JPH0346868B2 - - Google Patents

Description

[Detailed Description of the Invention] (A) Field of Industrial Application The present invention relates to a Japanese sentence processing device, and particularly to Japanese sentence processing in a process such as, for example, responding to an inquiry in Japanese and generating a command corresponding to the inquiry. In the device, the possibility and strength of modification relationships between clauses are defined on the world model, and this is used to determine the modification relations between clauses. The present invention relates to a Japanese sentence processing device that defines at least three groups of verbs in terms of handling, and examines modification relationships between adjacent or nearby clauses corresponding to each group.

(B) Prior Art and Problems to be Solved by the Invention Conventionally, Japanese sentence processing devices have been considered that allow a data processing system to understand Japanese sentences. In such cases, it would be unreasonable to try to determine the modifying relationships between clauses based only on grammatical rules, so the use of semantic information should be considered. When using such semantic information, it is necessary to use it under clear principles. Therefore, for example, a ``sales world model'' is prepared in a field related to ``sales,'' and a ``university world model'' is prepared in a field related to ``university.'' is prepared, the semantic relationships between words are defined within the world model, and by using the world model, the above semantic information is introduced under a predetermined principle. This is being taken into account.

Although the above device using a world model is extremely effective, one technical problem is how to handle verbs. In other words, the world model itself has been known for a long time, and for example, if a class object is related to a noun, such as preparing a ``price object'' as a class object corresponding to a noun such as the word ``price'' or ``price.'' However, there may be cases where it is convenient to group together verbs such as ``sell'' or ``sell'' as a ``sales object'' and treat them as class objects on the world model.

(C) Means for Solving the Problem In order to solve the above-mentioned points, the present invention deals with the handling of verbs on the world model, at least three types such as class verbs, arc verbs, and general verbs. A Japanese sentence processing device is provided that divides sentences into groups and performs processing corresponding to each group. Therefore, the Japanese sentence processing device of the present invention divides the input Japanese sentence into clauses, and performs an understanding process on the input Japanese sentence based on the modification relationship between the clauses. A world model that prepares objects corresponding to at least nouns and some verbs for fields related to the input Japanese sentence, and expresses attribute relationships and superior/subordinate relationships between objects. and prepare a dictionary that describes the correspondence between the words that make up the Japanese sentence and the objects, and index the dictionary using the segmented phrases,
Using the correspondence obtained based on the dictionary,
and an analysis rule processing unit that refers to the world model and examines modification relationships in Japanese sentences, and the analysis rule processing unit is configured to refer to the world model and perform a process of combining a plurality of clauses. The verb modification processing mechanism determines whether the verb currently being processed belongs to the first group of verbs described on the world model as the object, or whether the verb modification processing mechanism Does it belong to the second group of verbs assigned to attributes between objects on the model?
A third group that does not belong to either the first or second group above.
It is characterized by being structured to examine modifying relationships with adjacent or nearby clauses based on whether the verb belongs to a group. This will be explained below with reference to the drawings.

(D) Embodiment Fig. 1 shows the configuration of an embodiment of the present invention, Fig. 2 A and B
are explanatory diagrams explaining the world model, Figure 3 A,
B shows an embodiment of the configuration of a dictionary used in the present invention, FIGS. 4A to 4D show an embodiment of processing up to clause integration processing when examining the modification relationships between clauses, and FIG. FIG. 1 shows an embodiment of processing in the verb modification relation mechanism shown in FIG. 6, an example of processing to which the present invention is applied, and FIG. 7 shows an explanatory diagram illustrating one of the advantages of using arc verbs.

In FIG. 1, 1 represents an input section into which a Japanese sentence such as "Please tell me the price of the product sold on June 18th" is input, for example. 2 is a word segmentation processing unit that stores, for example, the price of the product sold on June 18th.
The input Japanese sentence is divided into words, such as /tell me/please. Reference numeral 3 is a dictionary in which information such as parts of speech (Vb, N, etc.) and objects listed in the world model are described corresponding to words, as will be described later with reference to FIG. Reference numeral 4 denotes an instantiation processing unit, which performs processing to obtain each instance as described later with reference to FIG. 4B. Reference numeral 5 indicates a clause integration processing unit, 6 indicates an analysis rule processing unit, 7 indicates a world model, 8 indicates a noun clause relation modification processing mechanism, and 9 indicates a verb modification processing mechanism.

World model 7 defines entities in the target field and the relationships that exist between them as objects (classes).
Figure 2A shows an example of a sales world model corresponding to a field related to sales, and Figure 2B shows a sales world model corresponding to a field related to sales. Examples of university world models corresponding to fields related to are shown. Furthermore, the second
In the illustrated world model, "store" is circled with a circle.
and "products" are objects (class objects are abbreviated as objects), solid line arrows in the diagram indicate that objects have attribute relationships with each other, and dotted line arrows in the diagram indicate that objects have a superior/subordinate relationship with each other. It shows. Also the second
The "top object" shown in FIG. A may be considered to be the object corresponding to the top of all objects in the world model.

In the world model used in the present invention, some verbs are grouped into objects and described on the world model, such as the "sales object" shown in FIG. 2A. Of course, some verbs are assigned in accordance with attribute relationships, such as the verbs ``belong,''``live,'' and ``teach,'' shown in FIG. 2B. Although it is not clear in Figure 2, under the world model there is a direct relationship with general objects, such as the verb ``teach'' under the ``sales world model'' related to Figure 2A. There are also a small number of verb groups that are treated in relation to the topmost illustrated object.

In this specification, verbs that are treated as objects, such as "sell" in the illustrated world model of FIG. 2A, are called "class verbs." Also, verbs that are assigned in correspondence with attribute relationships, such as "belong" in the illustrated world model of FIG. 2B, are called "arc verbs." Furthermore, common verbs such as the verb "teach" in the world model shown in FIG. 2A are called "general verbs." By classifying and handling verbs in this way, the verbs can be handled efficiently when checking for modification relationships.

When a world model is prepared in each field in the form shown in FIG. 2A or B, a dictionary is created correspondingly in the form shown in FIG. 3A or B. FIG. 3A shows an example of a dictionary corresponding to the "sales world model" shown in FIG. 2A. For example, the word "retail store" is a noun N and is clarified to correspond to a "store object." It is also clear that the word ``sell'' is a class verb (Vb class), corresponds to ``sales object,'' and is used in conjunction with clauses such as ``product'' and ``store.'' Furthermore, it is made clear that the word ``teach'' is a general verb (Vb general) and is used in the form ``**wo''. Furthermore, the illustration in FIG. 3B clarifies that the word "belongs to" is an arc verb (Vb arc) and is used in conjunction with phrases such as "professor" and "to an academic society."

In the configuration shown in FIG.
indexes the dictionary 3 for each classified word. As a result, when a Japanese sentence as shown in FIG. 4A is input, the dictionary shown in FIG. 3A is indexed, and dictionary 3 is indexed as shown in the additional information shown in FIG. 4A. The information obtained when doing so is added to the word. Then, the instantiation processing unit 4 shown in FIG. 1 registers each word as an instance (individual information) as indicated by the small circles shown in FIG. 4B. In this case, "June 18" corresponds to the "date object"
As shown in FIG. 4C, values such as "year", "month", and "day" are set in relation to the correspondingly created date instance.

The phrase integration processing unit 5 shown in FIG. 1 performs the integration as shown in FIG. ,
Extract phrases such as ``product'', ``price'', and ``please tell me''. The results are then transferred to the analysis rule processing section 6 shown in FIG.

The inside of the analysis rail processing section 6 is as shown in FIG.
Noun interclause modification processing mechanism 8 and verb modification processing mechanism 9
It is broadly divided into. The inter-noun clause modification processing mechanism 8 processes the modification relation between noun clauses, such as processing the modification relation between the noun clause "product" and the noun clause "name", such as "the name of the product". It is a mechanism that processes In the noun inter-clause modification processing mechanism 8, while indexing the world model 7,
() When two objects have a superior/subordinate relationship, the lower object is chosen. () When two objects have an attribute relationship, the originating object becomes the center of meaning. Applying the principle of centering, which is interpreted as (), the identification rule that combines two clauses that each contain two objects in a superior/subordinate relationship into one, and () that each contains two objects in an attribute relationship. A process of combining a plurality of phrases under one tree is performed using a joining rule that combines two phrases into one. However, since this processing itself is not directly related to the present invention, the detailed explanation will be left to the "Japanese Sentence Understanding Processing Device" filed on the same day as the present application.
In this specification, the explanation will be omitted.

The present invention directly addresses the problem of how to handle verbs in the process of examining modification relationships between clauses, and includes a verb modification processing mechanism in the analysis rule processing unit 6 shown in FIG. 9 is prepared.

As mentioned above, in the description of the world model 7 used in the present invention, verbs are defined as () class verbs, ()
I try to divide them into three groups: arc verbs and () general verbs.

In the case of this class verb, for example, regarding ``to sell'', in a left-horizontal sentence, such as ``on June 18th / at / sold / products'', multiple clauses on the left side, ``June 18th It may be related to phrases such as ``on the day'' or ``at the ○× store,'' and it may also be related to the phrase ``product'' on the right side. In addition, in the case of arc verbs, for example, ``belong to'' may be related to one clause on the left or one clause on the right, such as ``professor/belongs/to an academic society''. Furthermore, in the case of general verbs,
For example, it may be related to one clause on the left, such as ``Please tell me the price.'' Considering these matters, in one embodiment of the present invention, as shown in FIG. 5, the verb currently being processed is the class verb, the arc verb, or the general verb Depending on the crab,
The processing methods are different. In addition,
"Left noun modification processing" shown in Figure 5 is a process in which the noun clause located on the left side of the verb becomes an attribute of the verb on the world model, and if there is a case condition in the dictionary, the noun clause modifies the verb. This indicates that it will be processed as In addition, "right noun modification processing" is
This indicates that if a noun clause located on the right side of the verb has an attribute relationship with the verb on the world model, it is treated as being modified by the verb. When the process A in the figure is completed, the noun clauses for which the modification process has been completed are grouped into the class verb, and then the process of checking whether there is a noun clause that can be modified on the left side (process B in the diagram) is performed. In the case of process C in the figure, the modification process is performed only on one noun clause on the left side.

FIG. 6 shows an example of processing to which the present invention is applied. In the case shown, regarding the class verb "sold", "on June 18th" is modified by the left noun shown in Figure 5, and "of the product" is modified by the right noun shown in Figure 5. It has been shown that
Then, it is summarized as “Sold to...”
'of' is in a modifying relationship (modifying relationship between noun clauses) with the clause 'price', and then 'the price of... sold to...' which is combined into one, is the general verb clause 'teaching'. "Please" is combined with the general verb left noun modification relationship, and the whole is combined into one tree. That is, the modification relationships between clauses in the input Japanese sentence are clarified, and the Japanese sentence is understood by the data processing system.

In the case of the present invention, the arc verbs described in connection with the world model shown in FIG. 2B are also processed as shown in FIG. 5. For this reason, for example, if the world model shown in Fig. 2B further exists as shown in Fig. 7, and the input Japanese sentence is ``gives a lecture/speaks/person''. However, the meaning of the sentence should be made clearer by changing it to ``professor who gives a lecture,'' or ``person who listens to a lecture'' to ``student who listens to a lecture.'' It becomes possible.
These include "professor" and "student" as shown in Figure 7.
is in a subordinate relationship to "person", and "lecture" is associated with "speak" in the attribute relationship, or "professor" is associated with "listen" in the attribute relationship. This is to take advantage of the fact that the students are ``students.''

(E) Effects of the Invention As explained above, according to the present invention, handling of verbs becomes simpler and easier when processing Japanese sentences using a world model. The user can easily check the allowable range of sentences that can be input into the system on the world model, and the system developer can also easily expand the world model.

[Brief explanation of drawings]

Fig. 1 shows the configuration of an embodiment of the present invention, Fig. 2 A and B
are explanatory diagrams explaining the world model, Figure 3 A,
B shows an embodiment of the configuration of a dictionary used in the present invention, FIGS. 4A to 4D show an embodiment of the processing up to clause integration processing when examining the modification relationships between clauses, and FIG. FIG. 6 shows an example of processing to which the present invention is applied, and FIG. 7 shows an explanatory diagram illustrating one of the advantages of using arc verbs. In the figure, 1 is the input section, 2 is the word division processing section, 3 is the dictionary, 4 is the instantiation processing section, 5 is the clause integration processing section, 6 is the analysis rule processing section, 7 is the world model, and 8 is the noun clause interval Modification processing mechanism 9 represents a verb modification processing mechanism.

Claims (1)

[Scope of Claims] 1. A Japanese sentence processing device that divides an input Japanese sentence into clauses and performs an understanding process on the input Japanese sentence based on the modification relationship between the clauses, Prepare a world model that expresses attribute relationships and superior/subordinate relationships between objects by preparing objects corresponding to some verbs as well as objects corresponding to at least nouns in fields related to the above, and Prepare a dictionary that describes the correspondence between the words constituting the Japanese sentence and the objects, and index the dictionary according to the segmented clauses,
Using the correspondence obtained based on the dictionary,
and an analysis rule processing unit that refers to the world model and examines modification relationships in Japanese sentences, and the analysis rule processing unit is configured to refer to the world model and perform a process of combining a plurality of clauses. The verb modification processing mechanism determines whether the verb currently being processed belongs to the first group of verbs described on the world model as the object, or if the verb modification processing mechanism Does it belong to the second group of verbs assigned to attributes between objects on the model?
A third group that does not belong to either the first or second group above.
A Japanese sentence processing device characterized in that it is configured to check modification relationships with adjacent or nearby clauses based on whether the verb belongs to a group.