JPH0583950B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a method for determining semantic relationships between words, and in particular, expresses the semantic relationships between words as tree-structured data, and determines the meaning of the inclusion relationship between words. The present invention relates to an inter-word semantic relationship determination method for quickly determining the semantic inclusion relationship between words in a natural language processing system that performs semantic analysis of natural language data to be analyzed.

[Conventional technology]

In natural language processing systems such as machine translation, tree-structured data is generated in order to analyze syntax and express the semantic connections between words, the meaning of the syntax, etc. This tree structure data associates a word with each node, links each node, and expresses the semantic relationship between each word by the connection relationship of each node. In the next stage of semantic analysis of the semantic relationships between words, the tree structure data is searched to determine the semantic relationships between words. Search methods for analyzing tree-structured data include horizontal search methods, vertical search methods, and depth-first search methods. Further, as an analysis process for tree structure data, there is a method of adding information of upper nodes of the tree to lower nodes. However, in either process, each node is sequentially searched in the tree structure data analysis process, which requires a large amount of processing time in proportion to the number of nodes in the tree.

[Problem that the invention seeks to solve]

By the way, when performing semantic analysis of natural language data in which the semantic relationships between words are expressed as tree-structured data in a natural language processing system, the tree-structured data is analyzed to determine the semantic relationships. In the tree structure data analysis process, the nodes of the tree structure data are sequentially searched and the data of each node is analyzed. This process follows the pointer of the memory cell corresponding to the node and sequentially refers to the contents of the memory cell corresponding to the next node, so when the number of words to be handled increases, Processing time increases rapidly. Therefore, as the number of words to be handled increases, processing time increases.
There was a problem in that it was extremely difficult to perform semantic analysis processing in real time.

The present invention has been made to solve the above problems.

An object of the present invention is to provide a method for determining semantic relationships between words that can quickly determine semantic relationships between words expressed in tree-structured data in a natural language processing system.

The above and other objects and novel features of the present invention will become apparent from the description of this specification and the accompanying drawings.

[Means for solving problems]

In order to achieve the above object, in the present invention,
In a natural language processing system that associates each word with each node of a tree, links between each node, expresses the semantic relationship between words as tree-structured data, and performs semantic analysis of the semantic inclusion relationship between words. , Search each node of the tree structure data in advance by depth-first search, find the number of lower nodes indicating the range of nodes included under each node, and calculate this number of lower nodes according to the search order of each node. The present invention is characterized in that it includes a stored meaning table, and compares the number of subordinate nodes of each node stored in the meaning table with the search order of each node to determine the inclusion relationship of meanings between words.

[Effect]

According to the means, each word is associated with each node of a tree, each node is linked, and the tree-structured data expressing the semantic relationship between the words is searched in advance by a depth-first search. A meaning table is provided in which the number of lower nodes indicating the range of nodes included under each node is determined by searching each node, and the number of lower nodes is stored in correspondence with the search order of each node. This meaning table stores, as table data, the number of lower nodes included below each node in the tree structure data, corresponding to the search order of each node in the depth-first search. Therefore, according to this table data, the data on the number of lower nodes of a node in a certain rank is obtained from the table data in descending order from the table data of the relevant rank, and stores the data (number of lower nodes) of nodes in an inclusion relationship. It shows the number up to. Therefore, the inclusion relationship of each node can be determined by numerically comparing the magnitude relationship of the rankings of table data with the magnitude relationship of its search rankings. In other words, the target node to check the relationship with a certain node is
If the rank is higher by comparing the magnitude relationship of the ranks, and if the rank difference is smaller than the number of lower nodes by comparing the rank difference and the number of lower nodes of the target node stored corresponding to the rank, then , the target node is determined to be an upper node.

In this way, from tree-structured data expressing the semantic relationships between words, table data is generated that stores the number of subordinate nodes included below each node in accordance with the search order in depth-first search starting from the root node. Therefore, the inclusion relationship between words can be determined by referencing table data and performing comparison operations.
Determine the semantic relationships between words.

This makes it possible to avoid searching for tree-structured data for judgment in the process of determining semantic inclusion relationships, and quickly evaluates semantic inclusion relationships between words by referencing table data and performing two comparison operations. can be determined.

〔Example〕

Hereinafter, one embodiment of the present invention will be specifically described using the drawings.

FIG. 1 is a block diagram showing a schematic configuration of a processing system according to an embodiment of the present invention. In FIG. 1, 1 is a keyboard, 2 is a lexical analyzer,
3 is a word dictionary, 4 is a semantic analysis device, 5 is a semantic dictionary, 6 is an inter-word semantic relationship determination device, and 7 is a display device. The lexical analysis device 2 refers to the word dictionary 3 from the character string data input from the keyboard 1, and
Separate words. Each word is matched with a meaning dictionary 5 by a meaning analysis device 4, and the meaning of each word is analyzed. In the semantic analysis device 4, the semantic dictionary 5
With reference to , the part of speech of a word is determined, the modification relationship between an adjective and a noun is determined, and the dependency relationship between words is determined. That is, determination processing is performed on parts of speech with meanings of words, modification relationships between words from part of speech data, inclusion relationships between words, and the like.

In this way, the semantic analysis device 4 refers to the part of speech and semantic information of each word stored in the semantic dictionary 5, performs semantic analysis between each word, and determines the meaning between the words.
Determine the meaning of a sentence. In order to perform semantic analysis between each word, the semantic analysis device 4 uses an inter-word semantic relationship determining device 6 to analyze the hierarchical relationship between words,
Determines the inclusion relationship between words.

The semantic inclusion relationship between words is expressed, for example, as tree-structured data as shown in FIG. This tree structure data associates each word with each node of the tree, links each node, and expresses the semantic relationship between words. In the tree structure data in Figure 2, the word "meeting" is placed at the root node 11.
, the next lower node 12 linked to the root node 11 is associated with the word "board meeting", and the other next lower node 16 linked to the root node 11 is associated with the word "manager's meeting". Match the words. Furthermore, the next lower node 13 linked to the "executives meeting" node 12 is associated with the word "management policy meeting." In this way, in the tree structure data, each word is associated with each node of the tree based on the semantic relationships between the words, and the semantic relationships between the words are expressed by links between the nodes.

Therefore, in order to determine the semantic relationship of each input word, in which node position of the tree structure data the input word is located by referring to the tree structure data that expresses the semantic relationship of each word. By performing this analysis, the semantic relationships will be determined. This judgment process is performed here by the semantic analysis device 4. First, the semantic analysis device 4 analyzes the tree-structured data expressing the semantic relationship between words using the following procedure. Generate array data and generate a semantic table.

First, each node of the tree structure data is searched by depth-first search. The search order of this depth-first search is from the top node (root node) of the tree-structured data to the lower nodes, but if there are lower nodes, the lower nodes are given priority. The search is performed in the order in which the nodes on the left are searched first. For example, specifically, in the case of searching the tree-structured data in Figure 2, the search starts sequentially from the first root node "Meeting" node 11, and the nodes on the left are searched first, and then the lower nodes are searched. Proceed with the search in a search order that gives priority to the search. Therefore, the next search order is "Board Meeting" at the bottom left.
Node 12, then the "Management Policy Meeting" node further down
13.Next, the "First Half Management Policy Meeting" on the lower left
From Node 14, first, proceed sequentially from the upper node to the lower left. Then, when it reaches a node that has no lower nodes, it returns to the node one level above it,
Go to the next lower node to the right. In other words, when reaching the "first half management policy meeting" node 14, which is a node that has no lower nodes, it returns from this node to the "management policy meeting" node 13, which is a node one level higher, and
Next, the next lower node to the right is “Second Half Management Policy Meeting”
Proceed to node 15. When there are no unsearched lower nodes, the search is repeated by returning to the next higher node and proceeding to the lower node one level to the right. In this way,
When searching the tree structure data in Figure 2 sequentially, the search order is 1 "meeting" → 2 "executive meeting" → 3
"Management Policy Meeting" → 4 "First Half Management Policy Meeting" → 5
"Second half management policy meeting" → 6 "Manager's meeting" → 7 "Accounting manager's meeting" → 8 "Budget council" → 9 "Design manager's meeting" → 10 "Research meeting" → 11 "Process meeting"

By searching each node of this tree-structured data, the number of nodes included in the lower order of the node is determined for each node.

Next, as a result of the tree-structured data search process, each word and its search order are associated as entry numbers for each word that corresponds to each node in the tree-structured data, as shown in Figure 3. word table 3
Generates 0. In this word table 30, word meaning data such as the part of speech of each word is added to each word as necessary. This word table 30
When determining the semantic relationship between words, the entry number is determined from the input word.

Next, from the data on the number of nodes included below each node for each node found by searching the tree structure data, we calculate the number of self-nodes in the number of nodes included below each node, as shown in Figure 4. The meaning table 40 is generated which corresponds to the entry number corresponding to the search order of each node, with the number added by 1 as the number of lower nodes. Specifically, in this meaning table 40, the number of lower nodes indicating the range of nodes included under each node, including the self node, is stored in correspondence with the entry number corresponding to the search order. For example, in this example of tree structure data, the "meeting" node whose search order is 1 has 11 subordinate nodes, including itself, so in the meaning table 40, it corresponds to entry number 1. Therefore, the number of lower nodes is 11. In addition, in the "board meeting" node whose search order is 2, the number of subordinate nodes is 4, including itself, so in the meaning table 40, corresponding to the entry number 2, the number of subordinate nodes is 4. Stored. Similarly, the number of lower nodes of the "Management Policy Meeting" node with a search order of 3 is 3, so
The number of lower nodes is 3 corresponding to entry number 3. “First half management policy meeting” with a search ranking of 4
The "first half management policy meeting" node with a search order of 5 is the lowest node and the number of lower nodes is 1, so the number of lower nodes is 1 corresponding to entry number 4, and the number of lower nodes is 1 corresponding to entry number 5. Correspondingly, 1 is stored as the number of lower nodes. In this way, a meaning table 40 is generated in which the number of lower nodes is stored for the entry number corresponding to the search order.

In this way, the word table 30 and the meaning table 40 are generated in advance, and then, using the word table 30 and the meaning table 40, the inter-word semantic relation determining device 6 determines the meaning between the input words. Performs inclusion relationship determination processing.

FIG. 5 is a flowchart showing a process for determining the inclusion relationship of meanings between words. The process of determining the semantic inclusion relationship between input words will be described with reference to FIG.

When words A and B are input for which the inclusion relationship of meanings between words should be determined, in step 51, the word table 30 (FIG. 3) and the meaning table 40 (FIG. 4) are referred to. First, match the word A and word B with the word data in the word table 30, find the entry number E(A) for word A, and the entry number E(B) for word B. Based on the entry numbers E(A) and E(B), the corresponding numbers N(A) and N(B) of lower nodes are read from the meaning table 40, respectively. Next step 52
Then, compare entry numbers E(A) and E(B), and find that E(A)=
In the case of E(B), the process proceeds to step 53, where it is determined that A=B. That is, there is no inclusive relationship between word A and word B, and it is determined that they are of the same rank. In this case, word A and word B are the same word, and the same entry number is read from the word table 30.

On the other hand, in step 52, if E(A)<E(B), the process advances to step 54, where the difference in entry numbers [E
(B)-E(A)] is set as a variable D, and then in step 55, this variable D is compared with the number N(A) of lower nodes corresponding to the word A. In this comparison, if N(A)>D, the process proceeds to step 56 and it is determined that A⊃B. In other words, word A is in a relationship that includes word B,
Word A is determined to be higher in rank than word B. If N(A)>D is not true, the process proceeds to step 57, where it is determined that there is no relationship. In this case, word A and word B
There is no containment relationship between.

Further, in step 52, if E(A)>E(B), the process advances to step 58, and the difference in entry numbers [E
(A)-E(B)] is set as variable D, and then in step 59, this variable D is compared with the number of lower nodes N(B) corresponding to word B. In this comparison, if N(B)>D, the process proceeds to step 60 and it is determined that B⊃A. That is, word A is included in word B, and word A is determined to be subordinate to word B.
Also, if N(B) > D, proceed to step 57,
Determined to be unrelated. In this case, there is no inclusion relationship between word A and word B.

For example, the executive meeting 12 and manager meeting 16 in Figure 2
When considering the semantic relationship with , the entry number for the executive board meeting 12 = 2, the number of lower nodes = 4, and the entry number for the manager's meeting 16 = 6. In this case, only the number of lower nodes of the higher ranking entry number is required for comparison.

In step 54, the difference between the lower and higher entry numbers is calculated. Then, the result is “4=6
-2".

Next, in step 55, when the difference "4" is compared with the number of lower nodes of the board meeting 12 having a higher entry number rank, 4>4, so it is determined that there is no semantic relationship between the two.

Here, the reason why the rank difference is calculated in step 54 is to find out how many subordinate nodes the entry with a higher rank to be compared has under its own node, that is, to what extent. be.

In this example, board meeting 12 has entry number=
2. Since the number of lower nodes = 4, it has nodes with entry numbers 2, 3, 4, and 5. In step 55, it is checked whether the entry number "6" of the manager's meeting 16 is included among the entries with entry numbers 2, 3, 4, and 5. As a result, since they are not included, it is determined that there is no semantic relationship between the two.

In this way, the semantic inclusion relationship between words is
A high-speed determination can be made by reading the table data and comparing the numerical values in the magnitude relationship.

Although the present invention has been specifically described above based on Examples, it goes without saying that the present invention is not limited to the above-mentioned Examples, and can be modified in various ways without departing from the gist thereof.

〔Effect of the invention〕

As described above, according to the present invention, the process of determining the semantic inclusion relationship between two words can be performed by referring to table data and performing two numerical comparison operations, so that the process can be performed at high speed. . Even if the number of target words increases, the processing time does not increase, and the process of determining the semantic inclusion relationship between words can be performed in real time.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a schematic configuration of a processing system according to an embodiment of the present invention, FIG. 2 is a diagram illustrating tree-structured data expressing semantic relationships between words, and FIG. FIG. 4 is a diagram showing a word table in which the words of each node of the structural data are associated with their search rankings as entry numbers; FIG. The figure is a flowchart illustrating a process for determining a semantic inclusion relationship between words. In the figure, 1...keyboard, 2...lexical analyzer, 3...word dictionary, 4...semantic analyzer, 5...
...Semantic dictionary, 6...Semantic relationship determination device between words, 7
...Display device.

Claims (1)

[Claims] 1. Each word is associated with each node of a tree, each node is linked, the semantic relationship between words is expressed as tree structure data, and the semantic analysis of the inclusion relationship of meanings between words is performed. In a natural language processing system, each node of tree-structured data is searched in advance by depth-first search, the number of lower nodes indicating the range of nodes included under each node is determined, and this number of lower nodes is calculated as the number of lower nodes of each node. It has a meaning table stored in correspondence with the search order, and compares the number of subordinate nodes of each node stored in the meaning table with the search order of each node to determine the inclusion relationship of meanings between words. A method for determining semantic relationships between words.