the dictionary, including similarities between predicates and between cases, and outputs the matched headword, The purpose is to provide a system for searching words based on their meaning.

(Means for solving Llu) Means for solving the problem will be explained with reference to FIG.

In FIG. 1, a syntactic analysis unit 1 creates a semantic structure network by parsing an input sentence.

The semantic structure network 3 is a semantic structure network created by parsing the meanings of dictionary headwords.

The semantic collation unit 4 collates the predicates and cases-1 in the semantic structure network created from the input sentence by the syntactic analysis unit 1 and those that are similar to the case-1 with the bluish structure network 3 created from the dictionary. .

[Effect]

As shown in the first part, in the present invention, a syntactic analysis unit 1 parses an input sentence to create a semantic structure network, and the predicates and cases of this are compared with a semantic structure network 3 created from a dictionary. It performs matching including similar words and outputs the headwords of the matched words.

Therefore, 2. It becomes possible to search words (headwords) in a dictionary based on their meaning.

Embodiment] Next, the configuration and operation of an embodiment of the present invention will be sequentially explained in detail using FIGS. 1 to 6.

In FIG. 1, a syntactic analysis unit (input side) 1 creates a semantic structure network by parsing an input sentence input by a user. For example, regarding the user input sentence ``battle, lose, and run away'' in FIG. 3(a), word division and interlocking relationships are analyzed by syntactic analysis to create a semantic structure network in FIG. 3(b).

The syntactic analysis unit (dictionary side) 2 inputs the meaning of a word (headword) in the dictionary, parses the meaning, and creates a semantically structured non-sotowork 3. This created one is registered as the illustrated semantic structure network 3.

The semantic structure network (dictionary side) 3 is a network in which the meanings of words (headwords) in the dictionary are parsed and registered.

The semantic matching unit 4 uses the predicate of the semantic structure network created from the input sentence by the syntactic analysis unit (input side) (for example, the predicate ``defeated'' in the semantic structure network shown in FIG. 3(b)).
, "escape-1)", and cases (for example, the cases "subject body" and "objective case" in the semantic structure network in Figure 3 (b))
), and more similar ones (for example, Figure 6 (a) Similar predicates by searching predicate concept similar word data, Figure 6 (b) 849 superordinate concepts found by referring to the noun concept thesaurus) The purpose of this is to check whether the item falls under the following criteria.

Next, the entire operation of the configuration shown in FIG. 1 will be explained in detail in accordance with the order shown in the flowchart of FIG.

First, the creation of a semantic structure network on the dictionary side will be explained using FIG. 2(a).

(1) Extract the meaning of a word (headword) from the dictionary and input it to the syntactic analysis unit (dictionary side) 2 in Figure 1 (Figure 2 (a)
■).

+2+ The syntactic analysis unit (dictionary side) 2 into which the words of the dictionary are input in (1) uses syntactic analysis to analyze word division, interlocking relationships, etc., and creates a semantic structure network (Figure 2 (a)). O).

+31 (Register the semantic structure network created in step 21 as semantic structure network (dictionary side) 3 in Figure 1 (Figure 2 (a) 0).

By repeating the above steps, you can find the words (headwords) in the dictionary.
All of the meanings will be registered as semantic structure work (dictionary side) 3.

Next, the procedure for outputting a meaningful headword corresponding to a user input sentence will be explained in detail using FIG. 2(b).

il+ Inputs a user input sentence (for example, FIG. 3 (a) ``Fight and lose the fight and run away'') (FIG. 2 (b) ■).

+21 FII user input text (for example, Figure 3 (a)
A semantic structure network (for example, the semantic structure network in Figure 3 (b)) is parsed and a semantic structure network is created.
(Figure 2 (b) O).

[31 (For the predicates in the semantic structure network of the user input sentence created in step 21 (for example, the predicates of "defeat" and "run away" in the semantic structure network of FIG. 2 (b)), the flowchart in FIG. 2 (a) Repeat the matching with the predicate of the semantic structure network taken out from the semantic structure network (dictionary side) 3 created according to (described later using Figures 3 to 6), and if no matching is found, further refer to predicate concept similar word data as shown in Figure 6 (a) to check whether they are similar or not. If the cases are similar, move on to matching the next case.

(41 (For the semantic structure network that was matched by matching predicates (including similarities) in 31, cases (see Figure 6 (b) Noun concept thesaurus and including similar and similar Perform matching of (
Figure 2 (b) O). If it matches, proceed to the next matching degree calculation.

(Figure 2 (b)) Calculate the degree of matching for cases where predicates are matched using 5 + +31 and cases are matched using (4).
■) Output the headwords of words in descending order of matching degree. The degree of matching is ■ high when predicates match, low when matching is similar to or greater than, ■ high when all of multiple predicates match, low when some of them match, and ■ case subject. It is set in advance that case agreement is high and objective case agreement is low, and based on this criterion, the degree of agreement (matching degree) of the entire predicate is calculated.

As described above, when a user inputs a sentence expressing the meaning of a word that the user wants to know, for example, ``fight, lose, and run away'', the predicates and cases of the semantic structure network created by parsing this are created from the dictionary. Semantic structure 7
) Match the predicates of the work, including cases and similarities, and select words (headwords) from the dictionary in descending order of degree of matching, such as "defeat".
, "Rout", "Defeat Ko...", etc. This will be explained in detail below using Figures 1 through 6.

FIG. 3 shows a detailed illustration of the invention.

FIG. 3(a) shows an example of an input sentence/output sentence.

Here, the user input sentence in the upper row, ``fight, lose and run away'' is
This is an input sentence representing the meaning of a word that the user wants to know, and is input by the user into the syntactic analysis unit (input side) 1 in FIG.

The output sentences of the lower system are "defeat", "rout", "defeat"
is found and output (displayed) by the system of the present invention as a dictionary word (headword) that matches the meaning corresponding to the input sentence in the upper row.

FIG. 3(B) shows an example of a semantic structure network generated by parsing the user input sentence of FIG. 3(A) by the syntactic analysis unit 1 of FIG. This semantic structure network is centered around the predicates ``defeat'' and ``escape'' in the user input sentences, and expresses the subject, location, and object case (battle) in relation to each other using arrows as shown in the figure.

FIG. 4 shows a collation explanatory diagram of the present invention.

FIG. 4(a) shows an example of a semantic structure network of semantic matching input. This is the semantic structure network of the user question input on the left, the dictionary semantic structure network on the right,
In this case, first, we will compare the partial pattern of the predicate "yabureru" in the semantic structure network of the user question input shown on the left side of Figure 4 (b), and the right side. In this case, the predicate ``Yabureru'' and ``Kadaru-I'' are not equal, so the partial pattern matching (matching) of the predicate ``Kadaru'' in the semantic structure network of the dictionary shown in Figure 6, which will be described later, is performed. (a) Predicate concept g!! Check whether this predicate "obsession" is included in the group "Yabureru" in the A position word data, and since it is not included here, the -degree is set to None. Similarly, the fourth
Figure (a) Matching the predicate ``Yabureru'' in the semantic structure network of the user question input on the left with the predicate ``worry'' of the word ``carefree'' on the right, matching the predicate ``run away'' with the predicate ``stick to'', and the predicate `` A comparison is made between the predicate ``run away'' and the predicate ``worry,'' but none of them have a negative degree.

Next, Figure 4 (c) shows how the partial pattern of the semantic structure network of the user question input shown on the left side is compared with the predicate ``yabureru'' of the semantic structure network of the dictionary entry word ``defeat'' shown on the right side. In this case, the predicate ``yabureru'' matches, so we next check the cases. In this case, all possible cases are objective cases, and their contents (noun concepts) are both Part because it is a “battle”
The degree of matching of the entire predicate is calculated.

At this time, if the content of the cases is different, see Figure 6 (
b) Check the noun concept thesaurus to see if there is a superior-subordinate relationship, and if so, consider it a partial match.

FIG. 5 shows a detailed explanatory diagram of the present invention, and FIG.
An example of matching the semantic structure network of user question input on the left side is shown.

Figure 5 (a) shows the semantic structure network for the dictionary entry 1: defeat. The predicate of the semantic structure network of the word "defeat" is "
It was determined that the two matched because the words "to break down" and "to run away" matched, and the contents of the case and object case, "fighting and fighting," matched, and the subject's physique matched that of a human.

Figure 5 (B) shows the semantic structure network of the dictionary entry word ``Routou''. Similar to Figure 5 (B), this shows the content of the predicates ``Yabureru'', ``Escape'', case, and objective case. [Battle-1, the main physique matched that of a human, so it was determined that the two matched.

Figure 5 (c) shows the semantic structure network of the dictionary entry word ``defeat''.
Because the content of the case and object case, ``battle'', and the subject's physique matched that of a human, it was determined that the two partially matched.

The importance of the case is set in advance, for example, the subject of "defeat" is given high importance and the objective case is given low importance, and based on this criterion, the degree of agreement of the entire predicate is calculated. .

As mentioned above, for example, the predicates ``break'' and ``run away'' in the semantic structure network of the user question input on the left side of FIG. Furthermore, by searching the predicate concept synonym data in Figure 6 (a) and deeming it to be a match if it is similar, then by matching the case as well and checking l14EJ, for example, the meaning on the left side of Figure 4 (a) can be found. It is now possible to output such entries as dictionary headwords that match the structural network: Figure 5 (a) "Defeat", Figure 5 (B) "Rout", Figure 5 (C) [Defeat - 1], etc. . At this time, the predicates are output in descending order of the degree of matching of the entire predicate.

FIG. 6 shows a detailed illustration of the invention.

FIG. 6(a) shows an example of predicate concept similar word data. For example, the predicates ``meet'', ``meetj'', ``face-to-face'', ``face-to-face'', ``meeting'', ``conference'', etc. that correspond to the predicate ``meeting'' are registered in advance. As mentioned above, this predicate concept similar word data includes, for example, predicate 1 of the semantic structure network of the user question input on the left side of Fig. 4 (a), such as ``Yabureruko'', ``Kureru-I'', etc., and the semantic structure of the dictionary entry word. When matching with the predicate of the network, if there is no match, it is searched to see if the predicate of the headword of the dictionary to be queried exists in the group of predicates of the semantic structure network of the user question input, and if it exists This is to judge whether the predicate is similar to .

FIG. 6(b) shows an example of a noun concept thesaurus, in which similar noun concepts with case content, for example, "human being" are registered in advance in a hierarchical structure. This noun concept thesaurus is used for case matching, and when cases do not match, the noun concept thesaurus is used to check the contents of the case to be matched to determine whether or not they are similar.

〔Effect of the invention〕

As explained above, according to the present invention, the predicates and cases of the semantic structure network created from the input sentence are compared with the predicates and cases of the semantic structure network 3 of the meaning of the entry word of the dictionary, and further similar Since the system is configured to also collate and output the matching headword, it is possible to input the meaning and search for a headword (word) with this meaning from the dictionary and output it. As a result, in a system that searches words based on their meaning, it is possible to utilize syntactic analysis and semantic matching of natural language processing, and to improve the difficulty of dictionary lookup on the part of the writer of a document.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram of one embodiment of the present invention. FIG. 2 is a flowchart explaining the operation of the present invention.
4 is a detailed explanatory diagram of the present invention; FIG. 4 is a collated explanatory diagram of the present invention; and FIG. 6 is a similar explanatory diagram of the present invention. In the figure, 1: Syntactic analysis unit (input side) 2: Syntactic analysis unit (dictionary analysis 3: Semantic structure network (dictionary side): Semantic matching unit

Claims (2)

[Claims] (1) In a headword search system that searches headwords based on their meaning, a semantic structure network (3) is created by parsing the meanings of dictionary headwords, and a semantic structure network is created by parsing input sentences. The syntactic analysis unit (1) creates a syntactic analysis unit (1), and the predicates and cases in the semantic structure network created by this syntactic analysis unit (1) are sequentially compared with the above semantic structure network (3), and the headwords of the matches are A headword search system characterized by being configured to output. (2) In the above-mentioned comparison, the predicate concept similar word data and the noun concept thesaurus registered in advance are searched, and similar predicates and cases are also compared. Headword search system described in section 1).