JPH0567136A - Natural language interpreting method - Google Patents

Abstract

PURPOSE:To develop a simplified natural language interface. CONSTITUTION:An inquiry text is interpreted as meaning representing that (show the attribute value of a certain attribute in a certain attribute table) by expressing the target field of inquiry as the set of the attribute tables that is the set of pairs of the attribute name and the attribute value. Firstly, each word in the inquiry text is classified to the attribute name, the attribute value, and others, and the attribute name is set as corresponding attribute name, and it is stored as a conditional attribute value pair group setting the attribute and that attribute as a pair (17). Thence, all the attribute tables in which the attribute name of the conditional attribute value pair group and the pair of attribute values exist are selected, and the attribute values of the attribute names existing in the corresponding attribute names of those attribute tables are outputted as answers.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language sentence analysis method for realizing a query in a practical natural language in a limited area such as a natural language interface to a database.

[0002]

2. Description of the Related Art Currently, many natural language interface systems have been created mainly in the United States, but only a few have reached a practical level. There are two problems with the conventional natural language interface system.

Complexity of the target world Natural language systems must deal with knowledge with complex structures. When the knowledge structure becomes complicated, the semantic structure that must be extracted from the input sentence also becomes complicated. However, it is difficult to derive an accurate semantic structure from input sentences that are not always clear.

Invulnerability to non-grammaticalities In reality, people are often non-grammatical. However, many natural language interface systems do not handle non-grammatical sentences well.

The above problems seem to reduce the practicality of the conventional natural language interface system.
(References: Willenskiy et al., Ucy-A Progress Report, Wilensky R. et.a.
l. , "UC-A Progress Report
t ", Rep. UCB / CSD87 / 303, 198.
6).

The object of the present invention is to express the target world as a set of pairs of attribute names and attribute values, and also to abandon the concept of grammaticality itself in order to process even non-grammatical sentences. It is to provide a natural language interpretation method that solves the above problem.

[0007]

According to a first aspect of the present invention, an inquiry is made in a method of referring to an attribute value in an attribute table, which is a set of an attribute name of an attribute and a set of attribute values of the attribute, by an inquiry in natural language. Each word in the sentence is classified into three types, that is, the attribute name, the attribute value, and the word that is neither the attribute name nor the attribute value. For words classified in the attribute name, the attribute that is classified The names are collectively stored as a response attribute name, and for the words classified into the attribute values, the classified attribute values are grouped with the corresponding attribute name and stored as a condition attribute value set group. , After the classification and storage of all the constituent words in the inquiry sentence are completed, it is stored as the condition attribute value set group,
If all combinations of the attribute value and the attribute name exist in the attribute table, the attribute value of the attribute name stored as the response attribute name is output as a response to the inquiry, and the condition attribute value set Natural language that outputs as a response to the inquiry that there is no data to be inquired unless all pairs of the attribute value and the attribute name stored as a group are present in the attribute table It is an interpretation method.

A second aspect of the present invention provides the first invention with respect to a query in natural language by allowing the existence of a plurality of attribute tables each having the same attribute combination.
The attribute table in which the response attribute name and the condition attribute value set group are derived from the inquiry sentence by the natural language interpretation method of the invention, and all the pairs of the attribute value and the attribute name in the condition attribute value set group exist Is selected, and the attribute value of the attribute name in the response attribute name of the selected attribute table is output as a response to the inquiry.

A third aspect of the present invention allows a plurality of types of attribute tables each of which has a combination of different attributes to exist in the first aspect of the invention, so that in response to a natural language inquiry, the plurality of types of attribute tables are first described. The type of the attribute table to be searched is indicated from the attribute table, the inquiry sentence is interpreted by the natural language interpretation method of the second invention for the indicated type of attribute table, and if necessary, the attribute A natural language interpretation method for outputting an answer to the inquiry by repeating selection of a table type and interpretation of a query sentence by the natural language interpretation method of the second invention for the attribute table of the selected type. .. The fourth invention is the first
In the invention, there is an attribute name alias definition command for defining a name synonymous with the attribute name of the attribute table, and the attribute name alias definition command indicates the alias of the attribute name in the attribute table before the query sentence is interpreted. If the words in the input sentence are in the attribute name alias set group when the words in the inquiry sentence are classified, the aliases are grouped with the corresponding attribute name and stored as an attribute name alias set group. , The natural language interpretation method in which the attribute names are regarded as the same as the corresponding attribute names and are classified into the corresponding attribute names.

A fifth aspect of the present invention is the first aspect of the present invention, which has an attribute value alias definition command for defining a name synonymous with an attribute value in the attribute table, and the attribute value alias definition command is provided before interpreting a query statement. Asks for an alias of the attribute value in the attribute table, and stores the alias together with the corresponding attribute value as a group of attribute value alias groups. Is a natural language interpretation method in which if the word is in the attribute value alias set group, it is considered to be the same as the corresponding attribute value and is classified into the corresponding attribute value.

[0011]

In the first aspect of the present invention, the data to be inquired in natural language is represented by the attribute table. The attribute table is a set of pairs of attribute names and attribute values of some defined attributes. For example, the collection table data is represented by an attribute table as shown by 61 in FIG. By expressing the data to be searched for a query in natural language as an attribute table, in order to refer to the target data, "whether the attribute table contains the target data" and the attribute table stores the target data. If included,
The two points of "which attribute on the attribute table is the target data" should be derived from the inquiry sentence.

Therefore, in the first invention, each word in the inquiry sentence is classified into three, that is, an attribute name, an attribute value, and a word that is neither an attribute name nor an attribute value. For words classified as an attribute name, Save the attribute name as a response attribute name,
For the words classified into the attribute value, the attribute value and the attribute name corresponding to the attribute value are grouped together and stored as a group of conditional attribute value groups. For example, the collection attribute table 61
On the other hand, if the user gives a query sentence "Who is the author who is disqualified as a human?", "Author" is stored as the response attribute name and "(Book title, human disqualified)" is stored as the set of condition attribute values. It "Whether the attribute table contains the target data" is considered to be determined by the combination of the word classified into the attribute value and the attribute name corresponding to the attribute value into which the word is classified. It is considered that the word classified into the attribute name determines "whether the attribute is the target data". Therefore, if all the pairs of attribute values and attribute names stored as a group of condition attribute values exist in the attribute table, it is determined that the target data exists in the attribute table and stored as the response attribute name. Outputs the attribute value of the attribute name as a response to the query, and unless all the pairs of attribute value and attribute name stored as a set of conditional attribute value pairs exist in the attribute table, the target data is not stored in the attribute table. It is determined that the data does not exist, and the fact that the data to be inquired does not exist is output as the answer to the inquiry. In the case of the above example sentence, "(book title, human disqualification)" stored as a group of conditional attribute values exists in the attribute table 61, so it is determined that the target data exists in the attribute table 61, and the response of the attribute table 61 is returned. The answer is "Osamu Dazai", which is the value of "book" saved as the attribute name.

In the second invention, the object of inquiry in natural language is expressed as a plurality of attribute tables which are composed of the same combination of attributes but differ only in attribute value. For example,
In the case of the data of the collection, it becomes like 62 and 63 of FIG. By applying the analysis method of the first invention to each of these attribute tables, the answer to the inquiry in natural language is created. That is, a response attribute name and a condition attribute value set group are derived from a query sentence by the query interpretation method of the first aspect of the invention, and which attribute table is the target of the set of attribute value and attribute name in the condition attribute value set group. Considering that it is a condition that determines whether or not to have data, select all the attribute tables that have all the pairs of attribute values and attribute names in the condition attribute value set group, and select the attributes in the response attribute name of the selected attribute table. The attribute value of first name is output as the answer to the inquiry. For example, in the case of the inquiry sentence "Who is the author of human disqualification?" For the collection of the attribute table of the collection of books 62 and 63 in Fig. 6, "(book title, human disqualification) saved in the condition attribute value set group. Is selected, that is, the attribute table 62 is selected, and “Dazai Osamu”, which is the value of “author” stored as the response attribute name in the attribute table 62, is answered.

In the third invention, the existence of a plurality of types of attribute tables is further permitted. That is, the target of the inquiry in natural language is expressed as a set of a plurality of types of attribute tables each including a combination of different attributes. This expression method can handle a wide range of data. For example, if an attribute table of “writer” is introduced in addition to the attribute table of the collection as shown by 64 and 65 in FIG. 6, the inquiry sentence can be interpreted using the knowledge about the author. First, the user is asked to select the type of attribute table, the natural language interpretation method of the second invention is applied to the selected type of attribute table, and the selection of this type of attribute table is performed until the user approves. The inquiry sentence is processed by repeating the interpretation of the second invention. For example, if an inquiry sentence “where is the address of an author who has been disqualified as a human being” is given to the attribute tables 64 and 65 of the book collection and the writer, if the user first selects the attribute table of the book collection, According to the natural language interpretation method of the invention of claim 1, "human disqualified author = Osamu Dazai" is obtained from the attribute table 64, and if the user next selects the attribute table of the writer, the target data "Address of Osamu Dazai" is selected from the attribute table 65. = Aomori ”is obtained.

According to the fourth aspect of the invention, an attribute name alias definition command is prepared to ask the user to register an alias for the attribute name.
This makes it possible to refer to a target attribute by its alias even if you forget the attribute name. For example, if "writer" is registered as an alias of "author" by the attribute name alias definition command in the attribute table 61 of FIG. 6,
The inquiry sentence "Who is the disqualified human being?" Can also be correctly interpreted by the natural language interpretation method of the first invention.

A fifth aspect of the present invention prepares an attribute value alias definition command and performs the same processing as that of the fourth aspect on the attribute value. For example, if "Dazai" is registered as an alias for "Osamu Dazai" by the attribute value alias definition command in the attribute table 65 of the writer in FIG. 6, the inquiry sentence "Where is Dazai's address?" Can also be correctly interpreted by the natural language interpretation method of the first invention.

[0017]

EXAMPLES The first invention will be described with reference to examples. Prepare an attribute table consisting of three attributes: book title, author, and publisher. The attribute name of the attribute / book name is the book name, the attribute value is human disqualification, the attribute name of the attribute / author is author, the attribute value is Osamu Dazai,
Attribute-Publisher attribute name is publisher, and attribute value is Shinchosha. This attribute table is shown at 61 in FIG. For this attribute table 61, "who is the disqualified human being?"
The processing flow of the first invention will be described with reference to the processing of the inquiry sentence.

FIG. 1 shows a flowchart of the first invention. First, in block 11, the input sentence is morphologically analyzed and converted into a list of words. [Human disqualification,
Who is the author?]. Regarding the morphological analysis, the present invention does not claim any new method, and therefore the existing method is used. Next, in block 12, it is confirmed that the list is not empty, and in block 13, the first word "human disqualification" is taken out. Since "human disqualification" is an attribute value, block 14 fails and block 16 succeeds. Therefore, in block 17, "human disqualification" is paired with its attribute name "book name" to create a group of conditional attribute values (book name, Saved as human disqualification). Next, returning to block 12 again, it is confirmed that the list is not empty, a new head word is extracted from the list in block 13, and the same processing as described above is repeated until the list becomes empty. Neither attribute name nor attribute value of "no", "author" is attribute name, and "ha", "who", "is", and "ka" are neither attribute name nor attribute value. “Author” is stored as the name, and “(book title, human disqualification)” is stored as the group of conditional attribute values. Next, in block 18, since the set of the attribute value and the attribute name stored as the condition attribute value set group is “(book title, human disqualification)” and exists in the attribute table 61, the attribute table 6
Since 1 contains the target data and the attribute name saved as the response attribute name is the author, in block 19,
“Osamu Dazai”, which is the value of the attribute / author in the attribute table 61, is output as an answer. If the set of attribute values and the attribute name stored as the set of condition attribute value sets does not exist in the attribute table, a response indicating that the target data does not exist is output in block 20.

FIG. 2 shows a flowchart of the second invention. As an attribute table consisting of three attributes of a book title, an author, and a publisher, FIG. 6 shows an attribute table 62 "Book title: disqualified human, author: Osamu Dazai, publisher: Company A" and "Book title:
It is assumed that an attribute table 63 “Kinkakuji, author: Yukio Mishima, publisher: Company B” is prepared, and the object of inquiry in natural language is the attribute table 62 and the attribute table 63. The processing of the inquiry sentence "who is the author who is disqualified as a human being" is explained. First, in block 21, one attribute table is retrieved. It is assumed that the attribute table 62 is taken out. Next, in block 22, the process of the first invention is performed on the attribute table 62. "Osamu Dazai" is output as an answer. Next, in block 23,
See if there are any attribute tables that have not yet been processed. Since the attribute table 63 is an attribute table that has not been processed yet, the process returns to the block 21 again and the attribute table 6
3 is taken out, and in block 22, the process of the first invention is performed on the attribute table 63. It is represented by not outputting any answer to the effect that "target data does not exist". Proceed to block 23, confirm that there is no unprocessed attribute table, and end.

FIG. 3 shows a flowchart of the third invention. Attribute table 64 of type 1 consisting of three attributes of book title, author, and publisher in FIG. 6 "type of book: name disqualified by human, author: Osamu Dazai, publisher: company A", and two attributes of name and address The processing of the inquiry sentence "Where is the address of the author who is disqualified as a human being" is explained for the object of the inquiry, which is a collection of the attribute table 65 "name: Osamu Dazai, address: Aomori". First, in block 31, the type of attribute table is selected according to a user's instruction. It is assumed that the type 1 is selected. Next, in block 32, the query sentence is interpreted according to the second invention with respect to the selected type 1 attribute table. “Author = Osamu Dazai” is obtained from the attribute table 64, and “author = Osamu Dazai” is output. Next, at block 33, the user is queried for exit. If the user wishes to continue the process here, it returns to block 31 again,
Since "author = Osamu Dazai" is obtained, the sentence "where is the address of Osamu Dazai" converted from "the author of disqualified human being" in the original inquiry sentence to "Osamu Dazai"? Now, assuming that type 2 is selected as the type of attribute table, in block 32, the interpretation method of the second invention is applied,
“Address = Aomori” is obtained from the attribute table 65, “address = Aomori” is output as an answer, and the process ends with the user's approval in block 33.

FIG. 4 shows a flowchart of the fourth invention. In FIG. 6, an attribute table 61 "book name: human disqualification, author: Osamu Dazai, publisher: company A", which is composed of three attributes of a book name, an author, and a publisher, is prepared. This attribute table 61
On the other hand, first, a case will be described as an example in which the user is instructed by the attribute name alias definition command to be “author = author” and then the inquiry sentence “who is the disqualified author” is processed. First, it is assumed that the user gives an input of “author = author”. In block 41, the input is determined to be the definition of an alias, and in block 42, the attribute name alias definition command saves (author, writer) as a set of attribute name alias groups, and ends. Next, suppose that the user inputs a query sentence "who is the disqualified author?". In block 41, the input is not defined as an alias, and in block 43 the input sentence is converted into a list. [Who is the author's disqualification?]. For each word in this list,
In blocks 44 to 411, the same processing as in the first aspect of the invention is performed. However, the following process is performed for the "writer" in the list. Since "writer" exists in the attribute name alias group group as (author, writer), block 46 is successful, and "author" is saved as a response attribute name in block 47. When the processing of each word in the list is completed, the processing of the output unit of FIG. 1 is performed in block 412, "Dazai Osamu" is output, and the processing ends.

FIG. 5 shows a flowchart of the fifth invention. In FIG. 6, an attribute table 65 “name: Osamu Dazai, address: Aomori” having two attributes of name and address is prepared. For this attribute table 65, an example will be described in which, first, the attribute value alias definition command instructs the user to “Dazai Osamu = Dazai”, and then the inquiry sentence “where is Dazai's address” is processed. To do. First, it is assumed that the user inputs “Dazai Osamu = Dazai”. In block 51, it is determined that the input is a definition of an alias, and in block 52, the attribute value alias definition command saves (Dazai Osamu, Dazai) as an attribute value alias group set, and ends. Next, it is assumed that the user inputs an inquiry sentence "Where is the address of Osamu Dazai?" In block 51, said input is not an alias definition,
Proceeding to block 53, the input sentence is converted into a list. For each word in the list [where is Dazai's, address, where is?], From block 54 to block 511, the same processing as in the first invention is performed. However,
The following process is performed for "Dazai" in the list.
Since "Dazai" exists in the attribute value alias group group as (Dazai Osamu, Dazai), block 56 is successful, and in block 57, "(name, Dazai Osamu)" is saved as a conditional attribute value group group. After processing each word in the list, block 5
In 12, the processing of the output unit of FIG. 1 is performed, “Aomori” is output, and the processing ends.

[0023]

According to the present invention, a query in natural language can be interpreted by a simple method, which is an effective means for putting a natural language interface system into practical use.

[Brief description of drawings]

FIG. 1 is a flowchart showing an embodiment of the first invention.

FIG. 2 is a flow chart showing an embodiment of the second invention.

FIG. 3 is a flowchart showing an embodiment of the third invention.

FIG. 4 is a flowchart showing an embodiment of the fourth invention.

FIG. 5 is a flow chart showing an embodiment of the fifth invention.

FIG. 6 is an auxiliary conceptual diagram for explaining an embodiment of the first, second, third, fourth and fifth inventions.

[Explanation of symbols]

 61-65 attribute table

Claims (5)

[Claims] 1. A method of referring to a certain attribute value in an attribute table, which is a set of a set of an attribute name of an attribute and an attribute value of the attribute, by a natural language inquiry, wherein each word in the inquiry sentence is referred to as the attribute name and the Attribute values and words that are neither the attribute name nor the attribute value are classified into three types, and for words classified into the attribute name, the classified attribute names are collectively stored as a response attribute name. For the words classified into the attribute values, the classified attribute values are grouped together with the corresponding attribute name and stored as a group of conditional attribute values, and all the constituent words in the inquiry sentence are stored. After the classification and storage are completed, if all the combinations of the attribute value and the attribute name stored as the condition attribute value set group exist in the attribute table, they are stored as the response attribute name. The attribute value of the attribute name is output as a response to the inquiry, and if all the pairs of the attribute value and the attribute name stored as the condition attribute value pair group do not exist in the attribute table, A natural language interpretation method for outputting, as an answer to the inquiry, the fact that there is no data to be inquired about. 2. The natural language interpreting method according to claim 1, wherein the natural language interpreting method according to claim 1 allows the existence of a plurality of attribute tables having the same combination of attributes to respond to a query in natural language. The response attribute name and the condition attribute value set group are derived from the inquiry sentence, and all the attribute tables in which the set of the attribute value and the attribute name in the condition attribute value set group all exist are selected and selected. A natural language interpretation method for outputting the attribute value of the attribute name in the attribute response attribute name of the attribute table as a response to the inquiry. 3. The natural language interpretation method according to claim 1, wherein a plurality of types of attribute tables each of which has a combination of different attributes are allowed to exist, so that in response to a natural language inquiry, the plurality of types of attribute tables From the attribute table, the type of the attribute table to be searched is indicated, the inquiry sentence is interpreted by the natural language interpretation method according to claim 2 for the attribute table of the indicated type, and if necessary, the 3. A natural language interpretation method for outputting a response to the inquiry by repeating the selection of the type of attribute table and the interpretation of the inquiry sentence by the natural language interpretation method according to claim 2 for the selected type of attribute table. 4. An attribute name alias definition command for defining a name synonymous with an attribute name in the attribute table, wherein the attribute name alias definition command is executed by the attribute name alias definition command before interpreting the inquiry sentence. , The aliases are grouped together with the corresponding attribute name and stored as an attribute name alias group group, and when classifying the words in the inquiry sentence, the words in the input sentence are added to the attribute name alias group group. The natural language interpretation method according to claim 1, wherein if there is, the attribute name is considered to be the same as the corresponding attribute name and is classified into the corresponding attribute name. 5. An attribute value alias definition command for defining a value synonymous with an attribute value in the attribute table, the alias of the attribute value in the attribute table being defined by the attribute value alias definition command before interpreting a query statement. , The aliases are grouped with corresponding attribute values and stored as an attribute value alias group set, and when the words in the inquiry sentence are classified, the words in the input sentence are added to the attribute value alias group group. The natural language interpretation method according to claim 1, wherein if there is, the attribute value is considered to be the same as the corresponding attribute value and is classified into the corresponding attribute value.